Patchwork [09/27] UBUNTU: SAUCE: fsl_qbman: Add drivers for Freescale DPAA Qman & Bman

login
register
mail settings
Submitter Benjamin Collins
Date July 13, 2011, 1:38 p.m.
Message ID <a7f4f3a5000b82cc7b35398878a1bb73f7a045c6.1339455421.git.bcollins@ubuntu.com>
Download mbox | patch
Permalink /patch/164270/
State New
Headers show

Comments

Benjamin Collins - July 13, 2011, 1:38 p.m.
This patch is being maintained and will eventually be merged upstream by
Freescale directly. The powerpc-e500mc flavour uses this.

Signed-off-by: Geoff Thorpe <Geoff.Thorpe@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Jeffrey Ladouceur <Jeffrey.Ladouceur@freescale.com>
Signed-off-by: Haiying Wang <Haiying.Wang@freescale.com>
Signed-off-by: Ben Collins <bcollins@ubuntu.com>
---
 arch/powerpc/Kconfig                            |    5 +
 arch/powerpc/platforms/85xx/Kconfig             |    3 +
 drivers/staging/Kconfig                         |    2 +
 drivers/staging/Makefile                        |    1 +
 drivers/staging/fsl_qbman/Kconfig               |  278 +++
 drivers/staging/fsl_qbman/Makefile              |   23 +
 drivers/staging/fsl_qbman/bman_config.c         |  688 +++++++
 drivers/staging/fsl_qbman/bman_debugfs.c        |  121 ++
 drivers/staging/fsl_qbman/bman_driver.c         |  586 ++++++
 drivers/staging/fsl_qbman/bman_high.c           | 1044 ++++++++++
 drivers/staging/fsl_qbman/bman_low.h            |  494 +++++
 drivers/staging/fsl_qbman/bman_private.h        |  149 ++
 drivers/staging/fsl_qbman/bman_test.c           |   57 +
 drivers/staging/fsl_qbman/bman_test.h           |   92 +
 drivers/staging/fsl_qbman/bman_test_high.c      |  181 ++
 drivers/staging/fsl_qbman/bman_test_thresh.c    |  197 ++
 drivers/staging/fsl_qbman/dpa_sys.h             |  324 ++++
 drivers/staging/fsl_qbman/dpa_uio.c             |  231 +++
 drivers/staging/fsl_qbman/qman_config.c         | 1010 ++++++++++
 drivers/staging/fsl_qbman/qman_debugfs.c        | 1311 +++++++++++++
 drivers/staging/fsl_qbman/qman_driver.c         |  541 ++++++
 drivers/staging/fsl_qbman/qman_fqalloc.c        |  264 +++
 drivers/staging/fsl_qbman/qman_high.c           | 2361 +++++++++++++++++++++++
 drivers/staging/fsl_qbman/qman_low.h            | 1186 ++++++++++++
 drivers/staging/fsl_qbman/qman_private.h        |  292 +++
 drivers/staging/fsl_qbman/qman_test.c           |   61 +
 drivers/staging/fsl_qbman/qman_test.h           |   85 +
 drivers/staging/fsl_qbman/qman_test_errata.c    |  248 +++
 drivers/staging/fsl_qbman/qman_test_high.c      |  222 +++
 drivers/staging/fsl_qbman/qman_test_hotpotato.c |  497 +++++
 drivers/staging/fsl_qbman/qman_utility.c        |  131 ++
 include/linux/fsl_bman.h                        |  471 +++++
 include/linux/fsl_qman.h                        | 1761 +++++++++++++++++
 33 files changed, 14917 insertions(+)
 create mode 100644 drivers/staging/fsl_qbman/Kconfig
 create mode 100644 drivers/staging/fsl_qbman/Makefile
 create mode 100644 drivers/staging/fsl_qbman/bman_config.c
 create mode 100644 drivers/staging/fsl_qbman/bman_debugfs.c
 create mode 100644 drivers/staging/fsl_qbman/bman_driver.c
 create mode 100644 drivers/staging/fsl_qbman/bman_high.c
 create mode 100644 drivers/staging/fsl_qbman/bman_low.h
 create mode 100644 drivers/staging/fsl_qbman/bman_private.h
 create mode 100644 drivers/staging/fsl_qbman/bman_test.c
 create mode 100644 drivers/staging/fsl_qbman/bman_test.h
 create mode 100644 drivers/staging/fsl_qbman/bman_test_high.c
 create mode 100644 drivers/staging/fsl_qbman/bman_test_thresh.c
 create mode 100644 drivers/staging/fsl_qbman/dpa_sys.h
 create mode 100644 drivers/staging/fsl_qbman/dpa_uio.c
 create mode 100644 drivers/staging/fsl_qbman/qman_config.c
 create mode 100644 drivers/staging/fsl_qbman/qman_debugfs.c
 create mode 100644 drivers/staging/fsl_qbman/qman_driver.c
 create mode 100644 drivers/staging/fsl_qbman/qman_fqalloc.c
 create mode 100644 drivers/staging/fsl_qbman/qman_high.c
 create mode 100644 drivers/staging/fsl_qbman/qman_low.h
 create mode 100644 drivers/staging/fsl_qbman/qman_private.h
 create mode 100644 drivers/staging/fsl_qbman/qman_test.c
 create mode 100644 drivers/staging/fsl_qbman/qman_test.h
 create mode 100644 drivers/staging/fsl_qbman/qman_test_errata.c
 create mode 100644 drivers/staging/fsl_qbman/qman_test_high.c
 create mode 100644 drivers/staging/fsl_qbman/qman_test_hotpotato.c
 create mode 100644 drivers/staging/fsl_qbman/qman_utility.c
 create mode 100644 include/linux/fsl_bman.h
 create mode 100644 include/linux/fsl_qman.h

Patch

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index ac7cd68..e058e4a 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -708,6 +708,11 @@  config FSL_PAMU
 	help
 	  Freescale PAMU/IOMMU support
 
+config HAS_FSL_QBMAN
+	bool "Datapath Acceleration Queue and Buffer management"
+	help
+	  Datapath Acceleration Queue and Buffer management
+
 # Yes MCA RS/6000s exist but Linux-PPC does not currently support any
 config MCA
 	bool
diff --git a/arch/powerpc/platforms/85xx/Kconfig b/arch/powerpc/platforms/85xx/Kconfig
index 87306fa..d42394e 100644
--- a/arch/powerpc/platforms/85xx/Kconfig
+++ b/arch/powerpc/platforms/85xx/Kconfig
@@ -221,6 +221,7 @@  config P3041_DS
 	select HAS_RAPIDIO
 	select PPC_EPAPR_HV_PIC
 	select HAS_FSL_PAMU
+	select HAS_FSL_QBMAN
 	help
 	  This option enables support for the P3041 DS board
 
@@ -248,6 +249,7 @@  config P4080_DS
 	select HAS_RAPIDIO
 	select PPC_EPAPR_HV_PIC
 	select HAS_FSL_PAMU
+	select HAS_FSL_QBMAN
 	help
 	  This option enables support for the P4080 DS board
 
@@ -265,6 +267,7 @@  config P5020_DS
 	select HAS_RAPIDIO
 	select PPC_EPAPR_HV_PIC
 	select HAS_FSL_PAMU
+	select HAS_FSL_QBMAN
 	help
 	  This option enables support for the P5020 DS board
 
diff --git a/drivers/staging/Kconfig b/drivers/staging/Kconfig
index 05e33c7..5a76c35 100644
--- a/drivers/staging/Kconfig
+++ b/drivers/staging/Kconfig
@@ -132,4 +132,6 @@  source "drivers/staging/ipack/Kconfig"
 
 source "drivers/staging/gdm72xx/Kconfig"
 
+source "drivers/staging/fsl_qbman/Kconfig"
+
 endif # STAGING
diff --git a/drivers/staging/Makefile b/drivers/staging/Makefile
index 3b32dd9..298af85 100644
--- a/drivers/staging/Makefile
+++ b/drivers/staging/Makefile
@@ -58,3 +58,4 @@  obj-$(CONFIG_RAMSTER)		+= ramster/
 obj-$(CONFIG_USB_WPAN_HCD)	+= ozwpan/
 obj-$(CONFIG_USB_G_CCG)		+= ccg/
 obj-$(CONFIG_WIMAX_GDM72XX)	+= gdm72xx/
+obj-$(CONFIG_FSL_DPA)		+= fsl_qbman/
diff --git a/drivers/staging/fsl_qbman/Kconfig b/drivers/staging/fsl_qbman/Kconfig
new file mode 100644
index 0000000..3321379
--- /dev/null
+++ b/drivers/staging/fsl_qbman/Kconfig
@@ -0,0 +1,278 @@ 
+config FSL_DPA
+	bool "Freescale Datapath Queue and Buffer management"
+	depends on HAS_FSL_QBMAN
+	default y
+	select FSL_QMAN_FQ_LOOKUP if PPC64
+
+menu "Freescale Datapath QMan/BMan options"
+	depends on FSL_DPA
+
+config FSL_DPA_CHECKING
+	bool "additional driver checking"
+	default n
+	---help---
+	  Compiles in additional checks to sanity-check the drivers and any
+	  use of it by other code. Not recommended for performance.
+
+config FSL_DPA_HAVE_IRQ
+	bool
+	default y
+
+config FSL_DPA_CAN_WAIT
+	bool
+	default y
+
+config FSL_DPA_CAN_WAIT_SYNC
+	bool
+	default y
+
+config FSL_DPA_PIRQ_FAST
+	bool "fast-path processing via IRQ"
+	depends on FSL_DPA_HAVE_IRQ
+	default y
+	---help---
+	  By default, configure fast-path handling to be triggered by IRQ.
+	  Drivers can modify this at run-time, but for linux it makes sense
+	  for all handling to be triggered by IRQ, unless a special
+	  run-to-completion application is being built.
+
+config FSL_DPA_PIRQ_SLOW
+	bool "slow-path processing via IRQ"
+	depends on FSL_DPA_HAVE_IRQ
+	default y
+	---help---
+	  By default, configure slow-path handling to be triggered by IRQ.
+	  Drivers can modify this at run-time, but for linux it makes sense
+	  for all handling to be triggered by IRQ, unless a special
+	  run-to-completion application is being built.
+
+config FSL_DPA_PORTAL_SHARE
+	bool "allow portals to be affine to multiple CPUs"
+	default y
+	---help---
+	  If the kernel, via its device-tree, will be required to use a
+	  single Qman or Bman portal for multiple CPUs, then select this
+	  option. Note however that compiling with this option will result
+	  in a very slight (but non-zero) overhead even when the feature
+	  is not being used.
+
+config FSL_DPA_UIO
+	tristate "Export USDPAA portals via UIO"
+	depends on UIO
+	default y
+	---help---
+	  Portals are exported as UIO devices for use by USDPAA (User
+	  Space DataPath Acceleration Architecture) applications.
+
+config FSL_BMAN
+	bool "Freescale Buffer Manager support"
+	default y
+
+if FSL_BMAN
+
+config FSL_BMAN_PORTAL
+	bool "Bman portal support"
+	default y
+	---help---
+	  Compiles support to detect and support Bman software corenet portals
+	  (as provided by the device-tree).
+
+config FSL_BMAN_CONFIG
+	bool "Bman device management"
+	default y
+	---help---
+	  If this linux image is running natively, you need this option. If this
+	  linux image is running as a guest OS under the hypervisor, only one
+	  guest OS ("the control plane") needs this option.
+
+config FSL_BMAN_TEST
+	tristate "Bman self-tests"
+	depends on FSL_BMAN_PORTAL
+	default n
+	---help---
+	  This option compiles self-test code for Bman.
+
+config FSL_BMAN_TEST_HIGH
+	bool "Bman high-level self-test"
+	depends on FSL_BMAN_TEST
+	default y
+	---help---
+	  This requires the presence of cpu-affine portals, and performs
+	  high-level API testing with them (whichever portal(s) are affine to
+	  the cpu(s) the test executes on).
+
+config FSL_BMAN_TEST_THRESH
+	bool "Bman threshold test"
+	depends on FSL_BMAN_TEST
+	default y
+	---help---
+	  Multi-threaded (SMP) test of Bman pool depletion. A pool is seeded
+	  before multiple threads (one per cpu) create pool objects to track
+	  depletion state changes. The pool is then drained to empty by a
+	  "drainer" thread, and the other threads that they observe exactly
+	  the depletion state changes that are expected.
+
+config FSL_BMAN_DEBUGFS
+	tristate "Bman debugfs interface"
+	depends on FSL_BMAN_PORTAL && DEBUG_FS
+	default y
+	---help---
+	This option compiles bman debugfs code for Bman.
+
+endif # FSL_BMAN
+
+config FSL_QMAN
+	bool "Freescale Queue Manager support"
+	default y
+
+if FSL_QMAN
+
+config FSL_QMAN_PORTAL
+	bool "Qman portal support"
+	default y
+	---help---
+	  Compiles support to detect and support Qman software corenet portals
+	  (as provided by the device-tree).
+
+config FSL_QMAN_BUG_AND_FEATURE_REV1
+	bool "workarounds for errata and missing features in p4080 rev1"
+	depends on FSL_QMAN_PORTAL
+	default y
+	---help---
+	  If this option is selected, the driver will be compiled with
+	  workarounds for errata as well as feature limitations (relative to
+	  more recent parts) of p4080 rev1. On unaffected revisions, this
+	  support incurs only a negligable overhead, typically only a couple of
+	  instructions per non-fast-path operation (the fast-path operations are
+	  unaffected).
+
+	  If in doubt, say Y.
+
+config FSL_QMAN_POLL_LIMIT
+	int
+	default 32
+
+config FSL_QMAN_PORTAL_DISABLEAUTO_DCA
+	bool "disable discrete-consumption support on cpu-affine portals"
+	default n
+	---help---
+	  By default, auto-initialised cpu-affine portals support
+	  discrete-consumption acknowledgements, but this may be unimplemented
+	  in the simulation model.
+
+config FSL_QMAN_NULL_FQ_DEMUX
+	bool "support NULL demux handlers"
+	default y
+	---help---
+	  Normally, incoming frame dequeues and messages from Qman to a software
+	  portal provide a direct demux to the owner of the corresponding FQ.
+	  However, exotic "zero-conf" possibilities can be supported if other
+	  cpus (or operating systems) can schedule "NULL" FQs to a receiver's
+	  portal. If this option is selected, the driver will support this
+	  feature, but it adds a small overhead to the hottest-path in the
+	  driver, so it should be disabled unless needed.
+
+config FSL_QMAN_DQRR_PREFETCHING
+	bool "support prefetching or DQRR (if stashing disabled)"
+	default y
+	---help---
+	  Normally, portals are configured to stash DQRR entries, but if this is
+	  not the case, then s/w needs to invalidate and prefetch ring entries.
+	  Selecting this option supports both cases and chooses the best one at
+	  run-time, but introduces a small overhead in some hot paths of the
+	  driver, so disable it if you know stashing will be configured.
+
+config FSL_QMAN_CONFIG
+	bool "Qman device management"
+	default y
+	---help---
+	  If this linux image is running natively, you need this option. If this
+	  linux image is running as a guest OS under the hypervisor, only one
+	  guest OS ("the control plane") needs this option.
+
+config FSL_QMAN_TEST
+	tristate "Qman self-tests"
+	depends on FSL_QMAN_PORTAL
+	default n
+	---help---
+	  This option compiles self-test code for Qman.
+
+config FSL_QMAN_TEST_STASH_POTATO
+	bool "Qman 'hot potato' data-stashing self-test"
+	depends on FSL_QMAN_TEST
+	default y
+	---help---
+	  This performs a "hot potato" style test enqueuing/dequeuing a frame
+	  across a series of FQs scheduled to different portals (and cpus), with
+	  DQRR, data and context stashing always on.
+
+config FSL_QMAN_TEST_HIGH
+	bool "Qman high-level self-test"
+	depends on FSL_QMAN_TEST
+	default y
+	---help---
+	  This requires the presence of cpu-affine portals, and performs
+	  high-level API testing with them (whichever portal(s) are affine to
+	  the cpu(s) the test executes on).
+
+config FSL_QMAN_TEST_ERRATA
+	bool "Qman errata-handling self-test"
+	depends on FSL_QMAN_TEST
+	default y
+	---help---
+	  This requires the presence of cpu-affine portals, and performs
+	  testing that handling for known hardware-errata is correct.
+
+config FSL_QMAN_DEBUGFS
+	tristate "Qman debugfs interface"
+	depends on FSL_QMAN_PORTAL
+	default y
+	---help---
+	This option compiles qman debugfs code for Qman.
+
+# H/w settings that can be hard-coded for now.
+config FSL_QMAN_FQD_SZ
+	int "Size of Frame Queue Descriptor region"
+	default 9
+	---help---
+	This is the size of the FQD region defined as: PAGE_SIZE * (2^value)
+	ex: 9 => PAGE_SIZE * (2^9)
+
+# Corenet initiator settings. Stash request queues are 4-deep to match cores'
+# ability to snart. Stash priority is 3, other priorities are 2.
+config FSL_QMAN_CI_SCHED_CFG_SRCCIV
+	int
+	depends on FSL_QMAN_CONFIG
+	default 4
+config FSL_QMAN_CI_SCHED_CFG_SRQ_W
+	int
+	depends on FSL_QMAN_CONFIG
+	default 3
+config FSL_QMAN_CI_SCHED_CFG_RW_W
+	int
+	depends on FSL_QMAN_CONFIG
+	default 2
+config FSL_QMAN_CI_SCHED_CFG_BMAN_W
+	int
+	depends on FSL_QMAN_CONFIG
+	default 2
+
+# portal interrupt settings
+config FSL_QMAN_PIRQ_DQRR_ITHRESH
+	int
+	default 12
+config FSL_QMAN_PIRQ_MR_ITHRESH
+	int
+	default 4
+config FSL_QMAN_PIRQ_IPERIOD
+	int
+	default 100
+
+# 64 bit kernel support
+config FSL_QMAN_FQ_LOOKUP
+	bool
+	default n
+
+endif # FSL_QMAN
+
+endmenu
diff --git a/drivers/staging/fsl_qbman/Makefile b/drivers/staging/fsl_qbman/Makefile
new file mode 100644
index 0000000..334055f
--- /dev/null
+++ b/drivers/staging/fsl_qbman/Makefile
@@ -0,0 +1,23 @@ 
+# Bman
+obj-$(CONFIG_FSL_BMAN_CONFIG)	+= bman_config.o bman_driver.o
+obj-$(CONFIG_FSL_BMAN_PORTAL)	+= bman_high.o
+obj-$(CONFIG_FSL_BMAN_TEST)	+= bman_tester.o
+obj-$(CONFIG_FSL_BMAN_DEBUGFS)  += bman_debugfs_interface.o
+bman_tester-y			 = bman_test.o
+bman_tester-$(CONFIG_FSL_BMAN_TEST_HIGH) += bman_test_high.o
+bman_tester-$(CONFIG_FSL_BMAN_TEST_THRESH) += bman_test_thresh.o
+bman_debugfs_interface-y	 = bman_debugfs.o
+
+# Qman
+obj-$(CONFIG_FSL_QMAN)		+= qman_utility.o qman_fqalloc.o
+obj-$(CONFIG_FSL_QMAN_CONFIG)	+= qman_config.o qman_driver.o
+obj-$(CONFIG_FSL_QMAN_PORTAL)	+= qman_high.o
+obj-$(CONFIG_FSL_QMAN_TEST)	+= qman_tester.o
+qman_tester-y			 = qman_test.o qman_test_hotpotato.o \
+					qman_test_high.o
+qman_tester-$(CONFIG_FSL_QMAN_TEST_ERRATA) += qman_test_errata.o
+obj-$(CONFIG_FSL_QMAN_DEBUGFS) += qman_debugfs_interface.o
+qman_debugfs_interface-y	= qman_debugfs.o
+
+# USDPAA
+obj-$(CONFIG_FSL_DPA_UIO)	+= dpa_uio.o
diff --git a/drivers/staging/fsl_qbman/bman_config.c b/drivers/staging/fsl_qbman/bman_config.c
new file mode 100644
index 0000000..66b8323
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_config.c
@@ -0,0 +1,688 @@ 
+/* Copyright (c) 2009-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef CONFIG_SMP
+#include <linux/smp.h>	/* get_hard_smp_processor_id() */
+#endif
+
+#include "bman_private.h"
+
+/* Last updated for v00.79 of the BG */
+
+struct bman;
+
+/* Register offsets */
+#define REG_POOL_SWDET(n)	(0x0000 + ((n) * 0x04))
+#define REG_POOL_HWDET(n)	(0x0100 + ((n) * 0x04))
+#define REG_POOL_SWDXT(n)	(0x0200 + ((n) * 0x04))
+#define REG_POOL_HWDXT(n)	(0x0300 + ((n) * 0x04))
+#define REG_POOL_CONTENT(n)	(0x0600 + ((n) * 0x04))
+#define REG_FBPR_FPC		0x0800
+#define REG_ECSR		0x0a00
+#define REG_ECIR		0x0a04
+#define REG_EADR		0x0a08
+#define REG_EDATA(n)		(0x0a10 + ((n) * 0x04))
+#define REG_SBEC(n)		(0x0a80 + ((n) * 0x04))
+#define REG_IP_REV_1		0x0bf8
+#define REG_IP_REV_2		0x0bfc
+#define REG_FBPR_BARE		0x0c00
+#define REG_FBPR_BAR		0x0c04
+#define REG_FBPR_AR		0x0c10
+#define REG_SRCIDR		0x0d04
+#define REG_LIODNR		0x0d08
+#define REG_ERR_ISR		0x0e00	/* + "enum bm_isr_reg" */
+
+/* Used by all error interrupt registers except 'inhibit' */
+#define BM_EIRQ_IVCI	0x00000010	/* Invalid Command Verb */
+#define BM_EIRQ_FLWI	0x00000008	/* FBPR Low Watermark */
+#define BM_EIRQ_MBEI	0x00000004	/* Multi-bit ECC Error */
+#define BM_EIRQ_SBEI	0x00000002	/* Single-bit ECC Error */
+#define BM_EIRQ_BSCN	0x00000001	/* pool State Change Notification */
+
+/* BMAN_ECIR valid error bit */
+#define PORTAL_ECSR_ERR	(BM_EIRQ_IVCI)
+
+union bman_ecir {
+	u32 ecir_raw;
+	struct {
+		u32 __reserved1:4;
+		u32 portal_num:4;
+		u32 __reserved2:12;
+		u32 numb:4;
+		u32 __reserved3:2;
+		u32 pid:6;
+	} __packed info;
+};
+
+union bman_eadr {
+	u32 eadr_raw;
+	struct {
+		u32 __reserved1:5;
+		u32 memid:3;
+		u32 __reserved2:14;
+		u32 eadr:10;
+	} __packed info;
+};
+
+struct bman_hwerr_txt {
+	u32 mask;
+	const char *txt;
+};
+
+#define BMAN_HWE_TXT(a, b) { .mask = BM_EIRQ_##a, .txt = b }
+
+static const struct bman_hwerr_txt bman_hwerr_txts[] = {
+	BMAN_HWE_TXT(IVCI, "Invalid Command Verb"),
+	BMAN_HWE_TXT(FLWI, "FBPR Low Watermark"),
+	BMAN_HWE_TXT(MBEI, "Multi-bit ECC Error"),
+	BMAN_HWE_TXT(SBEI, "Single-bit ECC Error"),
+	BMAN_HWE_TXT(BSCN, "Pool State Change Notification"),
+};
+#define BMAN_HWE_COUNT (sizeof(bman_hwerr_txts)/sizeof(struct bman_hwerr_txt))
+
+struct bman_error_info_mdata {
+	u16 addr_mask;
+	u16 bits;
+	const char *txt;
+};
+
+#define BMAN_ERR_MDATA(a, b, c) { .addr_mask = a, .bits = b, .txt = c}
+static const struct bman_error_info_mdata error_mdata[] = {
+	BMAN_ERR_MDATA(0x03FF, 192, "Stockpile memory"),
+	BMAN_ERR_MDATA(0x00FF, 256, "SW portal ring memory port 1"),
+	BMAN_ERR_MDATA(0x00FF, 256, "SW portal ring memory port 2"),
+};
+#define BMAN_ERR_MDATA_COUNT \
+	(sizeof(error_mdata)/sizeof(struct bman_error_info_mdata))
+
+/* Add this in Kconfig */
+#define BMAN_ERRS_TO_UNENABLE (BM_EIRQ_FLWI)
+
+/**
+ * bm_err_isr_<reg>_<verb> - Manipulate global interrupt registers
+ * @v: for accessors that write values, this is the 32-bit value
+ *
+ * Manipulates BMAN_ERR_ISR, BMAN_ERR_IER, BMAN_ERR_ISDR, BMAN_ERR_IIR. All
+ * manipulations except bm_err_isr_[un]inhibit() use 32-bit masks composed of
+ * the BM_EIRQ_*** definitions. Note that "bm_err_isr_enable_write" means
+ * "write the enable register" rather than "enable the write register"!
+ */
+#define bm_err_isr_status_read(bm)	__bm_err_isr_read(bm, bm_isr_status)
+#define bm_err_isr_status_clear(bm, m)	__bm_err_isr_write(bm, bm_isr_status,m)
+#define bm_err_isr_enable_read(bm)	__bm_err_isr_read(bm, bm_isr_enable)
+#define bm_err_isr_enable_write(bm, v)	__bm_err_isr_write(bm, bm_isr_enable,v)
+#define bm_err_isr_disable_read(bm)	__bm_err_isr_read(bm, bm_isr_disable)
+#define bm_err_isr_disable_write(bm, v)	__bm_err_isr_write(bm, bm_isr_disable,v)
+#define bm_err_isr_inhibit(bm)		__bm_err_isr_write(bm, bm_isr_inhibit,1)
+#define bm_err_isr_uninhibit(bm)	__bm_err_isr_write(bm, bm_isr_inhibit,0)
+
+/*
+ * TODO: unimplemented registers
+ *
+ * BMAN_POOLk_SDCNT, BMAN_POOLk_HDCNT, BMAN_FULT,
+ * BMAN_VLDPL, BMAN_EECC, BMAN_SBET, BMAN_EINJ
+ */
+
+/* Encapsulate "struct bman *" as a cast of the register space address. */
+
+static struct bman *bm_create(void *regs)
+{
+	return (struct bman *)regs;
+}
+
+static inline u32 __bm_in(struct bman *bm, u32 offset)
+{
+	return in_be32((void *)bm + offset);
+}
+static inline void __bm_out(struct bman *bm, u32 offset, u32 val)
+{
+	out_be32((void *)bm + offset, val);
+}
+#define bm_in(reg)		__bm_in(bm, REG_##reg)
+#define bm_out(reg, val)	__bm_out(bm, REG_##reg, val)
+
+static u32 __bm_err_isr_read(struct bman *bm, enum bm_isr_reg n)
+{
+	return __bm_in(bm, REG_ERR_ISR + (n << 2));
+}
+
+static void __bm_err_isr_write(struct bman *bm, enum bm_isr_reg n, u32 val)
+{
+	__bm_out(bm, REG_ERR_ISR + (n << 2), val);
+}
+
+#if 0
+static void bm_get_details(struct bman *bm, u8 *int_options, u8 *errata,
+			u8 *conf_options)
+{
+	u32 v = bm_in(IP_REV_1);
+	*int_options = (v >> 16) & 0xff;
+	*errata = (v >> 8) & 0xff;
+	*conf_options = v & 0xff;
+}
+
+static u8 bm_get_corenet_sourceid(struct bman *bm)
+{
+	return bm_in(SRCIDR);
+}
+
+static void bm_set_liodn(struct bman *bm, u16 liodn)
+{
+	bm_out(LIODNR, liodn & 0xfff);
+}
+
+#endif
+
+static void bm_get_version(struct bman *bm, u16 *id, u8 *major, u8 *minor)
+{
+	u32 v = bm_in(IP_REV_1);
+	*id = (v >> 16);
+	*major = (v >> 8) & 0xff;
+	*minor = v & 0xff;
+}
+
+static u32 __generate_thresh(u32 val, int roundup)
+{
+	u32 e = 0;	/* co-efficient, exponent */
+	int oddbit = 0;
+	while(val > 0xff) {
+		oddbit = val & 1;
+		val >>= 1;
+		e++;
+		if(roundup && oddbit)
+			val++;
+	}
+	DPA_ASSERT(e < 0x10);
+	return (val | (e << 8));
+}
+
+static void bm_set_pool(struct bman *bm, u8 pool, u32 swdet, u32 swdxt,
+			u32 hwdet, u32 hwdxt)
+{
+	DPA_ASSERT(pool < bman_pool_max);
+	bm_out(POOL_SWDET(pool), __generate_thresh(swdet, 0));
+	bm_out(POOL_SWDXT(pool), __generate_thresh(swdxt, 1));
+	bm_out(POOL_HWDET(pool), __generate_thresh(hwdet, 0));
+	bm_out(POOL_HWDXT(pool), __generate_thresh(hwdxt, 1));
+}
+
+static void bm_set_memory(struct bman *bm, u64 ba, int prio, u32 size)
+{
+	u32 exp = ilog2(size);
+	/* choke if size isn't within range */
+	DPA_ASSERT((size >= 4096) && (size <= 1073741824) &&
+			is_power_of_2(size));
+	/* choke if '[e]ba' has lower-alignment than 'size' */
+	DPA_ASSERT(!(ba & (size - 1)));
+	bm_out(FBPR_BARE, upper_32_bits(ba));
+	bm_out(FBPR_BAR, lower_32_bits(ba));
+	bm_out(FBPR_AR, (prio ? 0x40000000 : 0) | (exp - 1));
+}
+
+/*****************/
+/* Config driver */
+/*****************/
+
+/* We support only one of these. */
+static struct bman *bm;
+static struct device_node *bm_node;
+
+/* TODO: Kconfig these? */
+#define DEFAULT_FBPR_SZ	(PAGE_SIZE << 12)
+
+/* Parse the <name> property to extract the memory location and size and
+ * memblock_reserve() it. If it isn't supplied, memblock_alloc() the default size. */
+static __init int parse_mem_property(struct device_node *node, const char *name,
+				dma_addr_t *addr, size_t *sz, int zero)
+{
+	const u32 *pint;
+	int ret;
+
+	pint = of_get_property(node, name, &ret);
+	if (!pint || (ret != 16)) {
+		pr_info("No %s property '%s', using memblock_alloc(%016zx)\n",
+				node->full_name, name, *sz);
+		*addr = memblock_alloc(*sz, *sz);
+		if (zero)
+			memset(phys_to_virt(*addr), 0, *sz);
+		return 0;
+	}
+	pr_info("Using %s property '%s'\n", node->full_name, name);
+	/* If using a "zero-pma", don't try to zero it, even if you asked */
+	if (zero && of_find_property(node, "zero-pma", &ret)) {
+		pr_info("  it's a 'zero-pma', not zeroing from s/w\n");
+		zero = 0;
+	}
+	*addr = ((u64)pint[0] << 32) | (u64)pint[1];
+	*sz = ((u64)pint[2] << 32) | (u64)pint[3];
+	/* Keep things simple, it's either all in the DRAM range or it's all
+	 * outside. */
+	if (*addr < memblock_end_of_DRAM()) {
+		BUG_ON((u64)*addr + (u64)*sz > memblock_end_of_DRAM());
+		if (memblock_reserve(*addr, *sz) < 0) {
+			pr_err("Failed to reserve %s\n", name);
+			return -ENOMEM;
+		}
+		if (zero)
+			memset(phys_to_virt(*addr), 0, *sz);
+	} else if (zero) {
+		/* map as cacheable, non-guarded */
+		void *tmpp = ioremap_prot(*addr, *sz, 0);
+		memset(tmpp, 0, *sz);
+		iounmap(tmpp);
+	}
+	return 0;
+}
+
+static int __init fsl_bman_init(struct device_node *node)
+{
+	struct resource res;
+	u32 __iomem *regs;
+	const char *s;
+	dma_addr_t fbpr_a = 0; /* gcc doesn't know this is unnecessary */
+	size_t fbpr_sz = DEFAULT_FBPR_SZ;
+	int ret, standby = 0;
+	u16 id;
+	u8 major, minor;
+
+	ret = of_address_to_resource(node, 0, &res);
+	if (ret) {
+		pr_err("Can't get %s property 'reg'\n",
+				node->full_name);
+		return ret;
+	}
+	s = of_get_property(node, "fsl,hv-claimable", &ret);
+	if (s && !strcmp(s, "standby"))
+		standby = 1;
+	if (!standby) {
+		ret = parse_mem_property(node, "fsl,bman-fbpr",
+					&fbpr_a, &fbpr_sz, 0);
+		BUG_ON(ret);
+	}
+	/* Global configuration */
+	regs = ioremap(res.start, res.end - res.start + 1);
+	bm = bm_create(regs);
+	BUG_ON(!bm);
+	bm_node = node;
+	bm_get_version(bm, &id, &major, &minor);
+	pr_info("Bman ver:%04x,%02x,%02x\n", id, major, minor);
+	if ((major == 1) && (minor == 0)) {
+		bman_ip_rev = BMAN_REV10;
+		bman_pool_max = 64;
+	} else if ((major == 2) && (minor == 0)) {
+		bman_ip_rev = BMAN_REV20;
+		bman_pool_max = 8;
+	} else {
+		pr_warning("unknown Bman version, default to rev1.0\n");
+	}
+
+	if (standby) {
+		pr_info("  -> in standby mode\n");
+		return 0;
+	}
+	/* FBPR memory */
+	bm_set_memory(bm, fbpr_a, 0, fbpr_sz);
+	return 0;
+}
+
+int bman_have_ccsr(void)
+{
+	return (bm ? 1 : 0);
+}
+
+int bm_pool_set(u32 bpid, const u32 *thresholds)
+{
+	if (!bm)
+		return -ENODEV;
+	bm_set_pool(bm, bpid, thresholds[0], thresholds[1],
+		thresholds[2], thresholds[3]);
+	return 0;
+}
+EXPORT_SYMBOL(bm_pool_set);
+
+__init void bman_init_early(void)
+{
+	struct device_node *dn;
+	for_each_compatible_node(dn, NULL, "fsl,bman") {
+		if (bm)
+			pr_err("%s: only one 'fsl,bman' allowed\n",
+				dn->full_name);
+		else {
+			int ret = fsl_bman_init(dn);
+			BUG_ON(ret);
+		}
+	}
+}
+
+static void log_edata_bits(u32 bit_count)
+{
+	u32 i, j, mask = 0xffffffff;
+
+	pr_warning("Bman ErrInt, EDATA:\n");
+	i = bit_count/32;
+	if (bit_count%32) {
+		i++;
+		mask = ~(mask << bit_count%32);
+	}
+	j = 16-i;
+	pr_warning("  0x%08x\n", bm_in(EDATA(j)) & mask);
+	j++;
+	for (; j < 16; j++)
+		pr_warning("  0x%08x\n", bm_in(EDATA(j)));
+}
+
+static void log_additional_error_info(u32 isr_val, u32 ecsr_val)
+{
+	union bman_ecir ecir_val;
+	union bman_eadr eadr_val;
+
+	ecir_val.ecir_raw = bm_in(ECIR);
+	/* Is portal info valid */
+	if (ecsr_val & PORTAL_ECSR_ERR) {
+		pr_warning("Bman ErrInt: SWP id %d, numb %d, pid %d\n",
+			ecir_val.info.portal_num, ecir_val.info.numb,
+			ecir_val.info.pid);
+	}
+	if (ecsr_val & (BM_EIRQ_SBEI|BM_EIRQ_MBEI)) {
+		eadr_val.eadr_raw = bm_in(EADR);
+		pr_warning("Bman ErrInt: EADR Memory: %s, 0x%x\n",
+			error_mdata[eadr_val.info.memid].txt,
+			error_mdata[eadr_val.info.memid].addr_mask
+				& eadr_val.info.eadr);
+		log_edata_bits(error_mdata[eadr_val.info.memid].bits);
+	}
+}
+
+/* Bman interrupt handler */
+static irqreturn_t bman_isr(int irq, void *ptr)
+{
+	u32 isr_val, ier_val, ecsr_val, isr_mask, i;
+
+	ier_val = bm_err_isr_enable_read(bm);
+	isr_val = bm_err_isr_status_read(bm);
+	ecsr_val = bm_in(ECSR);
+	isr_mask = isr_val & ier_val;
+
+	if (!isr_mask)
+		return IRQ_NONE;
+	for (i = 0; i < BMAN_HWE_COUNT; i++) {
+		if (bman_hwerr_txts[i].mask & isr_mask) {
+			pr_warning("Bman ErrInt: %s\n", bman_hwerr_txts[i].txt);
+			if (bman_hwerr_txts[i].mask & ecsr_val) {
+				log_additional_error_info(isr_mask, ecsr_val);
+				/* Re-arm error capture registers */
+				bm_out(ECSR, ecsr_val);
+			}
+			if (bman_hwerr_txts[i].mask & BMAN_ERRS_TO_UNENABLE) {
+				pr_devel("Bman un-enabling error 0x%x\n",
+					bman_hwerr_txts[i].mask);
+				ier_val &= ~bman_hwerr_txts[i].mask;
+				bm_err_isr_enable_write(bm, ier_val);
+			}
+		}
+	}
+	bm_err_isr_status_clear(bm, isr_val);
+	return IRQ_HANDLED;
+}
+
+static int __bind_irq(void)
+{
+	int ret, err_irq;
+
+	err_irq = of_irq_to_resource(bm_node, 0, NULL);
+	if (err_irq == NO_IRQ) {
+		pr_info("Can't get %s property '%s'\n", bm_node->full_name,
+			"interrupts");
+		return -ENODEV;
+	}
+	ret = request_irq(err_irq, bman_isr, IRQF_SHARED, "bman-err", bm_node);
+	if (ret)  {
+		pr_err("request_irq() failed %d for '%s'\n", ret,
+			bm_node->full_name);
+		return -ENODEV;
+	}
+	/* Disable Buffer Pool State Change */
+	bm_err_isr_disable_write(bm, BM_EIRQ_BSCN);
+	/* Write-to-clear any stale bits, (eg. starvation being asserted prior
+	 * to resource allocation during driver init). */
+	bm_err_isr_status_clear(bm, 0xffffffff);
+	/* Enable Error Interrupts */
+	bm_err_isr_enable_write(bm, 0xffffffff);
+	return 0;
+}
+
+/* Initialise Error Interrupt Handler */
+int bman_init_error_int(struct device_node *node)
+{
+	if (!bman_have_ccsr())
+		return 0;
+	if (node != bm_node)
+		return -EINVAL;
+	return __bind_irq();
+}
+
+#ifdef CONFIG_SYSFS
+
+#define DRV_NAME "fsl-bman"
+
+static ssize_t show_fbpr_fpc(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%u\n", bm_in(FBPR_FPC));
+};
+
+static ssize_t show_pool_count(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	u32 data;
+	int i;
+
+	if (!sscanf(dev_attr->attr.name, "%d", &i))
+		return -EINVAL;
+	data = bm_in(POOL_CONTENT(i));
+	return snprintf(buf, PAGE_SIZE, "%d\n", data);
+};
+
+static ssize_t show_err_isr(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "0x%08x\n", bm_in(ERR_ISR));
+};
+
+static ssize_t show_sbec(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	int i;
+
+	if (!sscanf(dev_attr->attr.name, "sbec_%d", &i))
+		return -EINVAL;
+	return snprintf(buf, PAGE_SIZE, "%u\n", bm_in(SBEC(i)));
+};
+
+static DEVICE_ATTR(err_isr, S_IRUSR, show_err_isr, NULL);
+static DEVICE_ATTR(fbpr_fpc, S_IRUSR, show_fbpr_fpc, NULL);
+
+/* Didn't use DEVICE_ATTR as 64 of this would be required.
+ * Initialize them when needed. */
+static char *name_attrs_pool_count; /* "xx" + null-terminator */
+static struct device_attribute *dev_attr_buffer_pool_count;
+
+static DEVICE_ATTR(sbec_0, S_IRUSR, show_sbec, NULL);
+static DEVICE_ATTR(sbec_1, S_IRUSR, show_sbec, NULL);
+
+static struct attribute *bman_dev_attributes[] = {
+	&dev_attr_fbpr_fpc.attr,
+	&dev_attr_err_isr.attr,
+	NULL
+};
+
+static struct attribute *bman_dev_ecr_attributes[] = {
+	&dev_attr_sbec_0.attr,
+	&dev_attr_sbec_1.attr,
+	NULL
+};
+
+static struct attribute **bman_dev_pool_count_attributes;
+
+
+/* root level */
+static const struct attribute_group bman_dev_attr_grp = {
+	.name = NULL,
+	.attrs = bman_dev_attributes
+};
+static const struct attribute_group bman_dev_ecr_grp = {
+	.name = "error_capture",
+	.attrs = bman_dev_ecr_attributes
+};
+static struct attribute_group bman_dev_pool_countent_grp = {
+	.name = "pool_count",
+};
+
+static int of_fsl_bman_remove(struct platform_device *ofdev)
+{
+	sysfs_remove_group(&ofdev->dev.kobj, &bman_dev_attr_grp);
+	return 0;
+};
+
+static int __devinit of_fsl_bman_probe(struct platform_device *ofdev)
+{
+	int ret, i;
+
+	ret = sysfs_create_group(&ofdev->dev.kobj, &bman_dev_attr_grp);
+	if (ret)
+		goto done;
+	ret = sysfs_create_group(&ofdev->dev.kobj, &bman_dev_ecr_grp);
+	if (ret)
+		goto del_group_0;
+
+	name_attrs_pool_count = kmalloc(sizeof(char) * bman_pool_max * 3,
+								 GFP_KERNEL);
+	if (!name_attrs_pool_count) {
+		pr_err("Can't alloc name_attrs_pool_count\n");
+		goto del_group_1;
+	}
+
+	dev_attr_buffer_pool_count = kmalloc(sizeof(struct device_attribute) *
+					bman_pool_max, GFP_KERNEL);
+	if (!dev_attr_buffer_pool_count) {
+		pr_err("Can't alloc dev_attr-buffer_pool_count\n");
+		goto del_group_2;
+	}
+
+	bman_dev_pool_count_attributes = kmalloc(sizeof(struct attribute *) *
+			 (bman_pool_max + 1), GFP_KERNEL);
+	if (!bman_dev_pool_count_attributes) {
+			pr_err("can't alloc bman_dev_pool_count_attributes\n");
+			goto del_group_3;
+	}
+
+	for (i = 0; i < (bman_pool_max + 1); i++) {
+		bman_dev_pool_count_attributes[i] =
+			kmalloc(sizeof(struct attribute), GFP_KERNEL);
+		if (!bman_dev_pool_count_attributes[i]) {
+			pr_err("cannot alloc for each"
+				" bman_dev_pool_count_attributes\n");
+			goto del_group_3;
+		}
+	}
+
+	for (i = 0; i < bman_pool_max; i++) {
+		ret = scnprintf((name_attrs_pool_count + i * 3), 3, "%d", i);
+		if (!ret)
+			goto del_group_4;
+		dev_attr_buffer_pool_count[i].attr.name =
+			(name_attrs_pool_count + i * 3);
+		dev_attr_buffer_pool_count[i].attr.mode = S_IRUSR;
+		dev_attr_buffer_pool_count[i].show = show_pool_count;
+		bman_dev_pool_count_attributes[i] =
+			&dev_attr_buffer_pool_count[i].attr;
+	}
+	bman_dev_pool_count_attributes[bman_pool_max] = NULL;
+
+	bman_dev_pool_countent_grp.attrs = bman_dev_pool_count_attributes;
+
+	ret = sysfs_create_group(&ofdev->dev.kobj, &bman_dev_pool_countent_grp);
+	if (ret)
+		goto del_group_4;
+
+	goto done;
+
+del_group_4:
+	for (i = 0; i < (bman_pool_max + 1); i++)
+		kfree(bman_dev_pool_count_attributes[i]);
+	kfree(bman_dev_pool_count_attributes);
+del_group_3:
+	kfree(dev_attr_buffer_pool_count);
+del_group_2:
+	kfree(name_attrs_pool_count);
+del_group_1:
+	sysfs_remove_group(&ofdev->dev.kobj, &bman_dev_ecr_grp);
+del_group_0:
+	sysfs_remove_group(&ofdev->dev.kobj, &bman_dev_attr_grp);
+done:
+	if (ret)
+		dev_err(&ofdev->dev,
+			"Cannot create dev attributes ret=%d\n", ret);
+	return ret;
+};
+
+static struct of_device_id of_fsl_bman_ids[] = {
+	{
+		.compatible = "fsl,bman",
+	},
+	{}
+};
+MODULE_DEVICE_TABLE(of, of_fsl_bman_ids);
+
+static struct platform_driver of_fsl_bman_driver = {
+	.driver = {
+		.owner = THIS_MODULE,
+		.name = DRV_NAME,
+		.of_match_table = of_fsl_bman_ids,
+	},
+	.probe = of_fsl_bman_probe,
+	.remove = __devexit_p(of_fsl_bman_remove),
+};
+
+static int bman_ctrl_init(void)
+{
+	return platform_driver_register(&of_fsl_bman_driver);
+}
+
+static void bman_ctrl_exit(void)
+{
+	platform_driver_unregister(&of_fsl_bman_driver);
+}
+
+module_init(bman_ctrl_init);
+module_exit(bman_ctrl_exit);
+
+#endif /* CONFIG_SYSFS */
diff --git a/drivers/staging/fsl_qbman/bman_debugfs.c b/drivers/staging/fsl_qbman/bman_debugfs.c
new file mode 100644
index 0000000..191711e
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_debugfs.c
@@ -0,0 +1,121 @@ 
+/* Copyright 2010-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include <linux/module.h>
+#include <linux/fsl_bman.h>
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
+#include <linux/uaccess.h>
+
+static struct dentry *dfs_root; /* debugfs root directory */
+
+/*******************************************************************************
+ *  Query Buffer Pool State
+ ******************************************************************************/
+static int query_bp_state_show(struct seq_file *file, void *offset)
+{
+	int ret;
+	struct bm_pool_state state;
+	int i, j;
+	u32 mask;
+
+	memset(&state, 0, sizeof(struct bm_pool_state));
+	ret = bman_query_pools(&state);
+	if (ret) {
+		seq_printf(file, "Error %d\n", ret);
+		return 0;
+	}
+	seq_printf(file, "bp_id  free_buffers_avail  bp_depleted\n");
+	for (i = 0; i < 2; i++) {
+		mask = 0x80000000;
+		for (j = 0; j < 32; j++) {
+			seq_printf(file,
+			 "  %-2u           %-3s             %-3s\n",
+			 (i*32)+j,
+			 (state.as.state.__state[i] & mask) ? "no" : "yes",
+			 (state.ds.state.__state[i] & mask) ? "yes" : "no");
+			 mask >>= 1;
+		}
+	}
+	return 0;
+}
+
+static int query_bp_state_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, query_bp_state_show, NULL);
+}
+
+static const struct file_operations query_bp_state_fops = {
+	.owner          = THIS_MODULE,
+	.open		= query_bp_state_open,
+	.read           = seq_read,
+	.release	= single_release,
+};
+
+static int __init bman_debugfs_module_init(void)
+{
+	int ret = 0;
+	struct dentry *d;
+
+	dfs_root = debugfs_create_dir("bman", NULL);
+
+	if (dfs_root == NULL) {
+		ret = -ENOMEM;
+		pr_err("Cannot create bman debugfs dir\n");
+		goto _return;
+	}
+	d = debugfs_create_file("query_bp_state",
+		S_IRUGO,
+		dfs_root,
+		NULL,
+		&query_bp_state_fops);
+	if (d == NULL) {
+		ret = -ENOMEM;
+		pr_err("Cannot create query_bp_state\n");
+		goto _return;
+	}
+	return 0;
+
+_return:
+	if (dfs_root)
+		debugfs_remove_recursive(dfs_root);
+	return ret;
+}
+
+static void __exit bman_debugfs_module_exit(void)
+{
+	debugfs_remove_recursive(dfs_root);
+}
+
+
+module_init(bman_debugfs_module_init);
+module_exit(bman_debugfs_module_exit);
+MODULE_LICENSE("Dual BSD/GPL");
+
diff --git a/drivers/staging/fsl_qbman/bman_driver.c b/drivers/staging/fsl_qbman/bman_driver.c
new file mode 100644
index 0000000..9473bcf
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_driver.c
@@ -0,0 +1,586 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "bman_private.h"
+
+/*
+ * Global variables of the max portal/pool number this bman version supported
+ */
+u16 bman_ip_rev;
+EXPORT_SYMBOL(bman_ip_rev);
+u16 bman_pool_max;
+EXPORT_SYMBOL(bman_pool_max);
+
+/*****************/
+/* Portal driver */
+/*****************/
+
+/* Compatibility behaviour (when no bpool-range is present) is that;
+ * (a) on a control plane, all pools that aren't explicitly mentioned in the dtb
+ *     are available for allocation,
+ * (b) on a non-control plane, there is never any allocation possible at all.
+ *
+ * New behaviour is that if any "fsl,bpool-range" nodes are declared, they
+ * declare what is available for allocation, and this is independent of which
+ * pools are/aren't mentioned in the dtb. Eg. to stipulate that no allocation is
+ * possible, a fsl,bpool-range should be specified with zero items in it.
+ *
+ * This "pools" struct contains the allocator, and "explicit allocator"
+ * indicates whether the range is seeded explicitly (via at least one range) or
+ * implicitly (by being the set of pools that aren't declared).
+ */
+static struct bman_depletion pools;
+static u8 num_pools;
+static DEFINE_SPINLOCK(pools_lock);
+static int explicit_allocator;
+
+static struct dpa_uio_class bman_uio = {
+	.list = LIST_HEAD_INIT(bman_uio.list),
+	.dev_prefix = "bman-uio-"
+};
+const struct dpa_uio_class *dpa_uio_bman(void)
+{
+	return &bman_uio;
+}
+EXPORT_SYMBOL(dpa_uio_bman);
+
+static int __bm_pool_add(u32 bpid, u32 *cfg, int triplets)
+{
+	u64 total = 0;
+	BUG_ON(bpid >= bman_pool_max);
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	while (triplets--) {
+		struct bman_pool_params params = {
+			.bpid = bpid,
+			.flags = BMAN_POOL_FLAG_ONLY_RELEASE
+		};
+		u64 c = ((u64)cfg[0] << 32) | cfg[1];
+		u64 d = ((u64)cfg[2] << 32) | cfg[3];
+		u64 b = ((u64)cfg[4] << 32) | cfg[5];
+		struct bman_pool *pobj = bman_new_pool(&params);
+		if (!pobj)
+			return -ENOMEM;
+		while (c) {
+			struct bm_buffer bufs[8];
+			int ret, num_bufs = 0;
+			do {
+				BUG_ON(b > 0xffffffffffffull);
+				bufs[num_bufs].bpid = bpid;
+				bm_buffer_set64(&bufs[num_bufs++], b);
+				b += d;
+			} while (--c && (num_bufs < 8));
+			ret = bman_release(pobj, bufs, num_bufs,
+					BMAN_RELEASE_FLAG_WAIT);
+			if (ret)
+				panic("Seeding reserved buffer pool failed\n");
+			total += num_bufs;
+		}
+		bman_free_pool(pobj);
+		cfg += 6;
+	}
+#endif
+	/* Remove this pool from the allocator (by treating its declaration as
+	 * an implicit "reservation") iff the allocator is *not* being set up
+	 * explicitly defined via "bpool-range" nodes. */
+	if (!explicit_allocator && !bman_depletion_get(&pools, bpid)) {
+		bman_depletion_set(&pools, bpid);
+		num_pools++;
+	}
+	if (total)
+		pr_info("Bman: reserved bpid %d, seeded %lld items\n", bpid,
+			total);
+	else
+		pr_info("Bman: reserved bpid %d\n", bpid);
+	return 0;
+}
+
+int bm_pool_new(u32 *bpid)
+{
+	int ret = 0, b = bman_pool_max;
+	spin_lock(&pools_lock);
+	if (num_pools >= bman_pool_max)
+		ret = -ENOMEM;
+	else {
+		while (b-- && bman_depletion_get(&pools, b))
+			;
+		BUG_ON(b < 0);
+		bman_depletion_set(&pools, b);
+		*bpid = b;
+		num_pools++;
+	}
+	spin_unlock(&pools_lock);
+	return ret;
+}
+EXPORT_SYMBOL(bm_pool_new);
+
+void bm_pool_free(u32 bpid)
+{
+	spin_lock(&pools_lock);
+	BUG_ON(bpid >= bman_pool_max);
+	BUG_ON(!bman_depletion_get(&pools, bpid));
+	bman_depletion_unset(&pools, bpid);
+	num_pools--;
+	spin_unlock(&pools_lock);
+}
+EXPORT_SYMBOL(bm_pool_free);
+
+#ifdef CONFIG_FSL_BMAN_PORTAL
+/* To understand the use of this structure and the flow of operation for all
+ * this portal-setup code, please see qman_driver.c. The Bman case is much the
+ * same, but simpler (no Qman-specific fiddly bits). */
+struct affine_portal_data {
+	struct completion done;
+	const struct bm_portal_config *pconfig;
+	struct bman_portal *redirect;
+	int recovery_mode;
+	struct bman_portal *portal;
+};
+
+static __init int thread_init_affine_portal(void *__data)
+{
+	struct affine_portal_data *data = __data;
+	const struct bm_portal_config *pconfig = data->pconfig;
+	if (data->redirect)
+		data->portal = bman_create_affine_slave(data->redirect);
+	else {
+		data->portal = bman_create_affine_portal(pconfig,
+					data->recovery_mode);
+#ifdef CONFIG_FSL_DPA_PIRQ_SLOW
+		if (data->portal)
+			bman_irqsource_add(BM_PIRQ_RCRI | BM_PIRQ_BSCN);
+#endif
+	}
+	complete(&data->done);
+	return 0;
+}
+
+static __init struct bman_portal *init_affine_portal(
+					struct bm_portal_config *pconfig,
+					int cpu, struct bman_portal *redirect,
+					int recovery_mode)
+{
+	struct affine_portal_data data = {
+		.done = COMPLETION_INITIALIZER_ONSTACK(data.done),
+		.pconfig = pconfig,
+		.redirect = redirect,
+		.recovery_mode = recovery_mode,
+		.portal = NULL
+	};
+	struct task_struct *k = kthread_create(thread_init_affine_portal, &data,
+		"bman_affine%d", cpu);
+	int ret;
+	if (IS_ERR(k)) {
+		pr_err("Failed to init %sBman affine portal for cpu %d\n",
+			redirect ? "(slave) " : "", cpu);
+		return NULL;
+	}
+	kthread_bind(k, cpu);
+	wake_up_process(k);
+	wait_for_completion(&data.done);
+	ret = kthread_stop(k);
+	if (ret) {
+		pr_err("Bman portal initialisation failed, cpu %d, code %d\n",
+			cpu, ret);
+		return NULL;
+	}
+	if (data.portal)
+		pr_info("Bman portal %sinitialised, cpu %d\n",
+			redirect ? "(slave) " :
+			pconfig->public_cfg.is_shared ? "(shared) " : "", cpu);
+	return data.portal;
+}
+#endif
+
+static struct bm_portal_config * __init fsl_bman_portal_init(
+						struct device_node *node)
+{
+	struct bm_portal_config *pcfg;
+	const u32 *index;
+	const phandle *ph = NULL;
+	int irq, ret;
+
+	pcfg = kmalloc(sizeof(*pcfg), GFP_KERNEL);
+	if (!pcfg) {
+		pr_err("can't allocate portal config");
+		return NULL;
+	}
+
+	if (of_device_is_compatible(node, "fsl,bman-portal-1.0")) {
+		bman_ip_rev = BMAN_REV10;
+		bman_pool_max = 64;
+	} else if (of_device_is_compatible(node, "fsl,bman-portal-2.0")) {
+		bman_ip_rev = BMAN_REV20;
+		bman_pool_max = 8;
+	}
+
+	ret = of_address_to_resource(node, BM_ADDR_CE,
+				&pcfg->addr_phys[BM_ADDR_CE]);
+	if (ret) {
+		pr_err("Can't get %s property 'reg::CE'\n", node->full_name);
+		goto err;
+	}
+	ret = of_address_to_resource(node, BM_ADDR_CI,
+				&pcfg->addr_phys[BM_ADDR_CI]);
+	if (ret) {
+		pr_err("Can't get %s property 'reg::CI'\n", node->full_name);
+		goto err;
+	}
+	index = of_get_property(node, "cell-index", &ret);
+	if (!index || (ret != 4)) {
+		pr_err("Can't get %s property '%s'\n", node->full_name,
+			"cell-index");
+		goto err;
+	}
+	ph = of_get_property(node, "cpu-handle", &ret);
+	if (ph) {
+		if (ret != sizeof(phandle)) {
+			pr_err("Malformed %s property '%s'\n", node->full_name,
+				"cpu-handle");
+			goto err;
+		}
+		ret = check_cpu_phandle(*ph);
+		if (ret < 0)
+			goto err;
+		pcfg->public_cfg.cpu = ret;
+	} else
+		pcfg->public_cfg.cpu = -1;
+
+	irq = irq_of_parse_and_map(node, 0);
+	if (irq == NO_IRQ) {
+		pr_err("Can't get %s property 'interrupts'\n", node->full_name);
+		goto err;
+	}
+	pcfg->public_cfg.irq = irq;
+	pcfg->public_cfg.index = *index;
+	bman_depletion_fill(&pcfg->public_cfg.mask);
+
+	if (of_get_property(node, "fsl,usdpaa-portal", &ret)) {
+		struct dpa_uio_portal *u = kmalloc(sizeof(*u), GFP_KERNEL);
+		if (!u)
+			goto err;
+		u->type = dpa_uio_portal_bman;
+		u->bm_cfg = pcfg;
+		list_add_tail(&u->node, &bman_uio.list);
+		/* Return NULL, otherwise the kernel may share it on CPUs that
+		 * don't have their own portals, which would be ... *bad*. */
+		return NULL;
+	}
+
+	/* Map the portals now we know they aren't for UIO (the UIO code doesn't
+	 * need the CE mapping, and so will do its own CI-only mapping). */
+	pcfg->addr_virt[BM_ADDR_CE] = ioremap_prot(
+				pcfg->addr_phys[BM_ADDR_CE].start,
+				resource_size(&pcfg->addr_phys[BM_ADDR_CE]),
+				0);
+	pcfg->addr_virt[BM_ADDR_CI] = ioremap_prot(
+				pcfg->addr_phys[BM_ADDR_CI].start,
+				resource_size(&pcfg->addr_phys[BM_ADDR_CI]),
+				_PAGE_GUARDED | _PAGE_NO_CACHE);
+	return pcfg;
+err:
+	kfree(pcfg);
+	return NULL;
+}
+
+static void __init fsl_bman_portal_destroy(struct bm_portal_config *pcfg)
+{
+	iounmap(pcfg->addr_virt[BM_ADDR_CE]);
+	iounmap(pcfg->addr_virt[BM_ADDR_CI]);
+	kfree(pcfg);
+}
+
+static int __init fsl_bpool_init(struct device_node *node)
+{
+	int ret;
+	u32 *cfg = NULL, *thresh;
+	struct device_node *tmp_node;
+	u32 *bpid = (u32 *)of_get_property(node, "fsl,bpid", &ret);
+	if (!bpid || (ret!= 4)) {
+		pr_err("Can't get %s property 'fsl,bpid'\n", node->full_name);
+		return -ENODEV;
+	}
+	thresh = (u32 *)of_get_property(node, "fsl,bpool-thresholds", &ret);
+	if (thresh) {
+		if (ret != 16) {
+			pr_err("Invalid %s property '%s'\n",
+				node->full_name, "fsl,bpool-thresholds");
+			return -ENODEV;
+		}
+#ifndef CONFIG_FSL_BMAN_CONFIG
+		pr_err("Ignoring %s property '%s', no CCSR support\n",
+			node->full_name, "fsl,bpool-thresholds");
+#endif
+	}
+	/* If rebooted, we should not re-seed any pools via bpool-cfg. */
+	/* TODO: parsing hypervisor fields to determine qualitative things like
+	 * "was I rebooted" should probably be wrapped in fsl_hypervisor.h. */
+	tmp_node = of_find_node_by_name(NULL, "hypervisor");
+	if (!tmp_node || !of_find_property(tmp_node, "fsl,hv-stopped-by",
+						&ret))
+		cfg = (u32 *)of_get_property(node, "fsl,bpool-cfg", &ret);
+	if (cfg && (!ret || (ret % 24))) {
+		pr_err("Invalid %s property '%s'\n", node->full_name,
+			"fsl,bpool-cfg");
+		return -ENODEV;
+	}
+	if (cfg)
+		ret = __bm_pool_add(*bpid, cfg, ret / 24);
+	else
+		ret = __bm_pool_add(*bpid, NULL, 0);
+	if (ret) {
+		pr_err("Can't reserve bpid %d from node %s\n", *bpid,
+			node->full_name);
+		return ret;
+	}
+#ifdef CONFIG_FSL_BMAN_CONFIG
+	if (thresh) {
+		ret = bm_pool_set(*bpid, thresh);
+		if (ret)
+			pr_err("No CCSR node for %s property '%s'\n",
+				node->full_name, "fsl,bpool-thresholds");
+	}
+#endif
+	return ret;
+}
+
+static int __init fsl_bpool_range_init(struct device_node *node,
+					int recovery_mode)
+{
+	int ret, warned = 0;
+	u32 bpid;
+	u32 *range = (u32 *)of_get_property(node, "fsl,bpool-range", &ret);
+	if (!range) {
+		pr_err("No 'fsl,bpool-range' property in node %s\n",
+			node->full_name);
+		return -EINVAL;
+	}
+	if (ret != 8) {
+		pr_err("'fsl,bpool-range' is not a 2-cell range in node %s\n",
+			node->full_name);
+		return -EINVAL;
+	}
+	for (bpid = range[0]; bpid < (range[0] + range[1]); bpid++) {
+		if (bpid >= bman_pool_max) {
+			pr_err("BPIDs out of range in node %s\n",
+				node->full_name);
+			return -EINVAL;
+		}
+		if (!bman_depletion_get(&pools, bpid)) {
+			if (!warned) {
+				warned = 1;
+				pr_err("BPID overlap in node %s, ignoring\n",
+					node->full_name);
+			}
+		} else {
+			bman_depletion_unset(&pools, bpid);
+			num_pools--;
+		}
+	}
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	/* If in recovery mode *and* we are using a private BPID allocation
+	 * range, then automatically clean up all BPIDs in that range so we can
+	 * automatically exit recovery mode too. */
+	if (recovery_mode) {
+		for (bpid = range[0]; bpid < (range[0] + range[1]); bpid++) {
+			ret = bman_recovery_cleanup_bpid(bpid);
+			if (ret) {
+				pr_err("Failed to recovery BPID %d\n", bpid);
+				return ret;
+			}
+		}
+	}
+#else
+	BUG_ON(recovery_mode);
+#endif
+	pr_info("Bman: BPID allocator includes range %d:%d%s\n",
+		range[0], range[1], recovery_mode ? " (recovered)" : "");
+	return 0;
+}
+
+#ifdef CONFIG_FSL_BMAN_PORTAL
+static __init int __leave_recovery(void *__data)
+{
+	struct completion *done = __data;
+	bman_recovery_exit_local();
+	complete(done);
+	return 0;
+}
+
+int bman_recovery_exit(void)
+{
+	struct completion done = COMPLETION_INITIALIZER_ONSTACK(done);
+	unsigned int cpu;
+
+	for_each_cpu(cpu, bman_affine_cpus()) {
+		struct task_struct *k = kthread_create(__leave_recovery, &done,
+						"bman_recovery");
+		int ret;
+		if (IS_ERR(k)) {
+			pr_err("Thread failure (recovery) on cpu %d\n", cpu);
+			return -ENOMEM;
+		}
+		kthread_bind(k, cpu);
+		wake_up_process(k);
+		wait_for_completion(&done);
+		ret = kthread_stop(k);
+		if (ret) {
+			pr_err("Failed to exit recovery on cpu %d\n", cpu);
+			return ret;
+		}
+		pr_info("Bman portal exited recovery, cpu %d\n", cpu);
+	}
+	return 0;
+}
+EXPORT_SYMBOL(bman_recovery_exit);
+#endif
+
+static __init int bman_init(void)
+{
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	struct cpumask primary_cpus = *cpu_none_mask;
+	struct cpumask slave_cpus = *cpu_online_mask;
+	struct bman_portal *sharing_portal = NULL;
+	int sharing_cpu = -1;
+#endif
+	struct device_node *dn;
+	struct bm_portal_config *pcfg;
+	int ret, recovery_mode = 0;
+	LIST_HEAD(cfg_list);
+
+	for_each_compatible_node(dn, NULL, "fsl,bman") {
+		if (!bman_init_error_int(dn))
+			pr_info("Bman err interrupt handler present\n");
+		else
+			pr_err("Bman err interrupt handler missing\n");
+	}
+	if (!bman_have_ccsr()) {
+		/* If there's no CCSR, our bpid allocator is empty unless
+		 * fsl,bpool-range nodes are used. */
+		bman_depletion_fill(&pools);
+		num_pools = bman_pool_max;
+	}
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	if (fsl_dpa_should_recover())
+		recovery_mode = 1;
+	for_each_compatible_node(dn, NULL, "fsl,bman-portal") {
+		pcfg = fsl_bman_portal_init(dn);
+		if (pcfg) {
+			if (pcfg->public_cfg.cpu >= 0) {
+				cpumask_set_cpu(pcfg->public_cfg.cpu,
+						&primary_cpus);
+				list_add(&pcfg->list, &cfg_list);
+			} else
+				fsl_bman_portal_destroy(pcfg);
+		}
+	}
+	/* only consider "online" CPUs */
+	cpumask_and(&primary_cpus, &primary_cpus, cpu_online_mask);
+	if (cpumask_empty(&primary_cpus))
+		/* No portals, we're done */
+		return 0;
+	if (!cpumask_subset(cpu_online_mask, &primary_cpus)) {
+		/* Need to do some sharing. In lieu of anything more scientific
+		 * (or configurable), we pick the last-most CPU that has a
+		 * portal and share that one. */
+		int next = cpumask_first(&primary_cpus);
+		while (next < nr_cpu_ids) {
+			sharing_cpu = next;
+			next = cpumask_next(next, &primary_cpus);
+		}
+	}
+	/* Parsing is done and sharing decisions are made, now initialise the
+	 * portals and determine which "slave" CPUs are left over. */
+	list_for_each_entry(pcfg, &cfg_list, list) {
+		struct bman_portal *p;
+		int is_shared = (!sharing_portal && (sharing_cpu >= 0) &&
+				(pcfg->public_cfg.cpu == sharing_cpu));
+		pcfg->public_cfg.is_shared = is_shared;
+		/* If it's not mapped to a CPU, or another portal is already
+		 * initialised to the same CPU, skip this portal. */
+		if (pcfg->public_cfg.cpu < 0 || !cpumask_test_cpu(
+					pcfg->public_cfg.cpu, &slave_cpus))
+			continue;
+		p = init_affine_portal(pcfg, pcfg->public_cfg.cpu, NULL,
+					recovery_mode);
+		if (p) {
+			if (is_shared)
+				sharing_portal = p;
+			cpumask_clear_cpu(pcfg->public_cfg.cpu, &slave_cpus);
+		}
+	}
+
+	if (sharing_portal) {
+		int loop;
+		for_each_cpu(loop, &slave_cpus) {
+			struct bman_portal *p = init_affine_portal(NULL, loop,
+					sharing_portal, recovery_mode);
+			if (!p)
+				pr_err("Failed slave Bman portal for cpu %d\n",
+					loop);
+		}
+	}
+#else
+	for_each_compatible_node(dn, NULL, "fsl,bman-portal") {
+		pcfg = fsl_bman_portal_init(dn);
+		if (pcfg)
+			/* No kernel portal support, so if USDPAA didn't consume
+			 * the portal, we've no other use for it. */
+			fsl_bman_portal_destroy(pcfg);
+	}
+#endif
+	for_each_compatible_node(dn, NULL, "fsl,bpool-range") {
+		if (!explicit_allocator) {
+			explicit_allocator = 1;
+			bman_depletion_fill(&pools);
+			num_pools = 64;
+		}
+		ret = fsl_bpool_range_init(dn, recovery_mode);
+		if (ret)
+			return ret;
+	}
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	/* If using private BPID allocation, exit recovery mode automatically
+	 * (ie. after automatic recovery) */
+	if (recovery_mode && explicit_allocator) {
+		ret = bman_recovery_exit();
+		if (ret)
+			return ret;
+	}
+#endif
+	for_each_compatible_node(dn, NULL, "fsl,bpool") {
+		ret = fsl_bpool_init(dn);
+		if (ret)
+			return ret;
+	}
+	pr_info("Bman portals initialised\n");
+	return 0;
+}
+subsys_initcall(bman_init);
diff --git a/drivers/staging/fsl_qbman/bman_high.c b/drivers/staging/fsl_qbman/bman_high.c
new file mode 100644
index 0000000..9d473c9
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_high.c
@@ -0,0 +1,1044 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "bman_low.h"
+
+/* Compilation constants */
+#define RCR_THRESH	2	/* reread h/w CI when running out of space */
+#define IRQNAME		"BMan portal %d"
+#define MAX_IRQNAME	16	/* big enough for "BMan portal %d" */
+
+struct bman_portal {
+	struct bm_portal p;
+	/* 2-element array. pools[0] is mask, pools[1] is snapshot. */
+	struct bman_depletion *pools;
+	int thresh_set;
+	unsigned long irq_sources;
+	u32 slowpoll;	/* only used when interrupts are off */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	struct bman_pool *rcri_owned; /* only 1 release WAIT_SYNC at a time */
+#endif
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	spinlock_t sharing_lock; /* only used if is_shared */
+	int is_shared;
+	struct bman_portal *sharing_redirect;
+#endif
+	/* When the cpu-affine portal is activated, this is non-NULL */
+	const struct bm_portal_config *config;
+	/* 64-entry hash-table of pool objects that are tracking depletion
+	 * entry/exit (ie. BMAN_POOL_FLAG_DEPLETION). This isn't fast-path, so
+	 * we're not fussy about cache-misses and so forth - whereas the above
+	 * members should all fit in one cacheline.
+	 * BTW, with 64 entries in the hash table and 64 buffer pools to track,
+	 * you'll never guess the hash-function ... */
+	struct bman_pool *cb[64];
+	char irqname[MAX_IRQNAME];
+};
+
+/* For an explanation of the locking, redirection, or affine-portal logic,
+ * please consult the Qman driver for details. This is the same, only simpler
+ * (no fiddly Qman-specific bits.) */
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+#define PORTAL_IRQ_LOCK(p, irqflags) \
+	do { \
+		if ((p)->is_shared) \
+			spin_lock_irqsave(&(p)->sharing_lock, irqflags); \
+		else \
+			local_irq_save(irqflags); \
+	} while (0)
+#define PORTAL_IRQ_UNLOCK(p, irqflags) \
+	do { \
+		if ((p)->is_shared) \
+			spin_unlock_irqrestore(&(p)->sharing_lock, irqflags); \
+		else \
+			local_irq_restore(irqflags); \
+	} while (0)
+#else
+#define PORTAL_IRQ_LOCK(p, irqflags) local_irq_save(irqflags)
+#define PORTAL_IRQ_UNLOCK(p, irqflags) local_irq_restore(irqflags)
+#endif
+
+static cpumask_t affine_mask;
+static DEFINE_SPINLOCK(affine_mask_lock);
+static DEFINE_PER_CPU(struct bman_portal, bman_affine_portal);
+static inline struct bman_portal *get_raw_affine_portal(void)
+{
+	return &get_cpu_var(bman_affine_portal);
+}
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+static inline struct bman_portal *get_affine_portal(void)
+{
+	struct bman_portal *p = get_raw_affine_portal();
+	if (p->sharing_redirect)
+		return p->sharing_redirect;
+	return p;
+}
+#else
+#define get_affine_portal() get_raw_affine_portal()
+#endif
+static inline void put_affine_portal(void)
+{
+	put_cpu_var(bman_affine_portal);
+}
+
+/* GOTCHA: this object type refers to a pool, it isn't *the* pool. There may be
+ * more than one such object per Bman buffer pool, eg. if different users of the
+ * pool are operating via different portals. */
+struct bman_pool {
+	struct bman_pool_params params;
+	/* Used for hash-table admin when using depletion notifications. */
+	struct bman_portal *portal;
+	struct bman_pool *next;
+	/* stockpile state - NULL unless BMAN_POOL_FLAG_STOCKPILE is set */
+	struct bm_buffer *sp;
+	unsigned int sp_fill;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	atomic_t in_use;
+#endif
+};
+
+/* (De)Registration of depletion notification callbacks */
+static void depletion_link(struct bman_portal *portal, struct bman_pool *pool)
+{
+	__maybe_unused unsigned long irqflags;
+	pool->portal = portal;
+	PORTAL_IRQ_LOCK(portal, irqflags);
+	pool->next = portal->cb[pool->params.bpid];
+	portal->cb[pool->params.bpid] = pool;
+	if (!pool->next)
+		/* First object for that bpid on this portal, enable the BSCN
+		 * mask bit. */
+		bm_isr_bscn_mask(&portal->p, pool->params.bpid, 1);
+	PORTAL_IRQ_UNLOCK(portal, irqflags);
+}
+static void depletion_unlink(struct bman_pool *pool)
+{
+	struct bman_pool *it, *last = NULL;
+	struct bman_pool **base = &pool->portal->cb[pool->params.bpid];
+	__maybe_unused unsigned long irqflags;
+	PORTAL_IRQ_LOCK(pool->portal, irqflags);
+	it = *base;	/* <-- gotcha, don't do this prior to the irq_save */
+	while (it != pool) {
+		last = it;
+		it = it->next;
+	}
+	if (!last)
+		*base = pool->next;
+	else
+		last->next = pool->next;
+	if (!last && !pool->next) {
+		/* Last object for that bpid on this portal, disable the BSCN
+		 * mask bit. */
+		bm_isr_bscn_mask(&pool->portal->p, pool->params.bpid, 0);
+		/* And "forget" that we last saw this pool as depleted */
+		bman_depletion_unset(&pool->portal->pools[1],
+					pool->params.bpid);
+	}
+	PORTAL_IRQ_UNLOCK(pool->portal, irqflags);
+}
+
+/* In the case that the application's core loop calls qman_poll() and
+ * bman_poll(), we ought to balance how often we incur the overheads of the
+ * slow-path poll. We'll use two decrementer sources. The idle decrementer
+ * constant is used when the last slow-poll detected no work to do, and the busy
+ * decrementer constant when the last slow-poll had work to do. */
+#define SLOW_POLL_IDLE 1000
+#define SLOW_POLL_BUSY 10
+static u32 __poll_portal_slow(struct bman_portal *p, u32 is);
+
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+/* Portal interrupt handler */
+static irqreturn_t portal_isr(__always_unused int irq, void *ptr)
+{
+	struct bman_portal *p = ptr;
+	u32 clear = p->irq_sources;
+	u32 is = bm_isr_status_read(&p->p) & p->irq_sources;
+	clear |= __poll_portal_slow(p, is);
+	bm_isr_status_clear(&p->p, clear);
+	return IRQ_HANDLED;
+}
+#endif
+
+struct bman_portal *bman_create_affine_portal(
+			const struct bm_portal_config *config,
+			int recovery_mode __maybe_unused)
+{
+	struct bman_portal *portal = get_raw_affine_portal();
+	struct bm_portal *__p = &portal->p;
+	const struct bman_depletion *pools = &config->public_cfg.mask;
+	int ret;
+
+	/* prep the low-level portal struct with the mapped addresses from the
+	 * config, everything that follows depends on it and "config" is more
+	 * for (de)reference... */
+	__p->addr.addr_ce = config->addr_virt[BM_ADDR_CE];
+	__p->addr.addr_ci = config->addr_virt[BM_ADDR_CI];
+	if (bm_rcr_init(__p, bm_rcr_pvb, bm_rcr_cce)) {
+		pr_err("Bman RCR initialisation failed\n");
+		goto fail_rcr;
+	}
+	if (bm_mc_init(__p)) {
+		pr_err("Bman MC initialisation failed\n");
+		goto fail_mc;
+	}
+	if (bm_isr_init(__p)) {
+		pr_err("Bman ISR initialisation failed\n");
+		goto fail_isr;
+	}
+	if (!pools)
+		portal->pools = NULL;
+	else {
+		u8 bpid = 0;
+		portal->pools = kmalloc(2 * sizeof(*pools), GFP_KERNEL);
+		if (!portal->pools)
+			goto fail_pools;
+		portal->pools[0] = *pools;
+		bman_depletion_init(portal->pools + 1);
+		while (bpid < bman_pool_max) {
+			/* Default to all BPIDs disabled, we enable as required
+			 * at run-time. */
+			bm_isr_bscn_mask(__p, bpid, 0);
+			bpid++;
+		}
+	}
+	portal->slowpoll = 0;
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	portal->rcri_owned = NULL;
+#endif
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	spin_lock_init(&portal->sharing_lock);
+	portal->is_shared = config->public_cfg.is_shared;
+	portal->sharing_redirect = NULL;
+#endif
+	memset(&portal->cb, 0, sizeof(portal->cb));
+	/* Write-to-clear any stale interrupt status bits */
+	bm_isr_disable_write(__p, 0xffffffff);
+	portal->irq_sources = 0;
+	bm_isr_enable_write(__p, portal->irq_sources);
+	bm_isr_status_clear(__p, 0xffffffff);
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, config->public_cfg.cpu);
+	if (request_irq(config->public_cfg.irq, portal_isr,
+			IRQF_NOBALANCING | IRQF_DISABLED, portal->irqname,
+			portal)) {
+		pr_err("request_irq() failed\n");
+		goto fail_irq;
+	}
+	if (config->public_cfg.cpu != -1) {
+		disable_irq(config->public_cfg.irq);
+		irq_set_affinity(config->public_cfg.irq,
+				 cpumask_of(config->public_cfg.cpu));
+	}
+	enable_irq(config->public_cfg.irq);
+	/* Enable the bits that make sense */
+	if (!recovery_mode)
+		bm_isr_uninhibit(__p);
+#endif
+	/* Need RCR to be empty before continuing */
+	bm_isr_disable_write(__p, ~BM_PIRQ_RCRI);
+	ret = bm_rcr_get_fill(__p);
+	if (ret) {
+		pr_err("Bman RCR unclean, need recovery\n");
+		goto fail_rcr_empty;
+	}
+	/* Success */
+	portal->config = config;
+	spin_lock(&affine_mask_lock);
+	cpumask_set_cpu(config->public_cfg.cpu, &affine_mask);
+	spin_unlock(&affine_mask_lock);
+	bm_isr_disable_write(__p, 0);
+	put_affine_portal();
+	return portal;
+fail_rcr_empty:
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+fail_affinity:
+	free_irq(config->public_cfg.irq, portal);
+fail_irq:
+#endif
+	if (portal->pools)
+		kfree(portal->pools);
+fail_pools:
+	bm_isr_finish(__p);
+fail_isr:
+	bm_mc_finish(__p);
+fail_mc:
+	bm_rcr_finish(__p);
+fail_rcr:
+	put_affine_portal();
+	return NULL;
+}
+
+struct bman_portal *bman_create_affine_slave(struct bman_portal *redirect)
+{
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	struct bman_portal *p = get_raw_affine_portal();
+	BUG_ON(p->config);
+	BUG_ON(p->is_shared);
+	BUG_ON(!redirect->config->public_cfg.is_shared);
+	p->irq_sources = 0;
+	p->sharing_redirect = redirect;
+	put_affine_portal();
+	return p;
+#else
+	BUG();
+	return NULL;
+#endif
+}
+
+const struct bm_portal_config *bman_destroy_affine_portal(void)
+{
+	struct bman_portal *bm = get_raw_affine_portal();
+	const struct bm_portal_config *pcfg;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (bm->sharing_redirect) {
+		bm->sharing_redirect = NULL;
+		put_affine_portal();
+		return NULL;
+	}
+	bm->is_shared = 0;
+#endif
+	pcfg = bm->config;
+	bm_rcr_cce_update(&bm->p);
+	bm_rcr_cce_update(&bm->p);
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	free_irq(pcfg->public_cfg.irq, bm);
+#endif
+	kfree(bm->pools);
+	bm_isr_finish(&bm->p);
+	bm_mc_finish(&bm->p);
+	bm_rcr_finish(&bm->p);
+	bm->config = NULL;
+	spin_lock(&affine_mask_lock);
+	cpumask_clear_cpu(pcfg->public_cfg.cpu, &affine_mask);
+	spin_unlock(&affine_mask_lock);
+	put_affine_portal();
+	return pcfg;
+}
+
+/* When release logic waits on available RCR space, we need a global waitqueue
+ * in the case of "affine" use (as the waits wake on different cpus which means
+ * different portals - so we can't wait on any per-portal waitqueue). */
+static DECLARE_WAIT_QUEUE_HEAD(affine_queue);
+
+static u32 __poll_portal_slow(struct bman_portal *p, u32 is)
+{
+	struct bman_depletion tmp;
+	u32 ret = is;
+
+	/* There is a gotcha to be aware of. If we do the query before clearing
+	 * the status register, we may miss state changes that occur between the
+	 * two. If we write to clear the status register before the query, the
+	 * cache-enabled query command may overtake the status register write
+	 * unless we use a heavyweight sync (which we don't want). Instead, we
+	 * write-to-clear the status register then *read it back* before doing
+	 * the query, hence the odd while loop with the 'is' accumulation. */
+	if (is & BM_PIRQ_BSCN) {
+		struct bm_mc_result *mcr;
+		__maybe_unused unsigned long irqflags;
+		unsigned int i, j;
+		u32 __is;
+		bm_isr_status_clear(&p->p, BM_PIRQ_BSCN);
+		while ((__is = bm_isr_status_read(&p->p)) & BM_PIRQ_BSCN) {
+			is |= __is;
+			bm_isr_status_clear(&p->p, BM_PIRQ_BSCN);
+		}
+		is &= ~BM_PIRQ_BSCN;
+		PORTAL_IRQ_LOCK(p, irqflags);
+		bm_mc_start(&p->p);
+		bm_mc_commit(&p->p, BM_MCC_VERB_CMD_QUERY);
+		while (!(mcr = bm_mc_result(&p->p)))
+			cpu_relax();
+		tmp = mcr->query.ds.state;
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+		for (i = 0; i < 2; i++) {
+			int idx = i * 32;
+			/* tmp is a mask of currently-depleted pools.
+			 * pools[0] is mask of those we care about.
+			 * pools[1] is our previous view (we only want to
+			 * be told about changes). */
+			tmp.__state[i] &= p->pools[0].__state[i];
+			if (tmp.__state[i] == p->pools[1].__state[i])
+				/* fast-path, nothing to see, move along */
+				continue;
+			for (j = 0; j <= 31; j++, idx++) {
+				struct bman_pool *pool = p->cb[idx];
+				int b4 = bman_depletion_get(&p->pools[1], idx);
+				int af = bman_depletion_get(&tmp, idx);
+				if (b4 == af)
+					continue;
+				while (pool) {
+					pool->params.cb(p, pool,
+						pool->params.cb_ctx, af);
+					pool = pool->next;
+				}
+			}
+		}
+		p->pools[1] = tmp;
+	}
+
+	if (is & BM_PIRQ_RCRI) {
+		__maybe_unused unsigned long irqflags;
+		PORTAL_IRQ_LOCK(p, irqflags);
+		bm_rcr_cce_update(&p->p);
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+		/* If waiting for sync, we only cancel the interrupt threshold
+		 * when the ring utilisation hits zero. */
+		if (p->rcri_owned) {
+			if (!bm_rcr_get_fill(&p->p)) {
+				p->rcri_owned = NULL;
+				bm_rcr_set_ithresh(&p->p, 0);
+			}
+		} else
+#endif
+		bm_rcr_set_ithresh(&p->p, 0);
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+		wake_up(&affine_queue);
+		bm_isr_status_clear(&p->p, BM_PIRQ_RCRI);
+		is &= ~BM_PIRQ_RCRI;
+	}
+
+	/* There should be no status register bits left undefined */
+	DPA_ASSERT(!is);
+	return ret;
+}
+
+const struct bman_portal_config *bman_get_portal_config(void)
+{
+	struct bman_portal *p = get_affine_portal();
+	const struct bman_portal_config *ret = &p->config->public_cfg;
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(bman_get_portal_config);
+
+u32 bman_irqsource_get(void)
+{
+	struct bman_portal *p = get_raw_affine_portal();
+	u32 ret = p->irq_sources & BM_PIRQ_VISIBLE;
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(bman_irqsource_get);
+
+int bman_irqsource_add(__maybe_unused u32 bits)
+{
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	struct bman_portal *p = get_raw_affine_portal();
+	int ret = 0;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (p->sharing_redirect)
+		ret = -EINVAL;
+	else
+#endif
+	{
+		__maybe_unused unsigned long irqflags;
+		PORTAL_IRQ_LOCK(p, irqflags);
+		set_bits(bits & BM_PIRQ_VISIBLE, &p->irq_sources);
+		bm_isr_enable_write(&p->p, p->irq_sources);
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+	}
+	put_affine_portal();
+	return ret;
+#else
+	pr_err("No Bman portal IRQ support, mustn't specify IRQ flags!");
+	return -EINVAL;
+#endif
+}
+EXPORT_SYMBOL(bman_irqsource_add);
+
+int bman_irqsource_remove(u32 bits)
+{
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	struct bman_portal *p = get_raw_affine_portal();
+	__maybe_unused unsigned long irqflags;
+	u32 ier;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (p->sharing_redirect) {
+		put_affine_portal();
+		return -EINVAL;
+	}
+#endif
+	/* Our interrupt handler only processes+clears status register bits that
+	 * are in p->irq_sources. As we're trimming that mask, if one of them
+	 * were to assert in the status register just before we remove it from
+	 * the enable register, there would be an interrupt-storm when we
+	 * release the IRQ lock. So we wait for the enable register update to
+	 * take effect in h/w (by reading it back) and then clear all other bits
+	 * in the status register. Ie. we clear them from ISR once it's certain
+	 * IER won't allow them to reassert. */
+	PORTAL_IRQ_LOCK(p, irqflags);
+	bits &= BM_PIRQ_VISIBLE;
+	clear_bits(bits, &p->irq_sources);
+	bm_isr_enable_write(&p->p, p->irq_sources);
+	ier = bm_isr_enable_read(&p->p);
+	/* Using "~ier" (rather than "bits" or "~p->irq_sources") creates a
+	 * data-dependency, ie. to protect against re-ordering. */
+	bm_isr_status_clear(&p->p, ~ier);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return 0;
+#else
+	pr_err("No Bman portal IRQ support, mustn't specify IRQ flags!");
+	return -EINVAL;
+#endif
+}
+EXPORT_SYMBOL(bman_irqsource_remove);
+
+const cpumask_t *bman_affine_cpus(void)
+{
+	return &affine_mask;
+}
+EXPORT_SYMBOL(bman_affine_cpus);
+
+u32 bman_poll_slow(void)
+{
+	struct bman_portal *p = get_raw_affine_portal();
+	u32 ret;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (unlikely(p->sharing_redirect))
+		ret = (u32)-1;
+	else
+#endif
+	{
+		u32 is = bm_isr_status_read(&p->p) & ~p->irq_sources;
+		ret = __poll_portal_slow(p, is);
+		bm_isr_status_clear(&p->p, ret);
+	}
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(bman_poll_slow);
+
+/* Legacy wrapper */
+void bman_poll(void)
+{
+	struct bman_portal *p = get_raw_affine_portal();
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (unlikely(p->sharing_redirect))
+		goto done;
+#endif
+	if (!(p->slowpoll--)) {
+		u32 is = bm_isr_status_read(&p->p) & ~p->irq_sources;
+		u32 active = __poll_portal_slow(p, is);
+		if (active)
+			p->slowpoll = SLOW_POLL_BUSY;
+		else
+			p->slowpoll = SLOW_POLL_IDLE;
+	}
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+done:
+#endif
+	put_affine_portal();
+}
+EXPORT_SYMBOL(bman_poll);
+
+int bman_recovery_cleanup_bpid(u32 bpid)
+{
+	struct bman_pool pool = {
+		.params = {
+			.bpid = bpid
+		}
+	};
+	struct bm_buffer bufs[8];
+	int ret = 0;
+	unsigned int num_bufs = 0;
+	do {
+		/* Acquire is all-or-nothing, so we drain in 8s, then in
+		 * 1s for the remainder. */
+		if (ret != 1)
+			ret = bman_acquire(&pool, bufs, 8, 0);
+		if (ret < 8)
+			ret = bman_acquire(&pool, bufs, 1, 0);
+		if (ret > 0)
+			num_bufs += ret;
+	} while (ret > 0);
+	if (num_bufs)
+		pr_info("Bman: BPID %d recovered (%d bufs)\n", bpid, num_bufs);
+	return 0;
+}
+EXPORT_SYMBOL(bman_recovery_cleanup_bpid);
+
+/* called from bman_driver.c::bman_recovery_exit() only (if exporting, use
+ * get_raw_affine_portal() and check for the "SLAVE" bit). */
+void bman_recovery_exit_local(void)
+{
+	struct bman_portal *p = get_affine_portal();
+	bm_isr_status_clear(&p->p, 0xffffffff);
+	bm_isr_uninhibit(&p->p);
+	put_affine_portal();
+}
+
+static const u32 zero_thresholds[4] = {0, 0, 0, 0};
+
+struct bman_pool *bman_new_pool(const struct bman_pool_params *params)
+{
+	struct bman_pool *pool = NULL;
+	u32 bpid;
+
+	if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID) {
+		int ret = bm_pool_new(&bpid);
+		if (ret)
+			return NULL;
+	} else {
+		if (params->bpid >= bman_pool_max)
+			return NULL;
+		bpid = params->bpid;
+	}
+#ifdef CONFIG_FSL_BMAN_CONFIG
+	if (params->flags & BMAN_POOL_FLAG_THRESH) {
+		int ret = bm_pool_set(bpid, params->thresholds);
+		if (ret)
+			goto err;
+	}
+#else
+	if (params->flags & BMAN_POOL_FLAG_THRESH)
+		goto err;
+#endif
+	pool = kmalloc(sizeof(*pool), GFP_KERNEL);
+	if (!pool)
+		goto err;
+	pool->sp = NULL;
+	pool->sp_fill = 0;
+	pool->params = *params;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	atomic_set(&pool->in_use, 1);
+#endif
+	if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID)
+		pool->params.bpid = bpid;
+	if (params->flags & BMAN_POOL_FLAG_STOCKPILE) {
+		pool->sp = kmalloc(sizeof(struct bm_buffer) * BMAN_STOCKPILE_SZ,
+					GFP_KERNEL);
+		if (!pool->sp)
+			goto err;
+	}
+	if (pool->params.flags & BMAN_POOL_FLAG_DEPLETION) {
+		struct bman_portal *p = get_affine_portal();
+		if (!p->pools || !bman_depletion_get(&p->pools[0], bpid)) {
+			pr_err("Depletion events disabled for bpid %d\n", bpid);
+			goto err;
+		}
+		depletion_link(p, pool);
+		put_affine_portal();
+	}
+	return pool;
+err:
+#ifdef CONFIG_FSL_BMAN_CONFIG
+	if (params->flags & BMAN_POOL_FLAG_THRESH)
+		bm_pool_set(bpid, zero_thresholds);
+#endif
+	if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID)
+		bm_pool_free(bpid);
+	if (pool) {
+		if (pool->sp)
+			kfree(pool->sp);
+		kfree(pool);
+	}
+	return NULL;
+}
+EXPORT_SYMBOL(bman_new_pool);
+
+void bman_free_pool(struct bman_pool *pool)
+{
+#ifdef CONFIG_FSL_BMAN_CONFIG
+	if (pool->params.flags & BMAN_POOL_FLAG_THRESH)
+		bm_pool_set(pool->params.bpid, zero_thresholds);
+#endif
+	if (pool->params.flags & BMAN_POOL_FLAG_DEPLETION)
+		depletion_unlink(pool);
+	if (pool->params.flags & BMAN_POOL_FLAG_STOCKPILE) {
+		if (pool->sp_fill)
+			pr_err("Stockpile not flushed, has %u in bpid %u.\n",
+				pool->sp_fill, pool->params.bpid);
+		kfree(pool->sp);
+		pool->sp = NULL;
+		pool->params.flags ^= BMAN_POOL_FLAG_STOCKPILE;
+	}
+	if (pool->params.flags & BMAN_POOL_FLAG_DYNAMIC_BPID) {
+		/* When releasing a BPID to the dynamic allocator, that pool
+		 * must be *empty*. This code makes it so by dropping everything
+		 * into the bit-bucket. This ignores whether or not it was a
+		 * mistake (or a leak) on the caller's part not to drain the
+		 * pool beforehand. */
+		struct bm_buffer bufs[8];
+		int ret = 0;
+		do {
+			/* Acquire is all-or-nothing, so we drain in 8s, then in
+			 * 1s for the remainder. */
+			if (ret != 1)
+				ret = bman_acquire(pool, bufs, 8, 0);
+			if (ret < 8)
+				ret = bman_acquire(pool, bufs, 1, 0);
+		} while (ret > 0);
+		bm_pool_free(pool->params.bpid);
+	}
+	kfree(pool);
+}
+EXPORT_SYMBOL(bman_free_pool);
+
+const struct bman_pool_params *bman_get_params(const struct bman_pool *pool)
+{
+	return &pool->params;
+}
+EXPORT_SYMBOL(bman_get_params);
+
+static noinline void update_rcr_ci(struct bman_portal *p, u8 avail)
+{
+	if (avail)
+		bm_rcr_cce_prefetch(&p->p);
+	else
+		bm_rcr_cce_update(&p->p);
+}
+
+int bman_rcr_is_empty(void)
+{
+	__maybe_unused unsigned long irqflags;
+	struct bman_portal *p = get_affine_portal();
+	u8 avail;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	update_rcr_ci(p, 0);
+	avail = bm_rcr_get_fill(&p->p);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return (avail == 0);
+}
+EXPORT_SYMBOL(bman_rcr_is_empty);
+
+static inline struct bm_rcr_entry *try_rel_start(struct bman_portal **p,
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+					__maybe_unused struct bman_pool *pool,
+#endif
+					__maybe_unused unsigned long *irqflags,
+					__maybe_unused u32 flags)
+{
+	struct bm_rcr_entry *r;
+	u8 avail;
+
+	*p = get_affine_portal();
+	PORTAL_IRQ_LOCK(*p, (*irqflags));
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) &&
+			(flags & BMAN_RELEASE_FLAG_WAIT_SYNC))) {
+		if ((*p)->rcri_owned) {
+			PORTAL_IRQ_UNLOCK(*p, (*irqflags));
+			put_affine_portal();
+			return NULL;
+		}
+		(*p)->rcri_owned = pool;
+	}
+#endif
+	avail = bm_rcr_get_avail(&(*p)->p);
+	if (avail < 2)
+		update_rcr_ci(*p, avail);
+	r = bm_rcr_start(&(*p)->p);
+	if (unlikely(!r)) {
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+		if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) &&
+				(flags & BMAN_RELEASE_FLAG_WAIT_SYNC)))
+			(*p)->rcri_owned = NULL;
+#endif
+		PORTAL_IRQ_UNLOCK(*p, (*irqflags));
+		put_affine_portal();
+	}
+	return r;
+}
+
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+static noinline struct bm_rcr_entry *__wait_rel_start(struct bman_portal **p,
+					struct bman_pool *pool,
+					__maybe_unused unsigned long *irqflags,
+					u32 flags)
+{
+	struct bm_rcr_entry *rcr = try_rel_start(p, pool, irqflags, flags);
+	if (!rcr)
+		bm_rcr_set_ithresh(&(*p)->p, 1);
+	return rcr;
+}
+
+static noinline struct bm_rcr_entry *wait_rel_start(struct bman_portal **p,
+					struct bman_pool *pool,
+					__maybe_unused unsigned long *irqflags,
+					u32 flags)
+{
+	struct bm_rcr_entry *rcr;
+#ifndef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	pool = NULL;
+#endif
+	if (flags & BMAN_RELEASE_FLAG_WAIT_INT)
+		wait_event_interruptible(affine_queue,
+			(rcr = __wait_rel_start(p, pool, irqflags, flags)));
+	else
+		wait_event(affine_queue,
+			(rcr = __wait_rel_start(p, pool, irqflags, flags)));
+	return rcr;
+}
+#endif
+
+/* to facilitate better copying of bufs into the ring without either (a) copying
+ * noise into the first byte (prematurely triggering the command), nor (b) being
+ * very inefficient by copying small fields using read-modify-write */
+struct overlay_bm_buffer {
+	u32 first;
+	u32 second;
+};
+
+static inline int __bman_release(struct bman_pool *pool,
+			const struct bm_buffer *bufs, u8 num, u32 flags)
+{
+	struct bman_portal *p;
+	struct bm_rcr_entry *r;
+	struct overlay_bm_buffer *o_dest;
+	struct overlay_bm_buffer *o_src = (struct overlay_bm_buffer *)&bufs[0];
+	__maybe_unused unsigned long irqflags;
+	u32 i = num - 1;
+
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+	if (flags & BMAN_RELEASE_FLAG_WAIT)
+		r = wait_rel_start(&p, pool, &irqflags, flags);
+	else
+		r = try_rel_start(&p, pool, &irqflags, flags);
+#else
+	r = try_rel_start(&p, &irqflags, flags);
+#endif
+	if (!r)
+		return -EBUSY;
+	/* We can copy all but the first entry, as this can trigger badness
+	 * with the valid-bit. Use the overlay to mask the verb byte. */
+	o_dest = (struct overlay_bm_buffer *)&r->bufs[0];
+	o_dest->first = (o_src->first & 0x0000ffff) |
+		(((u32)pool->params.bpid << 16) & 0x00ff0000);
+	o_dest->second = o_src->second;
+	if (i)
+		copy_words(&r->bufs[1], &bufs[1], i * sizeof(bufs[0]));
+	bm_rcr_pvb_commit(&p->p, BM_RCR_VERB_CMD_BPID_SINGLE |
+			(num & BM_RCR_VERB_BUFCOUNT_MASK));
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	/* if we wish to sync we need to set the threshold after h/w sees the
+	 * new ring entry. As we're mixing cache-enabled and cache-inhibited
+	 * accesses, this requires a heavy-weight sync. */
+	if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) &&
+			(flags & BMAN_RELEASE_FLAG_WAIT_SYNC))) {
+		hwsync();
+		bm_rcr_set_ithresh(&p->p, 1);
+	}
+#endif
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) &&
+			(flags & BMAN_RELEASE_FLAG_WAIT_SYNC))) {
+		if (flags & BMAN_RELEASE_FLAG_WAIT_INT)
+			wait_event_interruptible(affine_queue,
+					(p->rcri_owned != pool));
+		else
+			wait_event(affine_queue, (p->rcri_owned != pool));
+	}
+#endif
+	return 0;
+}
+
+int bman_release(struct bman_pool *pool, const struct bm_buffer *bufs, u8 num,
+			u32 flags)
+{
+	int ret = 0;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (!num || (num > 8))
+		return -EINVAL;
+	if (pool->params.flags & BMAN_POOL_FLAG_NO_RELEASE)
+		return -EINVAL;
+	if (!atomic_dec_and_test(&pool->in_use)) {
+		pr_crit("Parallel attempts to enter bman_released() detected.");
+		panic("only one instance of bman_released/acquired allowed");
+	}
+#endif
+	/* Without stockpile, this API is a pass-through to the h/w operation */
+	if (!(pool->params.flags & BMAN_POOL_FLAG_STOCKPILE)) {
+		ret = __bman_release(pool, bufs, num, flags);
+		goto release_done;
+	}
+	/* This needs some explanation. Adding the given buffers may take the
+	 * stockpile over the threshold, but in fact the stockpile may already
+	 * *be* over the threshold if a previous release-to-hw attempt had
+	 * failed. So we have 3 cases to cover;
+	 *   1. we add to the stockpile and don't hit the threshold,
+	 *   2. we add to the stockpile, hit the threshold and release-to-hw,
+	 *   3. we have to release-to-hw before adding to the stockpile
+	 *      (not enough room in the stockpile for case 2).
+	 * Our constraints on thresholds guarantee that in case 3, there must be
+	 * at least 8 bufs already in the stockpile, so all release-to-hw ops
+	 * are for 8 bufs. Despite all this, the API must indicate whether the
+	 * given buffers were taken off the caller's hands, irrespective of
+	 * whether a release-to-hw was attempted. */
+	while (num) {
+		/* Add buffers to stockpile if they fit */
+		if ((pool->sp_fill + num) < BMAN_STOCKPILE_SZ) {
+			copy_words(pool->sp + pool->sp_fill, bufs,
+				sizeof(struct bm_buffer) * num);
+			pool->sp_fill += num;
+			num = 0; /* --> will return success no matter what */
+		}
+		/* Do hw op if hitting the high-water threshold */
+		if ((pool->sp_fill + num) >= BMAN_STOCKPILE_HIGH) {
+			ret = __bman_release(pool,
+				pool->sp + (pool->sp_fill - 8), 8, flags);
+			if (ret) {
+				ret = (num ? ret : 0);
+				goto release_done;
+			}
+			pool->sp_fill -= 8;
+		}
+	}
+release_done:
+#ifdef CONFIG_FSL_DPA_CHECKING
+	atomic_inc(&pool->in_use);
+#endif
+	return ret;
+}
+EXPORT_SYMBOL(bman_release);
+
+static inline int __bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs,
+					u8 num)
+{
+	struct bman_portal *p = get_affine_portal();
+	struct bm_mc_command *mcc;
+	struct bm_mc_result *mcr;
+	__maybe_unused unsigned long irqflags;
+	int ret;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	mcc = bm_mc_start(&p->p);
+	mcc->acquire.bpid = pool->params.bpid;
+	bm_mc_commit(&p->p, BM_MCC_VERB_CMD_ACQUIRE |
+			(num & BM_MCC_VERB_ACQUIRE_BUFCOUNT));
+	while (!(mcr = bm_mc_result(&p->p)))
+		cpu_relax();
+	ret = mcr->verb & BM_MCR_VERB_ACQUIRE_BUFCOUNT;
+	if (bufs)
+		copy_words(&bufs[0], &mcr->acquire.bufs[0],
+				num * sizeof(bufs[0]));
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (ret != num)
+		ret = -ENOMEM;
+	return ret;
+}
+
+int bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs, u8 num,
+			u32 flags)
+{
+	int ret = 0;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (!num || (num > 8))
+		return -EINVAL;
+	if (pool->params.flags & BMAN_POOL_FLAG_ONLY_RELEASE)
+		return -EINVAL;
+	if (!atomic_dec_and_test(&pool->in_use)) {
+		pr_crit("Parallel attempts to enter bman_acquire() detected.");
+		panic("only one instance of bman_released/acquired allowed");
+	}
+#endif
+	/* Without stockpile, this API is a pass-through to the h/w operation */
+	if (!(pool->params.flags & BMAN_POOL_FLAG_STOCKPILE)) {
+		ret = __bman_acquire(pool, bufs, num);
+		goto acquire_done;
+	}
+	/* Only need a h/w op if we'll hit the low-water thresh */
+	if (!(flags & BMAN_ACQUIRE_FLAG_STOCKPILE) &&
+			(pool->sp_fill <= (BMAN_STOCKPILE_LOW + num))) {
+		/* refill stockpile with max amount, but if max amount
+		 * isn't available, try amount the user wants */
+		int bufcount = 8;
+		ret = __bman_acquire(pool, pool->sp + pool->sp_fill, bufcount);
+		if (ret < 0 && bufcount != num) {
+			bufcount = num;
+			/* Maybe buffer pool has less than 8 */
+			ret = __bman_acquire(pool, pool->sp + pool->sp_fill,
+						bufcount);
+		}
+		if (ret < 0)
+			goto hw_starved;
+		DPA_ASSERT(ret == bufcount);
+		pool->sp_fill += bufcount;
+	} else {
+hw_starved:
+		if (pool->sp_fill < num) {
+			ret = -ENOMEM;
+			goto acquire_done;
+		}
+	}
+	copy_words(bufs, pool->sp + (pool->sp_fill - num),
+		sizeof(struct bm_buffer) * num);
+	pool->sp_fill -= num;
+	ret = num;
+acquire_done:
+#ifdef CONFIG_FSL_DPA_CHECKING
+	atomic_inc(&pool->in_use);
+#endif
+	return ret;
+}
+EXPORT_SYMBOL(bman_acquire);
+
+int bman_flush_stockpile(struct bman_pool *pool, u32 flags)
+{
+	u8 num;
+	int ret;
+
+	while (pool->sp_fill) {
+		num = ((pool->sp_fill > 8) ? 8 : pool->sp_fill);
+		ret = __bman_release(pool, pool->sp + (pool->sp_fill - num),
+				     num, flags);
+		if (ret)
+			return ret;
+		pool->sp_fill -= num;
+	}
+	return 0;
+}
+EXPORT_SYMBOL(bman_flush_stockpile);
+
+int bman_query_pools(struct bm_pool_state *state)
+{
+	struct bman_portal *p = get_affine_portal();
+	struct bm_mc_result *mcr;
+	__maybe_unused unsigned long irqflags;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	bm_mc_start(&p->p);
+	bm_mc_commit(&p->p, BM_MCC_VERB_CMD_QUERY);
+	while (!(mcr = bm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & BM_MCR_VERB_CMD_MASK) == BM_MCR_VERB_CMD_QUERY);
+	*state = mcr->query;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return 0;
+}
+EXPORT_SYMBOL(bman_query_pools);
+
diff --git a/drivers/staging/fsl_qbman/bman_low.h b/drivers/staging/fsl_qbman/bman_low.h
new file mode 100644
index 0000000..262bae7
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_low.h
@@ -0,0 +1,494 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "bman_private.h"
+
+/***************************/
+/* Portal register assists */
+/***************************/
+
+/* Cache-inhibited register offsets */
+#define REG_RCR_PI_CINH		0x0000
+#define REG_RCR_CI_CINH		0x0004
+#define REG_RCR_ITR		0x0008
+#define REG_CFG			0x0100
+#define REG_SCN(n)		(0x0200 + ((n) << 2))
+#define REG_ISR			0x0e00
+
+/* Cache-enabled register offsets */
+#define CL_CR			0x0000
+#define CL_RR0			0x0100
+#define CL_RR1			0x0140
+#define CL_RCR			0x1000
+#define CL_RCR_PI_CENA		0x3000
+#define CL_RCR_CI_CENA		0x3100
+
+/* BTW, the drivers (and h/w programming model) already obtain the required
+ * synchronisation for portal accesses via lwsync(), hwsync(), and
+ * data-dependencies. Use of barrier()s or other order-preserving primitives
+ * simply degrade performance. Hence the use of the __raw_*() interfaces, which
+ * simply ensure that the compiler treats the portal registers as volatile (ie.
+ * non-coherent). */
+
+/* Cache-inhibited register access. */
+#define __bm_in(bm, o)		__raw_readl((bm)->addr_ci + (o))
+#define __bm_out(bm, o, val)	__raw_writel((val), (bm)->addr_ci + (o))
+#define bm_in(reg)		__bm_in(&portal->addr, REG_##reg)
+#define bm_out(reg, val)	__bm_out(&portal->addr, REG_##reg, val)
+
+/* Cache-enabled (index) register access */
+#define __bm_cl_touch_ro(bm, o) dcbt_ro((bm)->addr_ce + (o))
+#define __bm_cl_touch_rw(bm, o) dcbt_rw((bm)->addr_ce + (o))
+#define __bm_cl_in(bm, o)	__raw_readl((bm)->addr_ce + (o))
+#define __bm_cl_out(bm, o, val) \
+	do { \
+		u32 *__tmpclout = (bm)->addr_ce + (o); \
+		__raw_writel((val), __tmpclout); \
+		dcbf(__tmpclout); \
+	} while (0)
+#define __bm_cl_invalidate(bm, o) dcbi((bm)->addr_ce + (o))
+#define bm_cl_touch_ro(reg)	__bm_cl_touch_ro(&portal->addr, CL_##reg##_CENA)
+#define bm_cl_touch_rw(reg)	__bm_cl_touch_rw(&portal->addr, CL_##reg##_CENA)
+#define bm_cl_in(reg)		__bm_cl_in(&portal->addr, CL_##reg##_CENA)
+#define bm_cl_out(reg, val)	__bm_cl_out(&portal->addr, CL_##reg##_CENA, val)
+#define bm_cl_invalidate(reg) __bm_cl_invalidate(&portal->addr, CL_##reg##_CENA)
+
+/* Cyclic helper for rings. FIXME: once we are able to do fine-grain perf
+ * analysis, look at using the "extra" bit in the ring index registers to avoid
+ * cyclic issues. */
+static inline u8 cyc_diff(u8 ringsize, u8 first, u8 last)
+{
+	/* 'first' is included, 'last' is excluded */
+	if (first <= last)
+		return last - first;
+	return ringsize + last - first;
+}
+
+/* Portal modes.
+ *   Enum types;
+ *     pmode == production mode
+ *     cmode == consumption mode,
+ *   Enum values use 3 letter codes. First letter matches the portal mode,
+ *   remaining two letters indicate;
+ *     ci == cache-inhibited portal register
+ *     ce == cache-enabled portal register
+ *     vb == in-band valid-bit (cache-enabled)
+ */
+enum bm_rcr_pmode {		/* matches BCSP_CFG::RPM */
+	bm_rcr_pci = 0,		/* PI index, cache-inhibited */
+	bm_rcr_pce = 1,		/* PI index, cache-enabled */
+	bm_rcr_pvb = 2		/* valid-bit */
+};
+enum bm_rcr_cmode {		/* s/w-only */
+	bm_rcr_cci,		/* CI index, cache-inhibited */
+	bm_rcr_cce		/* CI index, cache-enabled */
+};
+
+
+/* ------------------------- */
+/* --- Portal structures --- */
+
+#define BM_RCR_SIZE		8
+
+struct bm_rcr {
+	struct bm_rcr_entry *ring, *cursor;
+	u8 ci, available, ithresh, vbit;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	u32 busy;
+	enum bm_rcr_pmode pmode;
+	enum bm_rcr_cmode cmode;
+#endif
+};
+
+struct bm_mc {
+	struct bm_mc_command *cr;
+	struct bm_mc_result *rr;
+	u8 rridx, vbit;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	enum {
+		/* Can only be _mc_start()ed */
+		mc_idle,
+		/* Can only be _mc_commit()ed or _mc_abort()ed */
+		mc_user,
+		/* Can only be _mc_retry()ed */
+		mc_hw
+	} state;
+#endif
+};
+
+struct bm_addr {
+	void __iomem *addr_ce;	/* cache-enabled */
+	void __iomem *addr_ci;	/* cache-inhibited */
+};
+
+struct bm_portal {
+	struct bm_addr addr;
+	struct bm_rcr rcr;
+	struct bm_mc mc;
+	struct bm_portal_config config;
+} ____cacheline_aligned;
+
+
+/* --------------- */
+/* --- RCR API --- */
+
+/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */
+#define RCR_CARRYCLEAR(p) \
+	(void *)((unsigned long)(p) & (~(unsigned long)(BM_RCR_SIZE << 6)))
+
+/* Bit-wise logic to convert a ring pointer to a ring index */
+static inline u8 RCR_PTR2IDX(struct bm_rcr_entry *e)
+{
+	return ((uintptr_t)e >> 6) & (BM_RCR_SIZE - 1);
+}
+
+/* Increment the 'cursor' ring pointer, taking 'vbit' into account */
+static inline void RCR_INC(struct bm_rcr *rcr)
+{
+	/* NB: this is odd-looking, but experiments show that it generates
+	 * fast code with essentially no branching overheads. We increment to
+	 * the next RCR pointer and handle overflow and 'vbit'. */
+	struct bm_rcr_entry *partial = rcr->cursor + 1;
+	rcr->cursor = RCR_CARRYCLEAR(partial);
+	if (partial != rcr->cursor)
+		rcr->vbit ^= BM_RCR_VERB_VBIT;
+}
+
+static inline int bm_rcr_init(struct bm_portal *portal, enum bm_rcr_pmode pmode,
+		__maybe_unused enum bm_rcr_cmode cmode)
+{
+	/* This use of 'register', as well as all other occurances, is because
+	 * it has been observed to generate much faster code with gcc than is
+	 * otherwise the case. */
+	register struct bm_rcr *rcr = &portal->rcr;
+	u32 cfg;
+	u8 pi;
+
+	rcr->ring = portal->addr.addr_ce + CL_RCR;
+	rcr->ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1);
+	pi = bm_in(RCR_PI_CINH) & (BM_RCR_SIZE - 1);
+	rcr->cursor = rcr->ring + pi;
+	rcr->vbit = (bm_in(RCR_PI_CINH) & BM_RCR_SIZE) ?  BM_RCR_VERB_VBIT : 0;
+	rcr->available = BM_RCR_SIZE - 1 - cyc_diff(BM_RCR_SIZE, rcr->ci, pi);
+	rcr->ithresh = bm_in(RCR_ITR);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	rcr->busy = 0;
+	rcr->pmode = pmode;
+	rcr->cmode = cmode;
+#endif
+	cfg = (bm_in(CFG) & 0xffffffe0) | (pmode & 0x3); /* BCSP_CFG::RPM */
+	bm_out(CFG, cfg);
+	return 0;
+}
+
+static inline void bm_rcr_finish(struct bm_portal *portal)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	u8 pi = bm_in(RCR_PI_CINH) & (BM_RCR_SIZE - 1);
+	u8 ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1);
+	DPA_ASSERT(!rcr->busy);
+	if (pi != RCR_PTR2IDX(rcr->cursor))
+		pr_crit("losing uncommited RCR entries\n");
+	if (ci != rcr->ci)
+		pr_crit("missing existing RCR completions\n");
+	if (rcr->ci != RCR_PTR2IDX(rcr->cursor))
+		pr_crit("RCR destroyed unquiesced\n");
+}
+
+static inline struct bm_rcr_entry *bm_rcr_start(struct bm_portal *portal)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	DPA_ASSERT(!rcr->busy);
+	if (!rcr->available)
+		return NULL;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	rcr->busy = 1;
+#endif
+	dcbz_64(rcr->cursor);
+	return rcr->cursor;
+}
+
+static inline void bm_rcr_abort(struct bm_portal *portal)
+{
+	__maybe_unused register struct bm_rcr *rcr = &portal->rcr;
+	DPA_ASSERT(rcr->busy);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	rcr->busy = 0;
+#endif
+}
+
+static inline struct bm_rcr_entry *bm_rcr_pend_and_next(
+					struct bm_portal *portal, u8 myverb)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	DPA_ASSERT(rcr->busy);
+	DPA_ASSERT(rcr->pmode != bm_rcr_pvb);
+	if (rcr->available == 1)
+		return NULL;
+	rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit;
+	dcbf_64(rcr->cursor);
+	RCR_INC(rcr);
+	rcr->available--;
+	dcbz_64(rcr->cursor);
+	return rcr->cursor;
+}
+
+static inline void bm_rcr_pci_commit(struct bm_portal *portal, u8 myverb)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	DPA_ASSERT(rcr->busy);
+	DPA_ASSERT(rcr->pmode == bm_rcr_pci);
+	rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit;
+	RCR_INC(rcr);
+	rcr->available--;
+	hwsync();
+	bm_out(RCR_PI_CINH, RCR_PTR2IDX(rcr->cursor));
+#ifdef CONFIG_FSL_DPA_CHECKING
+	rcr->busy = 0;
+#endif
+}
+
+static inline void bm_rcr_pce_prefetch(struct bm_portal *portal)
+{
+	__maybe_unused register struct bm_rcr *rcr = &portal->rcr;
+	DPA_ASSERT(rcr->pmode == bm_rcr_pce);
+	bm_cl_invalidate(RCR_PI);
+	bm_cl_touch_rw(RCR_PI);
+}
+
+static inline void bm_rcr_pce_commit(struct bm_portal *portal, u8 myverb)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	DPA_ASSERT(rcr->busy);
+	DPA_ASSERT(rcr->pmode == bm_rcr_pce);
+	rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit;
+	RCR_INC(rcr);
+	rcr->available--;
+	lwsync();
+	bm_cl_out(RCR_PI, RCR_PTR2IDX(rcr->cursor));
+#ifdef CONFIG_FSL_DPA_CHECKING
+	rcr->busy = 0;
+#endif
+}
+
+static inline void bm_rcr_pvb_commit(struct bm_portal *portal, u8 myverb)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	struct bm_rcr_entry *rcursor;
+	DPA_ASSERT(rcr->busy);
+	DPA_ASSERT(rcr->pmode == bm_rcr_pvb);
+	lwsync();
+	rcursor = rcr->cursor;
+	rcursor->__dont_write_directly__verb = myverb | rcr->vbit;
+	dcbf_64(rcursor);
+	RCR_INC(rcr);
+	rcr->available--;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	rcr->busy = 0;
+#endif
+}
+
+static inline u8 bm_rcr_cci_update(struct bm_portal *portal)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	u8 diff, old_ci = rcr->ci;
+	DPA_ASSERT(rcr->cmode == bm_rcr_cci);
+	rcr->ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1);
+	diff = cyc_diff(BM_RCR_SIZE, old_ci, rcr->ci);
+	rcr->available += diff;
+	return diff;
+}
+
+static inline void bm_rcr_cce_prefetch(struct bm_portal *portal)
+{
+	__maybe_unused register struct bm_rcr *rcr = &portal->rcr;
+	DPA_ASSERT(rcr->cmode == bm_rcr_cce);
+	bm_cl_touch_ro(RCR_CI);
+}
+
+static inline u8 bm_rcr_cce_update(struct bm_portal *portal)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	u8 diff, old_ci = rcr->ci;
+	DPA_ASSERT(rcr->cmode == bm_rcr_cce);
+	rcr->ci = bm_cl_in(RCR_CI) & (BM_RCR_SIZE - 1);
+	bm_cl_invalidate(RCR_CI);
+	diff = cyc_diff(BM_RCR_SIZE, old_ci, rcr->ci);
+	rcr->available += diff;
+	return diff;
+}
+
+static inline u8 bm_rcr_get_ithresh(struct bm_portal *portal)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	return rcr->ithresh;
+}
+
+static inline void bm_rcr_set_ithresh(struct bm_portal *portal, u8 ithresh)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	rcr->ithresh = ithresh;
+	bm_out(RCR_ITR, ithresh);
+}
+
+static inline u8 bm_rcr_get_avail(struct bm_portal *portal)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	return rcr->available;
+}
+
+static inline u8 bm_rcr_get_fill(struct bm_portal *portal)
+{
+	register struct bm_rcr *rcr = &portal->rcr;
+	return BM_RCR_SIZE - 1 - rcr->available;
+}
+
+
+/* ------------------------------ */
+/* --- Management command API --- */
+
+static inline int bm_mc_init(struct bm_portal *portal)
+{
+	register struct bm_mc *mc = &portal->mc;
+	mc->cr = portal->addr.addr_ce + CL_CR;
+	mc->rr = portal->addr.addr_ce + CL_RR0;
+	mc->rridx = (__raw_readb(&mc->cr->__dont_write_directly__verb) &
+			BM_MCC_VERB_VBIT) ?  0 : 1;
+	mc->vbit = mc->rridx ? BM_MCC_VERB_VBIT : 0;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_idle;
+#endif
+	return 0;
+}
+
+static inline void bm_mc_finish(struct bm_portal *portal)
+{
+	__maybe_unused register struct bm_mc *mc = &portal->mc;
+	DPA_ASSERT(mc->state == mc_idle);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (mc->state != mc_idle)
+		pr_crit("Losing incomplete MC command\n");
+#endif
+}
+
+static inline struct bm_mc_command *bm_mc_start(struct bm_portal *portal)
+{
+	register struct bm_mc *mc = &portal->mc;
+	DPA_ASSERT(mc->state == mc_idle);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_user;
+#endif
+	dcbz_64(mc->cr);
+	return mc->cr;
+}
+
+static inline void bm_mc_abort(struct bm_portal *portal)
+{
+	__maybe_unused register struct bm_mc *mc = &portal->mc;
+	DPA_ASSERT(mc->state == mc_user);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_idle;
+#endif
+}
+
+static inline void bm_mc_commit(struct bm_portal *portal, u8 myverb)
+{
+	register struct bm_mc *mc = &portal->mc;
+	struct bm_mc_result *rr = mc->rr + mc->rridx;
+	DPA_ASSERT(mc->state == mc_user);
+	lwsync();
+	mc->cr->__dont_write_directly__verb = myverb | mc->vbit;
+	dcbf(mc->cr);
+	dcbit_ro(rr);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_hw;
+#endif
+}
+
+static inline struct bm_mc_result *bm_mc_result(struct bm_portal *portal)
+{
+	register struct bm_mc *mc = &portal->mc;
+	struct bm_mc_result *rr = mc->rr + mc->rridx;
+	DPA_ASSERT(mc->state == mc_hw);
+	/* The inactive response register's verb byte always returns zero until
+	 * its command is submitted and completed. This includes the valid-bit,
+	 * in case you were wondering... */
+	if (!__raw_readb(&rr->verb)) {
+		dcbit_ro(rr);
+		return NULL;
+	}
+	mc->rridx ^= 1;
+	mc->vbit ^= BM_MCC_VERB_VBIT;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_idle;
+#endif
+	return rr;
+}
+
+
+/* ------------------------------------- */
+/* --- Portal interrupt register API --- */
+
+static inline int bm_isr_init(__always_unused struct bm_portal *portal)
+{
+	return 0;
+}
+
+static inline void bm_isr_finish(__always_unused struct bm_portal *portal)
+{
+}
+
+#define SCN_REG(bpid) REG_SCN((bpid) / 32)
+#define SCN_BIT(bpid) (0x80000000 >> (bpid & 31))
+static inline void bm_isr_bscn_mask(struct bm_portal *portal, u8 bpid,
+					int enable)
+{
+	u32 val;
+	DPA_ASSERT(bpid < bman_pool_max);
+	/* REG_SCN for bpid=0..31, REG_SCN+4 for bpid=32..63 */
+	val = __bm_in(&portal->addr, SCN_REG(bpid));
+	if (enable)
+		val |= SCN_BIT(bpid);
+	else
+		val &= ~SCN_BIT(bpid);
+	__bm_out(&portal->addr, SCN_REG(bpid), val);
+}
+
+static inline u32 __bm_isr_read(struct bm_portal *portal, enum bm_isr_reg n)
+{
+	return __bm_in(&portal->addr, REG_ISR + (n << 2));
+}
+
+static inline void __bm_isr_write(struct bm_portal *portal, enum bm_isr_reg n,
+					u32 val)
+{
+	__bm_out(&portal->addr, REG_ISR + (n << 2), val);
+}
diff --git a/drivers/staging/fsl_qbman/bman_private.h b/drivers/staging/fsl_qbman/bman_private.h
new file mode 100644
index 0000000..375da15
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_private.h
@@ -0,0 +1,149 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "dpa_sys.h"
+#include <linux/fsl_bman.h>
+
+/* Revision info (for errata and feature handling) */
+#define BMAN_REV10 0x0100
+#define BMAN_REV20 0x0200
+extern u16 bman_ip_rev; /* 0 if uninitialised, otherwise QMAN_REVx */
+
+/*
+ * Global variables of the max portal/pool number this bman version supported
+ */
+extern u16 bman_pool_max;
+
+/* used by CCSR and portal interrupt code */
+enum bm_isr_reg {
+	bm_isr_status = 0,
+	bm_isr_enable = 1,
+	bm_isr_disable = 2,
+	bm_isr_inhibit = 3
+};
+
+#define BM_ADDR_CE 0
+#define BM_ADDR_CI 1
+struct bm_portal_config {
+	/* Corenet portal addresses;
+	 * [0]==cache-enabled, [1]==cache-inhibited. */
+	__iomem void *addr_virt[2];
+	struct resource addr_phys[2];
+	/* Allow these to be joined in lists */
+	struct list_head list;
+	/* User-visible portal configuration settings */
+	struct bman_portal_config public_cfg;
+};
+
+#ifdef CONFIG_FSL_BMAN_CONFIG
+/* Hooks from bman_driver.c to bman_config.c */
+int bman_init_error_int(struct device_node *node);
+#endif
+
+/* Hooks from bman_driver.c in to bman_high.c */
+struct bman_portal *bman_create_affine_portal(
+			const struct bm_portal_config *config,
+			int recovery_mode);
+struct bman_portal *bman_create_affine_slave(struct bman_portal *redirect);
+const struct bm_portal_config *bman_destroy_affine_portal(void);
+void bman_recovery_exit_local(void);
+
+/* Pool logic in the portal driver, during initialisation, needs to know if
+ * there's access to CCSR or not (if not, it'll cripple the pool allocator). */
+#ifdef CONFIG_FSL_BMAN_CONFIG
+int bman_have_ccsr(void);
+#else
+#define bman_have_ccsr() 0
+#endif
+
+/* Stockpile build constants. The _LOW value: when bman_acquire() is called and
+ * the stockpile fill-level is <= _LOW, an acquire is attempted from h/w but it
+ * might fail (if the buffer pool is depleted). So this value provides some
+ * "stagger" in that the bman_acquire() function will only fail if lots of bufs
+ * are requested at once or if h/w has been tested a couple of times without
+ * luck. The _HIGH value: when bman_release() is called and the stockpile
+ * fill-level is >= _HIGH, a release is attempted to h/w but it might fail (if
+ * the release ring is full). So this value provides some "stagger" so that
+ * ring-access is retried a couple of times prior to the API returning a
+ * failure. The following *must* be true;
+ *   BMAN_STOCKPILE_HIGH-BMAN_STOCKPILE_LOW > 8
+ *     (to avoid thrashing)
+ *   BMAN_STOCKPILE_SZ >= 16
+ *     (as the release logic expects to either send 8 buffers to hw prior to
+ *     adding the given buffers to the stockpile or add the buffers to the
+ *     stockpile before sending 8 to hw, as the API must be an all-or-nothing
+ *     success/fail.)
+ */
+#define BMAN_STOCKPILE_SZ   16u /* number of bufs in per-pool cache */
+#define BMAN_STOCKPILE_LOW  2u  /* when fill is <= this, acquire from hw */
+#define BMAN_STOCKPILE_HIGH 14u /* when fill is >= this, release to hw */
+
+/*************************************************/
+/*   BMan s/w corenet portal, low-level i/face   */
+/*************************************************/
+
+/* Used by all portal interrupt registers except 'inhibit'. NB, some of these
+ * definitions are exported for use by the bman_irqsource_***() APIs, so are
+ * commented-out here. */
+#if 0
+#define BM_PIRQ_RCRI	0x00000002	/* RCR Ring (below threshold) */
+#define BM_PIRQ_BSCN	0x00000001	/* Buffer depletion State Change */
+#endif
+/* This mask contains all the "irqsource" bits visible to API users */
+#define BM_PIRQ_VISIBLE	(BM_PIRQ_RCRI | BM_PIRQ_BSCN)
+
+/* These are bm_<reg>_<verb>(). So for example, bm_disable_write() means "write
+ * the disable register" rather than "disable the ability to write". */
+#define bm_isr_status_read(bm)		__bm_isr_read(bm, bm_isr_status)
+#define bm_isr_status_clear(bm, m)	__bm_isr_write(bm, bm_isr_status, m)
+#define bm_isr_enable_read(bm)		__bm_isr_read(bm, bm_isr_enable)
+#define bm_isr_enable_write(bm, v)	__bm_isr_write(bm, bm_isr_enable, v)
+#define bm_isr_disable_read(bm)		__bm_isr_read(bm, bm_isr_disable)
+#define bm_isr_disable_write(bm, v)	__bm_isr_write(bm, bm_isr_disable, v)
+#define bm_isr_inhibit(bm)		__bm_isr_write(bm, bm_isr_inhibit, 1)
+#define bm_isr_uninhibit(bm)		__bm_isr_write(bm, bm_isr_inhibit, 0)
+
+/* Allocate/release an unreserved buffer pool id */
+int bm_pool_new(u32 *bpid);
+void bm_pool_free(u32 bpid);
+
+#ifdef CONFIG_FSL_BMAN_CONFIG
+/* Set depletion thresholds associated with a buffer pool. Requires that the
+ * operating system have access to Bman CCSR (ie. compiled in support and
+ * run-time access courtesy of the device-tree). */
+int bm_pool_set(u32 bpid, const u32 *thresholds);
+#define BM_POOL_THRESH_SW_ENTER 0
+#define BM_POOL_THRESH_SW_EXIT  1
+#define BM_POOL_THRESH_HW_ENTER 2
+#define BM_POOL_THRESH_HW_EXIT  3
+
+#endif /* CONFIG_FSL_BMAN_CONFIG */
+
diff --git a/drivers/staging/fsl_qbman/bman_test.c b/drivers/staging/fsl_qbman/bman_test.c
new file mode 100644
index 0000000..4cf29a6
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_test.c
@@ -0,0 +1,57 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "bman_test.h"
+
+MODULE_AUTHOR("Geoff Thorpe");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_DESCRIPTION("Bman testing");
+
+static int test_init(void)
+{
+#ifdef CONFIG_FSL_BMAN_TEST_HIGH
+	int loop = 1;
+	while (loop--)
+		bman_test_high();
+#endif
+#ifdef CONFIG_FSL_BMAN_TEST_THRESH
+	bman_test_thresh();
+#endif
+	return 0;
+}
+
+static void test_exit(void)
+{
+}
+
+module_init(test_init);
+module_exit(test_exit);
+
diff --git a/drivers/staging/fsl_qbman/bman_test.h b/drivers/staging/fsl_qbman/bman_test.h
new file mode 100644
index 0000000..b7acb4f
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_test.h
@@ -0,0 +1,92 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/io.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/interrupt.h>
+#include <linux/delay.h>
+#include <linux/kthread.h>
+
+#include <linux/fsl_bman.h>
+
+void bman_test_high(void);
+void bman_test_thresh(void);
+
+static inline void __hexdump(unsigned long start, unsigned long end,
+			unsigned long p, size_t sz, unsigned char *c)
+{
+	while (start < end) {
+		unsigned int pos = 0;
+		char buf[64];
+		int nl = 0;
+		pos += sprintf(buf + pos, "%08lx: ", start);
+		do {
+			if ((start < p) || (start >= (p + sz)))
+				pos += sprintf(buf + pos, "..");
+			else
+				pos += sprintf(buf + pos, "%02x", *(c++));
+			if (!(++start & 15)) {
+				buf[pos++] = '\n';
+				nl = 1;
+			} else {
+				nl = 0;
+				if(!(start & 1))
+					buf[pos++] = ' ';
+				if(!(start & 3))
+					buf[pos++] = ' ';
+			}
+		} while (start & 15);
+		if (!nl)
+			buf[pos++] = '\n';
+		buf[pos] = '\0';
+		pr_info("%s", buf);
+	}
+}
+static inline void hexdump(void *ptr, size_t sz)
+{
+	unsigned long p = (unsigned long)ptr;
+	unsigned long start = p & ~(unsigned long)15;
+	unsigned long end = (p + sz + 15) & ~(unsigned long)15;
+	unsigned char *c = ptr;
+	__hexdump(start, end, p, sz, c);
+}
+static inline void hexdump_by_cl(void *ptr, size_t sz)
+{
+	unsigned long p = (unsigned long)ptr;
+	unsigned long start = p & ~(unsigned long)63;
+	unsigned long end = (p + sz + 63) & ~(unsigned long)63;
+	unsigned char *c = ptr;
+	__hexdump(start, end, p, sz, c);
+}
+
diff --git a/drivers/staging/fsl_qbman/bman_test_high.c b/drivers/staging/fsl_qbman/bman_test_high.c
new file mode 100644
index 0000000..a9d5cad
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_test_high.c
@@ -0,0 +1,181 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "bman_test.h"
+#include "bman_private.h"
+
+/*************/
+/* constants */
+/*************/
+
+#define PORTAL_OPAQUE	(void *)0xf00dbeef
+#define POOL_OPAQUE	(void *)0xdeadabba
+#define NUM_BUFS	93
+#define LOOPS		3
+#define BMAN_TOKEN_MASK 0x00FFFFFFFFFFLLU
+
+/***************/
+/* global vars */
+/***************/
+
+static struct bman_pool *pool;
+static int depleted;
+static struct bm_buffer bufs_in[NUM_BUFS] ____cacheline_aligned;
+static struct bm_buffer bufs_out[NUM_BUFS] ____cacheline_aligned;
+static int bufs_received;
+
+/* Predeclare the callback so we can instantiate pool parameters */
+static void depletion_cb(struct bman_portal *, struct bman_pool *, void *, int);
+
+/**********************/
+/* internal functions */
+/**********************/
+
+static void bufs_init(void)
+{
+	int i;
+	for (i = 0; i < NUM_BUFS; i++)
+		bm_buffer_set64(&bufs_in[i], 0xfedc01234567LLU * i);
+	bufs_received = 0;
+}
+
+static inline int bufs_cmp(const struct bm_buffer *a, const struct bm_buffer *b)
+{
+	if (bman_ip_rev == BMAN_REV20) {
+		/* On SoCs with Bman revison 2.0, Bman only respects the 40
+		 * LS-bits of buffer addresses, masking off the upper 8-bits on
+		 * release commands. The API provides for 48-bit addresses
+		 * because some SoCs support all 48-bits. When generating
+		 * garbage addresses for testing, we either need to zero the
+		 * upper 8-bits when releasing to Bman (otherwise we'll be
+		 * disappointed when the buffers we acquire back from Bman
+		 * don't match), or we need to mask the upper 8-bits off when
+		 * comparing. We do the latter.
+		 */
+		if ((bm_buffer_get64(a) & BMAN_TOKEN_MASK)
+				< (bm_buffer_get64(b) & BMAN_TOKEN_MASK))
+			return -1;
+		if ((bm_buffer_get64(a) & BMAN_TOKEN_MASK)
+				> (bm_buffer_get64(b) & BMAN_TOKEN_MASK))
+			return 1;
+	} else {
+		if (bm_buffer_get64(a) < bm_buffer_get64(b))
+			return -1;
+		if (bm_buffer_get64(a) > bm_buffer_get64(b))
+			return 1;
+	}
+
+	return 0;
+}
+
+static void bufs_confirm(void)
+{
+	int i, j;
+	for (i = 0; i < NUM_BUFS; i++) {
+		int matches = 0;
+		for (j = 0; j < NUM_BUFS; j++)
+			if (!bufs_cmp(&bufs_in[i], &bufs_out[j]))
+				matches++;
+		BUG_ON(matches != 1);
+	}
+}
+
+/********/
+/* test */
+/********/
+
+static void depletion_cb(struct bman_portal *__portal, struct bman_pool *__pool,
+			void *pool_ctx, int __depleted)
+{
+	BUG_ON(__pool != pool);
+	BUG_ON(pool_ctx != POOL_OPAQUE);
+	depleted = __depleted;
+}
+
+void bman_test_high(void)
+{
+	struct bman_pool_params pparams = {
+		.flags = BMAN_POOL_FLAG_DEPLETION | BMAN_POOL_FLAG_DYNAMIC_BPID,
+		.cb = depletion_cb,
+		.cb_ctx = POOL_OPAQUE,
+	};
+	int i, loops = LOOPS;
+
+	bufs_init();
+
+	pr_info("BMAN:  --- starting high-level test ---\n");
+
+	pool = bman_new_pool(&pparams);
+	BUG_ON(!pool);
+
+	/*******************/
+	/* Release buffers */
+	/*******************/
+do_loop:
+	i = 0;
+	while (i < NUM_BUFS) {
+		u32 flags = BMAN_RELEASE_FLAG_WAIT;
+		int num = 8;
+		if ((i + num) > NUM_BUFS)
+			num = NUM_BUFS - i;
+		if ((i + num) == NUM_BUFS)
+			flags |= BMAN_RELEASE_FLAG_WAIT_SYNC;
+		if (bman_release(pool, bufs_in + i, num, flags))
+			panic("bman_release() failed\n");
+		i += num;
+	}
+
+	/*******************/
+	/* Acquire buffers */
+	/*******************/
+	while (i > 0) {
+		int tmp, num = 8;
+		if (num > i)
+			num = i;
+		tmp = bman_acquire(pool, bufs_out + i - num, num, 0);
+		BUG_ON(tmp != num);
+		i -= num;
+	}
+	i = bman_acquire(pool, NULL, 1, 0);
+	BUG_ON(i > 0);
+
+	bufs_confirm();
+
+	if (--loops)
+		goto do_loop;
+
+	/************/
+	/* Clean up */
+	/************/
+	bman_free_pool(pool);
+	pr_info("BMAN:  --- finished high-level test ---\n");
+}
+
diff --git a/drivers/staging/fsl_qbman/bman_test_thresh.c b/drivers/staging/fsl_qbman/bman_test_thresh.c
new file mode 100644
index 0000000..d556e6d
--- /dev/null
+++ b/drivers/staging/fsl_qbman/bman_test_thresh.c
@@ -0,0 +1,197 @@ 
+/* Copyright 2010-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "bman_test.h"
+
+/* Test constants */
+#define TEST_NUMBUFS	129728
+#define TEST_EXIT	129536
+#define TEST_ENTRY	129024
+
+struct affine_test_data {
+	struct task_struct *t;
+	int cpu;
+	int expect_affinity;
+	int drain;
+	int num_enter;
+	int num_exit;
+	struct list_head node;
+	struct completion wakethread;
+	struct completion wakeparent;
+};
+
+static void cb_depletion(struct bman_portal *portal,
+			struct bman_pool *pool,
+			void *opaque,
+			int depleted)
+{
+	struct affine_test_data *data = opaque;
+	int c = smp_processor_id();
+	pr_info("cb_depletion: bpid=%d, depleted=%d, cpu=%d, original=%d\n",
+		bman_get_params(pool)->bpid, depleted, c, data->cpu);
+	/* We should be executing on the CPU of the thread that owns the pool if
+	 * and that CPU has an affine portal (ie. it isn't slaved). */
+	BUG_ON((c != data->cpu) && data->expect_affinity);
+	BUG_ON((c == data->cpu) && !data->expect_affinity);
+	if (depleted)
+		data->num_enter++;
+	else
+		data->num_exit++;
+}
+
+/* Params used to set up a pool, this also dynamically allocates a BPID */
+struct bman_pool_params params_nocb = {
+	.flags = BMAN_POOL_FLAG_DYNAMIC_BPID | BMAN_POOL_FLAG_THRESH,
+	.thresholds = { TEST_ENTRY, TEST_EXIT, 0, 0 }
+};
+
+/* Params used to set up each cpu's pool with callbacks enabled */
+struct bman_pool_params params_cb = {
+	.bpid = 0, /* will be replaced to match pool_nocb */
+	.flags = BMAN_POOL_FLAG_DEPLETION,
+	.cb = cb_depletion
+};
+
+static struct bman_pool *pool_nocb;
+static LIST_HEAD(threads);
+
+static int affine_test(void *__data)
+{
+	struct bman_pool *pool;
+	struct affine_test_data *data = __data;
+	struct bman_pool_params my_params = params_cb;
+
+	pr_info("thread %d: starting\n", data->cpu);
+	/* create the pool */
+	my_params.cb_ctx = data;
+	pool = bman_new_pool(&my_params);
+	BUG_ON(!pool);
+	complete(&data->wakeparent);
+	wait_for_completion(&data->wakethread);
+	init_completion(&data->wakethread);
+
+	/* if we're the drainer, we get signalled for that */
+	if (data->drain) {
+		struct bm_buffer buf;
+		int ret;
+		pr_info("thread %d: draining...\n", data->cpu);
+		do {
+			ret = bman_acquire(pool, &buf, 1, 0);
+		} while (ret > 0);
+		pr_info("thread %d: draining done.\n", data->cpu);
+		complete(&data->wakeparent);
+		wait_for_completion(&data->wakethread);
+		init_completion(&data->wakethread);
+	}
+
+	/* cleanup */
+	bman_free_pool(pool);
+	while (!kthread_should_stop())
+		cpu_relax();
+	pr_info("thread %d: exiting\n", data->cpu);
+	return 0;
+}
+
+static struct affine_test_data *start_affine_test(int cpu, int drain)
+{
+	struct affine_test_data *data = kmalloc(sizeof(*data), GFP_KERNEL);
+
+	if (!data)
+		return NULL;
+	data->cpu = cpu;
+	data->expect_affinity = cpumask_test_cpu(cpu, bman_affine_cpus());
+	data->drain = drain;
+	data->num_enter = 0;
+	data->num_exit = 0;
+	init_completion(&data->wakethread);
+	init_completion(&data->wakeparent);
+	list_add_tail(&data->node, &threads);
+	data->t = kthread_create(affine_test, data, "threshtest%d", cpu);
+	BUG_ON(IS_ERR(data->t));
+	kthread_bind(data->t, cpu);
+	wake_up_process(data->t);
+	return data;
+}
+
+void bman_test_thresh(void)
+{
+	int loop = TEST_NUMBUFS;
+	int ret, num_cpus = 0;
+	struct affine_test_data *data, *drainer = NULL;
+
+	pr_info("bman_test_thresh: start\n");
+
+	/* allocate a BPID and seed it */
+	pool_nocb = bman_new_pool(&params_nocb);
+	BUG_ON(!pool_nocb);
+	while (loop--) {
+		struct bm_buffer buf;
+		bm_buffer_set64(&buf, 0x0badbeef + loop);
+		ret = bman_release(pool_nocb, &buf, 1,
+					BMAN_RELEASE_FLAG_WAIT);
+		BUG_ON(ret);
+	}
+	while (!bman_rcr_is_empty())
+		cpu_relax();
+	pr_info("bman_test_thresh: buffers are in\n");
+
+	/* create threads and wait for them to create pools */
+	params_cb.bpid = bman_get_params(pool_nocb)->bpid;
+	for_each_cpu(loop, cpu_online_mask) {
+		data = start_affine_test(loop, drainer ? 0 : 1);
+		BUG_ON(!data);
+		if (!drainer)
+			drainer = data;
+		num_cpus++;
+		wait_for_completion(&data->wakeparent);
+	}
+
+	/* signal the drainer to start draining */
+	complete(&drainer->wakethread);
+	wait_for_completion(&drainer->wakeparent);
+	init_completion(&drainer->wakeparent);
+
+	/* tear down */
+	list_for_each_entry_safe(data, drainer, &threads, node) {
+		complete(&data->wakethread);
+		ret = kthread_stop(data->t);
+		BUG_ON(ret);
+		list_del(&data->node);
+		/* check that we get the expected callbacks (and no others) */
+		BUG_ON(data->num_enter != 1);
+		BUG_ON(data->num_exit != 0);
+		kfree(data);
+	}
+	bman_free_pool(pool_nocb);
+
+	pr_info("bman_test_thresh: done\n");
+}
+
diff --git a/drivers/staging/fsl_qbman/dpa_sys.h b/drivers/staging/fsl_qbman/dpa_sys.h
new file mode 100644
index 0000000..25c48ef
--- /dev/null
+++ b/drivers/staging/fsl_qbman/dpa_sys.h
@@ -0,0 +1,324 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef DPA_SYS_H
+#define DPA_SYS_H
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/io.h>
+#include <linux/dma-mapping.h>
+#include <linux/bootmem.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/delay.h>
+#include <linux/of_platform.h>
+#include <linux/kthread.h>
+#include <linux/memblock.h>
+#include <linux/completion.h>
+#include <linux/log2.h>
+#include <linux/types.h>
+#include <linux/ioctl.h>
+#include <linux/miscdevice.h>
+#include <linux/uaccess.h>
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
+#include <linux/device.h>
+#include <linux/uio_driver.h>
+#include <asm/smp.h>
+#include <sysdev/fsl_soc.h>
+#include <linux/fsl_hypervisor.h>
+#include <linux/vmalloc.h>
+
+/* When copying aligned words or shorts, try to avoid memcpy() */
+#define CONFIG_TRY_BETTER_MEMCPY
+
+/* This takes a "phandle" and dereferences to the cpu device-tree node,
+ * returning the cpu index. Returns negative error codes. */
+static inline int check_cpu_phandle(phandle ph)
+{
+	const u32 *cpu_val;
+	struct device_node *tmp_node = of_find_node_by_phandle(ph);
+	int cpu, ret;
+
+	if (!tmp_node) {
+		pr_err("Bad 'cpu-handle'\n");
+		return -EINVAL;
+	}
+	cpu_val = of_get_property(tmp_node, "reg", &ret);
+	if (!cpu_val || (ret != sizeof(*cpu_val))) {
+		pr_err("Can't get %s property 'reg'\n", tmp_node->full_name);
+		return -ENODEV;
+	}
+	for_each_present_cpu(cpu) {
+		if (*cpu_val == get_hard_smp_processor_id(cpu))
+			goto done;
+	}
+	pr_err("Invalid cpu index %d in %s\n", *cpu_val, tmp_node->full_name);
+	return -ENODEV;
+done:
+	of_node_put(tmp_node);
+	return cpu;
+}
+
+/* Handle portals destined for USDPAA (user-space).
+ *
+ * The UIO layer is mostly Qman/Bman-agnostic, however the rest of the driver is
+ * separated along interface and implementation lines. So each Qman/Bman driver
+ * instantiates a dpa_uio_class and links dpa_uio_portal objects into it. If
+ * the dpa_uio driver module is built, it queries these two classes and creates
+ * the UIO devices accordingly.
+ */
+struct dpa_uio_class {
+	struct list_head list;
+	const char *dev_prefix;
+};
+struct dpa_uio_portal {
+	enum {
+		dpa_uio_portal_bman,
+		dpa_uio_portal_qman,
+	} type;
+	union {
+		const struct bm_portal_config *bm_cfg;
+		const struct qm_portal_config *qm_cfg;
+	};
+	struct list_head node;
+};
+const struct dpa_uio_class *dpa_uio_bman(void);
+const struct dpa_uio_class *dpa_uio_qman(void);
+
+/* These stubs are re-mapped to hypervisor+failover features in kernel trees
+ * that contain that support. */
+static inline int fsl_dpa_should_recover(void)
+{
+	return 0;
+}
+static inline int pamu_enable_liodn(struct device_node *n, int i)
+{
+	return 0;
+}
+/***********************/
+/* Misc inline assists */
+/***********************/
+
+/* TODO: NB, we currently assume that hwsync() and lwsync() imply compiler
+ * barriers and that dcb*() won't fall victim to compiler or execution
+ * reordering with respect to other code/instructions that manipulate the same
+ * cacheline. */
+#define hwsync() \
+	do { \
+		__asm__ __volatile__ ("sync" : : : "memory"); \
+	} while(0)
+#define lwsync() \
+	do { \
+		__asm__ __volatile__ (stringify_in_c(LWSYNC) : : : "memory"); \
+	} while(0)
+#define dcbf(p) \
+	do { \
+		__asm__ __volatile__ ("dcbf 0,%0" : : "r" (p) : "memory"); \
+	} while(0)
+#define dcbt_ro(p) \
+	do { \
+		__asm__ __volatile__ ("dcbt 0,%0" : : "r" (p)); \
+	} while(0)
+#define dcbt_rw(p) \
+	do { \
+		__asm__ __volatile__ ("dcbtst 0,%0" : : "r" (p)); \
+	} while(0)
+#define dcbi(p) dcbf(p)
+#ifdef CONFIG_PPC_E500MC
+#define dcbzl(p) \
+	do { \
+		__asm__ __volatile__ ("dcbzl 0,%0" : : "r" (p)); \
+	} while (0)
+#define dcbz_64(p) \
+	do { \
+		dcbzl(p); \
+	} while (0)
+#define dcbf_64(p) \
+	do { \
+		dcbf(p); \
+	} while (0)
+/* Commonly used combo */
+#define dcbit_ro(p) \
+	do { \
+		dcbi(p); \
+		dcbt_ro(p); \
+	} while (0)
+#else
+#define dcbz(p) \
+	do { \
+		__asm__ __volatile__ ("dcbz 0,%0" : : "r" (p)); \
+	} while (0)
+#define dcbz_64(p) \
+	do { \
+		dcbz((u32)p + 32);	\
+		dcbz(p);	\
+	} while (0)
+#define dcbf_64(p) \
+	do { \
+		dcbf((u32)p + 32); \
+		dcbf(p); \
+	} while (0)
+/* Commonly used combo */
+#define dcbit_ro(p) \
+	do { \
+		dcbi(p); \
+		dcbi((u32)p + 32); \
+		dcbt_ro(p); \
+		dcbt_ro((u32)p + 32); \
+	} while (0)
+#endif /* CONFIG_PPC_E500MC */
+
+static inline u64 mfatb(void)
+{
+	u32 hi, lo, chk;
+	do {
+		hi = mfspr(SPRN_ATBU);
+		lo = mfspr(SPRN_ATBL);
+		chk = mfspr(SPRN_ATBU);
+	} while (unlikely(hi != chk));
+	return ((u64)hi << 32) | (u64)lo;
+}
+
+#ifdef CONFIG_FSL_DPA_CHECKING
+#define DPA_ASSERT(x) \
+	do { \
+		if (!(x)) { \
+			pr_crit("ASSERT: (%s:%d) %s\n", __FILE__, __LINE__, \
+				__stringify_1(x)); \
+			dump_stack(); \
+			panic("assertion failure"); \
+		} \
+	} while(0)
+#else
+#define DPA_ASSERT(x)
+#endif
+
+/* memcpy() stuff - when you know alignments in advance */
+#ifdef CONFIG_TRY_BETTER_MEMCPY
+static inline void copy_words(void *dest, const void *src, size_t sz)
+{
+	u32 *__dest = dest;
+	const u32 *__src = src;
+	size_t __sz = sz >> 2;
+	BUG_ON((unsigned long)dest & 0x3);
+	BUG_ON((unsigned long)src & 0x3);
+	BUG_ON(sz & 0x3);
+	while (__sz--)
+		*(__dest++) = *(__src++);
+}
+static inline void copy_shorts(void *dest, const void *src, size_t sz)
+{
+	u16 *__dest = dest;
+	const u16 *__src = src;
+	size_t __sz = sz >> 1;
+	BUG_ON((unsigned long)dest & 0x1);
+	BUG_ON((unsigned long)src & 0x1);
+	BUG_ON(sz & 0x1);
+	while (__sz--)
+		*(__dest++) = *(__src++);
+}
+static inline void copy_bytes(void *dest, const void *src, size_t sz)
+{
+	u8 *__dest = dest;
+	const u8 *__src = src;
+	while (sz--)
+		*(__dest++) = *(__src++);
+}
+#else
+#define copy_words memcpy
+#define copy_shorts memcpy
+#define copy_bytes memcpy
+#endif
+
+/************/
+/* RB-trees */
+/************/
+
+/* We encapsulate RB-trees so that its easier to use non-linux forms in
+ * non-linux systems. This also encapsulates the extra plumbing that linux code
+ * usually provides when using RB-trees. This encapsulation assumes that the
+ * data type held by the tree is u32. */
+
+struct dpa_rbtree {
+	struct rb_root root;
+};
+#define DPA_RBTREE { .root = RB_ROOT }
+
+static inline void dpa_rbtree_init(struct dpa_rbtree *tree)
+{
+	tree->root = RB_ROOT;
+}
+
+#define IMPLEMENT_DPA_RBTREE(name, type, node_field, val_field) \
+static inline int name##_push(struct dpa_rbtree *tree, type *obj) \
+{ \
+	struct rb_node *parent = NULL, **p = &tree->root.rb_node; \
+	while (*p) { \
+		u32 item; \
+		parent = *p; \
+		item = rb_entry(parent, type, node_field)->val_field; \
+		if (obj->val_field < item) \
+			p = &parent->rb_left; \
+		else if (obj->val_field > item) \
+			p = &parent->rb_right; \
+		else \
+			return -EBUSY; \
+	} \
+	rb_link_node(&obj->node_field, parent, p); \
+	rb_insert_color(&obj->node_field, &tree->root); \
+	return 0; \
+} \
+static inline void name##_del(struct dpa_rbtree *tree, type *obj) \
+{ \
+	rb_erase(&obj->node_field, &tree->root); \
+} \
+static inline type *name##_find(struct dpa_rbtree *tree, u32 val) \
+{ \
+	type *ret; \
+	struct rb_node *p = tree->root.rb_node; \
+	while (p) { \
+		ret = rb_entry(p, type, node_field); \
+		if (val < ret->val_field) \
+			p = p->rb_left; \
+		else if (val > ret->val_field) \
+			p = p->rb_right; \
+		else \
+			return ret; \
+	} \
+	return NULL; \
+}
+
+#endif /* DPA_SYS_H */
+
diff --git a/drivers/staging/fsl_qbman/dpa_uio.c b/drivers/staging/fsl_qbman/dpa_uio.c
new file mode 100644
index 0000000..a5a6345
--- /dev/null
+++ b/drivers/staging/fsl_qbman/dpa_uio.c
@@ -0,0 +1,231 @@ 
+/* Copyright 2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "bman_private.h"
+#include "qman_private.h"
+
+static const char dpa_uio_version[] = "USDPAA UIO portal driver v0.2";
+
+static LIST_HEAD(uio_portal_list);
+
+struct dpa_uio_info {
+	atomic_t ref; /* exclusive, only one open() at a time */
+	struct uio_info uio;
+	void *addr_ci;
+	char name[16]; /* big enough for "qman-uio-xx" */
+	struct platform_device *pdev;
+	struct list_head node;
+};
+
+static int dpa_uio_open(struct uio_info *info, struct inode *inode)
+{
+	struct dpa_uio_info *i = container_of(info, struct dpa_uio_info, uio);
+	if (!atomic_dec_and_test(&i->ref)) {
+		atomic_inc(&i->ref);
+		return -EBUSY;
+	}
+	return 0;
+}
+
+static int dpa_uio_release(struct uio_info *info, struct inode *inode)
+{
+	struct dpa_uio_info *i = container_of(info, struct dpa_uio_info, uio);
+	atomic_inc(&i->ref);
+	return 0;
+}
+
+static int dpa_uio_mmap(struct uio_info *info, struct vm_area_struct *vma)
+{
+	struct uio_mem *mem;
+	struct dpa_uio_info *i = container_of(info, struct dpa_uio_info, uio);
+
+	if (vma->vm_pgoff == 0) {
+		/* CENA */
+		mem = &i->uio.mem[0];
+		vma->vm_page_prot &=
+			~(_PAGE_GUARDED | _PAGE_NO_CACHE | _PAGE_COHERENT);
+	} else if (vma->vm_pgoff == 1) {
+		/* CINH */
+		mem = &i->uio.mem[1];
+		vma->vm_page_prot |= _PAGE_GUARDED | _PAGE_NO_CACHE;
+	} else {
+		pr_err("%s: unknown mmap offset %d, rejecting\n",
+			i->name, (int)vma->vm_pgoff);
+		return -EINVAL;
+	}
+	if ((vma->vm_end - vma->vm_start) != mem->size) {
+		pr_err("%s: invalid mmap() size %d, expect %d\n",
+			i->name, (int)(vma->vm_end - vma->vm_start),
+			(int)mem->size);
+		return -EINVAL;
+	}
+	/* FIXME: UIO appears not to support sizeof(phys_addr_t) > sizeof(void*)
+	 * as mem->addr is 32-bit. Also, it would have been more natural (and in
+	 * keeping with UIO's design intent) to have used the UIO_MEM_PHYS type
+	 * for our two memory regions, and to rely on UIO's own mmap() handler
+	 * (by not declaring our own). Unfortunately UIO does not allow any
+	 * specification of pgprots and assumes cache-inhibited mappings for
+	 * anything physical (see drivers/uio/uio.c, eg. uio_mmap_physical()).
+	 * So UIO could use a couple of improvements as it is not saving us much
+	 * on the kernel nor the user side. The first would be to use PFN
+	 * instead of a raw base address in the uio_mem structs (same reason as
+	 * everywhere else, this covers 4096 times as much address space, and
+	 * why waste lower bits given it has to be page-aligned anyway?). The
+	 * second is to add a pgprot field to uio_mem to be used with _PHYS
+	 * mappings. (Or use a new _PHYS_PGPROT type, for backwards
+	 * compatibility?) */
+	/* Normally, we'd ">>PAGE_SHIFT" the mem->addr value here, but due to
+	 * the 36-bit issue, it is already stored as a PFN. */
+	return io_remap_pfn_range(vma, vma->vm_start, mem->addr, mem->size,
+				vma->vm_page_prot);
+}
+
+static irqreturn_t dpa_uio_irq_handler(int irq, struct uio_info *info)
+{
+	struct dpa_uio_info *i = container_of(info, struct dpa_uio_info, uio);
+	/* This is the only code outside the regular portal driver that
+	 * manipulates any portal register, so rather than breaking that
+	 * encapsulation I am simply hard-coding the offset to the inhibit
+	 * register here. */
+	out_be32(i->addr_ci + 0xe0c, ~(u32)0);
+	return IRQ_HANDLED;
+}
+
+static void __init dpa_uio_portal_init(struct dpa_uio_portal *p,
+				const struct dpa_uio_class *c)
+{
+	struct dpa_uio_info *info;
+	const struct resource *res;
+	u32 index;
+	int irq, ret;
+
+	/* allocate 'info' */
+	info = kzalloc(sizeof(*info), GFP_KERNEL);
+	if (!info)
+		return;
+	atomic_set(&info->ref, 1);
+	if (p->type == dpa_uio_portal_bman) {
+		res = &p->bm_cfg->addr_phys[0];
+		index = p->bm_cfg->public_cfg.index;
+		irq = p->bm_cfg->public_cfg.irq;
+	} else {
+		res = &p->qm_cfg->addr_phys[0];
+		index = p->qm_cfg->public_cfg.index;
+		irq = p->qm_cfg->public_cfg.irq;
+	}
+	/* We need to map the cache-inhibited region in the kernel for
+	 * interrupt-handling purposes. */
+	info->addr_ci = ioremap_prot(res[BM_ADDR_CI].start,
+				resource_size(&res[BM_ADDR_CI]),
+				_PAGE_GUARDED | _PAGE_NO_CACHE);
+	/* Name the UIO device according to the cell-index. It's supposed to be
+	 * unique for each device class (Qman/Bman), and is also a convenient
+	 * way for user-space to find the UIO device that corresponds to a given
+	 * portal device-tree node. */
+	sprintf(info->name, "%s%x", c->dev_prefix, index);
+	info->pdev = platform_device_alloc(info->name, -1);
+	if (!info->pdev) {
+		iounmap(info->addr_ci);
+		kfree(info);
+		pr_err("dpa_uio_portal: platform_device_alloc() failed\n");
+		return;
+	}
+	ret = platform_device_add(info->pdev);
+	if (ret) {
+		platform_device_put(info->pdev);
+		iounmap(info->addr_ci);
+		kfree(info);
+		pr_err("dpa_uio_portal: platform_device_add() failed\n");
+		return;
+	}
+	info->uio.name = info->name;
+	info->uio.version = dpa_uio_version;
+	/* Work around the 36-bit UIO issue by bit-shifting the addresses */
+	info->uio.mem[BM_ADDR_CE].name = "cena";
+	info->uio.mem[BM_ADDR_CE].addr = res[BM_ADDR_CE].start >> PAGE_SHIFT;
+	info->uio.mem[BM_ADDR_CE].size = resource_size(&res[BM_ADDR_CE]);
+	info->uio.mem[BM_ADDR_CI].name = "cinh";
+	info->uio.mem[BM_ADDR_CI].addr = res[BM_ADDR_CI].start >> PAGE_SHIFT;
+	info->uio.mem[BM_ADDR_CI].size = resource_size(&res[BM_ADDR_CI]);
+	info->uio.irq = irq;
+	info->uio.handler = dpa_uio_irq_handler;
+	info->uio.mmap = dpa_uio_mmap;
+	info->uio.open = dpa_uio_open;
+	info->uio.release = dpa_uio_release;
+	ret = uio_register_device(&info->pdev->dev, &info->uio);
+	if (ret) {
+		platform_device_del(info->pdev);
+		platform_device_put(info->pdev);
+		iounmap(info->addr_ci);
+		kfree(info);
+		pr_err("dpa_uio_portal: UIO registration failed\n");
+		return;
+	}
+	list_add_tail(&info->node, &uio_portal_list);
+	pr_info("USDPAA portal initialised, %s\n", info->name);
+}
+
+static int __init dpa_uio_init(void)
+{
+	const struct dpa_uio_class *classes[3], **c = classes;
+	classes[0] = dpa_uio_bman();
+	classes[1] = dpa_uio_qman();
+	classes[2] = NULL;
+	while (*c) {
+		struct dpa_uio_portal *p;
+		list_for_each_entry(p, &(*c)->list, node)
+			dpa_uio_portal_init(p, *c);
+		c++;
+	}
+	pr_info("USDPAA portal layer loaded\n");
+	return 0;
+}
+
+static void __exit dpa_uio_exit(void)
+{
+	struct dpa_uio_info *info, *tmp;
+	list_for_each_entry_safe(info, tmp, &uio_portal_list, node) {
+		list_del(&info->node);
+		uio_unregister_device(&info->uio);
+		platform_device_del(info->pdev);
+		platform_device_put(info->pdev);
+		iounmap(info->addr_ci);
+		pr_info("USDPAA portal removed, %s\n", info->name);
+		kfree(info);
+	}
+	pr_info("USDPAA portal layer unloaded\n");
+}
+
+
+module_init(dpa_uio_init)
+module_exit(dpa_uio_exit)
+MODULE_LICENSE("GPL");
+
diff --git a/drivers/staging/fsl_qbman/qman_config.c b/drivers/staging/fsl_qbman/qman_config.c
new file mode 100644
index 0000000..8b9e3f0
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_config.c
@@ -0,0 +1,1010 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef CONFIG_SMP
+#include <linux/smp.h>	/* get_hard_smp_processor_id() */
+#endif
+
+#include "qman_private.h"
+
+/* Last updated for v00.800 of the BG */
+
+/* Register offsets */
+#define REG_QCSP_PID_CFG(n)	(0x0000 + ((n) * 0x10))
+#define REG_QCSP_IO_CFG(n)	(0x0004 + ((n) * 0x10))
+#define REG_QCSP_DD_CFG(n)	(0x000c + ((n) * 0x10))
+#define REG_DD_CFG		0x0200
+#define REG_DCP_CFG(n)		(0x0300 + ((n) * 0x10))
+#define REG_DCP_DD_CFG(n)	(0x0304 + ((n) * 0x10))
+#define REG_DCP_DLM_AVG(n)	(0x030c + ((n) * 0x10))
+#define REG_PFDR_FPC		0x0400
+#define REG_PFDR_FP_HEAD	0x0404
+#define REG_PFDR_FP_TAIL	0x0408
+#define REG_PFDR_FP_LWIT	0x0410
+#define REG_PFDR_CFG		0x0414
+#define REG_SFDR_CFG		0x0500
+#define REG_SFDR_IN_USE		0x0504
+#define REG_WQ_CS_CFG(n)	(0x0600 + ((n) * 0x04))
+#define REG_WQ_DEF_ENC_WQID	0x0630
+#define REG_WQ_SC_DD_CFG(n)	(0x640 + ((n) * 0x04))
+#define REG_WQ_PC_DD_CFG(n)	(0x680 + ((n) * 0x04))
+#define REG_WQ_DC0_DD_CFG(n)	(0x6c0 + ((n) * 0x04))
+#define REG_WQ_DC1_DD_CFG(n)	(0x700 + ((n) * 0x04))
+#define REG_WQ_DCn_DD_CFG(n)	(0x6c0 + ((n) * 0x40)) /* n=2,3 */
+#define REG_CM_CFG		0x0800
+#define REG_ECSR		0x0a00
+#define REG_ECIR		0x0a04
+#define REG_EADR		0x0a08
+#define REG_EDATA(n)		(0x0a10 + ((n) * 0x04))
+#define REG_SBEC(n)		(0x0a80 + ((n) * 0x04))
+#define REG_MCR			0x0b00
+#define REG_MCP(n)		(0x0b04 + ((n) * 0x04))
+#define REG_HID_CFG		0x0bf0
+#define REG_IDLE_STAT		0x0bf4
+#define REG_IP_REV_1		0x0bf8
+#define REG_IP_REV_2		0x0bfc
+#define REG_FQD_BARE		0x0c00
+#define REG_PFDR_BARE		0x0c20
+#define REG_offset_BAR		0x0004	/* relative to REG_[FQD|PFDR]_BARE */
+#define REG_offset_AR		0x0010	/* relative to REG_[FQD|PFDR]_BARE */
+#define REG_QCSP_BARE		0x0c80
+#define REG_QCSP_BAR		0x0c84
+#define REG_CI_SCHED_CFG	0x0d00
+#define REG_SRCIDR		0x0d04
+#define REG_LIODNR		0x0d08
+#define REG_CI_RLM_AVG		0x0d14
+#define REG_ERR_ISR		0x0e00	/* + "enum qm_isr_reg" */
+
+/* Assists for QMAN_MCR */
+#define MCR_INIT_PFDR		0x01000000
+#define MCR_get_rslt(v)		(u8)((v) >> 24)
+#define MCR_rslt_idle(r)	(!rslt || (rslt >= 0xf0))
+#define MCR_rslt_ok(r)		(rslt == 0xf0)
+#define MCR_rslt_eaccess(r)	(rslt == 0xf8)
+#define MCR_rslt_inval(r)	(rslt == 0xff)
+
+struct qman;
+
+/* Follows WQ_CS_CFG0-5 */
+enum qm_wq_class {
+	qm_wq_portal = 0,
+	qm_wq_pool = 1,
+	qm_wq_fman0 = 2,
+	qm_wq_fman1 = 3,
+	qm_wq_caam = 4,
+	qm_wq_pme = 5,
+	qm_wq_first = qm_wq_portal,
+	qm_wq_last = qm_wq_pme
+};
+
+/* Follows FQD_[BARE|BAR|AR] and PFDR_[BARE|BAR|AR] */
+enum qm_memory {
+	qm_memory_fqd,
+	qm_memory_pfdr
+};
+
+/* Used by all error interrupt registers except 'inhibit' */
+#define QM_EIRQ_CIDE	0x20000000	/* Corenet Initiator Data Error */
+#define QM_EIRQ_CTDE	0x10000000	/* Corenet Target Data Error */
+#define QM_EIRQ_CITT	0x08000000	/* Corenet Invalid Target Transaction */
+#define QM_EIRQ_PLWI	0x04000000	/* PFDR Low Watermark */
+#define QM_EIRQ_MBEI	0x02000000	/* Multi-bit ECC Error */
+#define QM_EIRQ_SBEI	0x01000000	/* Single-bit ECC Error */
+#define QM_EIRQ_PEBI	0x00800000	/* PFDR Enqueues Blocked Interrupt */
+#define QM_EIRQ_ICVI	0x00010000	/* Invalid Command Verb */
+#define QM_EIRQ_IDDI	0x00000800	/* Invalid Dequeue (Direct-connect) */
+#define QM_EIRQ_IDFI	0x00000400	/* Invalid Dequeue FQ */
+#define QM_EIRQ_IDSI	0x00000200	/* Invalid Dequeue Source */
+#define QM_EIRQ_IDQI	0x00000100	/* Invalid Dequeue Queue */
+#define QM_EIRQ_IEOI	0x00000008	/* Invalid Enqueue Overflow */
+#define QM_EIRQ_IESI	0x00000004	/* Invalid Enqueue State */
+#define QM_EIRQ_IECI	0x00000002	/* Invalid Enqueue Channel */
+#define QM_EIRQ_IEQI	0x00000001	/* Invalid Enqueue Queue */
+
+/* QMAN_ECIR valid error bit */
+#define PORTAL_ECSR_ERR	(QM_EIRQ_IEQI | QM_EIRQ_IESI | QM_EIRQ_IEOI | \
+				QM_EIRQ_IDQI | QM_EIRQ_IDSI | QM_EIRQ_IDFI | \
+				QM_EIRQ_IDDI | QM_EIRQ_ICVI)
+#define FQID_ECSR_ERR	(QM_EIRQ_IEQI | QM_EIRQ_IECI | QM_EIRQ_IESI | \
+			QM_EIRQ_IEOI | QM_EIRQ_IDQI | QM_EIRQ_IDFI)
+
+union qman_ecir {
+	u32 ecir_raw;
+	struct {
+		u32 __reserved:2;
+		u32 portal_type:1;
+		u32 portal_num:5;
+		u32 fqid:24;
+	} __packed info;
+};
+
+union qman_eadr {
+	u32 eadr_raw;
+	struct {
+		u32 __reserved1:4;
+		u32 memid:4;
+		u32 __reserved2:12;
+		u32 eadr:12;
+	} __packed info;
+};
+
+struct qman_hwerr_txt {
+	u32 mask;
+	const char *txt;
+};
+
+#define QMAN_HWE_TXT(a, b) { .mask = QM_EIRQ_##a, .txt = b }
+
+static const struct qman_hwerr_txt qman_hwerr_txts[] = {
+	QMAN_HWE_TXT(CIDE, "Corenet Initiator Data Error"),
+	QMAN_HWE_TXT(CTDE, "Corenet Target Data Error"),
+	QMAN_HWE_TXT(CITT, "Corenet Invalid Target Transaction"),
+	QMAN_HWE_TXT(PLWI, "PFDR Low Watermark"),
+	QMAN_HWE_TXT(MBEI, "Multi-bit ECC Error"),
+	QMAN_HWE_TXT(SBEI, "Single-bit ECC Error"),
+	QMAN_HWE_TXT(PEBI, "PFDR Enqueues Blocked Interrupt"),
+	QMAN_HWE_TXT(ICVI, "Invalid Command Verb"),
+	QMAN_HWE_TXT(IDDI, "Invalid Dequeue (Direct-connect)"),
+	QMAN_HWE_TXT(IDFI, "Invalid Dequeue FQ"),
+	QMAN_HWE_TXT(IDSI, "Invalid Dequeue Source"),
+	QMAN_HWE_TXT(IDQI, "Invalid Dequeue Queue"),
+	QMAN_HWE_TXT(IEOI, "Invalid Enqueue Overflow"),
+	QMAN_HWE_TXT(IESI, "Invalid Enqueue State"),
+	QMAN_HWE_TXT(IECI, "Invalid Enqueue Channel"),
+	QMAN_HWE_TXT(IEQI, "Invalid Enqueue Queue")
+};
+#define QMAN_HWE_COUNT (sizeof(qman_hwerr_txts)/sizeof(struct qman_hwerr_txt))
+
+struct qman_error_info_mdata {
+	u16 addr_mask;
+	u16 bits;
+	const char *txt;
+};
+
+#define QMAN_ERR_MDATA(a, b, c) { .addr_mask = a, .bits = b, .txt = c}
+static const struct qman_error_info_mdata error_mdata[] = {
+	QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 0"),
+	QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 1"),
+	QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 2"),
+	QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 3"),
+	QMAN_ERR_MDATA(0x0FFF, 512, "FQD cache memory"),
+	QMAN_ERR_MDATA(0x07FF, 128, "SFDR memory"),
+	QMAN_ERR_MDATA(0x01FF, 72, "WQ context memory"),
+	QMAN_ERR_MDATA(0x00FF, 240, "CGR memory"),
+	QMAN_ERR_MDATA(0x00FF, 302, "Internal Order Restoration List memory"),
+	QMAN_ERR_MDATA(0x01FF, 256, "SW portal ring memory"),
+};
+#define QMAN_ERR_MDATA_COUNT \
+	(sizeof(error_mdata)/sizeof(struct qman_error_info_mdata))
+
+/* Add this in Kconfig */
+#define QMAN_ERRS_TO_UNENABLE (QM_EIRQ_PLWI | QM_EIRQ_PEBI)
+
+/**
+ * qm_err_isr_<reg>_<verb> - Manipulate global interrupt registers
+ * @v: for accessors that write values, this is the 32-bit value
+ *
+ * Manipulates QMAN_ERR_ISR, QMAN_ERR_IER, QMAN_ERR_ISDR, QMAN_ERR_IIR. All
+ * manipulations except qm_err_isr_[un]inhibit() use 32-bit masks composed of
+ * the QM_EIRQ_*** definitions. Note that "qm_err_isr_enable_write" means
+ * "write the enable register" rather than "enable the write register"!
+ */
+#define qm_err_isr_status_read(qm)	__qm_err_isr_read(qm, qm_isr_status)
+#define qm_err_isr_status_clear(qm, m)	__qm_err_isr_write(qm, qm_isr_status,m)
+#define qm_err_isr_enable_read(qm)	__qm_err_isr_read(qm, qm_isr_enable)
+#define qm_err_isr_enable_write(qm, v)	__qm_err_isr_write(qm, qm_isr_enable,v)
+#define qm_err_isr_disable_read(qm)	__qm_err_isr_read(qm, qm_isr_disable)
+#define qm_err_isr_disable_write(qm, v)	__qm_err_isr_write(qm, qm_isr_disable,v)
+#define qm_err_isr_inhibit(qm)		__qm_err_isr_write(qm, qm_isr_inhibit,1)
+#define qm_err_isr_uninhibit(qm)	__qm_err_isr_write(qm, qm_isr_inhibit,0)
+
+/*
+ * TODO: unimplemented registers
+ *
+ * Keeping a list here of Qman registers I have not yet covered;
+ * QCSP_DD_IHRSR, QCSP_DD_IHRFR, QCSP_DD_HASR,
+ * DCP_DD_IHRSR, DCP_DD_IHRFR, DCP_DD_HASR, CM_CFG,
+ * QMAN_EECC, QMAN_SBET, QMAN_EINJ, QMAN_SBEC0-12
+ */
+
+/* Encapsulate "struct qman *" as a cast of the register space address. */
+
+static struct qman *qm_create(void *regs)
+{
+	return (struct qman *)regs;
+}
+
+static inline u32 __qm_in(struct qman *qm, u32 offset)
+{
+	return in_be32((void *)qm + offset);
+}
+static inline void __qm_out(struct qman *qm, u32 offset, u32 val)
+{
+	out_be32((void *)qm + offset, val);
+}
+#define qm_in(reg)		__qm_in(qm, REG_##reg)
+#define qm_out(reg, val)	__qm_out(qm, REG_##reg, val)
+
+static u32 __qm_err_isr_read(struct qman *qm, enum qm_isr_reg n)
+{
+	return __qm_in(qm, REG_ERR_ISR + (n << 2));
+}
+
+static void __qm_err_isr_write(struct qman *qm, enum qm_isr_reg n, u32 val)
+{
+	__qm_out(qm, REG_ERR_ISR + (n << 2), val);
+}
+
+#if 0
+
+static void qm_set_portal(struct qman *qm, u8 swportalID,
+			u16 ec_tp_cfg, u16 ecd_tp_cfg)
+{
+	qm_out(QCSP_DD_CFG(swportalID),
+		((ec_tp_cfg & 0x1ff) << 16) | (ecd_tp_cfg & 0x1ff));
+}
+
+static void qm_set_ddebug(struct qman *qm, u8 mdd, u8 m_cfg)
+{
+	qm_out(DD_CFG, ((mdd & 0x3) << 4) | (m_cfg & 0xf));
+}
+
+static void qm_set_dc_ddebug(struct qman *qm, enum qm_dc_portal portal, u16 ecd_tp_cfg)
+{
+	qm_out(DCP_DD_CFG(portal), ecd_tp_cfg & 0x1ff);
+}
+
+static u32 qm_get_pfdr_free_pool_count(struct qman *qm)
+{
+	return qm_in(PFDR_FPC);
+}
+
+static void qm_get_pfdr_free_pool(struct qman *qm, u32 *head, u32 *tail)
+{
+	*head = qm_in(PFDR_FP_HEAD);
+	*tail = qm_in(PFDR_FP_TAIL);
+}
+
+static void qm_set_default_wq(struct qman *qm, u16 wqid)
+{
+	qm_out(WQ_DEF_ENC_WQID, wqid);
+}
+
+static void qm_set_channel_ddebug(struct qman *qm, enum qm_channel channel,
+				u16 tp_cfg)
+{
+	u32 offset;
+	int upperhalf = 0;
+	if ((channel >= qm_channel_swportal0) &&
+				(channel <= qm_channel_swportal9)) {
+		offset = (channel - qm_channel_swportal0);
+		upperhalf = offset & 0x1;
+		offset = REG_WQ_SC_DD_CFG(offset / 2);
+	} else if ((channel >= qm_channel_pool1) &&
+				(channel <= qm_channel_pool15)) {
+		offset = (channel + 1 - qm_channel_pool1);
+		upperhalf = offset & 0x1;
+		offset = REG_WQ_PC_DD_CFG(offset / 2);
+	} else if ((channel >= qm_channel_fman0_sp0) &&
+				(channel <= qm_channel_fman0_sp11)) {
+		offset = (channel - qm_channel_fman0_sp0);
+		upperhalf = offset & 0x1;
+		offset = REG_WQ_DC0_DD_CFG(offset / 2);
+	}
+	else if ((channel >= qm_channel_fman1_sp0) &&
+				(channel <= qm_channel_fman1_sp11)) {
+		offset = (channel - qm_channel_fman1_sp0);
+		upperhalf = offset & 0x1;
+		offset = REG_WQ_DC1_DD_CFG(offset / 2);
+	}
+	else if (channel == qm_channel_caam)
+		offset = REG_WQ_DCn_DD_CFG(2);
+	else if (channel == qm_channel_pme)
+		offset = REG_WQ_DCn_DD_CFG(3);
+	else {
+		pr_crit("Illegal qm_channel type %d\n", channel);
+		return;
+	}
+	__qm_out(qm, offset, upperhalf ? ((u32)tp_cfg << 16) : tp_cfg);
+}
+
+static void qm_get_details(struct qman *qm, u8 *int_options, u8 *errata,
+			u8 *conf_options)
+{
+	u32 v = qm_in(IP_REV_1);
+	*int_options = (v >> 16) & 0xff;
+	*errata = (v >> 8) & 0xff;
+	*conf_options = v & 0xff;
+}
+
+static void qm_set_corenet_bar(struct qman *qm, u16 eba, u32 ba)
+{
+	/* choke if 'ba' isn't properly aligned */
+	DPA_ASSERT(!(ba & 0x001fffff));
+	qm_out(QCSP_BARE, eba);
+	qm_out(QCSP_BAR, ba);
+}
+
+static u8 qm_get_corenet_sourceid(struct qman *qm)
+{
+	return qm_in(SRCIDR);
+}
+
+static u16 qm_get_liodn(struct qman *qm)
+{
+	return qm_in(LIODNR);
+}
+
+static void qm_set_congestion_config(struct qman *qm, u16 pres)
+{
+	qm_out(CM_CFG, pres);
+}
+
+#endif
+
+static void qm_set_dc(struct qman *qm, enum qm_dc_portal portal,
+			int ed, u8 sernd)
+{
+	DPA_ASSERT(!ed || (portal == qm_dc_portal_fman0) ||
+			(portal == qm_dc_portal_fman1));
+	qm_out(DCP_CFG(portal), (ed ? 0x100 : 0) | (sernd & 0x1f));
+}
+
+static void qm_set_wq_scheduling(struct qman *qm, enum qm_wq_class wq_class,
+			u8 cs_elev, u8 csw2, u8 csw3, u8 csw4, u8 csw5,
+			u8 csw6, u8 csw7)
+{
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+#define csw(x) \
+do { \
+	if (++x == 8) \
+		x = 7; \
+} while (0)
+	if (qman_ip_rev == QMAN_REV10) {
+		csw(csw2);
+		csw(csw3);
+		csw(csw4);
+		csw(csw5);
+		csw(csw6);
+		csw(csw7);
+	}
+#endif
+	qm_out(WQ_CS_CFG(wq_class), ((cs_elev & 0xff) << 24) |
+		((csw2 & 0x7) << 20) | ((csw3 & 0x7) << 16) |
+		((csw4 & 0x7) << 12) | ((csw5 & 0x7) << 8) |
+		((csw6 & 0x7) << 4) | (csw7 & 0x7));
+}
+
+static void qm_set_hid(struct qman *qm)
+{
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+	if (qman_ip_rev == QMAN_REV10)
+		qm_out(HID_CFG, 3);
+	else
+#endif
+	qm_out(HID_CFG, 0);
+}
+
+static void qm_set_corenet_initiator(struct qman *qm)
+{
+	qm_out(CI_SCHED_CFG,
+		0x80000000 | /* write srcciv enable */
+		(CONFIG_FSL_QMAN_CI_SCHED_CFG_SRCCIV << 24) |
+		(CONFIG_FSL_QMAN_CI_SCHED_CFG_SRQ_W << 8) |
+		(CONFIG_FSL_QMAN_CI_SCHED_CFG_RW_W << 4) |
+		CONFIG_FSL_QMAN_CI_SCHED_CFG_BMAN_W);
+}
+
+static void qm_get_version(struct qman *qm, u16 *id, u8 *major, u8 *minor)
+{
+	u32 v = qm_in(IP_REV_1);
+	*id = (v >> 16);
+	*major = (v >> 8) & 0xff;
+	*minor = v & 0xff;
+}
+
+static void qm_set_memory(struct qman *qm, enum qm_memory memory, u64 ba,
+			int enable, int prio, int stash, u32 size)
+{
+	u32 offset = (memory == qm_memory_fqd) ? REG_FQD_BARE : REG_PFDR_BARE;
+	u32 exp = ilog2(size);
+	/* choke if size isn't within range */
+	DPA_ASSERT((size >= 4096) && (size <= 1073741824) &&
+			is_power_of_2(size));
+	/* choke if 'ba' has lower-alignment than 'size' */
+	DPA_ASSERT(!(ba & (size - 1)));
+	__qm_out(qm, offset, upper_32_bits(ba));
+	__qm_out(qm, offset + REG_offset_BAR, lower_32_bits(ba));
+	__qm_out(qm, offset + REG_offset_AR,
+		(enable ? 0x80000000 : 0) |
+		(prio ? 0x40000000 : 0) |
+		(stash ? 0x20000000 : 0) |
+		(exp - 1));
+}
+
+static void qm_set_pfdr_threshold(struct qman *qm, u32 th, u8 k)
+{
+	qm_out(PFDR_FP_LWIT, th & 0xffffff);
+	qm_out(PFDR_CFG, k);
+}
+
+static void qm_set_sfdr_threshold(struct qman *qm, u16 th)
+{
+	qm_out(SFDR_CFG, th & 0x3ff);
+}
+
+static int qm_init_pfdr(struct qman *qm, u32 pfdr_start, u32 num)
+{
+	u8 rslt = MCR_get_rslt(qm_in(MCR));
+
+	DPA_ASSERT(pfdr_start && !(pfdr_start & 7) && !(num & 7) && num);
+	/* Make sure the command interface is 'idle' */
+	if(!MCR_rslt_idle(rslt))
+		panic("QMAN_MCR isn't idle");
+
+	/* Write the MCR command params then the verb */
+	qm_out(MCP(0), pfdr_start );
+	/* TODO: remove this - it's a workaround for a model bug that is
+	 * corrected in more recent versions. We use the workaround until
+	 * everyone has upgraded. */
+	qm_out(MCP(1), (pfdr_start + num - 16));
+	lwsync();
+	qm_out(MCR, MCR_INIT_PFDR);
+
+	/* Poll for the result */
+	do {
+		rslt = MCR_get_rslt(qm_in(MCR));
+	} while(!MCR_rslt_idle(rslt));
+	if (MCR_rslt_ok(rslt))
+		return 0;
+	if (MCR_rslt_eaccess(rslt))
+		return -EACCES;
+	if (MCR_rslt_inval(rslt))
+		return -EINVAL;
+	pr_crit("Unexpected result from MCR_INIT_PFDR: %02x\n", rslt);
+	return -ENOSYS;
+}
+
+/*****************/
+/* Config driver */
+/*****************/
+
+/* TODO: Kconfig these? */
+#define DEFAULT_FQD_SZ	(PAGE_SIZE << CONFIG_FSL_QMAN_FQD_SZ)
+#define DEFAULT_PFDR_SZ	(PAGE_SIZE << 12)
+
+/* We support only one of these */
+static struct qman *qm;
+static struct device_node *qm_node;
+
+/* Parse the <name> property to extract the memory location and size and
+ * memblock_reserve() it. If it isn't supplied, memblock_alloc() the default size. */
+static __init int parse_mem_property(struct device_node *node, const char *name,
+				dma_addr_t *addr, size_t *sz, int zero)
+{
+	const u32 *pint;
+	int ret;
+
+	pint = of_get_property(node, name, &ret);
+	if (!pint || (ret != 16)) {
+		pr_info("No %s property '%s', using memblock_alloc(%016zx)\n",
+				node->full_name, name, *sz);
+		*addr = memblock_alloc(*sz, *sz);
+		if (zero)
+			memset(phys_to_virt(*addr), 0, *sz);
+		return 0;
+	}
+	pr_info("Using %s property '%s'\n", node->full_name, name);
+	/* If using a "zero-pma", don't try to zero it, even if you asked */
+	if (zero && of_find_property(node, "zero-pma", &ret)) {
+		pr_info("  it's a 'zero-pma', not zeroing from s/w\n");
+		zero = 0;
+	}
+	*addr = ((u64)pint[0] << 32) | (u64)pint[1];
+	*sz = ((u64)pint[2] << 32) | (u64)pint[3];
+	/* Keep things simple, it's either all in the DRAM range or it's all
+	 * outside. */
+	if (*addr < memblock_end_of_DRAM()) {
+		BUG_ON((u64)*addr + (u64)*sz > memblock_end_of_DRAM());
+		if (memblock_reserve(*addr, *sz) < 0) {
+			pr_err("Failed to reserve %s\n", name);
+			return -ENOMEM;
+		}
+		if (zero)
+			memset(phys_to_virt(*addr), 0, *sz);
+	} else if (zero) {
+		/* map as cacheable, non-guarded */
+		void *tmpp = ioremap_prot(*addr, *sz, 0);
+		memset(tmpp, 0, *sz);
+		iounmap(tmpp);
+	}
+	return 0;
+}
+
+/* TODO:
+ * - there is obviously no handling of errors,
+ * - the calls to qm_set_memory() hard-code the priority and CPC-stashing for
+ *   both memory resources to zero.
+ */
+static int __init fsl_qman_init(struct device_node *node)
+{
+	struct resource res;
+	u32 __iomem *regs;
+	const char *s;
+	dma_addr_t fqd_a = 0, pfdr_a = 0;
+	size_t fqd_sz = DEFAULT_FQD_SZ, pfdr_sz = DEFAULT_PFDR_SZ;
+	int ret, standby = 0;
+	u16 id;
+	u8 major, minor;
+
+	ret = of_address_to_resource(node, 0, &res);
+	if (ret) {
+		pr_err("Can't get %s property '%s'\n", node->full_name, "reg");
+		return ret;
+	}
+	s = of_get_property(node, "fsl,hv-claimable", &ret);
+	if (s && !strcmp(s, "standby"))
+		standby = 1;
+	if (!standby) {
+		ret = parse_mem_property(node, "fsl,qman-fqd",
+					&fqd_a, &fqd_sz, 1);
+		BUG_ON(ret);
+		ret = parse_mem_property(node, "fsl,qman-pfdr",
+					&pfdr_a, &pfdr_sz, 0);
+		BUG_ON(ret);
+	}
+	/* Global configuration */
+	regs = ioremap(res.start, res.end - res.start + 1);
+	qm = qm_create(regs);
+	qm_node = node;
+	qm_get_version(qm, &id, &major, &minor);
+	pr_info("Qman ver:%04x,%02x,%02x\n", id, major, minor);
+	if (!qman_ip_rev) {
+		if ((major == 1) && (minor == 0))
+			qman_ip_rev = QMAN_REV10;
+		else if ((major == 1) && (minor == 1))
+			qman_ip_rev = QMAN_REV11;
+		else if	((major == 1) && (minor == 2))
+			qman_ip_rev = QMAN_REV12;
+		else if ((major == 2) && (minor == 0))
+			qman_ip_rev = QMAN_REV20;
+		else {
+			pr_warning("unknown Qman version, default to rev1.1\n");
+			qman_ip_rev = QMAN_REV11;
+		}
+	}
+
+	if (standby) {
+		pr_info("  -> in standby mode\n");
+		return 0;
+	}
+	/* FQD memory */
+	qm_set_memory(qm, qm_memory_fqd, fqd_a, 1, 0, 0, fqd_sz);
+	/* PFDR memory */
+	qm_set_memory(qm, qm_memory_pfdr, pfdr_a, 1, 0, 0, pfdr_sz);
+	qm_init_pfdr(qm, 8, pfdr_sz / 64 - 8);
+	/* thresholds */
+	qm_set_pfdr_threshold(qm, 512, 64);
+	qm_set_sfdr_threshold(qm, 128);
+	/* clear stale PEBI bit from interrupt status register */
+	qm_err_isr_status_clear(qm, QM_EIRQ_PEBI);
+	/* corenet initiator settings */
+	qm_set_corenet_initiator(qm);
+	/* HID settings */
+	qm_set_hid(qm);
+	/* Set scheduling weights to defaults */
+	for (ret = qm_wq_first; ret <= qm_wq_last; ret++)
+		qm_set_wq_scheduling(qm, ret, 0, 0, 0, 0, 0, 0, 0);
+	/* We are not prepared to accept ERNs for hardware enqueues */
+	qm_set_dc(qm, qm_dc_portal_fman0, 1, 0);
+	qm_set_dc(qm, qm_dc_portal_fman1, 1, 0);
+	return 0;
+}
+
+int qman_have_ccsr(void)
+{
+	return qm ? 1 : 0;
+}
+
+__init void qman_init_early(void)
+{
+	struct device_node *dn;
+	for_each_compatible_node(dn, NULL, "fsl,qman") {
+		if (qm)
+			pr_err("%s: only one 'fsl,qman' allowed\n",
+				dn->full_name);
+		else {
+			int ret = fsl_qman_init(dn);
+			BUG_ON(ret);
+		}
+	}
+}
+
+static void log_edata_bits(u32 bit_count)
+{
+	u32 i, j, mask = 0xffffffff;
+
+	pr_warning("Qman ErrInt, EDATA:\n");
+	i = bit_count/32;
+	if (bit_count%32) {
+		i++;
+		mask = ~(mask << bit_count%32);
+	}
+	j = 16-i;
+	pr_warning("  0x%08x\n", qm_in(EDATA(j)) & mask);
+	j++;
+	for (; j < 16; j++)
+		pr_warning("  0x%08x\n", qm_in(EDATA(j)));
+}
+
+static void log_additional_error_info(u32 isr_val, u32 ecsr_val)
+{
+	union qman_ecir ecir_val;
+	union qman_eadr eadr_val;
+
+	ecir_val.ecir_raw = qm_in(ECIR);
+	/* Is portal info valid */
+	if (ecsr_val & PORTAL_ECSR_ERR) {
+		pr_warning("Qman ErrInt: %s id %d\n",
+			(ecir_val.info.portal_type) ?
+			"DCP" : "SWP", ecir_val.info.portal_num);
+	}
+	if (ecsr_val & FQID_ECSR_ERR) {
+		pr_warning("Qman ErrInt: ecir.fqid 0x%x\n",
+			ecir_val.info.fqid);
+	}
+	if (ecsr_val & (QM_EIRQ_SBEI|QM_EIRQ_MBEI)) {
+		eadr_val.eadr_raw = qm_in(EADR);
+		pr_warning("Qman ErrInt: EADR Memory: %s, 0x%x\n",
+			error_mdata[eadr_val.info.memid].txt,
+			error_mdata[eadr_val.info.memid].addr_mask
+				& eadr_val.info.eadr);
+		log_edata_bits(error_mdata[eadr_val.info.memid].bits);
+	}
+}
+
+/* Qman interrupt handler */
+static irqreturn_t qman_isr(int irq, void *ptr)
+{
+	u32 isr_val, ier_val, ecsr_val, isr_mask, i;
+
+	ier_val = qm_err_isr_enable_read(qm);
+	isr_val = qm_err_isr_status_read(qm);
+	ecsr_val = qm_in(ECSR);
+	isr_mask = isr_val & ier_val;
+
+	if (!isr_mask)
+		return IRQ_NONE;
+	for (i = 0; i < QMAN_HWE_COUNT; i++) {
+		if (qman_hwerr_txts[i].mask & isr_mask) {
+			pr_warning("Qman ErrInt: %s\n", qman_hwerr_txts[i].txt);
+			if (qman_hwerr_txts[i].mask & ecsr_val) {
+				log_additional_error_info(isr_mask, ecsr_val);
+				/* Re-arm error capture registers */
+				qm_out(ECSR, ecsr_val);
+			}
+			if (qman_hwerr_txts[i].mask & QMAN_ERRS_TO_UNENABLE) {
+				pr_devel("Qman un-enabling error 0x%x\n",
+					qman_hwerr_txts[i].mask);
+				ier_val &= ~qman_hwerr_txts[i].mask;
+				qm_err_isr_enable_write(qm, ier_val);
+			}
+		}
+	}
+	qm_err_isr_status_clear(qm, isr_val);
+	return IRQ_HANDLED;
+}
+
+static int __bind_irq(void)
+{
+	int ret, err_irq;
+
+	err_irq = of_irq_to_resource(qm_node, 0, NULL);
+	if (err_irq == NO_IRQ) {
+		pr_info("Can't get %s property '%s'\n", qm_node->full_name,
+			"interrupts");
+		return -ENODEV;
+	}
+	ret = request_irq(err_irq, qman_isr, IRQF_SHARED, "qman-err", qm_node);
+	if (ret)  {
+		pr_err("request_irq() failed %d for '%s'\n", ret,
+			qm_node->full_name);
+		return -ENODEV;
+	}
+	/* Write-to-clear any stale bits, (eg. starvation being asserted prior
+	 * to resource allocation during driver init). */
+	qm_err_isr_status_clear(qm, 0xffffffff);
+	/* Enable Error Interrupts */
+	qm_err_isr_enable_write(qm, 0xffffffff);
+	return 0;
+}
+
+/* Initialise Error Interrupt Handler */
+int qman_init_error_int(struct device_node *node)
+{
+	if (!qman_have_ccsr())
+		return 0;
+	if (node != qm_node)
+		return -EINVAL;
+	return __bind_irq();
+}
+
+#define PID_CFG_LIODN_MASK 0x0fff0000
+void qman_liodn_fixup(enum qm_channel channel)
+{
+	static int done;
+	static u32 liodn_offset;
+	u32 before, after;
+	int idx = channel - qm_channel_swportal0;
+
+	if (!qman_have_ccsr())
+		return;
+	before = qm_in(QCSP_PID_CFG(idx));
+	if (!done) {
+		liodn_offset = before & PID_CFG_LIODN_MASK;
+		done = 1;
+		return;
+	}
+	after = (before & (~PID_CFG_LIODN_MASK)) | liodn_offset;
+	qm_out(QCSP_PID_CFG(idx), after);
+}
+
+#ifdef CONFIG_SYSFS
+
+#define DRV_NAME	"fsl-qman"
+
+static ssize_t show_pfdr_fpc(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(PFDR_FPC));
+};
+
+static ssize_t show_dlm_avg(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	u32 data;
+	int i;
+
+	if (!sscanf(dev_attr->attr.name, "dcp%d_dlm_avg", &i))
+		return -EINVAL;
+	data = qm_in(DCP_DLM_AVG(i));
+	return snprintf(buf, PAGE_SIZE, "%d.%08d\n", data>>8,
+			(data & 0x000000ff)*390625);
+};
+
+static ssize_t set_dlm_avg(struct device *dev,
+	struct device_attribute *dev_attr, const char *buf, size_t count)
+{
+	unsigned long val;
+	int i;
+
+	if (!sscanf(dev_attr->attr.name, "dcp%d_dlm_avg", &i))
+		return -EINVAL;
+	if (strict_strtoul(buf, 0, &val)) {
+		dev_dbg(dev, "invalid input %s\n", buf);
+		return -EINVAL;
+	}
+	qm_out(DCP_DLM_AVG(i), val);
+	return count;
+};
+
+static ssize_t show_pfdr_cfg(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(PFDR_CFG));
+};
+
+static ssize_t set_pfdr_cfg(struct device *dev,
+	struct device_attribute *dev_attr, const char *buf, size_t count)
+{
+	unsigned long val;
+
+	if (strict_strtoul(buf, 0, &val)) {
+		dev_dbg(dev, "invalid input %s\n", buf);
+		return -EINVAL;
+	}
+	qm_out(PFDR_CFG, val);
+	return count;
+};
+
+static ssize_t show_sfdr_in_use(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(SFDR_IN_USE));
+};
+
+static ssize_t show_idle_stat(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(IDLE_STAT));
+};
+
+static ssize_t show_ci_rlm_avg(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	u32 data = qm_in(CI_RLM_AVG);
+	return snprintf(buf, PAGE_SIZE, "%d.%08d\n", data>>8,
+			(data & 0x000000ff)*390625);
+};
+
+static ssize_t set_ci_rlm_avg(struct device *dev,
+	struct device_attribute *dev_attr, const char *buf, size_t count)
+{
+	unsigned long val;
+
+	if (strict_strtoul(buf, 0, &val)) {
+		dev_dbg(dev, "invalid input %s\n", buf);
+		return -EINVAL;
+	}
+	qm_out(CI_RLM_AVG, val);
+	return count;
+};
+
+static ssize_t show_err_isr(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "0x%08x\n", qm_in(ERR_ISR));
+};
+
+
+static ssize_t show_sbec(struct device *dev,
+	struct device_attribute *dev_attr, char *buf)
+{
+	int i;
+
+	if (!sscanf(dev_attr->attr.name, "sbec_%d", &i))
+		return -EINVAL;
+	return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(SBEC(i)));
+};
+
+static DEVICE_ATTR(pfdr_fpc, S_IRUSR, show_pfdr_fpc, NULL);
+static DEVICE_ATTR(pfdr_cfg, S_IRUSR, show_pfdr_cfg, set_pfdr_cfg);
+static DEVICE_ATTR(idle_stat, S_IRUSR, show_idle_stat, NULL);
+static DEVICE_ATTR(ci_rlm_avg, (S_IRUSR|S_IWUGO),
+		show_ci_rlm_avg, set_ci_rlm_avg);
+static DEVICE_ATTR(err_isr, S_IRUSR, show_err_isr, NULL);
+static DEVICE_ATTR(sfdr_in_use, S_IRUSR, show_sfdr_in_use, NULL);
+
+static DEVICE_ATTR(dcp0_dlm_avg, (S_IRUSR|S_IWUGO), show_dlm_avg, set_dlm_avg);
+static DEVICE_ATTR(dcp1_dlm_avg, (S_IRUSR|S_IWUGO), show_dlm_avg, set_dlm_avg);
+static DEVICE_ATTR(dcp2_dlm_avg, (S_IRUSR|S_IWUGO), show_dlm_avg, set_dlm_avg);
+static DEVICE_ATTR(dcp3_dlm_avg, (S_IRUSR|S_IWUGO), show_dlm_avg, set_dlm_avg);
+
+static DEVICE_ATTR(sbec_0, S_IRUSR, show_sbec, NULL);
+static DEVICE_ATTR(sbec_1, S_IRUSR, show_sbec, NULL);
+static DEVICE_ATTR(sbec_2, S_IRUSR, show_sbec, NULL);
+static DEVICE_ATTR(sbec_3, S_IRUSR, show_sbec, NULL);
+static DEVICE_ATTR(sbec_4, S_IRUSR, show_sbec, NULL);
+static DEVICE_ATTR(sbec_5, S_IRUSR, show_sbec, NULL);
+static DEVICE_ATTR(sbec_6, S_IRUSR, show_sbec, NULL);
+
+
+static struct attribute *qman_dev_attributes[] = {
+	&dev_attr_pfdr_fpc.attr,
+	&dev_attr_pfdr_cfg.attr,
+	&dev_attr_idle_stat.attr,
+	&dev_attr_ci_rlm_avg.attr,
+	&dev_attr_err_isr.attr,
+	&dev_attr_dcp0_dlm_avg.attr,
+	&dev_attr_dcp1_dlm_avg.attr,
+	&dev_attr_dcp2_dlm_avg.attr,
+	&dev_attr_dcp3_dlm_avg.attr,
+	/* sfdr_in_use will be added if necessary */
+	NULL
+};
+
+static struct attribute *qman_dev_ecr_attributes[] = {
+	&dev_attr_sbec_0.attr,
+	&dev_attr_sbec_1.attr,
+	&dev_attr_sbec_2.attr,
+	&dev_attr_sbec_3.attr,
+	&dev_attr_sbec_4.attr,
+	&dev_attr_sbec_5.attr,
+	&dev_attr_sbec_6.attr,
+	NULL
+};
+
+/* root level */
+static const struct attribute_group qman_dev_attr_grp = {
+	.name = NULL,
+	.attrs = qman_dev_attributes
+};
+static const struct attribute_group qman_dev_ecr_grp = {
+	.name = "error_capture",
+	.attrs = qman_dev_ecr_attributes
+};
+
+static int of_fsl_qman_remove(struct platform_device *ofdev)
+{
+	sysfs_remove_group(&ofdev->dev.kobj, &qman_dev_attr_grp);
+	return 0;
+};
+
+static int __devinit of_fsl_qman_probe(struct platform_device *ofdev)
+{
+	int ret;
+
+	ret = sysfs_create_group(&ofdev->dev.kobj, &qman_dev_attr_grp);
+	if (ret)
+		goto done;
+	if (qman_ip_rev != QMAN_REV10) {
+		ret = sysfs_add_file_to_group(&ofdev->dev.kobj,
+			&dev_attr_sfdr_in_use.attr, qman_dev_attr_grp.name);
+		if (ret)
+			goto del_group_0;
+	}
+	ret = sysfs_create_group(&ofdev->dev.kobj, &qman_dev_ecr_grp);
+	if (ret)
+		goto del_group_0;
+
+	goto done;
+
+del_group_0:
+	sysfs_remove_group(&ofdev->dev.kobj, &qman_dev_attr_grp);
+done:
+	if (ret)
+		dev_err(&ofdev->dev,
+				"Cannot create dev attributes ret=%d\n", ret);
+	return ret;
+};
+
+static struct of_device_id of_fsl_qman_ids[] = {
+	{
+		.compatible = "fsl,qman",
+	},
+	{}
+};
+MODULE_DEVICE_TABLE(of, of_fsl_qman_ids);
+
+static struct platform_driver of_fsl_qman_driver = {
+	.driver = {
+		.owner = THIS_MODULE,
+		.name = DRV_NAME,
+		.of_match_table = of_fsl_qman_ids,
+	},
+	.probe = of_fsl_qman_probe,
+	.remove      = __devexit_p(of_fsl_qman_remove),
+};
+
+static int qman_ctrl_init(void)
+{
+	return platform_driver_register(&of_fsl_qman_driver);
+}
+
+static void qman_ctrl_exit(void)
+{
+	platform_driver_unregister(&of_fsl_qman_driver);
+}
+
+module_init(qman_ctrl_init);
+module_exit(qman_ctrl_exit);
+
+#endif /* CONFIG_SYSFS */
diff --git a/drivers/staging/fsl_qbman/qman_debugfs.c b/drivers/staging/fsl_qbman/qman_debugfs.c
new file mode 100644
index 0000000..67fa240
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_debugfs.c
@@ -0,0 +1,1311 @@ 
+/* Copyright 2010-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include "qman_private.h"
+
+#define MAX_FQID (0x00ffffff)
+#define QM_FQD_BLOCK_SIZE     64
+#define QM_FQD_AR             (0xC10)
+
+static u32 fqid_max;
+static u64 qman_ccsr_start;
+static u64 qman_ccsr_size;
+
+static const char *state_txt[] = {
+	"Out of Service",
+	"Retired",
+	"Tentatively Scheduled",
+	"Truly Scheduled",
+	"Parked",
+	"Active, Active Held or Held Suspended",
+	"Unknown State 6",
+	"Unknown State 7",
+	NULL,
+};
+
+static const u8 fqd_states[] = {
+	QM_MCR_NP_STATE_OOS, QM_MCR_NP_STATE_RETIRED, QM_MCR_NP_STATE_TEN_SCHED,
+	QM_MCR_NP_STATE_TRU_SCHED, QM_MCR_NP_STATE_PARKED,
+	QM_MCR_NP_STATE_ACTIVE};
+static const u32 fqd_states_count = sizeof(fqd_states)/sizeof(u8);
+
+struct mask_to_text {
+	u16 mask;
+	const char *txt;
+};
+
+struct mask_filter_s {
+	u16 mask;
+	u8 filter;
+};
+
+static const struct mask_filter_s mask_filter[] = {
+	{QM_FQCTRL_PREFERINCACHE, 0},
+	{QM_FQCTRL_PREFERINCACHE, 1},
+	{QM_FQCTRL_HOLDACTIVE, 0},
+	{QM_FQCTRL_HOLDACTIVE, 1},
+	{QM_FQCTRL_AVOIDBLOCK, 0},
+	{QM_FQCTRL_AVOIDBLOCK, 1},
+	{QM_FQCTRL_FORCESFDR, 0},
+	{QM_FQCTRL_FORCESFDR, 1},
+	{QM_FQCTRL_CPCSTASH, 0},
+	{QM_FQCTRL_CPCSTASH, 1},
+	{QM_FQCTRL_CTXASTASHING, 0},
+	{QM_FQCTRL_CTXASTASHING, 1},
+	{QM_FQCTRL_ORP, 0},
+	{QM_FQCTRL_ORP, 1},
+	{QM_FQCTRL_TDE, 0},
+	{QM_FQCTRL_TDE, 1},
+	{QM_FQCTRL_CGE, 0},
+	{QM_FQCTRL_CGE, 1}
+};
+static const u32 mask_filter_count =
+	sizeof(mask_filter)/sizeof(struct mask_filter_s);
+
+static const struct mask_to_text fq_ctrl_text_list[] = {
+	{
+		.mask = QM_FQCTRL_PREFERINCACHE,
+		.txt = "Prefer in cache",
+	},
+	{
+		.mask = QM_FQCTRL_HOLDACTIVE,
+		.txt =  "Hold active in portal",
+	},
+	{
+		.mask = QM_FQCTRL_AVOIDBLOCK,
+		.txt = "Avoid Blocking",
+	},
+	{
+		.mask = QM_FQCTRL_FORCESFDR,
+		.txt = "High-priority SFDRs",
+	},
+	{
+		.mask = QM_FQCTRL_CPCSTASH,
+		.txt = "CPC Stash Enable",
+	},
+	{
+		.mask = QM_FQCTRL_CTXASTASHING,
+		.txt =  "Context-A stashing",
+	},
+	{
+		.mask = QM_FQCTRL_ORP,
+		.txt =  "ORP Enable",
+	},
+	{
+		.mask = QM_FQCTRL_TDE,
+		.txt = "Tail-Drop Enable",
+	},
+	{
+		.mask = QM_FQCTRL_CGE,
+		.txt = "Congestion Group Enable",
+	},
+	{
+		.mask = 0,
+		.txt = NULL,
+	}
+};
+
+static const char *get_fqd_ctrl_text(u16 mask)
+{
+	int i = 0;
+
+	while (fq_ctrl_text_list[i].txt != NULL) {
+		if (fq_ctrl_text_list[i].mask == mask)
+			return fq_ctrl_text_list[i].txt;
+		i++;
+	}
+	return NULL;
+}
+
+static const struct mask_to_text stashing_text_list[] = {
+	{
+		.mask = QM_STASHING_EXCL_CTX,
+		.txt = "FQ Ctx Stash"
+	},
+	{
+		.mask = QM_STASHING_EXCL_DATA,
+		.txt =  "Frame Data Stash",
+	},
+	{
+		.mask = QM_STASHING_EXCL_ANNOTATION,
+		.txt = "Frame Annotation Stash",
+	},
+	{
+		.mask = 0,
+		.txt = NULL,
+	},
+};
+
+static int user_input_convert(const char __user *user_buf, size_t count,
+				unsigned long *val)
+{
+	char buf[12];
+
+	if (count > sizeof(buf) - 1)
+		return -EINVAL;
+	if (copy_from_user(buf, user_buf, count))
+		return -EFAULT;
+	buf[count] = '\0';
+	if (strict_strtoul(buf, 0, val))
+		return -EINVAL;
+	return 0;
+}
+
+struct line_buffer_fq {
+	u32 buf[8];
+	u32 buf_cnt;
+	int line_cnt;
+};
+
+static void add_to_line_buffer(struct line_buffer_fq *line_buf, u32 fqid,
+			struct seq_file *file)
+{
+	line_buf->buf[line_buf->buf_cnt] = fqid;
+	line_buf->buf_cnt++;
+	if (line_buf->buf_cnt == 8) {
+		/* Buffer is full, flush it */
+		if (line_buf->line_cnt != 0)
+			seq_printf(file, ",\n");
+		seq_printf(file, "0x%06x,0x%06x,0x%06x,0x%06x,0x%06x,"
+			"0x%06x,0x%06x,0x%06x",
+			line_buf->buf[0], line_buf->buf[1], line_buf->buf[2],
+			line_buf->buf[3], line_buf->buf[4], line_buf->buf[5],
+			line_buf->buf[6], line_buf->buf[7]);
+		line_buf->buf_cnt = 0;
+		line_buf->line_cnt++;
+	}
+}
+
+static void flush_line_buffer(struct line_buffer_fq *line_buf,
+				struct seq_file *file)
+{
+	if (line_buf->buf_cnt) {
+		int y = 0;
+		if (line_buf->line_cnt != 0)
+			seq_printf(file, ",\n");
+		while (y != line_buf->buf_cnt) {
+			if (y+1 == line_buf->buf_cnt)
+				seq_printf(file, "0x%06x", line_buf->buf[y]);
+			else
+				seq_printf(file, "0x%06x,", line_buf->buf[y]);
+			y++;
+		}
+		line_buf->line_cnt++;
+	}
+	if (line_buf->line_cnt)
+		seq_printf(file, "\n");
+}
+
+static struct dentry *dfs_root; /* debugfs root directory */
+
+/*******************************************************************************
+ *  Query Frame Queue Non Programmable Fields
+ ******************************************************************************/
+struct query_fq_np_fields_data_s {
+	u32 fqid;
+};
+static struct query_fq_np_fields_data_s query_fq_np_fields_data = {
+	.fqid = 1,
+};
+
+static int query_fq_np_fields_show(struct seq_file *file, void *offset)
+{
+	int ret;
+	struct qm_mcr_queryfq_np np;
+	struct qman_fq fq;
+
+	fq.fqid = query_fq_np_fields_data.fqid;
+	ret = qman_query_fq_np(&fq, &np);
+	if (ret)
+		return ret;
+	/* Print state */
+	seq_printf(file, "Query FQ Non Programmable Fields Result fqid 0x%x\n",
+			fq.fqid);
+	seq_printf(file, " force eligible pending: %s\n",
+		(np.state & QM_MCR_NP_STATE_FE) ? "yes" : "no");
+	seq_printf(file, " retirement pending: %s\n",
+		(np.state & QM_MCR_NP_STATE_R) ? "yes" : "no");
+	seq_printf(file, " state: %s\n",
+		state_txt[np.state & QM_MCR_NP_STATE_MASK]);
+	seq_printf(file, " fq_link: 0x%x\n", np.fqd_link);
+	seq_printf(file, " odp_seq: %u\n", np.odp_seq);
+	seq_printf(file, " orp_nesn: %u\n", np.orp_nesn);
+	seq_printf(file, " orp_ea_hseq: %u\n", np.orp_ea_hseq);
+	seq_printf(file, " orp_ea_tseq: %u\n", np.orp_ea_tseq);
+	seq_printf(file, " orp_ea_hptr: 0x%x\n", np.orp_ea_hptr);
+	seq_printf(file, " orp_ea_tptr: 0x%x\n", np.orp_ea_tptr);
+	seq_printf(file, " pfdr_hptr: 0x%x\n", np.pfdr_hptr);
+	seq_printf(file, " pfdr_tptr: 0x%x\n", np.pfdr_tptr);
+	seq_printf(file, " is: ics_surp contains a %s\n",
+		(np.is) ? "deficit" : "surplus");
+	seq_printf(file, " ics_surp: %u\n", np.ics_surp);
+	seq_printf(file, " byte_cnt: %u\n", np.byte_cnt);
+	seq_printf(file, " frm_cnt: %u\n", np.frm_cnt);
+	seq_printf(file, " ra1_sfdr: 0x%x\n", np.ra1_sfdr);
+	seq_printf(file, " ra2_sfdr: 0x%x\n", np.ra2_sfdr);
+	seq_printf(file, " od1_sfdr: 0x%x\n", np.od1_sfdr);
+	seq_printf(file, " od2_sfdr: 0x%x\n", np.od2_sfdr);
+	seq_printf(file, " od3_sfdr: 0x%x\n", np.od3_sfdr);
+	return 0;
+}
+
+static int query_fq_np_fields_open(struct inode *inode,
+					struct file *file)
+{
+	return single_open(file, query_fq_np_fields_show, NULL);
+}
+
+static ssize_t query_fq_np_fields_write(struct file *f,
+			const char __user *buf, size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	if (val > MAX_FQID)
+		return -EINVAL;
+	query_fq_np_fields_data.fqid = (u32)val;
+	return count;
+}
+
+static const struct file_operations query_fq_np_fields_fops = {
+	.owner          = THIS_MODULE,
+	.open		= query_fq_np_fields_open,
+	.read           = seq_read,
+	.write		= query_fq_np_fields_write,
+	.release	= single_release,
+};
+
+/*******************************************************************************
+ *  Frame Queue Programmable Fields
+ ******************************************************************************/
+struct query_fq_fields_data_s {
+	u32 fqid;
+};
+
+static struct query_fq_fields_data_s query_fq_fields_data = {
+	.fqid = 1,
+};
+
+static int query_fq_fields_show(struct seq_file *file, void *offset)
+{
+	int ret;
+	struct qm_fqd fqd;
+	struct qman_fq fq;
+	int i = 0;
+
+	memset(&fqd, 0, sizeof(struct qm_fqd));
+	fq.fqid = query_fq_fields_data.fqid;
+	ret = qman_query_fq(&fq, &fqd);
+	if (ret)
+		return ret;
+	seq_printf(file, "Query FQ Programmable Fields Result fqid 0x%x\n",
+			fq.fqid);
+	seq_printf(file, " orprws: %u\n", fqd.orprws);
+	seq_printf(file, " oa: %u\n", fqd.oa);
+	seq_printf(file, " olws: %u\n", fqd.olws);
+
+	seq_printf(file, " cgid: %u\n", fqd.cgid);
+
+	if ((fqd.fq_ctrl & QM_FQCTRL_MASK) == 0)
+		seq_printf(file, " fq_ctrl: None\n");
+	else {
+		i = 0;
+		seq_printf(file, " fq_ctrl:\n");
+		while (fq_ctrl_text_list[i].txt != NULL) {
+			if ((fqd.fq_ctrl & QM_FQCTRL_MASK) &
+					fq_ctrl_text_list[i].mask)
+				seq_printf(file, "  %s\n",
+					fq_ctrl_text_list[i].txt);
+			i++;
+		}
+	}
+	seq_printf(file, " dest_channel: %u\n", fqd.dest.channel);
+	seq_printf(file, " dest_wq: %u\n", fqd.dest.wq);
+	seq_printf(file, " ics_cred: %u\n", fqd.ics_cred);
+	seq_printf(file, " td_mant: %u\n", fqd.td.mant);
+	seq_printf(file, " td_exp: %u\n", fqd.td.exp);
+
+	seq_printf(file, " ctx_b: 0x%x\n", fqd.context_b);
+
+	seq_printf(file, " ctx_a: 0x%llx\n", qm_fqd_stashing_get64(&fqd));
+	/* Any stashing configured */
+	if ((fqd.context_a.stashing.exclusive & 0x7) == 0)
+		seq_printf(file, " ctx_a_stash_exclusive: None\n");
+	else {
+		seq_printf(file, " ctx_a_stash_exclusive:\n");
+		i = 0;
+		while (stashing_text_list[i].txt != NULL) {
+			if ((fqd.fq_ctrl & 0x7) & stashing_text_list[i].mask)
+				seq_printf(file, "  %s\n",
+					stashing_text_list[i].txt);
+			i++;
+		}
+	}
+	seq_printf(file, " ctx_a_stash_annotation_cl: %u\n",
+			fqd.context_a.stashing.annotation_cl);
+	seq_printf(file, " ctx_a_stash_data_cl: %u\n",
+			fqd.context_a.stashing.data_cl);
+	seq_printf(file, " ctx_a_stash_context_cl: %u\n",
+			fqd.context_a.stashing.context_cl);
+	return 0;
+}
+
+static int query_fq_fields_open(struct inode *inode,
+					struct file *file)
+{
+	return single_open(file, query_fq_fields_show, NULL);
+}
+
+static ssize_t query_fq_fields_write(struct file *f,
+			const char __user *buf, size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	if (val > MAX_FQID)
+		return -EINVAL;
+	query_fq_fields_data.fqid = (u32)val;
+	return count;
+}
+
+static const struct file_operations query_fq_fields_fops = {
+	.owner          = THIS_MODULE,
+	.open		= query_fq_fields_open,
+	.read           = seq_read,
+	.write		= query_fq_fields_write,
+	.release	= single_release,
+};
+
+/*******************************************************************************
+ * Query WQ lengths
+ ******************************************************************************/
+struct query_wq_lengths_data_s {
+	union {
+		u16 channel_wq; /* ignores wq (3 lsbits) */
+		struct {
+			u16 id:13; /* enum qm_channel */
+			u16 __reserved:3;
+		} __packed channel;
+	};
+};
+static struct query_wq_lengths_data_s query_wq_lengths_data;
+static int query_wq_lengths_show(struct seq_file *file, void *offset)
+{
+	int ret;
+	struct qm_mcr_querywq wq;
+	int i;
+
+	memset(&wq, 0, sizeof(struct qm_mcr_querywq));
+	wq.channel.id = query_wq_lengths_data.channel.id;
+	ret = qman_query_wq(0, &wq);
+	if (ret)
+		return ret;
+	seq_printf(file, "Query Result For Channel: 0x%x\n", wq.channel.id);
+	for (i = 0; i < 8; i++)
+		/* mask out upper 4 bits since they are not part of length */
+		seq_printf(file, " wq%d_len : %u\n", i, wq.wq_len[i] & 0x0fff);
+	return 0;
+}
+
+static int query_wq_lengths_open(struct inode *inode,
+					struct file *file)
+{
+	return single_open(file, query_wq_lengths_show, NULL);
+}
+
+static ssize_t query_wq_lengths_write(struct file *f,
+			const char __user *buf, size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	if (val > 0xfff8)
+		return -EINVAL;
+	query_wq_lengths_data.channel.id = (u16)val;
+	return count;
+}
+
+static const struct file_operations query_wq_lengths_fops = {
+	.owner          = THIS_MODULE,
+	.open		= query_wq_lengths_open,
+	.read           = seq_read,
+	.write		= query_wq_lengths_write,
+	.release	= single_release,
+};
+
+/*******************************************************************************
+ *  Query CGR
+ ******************************************************************************/
+struct query_cgr_s {
+	u8 cgid;
+};
+static struct query_cgr_s query_cgr_data;
+
+static int query_cgr_show(struct seq_file *file, void *offset)
+{
+	int ret;
+	struct qm_mcr_querycgr cgrd;
+	struct qman_cgr cgr;
+
+	memset(&cgr, 0, sizeof(struct qm_mcr_querycgr));
+	cgr.cgrid = query_cgr_data.cgid;
+	ret = qman_query_cgr(&cgr, &cgrd);
+	if (ret)
+		return ret;
+	seq_printf(file, "Query CGR id 0x%x\n", cgr.cgrid);
+	seq_printf(file, " wr_parm_g MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n",
+		cgrd.cgr.wr_parm_g.MA, cgrd.cgr.wr_parm_g.Mn,
+		cgrd.cgr.wr_parm_g.SA, cgrd.cgr.wr_parm_g.Sn,
+		cgrd.cgr.wr_parm_g.Pn);
+
+	seq_printf(file, " wr_parm_y MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n",
+		cgrd.cgr.wr_parm_y.MA, cgrd.cgr.wr_parm_y.Mn,
+		cgrd.cgr.wr_parm_y.SA, cgrd.cgr.wr_parm_y.Sn,
+		cgrd.cgr.wr_parm_y.Pn);
+
+	seq_printf(file, " wr_parm_r MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n",
+		cgrd.cgr.wr_parm_r.MA, cgrd.cgr.wr_parm_r.Mn,
+		cgrd.cgr.wr_parm_r.SA, cgrd.cgr.wr_parm_r.Sn,
+		cgrd.cgr.wr_parm_r.Pn);
+
+	seq_printf(file, " wr_en_g: %u, wr_en_y: %u, we_en_r: %u\n",
+		cgrd.cgr.wr_en_g, cgrd.cgr.wr_en_y, cgrd.cgr.wr_en_r);
+
+	seq_printf(file, " cscn_en: %u\n", cgrd.cgr.cscn_en);
+	seq_printf(file, " cscn_targ: %u\n", cgrd.cgr.cscn_targ);
+	seq_printf(file, " cstd_en: %u\n", cgrd.cgr.cstd_en);
+	seq_printf(file, " cs: %u\n", cgrd.cgr.cs);
+
+	seq_printf(file, " cs_thresh_TA: %u, cs_thresh_Tn: %u\n",
+		cgrd.cgr.cs_thres.TA, cgrd.cgr.cs_thres.Tn);
+
+	if (qman_ip_rev != QMAN_REV10) {
+		seq_printf(file, " mode: %s\n",
+			(cgrd.cgr.mode & QMAN_CGR_MODE_FRAME) ?
+			"frame count" : "byte count");
+	}
+	seq_printf(file, " i_bcnt: %llu\n", qm_mcr_querycgr_i_get64(&cgrd));
+	seq_printf(file, " a_bcnt: %llu\n", qm_mcr_querycgr_a_get64(&cgrd));
+
+	return 0;
+}
+
+static int query_cgr_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, query_cgr_show, NULL);
+}
+
+static ssize_t query_cgr_write(struct file *f, const char __user *buf,
+				size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	if (val > 0xff)
+		return -EINVAL;
+	query_cgr_data.cgid = (u8)val;
+	return count;
+}
+
+static const struct file_operations query_cgr_fops = {
+	.owner          = THIS_MODULE,
+	.open		= query_cgr_open,
+	.read           = seq_read,
+	.write		= query_cgr_write,
+	.release	= single_release,
+};
+
+/*******************************************************************************
+ *  Test Write CGR
+ ******************************************************************************/
+struct test_write_cgr_s {
+	u64 i_bcnt;
+	u8 cgid;
+};
+static struct test_write_cgr_s test_write_cgr_data;
+
+static int testwrite_cgr_show(struct seq_file *file, void *offset)
+{
+	int ret;
+	struct qm_mcr_cgrtestwrite result;
+	struct qman_cgr cgr;
+	u64 i_bcnt;
+
+	memset(&cgr, 0, sizeof(struct qman_cgr));
+	memset(&result, 0, sizeof(struct qm_mcr_cgrtestwrite));
+	cgr.cgrid = test_write_cgr_data.cgid;
+	i_bcnt = test_write_cgr_data.i_bcnt;
+	ret = qman_testwrite_cgr(&cgr, i_bcnt, &result);
+	if (ret)
+		return ret;
+	seq_printf(file, "CGR Test Write CGR id 0x%x\n", cgr.cgrid);
+	seq_printf(file, " wr_parm_g MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n",
+		result.cgr.wr_parm_g.MA, result.cgr.wr_parm_g.Mn,
+		result.cgr.wr_parm_g.SA, result.cgr.wr_parm_g.Sn,
+		result.cgr.wr_parm_g.Pn);
+	seq_printf(file, " wr_parm_y MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n",
+		result.cgr.wr_parm_y.MA, result.cgr.wr_parm_y.Mn,
+		result.cgr.wr_parm_y.SA, result.cgr.wr_parm_y.Sn,
+		result.cgr.wr_parm_y.Pn);
+	seq_printf(file, " wr_parm_r MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n",
+		result.cgr.wr_parm_r.MA, result.cgr.wr_parm_r.Mn,
+		result.cgr.wr_parm_r.SA, result.cgr.wr_parm_r.Sn,
+		result.cgr.wr_parm_r.Pn);
+	seq_printf(file, " wr_en_g: %u, wr_en_y: %u, we_en_r: %u\n",
+		result.cgr.wr_en_g, result.cgr.wr_en_y, result.cgr.wr_en_r);
+	seq_printf(file, " cscn_en: %u\n", result.cgr.cscn_en);
+	seq_printf(file, " cscn_targ: %u\n", result.cgr.cscn_targ);
+	seq_printf(file, " cstd_en: %u\n", result.cgr.cstd_en);
+	seq_printf(file, " cs: %u\n", result.cgr.cs);
+	seq_printf(file, " cs_thresh_TA: %u, cs_thresh_Tn: %u\n",
+		result.cgr.cs_thres.TA, result.cgr.cs_thres.Tn);
+
+	/* Add Mode for Si 2 */
+	if (qman_ip_rev != QMAN_REV10) {
+		seq_printf(file, " mode: %s\n",
+			(result.cgr.mode & QMAN_CGR_MODE_FRAME) ?
+			"frame count" : "byte count");
+	}
+
+	seq_printf(file, " i_bcnt: %llu\n",
+		qm_mcr_cgrtestwrite_i_get64(&result));
+	seq_printf(file, " a_bcnt: %llu\n",
+		qm_mcr_cgrtestwrite_a_get64(&result));
+	seq_printf(file, " wr_prob_g: %u\n", result.wr_prob_g);
+	seq_printf(file, " wr_prob_y: %u\n", result.wr_prob_y);
+	seq_printf(file, " wr_prob_r: %u\n", result.wr_prob_r);
+	return 0;
+}
+
+static int testwrite_cgr_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, testwrite_cgr_show, NULL);
+}
+
+static const struct file_operations testwrite_cgr_fops = {
+	.owner          = THIS_MODULE,
+	.open		= testwrite_cgr_open,
+	.read           = seq_read,
+	.release	= single_release,
+};
+
+
+static int testwrite_cgr_ibcnt_show(struct seq_file *file, void *offset)
+{
+	seq_printf(file, "i_bcnt: %llu\n", test_write_cgr_data.i_bcnt);
+	return 0;
+}
+static int testwrite_cgr_ibcnt_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, testwrite_cgr_ibcnt_show, NULL);
+}
+
+static ssize_t testwrite_cgr_ibcnt_write(struct file *f, const char __user *buf,
+				size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	test_write_cgr_data.i_bcnt = val;
+	return count;
+}
+
+static const struct file_operations teswrite_cgr_ibcnt_fops = {
+	.owner          = THIS_MODULE,
+	.open		= testwrite_cgr_ibcnt_open,
+	.read           = seq_read,
+	.write		= testwrite_cgr_ibcnt_write,
+	.release	= single_release,
+};
+
+static int testwrite_cgr_cgrid_show(struct seq_file *file, void *offset)
+{
+	seq_printf(file, "cgrid: %u\n", (u32)test_write_cgr_data.cgid);
+	return 0;
+}
+static int testwrite_cgr_cgrid_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, testwrite_cgr_cgrid_show, NULL);
+}
+
+static ssize_t testwrite_cgr_cgrid_write(struct file *f, const char __user *buf,
+				size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	if (val > 0xff)
+		return -EINVAL;
+	test_write_cgr_data.cgid = (u8)val;
+	return count;
+}
+
+static const struct file_operations teswrite_cgr_cgrid_fops = {
+	.owner          = THIS_MODULE,
+	.open		= testwrite_cgr_cgrid_open,
+	.read           = seq_read,
+	.write		= testwrite_cgr_cgrid_write,
+	.release	= single_release,
+};
+
+/*******************************************************************************
+ *  Query Congestion State
+ ******************************************************************************/
+static int query_congestion_show(struct seq_file *file, void *offset)
+{
+	int ret;
+	struct qm_mcr_querycongestion cs;
+	int i, j, in_cong = 0;
+	u32 mask;
+
+	memset(&cs, 0, sizeof(struct qm_mcr_querycongestion));
+	ret = qman_query_congestion(&cs);
+	if (ret)
+		return ret;
+	seq_printf(file, "Query Congestion Result\n");
+	for (i = 0; i < 8; i++) {
+		mask = 0x80000000;
+		for (j = 0; j < 32; j++) {
+			if (cs.state.__state[i] & mask) {
+				in_cong = 1;
+				seq_printf(file, " cg %u: %s\n", (i*32)+j,
+					"in congestion");
+			}
+			mask >>= 1;
+		}
+	}
+	if (!in_cong)
+		seq_printf(file, " All congestion groups not congested.\n");
+	return 0;
+}
+
+static int query_congestion_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, query_congestion_show, NULL);
+}
+
+static const struct file_operations query_congestion_fops = {
+	.owner          = THIS_MODULE,
+	.open		= query_congestion_open,
+	.read           = seq_read,
+	.release	= single_release,
+};
+
+/*******************************************************************************
+ *  QMan register
+ ******************************************************************************/
+struct qman_register_s {
+	u32 val;
+};
+static struct qman_register_s qman_register_data;
+
+static void init_ccsrmempeek(void)
+{
+	struct device_node *dn;
+	const u32 *regaddr_p;
+
+	dn = of_find_compatible_node(NULL, NULL, "fsl,qman");
+	if (!dn) {
+		pr_info("No fsl,qman node\n");
+		return;
+	}
+	regaddr_p = of_get_address(dn, 0, &qman_ccsr_size, NULL);
+	if (!regaddr_p) {
+		of_node_put(dn);
+		return;
+	}
+	qman_ccsr_start = of_translate_address(dn, regaddr_p);
+	of_node_put(dn);
+}
+/* This function provides access to QMan ccsr memory map */
+static int qman_ccsrmempeek(u32 *val, u32 offset)
+{
+	void __iomem *addr;
+	u64 phys_addr;
+
+	if (!qman_ccsr_start)
+		return -EINVAL;
+
+	if (offset > (qman_ccsr_size - sizeof(u32)))
+		return -EINVAL;
+
+	phys_addr = qman_ccsr_start + offset;
+	addr = ioremap(phys_addr, sizeof(u32));
+	if (!addr) {
+		pr_err("ccsrmempeek, ioremap failed\n");
+		return -EINVAL;
+	}
+	*val = in_be32(addr);
+	iounmap(addr);
+	return 0;
+}
+
+static int qman_ccsrmempeek_show(struct seq_file *file, void *offset)
+{
+	u32 b;
+
+	qman_ccsrmempeek(&b, qman_register_data.val);
+	seq_printf(file, "QMan register offset = 0x%x\n",
+		   qman_register_data.val);
+	seq_printf(file, "value = 0x%08x\n", b);
+
+	return 0;
+}
+
+static int qman_ccsrmempeek_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, qman_ccsrmempeek_show, NULL);
+}
+
+static ssize_t qman_ccsrmempeek_write(struct file *f, const char __user *buf,
+				size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	/* multiple of 4 */
+	if (val > (qman_ccsr_size - sizeof(u32))) {
+		pr_info("Input 0x%lx > 0x%llx\n",
+			val, (qman_ccsr_size - sizeof(u32)));
+		return -EINVAL;
+	}
+	if (val & 0x3) {
+		pr_info("Input 0x%lx not multiple of 4\n", val);
+		return -EINVAL;
+	}
+	qman_register_data.val = val;
+	return count;
+}
+
+static const struct file_operations qman_ccsrmempeek_fops = {
+	.owner          = THIS_MODULE,
+	.open		= qman_ccsrmempeek_open,
+	.read           = seq_read,
+	.write		= qman_ccsrmempeek_write,
+};
+
+/*******************************************************************************
+ *  QMan state
+ ******************************************************************************/
+static int qman_fqd_state_show(struct seq_file *file, void *offset)
+{
+	struct qm_mcr_queryfq_np np;
+	struct qman_fq fq;
+	struct line_buffer_fq line_buf;
+	int ret, i;
+	u8 *state = file->private;
+	u32 qm_fq_state_cnt[fqd_states_count];
+
+	memset(qm_fq_state_cnt, 0, sizeof(qm_fq_state_cnt));
+	memset(&line_buf, 0, sizeof(line_buf));
+
+	seq_printf(file, "List of fq ids in state: %s\n", state_txt[*state]);
+
+	for (i = 1; i < fqid_max; i++) {
+		fq.fqid = i;
+		ret = qman_query_fq_np(&fq, &np);
+		if (ret)
+			return ret;
+		if (*state == (np.state & QM_MCR_NP_STATE_MASK))
+			add_to_line_buffer(&line_buf, fq.fqid, file);
+		/* Keep a summary count of all states */
+		if ((np.state & QM_MCR_NP_STATE_MASK) < fqd_states_count)
+			qm_fq_state_cnt[(np.state & QM_MCR_NP_STATE_MASK)]++;
+	}
+	flush_line_buffer(&line_buf, file);
+
+	for (i = 0; i < fqd_states_count; i++) {
+		seq_printf(file, "%s count = %u\n", state_txt[i],
+			   qm_fq_state_cnt[i]);
+	}
+	return 0;
+}
+
+static int qman_fqd_state_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, qman_fqd_state_show, inode->i_private);
+}
+
+static const struct file_operations qman_fqd_state_fops =  {
+	.owner          = THIS_MODULE,
+	.open		= qman_fqd_state_open,
+	.read           = seq_read,
+};
+
+static int qman_fqd_ctrl_show(struct seq_file *file, void *offset)
+{
+	struct qm_fqd fqd;
+	struct qman_fq fq;
+	u32 fq_en_cnt = 0, fq_di_cnt = 0;
+	int ret, i;
+	struct mask_filter_s *data = file->private;
+	const char *ctrl_txt = get_fqd_ctrl_text(data->mask);
+	struct line_buffer_fq line_buf;
+
+	memset(&line_buf, 0, sizeof(line_buf));
+	seq_printf(file, "List of fq ids with: %s :%s\n",
+		ctrl_txt, (data->filter) ? "enabled" : "disabled");
+	for (i = 1; i < fqid_max; i++) {
+		fq.fqid = i;
+		memset(&fqd, 0, sizeof(struct qm_fqd));
+		ret = qman_query_fq(&fq, &fqd);
+		if (ret)
+			return ret;
+		if (data->filter) {
+			if (fqd.fq_ctrl & data->mask)
+				add_to_line_buffer(&line_buf, fq.fqid, file);
+		} else {
+			if (!(fqd.fq_ctrl & data->mask))
+				add_to_line_buffer(&line_buf, fq.fqid, file);
+		}
+		if (fqd.fq_ctrl & data->mask)
+			fq_en_cnt++;
+		else
+			fq_di_cnt++;
+	}
+	flush_line_buffer(&line_buf, file);
+
+	seq_printf(file, "Total FQD with: %s :  enabled = %u\n",
+		   ctrl_txt, fq_en_cnt);
+	seq_printf(file, "Total FQD with: %s : disabled = %u\n",
+		   ctrl_txt, fq_di_cnt);
+	return 0;
+}
+
+/*******************************************************************************
+ *  QMan ctrl CGE, TDE, ORP, CTX, CPC, SFDR, BLOCK, HOLD, CACHE
+ ******************************************************************************/
+static int qman_fqd_ctrl_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, qman_fqd_ctrl_show, inode->i_private);
+}
+
+static const struct file_operations qman_fqd_ctrl_fops =  {
+	.owner          = THIS_MODULE,
+	.open		= qman_fqd_ctrl_open,
+	.read           = seq_read,
+};
+
+/*******************************************************************************
+ *  QMan ctrl summary
+ ******************************************************************************/
+/*******************************************************************************
+ *  QMan summary state
+ ******************************************************************************/
+static int qman_fqd_non_prog_summary_show(struct seq_file *file, void *offset)
+{
+	struct qm_mcr_queryfq_np np;
+	struct qman_fq fq;
+	int ret, i;
+	u32 qm_fq_state_cnt[fqd_states_count];
+
+	memset(qm_fq_state_cnt, 0, sizeof(qm_fq_state_cnt));
+
+	for (i = 1; i < fqid_max; i++) {
+		fq.fqid = i;
+		ret = qman_query_fq_np(&fq, &np);
+		if (ret)
+			return ret;
+		/* Keep a summary count of all states */
+		if ((np.state & QM_MCR_NP_STATE_MASK) < fqd_states_count)
+			qm_fq_state_cnt[(np.state & QM_MCR_NP_STATE_MASK)]++;
+	}
+
+	for (i = 0; i < fqd_states_count; i++) {
+		seq_printf(file, "%s count = %u\n", state_txt[i],
+			   qm_fq_state_cnt[i]);
+	}
+	return 0;
+}
+
+static int qman_fqd_prog_summary_show(struct seq_file *file, void *offset)
+{
+	struct qm_fqd fqd;
+	struct qman_fq fq;
+	int ret, i , j;
+	u32 qm_prog_cnt[mask_filter_count/2];
+
+	memset(qm_prog_cnt, 0, sizeof(qm_prog_cnt));
+
+	for (i = 1; i < fqid_max; i++) {
+		memset(&fqd, 0, sizeof(struct qm_fqd));
+		fq.fqid = i;
+		ret = qman_query_fq(&fq, &fqd);
+		if (ret)
+			return ret;
+		/* Keep a summary count of all states */
+		for (j = 0; j < mask_filter_count; j += 2)
+			if ((fqd.fq_ctrl & QM_FQCTRL_MASK) &
+					mask_filter[j].mask)
+				qm_prog_cnt[j/2]++;
+	}
+	for (i = 0; i < mask_filter_count/2; i++) {
+		seq_printf(file, "%s count = %u\n",
+			get_fqd_ctrl_text(mask_filter[i*2].mask),
+			   qm_prog_cnt[i]);
+	}
+	return 0;
+}
+
+static int qman_fqd_summary_show(struct seq_file *file, void *offset)
+{
+	int ret;
+
+	/* Display summary of non programmable fields */
+	ret = qman_fqd_non_prog_summary_show(file, offset);
+	if (ret)
+		return ret;
+	seq_printf(file, "-----------------------------------------\n");
+	/* Display programmable fields */
+	ret = qman_fqd_prog_summary_show(file, offset);
+	if (ret)
+		return ret;
+	return 0;
+}
+
+static int qman_fqd_summary_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, qman_fqd_summary_show, NULL);
+}
+
+static const struct file_operations qman_fqd_summary_fops =  {
+	.owner          = THIS_MODULE,
+	.open		= qman_fqd_summary_open,
+	.read           = seq_read,
+};
+
+/*******************************************************************************
+ *  QMan destination work queue
+ ******************************************************************************/
+struct qman_dest_wq_s {
+	u16 wq_id;
+};
+static struct qman_dest_wq_s qman_dest_wq_data = {
+	.wq_id = 0,
+};
+
+static int qman_fqd_dest_wq_show(struct seq_file *file, void *offset)
+{
+	struct qm_fqd fqd;
+	struct qman_fq fq;
+	int ret, i;
+	u16 *wq, wq_id = qman_dest_wq_data.wq_id;
+	struct line_buffer_fq line_buf;
+
+	memset(&line_buf, 0, sizeof(line_buf));
+	/* use vmalloc : need to allocate large memory region and don't
+	 * require the memory to be physically contiguous. */
+	wq = vmalloc(sizeof(u16) * (0xFFFF+1));
+	if (!wq)
+		return -ENOMEM;
+	memset(wq, 0, sizeof(u16) * (0xFFFF+1));
+
+	seq_printf(file, "List of fq ids with destination work queue id"
+			" = 0x%x\n", wq_id);
+
+	for (i = 1; i < fqid_max; i++) {
+		fq.fqid = i;
+		memset(&fqd, 0, sizeof(struct qm_fqd));
+		ret = qman_query_fq(&fq, &fqd);
+		if (ret) {
+			vfree(wq);
+			return ret;
+		}
+		if (wq_id == fqd.dest_wq)
+			add_to_line_buffer(&line_buf, fq.fqid, file);
+		wq[fqd.dest_wq]++;
+	}
+	flush_line_buffer(&line_buf, file);
+
+	seq_printf(file, "Summary of all FQD destination work queue values\n");
+	for (i = 0; i < 0xFFFF; i++) {
+		if (wq[i])
+			seq_printf(file, "Channel: 0x%x WQ: 0x%x WQ_ID: 0x%x, "
+				"count = %u\n", i >> 3, i & 0x3, i, wq[i]);
+	}
+	vfree(wq);
+	return 0;
+}
+
+static ssize_t qman_fqd_dest_wq_write(struct file *f, const char __user *buf,
+				      size_t count, loff_t *off)
+{
+	int ret;
+	unsigned long val;
+
+	ret = user_input_convert(buf, count, &val);
+	if (ret)
+		return ret;
+	if (val > 0xFFFF)
+		return -EINVAL;
+	qman_dest_wq_data.wq_id = val;
+	return count;
+}
+
+static int qman_fqd_dest_wq_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, qman_fqd_dest_wq_show, NULL);
+}
+
+static const struct file_operations qman_fqd_dest_wq_fops =  {
+	.owner          = THIS_MODULE,
+	.open		= qman_fqd_dest_wq_open,
+	.read           = seq_read,
+	.write		= qman_fqd_dest_wq_write,
+};
+
+/*******************************************************************************
+ *  QMan Intra-Class Scheduling Credit
+ ******************************************************************************/
+static int qman_fqd_cred_show(struct seq_file *file, void *offset)
+{
+	struct qm_fqd fqd;
+	struct qman_fq fq;
+	int ret, i;
+	u32 fq_cnt = 0;
+	struct line_buffer_fq line_buf;
+
+	memset(&line_buf, 0, sizeof(line_buf));
+	seq_printf(file, "List of fq ids with Intra-Class Scheduling Credit > 0"
+			"\n");
+
+	for (i = 1; i < fqid_max; i++) {
+		fq.fqid = i;
+		memset(&fqd, 0, sizeof(struct qm_fqd));
+		ret = qman_query_fq(&fq, &fqd);
+		if (ret)
+			return ret;
+		if (fqd.ics_cred > 0) {
+			add_to_line_buffer(&line_buf, fq.fqid, file);
+			fq_cnt++;
+		}
+	}
+	flush_line_buffer(&line_buf, file);
+
+	seq_printf(file, "Total FQD with ics_cred > 0 = %d\n", fq_cnt);
+	return 0;
+}
+
+static int qman_fqd_cred_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, qman_fqd_cred_show, NULL);
+}
+
+static const struct file_operations qman_fqd_cred_fops =  {
+	.owner          = THIS_MODULE,
+	.open		= qman_fqd_cred_open,
+	.read           = seq_read,
+};
+
+/* helper macros used in qman_debugfs_module_init */
+#define QMAN_DBGFS_ENTRY(name, mode, parent, data, fops) \
+	do { \
+		d = debugfs_create_file(name, \
+			mode, parent, \
+			data, \
+			fops); \
+		if (d == NULL) { \
+			ret = -ENOMEM; \
+			goto _return; \
+		} \
+	} while (0)
+
+/* dfs_root as parent */
+#define QMAN_DBGFS_ENTRY_ROOT(name, mode, data, fops) \
+	QMAN_DBGFS_ENTRY(name, mode, dfs_root, data, fops)
+
+/* fqd_root as parent */
+#define QMAN_DBGFS_ENTRY_FQDROOT(name, mode, data, fops) \
+	QMAN_DBGFS_ENTRY(name, mode, fqd_root, data, fops)
+
+/* fqd state */
+#define QMAN_DBGFS_ENTRY_FQDSTATE(name, index) \
+	QMAN_DBGFS_ENTRY_FQDROOT(name, S_IRUGO, \
+	(void *)&mask_filter[index], &qman_fqd_ctrl_fops)
+
+static int __init qman_debugfs_module_init(void)
+{
+	int ret = 0;
+	struct dentry *d, *fqd_root;
+	u32 reg;
+
+	fqid_max = 0;
+	init_ccsrmempeek();
+	if (qman_ccsr_start) {
+		if (!qman_ccsrmempeek(&reg, QM_FQD_AR)) {
+			/* extract the size of the FQD window */
+			reg = reg & 0x3f;
+			/* calculate valid frame queue descriptor range */
+			fqid_max = (1 << (reg + 1)) / QM_FQD_BLOCK_SIZE;
+		}
+	}
+	dfs_root = debugfs_create_dir("qman", NULL);
+	fqd_root = debugfs_create_dir("fqd", dfs_root);
+	if (dfs_root == NULL || fqd_root == NULL) {
+		ret = -ENOMEM;
+		pr_err("Cannot create qman/fqd debugfs dir\n");
+		goto _return;
+	}
+	if (fqid_max) {
+		QMAN_DBGFS_ENTRY_ROOT("ccsrmempeek", S_IRUGO | S_IWUGO,
+				NULL, &qman_ccsrmempeek_fops);
+	}
+	QMAN_DBGFS_ENTRY_ROOT("query_fq_np_fields", S_IRUGO | S_IWUGO,
+		&query_fq_np_fields_data, &query_fq_np_fields_fops);
+
+	QMAN_DBGFS_ENTRY_ROOT("query_fq_fields", S_IRUGO | S_IWUGO,
+		&query_fq_fields_data, &query_fq_fields_fops);
+
+	QMAN_DBGFS_ENTRY_ROOT("query_wq_lengths", S_IRUGO | S_IWUGO,
+		&query_wq_lengths_data, &query_wq_lengths_fops);
+
+	QMAN_DBGFS_ENTRY_ROOT("query_cgr", S_IRUGO | S_IWUGO,
+		&query_cgr_data, &query_cgr_fops);
+
+	QMAN_DBGFS_ENTRY_ROOT("query_congestion", S_IRUGO,
+		NULL, &query_congestion_fops);
+
+	QMAN_DBGFS_ENTRY_ROOT("testwrite_cgr", S_IRUGO,
+		NULL, &testwrite_cgr_fops);
+
+	QMAN_DBGFS_ENTRY_ROOT("testwrite_cgr_cgrid", S_IRUGO | S_IWUGO,
+		NULL, &teswrite_cgr_cgrid_fops);
+
+	QMAN_DBGFS_ENTRY_ROOT("testwrite_cgr_ibcnt", S_IRUGO | S_IWUGO,
+		NULL, &teswrite_cgr_ibcnt_fops);
+
+	/* Create files with fqd_root as parent */
+
+	QMAN_DBGFS_ENTRY_FQDROOT("stateoos", S_IRUGO,
+		(void *)&fqd_states[QM_MCR_NP_STATE_OOS], &qman_fqd_state_fops);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("state_retired", S_IRUGO,
+		(void *)&fqd_states[QM_MCR_NP_STATE_RETIRED],
+		&qman_fqd_state_fops);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("state_tentatively_sched", S_IRUGO,
+		(void *)&fqd_states[QM_MCR_NP_STATE_TEN_SCHED],
+		&qman_fqd_state_fops);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("state_truly_sched", S_IRUGO,
+		(void *)&fqd_states[QM_MCR_NP_STATE_TRU_SCHED],
+		&qman_fqd_state_fops);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("state_parked", S_IRUGO,
+		(void *)&fqd_states[QM_MCR_NP_STATE_PARKED],
+		&qman_fqd_state_fops);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("state_active", S_IRUGO,
+		(void *)&fqd_states[QM_MCR_NP_STATE_ACTIVE],
+		&qman_fqd_state_fops);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("cge_enable", 17);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("cge_disable", 16);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("tde_enable", 15);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("tde_disable", 14);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("orp_enable", 13);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("orp_disable", 12);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("ctx_a_stashing_enable", 11);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("ctx_a_stashing_disable", 10);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("cpc_enable", 9);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("cpc_disable", 8);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("sfdr_enable", 7);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("sfdr_disable", 6);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("avoid_blocking_enable", 5);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("avoid_blocking_disable", 4);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("hold_active_enable", 3);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("hold_active_disable", 2);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("prefer_in_cache_enable", 1);
+
+	QMAN_DBGFS_ENTRY_FQDSTATE("prefer_in_cache_disable", 0);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("summary", S_IRUGO,
+		NULL, &qman_fqd_summary_fops);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("wq", S_IRUGO | S_IWUGO,
+		NULL, &qman_fqd_dest_wq_fops);
+
+	QMAN_DBGFS_ENTRY_FQDROOT("cred", S_IRUGO,
+		NULL, &qman_fqd_cred_fops);
+
+	return 0;
+
+_return:
+	if (dfs_root)
+		debugfs_remove_recursive(dfs_root);
+	return ret;
+}
+
+static void __exit qman_debugfs_module_exit(void)
+{
+	debugfs_remove_recursive(dfs_root);
+}
+
+module_init(qman_debugfs_module_init);
+module_exit(qman_debugfs_module_exit);
+MODULE_LICENSE("Dual BSD/GPL");
+
diff --git a/drivers/staging/fsl_qbman/qman_driver.c b/drivers/staging/fsl_qbman/qman_driver.c
new file mode 100644
index 0000000..abce51b
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_driver.c
@@ -0,0 +1,541 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_private.h"
+
+/* Global variable containing revision id (even on non-control plane systems
+ * where CCSR isn't available) */
+u16 qman_ip_rev;
+EXPORT_SYMBOL(qman_ip_rev);
+
+/* size of the fqd region in bytes */
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+static u32 fqd_size = (PAGE_SIZE << CONFIG_FSL_QMAN_FQD_SZ);
+#endif
+
+/*****************/
+/* Portal driver */
+/*****************/
+
+static struct dpa_uio_class qman_uio = {
+	.list = LIST_HEAD_INIT(qman_uio.list),
+	.dev_prefix = "qman-uio-"
+};
+const struct dpa_uio_class *dpa_uio_qman(void)
+{
+	return &qman_uio;
+}
+EXPORT_SYMBOL(dpa_uio_qman);
+
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+/* Handlers for NULL portal callbacks (ie. where the contextB field, normally
+ * pointing to the corresponding FQ object, is NULL). */
+static enum qman_cb_dqrr_result null_cb_dqrr(struct qman_portal *qm,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dqrr)
+{
+	pr_warning("Ignoring unowned DQRR frame on portal %p.\n", qm);
+	return qman_cb_dqrr_consume;
+}
+static void null_cb_mr(struct qman_portal *qm, struct qman_fq *fq,
+			const struct qm_mr_entry *msg)
+{
+	pr_warning("Ignoring unowned MR msg on portal %p, verb 0x%02x.\n",
+			qm, msg->verb);
+}
+static const struct qman_fq_cb null_cb = {
+	.dqrr = null_cb_dqrr,
+	.ern = null_cb_mr,
+	.dc_ern = null_cb_mr,
+	.fqs = null_cb_mr
+};
+#endif
+
+#ifdef CONFIG_FSL_QMAN_PORTAL
+/* This structure carries parameters from the device-tree handling code that
+ * wants to set up a portal for use on 1 or more CPUs, and each temporary thread
+ * created to run on those CPUs. The 'portal' member is the return value. */
+struct affine_portal_data {
+	struct completion done;
+	const struct qm_portal_config *pconfig;
+	struct qman_portal *redirect;
+	int recovery_mode;
+	struct qman_portal *portal;
+};
+
+/* This function is called in a temporary thread for each CPU, to initialise the
+ * "affine" portal that the CPU should use. The thread is created and run from
+ * the init_affine_portal() bootstrapper. If the CPU has not been assigned its
+ * own portal, "redirect" will be non-NULL indicating it should share another
+ * CPU's portal (it becomes a "slave"). */
+static __init int thread_init_affine_portal(void *__data)
+{
+	struct affine_portal_data *data = __data;
+	const struct qm_portal_config *pconfig = data->pconfig;
+	if (data->redirect)
+		data->portal = qman_create_affine_slave(data->redirect);
+	else {
+		/* TODO: cgrs ?? */
+		data->portal = qman_create_affine_portal(pconfig, NULL,
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+				&null_cb,
+#endif
+				data->recovery_mode);
+		if (data->portal) {
+			u32 irq_sources = 0;
+			/* default: enable all (available) pool channels */
+			qman_static_dequeue_add(~0);
+			/* Determine what should be interrupt-vs-poll driven */
+#ifdef CONFIG_FSL_DPA_PIRQ_SLOW
+			irq_sources |= QM_PIRQ_EQCI | QM_PIRQ_EQRI |
+				QM_PIRQ_MRI | QM_PIRQ_CSCI;
+#endif
+#ifdef CONFIG_FSL_DPA_PIRQ_FAST
+			irq_sources |= QM_PIRQ_DQRI;
+#endif
+			qman_irqsource_add(irq_sources);
+		}
+	}
+	complete(&data->done);
+	return 0;
+}
+
+/* This function is just a bootstrap for running thread_init_affine_portal() on
+ * a given CPU. The parameters are passed in via the (void*) thread-arg (and
+ * results are received back) using the affine_portal_data struct. */
+static __init struct qman_portal *init_affine_portal(
+					const struct qm_portal_config *pconfig,
+					int cpu, struct qman_portal *redirect,
+					int recovery_mode)
+{
+	struct affine_portal_data data = {
+		.done = COMPLETION_INITIALIZER_ONSTACK(data.done),
+		.pconfig = pconfig,
+		.redirect = redirect,
+		.recovery_mode = recovery_mode,
+		.portal = NULL
+	};
+	struct task_struct *k = kthread_create(thread_init_affine_portal, &data,
+		"qman_affine%d", cpu);
+	int ret;
+	if (IS_ERR(k)) {
+		pr_err("Failed to init %sQman affine portal for cpu %d\n",
+			redirect ? "(slave) " : "", cpu);
+		return NULL;
+	}
+	kthread_bind(k, cpu);
+	wake_up_process(k);
+	wait_for_completion(&data.done);
+	ret = kthread_stop(k);
+	if (ret) {
+		pr_err("Qman portal initialisation failed, cpu %d, code %d\n",
+			cpu, ret);
+		return NULL;
+	}
+	if (data.portal)
+		pr_info("Qman portal %sinitialised, cpu %d\n",
+			redirect ? "(slave) " :
+			pconfig->public_cfg.is_shared ? "(shared) " : "", cpu);
+	return data.portal;
+}
+#endif
+
+/* Parses the device-tree node, extracts the configuration, and if appropriate
+ * initialises the portal for use on one or more CPUs. */
+static __init struct qm_portal_config *fsl_qman_portal_init(
+					struct device_node *node)
+{
+	struct qm_portal_config *pcfg;
+	const u32 *index, *channel;
+	const phandle *ph;
+	struct device_node *tmp_node;
+	int irq, ret, numpools;
+	u16 ip_rev = 0;
+
+	pcfg = kmalloc(sizeof(*pcfg), GFP_KERNEL);
+	if (!pcfg) {
+		pr_err("can't allocate portal config");
+		return NULL;
+	}
+
+	if (of_device_is_compatible(node, "fsl,qman-portal-1.0"))
+		ip_rev = QMAN_REV10;
+	else if (of_device_is_compatible(node, "fsl,qman-portal-1.1"))
+		ip_rev = QMAN_REV11;
+	else if	(of_device_is_compatible(node, "fsl,qman-portal-1.2"))
+		ip_rev = QMAN_REV12;
+	else if (of_device_is_compatible(node, "fsl,qman-portal-2.0"))
+		ip_rev = QMAN_REV20;
+
+	if (!qman_ip_rev) {
+		if (ip_rev)
+			qman_ip_rev = ip_rev;
+		else {
+			pr_warning("unknown Qman version, default to rev1.1\n");
+			qman_ip_rev = QMAN_REV11;
+		}
+	} else if (ip_rev && (qman_ip_rev != ip_rev))
+		pr_warning("Revision=0x%04x, but portal '%s' has 0x%04x\n",
+			qman_ip_rev, node->full_name, ip_rev);
+
+	ret = of_address_to_resource(node, QM_ADDR_CE,
+				&pcfg->addr_phys[QM_ADDR_CE]);
+	if (ret) {
+		pr_err("Can't get %s property '%s'\n", node->full_name,
+			"reg::CE");
+		goto err;
+	}
+	ret = of_address_to_resource(node, QM_ADDR_CI,
+				&pcfg->addr_phys[QM_ADDR_CI]);
+	if (ret) {
+		pr_err("Can't get %s property '%s'\n", node->full_name,
+			"reg::CI");
+		goto err;
+	}
+	index = of_get_property(node, "cell-index", &ret);
+	if (!index || (ret != 4)) {
+		pr_err("Can't get %s property '%s'\n", node->full_name,
+			"cell-index");
+		goto err;
+	}
+	channel = of_get_property(node, "fsl,qman-channel-id", &ret);
+	if (!channel || (ret != 4)) {
+		pr_err("Can't get %s property '%s'\n", node->full_name,
+			"fsl,qman-channel-id");
+		goto err;
+	}
+	if (*channel != (*index + qm_channel_swportal0))
+		pr_err("Warning: node %s has mismatched %s and %s\n",
+			node->full_name, "cell-index", "fsl,qman-channel-id");
+	pcfg->public_cfg.channel = *channel;
+	/* Parse cpu associations for this portal. This involves dereferencing
+	 * to the cpu device-tree nodes, but it also ensures we only try to work
+	 * with CPUs that exist. (Eg. under a hypervisor.) */
+	ph = of_get_property(node, "cpu-handle", &ret);
+	if (ph) {
+		if (ret != sizeof(phandle)) {
+			pr_err("Malformed %s property '%s'\n", node->full_name,
+				"cpu-handle");
+			return NULL;
+		}
+		ret = check_cpu_phandle(*ph);
+		if (ret < 0)
+			return NULL;
+		pcfg->public_cfg.cpu = ret;
+	} else
+		pcfg->public_cfg.cpu = -1;
+
+	ph = of_get_property(node, "fsl,qman-pool-channels", &ret);
+	if (ph && (ret % sizeof(phandle))) {
+		pr_err("Malformed %s property '%s'\n", node->full_name,
+			"fsl,qman-pool-channels");
+		goto err;
+	}
+	numpools = ph ? (ret / sizeof(phandle)) : 0;
+	irq = irq_of_parse_and_map(node, 0);
+	if (irq == NO_IRQ) {
+		pr_err("Can't get %s property '%s'\n", node->full_name,
+			"interrupts");
+		goto err;
+	}
+	pcfg->public_cfg.irq = irq;
+	pcfg->public_cfg.index = *index;
+	if (of_get_property(node, "fsl,hv-dma-handle", &ret))
+		pcfg->public_cfg.has_stashing = 1;
+	else if (qman_ip_rev == QMAN_REV20)
+		pcfg->public_cfg.has_stashing = 1;
+	else
+#ifdef CONFIG_FSL_PAMU
+		pcfg->public_cfg.has_stashing = 1;
+#else
+		pcfg->public_cfg.has_stashing = 0;
+#endif
+	pcfg->public_cfg.pools = 0;
+	pcfg->node = node;
+#ifdef CONFIG_FSL_QMAN_CONFIG
+	/* We need the same LIODN offset for all portals */
+	qman_liodn_fixup(pcfg->public_cfg.channel);
+#endif
+
+	if (of_get_property(node, "fsl,usdpaa-portal", &ret)) {
+		struct dpa_uio_portal *u = kmalloc(sizeof(*u), GFP_KERNEL);
+		if (!u)
+			goto err;
+		u->type = dpa_uio_portal_qman;
+		u->qm_cfg = pcfg;
+		list_add_tail(&u->node, &qman_uio.list);
+		/* Return NULL, otherwise the kernel may share it on CPUs that
+		 * don't have their own portals, which would be ... *bad*. */
+		 return NULL;
+	}
+
+	/* Map the portals now we know they aren't for UIO (the UIO code doesn't
+	 * need the CE mapping, and so will do its own CI-only mapping). */
+	pcfg->addr_virt[QM_ADDR_CE] = ioremap_prot(
+				pcfg->addr_phys[QM_ADDR_CE].start,
+				resource_size(&pcfg->addr_phys[QM_ADDR_CE]),
+				0);
+	pcfg->addr_virt[QM_ADDR_CI] = ioremap_prot(
+				pcfg->addr_phys[QM_ADDR_CI].start,
+				resource_size(&pcfg->addr_phys[QM_ADDR_CI]),
+				_PAGE_GUARDED | _PAGE_NO_CACHE);
+
+	while (numpools--) {
+		for_each_compatible_node(tmp_node, NULL,
+					 "fsl,qman-pool-channel") {
+			phandle *lph = (phandle *)of_get_property(tmp_node,
+				 "linux,phandle", &ret);
+			if (*lph == *ph) {
+				u32 *index = (u32 *)of_get_property(tmp_node,
+					"cell-index", &ret);
+				pcfg->public_cfg.pools |=
+					QM_SDQCR_CHANNELS_POOL(*index);
+			}
+		}
+		ph++;
+	}
+	if (pcfg->public_cfg.pools == 0)
+		panic("Unrecoverable error linking pool channels");
+
+	return pcfg;
+err:
+	kfree(pcfg);
+	return NULL;
+}
+
+static void __init fsl_qman_portal_destroy(struct qm_portal_config *pcfg)
+{
+	iounmap(pcfg->addr_virt[QM_ADDR_CE]);
+	iounmap(pcfg->addr_virt[QM_ADDR_CI]);
+	kfree(pcfg);
+}
+
+static __init int fsl_fqid_range_init(struct device_node *node,
+					int recovery_mode)
+{
+	int ret;
+	u32 *range = (u32 *)of_get_property(node, "fsl,fqid-range", &ret);
+	if (!range) {
+		pr_err("No 'fsl,fqid-range' property in node %s\n",
+			node->full_name);
+		return -EINVAL;
+	}
+	if (ret != 8) {
+		pr_err("'fsl,fqid-range' is not a 2-cell range in node %s\n",
+			node->full_name);
+		return -EINVAL;
+	}
+	qman_release_fqid_range(range[0], range[1]);
+#ifdef CONFIG_FSL_QMAN_PORTAL
+	/* If in recovery mode *and* we are using a private FQ allocation range,
+	 * then automatically clean up all FQs in that range so we can
+	 * automatically exit recovery mode too. */
+	if (recovery_mode) {
+		u32 fqid;
+		for (fqid = range[0]; fqid < (range[0] + range[1]); fqid++) {
+			ret = qman_recovery_cleanup_fq(fqid);
+			if (ret) {
+				pr_err("Failed to recovery FQID %d\n", fqid);
+				return ret;
+			}
+		}
+	}
+#else
+	BUG_ON(recovery_mode);
+#endif
+	pr_info("Qman: FQID allocator includes range %d:%d%s\n",
+		range[0], range[1], recovery_mode ? " (recovered)" : "");
+	return 0;
+}
+
+#ifdef CONFIG_FSL_QMAN_PORTAL
+static __init int __leave_recovery(void *__data)
+{
+	struct completion *done = __data;
+	qman_recovery_exit_local();
+	complete(done);
+	return 0;
+}
+
+int qman_recovery_exit(void)
+{
+	struct completion done = COMPLETION_INITIALIZER_ONSTACK(done);
+	unsigned int cpu;
+
+	for_each_cpu(cpu, qman_affine_cpus()) {
+		struct task_struct *k = kthread_create(__leave_recovery, &done,
+						"qman_recovery");
+		int ret;
+		if (IS_ERR(k)) {
+			pr_err("Thread failure (recovery) on cpu %d\n", cpu);
+			return -ENOMEM;
+		}
+		kthread_bind(k, cpu);
+		wake_up_process(k);
+		wait_for_completion(&done);
+		ret = kthread_stop(k);
+		if (ret) {
+			pr_err("Failed to exit recovery on cpu %d\n", cpu);
+			return ret;
+		}
+		pr_info("Qman portal exited recovery, cpu %d\n", cpu);
+	}
+	return 0;
+}
+EXPORT_SYMBOL(qman_recovery_exit);
+#endif
+
+/***************/
+/* Driver load */
+/***************/
+
+static __init int qman_init(void)
+{
+#ifdef CONFIG_FSL_QMAN_PORTAL
+	struct qman_cgr cgr;
+	struct cpumask primary_cpus = *cpu_none_mask;
+	struct cpumask slave_cpus = *cpu_online_mask;
+	struct qman_portal *sharing_portal = NULL;
+	int sharing_cpu = -1;
+#endif
+	struct device_node *dn;
+	struct qm_portal_config *pcfg;
+	int ret, use_bpid0 = 1, recovery_mode = 0;
+	LIST_HEAD(cfg_list);
+
+	for_each_compatible_node(dn, NULL, "fsl,qman") {
+		if (!qman_init_error_int(dn))
+			pr_info("Qman err interrupt handler present\n");
+		else
+			pr_err("Qman err interrupt handler missing\n");
+	}
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+	ret = qman_setup_fq_lookup_table(fqd_size/64);
+	if (ret)
+		return ret;
+#endif
+#ifdef CONFIG_FSL_QMAN_PORTAL
+	if (fsl_dpa_should_recover())
+		recovery_mode = 1;
+	for_each_compatible_node(dn, NULL, "fsl,qman-portal") {
+		pcfg = fsl_qman_portal_init(dn);
+		if (pcfg) {
+			if (pcfg->public_cfg.cpu >= 0) {
+				cpumask_set_cpu(pcfg->public_cfg.cpu,
+						&primary_cpus);
+				list_add(&pcfg->list, &cfg_list);
+			} else
+				fsl_qman_portal_destroy(pcfg);
+		}
+	}
+	/* only consider "online" CPUs */
+	cpumask_and(&primary_cpus, &primary_cpus, cpu_online_mask);
+	if (cpumask_empty(&primary_cpus))
+		/* No portals, we're done */
+		return 0;
+	if (!cpumask_subset(cpu_online_mask, &primary_cpus)) {
+		/* Need to do some sharing. In lieu of anything more scientific
+		 * (or configurable), we pick the last-most CPU that has a
+		 * portal and share that one. */
+		int next = cpumask_first(&primary_cpus);
+		while (next < nr_cpu_ids) {
+			sharing_cpu = next;
+			next = cpumask_next(next, &primary_cpus);
+		}
+	}
+	/* Parsing is done and sharing decisions are made, now initialise the
+	 * portals and determine which "slave" CPUs are left over. */
+	list_for_each_entry(pcfg, &cfg_list, list) {
+		struct qman_portal *p;
+		int is_shared = (!sharing_portal && (sharing_cpu >= 0) &&
+				(pcfg->public_cfg.cpu == sharing_cpu));
+		pcfg->public_cfg.is_shared = is_shared;
+		/* If it's not mapped to a CPU, or another portal is already
+		 * initialised to the same CPU, skip this portal. */
+		if (pcfg->public_cfg.cpu < 0 || !cpumask_test_cpu(
+					pcfg->public_cfg.cpu, &slave_cpus))
+			continue;
+		p = init_affine_portal(pcfg, pcfg->public_cfg.cpu, NULL,
+					recovery_mode);
+		if (p) {
+			if (is_shared)
+				sharing_portal = p;
+			cpumask_clear_cpu(pcfg->public_cfg.cpu, &slave_cpus);
+		}
+	}
+	if (sharing_portal) {
+		int loop;
+		for_each_cpu(loop, &slave_cpus) {
+			struct qman_portal *p = init_affine_portal(NULL, loop,
+					sharing_portal, recovery_mode);
+			if (!p)
+				pr_err("Failed slave Qman portal for cpu %d\n",
+					loop);
+		}
+	}
+#else
+	for_each_compatible_node(dn, NULL, "fsl,qman-portal") {
+		pcfg = fsl_qman_portal_init(dn);
+		if (pcfg)
+			/* No kernel portal support, so if USDPAA didn't consume
+			 * the portal, we've no other use for it. */
+			fsl_qman_portal_destroy(pcfg);
+	}
+#endif
+	for_each_compatible_node(dn, NULL, "fsl,fqid-range") {
+		use_bpid0 = 0;
+		ret = fsl_fqid_range_init(dn, recovery_mode);
+		if (ret)
+			return ret;
+	}
+#ifdef CONFIG_FSL_QMAN_PORTAL
+	/* If using private FQ allocation, exit recovery mode automatically (ie.
+	 * after automatic recovery) */
+	if (recovery_mode && !use_bpid0) {
+		ret = qman_recovery_exit();
+		if (ret)
+			return ret;
+	}
+	for (cgr.cgrid = 0; cgr.cgrid < __CGR_NUM; cgr.cgrid++) {
+		/* This is to ensure h/w-internal CGR memory is zeroed out. Note
+		 * that we do this for all conceivable CGRIDs, not all of which
+		 * are necessarily available on the underlying hardware version.
+		 * We ignore any errors for this reason. */
+		qman_modify_cgr(&cgr, QMAN_CGR_FLAG_USE_INIT, NULL);
+	}
+#endif
+	ret = fqalloc_init(use_bpid0);
+	if (ret)
+		return ret;
+	pr_info("Qman portals initialised\n");
+	return 0;
+}
+subsys_initcall(qman_init);
diff --git a/drivers/staging/fsl_qbman/qman_fqalloc.c b/drivers/staging/fsl_qbman/qman_fqalloc.c
new file mode 100644
index 0000000..e3f270e
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_fqalloc.c
@@ -0,0 +1,264 @@ 
+/* Copyright 2009-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_private.h"
+
+#include <linux/fsl_bman.h>
+
+/****************/
+/* FQ allocator */
+/****************/
+
+/* Global flag: use BPID==0 (fq_pool), or use the range-allocator? */
+static int use_bman;
+
+#ifdef CONFIG_FSL_BMAN_PORTAL
+static struct bman_pool *fq_pool;
+static const struct bman_pool_params fq_pool_params;
+#endif
+
+__init int fqalloc_init(int __use_bman)
+{
+	use_bman = __use_bman;
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	if (use_bman) {
+		fq_pool = bman_new_pool(&fq_pool_params);
+		if (!fq_pool)
+			return -ENOMEM;
+	}
+#else
+	BUG_ON(use_bman);
+#endif
+	return 0;
+}
+
+u32 qm_fq_new(void)
+{
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	struct bm_buffer buf;
+	int ret;
+#endif
+
+	if (!use_bman) {
+		u32 result;
+		if (qman_alloc_fqid(&result) < 0)
+			return 0;
+		return result;
+	}
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	BUG_ON(!fq_pool);
+	ret = bman_acquire(fq_pool, &buf, 1, 0);
+	if (ret != 1)
+		return 0;
+	return (u32)bm_buffer_get64(&buf);
+#else
+	BUG();
+#endif
+}
+EXPORT_SYMBOL(qm_fq_new);
+
+int qm_fq_free_flags(u32 fqid, __maybe_unused u32 flags)
+{
+#ifdef CONFIG_FSL_BMAN_PORTAL
+	struct bm_buffer buf;
+	u32 bflags = 0;
+	int ret;
+	bm_buffer_set64(&buf, fqid);
+#endif
+
+	if (!use_bman) {
+		qman_release_fqid(fqid);
+		return 0;
+	}
+#ifdef CONFIG_FSL_BMAN_PORTAL
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+	if (flags & QM_FQ_FREE_WAIT) {
+		bflags |= BMAN_RELEASE_FLAG_WAIT;
+		if (flags & BMAN_RELEASE_FLAG_WAIT_INT)
+			bflags |= BMAN_RELEASE_FLAG_WAIT_INT;
+		if (flags & BMAN_RELEASE_FLAG_WAIT_SYNC)
+			bflags |= BMAN_RELEASE_FLAG_WAIT_SYNC;
+	}
+#endif
+	ret = bman_release(fq_pool, &buf, 1, bflags);
+	return ret;
+#else
+	BUG();
+#endif
+}
+EXPORT_SYMBOL(qm_fq_free_flags);
+
+/* Global state for the allocator */
+static DEFINE_SPINLOCK(alloc_lock);
+static LIST_HEAD(alloc_list);
+
+/* The allocator is a (possibly-empty) list of these; */
+struct alloc_node {
+	struct list_head list;
+	u32 base;
+	u32 num;
+};
+
+/* #define FQRANGE_DEBUG */
+
+#ifdef FQRANGE_DEBUG
+#define DPRINT		pr_info
+static void DUMP(void)
+{
+	int off = 0;
+	char buf[256];
+	struct alloc_node *p;
+	list_for_each_entry(p, &alloc_list, list) {
+		if (off < 255)
+			off += snprintf(buf + off, 255-off, "{%d,%d}",
+				p->base, p->base + p->num - 1);
+	}
+	pr_info("%s\n", buf);
+}
+#else
+#define DPRINT(x...)	do { ; } while(0)
+#define DUMP()		do { ; } while(0)
+#endif
+
+int qman_alloc_fqid_range(u32 *result, u32 count, u32 align, int partial)
+{
+	struct alloc_node *i = NULL, *next_best = NULL;
+	u32 base, next_best_base = 0, num = 0, next_best_num = 0;
+	struct alloc_node *margin_left, *margin_right;
+
+	*result = (u32)-1;
+	DPRINT("alloc_range(%d,%d,%d)\n", count, align, partial);
+	DUMP();
+	/* If 'align' is 0, it should behave as though it was 1 */
+	if (!align)
+		align = 1;
+	margin_left = kmalloc(sizeof(*margin_left), GFP_KERNEL);
+	if (!margin_left)
+		goto err;
+	margin_right = kmalloc(sizeof(*margin_right), GFP_KERNEL);
+	if (!margin_right) {
+		kfree(margin_left);
+		goto err;
+	}
+	spin_lock_irq(&alloc_lock);
+	list_for_each_entry(i, &alloc_list, list) {
+		base = (i->base + align - 1) / align;
+		base *= align;
+		if ((base - i->base) >= i->num)
+			/* alignment is impossible, regardless of count */
+			continue;
+		num = i->num - (base - i->base);
+		if (num >= count) {
+			/* this one will do nicely */
+			num = count;
+			goto done;
+		}
+		if (num > next_best_num) {
+			next_best = i;
+			next_best_base = base;
+			next_best_num = num;
+		}
+	}
+	if (partial && next_best) {
+		i = next_best;
+		base = next_best_base;
+		num = next_best_num;
+	} else
+		i = NULL;
+done:
+	if (i) {
+		if (base != i->base) {
+			margin_left->base = i->base;
+			margin_left->num = base - i->base;
+			list_add_tail(&margin_left->list, &i->list);
+		} else
+			kfree(margin_left);
+		if ((base + num) < (i->base + i->num)) {
+			margin_right->base = base + num;
+			margin_right->num = (i->base + i->num) -
+						(base + num);
+			list_add(&margin_right->list, &i->list);
+		} else
+			kfree(margin_right);
+		list_del(&i->list);
+		kfree(i);
+		*result = base;
+	}
+	spin_unlock_irq(&alloc_lock);
+err:
+	DPRINT("returning %d\n", i ? num : -ENOMEM);
+	DUMP();
+	return i ? (int)num : -ENOMEM;
+}
+EXPORT_SYMBOL(qman_alloc_fqid_range);
+
+void qman_release_fqid_range(u32 fqid, u32 count)
+{
+	struct alloc_node *i, *node = kmalloc(sizeof(*node), GFP_KERNEL);
+	DPRINT("release_range(%d,%d)\n", fqid, count);
+	DUMP();
+	spin_lock_irq(&alloc_lock);
+	node->base = fqid;
+	node->num = count;
+	list_for_each_entry(i, &alloc_list, list) {
+		if (i->base >= node->base) {
+			list_add_tail(&node->list, &i->list);
+			goto done;
+		}
+	}
+	list_add_tail(&node->list, &alloc_list);
+done:
+	/* Merge to the left */
+	i = list_entry(node->list.prev, struct alloc_node, list);
+	if (node->list.prev != &alloc_list) {
+		BUG_ON((i->base + i->num) > node->base);
+		if ((i->base + i->num) == node->base) {
+			node->base = i->base;
+			node->num += i->num;
+			list_del(&i->list);
+			kfree(i);
+		}
+	}
+	/* Merge to the right */
+	i = list_entry(node->list.next, struct alloc_node, list);
+	if (node->list.next != &alloc_list) {
+		BUG_ON((node->base + node->num) > i->base);
+		if ((node->base + node->num) == i->base) {
+			node->num += i->num;
+			list_del(&i->list);
+			kfree(i);
+		}
+	}
+	spin_unlock_irq(&alloc_lock);
+	DUMP();
+}
+EXPORT_SYMBOL(qman_release_fqid_range);
+
diff --git a/drivers/staging/fsl_qbman/qman_high.c b/drivers/staging/fsl_qbman/qman_high.c
new file mode 100644
index 0000000..584d271
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_high.c
@@ -0,0 +1,2361 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_low.h"
+
+/* Compilation constants */
+#define DQRR_MAXFILL	15
+#define EQCR_ITHRESH	4	/* if EQCR congests, interrupt threshold */
+#define IRQNAME		"QMan portal %d"
+#define MAX_IRQNAME	16	/* big enough for "QMan portal %d" */
+
+/* Lock/unlock frame queues, subject to the "LOCKED" flag. This is about
+ * inter-processor locking only. Note, FQLOCK() is always called either under a
+ * local_irq_save() or from interrupt context - hence there's no need for irq
+ * protection (and indeed, attempting to nest irq-protection doesn't work, as
+ * the "irq en/disable" machinery isn't recursive...). */
+#define FQLOCK(fq) \
+	do { \
+		struct qman_fq *__fq478 = (fq); \
+		if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \
+			spin_lock(&__fq478->fqlock); \
+	} while(0)
+#define FQUNLOCK(fq) \
+	do { \
+		struct qman_fq *__fq478 = (fq); \
+		if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \
+			spin_unlock(&__fq478->fqlock); \
+	} while(0)
+
+static inline void fq_set(struct qman_fq *fq, u32 mask)
+{
+	set_bits(mask, &fq->flags);
+}
+static inline void fq_clear(struct qman_fq *fq, u32 mask)
+{
+	clear_bits(mask, &fq->flags);
+}
+static inline int fq_isset(struct qman_fq *fq, u32 mask)
+{
+	return fq->flags & mask;
+}
+static inline int fq_isclear(struct qman_fq *fq, u32 mask)
+{
+	return !(fq->flags & mask);
+}
+
+#define PORTAL_BITS_CI_PREFETCH	0x00020000	/* EQCR::CI prefetched */
+#define PORTAL_BITS_RECOVERY	0x00040000	/* recovery mode */
+
+struct qman_portal {
+	struct qm_portal p;
+	unsigned long bits; /* PORTAL_BITS_*** - dynamic, strictly internal */
+	unsigned long irq_sources;
+	u32 slowpoll;	/* only used when interrupts are off */
+	struct qman_fq *vdqcr_owned; /* only 1 volatile dequeue at a time */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	struct qman_fq *eqci_owned; /* only 1 enqueue WAIT_SYNC at a time */
+#endif
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	spinlock_t sharing_lock; /* only used if is_shared */
+	int is_shared;
+	struct qman_portal *sharing_redirect;
+#endif
+	u32 sdqcr;
+	int dqrr_disable_ref;
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+	/* If we receive a DQRR or MR ring entry for a "null" FQ, ie. for which
+	 * FQD::contextB is NULL rather than pointing to a FQ object, we use
+	 * these handlers. (This is not considered a fast-path mechanism.) */
+	struct qman_fq_cb null_cb;
+#endif
+	/* When the cpu-affine portal is activated, this is non-NULL */
+	const struct qm_portal_config *config;
+	/* This is needed for providing a non-NULL device to dma_map_***() */
+	struct platform_device *pdev;
+	struct dpa_rbtree retire_table;
+	char irqname[MAX_IRQNAME];
+	/* 2-element array. cgrs[0] is mask, cgrs[1] is snapshot. */
+	struct qman_cgrs *cgrs;
+	/* 256-element array, each is a linked-list of CSCN handlers. */
+	struct list_head cgr_cbs[256];
+	/* list lock */
+	spinlock_t cgr_lock;
+};
+
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+#define PORTAL_IRQ_LOCK(p, irqflags) \
+	do { \
+		if ((p)->is_shared) \
+			spin_lock_irqsave(&(p)->sharing_lock, irqflags); \
+		else \
+			local_irq_save(irqflags); \
+	} while (0)
+#define PORTAL_IRQ_UNLOCK(p, irqflags) \
+	do { \
+		if ((p)->is_shared) \
+			spin_unlock_irqrestore(&(p)->sharing_lock, irqflags); \
+		else \
+			local_irq_restore(irqflags); \
+	} while (0)
+#else
+#define PORTAL_IRQ_LOCK(p, irqflags) local_irq_save(irqflags)
+#define PORTAL_IRQ_UNLOCK(p, irqflags) local_irq_restore(irqflags)
+#endif
+
+static cpumask_t affine_mask;
+static DEFINE_SPINLOCK(affine_mask_lock);
+static DEFINE_PER_CPU(struct qman_portal, qman_affine_portal);
+/* "raw" gets the cpu-local struct whether it's a redirect or not. */
+static inline struct qman_portal *get_raw_affine_portal(void)
+{
+	return &get_cpu_var(qman_affine_portal);
+}
+/* For ops that can redirect, this obtains the portal to use */
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+static inline struct qman_portal *get_affine_portal(void)
+{
+	struct qman_portal *p = get_raw_affine_portal();
+	if (p->sharing_redirect)
+		return p->sharing_redirect;
+	return p;
+}
+#else
+#define get_affine_portal() get_raw_affine_portal()
+#endif
+/* For every "get", there must be a "put" */
+static inline void put_affine_portal(void)
+{
+	put_cpu_var(qman_affine_portal);
+}
+
+/* This gives a FQID->FQ lookup to cover the fact that we can't directly demux
+ * retirement notifications (the fact they are sometimes h/w-consumed means that
+ * contextB isn't always a s/w demux - and as we can't know which case it is
+ * when looking at the notification, we have to use the slow lookup for all of
+ * them). NB, it's possible to have multiple FQ objects refer to the same FQID
+ * (though at most one of them should be the consumer), so this table isn't for
+ * all FQs - FQs are added when retirement commands are issued, and removed when
+ * they complete, which also massively reduces the size of this table. */
+IMPLEMENT_DPA_RBTREE(fqtree, struct qman_fq, node, fqid);
+
+/* This is what everything can wait on, even if it migrates to a different cpu
+ * to the one whose affine portal it is waiting on. */
+static DECLARE_WAIT_QUEUE_HEAD(affine_queue);
+
+static inline int table_push_fq(struct qman_portal *p, struct qman_fq *fq)
+{
+	int ret = fqtree_push(&p->retire_table, fq);
+	if (ret)
+		pr_err("ERROR: double FQ-retirement %d\n", fq->fqid);
+	return ret;
+}
+
+static inline void table_del_fq(struct qman_portal *p, struct qman_fq *fq)
+{
+	fqtree_del(&p->retire_table, fq);
+}
+
+static inline struct qman_fq *table_find_fq(struct qman_portal *p, u32 fqid)
+{
+	return fqtree_find(&p->retire_table, fqid);
+}
+
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+static void **qman_fq_lookup_table;
+static size_t qman_fq_lookup_table_size;
+
+int qman_setup_fq_lookup_table(size_t num_entries)
+{
+	num_entries++;
+	/* Allocate 1 more entry since the first entry is not used */
+	qman_fq_lookup_table = vmalloc((num_entries * sizeof(void *)));
+	if (!qman_fq_lookup_table) {
+		pr_err("QMan: Could not allocate fq lookup table\n");
+		return -ENOMEM;
+	}
+	memset(qman_fq_lookup_table, 0, num_entries * sizeof(void *));
+	qman_fq_lookup_table_size = num_entries;
+	pr_info("QMan: Allocated lookup table at %p, entry count %lu\n",
+			qman_fq_lookup_table,
+			(unsigned long)qman_fq_lookup_table_size);
+	return 0;
+}
+
+/* global structure that maintains fq object mapping */
+static DEFINE_SPINLOCK(fq_hash_table_lock);
+
+static int find_empty_fq_table_entry(u32 *entry, struct qman_fq *fq)
+{
+	u32 i;
+
+	spin_lock(&fq_hash_table_lock);
+	/* Can't use index zero because this has special meaning
+	 * in context_b field. */
+	for (i = 1; i < qman_fq_lookup_table_size; i++) {
+		if (qman_fq_lookup_table[i] == NULL) {
+			*entry = i;
+			qman_fq_lookup_table[i] = fq;
+			spin_unlock(&fq_hash_table_lock);
+			return 0;
+		}
+	}
+	spin_unlock(&fq_hash_table_lock);
+	return -ENOMEM;
+}
+
+static void clear_fq_table_entry(u32 entry)
+{
+	spin_lock(&fq_hash_table_lock);
+	BUG_ON(entry >= qman_fq_lookup_table_size);
+	qman_fq_lookup_table[entry] = NULL;
+	spin_unlock(&fq_hash_table_lock);
+}
+
+static inline struct qman_fq *get_fq_table_entry(u32 entry)
+{
+	BUG_ON(entry >= qman_fq_lookup_table_size);
+	return  qman_fq_lookup_table[entry];
+}
+#endif
+
+/* In the case that slow- and fast-path handling are both done by qman_poll()
+ * (ie. because there is no interrupt handling), we ought to balance how often
+ * we do the fast-path poll versus the slow-path poll. We'll use two decrementer
+ * sources, so we call the fast poll 'n' times before calling the slow poll
+ * once. The idle decrementer constant is used when the last slow-poll detected
+ * no work to do, and the busy decrementer constant when the last slow-poll had
+ * work to do. */
+#define SLOW_POLL_IDLE   1000
+#define SLOW_POLL_BUSY   10
+static u32 __poll_portal_slow(struct qman_portal *p, u32 is);
+static inline unsigned int __poll_portal_fast(struct qman_portal *p,
+					unsigned int poll_limit);
+
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+/* Portal interrupt handler */
+static irqreturn_t portal_isr(__always_unused int irq, void *ptr)
+{
+	struct qman_portal *p = ptr;
+	u32 clear = QM_DQAVAIL_MASK | p->irq_sources;
+	u32 is = qm_isr_status_read(&p->p) & p->irq_sources;
+	/* DQRR-handling if it's interrupt-driven */
+	if (is & QM_PIRQ_DQRI)
+		__poll_portal_fast(p, CONFIG_FSL_QMAN_POLL_LIMIT);
+	/* Handling of anything else that's interrupt-driven */
+	clear |= __poll_portal_slow(p, is);
+	qm_isr_status_clear(&p->p, clear);
+	return IRQ_HANDLED;
+}
+#endif
+
+/* This inner version is used privately by qman_create_affine_portal(), as well
+ * as by the exported qman_stop_dequeues(). */
+static inline void qman_stop_dequeues_ex(struct qman_portal *p)
+{
+	unsigned long irqflags __maybe_unused;
+	PORTAL_IRQ_LOCK(p, irqflags);
+	if (!(p->dqrr_disable_ref++))
+		qm_dqrr_set_maxfill(&p->p, 0);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+}
+
+static int drain_mr_fqrni(struct qm_portal *p)
+{
+	const struct qm_mr_entry *msg;
+loop:
+	msg = qm_mr_current(p);
+	if (!msg) {
+		/* if MR was full and h/w had other FQRNI entries to produce, we
+		 * need to allow it time to produce those entries once the
+		 * existing entries are consumed. A worst-case situation
+		 * (fully-loaded system) means h/w sequencers may have to do 3-4
+		 * other things before servicing the portal's MR pump, each of
+		 * which (if slow) may take ~50 qman cycles (which is ~200
+		 * processor cycles). So rounding up and then multiplying this
+		 * worst-case estimate by a factor of 10, just to be
+		 * ultra-paranoid, goes as high as 10,000 cycles. NB, we consume
+		 * one entry at a time, so h/w has an opportunity to produce new
+		 * entries well before the ring has been fully consumed, so
+		 * we're being *really* paranoid here. */
+		u64 now, then = mfatb();
+		do {
+			now = mfatb();
+		} while ((then + 10000) > now);
+		msg = qm_mr_current(p);
+		if (!msg)
+			return 0;
+	}
+	if ((msg->verb & QM_MR_VERB_TYPE_MASK) != QM_MR_VERB_FQRNI)
+		/* We aren't draining anything but FQRNIs */
+		return -1;
+	qm_mr_next(p);
+	qm_mr_cci_consume(p, 1);
+	goto loop;
+}
+
+/* this is called from qman_create_affine_portal() if not initialising in
+ * recovery mode, otherwise from qman_recovery_exit_local() after recovery is
+ * done. */
+static void post_recovery(struct qman_portal *p __always_unused,
+			const struct qm_portal_config *config)
+{
+	struct device_node *tmp_node, *node = config->node;
+	/* Enable DMA on portal LIODNs (stashing) and those of its sub-nodes
+	 * (Fman TX and SEC/PME accelerators, where available). */
+	if (pamu_enable_liodn(node, -1))
+		/* If there's a PAMU problem, best to continue anyway and let
+		 * the corresponding traffic hit whatever problems it will hit,
+		 * than to fail portal initialisation and trigger a crash in
+		 * dependent code that has no relationship to the PAMU issue. */
+		pr_err("Failed to enable portal LIODN %s\n",
+			node->full_name);
+	for_each_child_of_node(node, tmp_node)
+		if (pamu_enable_liodn(tmp_node, -1))
+			pr_err("Failed to enable portal LIODN %s\n",
+				tmp_node->full_name);
+}
+
+struct qman_portal *qman_create_affine_portal(
+			const struct qm_portal_config *config,
+			const struct qman_cgrs *cgrs,
+			const struct qman_fq_cb *null_cb,
+			int recovery_mode)
+{
+	struct qman_portal *portal = get_raw_affine_portal();
+	struct qm_portal *__p = &portal->p;
+	char buf[16];
+	int ret;
+	u32 isdr;
+
+	/* A criteria for calling this function (from qman_driver.c) is that
+	 * we're already affine to the cpu and won't schedule onto another cpu.
+	 * This means we can put_affine_portal() and yet continue to use
+	 * "portal", which in turn means aspects of this routine can sleep. */
+	put_affine_portal();
+#ifndef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+	if (null_cb) {
+		pr_err("Driver does not support 'NULL FQ' callbacks\n");
+		return NULL;
+	}
+#endif
+	/* prep the low-level portal struct with the mapped addresses from the
+	 * config, everything that follows depends on it and "config" is more
+	 * for (de)reference... */
+	__p->addr.addr_ce = config->addr_virt[QM_ADDR_CE];
+	__p->addr.addr_ci = config->addr_virt[QM_ADDR_CI];
+	if (qm_eqcr_init(__p, qm_eqcr_pvb, qm_eqcr_cce)) {
+		pr_err("Qman EQCR initialisation failed\n");
+		goto fail_eqcr;
+	}
+#ifdef CONFIG_FSL_QMAN_PORTAL_DISABLEAUTO_DCA
+#define QM_DQRR_CMODE qm_dqrr_cci
+#else
+#define QM_DQRR_CMODE qm_dqrr_cdc
+#endif
+	/* for recovery mode, don't enable stashing yet */
+	if (qm_dqrr_init(__p, config, qm_dqrr_dpush, qm_dqrr_pvb,
+			recovery_mode ?  qm_dqrr_cci : QM_DQRR_CMODE,
+			DQRR_MAXFILL, recovery_mode)) {
+		pr_err("Qman DQRR initialisation failed\n");
+		goto fail_dqrr;
+	}
+	if (qm_mr_init(__p, qm_mr_pvb, qm_mr_cci)) {
+		pr_err("Qman MR initialisation failed\n");
+		goto fail_mr;
+	}
+	/* for recovery mode, quiesce SDQCR/VDQCR and drain DQRR+MR until h/w
+	 * wraps up anything it was doing (5ms is ample idle time). */
+	if (recovery_mode) {
+		const struct qm_dqrr_entry *dq;
+		const struct qm_mr_entry *msg;
+		int idle = 0;
+		/* quiesce SDQCR/VDQCR, then drain till h/w wraps up anything it
+		 * was doing (5ms is more than enough to ensure it's done). */
+		qm_dqrr_sdqcr_set(__p, 0);
+		qm_dqrr_vdqcr_set(__p, 0);
+drain_loop:
+		qm_dqrr_pvb_update(__p, 0);
+		dq = qm_dqrr_current(__p);
+		qm_mr_pvb_update(__p);
+		msg = qm_mr_current(__p);
+		if (dq) {
+			pr_warning("DQRR recovery: dumping dqrr %02x:%02x for "
+				"FQID %d\n", dq->verb & QM_DQRR_VERB_MASK,
+				dq->stat, dq->fqid);
+			qm_dqrr_next(__p);
+			qm_dqrr_cci_consume(__p, 1);
+		}
+		if (msg) {
+			pr_warning("MR recovery: dumping msg 0x%02x for "
+				"FQID %d\n", msg->verb & QM_MR_VERB_TYPE_MASK,
+				msg->fq.fqid);
+			qm_mr_next(__p);
+			qm_mr_cci_consume(__p, 1);
+		}
+		if (!dq && !msg) {
+			if (++idle < 5) {
+				msleep(1);
+				goto drain_loop;
+			}
+		} else {
+			idle = 0;
+			goto drain_loop;
+		}
+	}
+	if (qm_mc_init(__p)) {
+		pr_err("Qman MC initialisation failed\n");
+		goto fail_mc;
+	}
+	if (qm_isr_init(__p)) {
+		pr_err("Qman ISR initialisation failed\n");
+		goto fail_isr;
+	}
+	/* static interrupt-gating controls */
+	qm_dqrr_set_ithresh(__p, CONFIG_FSL_QMAN_PIRQ_DQRR_ITHRESH);
+	qm_mr_set_ithresh(__p, CONFIG_FSL_QMAN_PIRQ_MR_ITHRESH);
+	qm_isr_set_iperiod(__p, CONFIG_FSL_QMAN_PIRQ_IPERIOD);
+	portal->cgrs = kmalloc(2 * sizeof(*cgrs), GFP_KERNEL);
+	if (!portal->cgrs)
+		goto fail_cgrs;
+	/* initial snapshot is no-depletion */
+	qman_cgrs_init(&portal->cgrs[1]);
+	if (cgrs)
+		portal->cgrs[0] = *cgrs;
+	else
+		/* if the given mask is NULL, assume all CGRs can be seen */
+		qman_cgrs_fill(&portal->cgrs[0]);
+	for (ret = 0; ret < __CGR_NUM; ret++)
+		INIT_LIST_HEAD(&portal->cgr_cbs[ret]);
+	spin_lock_init(&portal->cgr_lock);
+	portal->bits = recovery_mode ? PORTAL_BITS_RECOVERY : 0;
+	portal->slowpoll = 0;
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	portal->eqci_owned = NULL;
+#endif
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	spin_lock_init(&portal->sharing_lock);
+	portal->is_shared = config->public_cfg.is_shared;
+	portal->sharing_redirect = NULL;
+#endif
+	portal->sdqcr = QM_SDQCR_SOURCE_CHANNELS | QM_SDQCR_COUNT_UPTO3 |
+			QM_SDQCR_DEDICATED_PRECEDENCE | QM_SDQCR_TYPE_PRIO_QOS |
+			QM_SDQCR_TOKEN_SET(0xab) | QM_SDQCR_CHANNELS_DEDICATED;
+	portal->dqrr_disable_ref = 0;
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+	if (null_cb)
+		portal->null_cb = *null_cb;
+	else
+		memset(&portal->null_cb, 0, sizeof(*null_cb));
+#endif
+	sprintf(buf, "qportal-%d", config->public_cfg.channel);
+	portal->pdev = platform_device_alloc(buf, -1);
+	if (!portal->pdev)
+		goto fail_devalloc;
+	if (dma_set_mask(&portal->pdev->dev, DMA_BIT_MASK(40)))
+		goto fail_devadd;
+	ret = platform_device_add(portal->pdev);
+	if (ret)
+		goto fail_devadd;
+	dpa_rbtree_init(&portal->retire_table);
+	isdr = 0xffffffff;
+	qm_isr_disable_write(__p, isdr);
+	portal->irq_sources = 0;
+	qm_isr_enable_write(__p, portal->irq_sources);
+	qm_isr_status_clear(__p, 0xffffffff);
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, config->public_cfg.cpu);
+	if (request_irq(config->public_cfg.irq, portal_isr,
+			IRQF_NOBALANCING | IRQF_DISABLED, portal->irqname,
+			portal)) {
+		pr_err("request_irq() failed\n");
+		goto fail_irq;
+	}
+	if (config->public_cfg.cpu != -1) {
+		disable_irq(config->public_cfg.irq);
+		irq_set_affinity(config->public_cfg.irq,
+				 cpumask_of(config->public_cfg.cpu));
+	}
+	enable_irq(config->public_cfg.irq);
+
+	if (recovery_mode) {
+		qm_isr_inhibit(__p);
+	} else {
+		post_recovery(portal, config);
+		qm_isr_uninhibit(__p);
+	}
+#endif
+	/* Need EQCR to be empty before continuing */
+	isdr ^= QM_PIRQ_EQCI;
+	qm_isr_disable_write(__p, isdr);
+	ret = qm_eqcr_get_fill(__p);
+	if (ret) {
+		pr_err("Qman EQCR unclean, need recovery\n");
+		goto fail_eqcr_empty;
+	}
+	isdr ^= (QM_PIRQ_DQRI | QM_PIRQ_MRI);
+	qm_isr_disable_write(__p, isdr);
+	if (qm_dqrr_current(__p) != NULL) {
+		pr_err("Qman DQRR unclean, need recovery\n");
+		goto fail_dqrr_mr_empty;
+	}
+	if (qm_mr_current(__p) != NULL) {
+		/* special handling, drain just in case it's a few FQRNIs */
+		if (drain_mr_fqrni(__p)) {
+			pr_err("Qman MR unclean, need recovery\n");
+			goto fail_dqrr_mr_empty;
+		}
+	}
+	/* Success */
+	portal->config = config;
+	spin_lock(&affine_mask_lock);
+	cpumask_set_cpu(config->public_cfg.cpu, &affine_mask);
+	spin_unlock(&affine_mask_lock);
+	qm_isr_disable_write(__p, 0);
+	/* Write a sane SDQCR */
+	qm_dqrr_sdqcr_set(__p, recovery_mode ? 0 : portal->sdqcr);
+	return portal;
+fail_dqrr_mr_empty:
+fail_eqcr_empty:
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+fail_affinity:
+	free_irq(config->public_cfg.irq, portal);
+fail_irq:
+#endif
+	platform_device_del(portal->pdev);
+fail_devadd:
+	platform_device_put(portal->pdev);
+fail_devalloc:
+	if (portal->cgrs)
+		kfree(portal->cgrs);
+fail_cgrs:
+	qm_isr_finish(__p);
+fail_isr:
+	qm_mc_finish(__p);
+fail_mc:
+	qm_mr_finish(__p);
+fail_mr:
+	qm_dqrr_finish(__p);
+fail_dqrr:
+	qm_eqcr_finish(__p);
+fail_eqcr:
+	put_affine_portal();
+	return NULL;
+}
+
+/* These checks are BUG_ON()s because the driver is already supposed to avoid
+ * these cases. */
+struct qman_portal *qman_create_affine_slave(struct qman_portal *redirect)
+{
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	struct qman_portal *p = get_raw_affine_portal();
+	/* Check that we don't already have our own portal */
+	BUG_ON(p->config);
+	/* Check that we aren't already slaving to another portal */
+	BUG_ON(p->is_shared);
+	/* Check that 'redirect' is prepared to have us */
+	BUG_ON(!redirect->config->public_cfg.is_shared);
+	/* These are the only elements to initialise when redirecting */
+	p->irq_sources = 0;
+	p->sharing_redirect = redirect;
+	put_affine_portal();
+	return p;
+#else
+	BUG();
+	return NULL;
+#endif
+}
+
+const struct qm_portal_config *qman_destroy_affine_portal(void)
+{
+	/* We don't want to redirect if we're a slave, use "raw" */
+	struct qman_portal *qm = get_raw_affine_portal();
+	const struct qm_portal_config *pcfg;
+	int cpu;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (qm->sharing_redirect) {
+		qm->sharing_redirect = NULL;
+		put_affine_portal();
+		return NULL;
+	}
+	qm->is_shared = 0;
+#endif
+	pcfg = qm->config;
+	cpu = pcfg->public_cfg.cpu;
+	/* NB we do this to "quiesce" EQCR. If we add enqueue-completions or
+	 * something related to QM_PIRQ_EQCI, this may need fixing.
+	 * Also, due to the prefetching model used for CI updates in the enqueue
+	 * path, this update will only invalidate the CI cacheline *after*
+	 * working on it, so we need to call this twice to ensure a full update
+	 * irrespective of where the enqueue processing was at when the teardown
+	 * began. */
+	qm_eqcr_cce_update(&qm->p);
+	qm_eqcr_cce_update(&qm->p);
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	free_irq(pcfg->public_cfg.irq, qm);
+#endif
+	kfree(qm->cgrs);
+	qm_isr_finish(&qm->p);
+	qm_mc_finish(&qm->p);
+	qm_mr_finish(&qm->p);
+	qm_dqrr_finish(&qm->p);
+	qm_eqcr_finish(&qm->p);
+	qm->config = NULL;
+	spin_lock(&affine_mask_lock);
+	cpumask_clear_cpu(cpu, &affine_mask);
+	spin_unlock(&affine_mask_lock);
+	put_affine_portal();
+	return pcfg;
+}
+
+const struct qman_portal_config *qman_get_portal_config(void)
+{
+	struct qman_portal *p = get_affine_portal();
+	const struct qman_portal_config *ret = &p->config->public_cfg;
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_get_portal_config);
+
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+void qman_get_null_cb(struct qman_fq_cb *null_cb)
+{
+	struct qman_portal *p = get_affine_portal();
+	*null_cb = p->null_cb;
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_get_null_cb);
+
+void qman_set_null_cb(const struct qman_fq_cb *null_cb)
+{
+	struct qman_portal *p = get_affine_portal();
+	p->null_cb = *null_cb;
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_set_null_cb);
+#endif
+
+/* Inline helper to reduce nesting in __poll_portal_slow() */
+static inline void fq_state_change(struct qman_portal *p, struct qman_fq *fq,
+				const struct qm_mr_entry *msg, u8 verb)
+{
+	FQLOCK(fq);
+	switch(verb) {
+	case QM_MR_VERB_FQRL:
+		DPA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_ORL));
+		fq_clear(fq, QMAN_FQ_STATE_ORL);
+		table_del_fq(p, fq);
+		break;
+	case QM_MR_VERB_FQRN:
+		DPA_ASSERT((fq->state == qman_fq_state_parked) ||
+			(fq->state == qman_fq_state_sched));
+		DPA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_CHANGING));
+		fq_clear(fq, QMAN_FQ_STATE_CHANGING);
+		if (msg->fq.fqs & QM_MR_FQS_NOTEMPTY)
+			fq_set(fq, QMAN_FQ_STATE_NE);
+		if (msg->fq.fqs & QM_MR_FQS_ORLPRESENT)
+			fq_set(fq, QMAN_FQ_STATE_ORL);
+		else
+			table_del_fq(p, fq);
+		fq->state = qman_fq_state_retired;
+		break;
+	case QM_MR_VERB_FQPN:
+		DPA_ASSERT(fq->state == qman_fq_state_sched);
+		DPA_ASSERT(fq_isclear(fq, QMAN_FQ_STATE_CHANGING));
+		fq->state = qman_fq_state_parked;
+	}
+	FQUNLOCK(fq);
+}
+
+static u32 __poll_portal_slow(struct qman_portal *p, u32 is)
+{
+	const struct qm_mr_entry *msg;
+
+	BUG_ON(p->bits & PORTAL_BITS_RECOVERY);
+
+	if (is & QM_PIRQ_CSCI) {
+		struct qman_cgrs rr, c;
+		struct qm_mc_result *mcr;
+		struct qman_cgr *cgr;
+		int i;
+		unsigned long irqflags __maybe_unused;
+
+		spin_lock_irqsave(&p->cgr_lock, irqflags);
+		qm_mc_start(&p->p);
+		qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION);
+		while (!(mcr = qm_mc_result(&p->p)))
+			cpu_relax();
+		/* mask out the ones I'm not interested in */
+		qman_cgrs_and(&rr, (const struct qman_cgrs *)
+			&mcr->querycongestion.state, &p->cgrs[0]);
+		/* check previous snapshot for delta, enter/exit congestion */
+		qman_cgrs_xor(&c, &rr, &p->cgrs[1]);
+		/* update snapshot */
+		qman_cgrs_cp(&p->cgrs[1], &rr);
+		/* Invoke callback */
+		qman_cgrs_for_each_1(i, &c)
+			list_for_each_entry(cgr, &p->cgr_cbs[i], node) {
+				if (cgr->cb)
+					cgr->cb(p, cgr, qman_cgrs_get(&rr, i));
+			}
+		spin_unlock_irqrestore(&p->cgr_lock, irqflags);
+	}
+
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	if (is & QM_PIRQ_EQCI) {
+		unsigned long irqflags;
+		PORTAL_IRQ_LOCK(p, irqflags);
+		p->eqci_owned = NULL;
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+		wake_up(&affine_queue);
+	}
+#endif
+
+	if (is & QM_PIRQ_EQRI) {
+		unsigned long irqflags __maybe_unused;
+		PORTAL_IRQ_LOCK(p, irqflags);
+		qm_eqcr_cce_update(&p->p);
+		qm_eqcr_set_ithresh(&p->p, 0);
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+		wake_up(&affine_queue);
+	}
+
+	if (is & QM_PIRQ_MRI) {
+		u8 num = 0;
+mr_loop:
+		qm_mr_pvb_update(&p->p);
+		msg = qm_mr_current(&p->p);
+		if (msg) {
+			u8 verb = msg->verb & QM_MR_VERB_TYPE_MASK;
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+			struct qman_fq *fq;
+#else
+			struct qman_fq *fq = (void *)(uintptr_t)msg->ern.tag;
+#endif
+			if (verb == QM_MR_VERB_FQRNI) {
+				; /* nada, we drop FQRNIs on the floor */
+			} else if ((verb == QM_MR_VERB_FQRN) ||
+					(verb == QM_MR_VERB_FQRL)) {
+				/* It's retirement related - need a lookup */
+				fq = table_find_fq(p, msg->fq.fqid);
+				if (!fq)
+					panic("unexpected FQ retirement");
+				fq_state_change(p, fq, msg, verb);
+				if (fq->cb.fqs)
+					fq->cb.fqs(p, fq, msg);
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+			} else if (likely(msg->ern.tag)) {
+				fq = get_fq_table_entry(msg->ern.tag);
+#else
+			} else if (likely(fq)) {
+#endif
+				/* As per the header note, this is the way to
+				 * determine if it's a s/w ERN or not. */
+				if (likely(!(verb & QM_MR_VERB_DC_ERN)))
+					fq->cb.ern(p, fq, msg);
+				else
+					fq->cb.dc_ern(p, fq, msg);
+			}
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+			else {
+				/* use portal default handlers for 'null's */
+				if (likely(!(verb & QM_MR_VERB_DC_ERN)))
+					p->null_cb.ern(p, NULL, msg);
+				else if (verb == QM_MR_VERB_DC_ERN)
+					p->null_cb.dc_ern(p, NULL, msg);
+				else if (p->null_cb.fqs)
+					p->null_cb.fqs(p, NULL, msg);
+			}
+#endif
+			num++;
+			qm_mr_next(&p->p);
+			goto mr_loop;
+		}
+		qm_mr_cci_consume(&p->p, num);
+	}
+
+	return is & (QM_PIRQ_CSCI | QM_PIRQ_EQCI | QM_PIRQ_EQRI | QM_PIRQ_MRI);
+}
+
+/* remove some slowish-path stuff from the "fast path" and make sure it isn't
+ * inlined. */
+static noinline void clear_vdqcr(struct qman_portal *p, struct qman_fq *fq)
+{
+	p->vdqcr_owned = NULL;
+	FQLOCK(fq);
+	fq_clear(fq, QMAN_FQ_STATE_VDQCR);
+	FQUNLOCK(fq);
+	wake_up(&affine_queue);
+}
+
+/* Look: no locks, no irq_save()s, no preempt_disable()s! :-) The only states
+ * that would conflict with other things if they ran at the same time on the
+ * same cpu are;
+ *
+ *   (i) setting/clearing vdqcr_owned, and
+ *  (ii) clearing the NE (Not Empty) flag.
+ *
+ * Both are safe. Because;
+ *
+ *   (i) this clearing can only occur after qman_volatile_dequeue() has set the
+ *       vdqcr_owned field (which it does before setting VDQCR), and
+ *       qman_volatile_dequeue() blocks interrupts and preemption while this is
+ *       done so that we can't interfere.
+ *  (ii) the NE flag is only cleared after qman_retire_fq() has set it, and as
+ *       with (i) that API prevents us from interfering until it's safe.
+ *
+ * The good thing is that qman_volatile_dequeue() and qman_retire_fq() run far
+ * less frequently (ie. per-FQ) than __poll_portal_fast() does, so the nett
+ * advantage comes from this function not having to "lock" anything at all.
+ *
+ * Note also that the callbacks are invoked at points which are safe against the
+ * above potential conflicts, but that this function itself is not re-entrant
+ * (this is because the function tracks one end of each FIFO in the portal and
+ * we do *not* want to lock that). So the consequence is that it is safe for
+ * user callbacks to call into any Qman API *except* qman_poll() (as that's the
+ * sole API that could be invoking the callback through this function).
+ */
+static inline unsigned int __poll_portal_fast(struct qman_portal *p,
+					unsigned int poll_limit)
+{
+	const struct qm_dqrr_entry *dq;
+	struct qman_fq *fq;
+	enum qman_cb_dqrr_result res;
+#ifdef CONFIG_FSL_QMAN_DQRR_PREFETCHING
+	int coherent = (p->config->public_cfg.has_stashing);
+#endif
+	unsigned int limit = 0;
+
+	BUG_ON(p->bits & PORTAL_BITS_RECOVERY);
+loop:
+#ifdef CONFIG_FSL_QMAN_DQRR_PREFETCHING
+	qm_dqrr_pvb_update(&p->p, coherent);
+#else
+	qm_dqrr_pvb_update(&p->p, 1);
+#endif
+	dq = qm_dqrr_current(&p->p);
+	if (!dq)
+		goto done;
+	if (dq->stat & QM_DQRR_STAT_UNSCHEDULED) {
+		/* VDQCR: don't trust contextB as the FQ may have been
+		 * configured for h/w consumption and we're draining it
+		 * post-retirement. */
+		fq = p->vdqcr_owned;
+		/* We only set QMAN_FQ_STATE_NE when retiring, so we only need
+		 * to check for clearing it when doing volatile dequeues. It's
+		 * one less thing to check in the critical path (SDQCR). */
+		if (dq->stat & QM_DQRR_STAT_FQ_EMPTY)
+			fq_clear(fq, QMAN_FQ_STATE_NE);
+		/* this is duplicated from the SDQCR code, but we have stuff to
+		 * do before *and* after this callback, and we don't want
+		 * multiple if()s in the critical path (SDQCR). */
+		res = fq->cb.dqrr(p, fq, dq);
+		if (res == qman_cb_dqrr_stop)
+			goto done;
+		/* Check for VDQCR completion */
+		if (dq->stat & QM_DQRR_STAT_DQCR_EXPIRED)
+			clear_vdqcr(p, fq);
+	} else {
+		/* SDQCR: contextB points to the FQ */
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+		fq = get_fq_table_entry(dq->contextB);
+#else
+		fq = (void *)(uintptr_t)dq->contextB;
+#endif
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+		if (unlikely(!fq)) {
+			/* use portal default handlers */
+			res = p->null_cb.dqrr(p, NULL, dq);
+			DPA_ASSERT(res == qman_cb_dqrr_consume);
+			res = qman_cb_dqrr_consume;
+		} else
+#endif
+		{
+			/* Now let the callback do its stuff */
+			res = fq->cb.dqrr(p, fq, dq);
+			/* The callback can request that we exit without
+			 * consuming this entry nor advancing; */
+			if (res == qman_cb_dqrr_stop)
+				goto done;
+		}
+	}
+	/* Interpret 'dq' from a driver perspective. */
+	/* Parking isn't possible unless HELDACTIVE was set. NB,
+	 * FORCEELIGIBLE implies HELDACTIVE, so we only need to
+	 * check for HELDACTIVE to cover both. */
+	DPA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) ||
+		(res != qman_cb_dqrr_park));
+#ifdef CONFIG_FSL_QMAN_PORTAL_DISABLEAUTO_DCA
+	if (res == qman_cb_dqrr_park)
+		/* The only thing to do for non-DCA is the park-request */
+		qm_dqrr_park_current(&p->p);
+	/* Move forward */
+	qm_dqrr_next(&p->p);
+	qm_dqrr_cci_consume(&p->p, 1);
+#else
+	/* Defer just means "skip it, I'll consume it myself later on" */
+	if (res != qman_cb_dqrr_defer)
+		qm_dqrr_cdc_consume_1ptr(&p->p, dq, (res == qman_cb_dqrr_park));
+	/* Move forward */
+	qm_dqrr_next(&p->p);
+#endif
+	/* Entry processed and consumed, increment our counter. The callback can
+	 * request that we exit after consuming the entry, and we also exit if
+	 * we reach our processing limit, so loop back only if neither of these
+	 * conditions is met. */
+	if ((++limit < poll_limit) && (res != qman_cb_dqrr_consume_stop))
+		goto loop;
+done:
+	return limit;
+}
+
+u32 qman_irqsource_get(void)
+{
+	/* "irqsource" and "poll" APIs mustn't redirect when sharing, they
+	 * should shut the user out if they are not the primary CPU hosting the
+	 * portal. That's why we use the "raw" interface. */
+	struct qman_portal *p = get_raw_affine_portal();
+	u32 ret = p->irq_sources & QM_PIRQ_VISIBLE;
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_irqsource_get);
+
+int qman_irqsource_add(u32 bits __maybe_unused)
+{
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	struct qman_portal *p = get_raw_affine_portal();
+	int ret = 0;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (p->sharing_redirect)
+		ret = -EINVAL;
+	else
+#endif
+	{
+		__maybe_unused unsigned long irqflags;
+		PORTAL_IRQ_LOCK(p, irqflags);
+		set_bits(bits & QM_PIRQ_VISIBLE, &p->irq_sources);
+		qm_isr_enable_write(&p->p, p->irq_sources);
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+	}
+	put_affine_portal();
+	return ret;
+#else
+	pr_err("No Qman portal IRQ support, mustn't specify IRQ flags!");
+	return -EINVAL;
+#endif
+}
+EXPORT_SYMBOL(qman_irqsource_add);
+
+int qman_irqsource_remove(u32 bits)
+{
+#ifdef CONFIG_FSL_DPA_HAVE_IRQ
+	struct qman_portal *p = get_raw_affine_portal();
+	__maybe_unused unsigned long irqflags;
+	u32 ier;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (p->sharing_redirect) {
+		put_affine_portal();
+		return -EINVAL;
+	}
+#endif
+	/* Our interrupt handler only processes+clears status register bits that
+	 * are in p->irq_sources. As we're trimming that mask, if one of them
+	 * were to assert in the status register just before we remove it from
+	 * the enable register, there would be an interrupt-storm when we
+	 * release the IRQ lock. So we wait for the enable register update to
+	 * take effect in h/w (by reading it back) and then clear all other bits
+	 * in the status register. Ie. we clear them from ISR once it's certain
+	 * IER won't allow them to reassert. */
+	PORTAL_IRQ_LOCK(p, irqflags);
+	bits &= QM_PIRQ_VISIBLE;
+	clear_bits(bits, &p->irq_sources);
+	qm_isr_enable_write(&p->p, p->irq_sources);
+	ier = qm_isr_enable_read(&p->p);
+	/* Using "~ier" (rather than "bits" or "~p->irq_sources") creates a
+	 * data-dependency, ie. to protect against re-ordering. */
+	qm_isr_status_clear(&p->p, ~ier);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return 0;
+#else
+	pr_err("No Qman portal IRQ support, mustn't specify IRQ flags!");
+	return -EINVAL;
+#endif
+}
+EXPORT_SYMBOL(qman_irqsource_remove);
+
+const cpumask_t *qman_affine_cpus(void)
+{
+	return &affine_mask;
+}
+EXPORT_SYMBOL(qman_affine_cpus);
+
+int qman_poll_dqrr(unsigned int limit)
+{
+	/* We need to fail when called for a "slave", so use "raw" */
+	struct qman_portal *p = get_raw_affine_portal();
+	int ret;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (unlikely(p->sharing_redirect))
+		ret = -EINVAL;
+	else
+#endif
+	{
+		BUG_ON(p->irq_sources & QM_PIRQ_DQRI);
+		ret = __poll_portal_fast(p, limit);
+	}
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_poll_dqrr);
+
+u32 qman_poll_slow(void)
+{
+	/* We need to fail when called for a "slave", so use "raw" */
+	struct qman_portal *p = get_raw_affine_portal();
+	u32 ret;
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (unlikely(p->sharing_redirect))
+		ret = (u32)-1;
+	else
+#endif
+	{
+		u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources;
+		ret = __poll_portal_slow(p, is);
+		qm_isr_status_clear(&p->p, ret);
+	}
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_poll_slow);
+
+/* Legacy wrapper */
+void qman_poll(void)
+{
+	struct qman_portal *p = get_raw_affine_portal();
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+	if (unlikely(p->sharing_redirect))
+		goto done;
+#endif
+	if ((~p->irq_sources) & QM_PIRQ_SLOW) {
+		if (!(p->slowpoll--)) {
+			u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources;
+			u32 active = __poll_portal_slow(p, is);
+			if (active) {
+				qm_isr_status_clear(&p->p, active);
+				p->slowpoll = SLOW_POLL_BUSY;
+			} else
+				p->slowpoll = SLOW_POLL_IDLE;
+		}
+	}
+	if ((~p->irq_sources) & QM_PIRQ_DQRI)
+		__poll_portal_fast(p, CONFIG_FSL_QMAN_POLL_LIMIT);
+#ifdef CONFIG_FSL_DPA_PORTAL_SHARE
+done:
+#endif
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_poll);
+
+/* Recovery processing. */
+static int recovery_poll_mr(struct qman_portal *p, u32 fqid)
+{
+	const struct qm_mr_entry *msg;
+	enum {
+		wait_for_fqrn,
+		wait_for_fqrl,
+		done
+	} state = wait_for_fqrn;
+	u8 v, fqs = 0;
+
+loop:
+	qm_mr_pvb_update(&p->p);
+	msg = qm_mr_current(&p->p);
+	if (!msg) {
+		cpu_relax();
+		goto loop;
+	}
+	v = msg->verb & QM_MR_VERB_TYPE_MASK;
+	/* all MR messages have "fqid" in the same place */
+	if (msg->fq.fqid != fqid) {
+ignore_msg:
+		pr_warning("recovery_poll_mr(), ignoring msg 0x%02x for "
+			"FQID %d\n", v, msg->fq.fqid);
+		goto next_msg;
+	}
+	if (state == wait_for_fqrn) {
+		if ((v != QM_MR_VERB_FQRN) && (v != QM_MR_VERB_FQRNI))
+			goto ignore_msg;
+		fqs = msg->fq.fqs;
+		if (!(fqs & QM_MR_FQS_ORLPRESENT))
+			state = done;
+		else
+			state = wait_for_fqrl;
+	} else {
+		if (v != QM_MR_VERB_FQRL)
+			goto ignore_msg;
+		state = done;
+	}
+next_msg:
+	qm_mr_next(&p->p);
+	qm_mr_cci_consume(&p->p, 1);
+	if (state != done)
+		goto loop;
+	return (fqs & QM_MR_FQS_NOTEMPTY) ? 1 : 0;
+}
+static unsigned int recovery_poll_dqrr(struct qman_portal *p, u32 fqid)
+{
+	const struct qm_dqrr_entry *dq;
+	u8 empty = 0, num_fds = 0;
+
+loop:
+	qm_dqrr_pvb_update(&p->p, 0);
+	dq = qm_dqrr_current(&p->p);
+	if (!dq) {
+		cpu_relax();
+		goto loop;
+	}
+	if (!(dq->stat & QM_DQRR_STAT_UNSCHEDULED)) {
+ignore_dqrr:
+		pr_warning("recovery_poll_dqrr(), ignoring dqrr %02x:%02x "
+			"for FQID %d\n",
+			dq->verb & QM_DQRR_VERB_MASK, dq->stat, dq->fqid);
+		goto next_dqrr;
+	}
+	if (dq->fqid != fqid)
+		goto ignore_dqrr;
+	if (dq->stat & QM_DQRR_STAT_FD_VALID)
+		num_fds++;
+	if (dq->stat & QM_DQRR_STAT_FQ_EMPTY)
+		empty = 1;
+next_dqrr:
+	qm_dqrr_next(&p->p);
+	qm_dqrr_cci_consume(&p->p, 1);
+	if (!empty)
+		goto loop;
+	return num_fds;
+}
+int qman_recovery_cleanup_fq(u32 fqid)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	int ret = 0;
+	unsigned int num_fds = 0;
+	const char *s;
+	u8 state;
+
+	/* Lock this whole flow down via the portal's "vdqcr" */
+	PORTAL_IRQ_LOCK(p, irqflags);
+	BUG_ON(!(p->bits & PORTAL_BITS_RECOVERY));
+	if (p->vdqcr_owned)
+		ret = -EBUSY;
+	else
+		p->vdqcr_owned = (void *)1;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	if (ret)
+		goto out;
+
+	/* Query the FQ's state */
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq.fqid = fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP);
+	if (mcr->result != QM_MCR_RESULT_OK) {
+		ret = -EIO;
+		goto out;
+	}
+	state = mcr->queryfq_np.state & QM_MCR_NP_STATE_MASK;
+
+	/* OOS: nothing to do */
+	if (state == QM_MCR_NP_STATE_OOS)
+		goto out;
+	/* Otherwise: must be retired */
+	if (state != QM_MCR_NP_STATE_RETIRED) {
+		mcc = qm_mc_start(&p->p);
+		mcc->alterfq.fqid = fqid;
+		qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_RETIRE);
+		while (!(mcr = qm_mc_result(&p->p)))
+			cpu_relax();
+		DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) ==
+				QM_MCR_VERB_ALTER_RETIRE);
+		if ((mcr->result != QM_MCR_RESULT_OK) &&
+				(mcr->result != QM_MCR_RESULT_PENDING)) {
+			ret = -EIO;
+			goto out;
+		}
+		ret = recovery_poll_mr(p, fqid);
+		if (!ret)
+			/* FQ empty */
+			goto oos;
+	}
+	/* Drain till empty */
+	qm_dqrr_vdqcr_set(&p->p, fqid & 0x00ffffff);
+	num_fds = recovery_poll_dqrr(p, fqid);
+
+oos:
+	mcc = qm_mc_start(&p->p);
+	mcc->alterfq.fqid = fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_OOS);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_OOS);
+	if (mcr->result != QM_MCR_RESULT_OK)
+		ret = -EIO;
+	/* done */
+	s = (state == QM_MCR_NP_STATE_RETIRED) ? "retired" :
+		(state == QM_MCR_NP_STATE_PARKED) ? "parked" : "scheduled";
+	pr_info("Qman: %s FQID %d recovered (%d frames)\n", s, fqid, num_fds);
+out:
+	PORTAL_IRQ_LOCK(p, irqflags);
+	p->vdqcr_owned = NULL;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_recovery_cleanup_fq);
+
+/* called from qman_driver.c::qman_recovery_exit() only (if exporting, use
+ * get_raw_affine_portal() and check for the "SLAVE" bit). */
+void qman_recovery_exit_local(void)
+{
+	struct qman_portal *p = get_affine_portal();
+	BUG_ON(!(p->bits & PORTAL_BITS_RECOVERY));
+	/* Reinitialise DQRR using expected settings */
+	qm_dqrr_finish(&p->p);
+	post_recovery(p, p->config);
+	clear_bits(PORTAL_BITS_RECOVERY, &p->bits);
+	if (qm_dqrr_init(&p->p, p->config, qm_dqrr_dpush, qm_dqrr_pvb,
+			QM_DQRR_CMODE, DQRR_MAXFILL, 0))
+		panic("Qman DQRR initialisation failed, recovery broken");
+	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
+	qm_isr_status_clear(&p->p, 0xffffffff);
+	qm_isr_uninhibit(&p->p);
+	put_affine_portal();
+}
+
+void qman_stop_dequeues(void)
+{
+	struct qman_portal *p = get_affine_portal();
+	qman_stop_dequeues_ex(p);
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_stop_dequeues);
+
+void qman_start_dequeues(void)
+{
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	PORTAL_IRQ_LOCK(p, irqflags);
+	DPA_ASSERT(p->dqrr_disable_ref > 0);
+	if (!(--p->dqrr_disable_ref))
+		qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_start_dequeues);
+
+void qman_static_dequeue_add(u32 pools)
+{
+	unsigned long irqflags __maybe_unused;
+	struct qman_portal *p = get_affine_portal();
+	PORTAL_IRQ_LOCK(p, irqflags);
+	pools &= p->config->public_cfg.pools;
+	p->sdqcr |= pools;
+	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_static_dequeue_add);
+
+void qman_static_dequeue_del(u32 pools)
+{
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	PORTAL_IRQ_LOCK(p, irqflags);
+	pools &= p->config->public_cfg.pools;
+	p->sdqcr &= ~pools;
+	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_static_dequeue_del);
+
+u32 qman_static_dequeue_get(void)
+{
+	struct qman_portal *p = get_affine_portal();
+	u32 ret = p->sdqcr;
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_static_dequeue_get);
+
+void qman_dca(struct qm_dqrr_entry *dq, int park_request)
+{
+	struct qman_portal *p = get_affine_portal();
+	qm_dqrr_cdc_consume_1ptr(&p->p, dq, park_request);
+	put_affine_portal();
+}
+EXPORT_SYMBOL(qman_dca);
+
+/*******************/
+/* Frame queue API */
+/*******************/
+
+static const char *mcr_result_str(u8 result)
+{
+	switch (result) {
+	case QM_MCR_RESULT_NULL:
+		return "QM_MCR_RESULT_NULL";
+	case QM_MCR_RESULT_OK:
+		return "QM_MCR_RESULT_OK";
+	case QM_MCR_RESULT_ERR_FQID:
+		return "QM_MCR_RESULT_ERR_FQID";
+	case QM_MCR_RESULT_ERR_FQSTATE:
+		return "QM_MCR_RESULT_ERR_FQSTATE";
+	case QM_MCR_RESULT_ERR_NOTEMPTY:
+		return "QM_MCR_RESULT_ERR_NOTEMPTY";
+	case QM_MCR_RESULT_PENDING:
+		return "QM_MCR_RESULT_PENDING";
+	case QM_MCR_RESULT_ERR_BADCOMMAND:
+		return "QM_MCR_RESULT_ERR_BADCOMMAND";
+	}
+	return "<unknown MCR result>";
+}
+
+int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq)
+{
+	struct qm_fqd fqd;
+	struct qm_mcr_queryfq_np np;
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p;
+	unsigned long irqflags __maybe_unused;
+
+	if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) {
+		fqid = qm_fq_new();
+		if (!fqid)
+			return -ENOMEM;
+	}
+	spin_lock_init(&fq->fqlock);
+	fq->fqid = fqid;
+	fq->flags = flags;
+	fq->state = qman_fq_state_oos;
+	fq->cgr_groupid = 0;
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+	if (unlikely(find_empty_fq_table_entry(&fq->key, fq)))
+		return -ENOMEM;
+#endif
+	if (!(flags & QMAN_FQ_FLAG_AS_IS) || (flags & QMAN_FQ_FLAG_NO_MODIFY))
+		return 0;
+	/* Everything else is AS_IS support */
+	p = get_affine_portal();
+	PORTAL_IRQ_LOCK(p, irqflags);
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq.fqid = fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ);
+	if (mcr->result != QM_MCR_RESULT_OK) {
+		pr_err("QUERYFQ failed: %s\n", mcr_result_str(mcr->result));
+		goto err;
+	}
+	fqd = mcr->queryfq.fqd;
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq_np.fqid = fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ_NP);
+	if (mcr->result != QM_MCR_RESULT_OK) {
+		pr_err("QUERYFQ_NP failed: %s\n", mcr_result_str(mcr->result));
+		goto err;
+	}
+	np = mcr->queryfq_np;
+	/* Phew, have queryfq and queryfq_np results, stitch together
+	 * the FQ object from those. */
+	fq->cgr_groupid = fqd.cgid;
+	switch (np.state & QM_MCR_NP_STATE_MASK) {
+	case QM_MCR_NP_STATE_OOS:
+		break;
+	case QM_MCR_NP_STATE_RETIRED:
+		fq->state = qman_fq_state_retired;
+		if (np.frm_cnt)
+			fq_set(fq, QMAN_FQ_STATE_NE);
+		break;
+	case QM_MCR_NP_STATE_TEN_SCHED:
+	case QM_MCR_NP_STATE_TRU_SCHED:
+	case QM_MCR_NP_STATE_ACTIVE:
+		fq->state = qman_fq_state_sched;
+		if (np.state & QM_MCR_NP_STATE_R)
+			fq_set(fq, QMAN_FQ_STATE_CHANGING);
+		break;
+	case QM_MCR_NP_STATE_PARKED:
+		fq->state = qman_fq_state_parked;
+		break;
+	default:
+		DPA_ASSERT(NULL == "invalid FQ state");
+	}
+	if (fqd.fq_ctrl & QM_FQCTRL_CGE)
+		fq->state |= QMAN_FQ_STATE_CGR_EN;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return 0;
+err:
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID)
+		qm_fq_free(fqid);
+	return -EIO;
+}
+EXPORT_SYMBOL(qman_create_fq);
+
+void qman_destroy_fq(struct qman_fq *fq, u32 flags __maybe_unused)
+{
+	/* We don't need to lock the FQ as it is a pre-condition that the FQ be
+	 * quiesced. Instead, run some checks. */
+	switch (fq->state) {
+	case qman_fq_state_parked:
+		DPA_ASSERT(flags & QMAN_FQ_DESTROY_PARKED);
+	case qman_fq_state_oos:
+		if (fq_isset(fq, QMAN_FQ_FLAG_DYNAMIC_FQID))
+			qm_fq_free(fq->fqid);
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+		clear_fq_table_entry(fq->key);
+#endif
+		return;
+	default:
+		break;
+	}
+	DPA_ASSERT(NULL == "qman_free_fq() on unquiesced FQ!");
+}
+EXPORT_SYMBOL(qman_destroy_fq);
+
+u32 qman_fq_fqid(struct qman_fq *fq)
+{
+	return fq->fqid;
+}
+EXPORT_SYMBOL(qman_fq_fqid);
+
+void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags)
+{
+	if (state)
+		*state = fq->state;
+	if (flags)
+		*flags = fq->flags;
+}
+EXPORT_SYMBOL(qman_fq_state);
+
+int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p;
+	unsigned long irqflags __maybe_unused;
+	u8 res, myverb = (flags & QMAN_INITFQ_FLAG_SCHED) ?
+		QM_MCC_VERB_INITFQ_SCHED : QM_MCC_VERB_INITFQ_PARKED;
+
+	DPA_ASSERT((fq->state == qman_fq_state_oos) ||
+		(fq->state == qman_fq_state_parked));
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY)))
+		return -EINVAL;
+#endif
+	if (opts && (opts->we_mask & QM_INITFQ_WE_OAC)) {
+		/* OAC not supported on rev1.0 */
+		if (unlikely(qman_ip_rev == QMAN_REV10))
+			return -EINVAL;
+		/* And can't be set at the same time as TDTHRESH */
+		if (opts->we_mask & QM_INITFQ_WE_TDTHRESH)
+			return -EINVAL;
+	}
+	/* Issue an INITFQ_[PARKED|SCHED] management command */
+	p = get_affine_portal();
+	PORTAL_IRQ_LOCK(p, irqflags);
+	FQLOCK(fq);
+	if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) ||
+			((fq->state != qman_fq_state_oos) &&
+				(fq->state != qman_fq_state_parked)))) {
+		FQUNLOCK(fq);
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+		put_affine_portal();
+		return -EBUSY;
+	}
+	mcc = qm_mc_start(&p->p);
+	if (opts)
+		mcc->initfq = *opts;
+	mcc->initfq.fqid = fq->fqid;
+	mcc->initfq.count = 0;
+	/* If INITFQ_FLAG_NULL is passed, contextB is set to zero. Otherwise,
+	 * if the FQ does *not* have the TO_DCPORTAL flag, contextB is set as a
+	 * demux pointer. Otherwise, TO_DCPORTAL is set, so the caller-provided
+	 * value is allowed to stand, don't overwrite it. */
+	if ((flags & QMAN_INITFQ_FLAG_NULL) ||
+			fq_isclear(fq, QMAN_FQ_FLAG_TO_DCPORTAL)) {
+		dma_addr_t phys_fq;
+		mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTB;
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+		mcc->initfq.fqd.context_b = (flags & QMAN_INITFQ_FLAG_NULL) ?
+						0 : fq->key;
+#else
+		mcc->initfq.fqd.context_b = (flags & QMAN_INITFQ_FLAG_NULL) ?
+						0 : (u32)(uintptr_t)fq;
+#endif
+		/* and the physical address - NB, if the user wasn't trying to
+		 * set CONTEXTA, clear the stashing settings. */
+		if (!(mcc->initfq.we_mask & QM_INITFQ_WE_CONTEXTA)) {
+			mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTA;
+			memset(&mcc->initfq.fqd.context_a, 0,
+				sizeof(&mcc->initfq.fqd.context_a));
+		} else {
+			phys_fq = dma_map_single(&p->pdev->dev, fq, sizeof(*fq),
+						DMA_TO_DEVICE);
+			qm_fqd_stashing_set64(&mcc->initfq.fqd, phys_fq);
+		}
+	}
+	if (flags & QMAN_INITFQ_FLAG_LOCAL) {
+		mcc->initfq.fqd.dest.channel = p->config->public_cfg.channel;
+		if (!(mcc->initfq.we_mask & QM_INITFQ_WE_DESTWQ)) {
+			mcc->initfq.we_mask |= QM_INITFQ_WE_DESTWQ;
+			mcc->initfq.fqd.dest.wq = 4;
+		}
+	}
+	qm_mc_commit(&p->p, myverb);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb);
+	res = mcr->result;
+	if (res != QM_MCR_RESULT_OK) {
+		FQUNLOCK(fq);
+		PORTAL_IRQ_UNLOCK(p, irqflags);
+		put_affine_portal();
+		return -EIO;
+	}
+	if (opts) {
+		if (opts->we_mask & QM_INITFQ_WE_FQCTRL) {
+			if (opts->fqd.fq_ctrl & QM_FQCTRL_CGE)
+				fq_set(fq, QMAN_FQ_STATE_CGR_EN);
+			else
+				fq_clear(fq, QMAN_FQ_STATE_CGR_EN);
+		}
+		if (opts->we_mask & QM_INITFQ_WE_CGID)
+			fq->cgr_groupid = opts->fqd.cgid;
+	}
+	fq->state = (flags & QMAN_INITFQ_FLAG_SCHED) ?
+			qman_fq_state_sched : qman_fq_state_parked;
+	FQUNLOCK(fq);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return 0;
+}
+EXPORT_SYMBOL(qman_init_fq);
+
+int qman_schedule_fq(struct qman_fq *fq)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p;
+	unsigned long irqflags __maybe_unused;
+	int ret = 0;
+	u8 res;
+
+	DPA_ASSERT(fq->state == qman_fq_state_parked);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY)))
+		return -EINVAL;
+#endif
+	/* Issue a ALTERFQ_SCHED management command */
+	p = get_affine_portal();
+	PORTAL_IRQ_LOCK(p, irqflags);
+	FQLOCK(fq);
+	if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) ||
+			(fq->state != qman_fq_state_parked))) {
+		ret = -EBUSY;
+		goto out;
+	}
+	mcc = qm_mc_start(&p->p);
+	mcc->alterfq.fqid = fq->fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_SCHED);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_SCHED);
+	res = mcr->result;
+	if (res != QM_MCR_RESULT_OK) {
+		ret = -EIO;
+		goto out;
+	}
+	fq->state = qman_fq_state_sched;
+out:
+	FQUNLOCK(fq);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_schedule_fq);
+
+int qman_retire_fq(struct qman_fq *fq, u32 *flags)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p;
+	unsigned long irqflags __maybe_unused;
+	int rval;
+	u8 res;
+
+	DPA_ASSERT((fq->state == qman_fq_state_parked) ||
+		(fq->state == qman_fq_state_sched));
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY)))
+		return -EINVAL;
+#endif
+	p = get_affine_portal();
+	PORTAL_IRQ_LOCK(p, irqflags);
+	FQLOCK(fq);
+	if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) ||
+			(fq->state == qman_fq_state_retired) ||
+				(fq->state == qman_fq_state_oos))) {
+		rval = -EBUSY;
+		goto out;
+	}
+	rval = table_push_fq(p, fq);
+	if (rval)
+		goto out;
+	mcc = qm_mc_start(&p->p);
+	mcc->alterfq.fqid = fq->fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_RETIRE);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_RETIRE);
+	res = mcr->result;
+	/* "Elegant" would be to treat OK/PENDING the same way; set CHANGING,
+	 * and defer the flags until FQRNI or FQRN (respectively) show up. But
+	 * "Friendly" is to process OK immediately, and not set CHANGING. We do
+	 * friendly, otherwise the caller doesn't necessarily have a fully
+	 * "retired" FQ on return even if the retirement was immediate. However
+	 * this does mean some code duplication between here and
+	 * fq_state_change(). */
+	if (likely(res == QM_MCR_RESULT_OK)) {
+		rval = 0;
+		/* Process 'fq' right away, we'll ignore FQRNI */
+		if (mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY)
+			fq_set(fq, QMAN_FQ_STATE_NE);
+		if (mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT)
+			fq_set(fq, QMAN_FQ_STATE_ORL);
+		else
+			table_del_fq(p, fq);
+		if (flags)
+			*flags = fq->flags;
+		fq->state = qman_fq_state_retired;
+		if (fq->cb.fqs) {
+			/* Another issue with supporting "immediate" retirement
+			 * is that we're forced to drop FQRNIs, because by the
+			 * time they're seen it may already be "too late" (the
+			 * fq may have been OOS'd and free()'d already). But if
+			 * the upper layer wants a callback whether it's
+			 * immediate or not, we have to fake a "MR" entry to
+			 * look like an FQRNI... */
+			struct qm_mr_entry msg;
+			msg.verb = QM_MR_VERB_FQRNI;
+			msg.fq.fqs = mcr->alterfq.fqs;
+			msg.fq.fqid = fq->fqid;
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+			msg.fq.contextB = fq->key;
+#else
+			msg.fq.contextB = (u32)(uintptr_t)fq;
+#endif
+			fq->cb.fqs(p, fq, &msg);
+		}
+	} else if (res == QM_MCR_RESULT_PENDING) {
+		rval = 1;
+		fq_set(fq, QMAN_FQ_STATE_CHANGING);
+	} else {
+		rval = -EIO;
+		table_del_fq(p, fq);
+	}
+out:
+	FQUNLOCK(fq);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return rval;
+}
+EXPORT_SYMBOL(qman_retire_fq);
+
+int qman_oos_fq(struct qman_fq *fq)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p;
+	unsigned long irqflags __maybe_unused;
+	int ret = 0;
+	u8 res;
+
+	DPA_ASSERT(fq->state == qman_fq_state_retired);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY)))
+		return -EINVAL;
+#endif
+	p = get_affine_portal();
+	PORTAL_IRQ_LOCK(p, irqflags);
+	FQLOCK(fq);
+	if (unlikely((fq_isset(fq, QMAN_FQ_STATE_BLOCKOOS)) ||
+			(fq->state != qman_fq_state_retired))) {
+		ret = -EBUSY;
+		goto out;
+	}
+	mcc = qm_mc_start(&p->p);
+	mcc->alterfq.fqid = fq->fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_OOS);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_OOS);
+	res = mcr->result;
+	if (res != QM_MCR_RESULT_OK) {
+		ret = -EIO;
+		goto out;
+	}
+	fq->state = qman_fq_state_oos;
+out:
+	FQUNLOCK(fq);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_oos_fq);
+
+int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	u8 res;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq.fqid = fq->fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ);
+	res = mcr->result;
+	if (res == QM_MCR_RESULT_OK)
+		*fqd = mcr->queryfq.fqd;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (res != QM_MCR_RESULT_OK)
+		return -EIO;
+	return 0;
+}
+EXPORT_SYMBOL(qman_query_fq);
+
+int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	u8 res;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq.fqid = fq->fqid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP);
+	res = mcr->result;
+	if (res == QM_MCR_RESULT_OK)
+		*np = mcr->queryfq_np;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (res != QM_MCR_RESULT_OK)
+		return -EIO;
+	return 0;
+}
+EXPORT_SYMBOL(qman_query_fq_np);
+
+int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	u8 res, myverb;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	myverb = (query_dedicated) ? QM_MCR_VERB_QUERYWQ_DEDICATED :
+				 QM_MCR_VERB_QUERYWQ;
+	mcc = qm_mc_start(&p->p);
+	mcc->querywq.channel.id = wq->channel.id;
+	qm_mc_commit(&p->p, myverb);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb);
+	res = mcr->result;
+	if (res == QM_MCR_RESULT_OK)
+		*wq = mcr->querywq;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (res != QM_MCR_RESULT_OK) {
+		pr_err("QUERYWQ failed: %s\n", mcr_result_str(res));
+		return -EIO;
+	}
+	return 0;
+}
+EXPORT_SYMBOL(qman_query_wq);
+
+int qman_testwrite_cgr(struct qman_cgr *cgr, u64 i_bcnt,
+			struct qm_mcr_cgrtestwrite *result)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	u8 res;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	mcc = qm_mc_start(&p->p);
+	mcc->cgrtestwrite.cgid = cgr->cgrid;
+	mcc->cgrtestwrite.i_bcnt_hi = (u8)(i_bcnt >> 32);
+	mcc->cgrtestwrite.i_bcnt_lo = (u32)i_bcnt;
+	qm_mc_commit(&p->p, QM_MCC_VERB_CGRTESTWRITE);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_CGRTESTWRITE);
+	res = mcr->result;
+	if (res == QM_MCR_RESULT_OK)
+		*result = mcr->cgrtestwrite;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (res != QM_MCR_RESULT_OK) {
+		pr_err("CGR TEST WRITE failed: %s\n", mcr_result_str(res));
+		return -EIO;
+	}
+	return 0;
+}
+EXPORT_SYMBOL(qman_testwrite_cgr);
+
+int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *cgrd)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	u8 res;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	mcc = qm_mc_start(&p->p);
+	mcc->querycgr.cgid = cgr->cgrid;
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCGR);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYCGR);
+	res = mcr->result;
+	if (res == QM_MCR_RESULT_OK)
+		*cgrd = mcr->querycgr;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (res != QM_MCR_RESULT_OK) {
+		pr_err("QUERY_CGR failed: %s\n", mcr_result_str(res));
+		return -EIO;
+	}
+	return 0;
+}
+EXPORT_SYMBOL(qman_query_cgr);
+
+int qman_query_congestion(struct qm_mcr_querycongestion *congestion)
+{
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	u8 res;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	qm_mc_start(&p->p);
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) ==
+			QM_MCC_VERB_QUERYCONGESTION);
+	res = mcr->result;
+	if (res == QM_MCR_RESULT_OK)
+		*congestion = mcr->querycongestion;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	if (res != QM_MCR_RESULT_OK) {
+		pr_err("QUERY_CONGESTION failed: %s\n", mcr_result_str(res));
+		return -EIO;
+	}
+	return 0;
+}
+EXPORT_SYMBOL(qman_query_congestion);
+
+/* internal function used as a wait_event() expression */
+static int set_vdqcr(struct qman_portal **p, struct qman_fq *fq, u32 vdqcr)
+{
+	unsigned long irqflags __maybe_unused;
+	int ret = -EBUSY;
+	*p = get_affine_portal();
+	PORTAL_IRQ_LOCK(*p, irqflags);
+	if (!(*p)->vdqcr_owned) {
+		FQLOCK(fq);
+		if (fq_isset(fq, QMAN_FQ_STATE_VDQCR))
+			goto escape;
+		fq_set(fq, QMAN_FQ_STATE_VDQCR);
+		FQUNLOCK(fq);
+		(*p)->vdqcr_owned = fq;
+		ret = 0;
+	}
+escape:
+	PORTAL_IRQ_UNLOCK(*p, irqflags);
+	if (!ret)
+		qm_dqrr_vdqcr_set(&(*p)->p, vdqcr);
+	put_affine_portal();
+	return ret;
+}
+
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+static int wait_vdqcr_start(struct qman_portal **p, struct qman_fq *fq,
+				u32 vdqcr, u32 flags)
+{
+	int ret = 0;
+	if (flags & QMAN_VOLATILE_FLAG_WAIT_INT)
+		ret = wait_event_interruptible(affine_queue,
+				!(ret = set_vdqcr(p, fq, vdqcr)));
+	else
+		wait_event(affine_queue, !(ret = set_vdqcr(p, fq, vdqcr)));
+	return ret;
+}
+#endif
+
+int qman_volatile_dequeue(struct qman_fq *fq, u32 flags __maybe_unused,
+				u32 vdqcr)
+{
+	struct qman_portal *p;
+	int ret;
+
+	DPA_ASSERT(!fq || (fq->state == qman_fq_state_parked) ||
+			(fq->state == qman_fq_state_retired));
+	DPA_ASSERT(!fq || !(vdqcr & QM_VDQCR_FQID_MASK));
+	DPA_ASSERT(!fq || !fq_isset(fq, QMAN_FQ_STATE_VDQCR));
+	if (fq)
+		vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid;
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+	if (flags & QMAN_VOLATILE_FLAG_WAIT)
+		ret = wait_vdqcr_start(&p, fq, vdqcr, flags);
+	else
+#endif
+		ret = set_vdqcr(&p, fq, vdqcr);
+	if (ret)
+		return ret;
+	/* VDQCR is set */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+	if (flags & QMAN_VOLATILE_FLAG_FINISH) {
+		if (flags & QMAN_VOLATILE_FLAG_WAIT_INT)
+			/* NB: don't propagate any error - the caller wouldn't
+			 * know whether the VDQCR was issued or not. A signal
+			 * could arrive after returning anyway, so the caller
+			 * can check signal_pending() if that's an issue. */
+			wait_event_interruptible(affine_queue,
+				!fq_isset(fq, QMAN_FQ_STATE_VDQCR));
+		else
+			wait_event(affine_queue,
+				!fq_isset(fq, QMAN_FQ_STATE_VDQCR));
+	}
+#endif
+	return 0;
+}
+EXPORT_SYMBOL(qman_volatile_dequeue);
+
+static noinline void update_eqcr_ci(struct qman_portal *p, u8 avail)
+{
+	if (avail)
+		qm_eqcr_cce_prefetch(&p->p);
+	else
+		qm_eqcr_cce_update(&p->p);
+}
+
+int qman_eqcr_is_empty(void)
+{
+	unsigned long irqflags __maybe_unused;
+	struct qman_portal *p = get_affine_portal();
+	u8 avail;
+
+	PORTAL_IRQ_LOCK(p, irqflags);
+	update_eqcr_ci(p, 0);
+	avail = qm_eqcr_get_fill(&p->p);
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return (avail == 0);
+}
+EXPORT_SYMBOL(qman_eqcr_is_empty);
+
+static inline struct qm_eqcr_entry *try_eq_start(struct qman_portal **p,
+					unsigned long *irqflags __maybe_unused,
+					struct qman_fq *fq,
+					const struct qm_fd *fd,
+					u32 flags)
+{
+	struct qm_eqcr_entry *eq;
+	u8 avail;
+
+	*p = get_affine_portal();
+	PORTAL_IRQ_LOCK(*p, (*irqflags));
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) &&
+			(flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) {
+		if ((*p)->eqci_owned) {
+			PORTAL_IRQ_UNLOCK(*p, (*irqflags));
+			put_affine_portal();
+			return NULL;
+		}
+		(*p)->eqci_owned = fq;
+	}
+#endif
+	avail = qm_eqcr_get_avail(&(*p)->p);
+	if (avail < 2)
+		update_eqcr_ci(*p, avail);
+	eq = qm_eqcr_start(&(*p)->p);
+	if (unlikely(!eq)) {
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+		if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) &&
+				(flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC)))
+			(*p)->eqci_owned = NULL;
+#endif
+		PORTAL_IRQ_UNLOCK(*p, (*irqflags));
+		put_affine_portal();
+		return NULL;
+	}
+	if (flags & QMAN_ENQUEUE_FLAG_DCA)
+		eq->dca = QM_EQCR_DCA_ENABLE |
+			((flags & QMAN_ENQUEUE_FLAG_DCA_PARK) ?
+					QM_EQCR_DCA_PARK : 0) |
+			((flags >> 8) & QM_EQCR_DCA_IDXMASK);
+	eq->fqid = fq->fqid;
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+	eq->tag = fq->key;
+#else
+	eq->tag = (u32)(uintptr_t)fq;
+#endif
+	/* From p4080 rev1 -> rev2, the FD struct's address went from 48-bit to
+	 * 40-bit but rev1 chips will still interpret it as 48-bit, meaning we
+	 * have to scrub the upper 8-bits, just in case the user left noise in
+	 * there. Doing this selectively via a run-time check of the h/w
+	 * revision (as we do for most errata, for example) is too slow in this
+	 * critical path code. The most inexpensive way to handle this is just
+	 * to reinterpret the FD as 4 32-bit words and to mask the first word
+	 * appropriately, irrespecitive of the h/w revision. The struct fields
+	 * corresponding to this word are;
+	 *     u8 dd:2;
+	 *     u8 liodn_offset:6;
+	 *     u8 bpid;
+	 *     u8 eliodn_offset:4;
+	 *     u8 __reserved:4;
+	 *     u8 addr_hi;
+	 * So we mask this word with 0xc0ff00ff, which implicitly scrubs out
+	 * liodn_offset, eliodn_offset, and __reserved - the latter two fields
+	 * are interpreted as the 8 msbits of the 48-bit address in the case of
+	 * rev1.
+	 */
+	{
+		const u32 *src = (const u32 *)fd;
+		u32 *dest = (u32 *)&eq->fd;
+		dest[0] = src[0] & 0xc0ff00ff;
+		dest[1] = src[1];
+		dest[2] = src[2];
+		dest[3] = src[3];
+	}
+	return eq;
+}
+
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+static noinline struct qm_eqcr_entry *__wait_eq_start(struct qman_portal **p,
+					unsigned long *irqflags __maybe_unused,
+					struct qman_fq *fq,
+					const struct qm_fd *fd,
+					u32 flags)
+{
+	struct qm_eqcr_entry *eq = try_eq_start(p, irqflags, fq, fd, flags);
+	if (!eq)
+		qm_eqcr_set_ithresh(&(*p)->p, EQCR_ITHRESH);
+	return eq;
+}
+static noinline struct qm_eqcr_entry *wait_eq_start(struct qman_portal **p,
+					unsigned long *irqflags __maybe_unused,
+					struct qman_fq *fq,
+					const struct qm_fd *fd,
+					u32 flags)
+{
+	struct qm_eqcr_entry *eq;
+	if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT)
+		wait_event_interruptible(affine_queue,
+			(eq = __wait_eq_start(p, irqflags, fq, fd, flags)));
+	else
+		wait_event(affine_queue,
+			(eq = __wait_eq_start(p, irqflags, fq, fd, flags)));
+	return eq;
+}
+#endif
+
+int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags)
+{
+	struct qman_portal *p;
+	struct qm_eqcr_entry *eq;
+	unsigned long irqflags __maybe_unused;
+
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+	if (flags & QMAN_ENQUEUE_FLAG_WAIT)
+		eq = wait_eq_start(&p, &irqflags, fq, fd, flags);
+	else
+#endif
+	eq = try_eq_start(&p, &irqflags, fq, fd, flags);
+	if (!eq)
+		return -EBUSY;
+	/* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */
+	qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE |
+		(flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT)));
+	/* Factor the below out, it's used from qman_enqueue_orp() too */
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) &&
+			(flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) {
+		if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT)
+			wait_event_interruptible(affine_queue,
+					(p->eqci_owned != fq));
+		else
+			wait_event(affine_queue, (p->eqci_owned != fq));
+	}
+#endif
+	return 0;
+}
+EXPORT_SYMBOL(qman_enqueue);
+
+int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags,
+			struct qman_fq *orp, u16 orp_seqnum)
+{
+	struct qman_portal *p;
+	struct qm_eqcr_entry *eq;
+	unsigned long irqflags __maybe_unused;
+
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+	if (flags & QMAN_ENQUEUE_FLAG_WAIT)
+		eq = wait_eq_start(&p, &irqflags, fq, fd, flags);
+	else
+#endif
+	eq = try_eq_start(&p, &irqflags, fq, fd, flags);
+	if (!eq)
+		return -EBUSY;
+	/* Process ORP-specifics here */
+	if (flags & QMAN_ENQUEUE_FLAG_NLIS)
+		orp_seqnum |= QM_EQCR_SEQNUM_NLIS;
+	else {
+		orp_seqnum &= ~QM_EQCR_SEQNUM_NLIS;
+		if (flags & QMAN_ENQUEUE_FLAG_NESN)
+			orp_seqnum |= QM_EQCR_SEQNUM_NESN;
+		else
+			/* No need to check 4 QMAN_ENQUEUE_FLAG_HOLE */
+			orp_seqnum &= ~QM_EQCR_SEQNUM_NESN;
+	}
+	eq->seqnum = orp_seqnum;
+	eq->orp = orp->fqid;
+	/* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */
+	qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_ORP |
+		((flags & (QMAN_ENQUEUE_FLAG_HOLE | QMAN_ENQUEUE_FLAG_NESN)) ?
+				0 : QM_EQCR_VERB_CMD_ENQUEUE) |
+		(flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT)));
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+	if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) &&
+			(flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) {
+		if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT)
+			wait_event_interruptible(affine_queue,
+					(p->eqci_owned != fq));
+		else
+			wait_event(affine_queue, (p->eqci_owned != fq));
+	}
+#endif
+	return 0;
+}
+EXPORT_SYMBOL(qman_enqueue_orp);
+
+int qman_modify_cgr(struct qman_cgr *cgr, u32 flags,
+			struct qm_mcc_initcgr *opts)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+	unsigned long irqflags __maybe_unused;
+	u8 res;
+	u8 verb = QM_MCC_VERB_MODIFYCGR;
+
+	/* frame mode not supported on rev1.0 */
+	if (unlikely(qman_ip_rev == QMAN_REV10)) {
+		if (opts && (opts->we_mask & QM_CGR_WE_MODE) &&
+				opts->cgr.mode == QMAN_CGR_MODE_FRAME) {
+			put_affine_portal();
+			return -EIO;
+		}
+	}
+	PORTAL_IRQ_LOCK(p, irqflags);
+	mcc = qm_mc_start(&p->p);
+	if (opts)
+		mcc->initcgr = *opts;
+	mcc->initcgr.cgid = cgr->cgrid;
+	if (flags & QMAN_CGR_FLAG_USE_INIT)
+		verb = QM_MCC_VERB_INITCGR;
+	qm_mc_commit(&p->p, verb);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == verb);
+	res = mcr->result;
+	PORTAL_IRQ_UNLOCK(p, irqflags);
+	put_affine_portal();
+	return (res == QM_MCR_RESULT_OK) ? 0 : -EIO;
+}
+EXPORT_SYMBOL(qman_modify_cgr);
+
+#define TARG_MASK(n) (0x80000000 >> (n->config->public_cfg.channel - \
+					qm_channel_swportal0))
+
+int qman_create_cgr(struct qman_cgr *cgr, u32 flags,
+			struct qm_mcc_initcgr *opts)
+{
+	unsigned long irqflags __maybe_unused;
+	struct qm_mcr_querycgr cgr_state;
+	struct qm_mcc_initcgr local_opts;
+	int ret;
+	struct qman_portal *p;
+
+	/* We have to check that the provided CGRID is within the limits of the
+	 * data-structures, for obvious reasons. However we'll let h/w take
+	 * care of determining whether it's within the limits of what exists on
+	 * the SoC. */
+	if (cgr->cgrid >= __CGR_NUM)
+		return -EINVAL;
+
+	p = get_affine_portal();
+
+	memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr));
+	cgr->chan = p->config->public_cfg.channel;
+	spin_lock_irqsave(&p->cgr_lock, irqflags);
+
+	/* if no opts specified and I'm not the first for this portal, just add
+	 * to the list */
+	if ((opts == NULL) && !list_empty(&p->cgr_cbs[cgr->cgrid]))
+		goto add_list;
+
+	ret = qman_query_cgr(cgr, &cgr_state);
+	if (ret)
+		goto release_lock;
+	if (opts)
+		local_opts = *opts;
+	/* Overwrite TARG */
+	local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ | TARG_MASK(p);
+	local_opts.we_mask |= QM_CGR_WE_CSCN_TARG;
+
+	/* send init if flags indicate so */
+	if (opts && (flags & QMAN_CGR_FLAG_USE_INIT))
+		ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, &local_opts);
+	else
+		ret = qman_modify_cgr(cgr, 0, &local_opts);
+	if (ret)
+		goto release_lock;
+add_list:
+	list_add(&cgr->node, &p->cgr_cbs[cgr->cgrid]);
+
+	/* Determine if newly added object requires its callback to be called */
+	ret = qman_query_cgr(cgr, &cgr_state);
+	if (ret) {
+		/* we can't go back, so proceed and return success, but screen
+		 * and wail to the log file */
+		pr_crit("CGR HW state partially modified\n");
+		ret = 0;
+		goto release_lock;
+	}
+	if (cgr->cb && cgr_state.cgr.cscn_en && qman_cgrs_get(&p->cgrs[1],
+							cgr->cgrid))
+		cgr->cb(p, cgr, 1);
+release_lock:
+	spin_unlock_irqrestore(&p->cgr_lock, irqflags);
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_create_cgr);
+
+int qman_delete_cgr(struct qman_cgr *cgr)
+{
+	unsigned long irqflags __maybe_unused;
+	struct qm_mcr_querycgr cgr_state;
+	struct qm_mcc_initcgr local_opts;
+	int ret = 0;
+	struct qman_portal *p = get_affine_portal();
+
+	if (cgr->chan != p->config->public_cfg.channel) {
+		pr_crit("Attempting to delete cgr from different portal "
+			"than it was create: create 0x%x, delete 0x%x\n",
+			cgr->chan, p->config->public_cfg.channel);
+		ret = -EINVAL;
+		goto put_portal;
+	}
+	memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr));
+	spin_lock_irqsave(&p->cgr_lock, irqflags);
+	list_del(&cgr->node);
+	/* If last in list, CSCN_TARG must be set accordingly */
+	if (!list_empty(&p->cgr_cbs[cgr->cgrid]))
+		goto release_lock;
+	ret = qman_query_cgr(cgr, &cgr_state);
+	if (ret)  {
+		/* add back to the list */
+		list_add(&cgr->node, &p->cgr_cbs[cgr->cgrid]);
+		goto release_lock;
+	}
+	/* Overwrite TARG */
+	local_opts.we_mask = QM_CGR_WE_CSCN_TARG;
+	local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ & ~(TARG_MASK(p));
+	ret = qman_modify_cgr(cgr, 0, &local_opts);
+	if (ret)
+		/* add back to the list */
+		list_add(&cgr->node, &p->cgr_cbs[cgr->cgrid]);
+release_lock:
+	spin_unlock_irqrestore(&p->cgr_lock, irqflags);
+put_portal:
+	put_affine_portal();
+	return ret;
+}
+EXPORT_SYMBOL(qman_delete_cgr);
+
diff --git a/drivers/staging/fsl_qbman/qman_low.h b/drivers/staging/fsl_qbman/qman_low.h
new file mode 100644
index 0000000..6caca90
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_low.h
@@ -0,0 +1,1186 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_private.h"
+
+/***************************/
+/* Portal register assists */
+/***************************/
+
+/* Cache-inhibited register offsets */
+#define REG_EQCR_PI_CINH	0x0000
+#define REG_EQCR_CI_CINH	0x0004
+#define REG_EQCR_ITR		0x0008
+#define REG_DQRR_PI_CINH	0x0040
+#define REG_DQRR_CI_CINH	0x0044
+#define REG_DQRR_ITR		0x0048
+#define REG_DQRR_DCAP		0x0050
+#define REG_DQRR_SDQCR		0x0054
+#define REG_DQRR_VDQCR		0x0058
+#define REG_DQRR_PDQCR		0x005c
+#define REG_MR_PI_CINH		0x0080
+#define REG_MR_CI_CINH		0x0084
+#define REG_MR_ITR		0x0088
+#define REG_CFG			0x0100
+#define REG_ISR			0x0e00
+#define REG_ITPR		0x0e14
+
+/* Cache-enabled register offsets */
+#define CL_EQCR			0x0000
+#define CL_DQRR			0x1000
+#define CL_MR			0x2000
+#define CL_EQCR_PI_CENA		0x3000
+#define CL_EQCR_CI_CENA		0x3100
+#define CL_DQRR_PI_CENA		0x3200
+#define CL_DQRR_CI_CENA		0x3300
+#define CL_MR_PI_CENA		0x3400
+#define CL_MR_CI_CENA		0x3500
+#define CL_CR			0x3800
+#define CL_RR0			0x3900
+#define CL_RR1			0x3940
+
+/* BTW, the drivers (and h/w programming model) already obtain the required
+ * synchronisation for portal accesses via lwsync(), hwsync(), and
+ * data-dependencies. Use of barrier()s or other order-preserving primitives
+ * simply degrade performance. Hence the use of the __raw_*() interfaces, which
+ * simply ensure that the compiler treats the portal registers as volatile (ie.
+ * non-coherent). */
+
+/* Cache-inhibited register access. */
+#define __qm_in(qm, o)		__raw_readl((qm)->addr_ci + (o))
+#define __qm_out(qm, o, val)	__raw_writel((val), (qm)->addr_ci + (o))
+#define qm_in(reg)		__qm_in(&portal->addr, REG_##reg)
+#define qm_out(reg, val)	__qm_out(&portal->addr, REG_##reg, val)
+
+/* Cache-enabled (index) register access */
+#define __qm_cl_touch_ro(qm, o) dcbt_ro((qm)->addr_ce + (o))
+#define __qm_cl_touch_rw(qm, o) dcbt_rw((qm)->addr_ce + (o))
+#define __qm_cl_in(qm, o)	__raw_readl((qm)->addr_ce + (o))
+#define __qm_cl_out(qm, o, val) \
+	do { \
+		u32 *__tmpclout = (qm)->addr_ce + (o); \
+		__raw_writel((val), __tmpclout); \
+		dcbf(__tmpclout); \
+	} while (0)
+#define __qm_cl_invalidate(qm, o) dcbi((qm)->addr_ce + (o))
+#define qm_cl_touch_ro(reg)	__qm_cl_touch_ro(&portal->addr, CL_##reg##_CENA)
+#define qm_cl_touch_rw(reg)	__qm_cl_touch_rw(&portal->addr, CL_##reg##_CENA)
+#define qm_cl_in(reg)		__qm_cl_in(&portal->addr, CL_##reg##_CENA)
+#define qm_cl_out(reg, val)	__qm_cl_out(&portal->addr, CL_##reg##_CENA, val)
+#define qm_cl_invalidate(reg) __qm_cl_invalidate(&portal->addr, CL_##reg##_CENA)
+
+/* Cache-enabled ring access */
+#define qm_cl(base, idx)	((void *)base + ((idx) << 6))
+
+/* Cyclic helper for rings. FIXME: once we are able to do fine-grain perf
+ * analysis, look at using the "extra" bit in the ring index registers to avoid
+ * cyclic issues. */
+static inline u8 cyc_diff(u8 ringsize, u8 first, u8 last)
+{
+	/* 'first' is included, 'last' is excluded */
+	if (first <= last)
+		return last - first;
+	return ringsize + last - first;
+}
+
+/* Portal modes.
+ *   Enum types;
+ *     pmode == production mode
+ *     cmode == consumption mode,
+ *     dmode == h/w dequeue mode.
+ *   Enum values use 3 letter codes. First letter matches the portal mode,
+ *   remaining two letters indicate;
+ *     ci == cache-inhibited portal register
+ *     ce == cache-enabled portal register
+ *     vb == in-band valid-bit (cache-enabled)
+ *     dc == DCA (Discrete Consumption Acknowledgement), DQRR-only
+ *   As for "enum qm_dqrr_dmode", it should be self-explanatory.
+ */
+enum qm_eqcr_pmode {		/* matches QCSP_CFG::EPM */
+	qm_eqcr_pci = 0,	/* PI index, cache-inhibited */
+	qm_eqcr_pce = 1,	/* PI index, cache-enabled */
+	qm_eqcr_pvb = 2		/* valid-bit */
+};
+enum qm_eqcr_cmode {		/* s/w-only */
+	qm_eqcr_cci,		/* CI index, cache-inhibited */
+	qm_eqcr_cce		/* CI index, cache-enabled */
+};
+enum qm_dqrr_dmode {		/* matches QCSP_CFG::DP */
+	qm_dqrr_dpush = 0,	/* SDQCR  + VDQCR */
+	qm_dqrr_dpull = 1	/* PDQCR */
+};
+enum qm_dqrr_pmode {		/* s/w-only */
+	qm_dqrr_pci,		/* reads DQRR_PI_CINH */
+	qm_dqrr_pce,		/* reads DQRR_PI_CENA */
+	qm_dqrr_pvb		/* reads valid-bit */
+};
+enum qm_dqrr_cmode {		/* matches QCSP_CFG::DCM */
+	qm_dqrr_cci = 0,	/* CI index, cache-inhibited */
+	qm_dqrr_cce = 1,	/* CI index, cache-enabled */
+	qm_dqrr_cdc = 2		/* Discrete Consumption Acknowledgement */
+};
+enum qm_mr_pmode {		/* s/w-only */
+	qm_mr_pci,		/* reads MR_PI_CINH */
+	qm_mr_pce,		/* reads MR_PI_CENA */
+	qm_mr_pvb		/* reads valid-bit */
+};
+enum qm_mr_cmode {		/* matches QCSP_CFG::MM */
+	qm_mr_cci = 0,		/* CI index, cache-inhibited */
+	qm_mr_cce = 1		/* CI index, cache-enabled */
+};
+
+
+/* ------------------------- */
+/* --- Portal structures --- */
+
+#define QM_EQCR_SIZE		8
+#define QM_DQRR_SIZE		16
+#define QM_MR_SIZE		8
+
+struct qm_eqcr {
+	struct qm_eqcr_entry *ring, *cursor;
+	u8 ci, available, ithresh, vbit;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	u32 busy;
+	enum qm_eqcr_pmode pmode;
+	enum qm_eqcr_cmode cmode;
+#endif
+};
+
+struct qm_dqrr {
+	const struct qm_dqrr_entry *ring, *cursor;
+	u8 pi, ci, fill, ithresh, vbit;
+#ifdef CONFIG_FSL_DPA_CHECKING
+#define QM_DQRR_FLAG_RE 0x01 /* Stash ring entries */
+#define QM_DQRR_FLAG_SE 0x02 /* Stash data */
+	u8 flags;
+	enum qm_dqrr_dmode dmode;
+	enum qm_dqrr_pmode pmode;
+	enum qm_dqrr_cmode cmode;
+#endif
+};
+
+struct qm_mr {
+	const struct qm_mr_entry *ring, *cursor;
+	u8 pi, ci, fill, ithresh, vbit;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	enum qm_mr_pmode pmode;
+	enum qm_mr_cmode cmode;
+#endif
+};
+
+struct qm_mc {
+	struct qm_mc_command *cr;
+	struct qm_mc_result *rr;
+	u8 rridx, vbit;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	enum {
+		/* Can be _mc_start()ed */
+		mc_idle,
+		/* Can be _mc_commit()ed or _mc_abort()ed */
+		mc_user,
+		/* Can only be _mc_retry()ed */
+		mc_hw
+	} state;
+#endif
+};
+
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+/* For workarounds that require storage. The struct alignment is required for
+ * cases where operations on "shadow" structs need the same alignment as is
+ * present on the corresponding h/w data structs (specifically, there is a
+ * zero-bit present above the range required to address the ring, so that
+ * iteration can be achieved by incrementing a ring pointer and clearing the
+ * carry-bit). The "portal" struct needs the same alignment because this type
+ * goes at its head, so it has a more radical alignment requirement if this
+ * structure is used. (NB: "64" instead of "L1_CACHE_BYTES", because this
+ * alignment relates to the h/w interface, not the CPU cache granularity!)*/
+#define QM_PORTAL_ALIGNMENT __attribute__((aligned(32 * 64)))
+struct qm_portal_bugs {
+	/* shadow MR ring, for QMAN9 workaround, 8-CL-aligned */
+	struct qm_mr_entry mr[QM_MR_SIZE];
+	/* shadow MC result, for QMAN6 and QMAN7 workarounds, CL-aligned */
+	struct qm_mc_result result;
+	/* boolean switch for QMAN7 workaround */
+	int initfq_and_sched;
+} QM_PORTAL_ALIGNMENT;
+#else
+#define QM_PORTAL_ALIGNMENT ____cacheline_aligned
+#endif
+
+struct qm_addr {
+	void __iomem *addr_ce;	/* cache-enabled */
+	void __iomem *addr_ci;	/* cache-inhibited */
+};
+
+struct qm_portal {
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+	struct qm_portal_bugs bugs;
+#endif
+	/* In the non-CONFIG_FSL_DPA_CHECKING case, the following stuff up to
+	 * and including 'mc' fits within a cacheline (yay!). The 'config' part
+	 * is setup-only, so isn't a cause for a concern. In other words, don't
+	 * rearrange this structure on a whim, there be dragons ... */
+	struct qm_addr addr;
+	struct qm_eqcr eqcr;
+	struct qm_dqrr dqrr;
+	struct qm_mr mr;
+	struct qm_mc mc;
+} QM_PORTAL_ALIGNMENT;
+
+
+/* ---------------- */
+/* --- EQCR API --- */
+
+/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */
+#define EQCR_CARRYCLEAR(p) \
+	(void *)((unsigned long)(p) & (~(unsigned long)(QM_EQCR_SIZE << 6)))
+
+/* Bit-wise logic to convert a ring pointer to a ring index */
+static inline u8 EQCR_PTR2IDX(struct qm_eqcr_entry *e)
+{
+	return ((uintptr_t)e >> 6) & (QM_EQCR_SIZE - 1);
+}
+
+/* Increment the 'cursor' ring pointer, taking 'vbit' into account */
+static inline void EQCR_INC(struct qm_eqcr *eqcr)
+{
+	/* NB: this is odd-looking, but experiments show that it generates fast
+	 * code with essentially no branching overheads. We increment to the
+	 * next EQCR pointer and handle overflow and 'vbit'. */
+	struct qm_eqcr_entry *partial = eqcr->cursor + 1;
+	eqcr->cursor = EQCR_CARRYCLEAR(partial);
+	if (partial != eqcr->cursor)
+		eqcr->vbit ^= QM_EQCR_VERB_VBIT;
+}
+
+static inline int qm_eqcr_init(struct qm_portal *portal,
+				enum qm_eqcr_pmode pmode,
+				__maybe_unused enum qm_eqcr_cmode cmode)
+{
+	/* This use of 'register', as well as all other occurances, is because
+	 * it has been observed to generate much faster code with gcc than is
+	 * otherwise the case. */
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	u32 cfg;
+	u8 pi;
+
+	eqcr->ring = portal->addr.addr_ce + CL_EQCR;
+	eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1);
+	qm_cl_invalidate(EQCR_CI);
+	pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1);
+	eqcr->cursor = eqcr->ring + pi;
+	eqcr->vbit = (qm_in(EQCR_PI_CINH) & QM_EQCR_SIZE) ?
+			QM_EQCR_VERB_VBIT : 0;
+	eqcr->available = QM_EQCR_SIZE - 1 -
+			cyc_diff(QM_EQCR_SIZE, eqcr->ci, pi);
+	eqcr->ithresh = qm_in(EQCR_ITR);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	eqcr->busy = 0;
+	eqcr->pmode = pmode;
+	eqcr->cmode = cmode;
+#endif
+	cfg = (qm_in(CFG) & 0x00ffffff) |
+		((pmode & 0x3) << 24);	/* QCSP_CFG::EPM */
+	qm_out(CFG, cfg);
+	return 0;
+}
+
+static inline void qm_eqcr_finish(struct qm_portal *portal)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	u8 pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1);
+	u8 ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1);
+
+	DPA_ASSERT(!eqcr->busy);
+	if (pi != EQCR_PTR2IDX(eqcr->cursor))
+		pr_crit("losing uncommited EQCR entries\n");
+	if (ci != eqcr->ci)
+		pr_crit("missing existing EQCR completions\n");
+	if (eqcr->ci != EQCR_PTR2IDX(eqcr->cursor))
+		pr_crit("EQCR destroyed unquiesced\n");
+}
+
+static inline struct qm_eqcr_entry *qm_eqcr_start(struct qm_portal *portal)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	DPA_ASSERT(!eqcr->busy);
+	if (!eqcr->available)
+		return NULL;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	eqcr->busy = 1;
+#endif
+	dcbz_64(eqcr->cursor);
+	return eqcr->cursor;
+}
+
+static inline void qm_eqcr_abort(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr;
+	DPA_ASSERT(eqcr->busy);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	eqcr->busy = 0;
+#endif
+}
+
+static inline struct qm_eqcr_entry *qm_eqcr_pend_and_next(
+					struct qm_portal *portal, u8 myverb)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	DPA_ASSERT(eqcr->busy);
+	DPA_ASSERT(eqcr->pmode != qm_eqcr_pvb);
+	if (eqcr->available == 1)
+		return NULL;
+	eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit;
+	dcbf(eqcr->cursor);
+	EQCR_INC(eqcr);
+	eqcr->available--;
+	dcbz_64(eqcr->cursor);
+	return eqcr->cursor;
+}
+
+#define EQCR_COMMIT_CHECKS(eqcr) \
+do { \
+	DPA_ASSERT(eqcr->busy); \
+	DPA_ASSERT(eqcr->cursor->orp == (eqcr->cursor->orp & 0x00ffffff)); \
+	DPA_ASSERT(eqcr->cursor->fqid == (eqcr->cursor->fqid & 0x00ffffff)); \
+} while(0)
+
+static inline void qm_eqcr_pci_commit(struct qm_portal *portal, u8 myverb)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	EQCR_COMMIT_CHECKS(eqcr);
+	DPA_ASSERT(eqcr->pmode == qm_eqcr_pci);
+	eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit;
+	EQCR_INC(eqcr);
+	eqcr->available--;
+	dcbf(eqcr->cursor);
+	hwsync();
+	qm_out(EQCR_PI_CINH, EQCR_PTR2IDX(eqcr->cursor));
+#ifdef CONFIG_FSL_DPA_CHECKING
+	eqcr->busy = 0;
+#endif
+}
+
+static inline void qm_eqcr_pce_prefetch(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr;
+	DPA_ASSERT(eqcr->pmode == qm_eqcr_pce);
+	qm_cl_invalidate(EQCR_PI);
+	qm_cl_touch_rw(EQCR_PI);
+}
+
+static inline void qm_eqcr_pce_commit(struct qm_portal *portal, u8 myverb)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	EQCR_COMMIT_CHECKS(eqcr);
+	DPA_ASSERT(eqcr->pmode == qm_eqcr_pce);
+	eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit;
+	EQCR_INC(eqcr);
+	eqcr->available--;
+	dcbf(eqcr->cursor);
+	lwsync();
+	qm_cl_out(EQCR_PI, EQCR_PTR2IDX(eqcr->cursor));
+#ifdef CONFIG_FSL_DPA_CHECKING
+	eqcr->busy = 0;
+#endif
+}
+
+static inline void qm_eqcr_pvb_commit(struct qm_portal *portal, u8 myverb)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	struct qm_eqcr_entry *eqcursor;
+	EQCR_COMMIT_CHECKS(eqcr);
+	DPA_ASSERT(eqcr->pmode == qm_eqcr_pvb);
+	lwsync();
+	eqcursor = eqcr->cursor;
+	eqcursor->__dont_write_directly__verb = myverb | eqcr->vbit;
+	dcbf(eqcursor);
+	EQCR_INC(eqcr);
+	eqcr->available--;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	eqcr->busy = 0;
+#endif
+}
+
+static inline u8 qm_eqcr_cci_update(struct qm_portal *portal)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	u8 diff, old_ci = eqcr->ci;
+	DPA_ASSERT(eqcr->cmode == qm_eqcr_cci);
+	eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1);
+	diff = cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci);
+	eqcr->available += diff;
+	return diff;
+}
+
+static inline void qm_eqcr_cce_prefetch(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr;
+	DPA_ASSERT(eqcr->cmode == qm_eqcr_cce);
+	qm_cl_touch_ro(EQCR_CI);
+}
+
+static inline u8 qm_eqcr_cce_update(struct qm_portal *portal)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	u8 diff, old_ci = eqcr->ci;
+	DPA_ASSERT(eqcr->cmode == qm_eqcr_cce);
+	eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1);
+	qm_cl_invalidate(EQCR_CI);
+	diff = cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci);
+	eqcr->available += diff;
+	return diff;
+}
+
+static inline u8 qm_eqcr_get_ithresh(struct qm_portal *portal)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	return eqcr->ithresh;
+}
+
+static inline void qm_eqcr_set_ithresh(struct qm_portal *portal, u8 ithresh)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	eqcr->ithresh = ithresh;
+	qm_out(EQCR_ITR, ithresh);
+}
+
+static inline u8 qm_eqcr_get_avail(struct qm_portal *portal)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	return eqcr->available;
+}
+
+static inline u8 qm_eqcr_get_fill(struct qm_portal *portal)
+{
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	return QM_EQCR_SIZE - 1 - eqcr->available;
+}
+
+
+/* ---------------- */
+/* --- DQRR API --- */
+
+/* FIXME: many possible improvements;
+ * - look at changing the API to use pointer rather than index parameters now
+ *   that 'cursor' is a pointer,
+ * - consider moving other parameters to pointer if it could help (ci)
+ */
+
+#define DQRR_CARRYCLEAR(p) \
+	(void *)((unsigned long)(p) & (~(unsigned long)(QM_DQRR_SIZE << 6)))
+
+static inline u8 DQRR_PTR2IDX(const struct qm_dqrr_entry *e)
+{
+	return ((uintptr_t)e >> 6) & (QM_DQRR_SIZE - 1);
+}
+
+static inline const struct qm_dqrr_entry *DQRR_INC(
+						const struct qm_dqrr_entry *e)
+{
+	return DQRR_CARRYCLEAR(e + 1);
+}
+
+static inline void qm_dqrr_set_maxfill(struct qm_portal *portal, u8 mf)
+{
+	qm_out(CFG, (qm_in(CFG) & 0xff0fffff) |
+		((mf & (QM_DQRR_SIZE - 1)) << 20));
+}
+
+static inline int qm_dqrr_init(struct qm_portal *portal,
+				const struct qm_portal_config *config,
+				enum qm_dqrr_dmode dmode,
+				__maybe_unused enum qm_dqrr_pmode pmode,
+				enum qm_dqrr_cmode cmode, u8 max_fill,
+				int disable_stash)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	u32 cfg;
+
+	if ((config->public_cfg.has_stashing) && (config->public_cfg.cpu == -1))
+		return -EINVAL;
+	/* Make sure the DQRR will be idle when we enable */
+	qm_out(DQRR_SDQCR, 0);
+	qm_out(DQRR_VDQCR, 0);
+	qm_out(DQRR_PDQCR, 0);
+	dqrr->ring = portal->addr.addr_ce + CL_DQRR;
+	dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1);
+	dqrr->ci = qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1);
+	dqrr->cursor = dqrr->ring + dqrr->ci;
+	dqrr->fill = cyc_diff(QM_DQRR_SIZE, dqrr->ci, dqrr->pi);
+	dqrr->vbit = (qm_in(DQRR_PI_CINH) & QM_DQRR_SIZE) ?
+			QM_DQRR_VERB_VBIT : 0;
+	dqrr->ithresh = qm_in(DQRR_ITR);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	dqrr->dmode = dmode;
+	dqrr->pmode = pmode;
+	dqrr->cmode = cmode;
+	dqrr->flags = 0;
+	if (!disable_stash)
+		dqrr->flags |= QM_DQRR_FLAG_RE | QM_DQRR_FLAG_SE;
+#endif
+	cfg = (qm_in(CFG) & 0xff000f00) |
+		((max_fill & (QM_DQRR_SIZE - 1)) << 20) | /* DQRR_MF */
+		((dmode & 1) << 18) |			/* DP */
+		((cmode & 3) << 16) |			/* DCM */
+		(disable_stash ? 0 :			/* RE+SE */
+			config->public_cfg.has_stashing ? 0xa0 : 0) |
+		(0 ? 0x40 : 0) |			/* Ignore RP */
+		(0 ? 0x10 : 0);				/* Ignore SP */
+	qm_out(CFG, cfg);
+	qm_dqrr_set_maxfill(portal, max_fill);
+	return 0;
+}
+
+static inline void qm_dqrr_finish(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if ((dqrr->cmode != qm_dqrr_cdc) &&
+			(dqrr->ci != DQRR_PTR2IDX(dqrr->cursor)))
+		pr_crit("Ignoring completed DQRR entries\n");
+#endif
+}
+
+static inline const struct qm_dqrr_entry *qm_dqrr_current(
+						struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	if (!dqrr->fill)
+		return NULL;
+	return dqrr->cursor;
+}
+
+static inline u8 qm_dqrr_cursor(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	return DQRR_PTR2IDX(dqrr->cursor);
+}
+
+static inline u8 qm_dqrr_next(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->fill);
+	dqrr->cursor = DQRR_INC(dqrr->cursor);
+	return --dqrr->fill;
+}
+
+static inline u8 qm_dqrr_pci_update(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	u8 diff, old_pi = dqrr->pi;
+	DPA_ASSERT(dqrr->pmode == qm_dqrr_pci);
+	dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1);
+	diff = cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi);
+	dqrr->fill += diff;
+	return diff;
+}
+
+static inline void qm_dqrr_pce_prefetch(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->pmode == qm_dqrr_pce);
+	qm_cl_invalidate(DQRR_PI);
+	qm_cl_touch_ro(DQRR_PI);
+}
+
+static inline u8 qm_dqrr_pce_update(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	u8 diff, old_pi = dqrr->pi;
+	DPA_ASSERT(dqrr->pmode == qm_dqrr_pce);
+	dqrr->pi = qm_cl_in(DQRR_PI) & (QM_DQRR_SIZE - 1);
+	diff = cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi);
+	dqrr->fill += diff;
+	return diff;
+}
+
+static inline void qm_dqrr_pvb_update(struct qm_portal *portal, int coherent)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	const struct qm_dqrr_entry *res = qm_cl(dqrr->ring, dqrr->pi);
+	DPA_ASSERT(dqrr->pmode == qm_dqrr_pvb);
+	/* when accessing 'verb', use __raw_readb() to ensure that compiler
+	 * inlining doesn't try to optimise out "excess reads". */
+	if ((__raw_readb(&res->verb) & QM_DQRR_VERB_VBIT) == dqrr->vbit) {
+		dqrr->pi = (dqrr->pi + 1) & (QM_DQRR_SIZE - 1);
+		if (!dqrr->pi)
+			dqrr->vbit ^= QM_DQRR_VERB_VBIT;
+		dqrr->fill++;
+		if (!coherent) {
+			DPA_ASSERT(!(dqrr->flags & QM_DQRR_FLAG_RE));
+			dcbit_ro(DQRR_INC(res));
+		}
+	} else if (!coherent) {
+		DPA_ASSERT(!(dqrr->flags & QM_DQRR_FLAG_RE));
+		dcbit_ro(res);
+	}
+}
+
+static inline void qm_dqrr_cci_consume(struct qm_portal *portal, u8 num)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cci);
+	dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1);
+	qm_out(DQRR_CI_CINH, dqrr->ci);
+}
+
+static inline void qm_dqrr_cci_consume_to_current(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cci);
+	dqrr->ci = DQRR_PTR2IDX(dqrr->cursor);
+	qm_out(DQRR_CI_CINH, dqrr->ci);
+}
+
+static inline void qm_dqrr_cce_prefetch(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cce);
+	qm_cl_invalidate(DQRR_CI);
+	qm_cl_touch_rw(DQRR_CI);
+}
+
+static inline void qm_dqrr_cce_consume(struct qm_portal *portal, u8 num)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cce);
+	dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1);
+	qm_cl_out(DQRR_CI, dqrr->ci);
+}
+
+static inline void qm_dqrr_cce_consume_to_current(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cce);
+	dqrr->ci = DQRR_PTR2IDX(dqrr->cursor);
+	qm_cl_out(DQRR_CI, dqrr->ci);
+}
+
+static inline void qm_dqrr_cdc_consume_1(struct qm_portal *portal, u8 idx,
+					int park)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	DPA_ASSERT(idx < QM_DQRR_SIZE);
+	qm_out(DQRR_DCAP, (0 << 8) |	/* S */
+		((park ? 1 : 0) << 6) |	/* PK */
+		idx);			/* DCAP_CI */
+}
+
+static inline void qm_dqrr_cdc_consume_1ptr(struct qm_portal *portal,
+					const struct qm_dqrr_entry *dq,
+					int park)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	u8 idx = DQRR_PTR2IDX(dq);
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	DPA_ASSERT((dqrr->ring + idx) == dq);
+	DPA_ASSERT(idx < QM_DQRR_SIZE);
+	qm_out(DQRR_DCAP, (0 << 8) |		/* DQRR_DCAP::S */
+		((park ? 1 : 0) << 6) |		/* DQRR_DCAP::PK */
+		idx);				/* DQRR_DCAP::DCAP_CI */
+}
+
+static inline void qm_dqrr_cdc_consume_n(struct qm_portal *portal, u16 bitmask)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	qm_out(DQRR_DCAP, (1 << 8) |		/* DQRR_DCAP::S */
+		((u32)bitmask << 16));		/* DQRR_DCAP::DCAP_CI */
+}
+
+static inline u8 qm_dqrr_cdc_cci(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	return qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1);
+}
+
+static inline void qm_dqrr_cdc_cce_prefetch(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	qm_cl_invalidate(DQRR_CI);
+	qm_cl_touch_ro(DQRR_CI);
+}
+
+static inline u8 qm_dqrr_cdc_cce(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	return qm_cl_in(DQRR_CI) & (QM_DQRR_SIZE - 1);
+}
+
+static inline u8 qm_dqrr_get_ci(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc);
+	return dqrr->ci;
+}
+
+static inline void qm_dqrr_park(struct qm_portal *portal, u8 idx)
+{
+	__maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc);
+	qm_out(DQRR_DCAP, (0 << 8) |		/* S */
+		(1 << 6) |			/* PK */
+		(idx & (QM_DQRR_SIZE - 1)));	/* DCAP_CI */
+}
+
+static inline void qm_dqrr_park_current(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc);
+	qm_out(DQRR_DCAP, (0 << 8) |		/* S */
+		(1 << 6) |			/* PK */
+		DQRR_PTR2IDX(dqrr->cursor));	/* DCAP_CI */
+}
+
+static inline void qm_dqrr_sdqcr_set(struct qm_portal *portal, u32 sdqcr)
+{
+	qm_out(DQRR_SDQCR, sdqcr);
+}
+
+static inline u32 qm_dqrr_sdqcr_get(struct qm_portal *portal)
+{
+	return qm_in(DQRR_SDQCR);
+}
+
+static inline void qm_dqrr_vdqcr_set(struct qm_portal *portal, u32 vdqcr)
+{
+	qm_out(DQRR_VDQCR, vdqcr);
+}
+
+static inline u32 qm_dqrr_vdqcr_get(struct qm_portal *portal)
+{
+	return qm_in(DQRR_VDQCR);
+}
+
+static inline void qm_dqrr_pdqcr_set(struct qm_portal *portal, u32 pdqcr)
+{
+	qm_out(DQRR_PDQCR, pdqcr);
+}
+
+static inline u32 qm_dqrr_pdqcr_get(struct qm_portal *portal)
+{
+	return qm_in(DQRR_PDQCR);
+}
+
+static inline u8 qm_dqrr_get_ithresh(struct qm_portal *portal)
+{
+	register struct qm_dqrr *dqrr = &portal->dqrr;
+	return dqrr->ithresh;
+}
+
+static inline void qm_dqrr_set_ithresh(struct qm_portal *portal, u8 ithresh)
+{
+	qm_out(DQRR_ITR, ithresh);
+}
+
+static inline u8 qm_dqrr_get_maxfill(struct qm_portal *portal)
+{
+	return (qm_in(CFG) & 0x00f00000) >> 20;
+}
+
+
+/* -------------- */
+/* --- MR API --- */
+
+#define MR_CARRYCLEAR(p) \
+	(void *)((unsigned long)(p) & (~(unsigned long)(QM_MR_SIZE << 6)))
+
+static inline u8 MR_PTR2IDX(const struct qm_mr_entry *e)
+{
+	return ((uintptr_t)e >> 6) & (QM_MR_SIZE - 1);
+}
+
+static inline const struct qm_mr_entry *MR_INC(const struct qm_mr_entry *e)
+{
+	return MR_CARRYCLEAR(e + 1);
+}
+
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+static inline void __mr_copy_and_fixup(struct qm_portal *p, u8 idx)
+{
+	if (qman_ip_rev == QMAN_REV10) {
+		struct qm_mr_entry *shadow = qm_cl(p->bugs.mr, idx);
+		struct qm_mr_entry *res = qm_cl(p->mr.ring, idx);
+		copy_words(shadow, res, sizeof(*res));
+		/* Bypass the QM_MR_RC_*** definitions, and check the byte value
+		 * directly to handle the erratum. */
+		if (shadow->ern.rc == 0x06)
+			shadow->ern.rc = 0x60;
+	}
+}
+#else
+#define __mr_copy_and_fixup(p, idx) do { ; } while (0)
+#endif
+
+static inline int qm_mr_init(struct qm_portal *portal, enum qm_mr_pmode pmode,
+		enum qm_mr_cmode cmode)
+{
+	register struct qm_mr *mr = &portal->mr;
+	u32 cfg;
+	int loop;
+
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+	if ((qman_ip_rev == QMAN_REV10) && (pmode != qm_mr_pvb)) {
+		pr_err("Qman is rev1, so QMAN9 workaround requires 'pvb'\n");
+		return -EINVAL;
+	}
+#endif
+	mr->ring = portal->addr.addr_ce + CL_MR;
+	mr->pi = qm_in(MR_PI_CINH) & (QM_MR_SIZE - 1);
+	mr->ci = qm_in(MR_CI_CINH) & (QM_MR_SIZE - 1);
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+	if (qman_ip_rev == QMAN_REV10)
+		/* Situate the cursor in the shadow ring */
+		mr->cursor = portal->bugs.mr + mr->ci;
+	else
+#endif
+	mr->cursor = mr->ring + mr->ci;
+	mr->fill = cyc_diff(QM_MR_SIZE, mr->ci, mr->pi);
+	mr->vbit = (qm_in(MR_PI_CINH) & QM_MR_SIZE) ? QM_MR_VERB_VBIT : 0;
+	mr->ithresh = qm_in(MR_ITR);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mr->pmode = pmode;
+	mr->cmode = cmode;
+#endif
+	/* Only new entries get the copy-and-fixup treatment from
+	 * qm_mr_pvb_update(), so perform it here for any stale entries. */
+	for (loop = 0; loop < mr->fill; loop++)
+		__mr_copy_and_fixup(portal, (mr->ci + loop) & (QM_MR_SIZE - 1));
+	cfg = (qm_in(CFG) & 0xfffff0ff) |
+		((cmode & 1) << 8);		/* QCSP_CFG:MM */
+	qm_out(CFG, cfg);
+	return 0;
+}
+
+static inline void qm_mr_finish(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	if (mr->ci != MR_PTR2IDX(mr->cursor))
+		pr_crit("Ignoring completed MR entries\n");
+}
+
+static inline const struct qm_mr_entry *qm_mr_current(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	if (!mr->fill)
+		return NULL;
+	return mr->cursor;
+}
+
+static inline u8 qm_mr_cursor(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	return MR_PTR2IDX(mr->cursor);
+}
+
+static inline u8 qm_mr_next(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	DPA_ASSERT(mr->fill);
+	mr->cursor = MR_INC(mr->cursor);
+	return --mr->fill;
+}
+
+static inline u8 qm_mr_pci_update(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	u8 diff, old_pi = mr->pi;
+	DPA_ASSERT(mr->pmode == qm_mr_pci);
+	mr->pi = qm_in(MR_PI_CINH);
+	diff = cyc_diff(QM_MR_SIZE, old_pi, mr->pi);
+	mr->fill += diff;
+	return diff;
+}
+
+static inline void qm_mr_pce_prefetch(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_mr *mr = &portal->mr;
+	DPA_ASSERT(mr->pmode == qm_mr_pce);
+	qm_cl_invalidate(MR_PI);
+	qm_cl_touch_ro(MR_PI);
+}
+
+static inline u8 qm_mr_pce_update(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	u8 diff, old_pi = mr->pi;
+	DPA_ASSERT(mr->pmode == qm_mr_pce);
+	mr->pi = qm_cl_in(MR_PI) & (QM_MR_SIZE - 1);
+	diff = cyc_diff(QM_MR_SIZE, old_pi, mr->pi);
+	mr->fill += diff;
+	return diff;
+}
+
+static inline void qm_mr_pvb_update(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	const struct qm_mr_entry *res = qm_cl(mr->ring, mr->pi);
+	DPA_ASSERT(mr->pmode == qm_mr_pvb);
+	/* when accessing 'verb', use __raw_readb() to ensure that compiler
+	 * inlining doesn't try to optimise out "excess reads". */
+	if ((__raw_readb(&res->verb) & QM_MR_VERB_VBIT) == mr->vbit) {
+		__mr_copy_and_fixup(portal, mr->pi);
+		mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1);
+		if (!mr->pi)
+			mr->vbit ^= QM_MR_VERB_VBIT;
+		mr->fill++;
+		res = MR_INC(res);
+	}
+	dcbit_ro(res);
+}
+
+static inline void qm_mr_cci_consume(struct qm_portal *portal, u8 num)
+{
+	register struct qm_mr *mr = &portal->mr;
+	DPA_ASSERT(mr->cmode == qm_mr_cci);
+	mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1);
+	qm_out(MR_CI_CINH, mr->ci);
+}
+
+static inline void qm_mr_cci_consume_to_current(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	DPA_ASSERT(mr->cmode == qm_mr_cci);
+	mr->ci = MR_PTR2IDX(mr->cursor);
+	qm_out(MR_CI_CINH, mr->ci);
+}
+
+static inline void qm_mr_cce_prefetch(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_mr *mr = &portal->mr;
+	DPA_ASSERT(mr->cmode == qm_mr_cce);
+	qm_cl_invalidate(MR_CI);
+	qm_cl_touch_rw(MR_CI);
+}
+
+static inline void qm_mr_cce_consume(struct qm_portal *portal, u8 num)
+{
+	register struct qm_mr *mr = &portal->mr;
+	DPA_ASSERT(mr->cmode == qm_mr_cce);
+	mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1);
+	qm_cl_out(MR_CI, mr->ci);
+}
+
+static inline void qm_mr_cce_consume_to_current(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	DPA_ASSERT(mr->cmode == qm_mr_cce);
+	mr->ci = MR_PTR2IDX(mr->cursor);
+	qm_cl_out(MR_CI, mr->ci);
+}
+
+static inline u8 qm_mr_get_ci(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	return mr->ci;
+}
+
+static inline u8 qm_mr_get_ithresh(struct qm_portal *portal)
+{
+	register struct qm_mr *mr = &portal->mr;
+	return mr->ithresh;
+}
+
+static inline void qm_mr_set_ithresh(struct qm_portal *portal, u8 ithresh)
+{
+	qm_out(MR_ITR, ithresh);
+}
+
+
+/* ------------------------------ */
+/* --- Management command API --- */
+
+static inline int qm_mc_init(struct qm_portal *portal)
+{
+	register struct qm_mc *mc = &portal->mc;
+	mc->cr = portal->addr.addr_ce + CL_CR;
+	mc->rr = portal->addr.addr_ce + CL_RR0;
+	mc->rridx = (__raw_readb(&mc->cr->__dont_write_directly__verb) &
+			QM_MCC_VERB_VBIT) ?  0 : 1;
+	mc->vbit = mc->rridx ? QM_MCC_VERB_VBIT : 0;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_idle;
+#endif
+	return 0;
+}
+
+static inline void qm_mc_finish(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_mc *mc = &portal->mc;
+	DPA_ASSERT(mc->state == mc_idle);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	if (mc->state != mc_idle)
+		pr_crit("Losing incomplete MC command\n");
+#endif
+}
+
+static inline struct qm_mc_command *qm_mc_start(struct qm_portal *portal)
+{
+	register struct qm_mc *mc = &portal->mc;
+	DPA_ASSERT(mc->state == mc_idle);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_user;
+#endif
+	dcbz_64(mc->cr);
+	return mc->cr;
+}
+
+static inline void qm_mc_abort(struct qm_portal *portal)
+{
+	__maybe_unused register struct qm_mc *mc = &portal->mc;
+	DPA_ASSERT(mc->state == mc_user);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_idle;
+#endif
+}
+
+static inline void qm_mc_commit(struct qm_portal *portal, u8 myverb)
+{
+	register struct qm_mc *mc = &portal->mc;
+	struct qm_mc_result *rr = mc->rr + mc->rridx;
+	DPA_ASSERT(mc->state == mc_user);
+	lwsync();
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+	if ((qman_ip_rev == QMAN_REV10) && ((myverb & QM_MCC_VERB_MASK) ==
+					QM_MCC_VERB_INITFQ_SCHED)) {
+		u32 fqid = mc->cr->initfq.fqid;
+		/* Do two commands to avoid the hw bug. Note, we poll locally
+		 * rather than using qm_mc_result() because from a DPA_CHECKING
+		 * perspective, we don't want to appear to have "finished" until
+		 * both commands are done. */
+		mc->cr->__dont_write_directly__verb = mc->vbit |
+					QM_MCC_VERB_INITFQ_PARKED;
+		dcbf(mc->cr);
+		portal->bugs.initfq_and_sched = 1;
+		do {
+			dcbit_ro(rr);
+		} while (!__raw_readb(&rr->verb));
+#ifdef CONFIG_FSL_DPA_CHECKING
+		mc->state = mc_idle;
+#endif
+		if (rr->result != QM_MCR_RESULT_OK) {
+#ifdef CONFIG_FSL_DPA_CHECKING
+			mc->state = mc_hw;
+#endif
+			return;
+		}
+		mc->rridx ^= 1;
+		mc->vbit ^= QM_MCC_VERB_VBIT;
+		rr = mc->rr + mc->rridx;
+		dcbz_64(mc->cr);
+		mc->cr->alterfq.fqid = fqid;
+		lwsync();
+		myverb = QM_MCC_VERB_ALTER_SCHED;
+	} else
+		portal->bugs.initfq_and_sched = 0;
+#endif
+	mc->cr->__dont_write_directly__verb = myverb | mc->vbit;
+	dcbf(mc->cr);
+	dcbit_ro(rr);
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_hw;
+#endif
+}
+
+static inline struct qm_mc_result *qm_mc_result(struct qm_portal *portal)
+{
+	register struct qm_mc *mc = &portal->mc;
+	struct qm_mc_result *rr = mc->rr + mc->rridx;
+	DPA_ASSERT(mc->state == mc_hw);
+	/* The inactive response register's verb byte always returns zero until
+	 * its command is submitted and completed. This includes the valid-bit,
+	 * in case you were wondering... */
+	if (!__raw_readb(&rr->verb)) {
+		dcbit_ro(rr);
+		return NULL;
+	}
+#ifdef CONFIG_FSL_QMAN_BUG_AND_FEATURE_REV1
+	if (qman_ip_rev == QMAN_REV10) {
+		if ((__raw_readb(&rr->verb) & QM_MCR_VERB_MASK) ==
+						QM_MCR_VERB_QUERYFQ) {
+			void *misplaced = (void *)rr + 50;
+			copy_words(&portal->bugs.result, rr, sizeof(*rr));
+			rr = &portal->bugs.result;
+			copy_shorts(&rr->queryfq.fqd.td, misplaced,
+				sizeof(rr->queryfq.fqd.td));
+		} else if (portal->bugs.initfq_and_sched) {
+			/* We split the user-requested command, make the final
+			 * result match the requested type. */
+			copy_words(&portal->bugs.result, rr, sizeof(*rr));
+			rr = &portal->bugs.result;
+			rr->verb = (rr->verb & QM_MCR_VERB_RRID) |
+					QM_MCR_VERB_INITFQ_SCHED;
+		}
+	}
+#endif
+	mc->rridx ^= 1;
+	mc->vbit ^= QM_MCC_VERB_VBIT;
+#ifdef CONFIG_FSL_DPA_CHECKING
+	mc->state = mc_idle;
+#endif
+	return rr;
+}
+
+
+/* ------------------------------------- */
+/* --- Portal interrupt register API --- */
+
+static inline int qm_isr_init(__always_unused struct qm_portal *portal)
+{
+	return 0;
+}
+
+static inline void qm_isr_finish(__always_unused struct qm_portal *portal)
+{
+}
+
+static inline void qm_isr_set_iperiod(struct qm_portal *portal, u16 iperiod)
+{
+	qm_out(ITPR, iperiod);
+}
+
+static inline u32 __qm_isr_read(struct qm_portal *portal, enum qm_isr_reg n)
+{
+	return __qm_in(&portal->addr, REG_ISR + (n << 2));
+}
+
+static inline void __qm_isr_write(struct qm_portal *portal, enum qm_isr_reg n,
+					u32 val)
+{
+	__qm_out(&portal->addr, REG_ISR + (n << 2), val);
+}
+
diff --git a/drivers/staging/fsl_qbman/qman_private.h b/drivers/staging/fsl_qbman/qman_private.h
new file mode 100644
index 0000000..fdda948
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_private.h
@@ -0,0 +1,292 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "dpa_sys.h"
+#include <linux/fsl_qman.h>
+
+#if !defined(CONFIG_FSL_QMAN_FQ_LOOKUP) && defined(CONFIG_PPC64)
+#error "_PPC64 requires _FSL_QMAN_FQ_LOOKUP"
+#endif
+
+	/* ----------------- */
+	/* Congestion Groups */
+	/* ----------------- */
+/* This wrapper represents a bit-array for the state of the 256 Qman congestion
+ * groups. Is also used as a *mask* for congestion groups, eg. so we ignore
+ * those that don't concern us. We harness the structure and accessor details
+ * already used in the management command to query congestion groups. */
+struct qman_cgrs {
+	struct __qm_mcr_querycongestion q;
+};
+static inline void qman_cgrs_init(struct qman_cgrs *c)
+{
+	memset(c, 0, sizeof(*c));
+}
+static inline void qman_cgrs_fill(struct qman_cgrs *c)
+{
+	memset(c, 0xff, sizeof(*c));
+}
+static inline int qman_cgrs_get(struct qman_cgrs *c, int num)
+{
+	return QM_MCR_QUERYCONGESTION(&c->q, num);
+}
+static inline void qman_cgrs_set(struct qman_cgrs *c, int num)
+{
+	c->q.__state[__CGR_WORD(num)] |= (0x80000000 >> __CGR_SHIFT(num));
+}
+static inline void qman_cgrs_unset(struct qman_cgrs *c, int num)
+{
+	c->q.__state[__CGR_WORD(num)] &= ~(0x80000000 >> __CGR_SHIFT(num));
+}
+static inline int qman_cgrs_next(struct qman_cgrs *c, int num)
+{
+	while ((++num < __CGR_NUM) && !qman_cgrs_get(c, num))
+		;
+	return num;
+}
+static inline void qman_cgrs_cp(struct qman_cgrs *dest,
+				const struct qman_cgrs *src)
+{
+	memcpy(dest, src, sizeof(*dest));
+}
+static inline void qman_cgrs_and(struct qman_cgrs *dest,
+			const struct qman_cgrs *a, const struct qman_cgrs *b)
+{
+	int ret;
+	u32 *_d = dest->q.__state;
+	const u32 *_a = a->q.__state;
+	const u32 *_b = b->q.__state;
+	for (ret = 0; ret < 8; ret++)
+		*(_d++) = *(_a++) & *(_b++);
+}
+static inline void qman_cgrs_xor(struct qman_cgrs *dest,
+			const struct qman_cgrs *a, const struct qman_cgrs *b)
+{
+	int ret;
+	u32 *_d = dest->q.__state;
+	const u32 *_a = a->q.__state;
+	const u32 *_b = b->q.__state;
+	for (ret = 0; ret < 8; ret++)
+		*(_d++) = *(_a++) ^ *(_b++);
+}
+
+#define qman_cgrs_for_each_1(cgr, cgrs) \
+	for ((cgr) = -1; (cgr) = qman_cgrs_next((cgrs), (cgr)),\
+					(cgr) < __CGR_NUM;)
+
+/* used by CCSR and portal interrupt code */
+enum qm_isr_reg {
+	qm_isr_status = 0,
+	qm_isr_enable = 1,
+	qm_isr_disable = 2,
+	qm_isr_inhibit = 3
+};
+
+#define QM_ADDR_CE 0
+#define QM_ADDR_CI 1
+struct qm_portal_config {
+	/* Corenet portal addresses;
+	 * [0]==cache-enabled, [1]==cache-inhibited. */
+	__iomem void *addr_virt[2];
+	struct resource addr_phys[2];
+	struct device_node *node;
+	/* Allow these to be joined in lists */
+	struct list_head list;
+	/* User-visible portal configuration settings */
+	struct qman_portal_config public_cfg;
+};
+
+/* Hooks for driver initialisation */
+__init int fqalloc_init(int use_bman);
+
+/* Revision info (for errata and feature handling) */
+#define QMAN_REV10 0x0100
+#define QMAN_REV11 0x0101
+#define QMAN_REV12 0x0102
+#define QMAN_REV20 0x0200
+extern u16 qman_ip_rev; /* 0 if uninitialised, otherwise QMAN_REVx */
+
+#ifdef CONFIG_FSL_QMAN_CONFIG
+/* Hooks from qman_driver.c to qman_config.c */
+int qman_init_error_int(struct device_node *node);
+void qman_liodn_fixup(enum qm_channel channel);
+#endif
+
+/* Hooks from qman_driver.c in to qman_high.c */
+struct qman_portal *qman_create_affine_portal(
+			const struct qm_portal_config *config,
+			const struct qman_cgrs *cgrs,
+			const struct qman_fq_cb *null_cb,
+			int recovery_mode);
+struct qman_portal *qman_create_affine_slave(struct qman_portal *redirect);
+const struct qm_portal_config *qman_destroy_affine_portal(void);
+void qman_recovery_exit_local(void);
+
+/* This CGR feature is supported by h/w and required by unit-tests and the
+ * debugfs hooks, so is implemented in the driver. However it allows an explicit
+ * corruption of h/w fields by s/w that are usually incorruptible (because the
+ * counters are usually maintained entirely within h/w). As such, we declare
+ * this API internally. */
+int qman_testwrite_cgr(struct qman_cgr *cgr, u64 i_bcnt,
+	struct qm_mcr_cgrtestwrite *result);
+
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+/* If the fq object pointer is greater than the size of context_b field,
+ * than a lookup table is required. */
+int qman_setup_fq_lookup_table(size_t num_entries);
+#endif
+
+/*************************************************/
+/*   QMan s/w corenet portal, low-level i/face   */
+/*************************************************/
+
+/* Note: most functions are only used by the high-level interface, so are
+ * inlined from qman_low.h. The stuff below is for use by other parts of the
+ * driver. */
+
+/* For qm_dqrr_sdqcr_set(); Choose one SOURCE. Choose one COUNT. Choose one
+ * dequeue TYPE. Choose TOKEN (8-bit).
+ * If SOURCE == CHANNELS,
+ *   Choose CHANNELS_DEDICATED and/or CHANNELS_POOL(n).
+ *   You can choose DEDICATED_PRECEDENCE if the portal channel should have
+ *   priority.
+ * If SOURCE == SPECIFICWQ,
+ *     Either select the work-queue ID with SPECIFICWQ_WQ(), or select the
+ *     channel (SPECIFICWQ_DEDICATED or SPECIFICWQ_POOL()) and specify the
+ *     work-queue priority (0-7) with SPECIFICWQ_WQ() - either way, you get the
+ *     same value.
+ */
+#define QM_SDQCR_SOURCE_CHANNELS	0x0
+#define QM_SDQCR_SOURCE_SPECIFICWQ	0x40000000
+#define QM_SDQCR_COUNT_EXACT1		0x0
+#define QM_SDQCR_COUNT_UPTO3		0x20000000
+#define QM_SDQCR_DEDICATED_PRECEDENCE	0x10000000
+#define QM_SDQCR_TYPE_MASK		0x03000000
+#define QM_SDQCR_TYPE_NULL		0x0
+#define QM_SDQCR_TYPE_PRIO_QOS		0x01000000
+#define QM_SDQCR_TYPE_ACTIVE_QOS	0x02000000
+#define QM_SDQCR_TYPE_ACTIVE		0x03000000
+#define QM_SDQCR_TOKEN_MASK		0x00ff0000
+#define QM_SDQCR_TOKEN_SET(v)		(((v) & 0xff) << 16)
+#define QM_SDQCR_TOKEN_GET(v)		(((v) >> 16) & 0xff)
+#define QM_SDQCR_CHANNELS_DEDICATED	0x00008000
+#if 0 /* These are defined in the external fsl_qman.h API */
+#define QM_SDQCR_CHANNELS_POOL_MASK	0x00007fff
+#define QM_SDQCR_CHANNELS_POOL(n)	(0x00008000 >> (n))
+#endif
+#define QM_SDQCR_SPECIFICWQ_MASK	0x000000f7
+#define QM_SDQCR_SPECIFICWQ_DEDICATED	0x00000000
+#define QM_SDQCR_SPECIFICWQ_POOL(n)	((n) << 4)
+#define QM_SDQCR_SPECIFICWQ_WQ(n)	(n)
+
+/* For qm_dqrr_vdqcr_set(); Choose one PRECEDENCE. EXACT is optional. Use
+ * NUMFRAMES(n) (6-bit) or NUMFRAMES_TILLEMPTY to fill in the frame-count. Use
+ * FQID(n) to fill in the frame queue ID. */
+#if 0 /* These are defined in the external fsl_qman.h API */
+#define QM_VDQCR_PRECEDENCE_VDQCR	0x0
+#define QM_VDQCR_PRECEDENCE_SDQCR	0x80000000
+#define QM_VDQCR_EXACT			0x40000000
+#define QM_VDQCR_NUMFRAMES_MASK		0x3f000000
+#define QM_VDQCR_NUMFRAMES_SET(n)	(((n) & 0x3f) << 24)
+#define QM_VDQCR_NUMFRAMES_GET(n)	(((n) >> 24) & 0x3f)
+#define QM_VDQCR_NUMFRAMES_TILLEMPTY	QM_VDQCR_NUMFRAMES_SET(0)
+#endif
+#define QM_VDQCR_FQID_MASK		0x00ffffff
+#define QM_VDQCR_FQID(n)		((n) & QM_VDQCR_FQID_MASK)
+
+/* For qm_dqrr_pdqcr_set(); Choose one MODE. Choose one COUNT.
+ * If MODE==SCHEDULED
+ *   Choose SCHEDULED_CHANNELS or SCHEDULED_SPECIFICWQ. Choose one dequeue TYPE.
+ *   If CHANNELS,
+ *     Choose CHANNELS_DEDICATED and/or CHANNELS_POOL() channels.
+ *     You can choose DEDICATED_PRECEDENCE if the portal channel should have
+ *     priority.
+ *   If SPECIFICWQ,
+ *     Either select the work-queue ID with SPECIFICWQ_WQ(), or select the
+ *     channel (SPECIFICWQ_DEDICATED or SPECIFICWQ_POOL()) and specify the
+ *     work-queue priority (0-7) with SPECIFICWQ_WQ() - either way, you get the
+ *     same value.
+ * If MODE==UNSCHEDULED
+ *     Choose FQID().
+ */
+#define QM_PDQCR_MODE_SCHEDULED		0x0
+#define QM_PDQCR_MODE_UNSCHEDULED	0x80000000
+#define QM_PDQCR_SCHEDULED_CHANNELS	0x0
+#define QM_PDQCR_SCHEDULED_SPECIFICWQ	0x40000000
+#define QM_PDQCR_COUNT_EXACT1		0x0
+#define QM_PDQCR_COUNT_UPTO3		0x20000000
+#define QM_PDQCR_DEDICATED_PRECEDENCE	0x10000000
+#define QM_PDQCR_TYPE_MASK		0x03000000
+#define QM_PDQCR_TYPE_NULL		0x0
+#define QM_PDQCR_TYPE_PRIO_QOS		0x01000000
+#define QM_PDQCR_TYPE_ACTIVE_QOS	0x02000000
+#define QM_PDQCR_TYPE_ACTIVE		0x03000000
+#define QM_PDQCR_CHANNELS_DEDICATED	0x00008000
+#define QM_PDQCR_CHANNELS_POOL(n)	(0x00008000 >> (n))
+#define QM_PDQCR_SPECIFICWQ_MASK	0x000000f7
+#define QM_PDQCR_SPECIFICWQ_DEDICATED	0x00000000
+#define QM_PDQCR_SPECIFICWQ_POOL(n)	((n) << 4)
+#define QM_PDQCR_SPECIFICWQ_WQ(n)	(n)
+#define QM_PDQCR_FQID(n)		((n) & 0xffffff)
+
+/* Used by all portal interrupt registers except 'inhibit'. NB, some of these
+ * definitions are exported for use by the qman_irqsource_***() APIs, so are
+ * commented-out here. */
+#define QM_PIRQ_DQAVAIL	0x0000ffff	/* Channels with frame availability */
+#if 0
+#define QM_PIRQ_CSCI	0x00100000	/* Congestion State Change */
+#define QM_PIRQ_EQCI	0x00080000	/* Enqueue Command Committed */
+#define QM_PIRQ_EQRI	0x00040000	/* EQCR Ring (below threshold) */
+#define QM_PIRQ_DQRI	0x00020000	/* DQRR Ring (non-empty) */
+#define QM_PIRQ_MRI	0x00010000	/* MR Ring (non-empty) */
+/* This mask contains all the interrupt sources that need handling except DQRI,
+ * ie. that if present should trigger slow-path processing. */
+#define QM_PIRQ_SLOW	(QM_PIRQ_CSCI | QM_PIRQ_EQCI | QM_PIRQ_EQRI | \
+			QM_PIRQ_MRI)
+#endif
+/* The DQAVAIL interrupt fields break down into these bits; */
+#define QM_DQAVAIL_PORTAL	0x8000		/* Portal channel */
+#define QM_DQAVAIL_POOL(n)	(0x8000 >> (n))	/* Pool channel, n==[1..15] */
+#define QM_DQAVAIL_MASK		0xffff
+/* This mask contains all the "irqsource" bits visible to API users */
+#define QM_PIRQ_VISIBLE	(QM_PIRQ_SLOW | QM_PIRQ_DQRI)
+
+/* These are qm_<reg>_<verb>(). So for example, qm_disable_write() means "write
+ * the disable register" rather than "disable the ability to write". */
+#define qm_isr_status_read(qm)		__qm_isr_read(qm, qm_isr_status)
+#define qm_isr_status_clear(qm, m)	__qm_isr_write(qm, qm_isr_status, m)
+#define qm_isr_enable_read(qm)		__qm_isr_read(qm, qm_isr_enable)
+#define qm_isr_enable_write(qm, v)	__qm_isr_write(qm, qm_isr_enable, v)
+#define qm_isr_disable_read(qm)		__qm_isr_read(qm, qm_isr_disable)
+#define qm_isr_disable_write(qm, v)	__qm_isr_write(qm, qm_isr_disable, v)
+/* TODO: unfortunate name-clash here, reword? */
+#define qm_isr_inhibit(qm)		__qm_isr_write(qm, qm_isr_inhibit, 1)
+#define qm_isr_uninhibit(qm)		__qm_isr_write(qm, qm_isr_inhibit, 0)
diff --git a/drivers/staging/fsl_qbman/qman_test.c b/drivers/staging/fsl_qbman/qman_test.c
new file mode 100644
index 0000000..248fc0e
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_test.c
@@ -0,0 +1,61 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_test.h"
+
+MODULE_AUTHOR("Geoff Thorpe");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_DESCRIPTION("Qman testing");
+
+static int test_init(void)
+{
+	int loop = 1;
+	while(loop--) {
+#ifdef CONFIG_FSL_QMAN_TEST_STASH_POTATO
+		qman_test_hotpotato();
+#endif
+#ifdef CONFIG_FSL_QMAN_TEST_HIGH
+		qman_test_high();
+#endif
+#ifdef CONFIG_FSL_QMAN_TEST_ERRATA
+		qman_test_errata();
+#endif
+	}
+	return 0;
+}
+
+static void test_exit(void)
+{
+}
+
+module_init(test_init);
+module_exit(test_exit);
+
diff --git a/drivers/staging/fsl_qbman/qman_test.h b/drivers/staging/fsl_qbman/qman_test.h
new file mode 100644
index 0000000..681751d
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_test.h
@@ -0,0 +1,85 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/io.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/interrupt.h>
+#include <linux/delay.h>
+
+#include <linux/fsl_qman.h>
+
+void qman_test_hotpotato(void);
+void qman_test_high(void);
+void qman_test_errata(void);
+void qman_test_fqrange(void);
+
+static inline void __hexdump(unsigned long start, unsigned long end,
+			unsigned long p, size_t sz, const unsigned char *c)
+{
+	while (start < end) {
+		unsigned int pos = 0;
+		char buf[64];
+		int nl = 0;
+		pos += sprintf(buf + pos, "%08lx: ", start);
+		do {
+			if ((start < p) || (start >= (p + sz)))
+				pos += sprintf(buf + pos, "..");
+			else
+				pos += sprintf(buf + pos, "%02x", *(c++));
+			if (!(++start & 15)) {
+				buf[pos++] = '\n';
+				nl = 1;
+			} else {
+				nl = 0;
+				if(!(start & 1))
+					buf[pos++] = ' ';
+				if(!(start & 3))
+					buf[pos++] = ' ';
+			}
+		} while (start & 15);
+		if (!nl)
+			buf[pos++] = '\n';
+		buf[pos] = '\0';
+		pr_info("%s", buf);
+	}
+}
+static inline void hexdump(const void *ptr, size_t sz)
+{
+	unsigned long p = (unsigned long)ptr;
+	unsigned long start = p & ~(unsigned long)15;
+	unsigned long end = (p + sz + 15) & ~(unsigned long)15;
+	const unsigned char *c = ptr;
+	__hexdump(start, end, p, sz, c);
+}
+
diff --git a/drivers/staging/fsl_qbman/qman_test_errata.c b/drivers/staging/fsl_qbman/qman_test_errata.c
new file mode 100644
index 0000000..26b2878
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_test_errata.c
@@ -0,0 +1,248 @@ 
+/* Copyright 2009-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_test.h"
+
+/* Waiting on a model fix from virtutech */
+#if 0
+/*********************/
+/* generic utilities */
+/*********************/
+
+static int do_enqueues(struct qman_fq *fq, const struct qm_fd *fds, int num)
+{
+	int ret = 0;
+	u32 flags = QMAN_ENQUEUE_FLAG_WAIT;
+	while (num-- && !ret) {
+		if (!num)
+			flags |= QMAN_ENQUEUE_FLAG_WAIT_SYNC;
+		pr_info("about to enqueue\n");
+		ret = qman_enqueue(fq, fds++, flags);
+	}
+	return ret;
+}
+
+/***************************/
+/* "tdthresh" test (QMAN6) */
+/***************************/
+
+/* First thresh == 201 * (2^21) == 421527552 (0x19200000) */
+#define THRESH_MANT	201
+#define THRESH_EXP	21
+
+/* first three equal thresh, fourth takes us over */
+static const struct qm_fd td_eq[] = {
+	QM_FD_FMT_20(0, 0x34, 0x87654321, QM_FD_SG, 0, 79321),
+	QM_FD_FMT_29(0, 0x34, 0x87654321, QM_FD_COMPOUND, 29923679),
+	QM_FD_FMT_29(0, 0x0d, 0xacadabba, QM_FD_CONTIG_BIG, 391524552),
+	QM_FD_FMT_20(0, 0x0b, 0x0fa10ada, QM_FD_CONTIG, 0, 1),
+	QM_FD_FMT_20(0, 0x0b, 0x0fa10ada, QM_FD_CONTIG, 0, 1),
+};
+
+struct tdthresh_fq {
+	struct qman_fq fq;
+	int got_ern;
+	int num_dqrr;
+};
+
+static enum qman_cb_dqrr_result cb_dqrr_tdthresh(struct qman_portal *p,
+					struct qman_fq *__fq,
+					const struct qm_dqrr_entry *dqrr)
+{
+	struct tdthresh_fq *t = (void *)__fq;
+	t->num_dqrr++;
+	return qman_cb_dqrr_consume;
+}
+
+static void cb_ern_tdthresh(struct qman_portal *p, struct qman_fq *__fq,
+					const struct qm_mr_entry *mr)
+{
+	struct tdthresh_fq *t = (void *)__fq;
+	t->got_ern = 1;
+}
+
+static void test_tdthresh(void)
+{
+	struct tdthresh_fq tdfq = {
+		.fq = {
+			.cb = {
+				.dqrr = cb_dqrr_tdthresh,
+				.ern = cb_ern_tdthresh
+			}
+		},
+		.got_ern = 0,
+		.num_dqrr = 0
+	};
+	struct qman_fq *fq = &tdfq.fq;
+	struct qm_mcc_initfq opts = {
+		.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_TDTHRESH,
+		.fqd = {
+			.fq_ctrl = QM_FQCTRL_TDE,
+			.td = {
+				.exp = THRESH_EXP,
+				.mant = THRESH_MANT,
+			}
+		}
+	};
+	struct qm_fqd fqd;
+	u32 flags;
+	int ret = qman_create_fq(0, QMAN_FQ_FLAG_DYNAMIC_FQID, fq);
+	BUG_ON(ret);
+	/* leave it parked, and set it for local dequeue (loopback) */
+	ret = qman_init_fq(fq, QMAN_INITFQ_FLAG_LOCAL, &opts);
+	BUG_ON(ret);
+	/* query it back and confirm everything is ok */
+	ret = qman_query_fq(fq, &fqd);
+	BUG_ON(ret);
+	if (fqd.fq_ctrl != opts.fqd.fq_ctrl) {
+		pr_err("queried fq_ctrl=%x, should be=%x\n", fqd.fq_ctrl,
+			opts.fqd.fq_ctrl);
+		panic("fail");
+	}
+	if (memcmp(&fqd.td, &opts.fqd.td, sizeof(fqd.td))) {
+		pr_err("queried td_thresh=%x:%x, should be=%x:%x\n",
+			fqd.td.exp, fqd.td.mant,
+			opts.fqd.td.exp, opts.fqd.td.mant);
+		panic("fail");
+	}
+	ret = do_enqueues(fq, td_eq, 3);
+	BUG_ON(ret);
+	pr_info("  tdthresh: eq[0..2] complete\n");
+	/* enqueues are flushed, so if Qman is going to throw an ERN, the irq
+	 * assertion will already be on its way. */
+	msleep(500);
+	BUG_ON(tdfq.got_ern);
+	pr_info("  tdthresh: eq <= thresh OK\n");
+	ret = do_enqueues(fq, td_eq + 3, 1);
+	BUG_ON(ret);
+	pr_info("  tdthresh: eq[3] complete\n");
+	/* enqueues are flushed, so if Qman is going to throw an ERN, the irq
+	 * assertion will already be on its way. */
+	msleep(500);
+	BUG_ON(tdfq.got_ern);
+	pr_info("  tdthresh: eq <= thresh OK\n");
+	ret = do_enqueues(fq, td_eq + 4, 1);
+	BUG_ON(ret);
+	pr_info("  tdthresh: eq[4] complete\n");
+	/* enqueues are flushed, so if Qman is going to throw an ERN, the irq
+	 * assertion will already be on its way. */
+	msleep(500);
+	BUG_ON(!tdfq.got_ern);
+	pr_info("  tdthresh: eq > thresh OK\n");
+	ret = qman_volatile_dequeue(fq,
+		QMAN_VOLATILE_FLAG_WAIT | QMAN_VOLATILE_FLAG_FINISH,
+		QM_VDQCR_NUMFRAMES_TILLEMPTY);
+	BUG_ON(ret);
+	BUG_ON(tdfq.num_dqrr != 4);
+	ret = qman_retire_fq(fq, &flags);
+	BUG_ON(ret);
+	BUG_ON(flags);
+	ret = qman_oos_fq(fq);
+	BUG_ON(ret);
+}
+
+/****************************/
+/* "ern code6" test (QMAN9) */
+/****************************/
+
+/* Dummy FD to enqueue out-of-sequence and generate an ERN */
+static const struct qm_fd c6_eq =
+	QM_FD_FMT_29(0, 0xba, 0xdeadbeef, QM_FD_CONTIG_BIG, 1234);
+
+struct code6_fq {
+	struct qman_fq fq;
+	struct qm_mr_entry mr;
+	struct completion got_ern;
+};
+
+static void cb_ern_code6(struct qman_portal *p, struct qman_fq *__fq,
+					const struct qm_mr_entry *mr)
+{
+	struct code6_fq *c = (void *)__fq;
+	memcpy(&c->mr, mr, sizeof(*mr));
+	complete(&c->got_ern);
+}
+
+static void test_ern_code6(void)
+{
+	struct code6_fq c6fq = {
+		.fq = {
+			.cb = {
+				.ern = cb_ern_code6
+			}
+		},
+		.got_ern = COMPLETION_INITIALIZER(c6fq.got_ern)
+	};
+	struct qman_fq *fq = &c6fq.fq;
+	struct qm_mcc_initfq opts = {
+		.we_mask = QM_INITFQ_WE_FQCTRL,
+		.fqd = {
+			.fq_ctrl = QM_FQCTRL_ORP
+		}
+	};
+	u32 flags;
+	int ret = qman_create_fq(0, QMAN_FQ_FLAG_DYNAMIC_FQID, fq);
+	BUG_ON(ret);
+	/* leave it parked, and set it for local dequeue (loopback) */
+	ret = qman_init_fq(fq, QMAN_INITFQ_FLAG_LOCAL, &opts);
+	BUG_ON(ret);
+	/* enqueue with ORP using a "too early" sequence number */
+	ret = qman_enqueue_orp(fq, &c6_eq,
+		QMAN_ENQUEUE_FLAG_WAIT | QMAN_ENQUEUE_FLAG_WAIT_SYNC, fq, 5);
+	BUG_ON(ret);
+	pr_info("  code6: eq complete\n");
+	ret = qman_retire_fq(fq, &flags);
+	BUG_ON(ret);
+	pr_info("  code6: retire complete, flags=%08x\n", flags);
+	BUG_ON(flags != QMAN_FQ_STATE_ORL);
+	wait_for_completion(&c6fq.got_ern);
+	pr_info("  code6: ERN, VERB=0x%02x, RC==0x%02x\n",
+		c6fq.mr.verb, c6fq.mr.ern.rc);
+	BUG_ON(c6fq.mr.verb & 0x20);
+	BUG_ON((c6fq.mr.ern.rc & QM_MR_RC_MASK) != QM_MR_RC_ORPWINDOW_RETIRED);
+	ret = qman_oos_fq(fq);
+	BUG_ON(ret);
+}
+
+void qman_test_errata(void)
+{
+	pr_info("Testing Qman errata handling ...\n");
+	test_tdthresh();
+	test_ern_code6();
+	pr_info("                              ... SUCCESS!\n");
+}
+#else
+void qman_test_errata(void)
+{
+	pr_info("Qman errata-handling test disabled, waiting on model fix\n");
+}
+#endif
+
diff --git a/drivers/staging/fsl_qbman/qman_test_high.c b/drivers/staging/fsl_qbman/qman_test_high.c
new file mode 100644
index 0000000..e394640
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_test_high.c
@@ -0,0 +1,222 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_test.h"
+
+/*************/
+/* constants */
+/*************/
+
+#define CGR_ID		27
+#define POOL_ID		2
+#define FQ_FLAGS	QMAN_FQ_FLAG_DYNAMIC_FQID
+#define NUM_ENQUEUES	10
+#define NUM_PARTIAL	4
+#define PORTAL_SDQCR	(QM_SDQCR_SOURCE_CHANNELS | \
+			QM_SDQCR_TYPE_PRIO_QOS | \
+			QM_SDQCR_TOKEN_SET(0x98) | \
+			QM_SDQCR_CHANNELS_DEDICATED | \
+			QM_SDQCR_CHANNELS_POOL(POOL_ID))
+#define PORTAL_OPAQUE	(void *)0xf00dbeef
+#define VDQCR_FLAGS	(QMAN_VOLATILE_FLAG_WAIT | QMAN_VOLATILE_FLAG_FINISH)
+
+/*************************************/
+/* Predeclarations (eg. for fq_base) */
+/*************************************/
+
+static enum qman_cb_dqrr_result cb_dqrr(struct qman_portal *,
+					struct qman_fq *,
+					const struct qm_dqrr_entry *);
+static void cb_ern(struct qman_portal *, struct qman_fq *,
+			const struct qm_mr_entry *);
+static void cb_dc_ern(struct qman_portal *, struct qman_fq *,
+			const struct qm_mr_entry *);
+static void cb_fqs(struct qman_portal *, struct qman_fq *,
+			const struct qm_mr_entry *);
+
+/***************/
+/* global vars */
+/***************/
+
+static struct qm_fd fd, fd_dq;
+static struct qman_fq fq_base = {
+	.cb.dqrr = cb_dqrr,
+	.cb.ern = cb_ern,
+	.cb.dc_ern = cb_dc_ern,
+	.cb.fqs = cb_fqs
+};
+static DECLARE_WAIT_QUEUE_HEAD(waitqueue);
+static int retire_complete, sdqcr_complete;
+
+/**********************/
+/* internal functions */
+/**********************/
+
+/* Helpers for initialising and "incrementing" a frame descriptor */
+static void fd_init(struct qm_fd *__fd)
+{
+	qm_fd_addr_set64(__fd, 0xabdeadbeefLLU);
+	__fd->format = qm_fd_contig_big;
+	__fd->length29 = 0x0000ffff;
+	__fd->cmd = 0xfeedf00d;
+}
+
+static void fd_inc(struct qm_fd *__fd)
+{
+	u64 t = qm_fd_addr_get64(__fd);
+	int z = t >> 40;
+	t <<= 1;
+	if (z)
+		t |= 1;
+	qm_fd_addr_set64(__fd, t);
+	__fd->length29--;
+	__fd->cmd++;
+}
+
+/* The only part of the 'fd' we can't memcmp() is the ppid */
+static int fd_cmp(const struct qm_fd *a, const struct qm_fd *b)
+{
+	int r = (qm_fd_addr_get64(a) == qm_fd_addr_get64(b)) ? 0 : -1;
+	if (!r)
+		r = a->format - b->format;
+	if (!r)
+		r = a->opaque - b->opaque;
+	if (!r)
+		r = a->cmd - b->cmd;
+	return r;
+}
+
+/********/
+/* test */
+/********/
+
+static void do_enqueues(struct qman_fq *fq)
+{
+	unsigned int loop;
+	for (loop = 0; loop < NUM_ENQUEUES; loop++) {
+		if (qman_enqueue(fq, &fd, QMAN_ENQUEUE_FLAG_WAIT |
+				(((loop + 1) == NUM_ENQUEUES) ?
+				QMAN_ENQUEUE_FLAG_WAIT_SYNC : 0)))
+			panic("qman_enqueue() failed\n");
+		fd_inc(&fd);
+	}
+}
+
+void qman_test_high(void)
+{
+	int flags, res;
+	struct qman_fq *fq = &fq_base;
+
+	pr_info("qman_test_high starting\n");
+	fd_init(&fd);
+	fd_init(&fd_dq);
+
+	/* Initialise (parked) FQ */
+	if (qman_create_fq(0, FQ_FLAGS, fq))
+		panic("qman_create_fq() failed\n");
+	if (qman_init_fq(fq, QMAN_INITFQ_FLAG_LOCAL, NULL))
+		panic("qman_init_fq() failed\n");
+
+	/* Do enqueues + VDQCR, twice. (Parked FQ) */
+	do_enqueues(fq);
+	pr_info("VDQCR (till-empty);\n");
+	if (qman_volatile_dequeue(fq, VDQCR_FLAGS,
+			QM_VDQCR_NUMFRAMES_TILLEMPTY))
+		panic("qman_volatile_dequeue() failed\n");
+	do_enqueues(fq);
+	pr_info("VDQCR (%d of %d);\n", NUM_PARTIAL, NUM_ENQUEUES);
+	if (qman_volatile_dequeue(fq, VDQCR_FLAGS,
+			QM_VDQCR_NUMFRAMES_SET(NUM_PARTIAL)))
+		panic("qman_volatile_dequeue() failed\n");
+	pr_info("VDQCR (%d of %d);\n", NUM_ENQUEUES - NUM_PARTIAL,
+					NUM_ENQUEUES);
+	if (qman_volatile_dequeue(fq, VDQCR_FLAGS,
+			QM_VDQCR_NUMFRAMES_SET(NUM_ENQUEUES - NUM_PARTIAL)))
+		panic("qman_volatile_dequeue() failed\n");
+
+	do_enqueues(fq);
+	pr_info("scheduled dequeue (till-empty)\n");
+	if (qman_schedule_fq(fq))
+		panic("qman_schedule_fq() failed\n");
+	wait_event(waitqueue, sdqcr_complete);
+
+	/* Retire and OOS the FQ */
+	res = qman_retire_fq(fq, &flags);
+	if (res < 0)
+		panic("qman_retire_fq() failed\n");
+	wait_event(waitqueue, retire_complete);
+	if (flags & QMAN_FQ_STATE_BLOCKOOS)
+		panic("leaking frames\n");
+	if (qman_oos_fq(fq))
+		panic("qman_oos_fq() failed\n");
+	qman_destroy_fq(fq, 0);
+	pr_info("qman_test_high finished\n");
+}
+
+static enum qman_cb_dqrr_result cb_dqrr(struct qman_portal *p,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dq)
+{
+	if (fd_cmp(&fd_dq, &dq->fd)) {
+		pr_err("BADNESS: dequeued frame doesn't match;\n");
+		BUG();
+	}
+	fd_inc(&fd_dq);
+	if (!(dq->stat & QM_DQRR_STAT_UNSCHEDULED) && !fd_cmp(&fd_dq, &fd)) {
+		sdqcr_complete = 1;
+		wake_up(&waitqueue);
+	}
+	return qman_cb_dqrr_consume;
+}
+
+static void cb_ern(struct qman_portal *p, struct qman_fq *fq,
+			const struct qm_mr_entry *msg)
+{
+	panic("cb_ern() unimplemented");
+}
+
+static void cb_dc_ern(struct qman_portal *p, struct qman_fq *fq,
+			const struct qm_mr_entry *msg)
+{
+	panic("cb_dc_ern() unimplemented");
+}
+
+static void cb_fqs(struct qman_portal *p, struct qman_fq *fq,
+			const struct qm_mr_entry *msg)
+{
+	u8 verb = (msg->verb & QM_MR_VERB_TYPE_MASK);
+	if ((verb != QM_MR_VERB_FQRN) && (verb != QM_MR_VERB_FQRNI))
+		panic("unexpected FQS message");
+	pr_info("Retirement message received\n");
+	retire_complete = 1;
+	wake_up(&waitqueue);
+}
+
diff --git a/drivers/staging/fsl_qbman/qman_test_hotpotato.c b/drivers/staging/fsl_qbman/qman_test_hotpotato.c
new file mode 100644
index 0000000..5daa431
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_test_hotpotato.c
@@ -0,0 +1,497 @@ 
+/* Copyright 2009-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/kthread.h>
+#include <linux/platform_device.h>
+#include <linux/dma-mapping.h>
+#include "qman_test.h"
+
+/* Algorithm:
+ *
+ * Each cpu will have HP_PER_CPU "handlers" set up, each of which incorporates
+ * an rx/tx pair of FQ objects (both of which are stashed on dequeue). The
+ * organisation of FQIDs is such that the HP_PER_CPU*NUM_CPUS handlers will
+ * shuttle a "hot potato" frame around them such that every forwarding action
+ * moves it from one cpu to another. (The use of more than one handler per cpu
+ * is to allow enough handlers/FQs to truly test the significance of caching -
+ * ie. when cache-expiries are occuring.)
+ *
+ * The "hot potato" frame content will be HP_NUM_WORDS*4 bytes in size, and the
+ * first and last words of the frame data will undergo a transformation step on
+ * each forwarding action. To achieve this, each handler will be assigned a
+ * 32-bit "mixer", that is produced using a 32-bit LFSR. When a frame is
+ * received by a handler, the mixer of the expected sender is XOR'd into all
+ * words of the entire frame, which is then validated against the original
+ * values. Then, before forwarding, the entire frame is XOR'd with the mixer of
+ * the current handler. Apart from validating that the frame is taking the
+ * expected path, this also provides some quasi-realistic overheads to each
+ * forwarding action - dereferencing *all* the frame data, computation, and
+ * conditional branching. There is a "special" handler designated to act as the
+ * instigator of the test by creating an enqueuing the "hot potato" frame, and
+ * to determine when the test has completed by counting HP_LOOPS iterations.
+ *
+ * Init phases:
+ *
+ * 1. prepare each cpu's 'hp_cpu' struct using on_each_cpu(,,1) and link them
+ *    into 'hp_cpu_list'. Specifically, set processor_id, allocate HP_PER_CPU
+ *    handlers and link-list them (but do no other handler setup).
+ *
+ * 2. scan over 'hp_cpu_list' HP_PER_CPU times, the first time sets each
+ *    hp_cpu's 'iterator' to point to its first handler. With each loop,
+ *    allocate rx/tx FQIDs and mixer values to the hp_cpu's iterator handler
+ *    and advance the iterator for the next loop. This includes a final fixup,
+ *    which connects the last handler to the first (and which is why phase 2
+ *    and 3 are separate).
+ *
+ * 3. scan over 'hp_cpu_list' HP_PER_CPU times, the first time sets each
+ *    hp_cpu's 'iterator' to point to its first handler. With each loop,
+ *    initialise FQ objects and advance the iterator for the next loop.
+ *    Moreover, do this initialisation on the cpu it applies to so that Rx FQ
+ *    initialisation targets the correct cpu.
+ */
+
+/* helper to run something on all cpus (can't use on_each_cpu(), as that invokes
+ * the fn from irq context, which is too restrictive). */
+struct bstrap {
+	void (*fn)(void);
+	atomic_t started;
+};
+static int bstrap_fn(void *__bstrap)
+{
+	struct bstrap *bstrap = __bstrap;
+	atomic_inc(&bstrap->started);
+	bstrap->fn();
+	while (!kthread_should_stop())
+		msleep(1);
+	return 0;
+}
+static int on_all_cpus(void (*fn)(void))
+{
+	int cpu;
+	for_each_cpu(cpu, cpu_online_mask) {
+		struct bstrap bstrap = {
+			.fn = fn,
+			.started = ATOMIC_INIT(0)
+		};
+		struct task_struct *k = kthread_create(bstrap_fn, &bstrap,
+			"hotpotato%d", cpu);
+		int ret;
+		if (IS_ERR(k))
+			return -ENOMEM;
+		kthread_bind(k, cpu);
+		wake_up_process(k);
+		/* If we call kthread_stop() before the "wake up" has had an
+		 * effect, then the thread may exit with -EINTR without ever
+		 * running the function. So poll until it's started before
+		 * requesting it to stop. */
+		while (!atomic_read(&bstrap.started))
+			msleep(10);
+		ret = kthread_stop(k);
+		if (ret)
+			return ret;
+	}
+	return 0;
+}
+
+struct hp_handler {
+
+	/* The following data is stashed when 'rx' is dequeued; */
+	/* -------------- */
+	/* The Rx FQ, dequeues of which will stash the entire hp_handler */
+	struct qman_fq rx;
+	/* The Tx FQ we should forward to */
+	struct qman_fq tx;
+	/* The value we XOR post-dequeue, prior to validating */
+	u32 rx_mixer;
+	/* The value we XOR pre-enqueue, after validating */
+	u32 tx_mixer;
+	/* what the hotpotato address should be on dequeue */
+	dma_addr_t addr;
+	u32 *frame_ptr;
+
+	/* The following data isn't (necessarily) stashed on dequeue; */
+	/* -------------- */
+	u32 fqid_rx, fqid_tx;
+	/* list node for linking us into 'hp_cpu' */
+	struct list_head node;
+	/* Just to check ... */
+	unsigned int processor_id;
+} ____cacheline_aligned;
+
+struct hp_cpu {
+	/* identify the cpu we run on; */
+	unsigned int processor_id;
+	/* root node for the per-cpu list of handlers */
+	struct list_head handlers;
+	/* list node for linking us into 'hp_cpu_list' */
+	struct list_head node;
+	/* when repeatedly scanning 'hp_list', each time linking the n'th
+	 * handlers together, this is used as per-cpu iterator state */
+	struct hp_handler *iterator;
+};
+
+/* Each cpu has one of these */
+static DEFINE_PER_CPU(struct hp_cpu, hp_cpus);
+
+/* links together the hp_cpu structs, in first-come first-serve order. */
+static LIST_HEAD(hp_cpu_list);
+static spinlock_t hp_lock = __SPIN_LOCK_UNLOCKED(hp_lock);
+
+static unsigned int hp_cpu_list_length;
+
+/* the "special" handler, that starts and terminates the test. */
+static struct hp_handler *special_handler;
+static int loop_counter;
+
+/* handlers are allocated out of this, so they're properly aligned. */
+static struct kmem_cache *hp_handler_slab;
+
+/* this is the frame data */
+static void *__frame_ptr;
+static u32 *frame_ptr;
+static dma_addr_t frame_dma;
+
+/* the main function waits on this */
+static DECLARE_WAIT_QUEUE_HEAD(queue);
+
+#define HP_PER_CPU 	2
+#define HP_LOOPS	8
+/* 80 bytes, like a small ethernet frame, and bleeds into a second cacheline */
+#define HP_NUM_WORDS	80
+/* First word of the LFSR-based frame data */
+#define HP_FIRST_WORD	0xabbaf00d
+
+static inline u32 do_lfsr(u32 prev)
+{
+	return (prev >> 1) ^ (-(prev & 1u) & 0xd0000001u);
+}
+
+static void allocate_frame_data(void)
+{
+	u32 lfsr = HP_FIRST_WORD;
+	int loop;
+	struct platform_device *pdev = platform_device_alloc("foobar", -1);
+	if (!pdev)
+		panic("platform_device_alloc() failed");
+	if (platform_device_add(pdev))
+		panic("platform_device_add() failed");
+	__frame_ptr = kmalloc(4 * HP_NUM_WORDS, GFP_KERNEL);
+	if (!__frame_ptr)
+		panic("kmalloc() failed");
+	frame_ptr = (void *)(((unsigned long)__frame_ptr + 63) &
+				~(unsigned long)63);
+	for (loop = 0; loop < HP_NUM_WORDS; loop++) {
+		frame_ptr[loop] = lfsr;
+		lfsr = do_lfsr(lfsr);
+	}
+	frame_dma = dma_map_single(&pdev->dev, frame_ptr, 4 * HP_NUM_WORDS,
+					DMA_BIDIRECTIONAL);
+	platform_device_del(pdev);
+	platform_device_put(pdev);
+}
+
+static void deallocate_frame_data(void)
+{
+	kfree(__frame_ptr);
+}
+
+static inline void process_frame_data(struct hp_handler *handler,
+				const struct qm_fd *fd)
+{
+	u32 *p = handler->frame_ptr;
+	u32 lfsr = HP_FIRST_WORD;
+	int loop;
+	if (qm_fd_addr_get64(fd) != handler->addr)
+		panic("bad frame address");
+	for (loop = 0; loop < HP_NUM_WORDS; loop++, p++) {
+		*p ^= handler->rx_mixer;
+		if (*p != lfsr)
+			panic("corrupt frame data");
+		*p ^= handler->tx_mixer;
+		lfsr = do_lfsr(lfsr);
+	}
+}
+
+static enum qman_cb_dqrr_result normal_dqrr(struct qman_portal *portal,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dqrr)
+{
+	struct hp_handler *handler = (struct hp_handler *)fq;
+
+	process_frame_data(handler, &dqrr->fd);
+	if (qman_enqueue(&handler->tx, &dqrr->fd, 0))
+		panic("qman_enqueue() failed");
+	return qman_cb_dqrr_consume;
+}
+
+static enum qman_cb_dqrr_result special_dqrr(struct qman_portal *portal,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dqrr)
+{
+	struct hp_handler *handler = (struct hp_handler *)fq;
+
+	process_frame_data(handler, &dqrr->fd);
+	if (++loop_counter < HP_LOOPS) {
+		if (qman_enqueue(&handler->tx, &dqrr->fd, 0))
+			panic("qman_enqueue() failed");
+	} else {
+		pr_info("Received final (%dth) frame\n", loop_counter);
+		wake_up(&queue);
+	}
+	return qman_cb_dqrr_consume;
+}
+
+static void create_per_cpu_handlers(void)
+{
+	struct hp_handler *handler;
+	int loop;
+	struct hp_cpu *hp_cpu = &__get_cpu_var(hp_cpus);
+
+	hp_cpu->processor_id = smp_processor_id();
+	spin_lock(&hp_lock);
+	list_add_tail(&hp_cpu->node, &hp_cpu_list);
+	hp_cpu_list_length++;
+	spin_unlock(&hp_lock);
+	INIT_LIST_HEAD(&hp_cpu->handlers);
+	for (loop = 0; loop < HP_PER_CPU; loop++) {
+		handler = kmem_cache_alloc(hp_handler_slab, GFP_KERNEL);
+		if (!handler)
+			panic("kmem_cache_alloc() failed");
+		handler->processor_id = hp_cpu->processor_id;
+		handler->addr = frame_dma;
+		handler->frame_ptr = frame_ptr;
+		list_add_tail(&handler->node, &hp_cpu->handlers);
+	}
+}
+
+static void destroy_per_cpu_handlers(void)
+{
+	struct list_head *loop, *tmp;
+	struct hp_cpu *hp_cpu = &__get_cpu_var(hp_cpus);
+
+	spin_lock(&hp_lock);
+	list_del(&hp_cpu->node);
+	spin_unlock(&hp_lock);
+	list_for_each_safe(loop, tmp, &hp_cpu->handlers) {
+		u32 flags;
+		struct hp_handler *handler = list_entry(loop, struct hp_handler,
+							node);
+		if (qman_retire_fq(&handler->rx, &flags))
+			panic("qman_retire_fq(rx) failed");
+		BUG_ON(flags & QMAN_FQ_STATE_BLOCKOOS);
+		if (qman_oos_fq(&handler->rx))
+			panic("qman_oos_fq(rx) failed");
+		qman_destroy_fq(&handler->rx, 0);
+		qman_destroy_fq(&handler->tx, 0);
+		qm_fq_free(handler->fqid_rx);
+		list_del(&handler->node);
+		kmem_cache_free(hp_handler_slab, handler);
+	}
+}
+
+static inline u8 num_cachelines(u32 offset)
+{
+	u8 res = (offset + (L1_CACHE_BYTES - 1))
+			 / (L1_CACHE_BYTES);
+	if (res > 3)
+		return 3;
+	return res;
+}
+#define STASH_DATA_CL \
+	num_cachelines(HP_NUM_WORDS * 4)
+#define STASH_CTX_CL \
+	num_cachelines(offsetof(struct hp_handler,fqid_rx))
+
+static void init_handler(void *__handler)
+{
+	struct qm_mcc_initfq opts;
+	struct hp_handler *handler = __handler;
+	BUG_ON(handler->processor_id != smp_processor_id());
+	/* Set up rx */
+	memset(&handler->rx, 0, sizeof(handler->rx));
+	if (handler == special_handler)
+		handler->rx.cb.dqrr = special_dqrr;
+	else
+		handler->rx.cb.dqrr = normal_dqrr;
+	if (qman_create_fq(handler->fqid_rx, 0, &handler->rx))
+		panic("qman_create_fq(rx) failed");
+	memset(&opts, 0, sizeof(opts));
+	opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA;
+	opts.fqd.fq_ctrl = QM_FQCTRL_CTXASTASHING;
+	opts.fqd.context_a.stashing.data_cl = STASH_DATA_CL;
+	opts.fqd.context_a.stashing.context_cl = STASH_CTX_CL;
+	if (qman_init_fq(&handler->rx, QMAN_INITFQ_FLAG_SCHED |
+				QMAN_INITFQ_FLAG_LOCAL, &opts))
+		panic("qman_init_fq(rx) failed");
+	/* Set up tx */
+	memset(&handler->tx, 0, sizeof(handler->tx));
+	if (qman_create_fq(handler->fqid_tx, QMAN_FQ_FLAG_NO_MODIFY,
+				&handler->tx))
+		panic("qman_create_fq(tx) failed");
+}
+
+static void init_phase2(void)
+{
+	int loop;
+	u32 fqid = 0;
+	u32 lfsr = 0xdeadbeef;
+	struct hp_cpu *hp_cpu;
+	struct hp_handler *handler;
+
+	for (loop = 0; loop < HP_PER_CPU; loop++) {
+		list_for_each_entry(hp_cpu, &hp_cpu_list, node) {
+			if (!loop)
+				hp_cpu->iterator = list_first_entry(
+						&hp_cpu->handlers,
+						struct hp_handler, node);
+			else
+				hp_cpu->iterator = list_entry(
+						hp_cpu->iterator->node.next,
+						struct hp_handler, node);
+			/* Rx FQID is the previous handler's Tx FQID */
+			hp_cpu->iterator->fqid_rx = fqid;
+			/* Allocate new FQID for Tx */
+			fqid = qm_fq_new();
+			if (!fqid)
+				panic("qm_fq_new() failed");
+			hp_cpu->iterator->fqid_tx = fqid;
+			/* Rx mixer is the previous handler's Tx mixer */
+			hp_cpu->iterator->rx_mixer = lfsr;
+			/* Get new mixer for Tx */
+			lfsr = do_lfsr(lfsr);
+			hp_cpu->iterator->tx_mixer = lfsr;
+		}
+	}
+	/* Fix up the first handler (fqid_rx==0, rx_mixer=0xdeadbeef) */
+	hp_cpu = list_first_entry(&hp_cpu_list, struct hp_cpu, node);
+	handler = list_first_entry(&hp_cpu->handlers, struct hp_handler, node);
+	BUG_ON((handler->fqid_rx != 0) || (handler->rx_mixer != 0xdeadbeef));
+	handler->fqid_rx = fqid;
+	handler->rx_mixer = lfsr;
+	/* and tag it as our "special" handler */
+	special_handler = handler;
+}
+
+static void init_phase3(void)
+{
+	int loop;
+	struct hp_cpu *hp_cpu;
+
+	for (loop = 0; loop < HP_PER_CPU; loop++) {
+		list_for_each_entry(hp_cpu, &hp_cpu_list, node) {
+			if (!loop)
+				hp_cpu->iterator = list_first_entry(
+						&hp_cpu->handlers,
+						struct hp_handler, node);
+			else
+				hp_cpu->iterator = list_entry(
+						hp_cpu->iterator->node.next,
+						struct hp_handler, node);
+			preempt_disable();
+			if (hp_cpu->processor_id == smp_processor_id())
+				init_handler(hp_cpu->iterator);
+			else
+				smp_call_function_single(hp_cpu->processor_id,
+					init_handler, hp_cpu->iterator, 1);
+			preempt_enable();
+		}
+	}
+}
+
+static void send_first_frame(void *ignore)
+{
+	u32 *p = special_handler->frame_ptr;
+	u32 lfsr = HP_FIRST_WORD;
+	int loop;
+	struct qm_fd fd;
+
+	BUG_ON(special_handler->processor_id != smp_processor_id());
+	memset(&fd, 0, sizeof(fd));
+	qm_fd_addr_set64(&fd, special_handler->addr);
+	fd.format = qm_fd_contig_big;
+	fd.length29 = HP_NUM_WORDS * 4;
+	for (loop = 0; loop < HP_NUM_WORDS; loop++, p++) {
+		if (*p != lfsr)
+			panic("corrupt frame data");
+		*p ^= special_handler->tx_mixer;
+		lfsr = do_lfsr(lfsr);
+	}
+	pr_info("Sending first frame\n");
+	if (qman_enqueue(&special_handler->tx, &fd, 0))
+		panic("qman_enqueue() failed");
+}
+
+void qman_test_hotpotato(void)
+{
+	if (cpumask_weight(cpu_online_mask) < 2) {
+		pr_info("qman_test_hotpotato, skip - only 1 CPU\n");
+		return;
+	}
+
+	pr_info("qman_test_hotpotato starting\n");
+
+	hp_cpu_list_length = 0;
+	loop_counter = 0;
+	hp_handler_slab = kmem_cache_create("hp_handler_slab",
+			sizeof(struct hp_handler), L1_CACHE_BYTES,
+			SLAB_HWCACHE_ALIGN, NULL);
+	if (!hp_handler_slab)
+		panic("kmem_cache_create() failed");
+
+	allocate_frame_data();
+
+	/* Init phase 1 */
+	pr_info("Creating %d handlers per cpu...\n", HP_PER_CPU);
+	if (on_all_cpus(create_per_cpu_handlers))
+		panic("on_each_cpu() failed");
+	pr_info("Number of cpus: %d, total of %d handlers\n",
+		hp_cpu_list_length, hp_cpu_list_length * HP_PER_CPU);
+
+	init_phase2();
+
+	init_phase3();
+
+	preempt_disable();
+	if (special_handler->processor_id == smp_processor_id())
+		send_first_frame(NULL);
+	else
+		smp_call_function_single(special_handler->processor_id,
+			send_first_frame, NULL, 1);
+	preempt_enable();
+
+	wait_event(queue, loop_counter == HP_LOOPS);
+	deallocate_frame_data();
+	if (on_all_cpus(destroy_per_cpu_handlers))
+		panic("on_each_cpu() failed");
+	kmem_cache_destroy(hp_handler_slab);
+	pr_info("qman_test_hotpotato finished\n");
+}
+
diff --git a/drivers/staging/fsl_qbman/qman_utility.c b/drivers/staging/fsl_qbman/qman_utility.c
new file mode 100644
index 0000000..a190f3d
--- /dev/null
+++ b/drivers/staging/fsl_qbman/qman_utility.c
@@ -0,0 +1,131 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "qman_private.h"
+
+/* ----------------- */
+/* --- FQID Pool --- */
+
+struct qman_fqid_pool {
+	/* Base and size of the FQID range */
+	u32 fqid_base;
+	u32 total;
+	/* Number of FQIDs currently "allocated" */
+	u32 used;
+	/* Allocation optimisation. When 'used<total', it is the index of an
+	 * available FQID. Otherwise there are no available FQIDs, and this
+	 * will be set when the next deallocation occurs. */
+	u32 next;
+	/* A bit-field representation of the FQID range. */
+	unsigned long *bits;
+};
+
+#define QLONG_BYTES	sizeof(unsigned long)
+#define QLONG_BITS	(QLONG_BYTES * 8)
+/* Number of 'longs' required for the given number of bits */
+#define QNUM_LONGS(b)	(((b) + QLONG_BITS - 1) / QLONG_BITS)
+/* Shorthand for the number of bytes of same (kmalloc, memset, etc) */
+#define QNUM_BYTES(b)	(QNUM_LONGS(b) * QLONG_BYTES)
+/* And in bits */
+#define QNUM_BITS(b)	(QNUM_LONGS(b) * QLONG_BITS)
+
+struct qman_fqid_pool *qman_fqid_pool_create(u32 fqid_start, u32 num)
+{
+	struct qman_fqid_pool *pool = kmalloc(sizeof(*pool), GFP_KERNEL);
+	unsigned int i;
+
+	BUG_ON(!num);
+	if (!pool)
+		return NULL;
+	pool->fqid_base = fqid_start;
+	pool->total = num;
+	pool->used = 0;
+	pool->next = 0;
+	pool->bits = kmalloc(QNUM_BYTES(num), GFP_KERNEL);
+	if (!pool->bits) {
+		kfree(pool);
+		return NULL;
+	}
+	memset(pool->bits, 0, QNUM_BYTES(num));
+	/* If num is not an even multiple of QLONG_BITS (or even 8, for
+	 * byte-oriented searching) then we fill the trailing bits with 1, to
+	 * make them look allocated (permanently). */
+	for (i = num + 1; i < QNUM_BITS(num); i++)
+		set_bit(i, pool->bits);
+	return pool;
+}
+EXPORT_SYMBOL(qman_fqid_pool_create);
+
+int qman_fqid_pool_destroy(struct qman_fqid_pool *pool)
+{
+	int ret = pool->used;
+	kfree(pool->bits);
+	kfree(pool);
+	return ret;
+}
+EXPORT_SYMBOL(qman_fqid_pool_destroy);
+
+int qman_fqid_pool_alloc(struct qman_fqid_pool *pool, u32 *fqid)
+{
+	int ret;
+	if (pool->used == pool->total)
+		return -ENOMEM;
+	*fqid = pool->fqid_base + pool->next;
+	ret = test_and_set_bit(pool->next, pool->bits);
+	BUG_ON(ret);
+	if (++pool->used == pool->total)
+		return 0;
+	pool->next = find_next_zero_bit(pool->bits, pool->total, pool->next);
+	if (pool->next >= pool->total)
+		pool->next = find_first_zero_bit(pool->bits, pool->total);
+	BUG_ON(pool->next >= pool->total);
+	return 0;
+}
+EXPORT_SYMBOL(qman_fqid_pool_alloc);
+
+void qman_fqid_pool_free(struct qman_fqid_pool *pool, u32 fqid)
+{
+	int ret;
+
+	fqid -= pool->fqid_base;
+	ret = test_and_clear_bit(fqid, pool->bits);
+	BUG_ON(!ret);
+	if (pool->used-- == pool->total)
+		pool->next = fqid;
+}
+EXPORT_SYMBOL(qman_fqid_pool_free);
+
+u32 qman_fqid_pool_used(struct qman_fqid_pool *pool)
+{
+	return pool->used;
+}
+EXPORT_SYMBOL(qman_fqid_pool_used);
+
diff --git a/include/linux/fsl_bman.h b/include/linux/fsl_bman.h
new file mode 100644
index 0000000..6976a8d
--- /dev/null
+++ b/include/linux/fsl_bman.h
@@ -0,0 +1,471 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef FSL_BMAN_H
+#define FSL_BMAN_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Last updated for v00.79 of the BG */
+
+/* Portal processing (interrupt) sources */
+#define BM_PIRQ_RCRI	0x00000002	/* RCR Ring (below threshold) */
+#define BM_PIRQ_BSCN	0x00000001	/* Buffer depletion State Change */
+
+/* This wrapper represents a bit-array for the depletion state of the 64 Bman
+ * buffer pools. */
+struct bman_depletion {
+	u32 __state[2];
+};
+#define BMAN_DEPLETION_EMPTY { { 0x00000000, 0x00000000 } }
+#define BMAN_DEPLETION_FULL { { 0xffffffff, 0xffffffff } }
+#define __bmdep_word(x) ((x) >> 5)
+#define __bmdep_shift(x) ((x) & 0x1f)
+#define __bmdep_bit(x) (0x80000000 >> __bmdep_shift(x))
+static inline void bman_depletion_init(struct bman_depletion *c)
+{
+	c->__state[0] = c->__state[1] = 0;
+}
+static inline void bman_depletion_fill(struct bman_depletion *c)
+{
+	c->__state[0] = c->__state[1] = ~0;
+}
+static inline int bman_depletion_get(const struct bman_depletion *c, u8 bpid)
+{
+	return c->__state[__bmdep_word(bpid)] & __bmdep_bit(bpid);
+}
+static inline void bman_depletion_set(struct bman_depletion *c, u8 bpid)
+{
+	c->__state[__bmdep_word(bpid)] |= __bmdep_bit(bpid);
+}
+static inline void bman_depletion_unset(struct bman_depletion *c, u8 bpid)
+{
+	c->__state[__bmdep_word(bpid)] &= ~__bmdep_bit(bpid);
+}
+
+/* ------------------------------------------------------- */
+/* --- Bman data structures (and associated constants) --- */
+
+/* Represents s/w corenet portal mapped data structures */
+struct bm_rcr_entry;	/* RCR (Release Command Ring) entries */
+struct bm_mc_command;	/* MC (Management Command) command */
+struct bm_mc_result;	/* MC result */
+
+/* Code-reduction, define a wrapper for 48-bit buffers. In cases where a buffer
+ * pool id specific to this buffer is needed (BM_RCR_VERB_CMD_BPID_MULTI,
+ * BM_MCC_VERB_ACQUIRE), the 'bpid' field is used. */
+struct bm_buffer {
+	union {
+		struct {
+			u8 __reserved1;
+			u8 bpid;
+			u16 hi; /* High 16-bits of 48-bit address */
+			u32 lo; /* Low 32-bits of 48-bit address */
+		};
+		struct {
+			u64 __notaddress:16;
+			u64 addr:48;
+		};
+	};
+} __attribute__((aligned(8)));
+static inline u64 bm_buffer_get64(const struct bm_buffer *buf)
+{
+	return buf->addr;
+}
+static inline dma_addr_t bm_buf_addr(const struct bm_buffer *buf)
+{
+	return (dma_addr_t)buf->addr;
+}
+/* Macro, so we compile better if 'v' isn't always 64-bit */
+/* Note: this first version is causing a noticable performance degradation,
+ * which needs analysis, so leaving it commented out for now. The second version
+ * achieves optimal performance. */
+#if 0
+#define bm_buffer_set64(buf, v) \
+	do { \
+		struct bm_buffer *__buf931 = (buf); \
+		__buf931->addr = v; \
+	} while (0)
+#else
+#define bm_buffer_set64(buf, v) \
+	do { \
+		struct bm_buffer *__buf931 = (buf); \
+		__buf931->hi = upper_32_bits(v); \
+		__buf931->lo = lower_32_bits(v); \
+	} while (0)
+#endif
+
+/* See 1.5.3.5.4: "Release Command" */
+struct bm_rcr_entry {
+	union {
+		struct {
+			u8 __dont_write_directly__verb;
+			u8 bpid; /* used with BM_RCR_VERB_CMD_BPID_SINGLE */
+			u8 __reserved1[62];
+		};
+		struct bm_buffer bufs[8];
+	};
+} __packed;
+#define BM_RCR_VERB_VBIT		0x80
+#define BM_RCR_VERB_CMD_MASK		0x70	/* one of two values; */
+#define BM_RCR_VERB_CMD_BPID_SINGLE	0x20
+#define BM_RCR_VERB_CMD_BPID_MULTI	0x30
+#define BM_RCR_VERB_BUFCOUNT_MASK	0x0f	/* values 1..8 */
+
+/* See 1.5.3.1: "Acquire Command" */
+/* See 1.5.3.2: "Query Command" */
+struct bm_mcc_acquire {
+	u8 bpid;
+	u8 __reserved1[62];
+} __packed;
+struct bm_mcc_query {
+	u8 __reserved2[63];
+} __packed;
+struct bm_mc_command {
+	u8 __dont_write_directly__verb;
+	union {
+		struct bm_mcc_acquire acquire;
+		struct bm_mcc_query query;
+	};
+} __packed;
+#define BM_MCC_VERB_VBIT		0x80
+#define BM_MCC_VERB_CMD_MASK		0x70	/* where the verb contains; */
+#define BM_MCC_VERB_CMD_ACQUIRE		0x10
+#define BM_MCC_VERB_CMD_QUERY		0x40
+#define BM_MCC_VERB_ACQUIRE_BUFCOUNT	0x0f	/* values 1..8 go here */
+
+/* See 1.5.3.3: "Acquire Reponse" */
+/* See 1.5.3.4: "Query Reponse" */
+struct bm_pool_state {
+	u8 __reserved1[32];
+	/* "availability state" and "depletion state" */
+	struct {
+		u8 __reserved1[8];
+		/* Access using bman_depletion_***() */
+		struct bman_depletion state;
+	} as, ds;
+};
+struct bm_mc_result {
+	union {
+		struct {
+			u8 verb;
+			u8 __reserved1[63];
+		};
+		union {
+			struct {
+				u8 __reserved1;
+				u8 bpid;
+				u8 __reserved2[62];
+			};
+			struct bm_buffer bufs[8];
+		} acquire;
+		struct bm_pool_state query;
+	};
+} __packed;
+#define BM_MCR_VERB_VBIT		0x80
+#define BM_MCR_VERB_CMD_MASK		BM_MCC_VERB_CMD_MASK
+#define BM_MCR_VERB_CMD_ACQUIRE		BM_MCC_VERB_CMD_ACQUIRE
+#define BM_MCR_VERB_CMD_QUERY		BM_MCC_VERB_CMD_QUERY
+#define BM_MCR_VERB_CMD_ERR_INVALID	0x60
+#define BM_MCR_VERB_CMD_ERR_ECC		0x70
+#define BM_MCR_VERB_ACQUIRE_BUFCOUNT	BM_MCC_VERB_ACQUIRE_BUFCOUNT /* 0..8 */
+/* Determine the "availability state" of pool 'p' from a query result 'r' */
+#define BM_MCR_QUERY_AVAILABILITY(r,p) bman_depletion_get(&r->query.as.state,p)
+/* Determine the "depletion state" of pool 'p' from a query result 'r' */
+#define BM_MCR_QUERY_DEPLETION(r,p) bman_depletion_get(&r->query.ds.state,p)
+
+/*******************************************************************/
+/* Managed (aka "shared" or "mux/demux") portal, high-level i/face */
+/*******************************************************************/
+
+	/* Portal and Buffer Pools */
+	/* ----------------------- */
+/* Represents a managed portal */
+struct bman_portal;
+
+/* This object type represents Bman buffer pools. */
+struct bman_pool;
+
+struct bman_portal_config {
+	/* This is used for any "core-affine" portals, ie. default portals
+	 * associated to the corresponding cpu. -1 implies that there is no core
+	 * affinity configured. */
+	int cpu;
+	/* portal interrupt line */
+	int irq;
+	/* the unique index of this portal */
+	u32 index;
+	/* Is this portal shared? (If so, it has coarser locking and demuxes
+	 * processing on behalf of other CPUs.) */
+	int is_shared;
+	/* These are the buffer pool IDs that may be used via this portal. */
+	struct bman_depletion mask;
+};
+
+/* This callback type is used when handling pool depletion entry/exit. The
+ * 'cb_ctx' value is the opaque value associated with the pool object in
+ * bman_new_pool(). 'depleted' is non-zero on depletion-entry, and zero on
+ * depletion-exit. */
+typedef void (*bman_cb_depletion)(struct bman_portal *bm,
+			struct bman_pool *pool, void *cb_ctx, int depleted);
+
+/* This struct specifies parameters for a bman_pool object. */
+struct bman_pool_params {
+	/* index of the buffer pool to encapsulate (0-63), ignored if
+	 * BMAN_POOL_FLAG_DYNAMIC_BPID is set. */
+	u32 bpid;
+	/* bit-mask of BMAN_POOL_FLAG_*** options */
+	u32 flags;
+	/* depletion-entry/exit callback, if BMAN_POOL_FLAG_DEPLETION is set */
+	bman_cb_depletion cb;
+	/* opaque user value passed as a parameter to 'cb' */
+	void *cb_ctx;
+	/* depletion-entry/exit thresholds, if BMAN_POOL_FLAG_THRESH is set. NB:
+	 * this is only allowed if BMAN_POOL_FLAG_DYNAMIC_BPID is used *and*
+	 * when run in the control plane (which controls Bman CCSR). This array
+	 * matches the definition of bm_pool_set(). */
+	u32 thresholds[4];
+};
+
+/* Flags to bman_new_pool() */
+#define BMAN_POOL_FLAG_NO_RELEASE    0x00000001 /* can't release to pool */
+#define BMAN_POOL_FLAG_ONLY_RELEASE  0x00000002 /* can only release to pool */
+#define BMAN_POOL_FLAG_DEPLETION     0x00000004 /* track depletion entry/exit */
+#define BMAN_POOL_FLAG_DYNAMIC_BPID  0x00000008 /* (de)allocate bpid */
+#define BMAN_POOL_FLAG_THRESH        0x00000010 /* set depletion thresholds */
+#define BMAN_POOL_FLAG_STOCKPILE     0x00000020 /* stockpile to reduce hw ops */
+
+/* Flags to bman_release() */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+#define BMAN_RELEASE_FLAG_WAIT       0x00000001 /* wait if RCR is full */
+#define BMAN_RELEASE_FLAG_WAIT_INT   0x00000002 /* if we wait, interruptible? */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+#define BMAN_RELEASE_FLAG_WAIT_SYNC  0x00000004 /* if wait, until consumed? */
+#endif
+#endif
+#define BMAN_RELEASE_FLAG_NOW        0x00000008 /* issue immediate release */
+
+/* Flags to bman_acquire() */
+#define BMAN_ACQUIRE_FLAG_STOCKPILE  0x00000001 /* no hw op, stockpile only */
+
+	/* Portal Management */
+	/* ----------------- */
+/**
+ * bman_get_portal_config - get portal configuration settings
+ *
+ * This returns a read-only view of the current cpu's affine portal settings.
+ */
+const struct bman_portal_config *bman_get_portal_config(void);
+
+/**
+ * bman_irqsource_get - return the portal work that is interrupt-driven
+ *
+ * Returns a bitmask of BM_PIRQ_**I processing sources that are currently
+ * enabled for interrupt handling on the current cpu's affine portal. These
+ * sources will trigger the portal interrupt and the interrupt handler (or a
+ * tasklet/bottom-half it defers to) will perform the corresponding processing
+ * work. The bman_poll_***() functions will only process sources that are not in
+ * this bitmask. If the current CPU is sharing a portal hosted on another CPU,
+ * this always returns zero.
+ */
+u32 bman_irqsource_get(void);
+
+/**
+ * bman_irqsource_add - add processing sources to be interrupt-driven
+ * @bits: bitmask of BM_PIRQ_**I processing sources
+ *
+ * Adds processing sources that should be interrupt-driven (rather than
+ * processed via bman_poll_***() functions). Returns zero for success, or
+ * -EINVAL if the current CPU is sharing a portal hosted on another CPU. */
+int bman_irqsource_add(u32 bits);
+
+/**
+ * bman_irqsource_remove - remove processing sources from being interrupt-driven
+ * @bits: bitmask of BM_PIRQ_**I processing sources
+ *
+ * Removes processing sources from being interrupt-driven, so that they will
+ * instead be processed via bman_poll_***() functions. Returns zero for success,
+ * or -EINVAL if the current CPU is sharing a portal hosted on another CPU. */
+int bman_irqsource_remove(u32 bits);
+
+/**
+ * bman_affine_cpus - return a mask of cpus that have affine portals
+ */
+const cpumask_t *bman_affine_cpus(void);
+
+/**
+ * bman_poll_slow - process anything that isn't interrupt-driven.
+ *
+ * This function does any portal processing that isn't interrupt-driven. If the
+ * current CPU is sharing a portal hosted on another CPU, this function will
+ * return -EINVAL, otherwise the return value is a bitmask of BM_PIRQ_* sources
+ * indicating what interrupt sources were actually processed by the call.
+ *
+ * NB, unlike the legacy wrapper bman_poll(), this function will
+ * deterministically check for the presence of portal processing work and do it,
+ * which implies some latency even if there's nothing to do. The bman_poll()
+ * wrapper on the other hand (like the qman_poll() wrapper) attenuates this by
+ * checking for (and doing) portal processing infrequently. Ie. such that
+ * qman_poll() and bman_poll() can be called from core-processing loops. Use
+ * bman_poll_slow() when you yourself are deciding when to incur the overhead of
+ * processing.
+ */
+u32 bman_poll_slow(void);
+
+/**
+ * bman_poll - process anything that isn't interrupt-driven.
+ *
+ * Dispatcher logic on a cpu can use this to trigger any maintenance of the
+ * affine portal. This function does whatever processing is not triggered by
+ * interrupts. This is a legacy wrapper that can be used in core-processing
+ * loops but mitigates the performance overhead of portal processing by
+ * adaptively bypassing true portal processing most of the time. (Processing is
+ * done once every 10 calls if the previous processing revealed that work needed
+ * to be done, or once very 1000 calls if the previous processing revealed no
+ * work needed doing.) If you wish to control this yourself, call
+ * bman_poll_slow() instead, which always checks for portal processing work.
+ */
+void bman_poll(void);
+
+/**
+ * bman_recovery_cleanup_bpid  - in recovery mode, cleanup a buffer pool
+ */
+int bman_recovery_cleanup_bpid(u32 bpid);
+
+/**
+ * bman_recovery_exit - leave recovery mode
+ */
+int bman_recovery_exit(void);
+
+/**
+ * bman_rcr_is_empty - Determine if portal's RCR is empty
+ *
+ * For use in situations where a cpu-affine caller needs to determine when all
+ * releases for the local portal have been processed by Bman but can't use the
+ * BMAN_RELEASE_FLAG_WAIT_SYNC flag to do this from the final bman_release().
+ * The function forces tracking of RCR consumption (which normally doesn't
+ * happen until release processing needs to find space to put new release
+ * commands), and returns zero if the ring still has unprocessed entries,
+ * non-zero if it is empty.
+ */
+int bman_rcr_is_empty(void);
+
+
+	/* Pool management */
+	/* --------------- */
+/**
+ * bman_new_pool - Allocates a Buffer Pool object
+ * @params: parameters specifying the buffer pool ID and behaviour
+ *
+ * Creates a pool object for the given @params. A portal and the depletion
+ * callback field of @params are only used if the BMAN_POOL_FLAG_DEPLETION flag
+ * is set. NB, the fields from @params are copied into the new pool object, so
+ * the structure provided by the caller can be released or reused after the
+ * function returns.
+ */
+struct bman_pool *bman_new_pool(const struct bman_pool_params *params);
+
+/**
+ * bman_free_pool - Deallocates a Buffer Pool object
+ * @pool: the pool object to release
+ *
+ */
+void bman_free_pool(struct bman_pool *pool);
+
+/**
+ * bman_get_params - Returns a pool object's parameters.
+ * @pool: the pool object
+ *
+ * The returned pointer refers to state within the pool object so must not be
+ * modified and can no longer be read once the pool object is destroyed.
+ */
+const struct bman_pool_params *bman_get_params(const struct bman_pool *pool);
+
+/**
+ * bman_release - Release buffer(s) to the buffer pool
+ * @pool: the buffer pool object to release to
+ * @bufs: an array of buffers to release
+ * @num: the number of buffers in @bufs (1-8)
+ * @flags: bit-mask of BMAN_RELEASE_FLAG_*** options
+ *
+ * Adds the given buffers to RCR entries. If the portal @p was created with the
+ * "COMPACT" flag, then it will be using a compaction algorithm to improve
+ * utilisation of RCR. As such, these buffers may join an existing ring entry
+ * and/or it may not be issued right away so as to allow future releases to join
+ * the same ring entry. Use the BMAN_RELEASE_FLAG_NOW flag to override this
+ * behaviour by committing the RCR entry (or entries) right away. If the RCR
+ * ring is full, the function will return -EBUSY unless BMAN_RELEASE_FLAG_WAIT
+ * is selected, in which case it will sleep waiting for space to become
+ * available in RCR. If the function receives a signal before such time (and
+ * BMAN_RELEASE_FLAG_WAIT_INT is set), the function returns -EINTR. Otherwise,
+ * it returns zero.
+ */
+int bman_release(struct bman_pool *pool, const struct bm_buffer *bufs, u8 num,
+			u32 flags);
+
+/**
+ * bman_acquire - Acquire buffer(s) from a buffer pool
+ * @pool: the buffer pool object to acquire from
+ * @bufs: array for storing the acquired buffers
+ * @num: the number of buffers desired (@bufs is at least this big)
+ *
+ * Issues an "Acquire" command via the portal's management command interface.
+ * The return value will be the number of buffers obtained from the pool, or a
+ * negative error code if a h/w error or pool starvation was encountered.
+ */
+int bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs, u8 num,
+			u32 flags);
+
+/**
+ * bman_flush_stockpile - Flush stockpile buffer(s) to the buffer pool
+ * @pool: the buffer pool object the stockpile belongs
+ * @flags: bit-mask of BMAN_RELEASE_FLAG_*** options
+ *
+ * Adds stockpile buffers to RCR entries until the stockpile is empty.
+ * The return value will be a negative error code if a h/w error occured.
+ * If BMAN_RELEASE_FLAG_NOW flag is passed and RCR ring is full,
+ * -EAGAIN will be returned.
+ */
+int bman_flush_stockpile(struct bman_pool *pool, u32 flags);
+
+/**
+ * bman_query_pools - Query all buffer pool states
+ * @state: storage for the queried availability and depletion states
+ */
+int bman_query_pools(struct bm_pool_state *state);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* FSL_BMAN_H */
diff --git a/include/linux/fsl_qman.h b/include/linux/fsl_qman.h
new file mode 100644
index 0000000..8a95f3b
--- /dev/null
+++ b/include/linux/fsl_qman.h
@@ -0,0 +1,1761 @@ 
+/* Copyright 2008-2011 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef FSL_QMAN_H
+#define FSL_QMAN_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Last updated for v00.800 of the BG */
+
+/* Hardware constants */
+enum qm_channel {
+	qm_channel_swportal0 = 0, qm_channel_swportal1, qm_channel_swportal2,
+	qm_channel_swportal3, qm_channel_swportal4, qm_channel_swportal5,
+	qm_channel_swportal6, qm_channel_swportal7, qm_channel_swportal8,
+	qm_channel_swportal9,
+	qm_channel_pool1 = 0x21, qm_channel_pool2, qm_channel_pool3,
+	qm_channel_pool4, qm_channel_pool5, qm_channel_pool6,
+	qm_channel_pool7, qm_channel_pool8, qm_channel_pool9,
+	qm_channel_pool10, qm_channel_pool11, qm_channel_pool12,
+	qm_channel_pool13, qm_channel_pool14, qm_channel_pool15,
+	qm_channel_fman0_sp0 = 0x40, qm_channel_fman0_sp1, qm_channel_fman0_sp2,
+	qm_channel_fman0_sp3, qm_channel_fman0_sp4, qm_channel_fman0_sp5,
+	qm_channel_fman0_sp6, qm_channel_fman0_sp7, qm_channel_fman0_sp8,
+	qm_channel_fman0_sp9, qm_channel_fman0_sp10, qm_channel_fman0_sp11,
+	qm_channel_fman1_sp0 = 0x60, qm_channel_fman1_sp1, qm_channel_fman1_sp2,
+	qm_channel_fman1_sp3, qm_channel_fman1_sp4, qm_channel_fman1_sp5,
+	qm_channel_fman1_sp6, qm_channel_fman1_sp7, qm_channel_fman1_sp8,
+	qm_channel_fman1_sp9, qm_channel_fman1_sp10, qm_channel_fman1_sp11,
+	qm_channel_caam = 0x80,
+	qm_channel_pme = 0xa0,
+};
+enum qm_dc_portal {
+	qm_dc_portal_fman0 = 0,
+	qm_dc_portal_fman1 = 1,
+	qm_dc_portal_caam = 2,
+	qm_dc_portal_pme = 3
+};
+
+/* Portal processing (interrupt) sources */
+#define QM_PIRQ_CSCI	0x00100000	/* Congestion State Change */
+#define QM_PIRQ_EQCI	0x00080000	/* Enqueue Command Committed */
+#define QM_PIRQ_EQRI	0x00040000	/* EQCR Ring (below threshold) */
+#define QM_PIRQ_DQRI	0x00020000	/* DQRR Ring (non-empty) */
+#define QM_PIRQ_MRI	0x00010000	/* MR Ring (non-empty) */
+/* This mask contains all the interrupt sources that need handling except DQRI,
+ * ie. that if present should trigger slow-path processing. */
+#define QM_PIRQ_SLOW	(QM_PIRQ_CSCI | QM_PIRQ_EQCI | QM_PIRQ_EQRI | \
+			QM_PIRQ_MRI)
+
+
+/* ------------------------ */
+/* --- FQ allocator API --- */
+
+/* Flags to qm_fq_free_flags() */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+#define QM_FQ_FREE_WAIT       0x00000001 /* wait if RCR is full */
+#define QM_FQ_FREE_WAIT_INT   0x00000002 /* if wait, interruptible? */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+#define QM_FQ_FREE_WAIT_SYNC  0x00000004 /* if wait, until consumed? */
+#endif
+#endif
+
+/* Allocate an unused FQID from the FQ allocator, returns zero for failure */
+u32 qm_fq_new(void);
+/* Release a FQID back to the FQ allocator */
+int qm_fq_free_flags(u32 fqid, u32 flags);
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+static inline void qm_fq_free(u32 fqid)
+{
+	if (qm_fq_free_flags(fqid, QM_FQ_FREE_WAIT))
+		BUG();
+}
+#else
+#define qm_fq_free(fqid) qm_fq_free_flags(fqid, 0)
+#endif
+
+/* For qman_static_dequeue_*** APIs */
+#define QM_SDQCR_CHANNELS_POOL_MASK	0x00007fff
+/* for n in [1,15] */
+#define QM_SDQCR_CHANNELS_POOL(n)	(0x00008000 >> (n))
+/* for conversion from n of type "enum qm_channel" */
+#define QM_SDQCR_CHANNELS_POOL_CONV(c) \
+	QM_SDQCR_CHANNELS_POOL((c) + 1 - qm_channel_pool1)
+
+/* For qman_volatile_dequeue(); Choose one PRECEDENCE. EXACT is optional. Use
+ * NUMFRAMES(n) (6-bit) or NUMFRAMES_TILLEMPTY to fill in the frame-count. Use
+ * FQID(n) to fill in the frame queue ID. */
+#define QM_VDQCR_PRECEDENCE_VDQCR	0x0
+#define QM_VDQCR_PRECEDENCE_SDQCR	0x80000000
+#define QM_VDQCR_EXACT			0x40000000
+#define QM_VDQCR_NUMFRAMES_MASK		0x3f000000
+#define QM_VDQCR_NUMFRAMES_SET(n)	(((n) & 0x3f) << 24)
+#define QM_VDQCR_NUMFRAMES_GET(n)	(((n) >> 24) & 0x3f)
+#define QM_VDQCR_NUMFRAMES_TILLEMPTY	QM_VDQCR_NUMFRAMES_SET(0)
+
+
+/* ------------------------------------------------------- */
+/* --- Qman data structures (and associated constants) --- */
+
+/* Represents s/w corenet portal mapped data structures */
+struct qm_eqcr_entry;	/* EQCR (EnQueue Command Ring) entries */
+struct qm_dqrr_entry;	/* DQRR (DeQueue Response Ring) entries */
+struct qm_mr_entry;	/* MR (Message Ring) entries */
+struct qm_mc_command;	/* MC (Management Command) command */
+struct qm_mc_result;	/* MC result */
+
+/* See David Lapp's "Frame formats" document, "dpateam", Jan 07, 2008 */
+#define QM_FD_FORMAT_SG		0x4
+#define QM_FD_FORMAT_LONG	0x2
+#define QM_FD_FORMAT_COMPOUND	0x1
+enum qm_fd_format {
+	/* 'contig' implies a contiguous buffer, whereas 'sg' implies a
+	 * scatter-gather table. 'big' implies a 29-bit length with no offset
+	 * field, otherwise length is 20-bit and offset is 9-bit. 'compound'
+	 * implies a s/g-like table, where each entry itself represents a frame
+	 * (contiguous or scatter-gather) and the 29-bit "length" is
+	 * interpreted purely for congestion calculations, ie. a "congestion
+	 * weight". */
+	qm_fd_contig = 0,
+	qm_fd_contig_big = QM_FD_FORMAT_LONG,
+	qm_fd_sg = QM_FD_FORMAT_SG,
+	qm_fd_sg_big = QM_FD_FORMAT_SG | QM_FD_FORMAT_LONG,
+	qm_fd_compound = QM_FD_FORMAT_COMPOUND
+};
+
+/* Capitalised versions are un-typed but can be used in static expressions */
+#define QM_FD_CONTIG	0
+#define QM_FD_CONTIG_BIG QM_FD_FORMAT_LONG
+#define QM_FD_SG	QM_FD_FORMAT_SG
+#define QM_FD_SG_BIG	(QM_FD_FORMAT_SG | QM_FD_FORMAT_LONG)
+#define QM_FD_COMPOUND	QM_FD_FORMAT_COMPOUND
+
+/* See 1.5.1.1: "Frame Descriptor (FD)" */
+struct qm_fd {
+	union {
+		struct {
+			u8 dd:2;	/* dynamic debug */
+			u8 liodn_offset:6;
+			u8 bpid:8;	/* Buffer Pool ID */
+			u8 eliodn_offset:4;
+			u8 __reserved:4;
+			u8 addr_hi;	/* high 8-bits of 40-bit address */
+			u32 addr_lo;	/* low 32-bits of 40-bit address */
+		};
+		struct {
+			u64 __notaddress:24;
+			/* More efficient address accessor */
+			u64 addr:40;
+		};
+		u64 opaque_addr;
+	};
+	/* The 'format' field indicates the interpretation of the remaining 29
+	 * bits of the 32-bit word. For packing reasons, it is duplicated in the
+	 * other union elements. Note, union'd structs are difficult to use with
+	 * static initialisation under gcc, in which case use the "opaque" form
+	 * with one of the macros. */
+	union {
+		/* For easier/faster copying of this part of the fd (eg. from a
+		 * DQRR entry to an EQCR entry) copy 'opaque' */
+		u32 opaque;
+		/* If 'format' is _contig or _sg, 20b length and 9b offset */
+		struct {
+			enum qm_fd_format format:3;
+			u16 offset:9;
+			u32 length20:20;
+		};
+		/* If 'format' is _contig_big or _sg_big, 29b length */
+		struct {
+			enum qm_fd_format _format1:3;
+			u32 length29:29;
+		};
+		/* If 'format' is _compound, 29b "congestion weight" */
+		struct {
+			enum qm_fd_format _format2:3;
+			u32 cong_weight:29;
+		};
+	};
+	union {
+		u32 cmd;
+		u32 status;
+	};
+} __attribute__((aligned(8)));
+#define QM_FD_DD_NULL		0x00
+#define QM_FD_PID_MASK		0x3f
+static inline u64 qm_fd_addr_get64(const struct qm_fd *fd)
+{
+	return fd->addr;
+}
+
+static inline dma_addr_t qm_fd_addr(const struct qm_fd *fd)
+{
+	return (dma_addr_t)fd->addr;
+}
+/* Macro, so we compile better if 'v' isn't always 64-bit */
+#define qm_fd_addr_set64(fd, v) \
+	do { \
+		struct qm_fd *__fd931 = (fd); \
+		__fd931->addr = v; \
+	} while (0)
+
+/* For static initialisation of FDs (which is complicated by the use of unions
+ * in "struct qm_fd"), use the following macros. Note that;
+ * - 'dd', 'pid' and 'bpid' are ignored because there's no static initialisation
+ *   use-case),
+ * - use capitalised QM_FD_*** formats for static initialisation.
+ */
+#define QM_FD_FMT_20(cmd, addr_hi, addr_lo, fmt, off, len) \
+	{ 0, 0, 0, 0, 0, addr_hi, addr_lo, \
+	{ (((fmt)&0x7) << 29) | (((off)&0x1ff) << 20) | ((len)&0xfffff) }, \
+	{ cmd } }
+#define QM_FD_FMT_29(cmd, addr_hi, addr_lo, fmt, len) \
+	{ 0, 0, 0, 0, 0, addr_hi, addr_lo, \
+	{ (((fmt)&0x7) << 29) | ((len)&0x1fffffff) }, \
+	{ cmd } }
+
+/* See 2.2.1.3 Multi-Core Datapath Acceleration Architecture */
+struct qm_sg_entry {
+	union {
+		struct {
+			u8 __reserved1[3];
+			u8 addr_hi;	/* high 8-bits of 40-bit address */
+			u32 addr_lo;	/* low 32-bits of 40-bit address */
+		};
+		struct {
+			u64 __notaddress:24;
+			u64 addr:40;
+		};
+	};
+	u32 extension:1;	/* Extension bit */
+	u32 final:1; 		/* Final bit */
+	u32 length:30;
+	u8 __reserved2;
+	u8 bpid;
+	u16 __reserved3:3;
+	u16 offset:13;
+} __packed;
+static inline u64 qm_sg_entry_get64(const struct qm_sg_entry *sg)
+{
+	return sg->addr;
+}
+static inline dma_addr_t qm_sg_addr(const struct qm_sg_entry *sg)
+{
+	return (dma_addr_t)sg->addr;
+}
+/* Macro, so we compile better if 'v' isn't always 64-bit */
+#define qm_sg_entry_set64(sg, v) \
+	do { \
+		struct qm_sg_entry *__sg931 = (sg); \
+		__sg931->addr = v; \
+	} while (0)
+
+/* See 1.5.8.1: "Enqueue Command" */
+struct qm_eqcr_entry {
+	u8 __dont_write_directly__verb;
+	u8 dca;
+	u16 seqnum;
+	u32 orp;	/* 24-bit */
+	u32 fqid;	/* 24-bit */
+	u32 tag;
+	struct qm_fd fd;
+	u8 __reserved3[32];
+} __packed;
+#define QM_EQCR_VERB_VBIT		0x80
+#define QM_EQCR_VERB_CMD_MASK		0x61	/* but only one value; */
+#define QM_EQCR_VERB_CMD_ENQUEUE	0x01
+#define QM_EQCR_VERB_COLOUR_MASK	0x18	/* 4 possible values; */
+#define QM_EQCR_VERB_COLOUR_GREEN	0x00
+#define QM_EQCR_VERB_COLOUR_YELLOW	0x08
+#define QM_EQCR_VERB_COLOUR_RED		0x10
+#define QM_EQCR_VERB_COLOUR_OVERRIDE	0x18
+#define QM_EQCR_VERB_INTERRUPT		0x04	/* on command consumption */
+#define QM_EQCR_VERB_ORP		0x02	/* enable order restoration */
+#define QM_EQCR_DCA_ENABLE		0x80
+#define QM_EQCR_DCA_PARK		0x40
+#define QM_EQCR_DCA_IDXMASK		0x0f	/* "DQRR::idx" goes here */
+#define QM_EQCR_SEQNUM_NESN		0x8000	/* Advance NESN */
+#define QM_EQCR_SEQNUM_NLIS		0x4000	/* More fragments to come */
+#define QM_EQCR_SEQNUM_SEQMASK		0x3fff	/* sequence number goes here */
+#define QM_EQCR_FQID_NULL		0	/* eg. for an ORP seqnum hole */
+
+/* See 1.5.8.2: "Frame Dequeue Response" */
+struct qm_dqrr_entry {
+	u8 verb;
+	u8 stat;
+	u16 seqnum;	/* 15-bit */
+	u8 tok;
+	u8 __reserved2[3];
+	u32 fqid;	/* 24-bit */
+	u32 contextB;
+	struct qm_fd fd;
+	u8 __reserved4[32];
+};
+#define QM_DQRR_VERB_VBIT		0x80
+#define QM_DQRR_VERB_MASK		0x7f	/* where the verb contains; */
+#define QM_DQRR_VERB_FRAME_DEQUEUE	0x60	/* "this format" */
+#define QM_DQRR_STAT_FQ_EMPTY		0x80	/* FQ empty */
+#define QM_DQRR_STAT_FQ_HELDACTIVE	0x40	/* FQ held active */
+#define QM_DQRR_STAT_FQ_FORCEELIGIBLE	0x20	/* FQ was force-eligible'd */
+#define QM_DQRR_STAT_FD_VALID		0x10	/* has a non-NULL FD */
+#define QM_DQRR_STAT_UNSCHEDULED	0x02	/* Unscheduled dequeue */
+#define QM_DQRR_STAT_DQCR_EXPIRED	0x01	/* VDQCR or PDQCR expired*/
+
+/* See 1.5.8.3: "ERN Message Response" */
+/* See 1.5.8.4: "FQ State Change Notification" */
+struct qm_mr_entry {
+	u8 verb;
+	union {
+		struct {
+			u8 dca;
+			u16 seqnum;
+			u8 rc;		/* Rejection Code */
+			u32 orp:24;
+			u32 fqid;	/* 24-bit */
+			u32 tag;
+			struct qm_fd fd;
+		} __packed ern;
+		struct {
+			u8 colour:2;	/* See QM_MR_DCERN_COLOUR_* */
+			u8 __reserved1:4;
+			enum qm_dc_portal portal:2;
+			u16 __reserved2;
+			u8 rc;		/* Rejection Code */
+			u32 __reserved3:24;
+			u32 fqid;	/* 24-bit */
+			u32 tag;
+			struct qm_fd fd;
+		} __packed dcern;
+		struct {
+			u8 fqs;		/* Frame Queue Status */
+			u8 __reserved1[6];
+			u32 fqid;	/* 24-bit */
+			u32 contextB;
+			u8 __reserved2[16];
+		} __packed fq;		/* FQRN/FQRNI/FQRL/FQPN */
+	};
+	u8 __reserved2[32];
+} __packed;
+#define QM_MR_VERB_VBIT			0x80
+/* The "ern" VERB bits match QM_EQCR_VERB_*** so aren't reproduced here. ERNs
+ * originating from direct-connect portals ("dcern") use 0x20 as a verb which
+ * would be invalid as a s/w enqueue verb. A s/w ERN can be distinguished from
+ * the other MR types by noting if the 0x20 bit is unset. */
+#define QM_MR_VERB_TYPE_MASK		0x27
+#define QM_MR_VERB_DC_ERN		0x20
+#define QM_MR_VERB_FQRN			0x21
+#define QM_MR_VERB_FQRNI		0x22
+#define QM_MR_VERB_FQRL			0x23
+#define QM_MR_VERB_FQPN			0x24
+#define QM_MR_RC_MASK			0xf0	/* contains one of; */
+#define QM_MR_RC_CGR_TAILDROP		0x00
+#define QM_MR_RC_WRED			0x10
+#define QM_MR_RC_ERROR			0x20
+#define QM_MR_RC_ORPWINDOW_EARLY	0x30
+#define QM_MR_RC_ORPWINDOW_LATE		0x40
+#define QM_MR_RC_FQ_TAILDROP		0x50
+#define QM_MR_RC_ORPWINDOW_RETIRED	0x60
+#define QM_MR_FQS_ORLPRESENT		0x02	/* ORL fragments to come */
+#define QM_MR_FQS_NOTEMPTY		0x01	/* FQ has enqueued frames */
+#define QM_MR_DCERN_COLOUR_GREEN	0x00
+#define QM_MR_DCERN_COLOUR_YELLOW	0x01
+#define QM_MR_DCERN_COLOUR_RED		0x02
+#define QM_MR_DCERN_COLOUR_OVERRIDE	0x03
+
+/* An identical structure of FQD fields is present in the "Init FQ" command and
+ * the "Query FQ" result, it's suctioned out into the "struct qm_fqd" type.
+ * Within that, the 'stashing' and 'taildrop' pieces are also factored out, the
+ * latter has two inlines to assist with converting to/from the mant+exp
+ * representation. */
+struct qm_fqd_stashing {
+	/* See QM_STASHING_EXCL_<...> */
+	u8 exclusive;
+	u8 __reserved1:2;
+	/* Numbers of cachelines */
+	u8 annotation_cl:2;
+	u8 data_cl:2;
+	u8 context_cl:2;
+} __packed;
+struct qm_fqd_taildrop {
+	u16 __reserved1:3;
+	u16 mant:8;
+	u16 exp:5;
+} __packed;
+struct qm_fqd_oac {
+	/* See QM_OAC_<...> */
+	u8 oac:2; /* "Overhead Accounting Control" */
+	u8 __reserved1:6;
+	/* Two's-complement value (-128 to +127) */
+	signed char oal; /* "Overhead Accounting Length" */
+} __packed;
+struct qm_fqd {
+	union {
+		u8 orpc;
+		struct {
+			u8 __reserved1:2;
+			u8 orprws:3;
+			u8 oa:1;
+			u8 olws:2;
+		} __packed;
+	};
+	u8 cgid;
+	u16 fq_ctrl;	/* See QM_FQCTRL_<...> */
+	union {
+		u16 dest_wq;
+		struct {
+			u16 channel:13; /* enum qm_channel */
+			u16 wq:3;
+		} __packed dest;
+	};
+	u16 __reserved2:1;
+	u16 ics_cred:15;
+	/* For "Initialize Frame Queue" commands, the write-enable mask
+	 * determines whether 'td' or 'oac_init' is observed. For query
+	 * commands, this field is always 'td', and 'oac_query' (below) reflects
+	 * the Overhead ACcounting values. */
+	union {
+		struct qm_fqd_taildrop td;
+		struct qm_fqd_oac oac_init;
+	};
+	u32 context_b;
+	union {
+		/* Treat it as 64-bit opaque */
+		u64 opaque;
+		struct {
+			u32 hi;
+			u32 lo;
+		};
+		/* Treat it as s/w portal stashing config */
+		/* See 1.5.6.7.1: "FQD Context_A field used for [...] */
+		struct {
+			struct qm_fqd_stashing stashing;
+			/* 48-bit address of FQ context to
+			 * stash, must be cacheline-aligned */
+			u16 context_hi;
+			u32 context_lo;
+		} __packed;
+	} context_a;
+	struct qm_fqd_oac oac_query;
+} __packed;
+/* 64-bit converters for context_hi/lo */
+static inline u64 qm_fqd_stashing_get64(const struct qm_fqd *fqd)
+{
+	return ((u64)fqd->context_a.context_hi << 32) |
+		(u64)fqd->context_a.context_lo;
+}
+static inline dma_addr_t qm_fqd_stashing_addr(const struct qm_fqd *fqd)
+{
+	return (dma_addr_t)qm_fqd_stashing_get64(fqd);
+}
+static inline u64 qm_fqd_context_a_get64(const struct qm_fqd *fqd)
+{
+	return ((u64)fqd->context_a.hi << 32) |
+		(u64)fqd->context_a.lo;
+}
+/* Macro, so we compile better when 'v' isn't necessarily 64-bit */
+#define qm_fqd_stashing_set64(fqd, v) \
+	do { \
+		struct qm_fqd *__fqd931 = (fqd); \
+		__fqd931->context_a.context_hi = upper_32_bits(v); \
+		__fqd931->context_a.context_lo = lower_32_bits(v); \
+	} while (0)
+#define qm_fqd_context_a_set64(fqd, v) \
+	do { \
+		struct qm_fqd *__fqd931 = (fqd); \
+		__fqd931->context_a.hi = upper_32_bits(v); \
+		__fqd931->context_a.lo = lower_32_bits(v); \
+	} while (0)
+/* convert a threshold value into mant+exp representation */
+static inline int qm_fqd_taildrop_set(struct qm_fqd_taildrop *td, u32 val,
+					int roundup)
+{
+	u32 e = 0;
+	int oddbit = 0;
+	if (val > 0xe0000000)
+		return -ERANGE;
+	while (val > 0xff) {
+		oddbit = val & 1;
+		val >>= 1;
+		e++;
+		if (roundup && oddbit)
+			val++;
+	}
+	td->exp = e;
+	td->mant = val;
+	return 0;
+}
+/* and the other direction */
+static inline u32 qm_fqd_taildrop_get(const struct qm_fqd_taildrop *td)
+{
+	return (u32)td->mant << td->exp;
+}
+
+/* See 1.5.2.2: "Frame Queue Descriptor (FQD)" */
+/* Frame Queue Descriptor (FQD) field 'fq_ctrl' uses these constants */
+#define QM_FQCTRL_MASK		0x07ff	/* 'fq_ctrl' flags; */
+#define QM_FQCTRL_CGE		0x0400	/* Congestion Group Enable */
+#define QM_FQCTRL_TDE		0x0200	/* Tail-Drop Enable */
+#define QM_FQCTRL_ORP		0x0100	/* ORP Enable */
+#define QM_FQCTRL_CTXASTASHING	0x0080	/* Context-A stashing */
+#define QM_FQCTRL_CPCSTASH	0x0040	/* CPC Stash Enable */
+#define QM_FQCTRL_FORCESFDR	0x0008	/* High-priority SFDRs */
+#define QM_FQCTRL_AVOIDBLOCK	0x0004	/* Don't block active */
+#define QM_FQCTRL_HOLDACTIVE	0x0002	/* Hold active in portal */
+#define QM_FQCTRL_PREFERINCACHE	0x0001	/* Aggressively cache FQD */
+#define QM_FQCTRL_LOCKINCACHE	QM_FQCTRL_PREFERINCACHE /* older naming */
+
+/* See 1.5.6.7.1: "FQD Context_A field used for [...] */
+/* Frame Queue Descriptor (FQD) field 'CONTEXT_A' uses these constants */
+#define QM_STASHING_EXCL_ANNOTATION	0x04
+#define QM_STASHING_EXCL_DATA		0x02
+#define QM_STASHING_EXCL_CTX		0x01
+
+/* See 1.5.5.3: "Intra Class Scheduling" */
+/* FQD field 'OAC' (Overhead ACcounting) uses these constants */
+#define QM_OAC_ICS		0x2 /* Accounting for Intra-Class Scheduling */
+#define QM_OAC_CG		0x1 /* Accounting for Congestion Groups */
+
+/* See 1.5.8.4: "FQ State Change Notification" */
+/* This struct represents the 32-bit "WR_PARM_[GYR]" parameters in CGR fields
+ * and associated commands/responses. The WRED parameters are calculated from
+ * these fields as follows;
+ *   MaxTH = MA * (2 ^ Mn)
+ *   Slope = SA / (2 ^ Sn)
+ *    MaxP = 4 * (Pn + 1)
+ */
+struct qm_cgr_wr_parm {
+	union {
+		u32 word;
+		struct {
+			u32 MA:8;
+			u32 Mn:5;
+			u32 SA:7; /* must be between 64-127 */
+			u32 Sn:6;
+			u32 Pn:6;
+		} __packed;
+	};
+} __packed;
+/* This struct represents the 13-bit "CS_THRES" CGR field. In the corresponding
+ * management commands, this is padded to a 16-bit structure field, so that's
+ * how we represent it here. The congestion state threshold is calculated from
+ * these fields as follows;
+ *   CS threshold = TA * (2 ^ Tn)
+ */
+struct qm_cgr_cs_thres {
+	u16 __reserved:3;
+	u16 TA:8;
+	u16 Tn:5;
+} __packed;
+/* This identical structure of CGR fields is present in the "Init/Modify CGR"
+ * commands and the "Query CGR" result. It's suctioned out here into its own
+ * struct. */
+struct __qm_mc_cgr {
+	struct qm_cgr_wr_parm wr_parm_g;
+	struct qm_cgr_wr_parm wr_parm_y;
+	struct qm_cgr_wr_parm wr_parm_r;
+	u8 wr_en_g;	/* boolean, use QM_CGR_EN */
+	u8 wr_en_y;	/* boolean, use QM_CGR_EN */
+	u8 wr_en_r;	/* boolean, use QM_CGR_EN */
+	u8 cscn_en;	/* boolean, use QM_CGR_EN */
+	u32 cscn_targ;	/* use QM_CGR_TARG_* */
+	u8 cstd_en;	/* boolean, use QM_CGR_EN */
+	u8 cs;		/* boolean, only used in query response */
+	struct qm_cgr_cs_thres cs_thres; /* use qm_cgr_cs_thres_set64() */
+	u8 mode;	/* QMAN_CGR_MODE_FRAME not supported in rev1.0 */
+} __packed;
+#define QM_CGR_EN		0x01 /* For wr_en_*, cscn_en, cstd_en */
+#define QM_CGR_TARG_PORTAL(n)	(0x80000000 >> (n)) /* s/w portal, 0-9 */
+#define QM_CGR_TARG_FMAN0	0x00200000 /* direct-connect portal: fman0 */
+#define QM_CGR_TARG_FMAN1	0x00100000 /*                      : fman1 */
+/* Convert CGR thresholds to/from "cs_thres" format */
+static inline u64 qm_cgr_cs_thres_get64(const struct qm_cgr_cs_thres *th)
+{
+	return (u64)th->TA << th->Tn;
+}
+static inline int qm_cgr_cs_thres_set64(struct qm_cgr_cs_thres *th, u64 val,
+					int roundup)
+{
+	u32 e = 0;
+	int oddbit = 0;
+	while (val > 0xff) {
+		oddbit = val & 1;
+		val >>= 1;
+		e++;
+		if (roundup && oddbit)
+			val++;
+	}
+	th->Tn = e;
+	th->TA = val;
+	return 0;
+}
+
+/* See 1.5.8.5.1: "Initialize FQ" */
+/* See 1.5.8.5.2: "Query FQ" */
+/* See 1.5.8.5.3: "Query FQ Non-Programmable Fields" */
+/* See 1.5.8.5.4: "Alter FQ State Commands " */
+/* See 1.5.8.6.1: "Initialize/Modify CGR" */
+/* See 1.5.8.6.2: "CGR Test Write" */
+/* See 1.5.8.6.3: "Query CGR" */
+/* See 1.5.8.6.4: "Query Congestion Group State" */
+struct qm_mcc_initfq {
+	u8 __reserved1;
+	u16 we_mask;	/* Write Enable Mask */
+	u32 fqid;	/* 24-bit */
+	u16 count;	/* Initialises 'count+1' FQDs */
+	struct qm_fqd fqd; /* the FQD fields go here */
+	u8 __reserved3[30];
+} __packed;
+struct qm_mcc_queryfq {
+	u8 __reserved1[3];
+	u32 fqid;	/* 24-bit */
+	u8 __reserved2[56];
+} __packed;
+struct qm_mcc_queryfq_np {
+	u8 __reserved1[3];
+	u32 fqid;	/* 24-bit */
+	u8 __reserved2[56];
+} __packed;
+struct qm_mcc_alterfq {
+	u8 __reserved1[3];
+	u32 fqid;	/* 24-bit */
+	u8 __reserved2[56];
+} __packed;
+struct qm_mcc_initcgr {
+	u8 __reserved1;
+	u16 we_mask;	/* Write Enable Mask */
+	struct __qm_mc_cgr cgr;	/* CGR fields */
+	u8 __reserved2[2];
+	u8 cgid;
+	u8 __reserved4[32];
+} __packed;
+struct qm_mcc_cgrtestwrite {
+	u8 __reserved1[2];
+	u8 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */
+	u32 i_bcnt_lo;	/* low 32-bits of 40-bit */
+	u8 __reserved2[23];
+	u8 cgid;
+	u8 __reserved3[32];
+} __packed;
+struct qm_mcc_querycgr {
+	u8 __reserved1[30];
+	u8 cgid;
+	u8 __reserved2[32];
+} __packed;
+struct qm_mcc_querycongestion {
+	u8 __reserved[63];
+} __packed;
+struct qm_mcc_querywq {
+	u8 __reserved;
+	/* select channel if verb != QUERYWQ_DEDICATED */
+	union {
+		u16 channel_wq; /* ignores wq (3 lsbits) */
+		struct {
+			u16 id:13; /* enum qm_channel */
+			u16 __reserved1:3;
+		} __packed channel;
+	};
+	u8 __reserved2[60];
+} __packed;
+struct qm_mc_command {
+	u8 __dont_write_directly__verb;
+	union {
+		struct qm_mcc_initfq initfq;
+		struct qm_mcc_queryfq queryfq;
+		struct qm_mcc_queryfq_np queryfq_np;
+		struct qm_mcc_alterfq alterfq;
+		struct qm_mcc_initcgr initcgr;
+		struct qm_mcc_cgrtestwrite cgrtestwrite;
+		struct qm_mcc_querycgr querycgr;
+		struct qm_mcc_querycongestion querycongestion;
+		struct qm_mcc_querywq querywq;
+	};
+} __packed;
+#define QM_MCC_VERB_VBIT		0x80
+#define QM_MCC_VERB_MASK		0x7f	/* where the verb contains; */
+#define QM_MCC_VERB_INITFQ_PARKED	0x40
+#define QM_MCC_VERB_INITFQ_SCHED	0x41
+#define QM_MCC_VERB_QUERYFQ		0x44
+#define QM_MCC_VERB_QUERYFQ_NP		0x45	/* "non-programmable" fields */
+#define QM_MCC_VERB_QUERYWQ		0x46
+#define QM_MCC_VERB_QUERYWQ_DEDICATED	0x47
+#define QM_MCC_VERB_ALTER_SCHED		0x48	/* Schedule FQ */
+#define QM_MCC_VERB_ALTER_FE		0x49	/* Force Eligible FQ */
+#define QM_MCC_VERB_ALTER_RETIRE	0x4a	/* Retire FQ */
+#define QM_MCC_VERB_ALTER_OOS		0x4b	/* Take FQ out of service */
+#define QM_MCC_VERB_INITCGR		0x50
+#define QM_MCC_VERB_MODIFYCGR		0x51
+#define QM_MCC_VERB_CGRTESTWRITE	0x52
+#define QM_MCC_VERB_QUERYCGR		0x58
+#define QM_MCC_VERB_QUERYCONGESTION	0x59
+/* INITFQ-specific flags */
+#define QM_INITFQ_WE_MASK		0x01ff	/* 'Write Enable' flags; */
+#define QM_INITFQ_WE_OAC		0x0100
+#define QM_INITFQ_WE_ORPC		0x0080
+#define QM_INITFQ_WE_CGID		0x0040
+#define QM_INITFQ_WE_FQCTRL		0x0020
+#define QM_INITFQ_WE_DESTWQ		0x0010
+#define QM_INITFQ_WE_ICSCRED		0x0008
+#define QM_INITFQ_WE_TDTHRESH		0x0004
+#define QM_INITFQ_WE_CONTEXTB		0x0002
+#define QM_INITFQ_WE_CONTEXTA		0x0001
+/* INITCGR/MODIFYCGR-specific flags */
+#define QM_CGR_WE_MASK			0x07ff	/* 'Write Enable Mask'; */
+#define QM_CGR_WE_WR_PARM_G		0x0400
+#define QM_CGR_WE_WR_PARM_Y		0x0200
+#define QM_CGR_WE_WR_PARM_R		0x0100
+#define QM_CGR_WE_WR_EN_G		0x0080
+#define QM_CGR_WE_WR_EN_Y		0x0040
+#define QM_CGR_WE_WR_EN_R		0x0020
+#define QM_CGR_WE_CSCN_EN		0x0010
+#define QM_CGR_WE_CSCN_TARG		0x0008
+#define QM_CGR_WE_CSTD_EN		0x0004
+#define QM_CGR_WE_CS_THRES		0x0002
+#define QM_CGR_WE_MODE			0x0001
+
+/* See 1.5.8.5.1: "Initialize FQ" */
+/* See 1.5.8.5.2: "Query FQ" */
+/* See 1.5.8.5.3: "Query FQ Non-Programmable Fields" */
+/* See 1.5.8.5.4: "Alter FQ State Commands " */
+/* See 1.5.8.6.1: "Initialize/Modify CGR" */
+/* See 1.5.8.6.2: "CGR Test Write" */
+/* See 1.5.8.6.3: "Query CGR" */
+/* See 1.5.8.6.4: "Query Congestion Group State" */
+struct qm_mcr_initfq {
+	u8 __reserved1[62];
+} __packed;
+struct qm_mcr_queryfq {
+	u8 __reserved1[8];
+	struct qm_fqd fqd;	/* the FQD fields are here */
+	u8 __reserved2[30];
+} __packed;
+struct qm_mcr_queryfq_np {
+	u8 __reserved1;
+	u8 state;	/* QM_MCR_NP_STATE_*** */
+	u8 __reserved2;
+	u32 fqd_link:24;
+	u16 __reserved3:2;
+	u16 odp_seq:14;
+	u16 __reserved4:2;
+	u16 orp_nesn:14;
+	u16 __reserved5:1;
+	u16 orp_ea_hseq:15;
+	u16 __reserved6:1;
+	u16 orp_ea_tseq:15;
+	u8 __reserved7;
+	u32 orp_ea_hptr:24;
+	u8 __reserved8;
+	u32 orp_ea_tptr:24;
+	u8 __reserved9;
+	u32 pfdr_hptr:24;
+	u8 __reserved10;
+	u32 pfdr_tptr:24;
+	u8 __reserved11[5];
+	u8 __reserved12:7;
+	u8 is:1;
+	u16 ics_surp;
+	u32 byte_cnt;
+	u8 __reserved13;
+	u32 frm_cnt:24;
+	u32 __reserved14;
+	u16 ra1_sfdr;	/* QM_MCR_NP_RA1_*** */
+	u16 ra2_sfdr;	/* QM_MCR_NP_RA2_*** */
+	u16 __reserved15;
+	u16 od1_sfdr;	/* QM_MCR_NP_OD1_*** */
+	u16 od2_sfdr;	/* QM_MCR_NP_OD2_*** */
+	u16 od3_sfdr;	/* QM_MCR_NP_OD3_*** */
+} __packed;
+struct qm_mcr_alterfq {
+	u8 fqs;		/* Frame Queue Status */
+	u8 __reserved1[61];
+} __packed;
+struct qm_mcr_initcgr {
+	u8 __reserved1[62];
+} __packed;
+struct qm_mcr_cgrtestwrite {
+	u16 __reserved1;
+	struct __qm_mc_cgr cgr; /* CGR fields */
+	u8 __reserved2[3];
+	u32 __reserved3:24;
+	u32 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */
+	u32 i_bcnt_lo;	/* low 32-bits of 40-bit */
+	u32 __reserved4:24;
+	u32 a_bcnt_hi:8;/* high 8-bits of 40-bit "Average" */
+	u32 a_bcnt_lo;	/* low 32-bits of 40-bit */
+	u16 lgt;	/* Last Group Tick */
+	u16 wr_prob_g;
+	u16 wr_prob_y;
+	u16 wr_prob_r;
+	u8 __reserved5[8];
+} __packed;
+struct qm_mcr_querycgr {
+	u16 __reserved1;
+	struct __qm_mc_cgr cgr; /* CGR fields */
+	u8 __reserved2[3];
+	u32 __reserved3:24;
+	u32 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */
+	u32 i_bcnt_lo;	/* low 32-bits of 40-bit */
+	u32 __reserved4:24;
+	u32 a_bcnt_hi:8;/* high 8-bits of 40-bit "Average" */
+	u32 a_bcnt_lo;	/* low 32-bits of 40-bit */
+	u8 __reserved5[16];
+} __packed;
+static inline u64 qm_mcr_querycgr_i_get64(const struct qm_mcr_querycgr *q)
+{
+	return ((u64)q->i_bcnt_hi << 32) | (u64)q->i_bcnt_lo;
+}
+static inline u64 qm_mcr_querycgr_a_get64(const struct qm_mcr_querycgr *q)
+{
+	return ((u64)q->a_bcnt_hi << 32) | (u64)q->a_bcnt_lo;
+}
+static inline u64 qm_mcr_cgrtestwrite_i_get64(
+					const struct qm_mcr_cgrtestwrite *q)
+{
+	return ((u64)q->i_bcnt_hi << 32) | (u64)q->i_bcnt_lo;
+}
+static inline u64 qm_mcr_cgrtestwrite_a_get64(
+					const struct qm_mcr_cgrtestwrite *q)
+{
+	return ((u64)q->a_bcnt_hi << 32) | (u64)q->a_bcnt_lo;
+}
+/* Macro, so we compile better if 'v' isn't always 64-bit */
+#define qm_mcr_querycgr_i_set64(q, v) \
+	do { \
+		struct qm_mcr_querycgr *__q931 = (fd); \
+		__q931->i_bcnt_hi = upper_32_bits(v); \
+		__q931->i_bcnt_lo = lower_32_bits(v); \
+	} while (0)
+#define qm_mcr_querycgr_a_set64(q, v) \
+	do { \
+		struct qm_mcr_querycgr *__q931 = (fd); \
+		__q931->a_bcnt_hi = upper_32_bits(v); \
+		__q931->a_bcnt_lo = lower_32_bits(v); \
+	} while (0)
+struct __qm_mcr_querycongestion {
+	u32 __state[8];
+};
+struct qm_mcr_querycongestion {
+	u8 __reserved[30];
+	/* Access this struct using QM_MCR_QUERYCONGESTION() */
+	struct __qm_mcr_querycongestion state;
+} __packed;
+struct qm_mcr_querywq {
+	union {
+		u16 channel_wq; /* ignores wq (3 lsbits) */
+		struct {
+			u16 id:13; /* enum qm_channel */
+			u16 __reserved:3;
+		} __packed channel;
+	};
+	u8 __reserved[28];
+	u32 wq_len[8];
+} __packed;
+struct qm_mc_result {
+	u8 verb;
+	u8 result;
+	union {
+		struct qm_mcr_initfq initfq;
+		struct qm_mcr_queryfq queryfq;
+		struct qm_mcr_queryfq_np queryfq_np;
+		struct qm_mcr_alterfq alterfq;
+		struct qm_mcr_initcgr initcgr;
+		struct qm_mcr_cgrtestwrite cgrtestwrite;
+		struct qm_mcr_querycgr querycgr;
+		struct qm_mcr_querycongestion querycongestion;
+		struct qm_mcr_querywq querywq;
+	};
+} __packed;
+#define QM_MCR_VERB_RRID		0x80
+#define QM_MCR_VERB_MASK		QM_MCC_VERB_MASK
+#define QM_MCR_VERB_INITFQ_PARKED	QM_MCC_VERB_INITFQ_PARKED
+#define QM_MCR_VERB_INITFQ_SCHED	QM_MCC_VERB_INITFQ_SCHED
+#define QM_MCR_VERB_QUERYFQ		QM_MCC_VERB_QUERYFQ
+#define QM_MCR_VERB_QUERYFQ_NP		QM_MCC_VERB_QUERYFQ_NP
+#define QM_MCR_VERB_QUERYWQ		QM_MCC_VERB_QUERYWQ
+#define QM_MCR_VERB_QUERYWQ_DEDICATED	QM_MCC_VERB_QUERYWQ_DEDICATED
+#define QM_MCR_VERB_ALTER_SCHED		QM_MCC_VERB_ALTER_SCHED
+#define QM_MCR_VERB_ALTER_FE		QM_MCC_VERB_ALTER_FE
+#define QM_MCR_VERB_ALTER_RETIRE	QM_MCC_VERB_ALTER_RETIRE
+#define QM_MCR_VERB_ALTER_OOS		QM_MCC_VERB_ALTER_OOS
+#define QM_MCR_RESULT_NULL		0x00
+#define QM_MCR_RESULT_OK		0xf0
+#define QM_MCR_RESULT_ERR_FQID		0xf1
+#define QM_MCR_RESULT_ERR_FQSTATE	0xf2
+#define QM_MCR_RESULT_ERR_NOTEMPTY	0xf3	/* OOS fails if FQ is !empty */
+#define QM_MCR_RESULT_ERR_BADCHANNEL	0xf4
+#define QM_MCR_RESULT_PENDING		0xf8
+#define QM_MCR_RESULT_ERR_BADCOMMAND	0xff
+#define QM_MCR_NP_STATE_FE		0x10
+#define QM_MCR_NP_STATE_R		0x08
+#define QM_MCR_NP_STATE_MASK		0x07	/* Reads FQD::STATE; */
+#define QM_MCR_NP_STATE_OOS		0x00
+#define QM_MCR_NP_STATE_RETIRED		0x01
+#define QM_MCR_NP_STATE_TEN_SCHED	0x02
+#define QM_MCR_NP_STATE_TRU_SCHED	0x03
+#define QM_MCR_NP_STATE_PARKED		0x04
+#define QM_MCR_NP_STATE_ACTIVE		0x05
+#define QM_MCR_NP_PTR_MASK		0x07ff	/* for RA[12] & OD[123] */
+#define QM_MCR_NP_RA1_NRA(v)		(((v) >> 14) & 0x3)	/* FQD::NRA */
+#define QM_MCR_NP_RA2_IT(v)		(((v) >> 14) & 0x1)	/* FQD::IT */
+#define QM_MCR_NP_OD1_NOD(v)		(((v) >> 14) & 0x3)	/* FQD::NOD */
+#define QM_MCR_NP_OD3_NPC(v)		(((v) >> 14) & 0x3)	/* FQD::NPC */
+#define QM_MCR_FQS_ORLPRESENT		0x02	/* ORL fragments to come */
+#define QM_MCR_FQS_NOTEMPTY		0x01	/* FQ has enqueued frames */
+/* This extracts the state for congestion group 'n' from a query response.
+ * Eg.
+ *   u8 cgr = [...];
+ *   struct qm_mc_result *res = [...];
+ *   printf("congestion group %d congestion state: %d\n", cgr,
+ *       QM_MCR_QUERYCONGESTION(&res->querycongestion.state, cgr));
+ */
+#define __CGR_WORD(num)		(num >> 5)
+#define __CGR_SHIFT(num)	(num & 0x1f)
+#define __CGR_NUM		(sizeof(struct __qm_mcr_querycongestion) << 3)
+static inline int QM_MCR_QUERYCONGESTION(struct __qm_mcr_querycongestion *p,
+					u8 cgr)
+{
+	return p->__state[__CGR_WORD(cgr)] & (0x80000000 >> __CGR_SHIFT(cgr));
+}
+
+
+/*********************/
+/* Utility interface */
+/*********************/
+
+/* Represents an allocator over a range of FQIDs. NB, accesses are not locked,
+ * spinlock them yourself if needed. */
+struct qman_fqid_pool;
+
+/* Create/destroy a FQID pool, num must be a multiple of 32. NB, _destroy()
+ * always succeeds, but returns non-zero if there were "leaked" FQID
+ * allocations. */
+struct qman_fqid_pool *qman_fqid_pool_create(u32 fqid_start, u32 num);
+int qman_fqid_pool_destroy(struct qman_fqid_pool *pool);
+/* Alloc/free a FQID from the range. _alloc() returns zero for success. */
+int qman_fqid_pool_alloc(struct qman_fqid_pool *pool, u32 *fqid);
+void qman_fqid_pool_free(struct qman_fqid_pool *pool, u32 fqid);
+u32 qman_fqid_pool_used(struct qman_fqid_pool *pool);
+
+/*******************************************************************/
+/* Managed (aka "shared" or "mux/demux") portal, high-level i/face */
+/*******************************************************************/
+
+	/* Portal and Frame Queues */
+	/* ----------------------- */
+/* Represents a managed portal */
+struct qman_portal;
+
+/* This object type represents Qman frame queue descriptors (FQD), it is
+ * cacheline-aligned, and initialised by qman_create_fq(). The structure is
+ * defined further down. */
+struct qman_fq;
+
+/* This object type represents a Qman congestion group, it is defined further
+ * down. */
+struct qman_cgr;
+
+struct qman_portal_config {
+	/* If the caller enables DQRR stashing (and thus wishes to operate the
+	 * portal from only one cpu), this is the logical CPU that the portal
+	 * will stash to. Whether stashing is enabled or not, this setting is
+	 * also used for any "core-affine" portals, ie. default portals
+	 * associated to the corresponding cpu. -1 implies that there is no core
+	 * affinity configured. */
+	int cpu;
+	/* portal interrupt line */
+	int irq;
+	/* the unique index of this portal */
+	u32 index;
+	/* Is this portal shared? (If so, it has coarser locking and demuxes
+	 * processing on behalf of other CPUs.) */
+	int is_shared;
+	/* The portal's dedicated channel id, use this value for initialising
+	 * frame queues to target this portal when scheduled. */
+	enum qm_channel channel;
+	/* A mask of which pool channels this portal has dequeue access to
+	 * (using QM_SDQCR_CHANNELS_POOL(n) for the bitmask) */
+	u32 pools;
+	/* Is stashing available? (Depends on PAMU config) */
+	int has_stashing;
+};
+
+/* This enum, and the callback type that returns it, are used when handling
+ * dequeued frames via DQRR. Note that for "null" callbacks registered with the
+ * portal object (for handling dequeues that do not demux because contextB is
+ * NULL), the return value *MUST* be qman_cb_dqrr_consume. */
+enum qman_cb_dqrr_result {
+	/* DQRR entry can be consumed */
+	qman_cb_dqrr_consume,
+	/* Like _consume, but requests parking - FQ must be held-active */
+	qman_cb_dqrr_park,
+	/* Does not consume, for DCA mode only. This allows out-of-order
+	 * consumes by explicit calls to qman_dca() and/or the use of implicit
+	 * DCA via EQCR entries. */
+	qman_cb_dqrr_defer,
+	/* Stop processing without consuming this ring entry. Exits the current
+	 * qman_poll_dqrr() or interrupt-handling, as appropriate. If within an
+	 * interrupt handler, the callback would typically call
+	 * qman_irqsource_remove(QM_PIRQ_DQRI) before returning this value,
+	 * otherwise the interrupt will reassert immediately. */
+	qman_cb_dqrr_stop,
+	/* Like qman_cb_dqrr_stop, but consumes the current entry. */
+	qman_cb_dqrr_consume_stop
+};
+typedef enum qman_cb_dqrr_result (*qman_cb_dqrr)(struct qman_portal *qm,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dqrr);
+
+/* This callback type is used when handling ERNs, FQRNs and FQRLs via MR. They
+ * are always consumed after the callback returns. */
+typedef void (*qman_cb_mr)(struct qman_portal *qm, struct qman_fq *fq,
+				const struct qm_mr_entry *msg);
+
+/* s/w-visible states. Ie. tentatively scheduled + truly scheduled + active +
+ * held-active + held-suspended are just "sched". Things like "retired" will not
+ * be assumed until it is complete (ie. QMAN_FQ_STATE_CHANGING is set until
+ * then, to indicate it's completing and to gate attempts to retry the retire
+ * command). Note, park commands do not set QMAN_FQ_STATE_CHANGING because it's
+ * technically impossible in the case of enqueue DCAs (which refer to DQRR ring
+ * index rather than the FQ that ring entry corresponds to), so repeated park
+ * commands are allowed (if you're silly enough to try) but won't change FQ
+ * state, and the resulting park notifications move FQs from "sched" to
+ * "parked". */
+enum qman_fq_state {
+	qman_fq_state_oos,
+	qman_fq_state_parked,
+	qman_fq_state_sched,
+	qman_fq_state_retired
+};
+
+/* Frame queue objects (struct qman_fq) are stored within memory passed to
+ * qman_create_fq(), as this allows stashing of caller-provided demux callback
+ * pointers at no extra cost to stashing of (driver-internal) FQ state. If the
+ * caller wishes to add per-FQ state and have it benefit from dequeue-stashing,
+ * they should;
+ *
+ * (a) extend the qman_fq structure with their state; eg.
+ *
+ *     // myfq is allocated and driver_fq callbacks filled in;
+ *     struct my_fq {
+ *         struct qman_fq base;
+ *         int an_extra_field;
+ *         [ ... add other fields to be associated with each FQ ...]
+ *     } *myfq = some_my_fq_allocator();
+ *     struct qman_fq *fq = qman_create_fq(fqid, flags, &myfq->base);
+ *
+ *     // in a dequeue callback, access extra fields from 'fq' via a cast;
+ *     struct my_fq *myfq = (struct my_fq *)fq;
+ *     do_something_with(myfq->an_extra_field);
+ *     [...]
+ *
+ * (b) when and if configuring the FQ for context stashing, specify how ever
+ *     many cachelines are required to stash 'struct my_fq', to accelerate not
+ *     only the Qman driver but the callback as well.
+ */
+
+struct qman_fq_cb {
+	qman_cb_dqrr dqrr;      /* for dequeued frames */
+	qman_cb_mr ern;         /* for s/w ERNs */
+	qman_cb_mr dc_ern;      /* for diverted h/w ERNs */
+	qman_cb_mr fqs;         /* frame-queue state changes*/
+};
+
+struct qman_fq {
+	/* Caller of qman_create_fq() provides these demux callbacks */
+	struct qman_fq_cb cb;
+	/* These are internal to the driver, don't touch. In particular, they
+	 * may change, be removed, or extended (so you shouldn't rely on
+	 * sizeof(qman_fq) being a constant). */
+	spinlock_t fqlock;
+	u32 fqid;
+	volatile unsigned long flags;
+	enum qman_fq_state state;
+	int cgr_groupid;
+	struct rb_node node;
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+	u32 key;
+#endif
+};
+
+/* This callback type is used when handling congestion group entry/exit.
+ * 'congested' is non-zero on congestion-entry, and zero on congestion-exit. */
+typedef void (*qman_cb_cgr)(struct qman_portal *qm,
+			struct qman_cgr *cgr, int congested);
+
+struct qman_cgr {
+	/* Set these prior to qman_create_cgr() */
+	u32 cgrid; /* 0..255, but u32 to allow specials like -1, 256, etc.*/
+	qman_cb_cgr cb;
+	/* These are private to the driver */
+	enum qm_channel chan; /* portal channel this object is created on */
+	struct list_head node;
+};
+
+/* Flags to qman_create_fq() */
+#define QMAN_FQ_FLAG_NO_ENQUEUE      0x00000001 /* can't enqueue */
+#define QMAN_FQ_FLAG_NO_MODIFY       0x00000002 /* can only enqueue */
+#define QMAN_FQ_FLAG_TO_DCPORTAL     0x00000004 /* consumed by CAAM/PME/Fman */
+#define QMAN_FQ_FLAG_LOCKED          0x00000008 /* multi-core locking */
+#define QMAN_FQ_FLAG_AS_IS           0x00000010 /* query h/w state */
+#define QMAN_FQ_FLAG_DYNAMIC_FQID    0x00000020 /* (de)allocate fqid */
+
+/* Flags to qman_destroy_fq() */
+#define QMAN_FQ_DESTROY_PARKED       0x00000001 /* FQ can be parked or OOS */
+
+/* Flags from qman_fq_state() */
+#define QMAN_FQ_STATE_CHANGING       0x80000000 /* 'state' is changing */
+#define QMAN_FQ_STATE_NE             0x40000000 /* retired FQ isn't empty */
+#define QMAN_FQ_STATE_ORL            0x20000000 /* retired FQ has ORL */
+#define QMAN_FQ_STATE_BLOCKOOS       0xe0000000 /* if any are set, no OOS */
+#define QMAN_FQ_STATE_CGR_EN         0x10000000 /* CGR enabled */
+#define QMAN_FQ_STATE_VDQCR          0x08000000 /* being volatile dequeued */
+
+/* Flags to qman_init_fq() */
+#define QMAN_INITFQ_FLAG_SCHED       0x00000001 /* schedule rather than park */
+#define QMAN_INITFQ_FLAG_NULL        0x00000002 /* zero 'contextB', no demux */
+#define QMAN_INITFQ_FLAG_LOCAL       0x00000004 /* set dest portal */
+
+/* Flags to qman_volatile_dequeue() */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+#define QMAN_VOLATILE_FLAG_WAIT      0x00000001 /* wait if VDQCR is in use */
+#define QMAN_VOLATILE_FLAG_WAIT_INT  0x00000002 /* if wait, interruptible? */
+#define QMAN_VOLATILE_FLAG_FINISH    0x00000004 /* wait till VDQCR completes */
+#endif
+
+/* Flags to qman_enqueue(). NB, the strange numbering is to align with hardware,
+ * bit-wise. (NB: the PME API is sensitive to these precise numberings too, so
+ * any change here should be audited in PME.) */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT
+#define QMAN_ENQUEUE_FLAG_WAIT       0x00010000 /* wait if EQCR is full */
+#define QMAN_ENQUEUE_FLAG_WAIT_INT   0x00020000 /* if wait, interruptible? */
+#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC
+#define QMAN_ENQUEUE_FLAG_WAIT_SYNC  0x00000004 /* if wait, until consumed? */
+#endif
+#endif
+#define QMAN_ENQUEUE_FLAG_WATCH_CGR  0x00080000 /* watch congestion state */
+#define QMAN_ENQUEUE_FLAG_DCA        0x00008000 /* perform enqueue-DCA */
+#define QMAN_ENQUEUE_FLAG_DCA_PARK   0x00004000 /* If DCA, requests park */
+#define QMAN_ENQUEUE_FLAG_DCA_PTR(p)		/* If DCA, p is DQRR entry */ \
+		(((u32)(p) << 2) & 0x00000f00)
+#define QMAN_ENQUEUE_FLAG_C_GREEN    0x00000000 /* choose one C_*** flag */
+#define QMAN_ENQUEUE_FLAG_C_YELLOW   0x00000008
+#define QMAN_ENQUEUE_FLAG_C_RED      0x00000010
+#define QMAN_ENQUEUE_FLAG_C_OVERRIDE 0x00000018
+/* For the ORP-specific qman_enqueue_orp() variant;
+ * - this flag indicates "Not Last In Sequence", ie. all but the final fragment
+ *   of a frame. */
+#define QMAN_ENQUEUE_FLAG_NLIS       0x01000000
+/* - this flag performs no enqueue but fills in an ORP sequence number that
+ *   would otherwise block it (eg. if a frame has been dropped). */
+#define QMAN_ENQUEUE_FLAG_HOLE       0x02000000
+/* - this flag performs no enqueue but advances NESN to the given sequence
+ *   number. */
+#define QMAN_ENQUEUE_FLAG_NESN       0x04000000
+
+/* Flags to qman_modify_cgr() */
+#define QMAN_CGR_FLAG_USE_INIT       0x00000001
+#define QMAN_CGR_MODE_FRAME          0x00000001
+
+	/* Portal Management */
+	/* ----------------- */
+/**
+ * qman_get_portal_config - get portal configuration settings
+ *
+ * This returns a read-only view of the current cpu's affine portal settings.
+ */
+const struct qman_portal_config *qman_get_portal_config(void);
+
+#ifdef CONFIG_FSL_QMAN_NULL_FQ_DEMUX
+/**
+ * qman_get_null_cb - get callbacks currently used for "null" frame queues
+ *
+ * Copies the callbacks used for the affine portal of the current cpu.
+ */
+void qman_get_null_cb(struct qman_fq_cb *null_cb);
+
+/**
+ * qman_set_null_cb - set callbacks to use for "null" frame queues
+ *
+ * Sets the callbacks to use for the affine portal of the current cpu, whenever
+ * a DQRR or MR entry refers to a "null" FQ object. (Eg. zero-conf messaging.)
+ */
+void qman_set_null_cb(const struct qman_fq_cb *null_cb);
+#endif
+
+/**
+ * qman_irqsource_get - return the portal work that is interrupt-driven
+ *
+ * Returns a bitmask of QM_PIRQ_**I processing sources that are currently
+ * enabled for interrupt handling on the current cpu's affine portal. These
+ * sources will trigger the portal interrupt and the interrupt handler (or a
+ * tasklet/bottom-half it defers to) will perform the corresponding processing
+ * work. The qman_poll_***() functions will only process sources that are not in
+ * this bitmask. If the current CPU is sharing a portal hosted on another CPU,
+ * this always returns zero.
+ */
+u32 qman_irqsource_get(void);
+
+/**
+ * qman_irqsource_add - add processing sources to be interrupt-driven
+ * @bits: bitmask of QM_PIRQ_**I processing sources
+ *
+ * Adds processing sources that should be interrupt-driven (rather than
+ * processed via qman_poll_***() functions). Returns zero for success, or
+ * -EINVAL if the current CPU is sharing a portal hosted on another CPU.
+ */
+int qman_irqsource_add(u32 bits);
+
+/**
+ * qman_irqsource_remove - remove processing sources from being interrupt-driven
+ * @bits: bitmask of QM_PIRQ_**I processing sources
+ *
+ * Removes processing sources from being interrupt-driven, so that they will
+ * instead be processed via qman_poll_***() functions. Returns zero for success,
+ * or -EINVAL if the current CPU is sharing a portal hosted on another CPU.
+ */
+int qman_irqsource_remove(u32 bits);
+
+/**
+ * qman_affine_cpus - return a mask of cpus that have affine portals
+ */
+const cpumask_t *qman_affine_cpus(void);
+
+/**
+ * qman_poll_dqrr - process DQRR (fast-path) entries
+ * @limit: the maximum number of DQRR entries to process
+ *
+ * Use of this function requires that DQRR processing not be interrupt-driven.
+ * Ie. the value returned by qman_irqsource_get() should not include
+ * QM_PIRQ_DQRI. If the current CPU is sharing a portal hosted on another CPU,
+ * this function will return -EINVAL, otherwise the return value is >=0 and
+ * represents the number of DQRR entries processed.
+ */
+int qman_poll_dqrr(unsigned int limit);
+
+/**
+ * qman_poll_slow - process anything (except DQRR) that isn't interrupt-driven.
+ *
+ * This function does any portal processing that isn't interrupt-driven. If the
+ * current CPU is sharing a portal hosted on another CPU, this function will
+ * return (u32)-1, otherwise the return value is a bitmask of QM_PIRQ_* sources
+ * indicating what interrupt sources were actually processed by the call.
+ */
+u32 qman_poll_slow(void);
+
+/**
+ * qman_poll - legacey wrapper for qman_poll_dqrr() and qman_poll_slow()
+ *
+ * Dispatcher logic on a cpu can use this to trigger any maintenance of the
+ * affine portal. There are two classes of portal processing in question;
+ * fast-path (which involves demuxing dequeue ring (DQRR) entries and tracking
+ * enqueue ring (EQCR) consumption), and slow-path (which involves EQCR
+ * thresholds, congestion state changes, etc). This function does whatever
+ * processing is not triggered by interrupts.
+ *
+ * Note, if DQRR and some slow-path processing are poll-driven (rather than
+ * interrupt-driven) then this function uses a heuristic to determine how often
+ * to run slow-path processing - as slow-path processing introduces at least a
+ * minimum latency each time it is run, whereas fast-path (DQRR) processing is
+ * close to zero-cost if there is no work to be done. Applications can tune this
+ * behaviour themselves by using qman_poll_dqrr() and qman_poll_slow() directly
+ * rather than going via this wrapper.
+ */
+void qman_poll(void);
+
+/**
+ * qman_recovery_cleanup_fq - in recovery mode, cleanup a FQ of unknown state
+ */
+int qman_recovery_cleanup_fq(u32 fqid);
+
+/**
+ * qman_recovery_exit - leave recovery mode
+ */
+int qman_recovery_exit(void);
+
+/**
+ * qman_stop_dequeues - Stop h/w dequeuing to the s/w portal
+ *
+ * Disables DQRR processing of the portal. This is reference-counted, so
+ * qman_start_dequeues() must be called as many times as qman_stop_dequeues() to
+ * truly re-enable dequeuing.
+ */
+void qman_stop_dequeues(void);
+
+/**
+ * qman_start_dequeues - (Re)start h/w dequeuing to the s/w portal
+ *
+ * Enables DQRR processing of the portal. This is reference-counted, so
+ * qman_start_dequeues() must be called as many times as qman_stop_dequeues() to
+ * truly re-enable dequeuing.
+ */
+void qman_start_dequeues(void);
+
+/**
+ * qman_static_dequeue_add - Add pool channels to the portal SDQCR
+ * @pools: bit-mask of pool channels, using QM_SDQCR_CHANNELS_POOL(n)
+ *
+ * Adds a set of pool channels to the portal's static dequeue command register
+ * (SDQCR). The requested pools are limited to those the portal has dequeue
+ * access to.
+ */
+void qman_static_dequeue_add(u32 pools);
+
+/**
+ * qman_static_dequeue_del - Remove pool channels from the portal SDQCR
+ * @pools: bit-mask of pool channels, using QM_SDQCR_CHANNELS_POOL(n)
+ *
+ * Removes a set of pool channels from the portal's static dequeue command
+ * register (SDQCR). The requested pools are limited to those the portal has
+ * dequeue access to.
+ */
+void qman_static_dequeue_del(u32 pools);
+
+/**
+ * qman_static_dequeue_get - return the portal's current SDQCR
+ *
+ * Returns the portal's current static dequeue command register (SDQCR). The
+ * entire register is returned, so if only the currently-enabled pool channels
+ * are desired, mask the return value with QM_SDQCR_CHANNELS_POOL_MASK.
+ */
+u32 qman_static_dequeue_get(void);
+
+/**
+ * qman_dca - Perform a Discrete Consumption Acknowledgement
+ * @dq: the DQRR entry to be consumed
+ * @park_request: indicates whether the held-active @fq should be parked
+ *
+ * Only allowed in DCA-mode portals, for DQRR entries whose handler callback had
+ * previously returned 'qman_cb_dqrr_defer'. NB, as with the other APIs, this
+ * does not take a 'portal' argument but implies the core affine portal from the
+ * cpu that is currently executing the function. For reasons of locking, this
+ * function must be called from the same CPU as that which processed the DQRR
+ * entry in the first place.
+ */
+void qman_dca(struct qm_dqrr_entry *dq, int park_request);
+
+/**
+ * qman_eqcr_is_empty - Determine if portal's EQCR is empty
+ *
+ * For use in situations where a cpu-affine caller needs to determine when all
+ * enqueues for the local portal have been processed by Qman but can't use the
+ * QMAN_ENQUEUE_FLAG_WAIT_SYNC flag to do this from the final qman_enqueue().
+ * The function forces tracking of EQCR consumption (which normally doesn't
+ * happen until enqueue processing needs to find space to put new enqueue
+ * commands), and returns zero if the ring still has unprocessed entries,
+ * non-zero if it is empty.
+ */
+int qman_eqcr_is_empty(void);
+
+	/* FQ management */
+	/* ------------- */
+/**
+ * qman_create_fq - Allocates a FQ
+ * @fqid: the index of the FQD to encapsulate, must be "Out of Service"
+ * @flags: bit-mask of QMAN_FQ_FLAG_*** options
+ * @fq: memory for storing the 'fq', with callbacks filled in
+ *
+ * Creates a frame queue object for the given @fqid, unless the
+ * QMAN_FQ_FLAG_DYNAMIC_FQID flag is set in @flags, in which case a FQID is
+ * dynamically allocated (or the function fails if none are available). Once
+ * created, the caller should not touch the memory at 'fq' except as extended to
+ * adjacent memory for user-defined fields (see the definition of "struct
+ * qman_fq" for more info). NO_MODIFY is only intended for enqueuing to
+ * pre-existing frame-queues that aren't to be otherwise interfered with, it
+ * prevents all other modifications to the frame queue. The TO_DCPORTAL flag
+ * causes the driver to honour any contextB modifications requested in the
+ * qm_init_fq() API, as this indicates the frame queue will be consumed by a
+ * direct-connect portal (PME, CAAM, or Fman). When frame queues are consumed by
+ * software portals, the contextB field is controlled by the driver and can't be
+ * modified by the caller. If the AS_IS flag is specified, management commands
+ * will be used on portal @p to query state for frame queue @fqid and construct
+ * a frame queue object based on that, rather than assuming/requiring that it be
+ * Out of Service.
+ */
+int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq);
+
+/**
+ * qman_destroy_fq - Deallocates a FQ
+ * @fq: the frame queue object to release
+ * @flags: bit-mask of QMAN_FQ_FREE_*** options
+ *
+ * The memory for this frame queue object ('fq' provided in qman_create_fq()) is
+ * not deallocated but the caller regains ownership, to do with as desired. The
+ * FQ must be in the 'out-of-service' state unless the QMAN_FQ_FREE_PARKED flag
+ * is specified, in which case it may also be in the 'parked' state.
+ */
+void qman_destroy_fq(struct qman_fq *fq, u32 flags);
+
+/**
+ * qman_fq_fqid - Queries the frame queue ID of a FQ object
+ * @fq: the frame queue object to query
+ */
+u32 qman_fq_fqid(struct qman_fq *fq);
+
+/**
+ * qman_fq_state - Queries the state of a FQ object
+ * @fq: the frame queue object to query
+ * @state: pointer to state enum to return the FQ scheduling state
+ * @flags: pointer to state flags to receive QMAN_FQ_STATE_*** bitmask
+ *
+ * Queries the state of the FQ object, without performing any h/w commands.
+ * This captures the state, as seen by the driver, at the time the function
+ * executes.
+ */
+void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags);
+
+/**
+ * qman_init_fq - Initialises FQ fields, leaves the FQ "parked" or "scheduled"
+ * @fq: the frame queue object to modify, must be 'parked' or new.
+ * @flags: bit-mask of QMAN_INITFQ_FLAG_*** options
+ * @opts: the FQ-modification settings, as defined in the low-level API
+ *
+ * The @opts parameter comes from the low-level portal API. Select
+ * QMAN_INITFQ_FLAG_SCHED in @flags to cause the frame queue to be scheduled
+ * rather than parked. Select QMAN_INITFQ_FLAG_NULL in @flags to configure a
+ * frame queue that will not demux to a 'struct qman_fq' object when dequeued
+ * frames or messages arrive at a software portal, but which will instead
+ * trigger the portal's 'null_cb' callbacks (see qman_create_portal()). NB,
+ * @opts can be NULL.
+ *
+ * Note that some fields and options within @opts may be ignored or overwritten
+ * by the driver;
+ * 1. the 'count' and 'fqid' fields are always ignored (this operation only
+ * affects one frame queue: @fq).
+ * 2. the QM_INITFQ_WE_CONTEXTB option of the 'we_mask' field and the associated
+ * 'fqd' structure's 'context_b' field are sometimes overwritten;
+ *   - if @flags contains QMAN_INITFQ_FLAG_NULL, then context_b is initialised
+ *     to zero by the driver,
+ *   - if @fq was not created with QMAN_FQ_FLAG_TO_DCPORTAL, then context_b is
+ *     initialised to a value used by the driver for demux.
+ *   - if context_b is initialised for demux, so is context_a in case stashing
+ *     is requested (see item 4).
+ * (So caller control of context_b is only possible for TO_DCPORTAL frame queue
+ * objects.)
+ * 3. if @flags contains QMAN_INITFQ_FLAG_LOCAL, the 'fqd' structure's
+ * 'dest::channel' field will be overwritten to match the portal used to issue
+ * the command. If the WE_DESTWQ write-enable bit had already been set by the
+ * caller, the channel workqueue will be left as-is, otherwise the write-enable
+ * bit is set and the workqueue is set to a default of 4. If the "LOCAL" flag
+ * isn't set, the destination channel/workqueue fields and the write-enable bit
+ * are left as-is.
+ * 4. if the driver overwrites context_a/b for demux, then if
+ * QM_INITFQ_WE_CONTEXTA is set, the driver will only overwrite
+ * context_a.address fields and will leave the stashing fields provided by the
+ * user alone, otherwise it will zero out the context_a.stashing fields.
+ */
+int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts);
+
+/**
+ * qman_schedule_fq - Schedules a FQ
+ * @fq: the frame queue object to schedule, must be 'parked'
+ *
+ * Schedules the frame queue, which must be Parked, which takes it to
+ * Tentatively-Scheduled or Truly-Scheduled depending on its fill-level.
+ */
+int qman_schedule_fq(struct qman_fq *fq);
+
+/**
+ * qman_retire_fq - Retires a FQ
+ * @fq: the frame queue object to retire
+ * @flags: FQ flags (as per qman_fq_state) if retirement completes immediately
+ *
+ * Retires the frame queue. This returns zero if it succeeds immediately, +1 if
+ * the retirement was started asynchronously, otherwise it returns negative for
+ * failure. When this function returns zero, @flags is set to indicate whether
+ * the retired FQ is empty and/or whether it has any ORL fragments (to show up
+ * as ERNs). Otherwise the corresponding flags will be known when a subsequent
+ * FQRN message shows up on the portal's message ring.
+ *
+ * NB, if the retirement is asynchronous (the FQ was in the Truly Scheduled or
+ * Active state), the completion will be via the message ring as a FQRN - but
+ * the corresponding callback may occur before this function returns!! Ie. the
+ * caller should be prepared to accept the callback as the function is called,
+ * not only once it has returned.
+ */
+int qman_retire_fq(struct qman_fq *fq, u32 *flags);
+
+/**
+ * qman_oos_fq - Puts a FQ "out of service"
+ * @fq: the frame queue object to be put out-of-service, must be 'retired'
+ *
+ * The frame queue must be retired and empty, and if any order restoration list
+ * was released as ERNs at the time of retirement, they must all be consumed.
+ */
+int qman_oos_fq(struct qman_fq *fq);
+
+/**
+ * qman_query_fq - Queries FQD fields (via h/w query command)
+ * @fq: the frame queue object to be queried
+ * @fqd: storage for the queried FQD fields
+ */
+int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd);
+
+/**
+ * qman_query_fq_np - Queries non-programmable FQD fields
+ * @fq: the frame queue object to be queried
+ * @np: storage for the queried FQD fields
+ */
+int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np);
+
+/**
+ * qman_query_wq - Queries work queue lengths
+ * @query_dedicated: If non-zero, query length of WQs in the channel dedicated
+ *		to this software portal. Otherwise, query length of WQs in a
+ *		channel  specified in wq.
+ * @wq: storage for the queried WQs lengths. Also specified the channel to
+ *	to query if query_dedicated is zero.
+ */
+int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq);
+
+/**
+ * qman_volatile_dequeue - Issue a volatile dequeue command
+ * @fq: the frame queue object to dequeue from (or NULL)
+ * @flags: a bit-mask of QMAN_VOLATILE_FLAG_*** options
+ * @vdqcr: bit mask of QM_VDQCR_*** options, as per qm_dqrr_vdqcr_set()
+ *
+ * Attempts to lock access to the portal's VDQCR volatile dequeue functionality.
+ * The function will block and sleep if QMAN_VOLATILE_FLAG_WAIT is specified and
+ * the VDQCR is already in use, otherwise returns non-zero for failure. If
+ * QMAN_VOLATILE_FLAG_FINISH is specified, the function will only return once
+ * the VDQCR command has finished executing (ie. once the callback for the last
+ * DQRR entry resulting from the VDQCR command has been called). If not using
+ * the FINISH flag, completion can be determined either by detecting the
+ * presence of the QM_DQRR_STAT_UNSCHEDULED and QM_DQRR_STAT_DQCR_EXPIRED bits
+ * in the "stat" field of the "struct qm_dqrr_entry" passed to the FQ's dequeue
+ * callback, or by waiting for the QMAN_FQ_STATE_VDQCR bit to disappear from the
+ * "flags" retrieved from qman_fq_state(). If @fq is non-NULL, the corresponding
+ * FQID will be substituted in to the VDQCR command, otherwise it is assumed
+ * that @vdqcr already contains the FQID to dequeue from.
+ */
+int qman_volatile_dequeue(struct qman_fq *fq, u32 flags, u32 vdqcr);
+
+/**
+ * qman_enqueue - Enqueue a frame to a frame queue
+ * @fq: the frame queue object to enqueue to
+ * @fd: a descriptor of the frame to be enqueued
+ * @flags: bit-mask of QMAN_ENQUEUE_FLAG_*** options
+ *
+ * Fills an entry in the EQCR of portal @qm to enqueue the frame described by
+ * @fd. The descriptor details are copied from @fd to the EQCR entry, the 'pid'
+ * field is ignored. The return value is non-zero on error, such as ring full
+ * (and FLAG_WAIT not specified), congestion avoidance (FLAG_WATCH_CGR
+ * specified), etc. If the ring is full and FLAG_WAIT is specified, this
+ * function will block. If FLAG_INTERRUPT is set, the EQCI bit of the portal
+ * interrupt will assert when Qman consumes the EQCR entry (subject to "status
+ * disable", "enable", and "inhibit" registers). If FLAG_DCA is set, Qman will
+ * perform an implied "discrete consumption acknowledgement" on the dequeue
+ * ring's (DQRR) entry, at the ring index specified by the FLAG_DCA_IDX(x)
+ * macro. (As an alternative to issuing explicit DCA actions on DQRR entries,
+ * this implicit DCA can delay the release of a "held active" frame queue
+ * corresponding to a DQRR entry until Qman consumes the EQCR entry - providing
+ * order-preservation semantics in packet-forwarding scenarios.) If FLAG_DCA is
+ * set, then FLAG_DCA_PARK can also be set to imply that the DQRR consumption
+ * acknowledgement should "park request" the "held active" frame queue. Ie.
+ * when the portal eventually releases that frame queue, it will be left in the
+ * Parked state rather than Tentatively Scheduled or Truly Scheduled. If the
+ * portal is watching congestion groups, the QMAN_ENQUEUE_FLAG_WATCH_CGR flag
+ * is requested, and the FQ is a member of a congestion group, then this
+ * function returns -EAGAIN if the congestion group is currently congested.
+ * Note, this does not eliminate ERNs, as the async interface means we can be
+ * sending enqueue commands to an un-congested FQ that becomes congested before
+ * the enqueue commands are processed, but it does minimise needless thrashing
+ * of an already busy hardware resource by throttling many of the to-be-dropped
+ * enqueues "at the source".
+ */
+int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags);
+
+/**
+ * qman_enqueue_orp - Enqueue a frame to a frame queue using an ORP
+ * @fq: the frame queue object to enqueue to
+ * @fd: a descriptor of the frame to be enqueued
+ * @flags: bit-mask of QMAN_ENQUEUE_FLAG_*** options
+ * @orp: the frame queue object used as an order restoration point.
+ * @orp_seqnum: the sequence number of this frame in the order restoration path
+ *
+ * Similar to qman_enqueue(), but with the addition of an Order Restoration
+ * Point (@orp) and corresponding sequence number (@orp_seqnum) for this
+ * enqueue operation to employ order restoration. Each frame queue object acts
+ * as an Order Definition Point (ODP) by providing each frame dequeued from it
+ * with an incrementing sequence number, this value is generally ignored unless
+ * that sequence of dequeued frames will need order restoration later. Each
+ * frame queue object also encapsulates an Order Restoration Point (ORP), which
+ * is a re-assembly context for re-ordering frames relative to their sequence
+ * numbers as they are enqueued. The ORP does not have to be within the frame
+ * queue that receives the enqueued frame, in fact it is usually the frame
+ * queue from which the frames were originally dequeued. For the purposes of
+ * order restoration, multiple frames (or "fragments") can be enqueued for a
+ * single sequence number by setting the QMAN_ENQUEUE_FLAG_NLIS flag for all
+ * enqueues except the final fragment of a given sequence number. Ordering
+ * between sequence numbers is guaranteed, even if fragments of different
+ * sequence numbers are interlaced with one another. Fragments of the same
+ * sequence number will retain the order in which they are enqueued. If no
+ * enqueue is to performed, QMAN_ENQUEUE_FLAG_HOLE indicates that the given
+ * sequence number is to be "skipped" by the ORP logic (eg. if a frame has been
+ * dropped from a sequence), or QMAN_ENQUEUE_FLAG_NESN indicates that the given
+ * sequence number should become the ORP's "Next Expected Sequence Number".
+ *
+ * Side note: a frame queue object can be used purely as an ORP, without
+ * carrying any frames at all. Care should be taken not to deallocate a frame
+ * queue object that is being actively used as an ORP, as a future allocation
+ * of the frame queue object may start using the internal ORP before the
+ * previous use has finished.
+ */
+int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags,
+			struct qman_fq *orp, u16 orp_seqnum);
+
+/**
+ * qman_alloc_fqid_range - Allocate a contiguous range of FQIDs
+ * @result: is set by the API to the base FQID of the allocated range
+ * @count: the number of FQIDs required
+ * @align: required alignment of the allocated range
+ * @partial: non-zero if the API can return fewer than @count FQIDs
+
+ * Returns the number of frame queues allocated, or a negative error code. If
+ * @partial is non zero, the allocation request may return a smaller range of
+ * FQs than requested (though alignment will be as requested). If @partial is
+ * zero, the return value will either be 'count' or negative.
+ */
+int qman_alloc_fqid_range(u32 *result, u32 count, u32 align, int partial);
+static inline int qman_alloc_fqid(u32 *result)
+{
+	return qman_alloc_fqid_range(result, 1, 0, 0);
+}
+
+/**
+ * qman_release_fqid_range - Release the specified range of frame queue IDs
+ * @fqid: the base FQID of the range to deallocate
+ * @count: the number of FQIDs in the range
+ *
+ * This function can also be used to seed the allocator with ranges of FQIDs
+ * that it can subsequently use. Returns zero for success.
+ */
+void qman_release_fqid_range(u32 fqid, unsigned int count);
+static inline void qman_release_fqid(u32 fqid)
+{
+	qman_release_fqid_range(fqid, 1);
+}
+
+	/* CGR management */
+	/* -------------- */
+/**
+ * qman_create_cgr - Register a congestion group object
+ * @cgr: the 'cgr' object, with fields filled in
+ * @flags: QMAN_CGR_FLAG_* values
+ * @opts: optional state of CGR settings
+ *
+ * Registers this object to receiving congestion entry/exit callbacks on the
+ * portal affine to the cpu portal on which this API is executed. If opts is
+ * NULL then only the callback (cgr->cb) function is registered. If @flags
+ * contains QMAN_CGR_FLAG_USE_INIT, then an init hw command (which will reset
+ * any unspecified parameters) will be used rather than a modify hw hardware
+ * (which only modifies the specified parameters).
+ */
+int qman_create_cgr(struct qman_cgr *cgr, u32 flags,
+			struct qm_mcc_initcgr *opts);
+
+/**
+ * qman_delete_cgr - Deregisters a congestion group object
+ * @cgr: the 'cgr' object to deregister
+ *
+ * "Unplugs" this CGR object from the portal affine to the cpu on which this API
+ * is executed. This must be excuted on the same affine portal on which it was
+ * created.
+ */
+int qman_delete_cgr(struct qman_cgr *cgr);
+
+/**
+ * qman_modify_cgr - Modify CGR fields
+ * @cgr: the 'cgr' object to modify
+ * @flags: QMAN_CGR_FLAG_* values
+ * @opts: the CGR-modification settings
+ *
+ * The @opts parameter comes from the low-level portal API, and can be NULL.
+ * Note that some fields and options within @opts may be ignored or overwritten
+ * by the driver, in particular the 'cgrid' field is ignored (this operation
+ * only affects the given CGR object). If @flags contains
+ * QMAN_CGR_FLAG_USE_INIT, then an init hw command (which will reset any
+ * unspecified parameters) will be used rather than a modify hw hardware (which
+ * only modifies the specified parameters).
+ */
+int qman_modify_cgr(struct qman_cgr *cgr, u32 flags,
+			struct qm_mcc_initcgr *opts);
+
+/**
+* qman_query_cgr - Queries CGR fields
+* @cgr: the 'cgr' object to query
+* @result: storage for the queried congestion group record
+*/
+int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *result);
+
+/**
+ * qman_query_congestion - Queries the state of all congestion groups
+ * @congestion: storage for the queried state of all congestion groups
+ */
+int qman_query_congestion(struct qm_mcr_querycongestion *congestion);
+
+	/* Helpers */
+	/* ------- */
+/**
+ * qman_poll_fq_for_init - Check if an FQ has been initialised from OOS
+ * @fqid: the FQID that will be initialised by other s/w
+ *
+ * In many situations, a FQID is provided for communication between s/w
+ * entities, and whilst the consumer is responsible for initialising and
+ * scheduling the FQ, the producer(s) generally create a wrapper FQ object using
+ * and only call qman_enqueue() (no FQ initialisation, scheduling, etc). Ie;
+ *     qman_create_fq(..., QMAN_FQ_FLAG_NO_MODIFY, ...);
+ * However, data can not be enqueued to the FQ until it is initialised out of
+ * the OOS state - this function polls for that condition. It is particularly
+ * useful for users of IPC functions - each endpoint's Rx FQ is the other
+ * endpoint's Tx FQ, so each side can initialise and schedule their Rx FQ object
+ * and then use this API on the (NO_MODIFY) Tx FQ object in order to
+ * synchronise. The function returns zero for success, +1 if the FQ is still in
+ * the OOS state, or negative if there was an error.
+ */
+static inline int qman_poll_fq_for_init(struct qman_fq *fq)
+{
+	struct qm_mcr_queryfq_np np;
+	int err;
+	err = qman_query_fq_np(fq, &np);
+	if (err)
+		return err;
+	if ((np.state & QM_MCR_NP_STATE_MASK) == QM_MCR_NP_STATE_OOS)
+		return 1;
+	return 0;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* FSL_QMAN_H */
+