Patchwork DMA/RaidEngine: Enable FSL RaidEngine

login
register
mail settings
Submitter b29237@freescale.com
Date Aug. 22, 2012, 6:24 a.m.
Message ID <1345616661-8550-1-git-send-email-b29237@freescale.com>
Download mbox | patch
Permalink /patch/179247/
State Superseded
Headers show

Comments

b29237@freescale.com - Aug. 22, 2012, 6:24 a.m.
From: Xuelin Shi <b29237@freescale.com>

The RaidEngine is a new FSL hardware that used as hardware acceration
for RAID5/6.

This patch enables the RaidEngine functionality and provides hardware
offloading capability for memcpy, xor and raid6 pq computation. It works
under dmaengine control with async_layer interface.

Signed-off-by: Harninder Rai <harninder.rai@freescale.com>
Signed-off-by: Naveen Burmi <naveenburmi@freescale.com>
Signed-off-by: Xuelin Shi <b29237@freescale.com>
---
 arch/powerpc/boot/dts/fsl/p5020si-post.dtsi    |    1 +
 arch/powerpc/boot/dts/fsl/p5020si-pre.dtsi     |    6 +
 arch/powerpc/boot/dts/fsl/qoriq-raid1.0-0.dtsi |   85 ++
 drivers/dma/Kconfig                            |   14 +
 drivers/dma/Makefile                           |    1 +
 drivers/dma/fsl_raid.c                         | 1090 ++++++++++++++++++++++++
 drivers/dma/fsl_raid.h                         |  294 +++++++
 7 files changed, 1491 insertions(+), 0 deletions(-)
 create mode 100644 arch/powerpc/boot/dts/fsl/qoriq-raid1.0-0.dtsi
 create mode 100644 drivers/dma/fsl_raid.c
 create mode 100644 drivers/dma/fsl_raid.h
Kumar Gala - Aug. 22, 2012, 1:37 p.m.
On Aug 22, 2012, at 1:24 AM, <b29237@freescale.com> <b29237@freescale.com> wrote:

> From: Xuelin Shi <b29237@freescale.com>
> 
> The RaidEngine is a new FSL hardware that used as hardware acceration
> for RAID5/6.
> 
> This patch enables the RaidEngine functionality and provides hardware
> offloading capability for memcpy, xor and raid6 pq computation. It works
> under dmaengine control with async_layer interface.
> 
> Signed-off-by: Harninder Rai <harninder.rai@freescale.com>
> Signed-off-by: Naveen Burmi <naveenburmi@freescale.com>
> Signed-off-by: Xuelin Shi <b29237@freescale.com>
> ---
> arch/powerpc/boot/dts/fsl/p5020si-post.dtsi    |    1 +
> arch/powerpc/boot/dts/fsl/p5020si-pre.dtsi     |    6 +
> arch/powerpc/boot/dts/fsl/qoriq-raid1.0-0.dtsi |   85 ++
> drivers/dma/Kconfig                            |   14 +
> drivers/dma/Makefile                           |    1 +
> drivers/dma/fsl_raid.c                         | 1090 ++++++++++++++++++++++++
> drivers/dma/fsl_raid.h                         |  294 +++++++
> 7 files changed, 1491 insertions(+), 0 deletions(-)
> create mode 100644 arch/powerpc/boot/dts/fsl/qoriq-raid1.0-0.dtsi
> create mode 100644 drivers/dma/fsl_raid.c
> create mode 100644 drivers/dma/fsl_raid.h

Where is the dts binding spec related to qoriq-raid1.0-0.dtsi ?

We should split the device tree parts into a separate patch as they can be reviewed & accepted separate from the driver code.

- k

Patch

diff --git a/arch/powerpc/boot/dts/fsl/p5020si-post.dtsi b/arch/powerpc/boot/dts/fsl/p5020si-post.dtsi
index 64b6abe..5d7205b 100644
--- a/arch/powerpc/boot/dts/fsl/p5020si-post.dtsi
+++ b/arch/powerpc/boot/dts/fsl/p5020si-post.dtsi
@@ -354,4 +354,5 @@ 
 /include/ "qoriq-sata2-0.dtsi"
 /include/ "qoriq-sata2-1.dtsi"
 /include/ "qoriq-sec4.2-0.dtsi"
+/include/ "qoriq-raid1.0-0.dtsi"
 };
diff --git a/arch/powerpc/boot/dts/fsl/p5020si-pre.dtsi b/arch/powerpc/boot/dts/fsl/p5020si-pre.dtsi
index ae823a4..d54cd90 100644
--- a/arch/powerpc/boot/dts/fsl/p5020si-pre.dtsi
+++ b/arch/powerpc/boot/dts/fsl/p5020si-pre.dtsi
@@ -70,6 +70,12 @@ 
 		rtic_c = &rtic_c;
 		rtic_d = &rtic_d;
 		sec_mon = &sec_mon;
+
+		raideng = &raideng;
+		raideng_jr0 = &raideng_jr0;
+		raideng_jr1 = &raideng_jr1;
+		raideng_jr2 = &raideng_jr2;
+		raideng_jr3 = &raideng_jr3;
 	};
 
 	cpus {
diff --git a/arch/powerpc/boot/dts/fsl/qoriq-raid1.0-0.dtsi b/arch/powerpc/boot/dts/fsl/qoriq-raid1.0-0.dtsi
new file mode 100644
index 0000000..8d2e8aa
--- /dev/null
+++ b/arch/powerpc/boot/dts/fsl/qoriq-raid1.0-0.dtsi
@@ -0,0 +1,85 @@ 
+/*
+ * QorIQ RAID 1.0 device tree stub [ controller @ offset 0x320000 ]
+ *
+ * Copyright 2012 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.
+ */
+
+raideng: raideng@320000 {
+	compatible = "fsl,raideng-v1.0";
+	#address-cells = <1>;
+	#size-cells = <1>;
+	reg = <0x320000 0x10000>;
+	ranges = <0 0x320000 0x10000>;
+
+	raideng_jq0@1000 {
+		compatible = "fsl,raideng-v1.0-job-queue";
+		#address-cells = <1>;
+		#size-cells = <1>;
+		reg = <0x1000 0x1000>;
+		ranges = <0x0 0x1000 0x1000>;
+
+		raideng_jr0: jr@0 {
+			compatible = "fsl,raideng-v1.0-job-ring", "fsl,raideng-v1.0-hp-ring";
+			reg = <0x0 0x400>;
+			interrupts = <139 2 0 0>;
+			interrupt-parent = <&mpic>;
+		};
+
+		raideng_jr1: jr@400 {
+			compatible = "fsl,raideng-v1.0-job-ring", "fsl,raideng-v1.0-lp-ring";
+			reg = <0x400 0x400>;
+			interrupts = <140 2 0 0>;
+			interrupt-parent = <&mpic>;
+		};
+	};
+
+	raideng_jq1@2000 {
+		compatible = "fsl,raideng-v1.0-job-queue";
+		#address-cells = <1>;
+		#size-cells = <1>;
+		reg = <0x2000 0x1000>;
+		ranges = <0x0 0x2000 0x1000>;
+
+		raideng_jr2: jr@0 {
+			compatible = "fsl,raideng-v1.0-job-ring", "fsl,raideng-v1.0-hp-ring";
+			reg = <0x0 0x400>;
+			interrupts = <141 2 0 0>;
+			interrupt-parent = <&mpic>;
+		};
+
+		raideng_jr3: jr@400 {
+			compatible = "fsl,raideng-v1.0-job-ring", "fsl,raideng-v1.0-lp-ring";
+			reg = <0x400 0x400>;
+			interrupts = <142 2 0 0>;
+			interrupt-parent = <&mpic>;
+		};
+	};
+};
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index d06ea29..1b94c8e 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -292,6 +292,20 @@  config DMA_OMAP
 	select DMA_ENGINE
 	select DMA_VIRTUAL_CHANNELS
 
+config FSL_RAID
+        tristate "Freescale RAID Engine Device Driver"
+        depends on FSL_SOC && !FSL_DMA
+        select DMA_ENGINE
+        select ASYNC_TX_ENABLE_CHANNEL_SWITCH
+        select ASYNC_MEMCPY
+        select ASYNC_XOR
+        select ASYNC_PQ
+        ---help---
+          Enable support for Freescale RAID Engine. RAID Engine is
+          available on some QorIQ SoCs (like P5020). It has
+          the capability to offload RAID5/RAID6 operations from CPU.
+          RAID5 is XOR and memcpy. RAID6 is P/Q and memcpy
+
 config DMA_ENGINE
 	bool
 
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index 4cf6b12..b3e7702 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -32,3 +32,4 @@  obj-$(CONFIG_EP93XX_DMA) += ep93xx_dma.o
 obj-$(CONFIG_DMA_SA11X0) += sa11x0-dma.o
 obj-$(CONFIG_MMP_TDMA) += mmp_tdma.o
 obj-$(CONFIG_DMA_OMAP) += omap-dma.o
+obj-$(CONFIG_FSL_RAID) += fsl_raid.o
diff --git a/drivers/dma/fsl_raid.c b/drivers/dma/fsl_raid.c
new file mode 100644
index 0000000..ec860fd
--- /dev/null
+++ b/drivers/dma/fsl_raid.c
@@ -0,0 +1,1090 @@ 
+/*
+ * drivers/dma/fsl_raid.c
+ *
+ * Freescale RAID Engine device driver
+ *
+ * Author:
+ *	Harninder Rai <harninder.rai@freescale.com>
+ *	Naveen Burmi <naveenburmi@freescale.com>
+ *	Xuelin Shi <b29237@freescale.com>
+ *
+ * Copyright (c) 2010-2012 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.
+ *
+ * Theory of operation:
+ *
+ * General capabilities:
+ *	RAID Engine (RE) block is capable of offloading XOR, memcpy and P/Q
+ *	calculations required in RAID5 and RAID6 operations. RE driver
+ *	registers with Linux's ASYNC layer as dma driver. RE hardware
+ *	maintains strict ordering of the requests through chained
+ *	command queueing.
+ *
+ * Data flow:
+ *	Software RAID layer of Linux (MD layer) maintains RAID partitions,
+ *	strips, stripes etc. It sends requests to the underlying AYSNC layer
+ *	which further passes it to RE driver. ASYNC layer decides which request
+ *	goes to which job ring of RE hardware. For every request processed by
+ *	RAID Engine, driver gets an interrupt unless coalescing is set. The
+ *	per job ring interrupt handler checks the status register for errors,
+ *	clears the interrupt and schedules a tasklet. Main request processing
+ *	is done in tasklet. A software shadow copy of the HW ring is kept to
+ *	maintain virtual to physical translation. Based on the internal indexes
+ *	maintained, the tasklet picks the descriptor address from shadow copy,
+ *	updates the corresponding cookie, updates the outbound ring job removed
+ *	register in RE hardware and eventually calls the callback function. This
+ *	callback function gets passed as part of request from MD layer.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/of_platform.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmapool.h>
+#include <linux/dmaengine.h>
+#include <linux/io.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+
+#include "fsl_raid.h"
+
+#define MAX_RE_JRS		4
+#define MAX_XOR_SRCS		16
+#define MAX_PQ_SRCS		16
+#define MAX_INITIAL_DESCS	256
+#define FRAME_FORMAT		0x1
+#define MAX_DATA_LENGTH		(1024*1024)
+
+#define to_fsl_re_dma_desc(tx) container_of(tx, \
+		struct fsl_re_dma_async_tx_desc, async_tx)
+
+struct re_drv_private {
+	struct device *dev;
+	u8 total_jrs;
+	struct dma_device dma_dev;
+	struct re_ctrl *re_regs;
+	struct re_jr *re_jrs[MAX_RE_JRS];
+};
+
+static struct kmem_cache *re_sw_desc_cache;
+
+/* Add descriptors into per jr software queue - submit_q */
+static dma_cookie_t re_jr_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+	struct fsl_re_dma_async_tx_desc *desc = NULL;
+	struct re_jr *jr = NULL;
+	dma_cookie_t cookie;
+
+	desc = container_of(tx, struct fsl_re_dma_async_tx_desc, async_tx);
+	jr = container_of(tx->chan, struct re_jr, chan);
+
+	spin_lock_bh(&jr->desc_lock);
+
+	jr->timer.data = (unsigned long)tx->chan;
+	cookie = jr->chan.cookie + 1;
+	if (cookie < 0)
+		cookie = 1;
+
+	desc->async_tx.cookie = cookie;
+	jr->chan.cookie = desc->async_tx.cookie;
+	list_add_tail(&desc->node, &jr->submit_q);
+
+	if (!timer_pending(&jr->timer))
+		add_timer(&jr->timer);
+
+	spin_unlock_bh(&jr->desc_lock);
+
+	return cookie;
+}
+
+static void re_jr_unmap_dest_src(struct fsl_re_dma_async_tx_desc *desc)
+{
+	int i, j;
+	struct cmpnd_frame *cf;
+	dma_addr_t dest1 = 0, dest2 = 0, src;
+	struct device *dev;
+	enum dma_ctrl_flags flags;
+	enum dma_data_direction dir;
+
+	BUG_ON(!desc);
+	cf = desc->cf_addr;
+	dest1 = cf[1].address;
+	j = 2;
+	if (desc->dest_cnt == 2) {
+		dest2 = cf[2].address;
+		j = 3;
+	}
+	dev = desc->jr->chan.device->dev;
+	flags = desc->async_tx.flags;
+	if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
+		if (desc->cdb_opcode == RE_MOVE_OPCODE)
+			dir = DMA_FROM_DEVICE;
+		else
+			dir = DMA_BIDIRECTIONAL;
+
+		dma_unmap_page(dev, dest1, desc->dma_len, dir);
+
+		if (dest2)
+			dma_unmap_page(dev, dest2, desc->dma_len, dir);
+	}
+
+	if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
+		dir = DMA_TO_DEVICE;
+		for (i = j; i < desc->src_cnt+j; i++) {
+			src = cf[i].address;
+			if (src == dest1 || src == dest2)
+				continue;
+			dma_unmap_page(dev, src, desc->dma_len, dir);
+		}
+	}
+}
+
+static void re_jr_desc_done(struct fsl_re_dma_async_tx_desc *desc)
+{
+	struct re_jr *dma_jr = desc->jr;
+	dma_async_tx_callback callback;
+	void *callback_param;
+
+	dma_run_dependencies(&desc->async_tx);
+
+	spin_lock_bh(&dma_jr->desc_lock);
+	if (dma_jr->completed_cookie < desc->async_tx.cookie) {
+		dma_jr->completed_cookie = desc->async_tx.cookie;
+		if (dma_jr->completed_cookie == DMA_MAX_COOKIE)
+			dma_jr->completed_cookie = DMA_MIN_COOKIE;
+	}
+	spin_unlock_bh(&dma_jr->desc_lock);
+
+	callback = desc->async_tx.callback;
+	callback_param = desc->async_tx.callback_param;
+
+	re_jr_unmap_dest_src(desc);
+
+	/* Free the cf/cdb address stored in descs
+	 * unconditionally. These pointers are only
+	 * required for RE driver's housekeeping
+	 */
+	dma_pool_free(dma_jr->desc_pool, desc->cf_addr, desc->cf_paddr);
+	dma_pool_free(dma_jr->desc_pool, desc->cdb_addr, desc->cdb_paddr);
+
+	if (callback)
+		callback(callback_param);
+}
+
+/*
+ * Get the virtual address of software desc from virt_addr.
+ * Storing the address of software desc like this makes the
+ * order of alogorithm as O(1)
+ */
+static void re_jr_dequeue(unsigned long data)
+{
+	struct device *dev = (struct device *)data;
+	struct re_jr *jr = dev_get_drvdata(dev);
+	struct fsl_re_dma_async_tx_desc *desc;
+	unsigned int count = 0;
+	struct fsl_re_dma_async_tx_desc *ack_desc = NULL, *_ack_desc = NULL;
+
+	while ((count =
+		RE_JR_OUB_SLOT_FULL(in_be32(&jr->jrregs->oubring_slot_full)))) {
+		while (count--) {
+			spin_lock_bh(&jr->desc_lock);
+
+			/* Wrap around index */
+			jr->oub_count &= RING_SIZE-1;
+
+			desc = (struct fsl_re_dma_async_tx_desc *)
+				jr->virt_arry[jr->oub_count].virt_addr;
+
+			/* Re-initialise so that it can be reused */
+			jr->virt_arry[jr->oub_count].virt_addr = 0;
+			jr->virt_arry[jr->oub_count++].phys_addr = 0;
+
+			/* One job processed */
+			out_be32(&jr->jrregs->oubring_job_rmvd,
+						RE_JR_OUB_JOB_REMOVE);
+
+			list_add_tail(&desc->node, &jr->ack_q);
+
+			spin_unlock_bh(&jr->desc_lock);
+
+			re_jr_desc_done(desc);
+		}
+	}
+
+	/* To save memory, parse the ack_q and free up descs */
+	list_for_each_entry_safe(ack_desc, _ack_desc, &jr->ack_q, node) {
+		if (async_tx_test_ack(&ack_desc->async_tx)) {
+			list_del(&ack_desc->node);
+			kmem_cache_free(re_sw_desc_cache, ack_desc);
+		}
+	}
+}
+
+/* Per Job Ring interrupt handler */
+static irqreturn_t re_jr_interrupt(int irq, void *data)
+{
+	struct device *dev = data;
+	struct re_jr *jr = dev_get_drvdata(dev);
+	u32 irqstate;
+
+	irqstate = in_be32(&jr->jrregs->jr_interrupt_status);
+	if (!irqstate)
+		return IRQ_NONE;
+
+	/*
+	 * There's no way in upper layer (read MD layer) to recover from
+	 * error conditions except restart everything. In long term we
+	 * need to do something more than just crashing
+	 */
+	if (irqstate & RE_JR_ERROR) {
+		dev_err(dev, "%s: jr error irqstate: %x\n", __func__,
+			irqstate);
+		BUG();
+	}
+
+	/* Clear interrupt */
+	out_be32(&jr->jrregs->jr_interrupt_status, RE_JR_CLEAR_INT);
+
+	tasklet_schedule(&jr->irqtask);
+
+	return IRQ_HANDLED;
+}
+
+static int re_jr_alloc_chan_resources(struct dma_chan *chan)
+{
+	struct re_jr *jr = container_of(chan, struct re_jr, chan);
+
+	jr->desc_pool = dma_pool_create("re_jr_desc_pool", jr->dev,
+					RE_CF_CDB_SIZE, RE_CF_CDB_ALIGN, 0);
+
+	if (!jr->desc_pool) {
+		pr_err("%s:No memory for re_jr_desc_pool\n", __func__);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+/* This function is just to please the ASYNC layer */
+static void re_jr_free_chan_resources(struct dma_chan *chan)
+{
+	struct re_jr *jr = container_of(chan, struct re_jr, chan);
+	dma_pool_destroy(jr->desc_pool);
+}
+
+static enum dma_status re_jr_tx_status(struct dma_chan *chan,
+		dma_cookie_t cookie, struct dma_tx_state *txstate)
+{
+	struct re_jr *jr = NULL;
+	dma_cookie_t last_used;
+	dma_cookie_t last_complete;
+
+	jr = container_of(chan, struct re_jr, chan);
+	last_used = chan->cookie;
+	smp_mb();
+	last_complete = jr->completed_cookie;
+
+	dma_set_tx_state(txstate, last_complete, last_used, 0);
+
+	return dma_async_is_complete(cookie, last_complete, last_used);
+}
+
+
+/* Copy descriptor from per jr software queue into hardware job ring */
+void re_jr_issue_pending(struct dma_chan *chan)
+{
+	struct re_jr *jr = NULL;
+	struct fsl_re_dma_async_tx_desc *desc, *_desc;
+
+	jr = container_of(chan, struct re_jr, chan);
+	if (timer_pending(&jr->timer))
+		del_timer_sync(&jr->timer);
+
+	spin_lock_bh(&jr->desc_lock);
+
+	list_for_each_entry_safe(desc, _desc, &jr->submit_q, node) {
+
+		if (!in_be32(&jr->jrregs->inbring_slot_avail))
+			goto out_unlock;
+
+		/* Wrap around ring index */
+		jr->inb_count &= RING_SIZE-1;
+
+		if (jr->virt_arry[jr->inb_count].phys_addr != 0)
+			goto out_unlock;
+
+		jr->virt_arry[jr->inb_count].phys_addr =
+			(phys_addr_t)desc->hwdesc.address;
+		jr->virt_arry[jr->inb_count++].virt_addr = (unsigned long)desc;
+
+		/* Wrap around ring index */
+		jr->inb_ring_index &= RING_SIZE-1;
+
+		/* Copy frame descriptor into job ring */
+		memcpy(&jr->inb_ring_virt_addr[jr->inb_ring_index],
+			&desc->hwdesc, sizeof(struct jr_hw_desc));
+
+		smp_wmb();
+
+		jr->inb_ring_index++;
+		list_del_init(&desc->node);
+
+		wmb();
+
+		/* One job has been added into job ring */
+		out_be32(&jr->jrregs->inbring_add_job, RE_JR_INB_JOB_ADD);
+	}
+out_unlock:
+	spin_unlock_bh(&jr->desc_lock);
+}
+
+/* Per Job Ring timer handler */
+static void raide_timer_handler(unsigned long data)
+{
+	struct dma_chan *chan = NULL;
+	chan = (struct dma_chan *)data;
+
+	re_jr_issue_pending(chan);
+}
+
+static void fill_cfd_frame(struct cmpnd_frame *cf, u8 index,
+		size_t length, dma_addr_t addr, bool final)
+{
+	cf[index].extension = 0;
+	cf[index].final = final;
+	cf[index].rsvd1 = 0;
+	cf[index].rsvd3 = 0;
+	cf[index].rsvd4 = 0;
+	cf[index].rsvd5 = 0;
+	cf[index].length = length;
+	cf[index].offset = 0;
+	cf[index].address = addr;
+
+}
+
+static struct fsl_re_dma_async_tx_desc *re_jr_alloc_desc(struct re_jr *jr,
+					int cf_num, unsigned long flags)
+{
+	struct fsl_re_dma_async_tx_desc *desc = NULL;
+	struct jr_hw_desc *hw_desc = NULL;
+	struct cmpnd_frame *cf;
+	dma_addr_t paddr;
+
+	desc = kmem_cache_alloc(re_sw_desc_cache, GFP_KERNEL);
+	if (!desc)
+		return NULL;
+
+	memset(desc, 0, sizeof(*desc));
+	desc->async_tx.tx_submit = re_jr_tx_submit;
+	dma_async_tx_descriptor_init(&desc->async_tx, &jr->chan);
+	INIT_LIST_HEAD(&desc->node);
+	hw_desc = &desc->hwdesc;
+	hw_desc->format = FRAME_FORMAT;
+
+	cf = dma_pool_alloc(jr->desc_pool, GFP_ATOMIC, &paddr);
+	if (!cf) {
+		kmem_cache_free(re_sw_desc_cache, desc);
+		return NULL;
+	}
+
+	hw_desc->address = paddr;
+	desc->cf_addr = cf;
+	desc->cf_paddr = paddr;
+	desc->cf_len = sizeof(struct cmpnd_frame)*cf_num;
+	memset(cf, 0, desc->cf_len);
+
+	desc->jr = jr;
+	desc->async_tx.cookie = -EBUSY;
+	desc->async_tx.flags = flags;
+
+	return desc;
+}
+
+static void re_jr_free_desc(struct fsl_re_dma_async_tx_desc *desc)
+{
+	struct re_jr *jr = NULL;
+
+	if (desc == NULL)
+		return;
+
+	jr = desc->jr;
+
+	dma_pool_free(jr->desc_pool, desc->cf_addr, desc->cf_paddr);
+	kmem_cache_free(re_sw_desc_cache, desc);
+
+}
+
+static struct dma_async_tx_descriptor *re_jr_prep_genq(
+		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
+		unsigned int src_cnt, const unsigned char *scf, size_t len,
+		unsigned long flags)
+{
+	struct re_jr *jr = NULL;
+	struct fsl_re_dma_async_tx_desc *desc = NULL;
+	struct xor_cdb *xor = NULL;
+	unsigned int cfs_reqd = src_cnt + 2; /* CDB+dest+src_cnt */
+	struct cmpnd_frame *cf;
+	int i = 0, j = 0;
+	dma_addr_t paddr;
+
+	if (len > MAX_DATA_LENGTH) {
+		pr_err("%s: Length greater than %d not supported\n",
+				__func__, MAX_DATA_LENGTH);
+		return NULL;
+	}
+	jr = container_of(chan, struct re_jr, chan);
+	desc = re_jr_alloc_desc(jr, cfs_reqd, flags);
+	if (!desc)
+		return ERR_PTR(-ENOMEM);
+
+	desc->dma_len = len;
+	desc->dest_cnt = 1;
+	desc->src_cnt = src_cnt;
+
+	/* Allocate memory for GenQ CDB */
+	xor = dma_pool_alloc(jr->desc_pool, GFP_ATOMIC, &paddr);
+	if (!xor) {
+		re_jr_free_desc(desc);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	desc->cdb_addr = xor;
+	desc->cdb_paddr = paddr;
+	desc->cdb_len = sizeof(struct xor_cdb);
+	memset(xor, 0, desc->cdb_len);
+	desc->cdb_opcode = RE_XOR_OPCODE;
+
+	/* Filling xor CDB */
+	xor->opcode = RE_XOR_OPCODE;
+	xor->nrcs = (src_cnt - 1);
+	xor->blk_size = RE_BLOCK_SIZE;
+	xor->cache_attrib = CACHEABLE_INPUT_OUTPUT;
+	xor->buffer_attrib = BUFFERABLE_OUTPUT;
+	xor->error_attrib = INTERRUPT_ON_ERROR;
+	xor->data_depend = DATA_DEPENDENCY;
+	xor->dpi = ENABLE_DPI;
+
+	if (scf != NULL) {
+		/* compute q = src0*coef0^src1*coef1^..., * is GF(8) mult */
+		for (i = 0; i < src_cnt; i++)
+			xor->gfm[i] = scf[i];
+	} else {
+		/* compute P, that is XOR all srcs */
+		for (i = 0; i < src_cnt; i++)
+			xor->gfm[i] = 1;
+	}
+
+	/* Filling frame 0 of compound frame descriptor with CDB */
+	cf = desc->cf_addr;
+	fill_cfd_frame(cf, 0, desc->cdb_len, paddr, 0);
+
+	/* Fill CFD's 1st frame with dest buffer */
+	fill_cfd_frame(cf, 1, len, dest, 0);
+
+	/* Fill CFD's rest of the frames with source buffers */
+	for (i = 2, j = 0; j < src_cnt; i++, j++)
+		fill_cfd_frame(cf, i, len, src[j], 0);
+
+	/* Setting the final bit in the last source buffer frame in CFD */
+	cf[i - 1].final = 1;
+
+	return &desc->async_tx;
+}
+
+/*
+ * Prep function for P parity calculation.In RAID Engine terminology,
+ * XOR calculation is called GenQ calculation done through GenQ command
+ */
+static struct dma_async_tx_descriptor *re_jr_prep_dma_xor(
+		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
+		unsigned int src_cnt, size_t len, unsigned long flags)
+{
+	/* NULL let genq take all coef as 1 */
+	return re_jr_prep_genq(chan, dest, src, src_cnt, NULL, len, flags);
+}
+
+/*
+ * Prep function for P/Q parity calculation.In RAID Engine terminology,
+ * P/Q calculation is called GenQQ done through GenQQ command
+ */
+static struct dma_async_tx_descriptor *re_jr_prep_pq(
+		struct dma_chan *chan, dma_addr_t *dest, dma_addr_t *src,
+		unsigned int src_cnt, const unsigned char *scf, size_t len,
+		unsigned long flags)
+{
+	struct re_jr *jr = NULL;
+	struct fsl_re_dma_async_tx_desc *desc = NULL;
+	struct pq_cdb *pq = NULL;
+	u8 cfs_reqd = src_cnt + 3; /* CDB+P+Q+src_cnt */
+	struct cmpnd_frame *cf;
+	u8 *p;
+	int gfmq_len, i, j;
+	dma_addr_t paddr;
+
+	if (len > MAX_DATA_LENGTH) {
+		pr_err("%s: Length greater than %d not supported\n",
+				__func__, MAX_DATA_LENGTH);
+		return NULL;
+	}
+
+	/*
+	 * RE requires at least 2 sources, if given only one source, we pass
+	 * the second source same as the first one.
+	 * With only one source, generate P is meaningless, only care generate
+	 * Q.
+	 */
+	if (src_cnt == 1) {
+		struct dma_async_tx_descriptor *tx = NULL;
+		dma_addr_t dma_src[2];
+		unsigned char coef[2];
+		dma_src[0] = *src;
+		coef[0] = *scf;
+		dma_src[1] = *src;
+		coef[1] = 0;
+		tx = re_jr_prep_genq(chan, dest[1], dma_src, 2, coef, len,
+				flags);
+		if (tx) {
+			desc = to_fsl_re_dma_desc(tx);
+			desc->src_cnt = 1;
+		}
+		return tx;
+	}
+
+	/*
+	 * During RAID6 array creation, Linux's MD layer gets P and Q
+	 * calculated separately in two steps. But our RAID Engine has
+	 * the capability to calculate both P and Q with a single command
+	 * Hence to merge well with MD layer, we need to provide a hook
+	 * here and call re_jq_prep_genq() function
+	 */
+
+	if (flags & DMA_PREP_PQ_DISABLE_P)
+		return re_jr_prep_genq(chan, dest[1], src, src_cnt,
+				scf, len, flags);
+
+	jr = container_of(chan, struct re_jr, chan);
+	desc = re_jr_alloc_desc(jr, cfs_reqd, flags);
+	if (!desc)
+		return ERR_PTR(-ENOMEM);
+
+	desc->dma_len = len;
+	desc->dest_cnt = 2;
+	desc->src_cnt = src_cnt;
+
+	/* Filling frame 0 of compound frame descriptor with CDB */
+	pq = dma_pool_alloc(jr->desc_pool, GFP_ATOMIC, &paddr);
+	if (!pq) {
+		re_jr_free_desc(desc);
+		return ERR_PTR(-ENOMEM);
+	}
+	desc->cdb_addr = pq;
+	desc->cdb_paddr = paddr;
+	desc->cdb_len = sizeof(struct pq_cdb);
+	memset(pq, 0, desc->cdb_len);
+	desc->cdb_opcode = RE_PQ_OPCODE;
+
+	/* Filling GenQQ CDB */
+	pq->opcode = RE_PQ_OPCODE;
+	pq->excl_enable = 0x0; /* Don't exclude for Q1/Q2 parity */
+	pq->excl_q1 = 0x0;
+	pq->excl_q2 = 0x0;
+	pq->blk_size = RE_BLOCK_SIZE;
+	pq->cache_attrib = CACHEABLE_INPUT_OUTPUT;
+	pq->buffer_attrib = BUFFERABLE_OUTPUT;
+	pq->error_attrib = INTERRUPT_ON_ERROR;
+	pq->data_depend = DATA_DEPENDENCY;
+	pq->dpi = ENABLE_DPI;
+	pq->nrcs = (src_cnt - 1);
+
+	p = pq->gfm_q1;
+	/* Init gfm_q1[] */
+	for (i = 0; i < src_cnt; i++)
+		p[i] = 1;
+
+	/* Align gfm[] to 32bit */
+	gfmq_len = ((src_cnt+3)/4)*4;
+
+	/* Init gfm_q2[] */
+	p += gfmq_len;
+	for (i = 0; i < src_cnt; i++)
+		p[i] = scf[i];
+
+	/* Filling frame 0 of compound frame descriptor with CDB */
+	cf = desc->cf_addr;
+	fill_cfd_frame(cf, 0, desc->cdb_len, paddr, 0);
+
+	/* Fill CFD's 1st & 2nd frame with dest buffers */
+	for (i = 1, j = 0; i < 3; i++, j++)
+		fill_cfd_frame(cf, i, len, dest[j], 0);
+
+	/* Fill CFD's rest of the frames with source buffers */
+	for (i = 3, j = 0; j < src_cnt; i++, j++)
+		fill_cfd_frame(cf, i, len, src[j], 0);
+
+	/* Setting the final bit in the last source buffer frame in CFD */
+	cf[i - 1].final = 1;
+
+	return &desc->async_tx;
+}
+
+/*
+ * Prep function for memcpy. In RAID Engine, memcpy is done through MOVE
+ * command. Logic of this function will need to be modified once multipage
+ * support is added in Linux's MD/ASYNC Layer
+ */
+static struct dma_async_tx_descriptor *re_jr_prep_memcpy(
+		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
+		size_t len, unsigned long flags)
+{
+	struct re_jr *jr = NULL;
+	struct fsl_re_dma_async_tx_desc *desc = NULL;
+	size_t length = 0;
+	unsigned int cfs_reqd = 3; /* CDB+dest+src */
+	struct cmpnd_frame *cf = NULL;
+	struct move_cdb *move = NULL;
+	dma_addr_t paddr;
+
+	jr = container_of(chan, struct re_jr, chan);
+
+	if (len > MAX_DATA_LENGTH) {
+		pr_err("%s: Length greater than %d not supported\n",
+				__func__, MAX_DATA_LENGTH);
+		return NULL;
+	}
+
+	desc = re_jr_alloc_desc(jr, cfs_reqd, flags);
+	if (!desc)
+		return ERR_PTR(-ENOMEM);
+
+	desc->dma_len = len;
+	desc->src_cnt = 1;
+	desc->dest_cnt = 1;
+
+	move = dma_pool_alloc(jr->desc_pool, GFP_ATOMIC, &paddr);
+	if (!move) {
+		re_jr_free_desc(desc);
+		return ERR_PTR(-ENOMEM);
+	}
+	desc->cdb_addr = move;
+	desc->cdb_paddr = paddr;
+	desc->cdb_len = sizeof(struct move_cdb);
+	memset(move, 0, desc->cdb_len);
+	desc->cdb_opcode = RE_MOVE_OPCODE;
+
+	/* Filling move CDB */
+	move->opcode = RE_MOVE_OPCODE; /* Unicast move */
+	move->blk_size = RE_BLOCK_SIZE;
+	move->cache_attrib = CACHEABLE_INPUT_OUTPUT;
+	move->buffer_attrib = BUFFERABLE_OUTPUT;
+	move->error_attrib = INTERRUPT_ON_ERROR;
+	move->data_depend = DATA_DEPENDENCY;
+	move->dpi = ENABLE_DPI;
+
+	/* Filling frame 0 of CFD with move CDB */
+	cf = desc->cf_addr;
+	fill_cfd_frame(cf, 0, desc->cdb_len, paddr, 0);
+
+	length = min_t(size_t, len, MAX_DATA_LENGTH);
+
+	/* Fill CFD's 1st frame with dest buffer */
+	fill_cfd_frame(cf, 1, length, dest, 0);
+
+	/* Fill CFD's 2nd frame with src buffer */
+	fill_cfd_frame(cf, 2, length, src, 1);
+
+	return &desc->async_tx;
+}
+
+/*
+ * Job ring probe function. This function gets called for each detected
+ * job ring in the system
+ */
+int re_jr_probe(struct platform_device *ofdev,
+		struct device_node *np, u8 q, u32 *off)
+{
+	struct device *dev = NULL;
+	struct re_drv_private *repriv = NULL;
+	struct re_jr *jr = NULL;
+	struct dma_device *dma_dev = NULL;
+	u32 *ptr = NULL;
+	u32 status;
+	int ret = 0;
+	int k = 0;
+	struct platform_device *jr_ofdev = NULL;
+
+	dev = &ofdev->dev;
+	repriv = dev_get_drvdata(dev);
+	dma_dev = &repriv->dma_dev;
+
+	jr = kzalloc(sizeof(struct re_jr), GFP_KERNEL);
+	if (!jr) {
+		dev_err(dev, "%s: No free memory for allocating JR struct\n",
+			__func__);
+		return -ENOMEM;
+	}
+
+	jr_ofdev = of_platform_device_create(np, NULL, dev);
+	if (jr_ofdev == NULL) {
+		dev_err(dev, "%s: Not able to create ofdev for jr %d\n",
+			__func__, q);
+		ret = -EINVAL;
+		goto err_free_2;
+	}
+	dev_set_drvdata(&jr_ofdev->dev, jr);
+
+	ptr = (u32 *)of_get_property(np, "reg", NULL);
+	if (!ptr) {
+		dev_err(dev, "%s: Reg property not found in JR number %d\n",
+			__func__, q);
+		ret = -ENODEV;
+		goto err_free_2;
+	}
+
+	jr->jrregs = (struct jr_config_regs *)((u8 *)repriv->re_regs +
+			*off + *ptr);
+
+	jr->irq = irq_of_parse_and_map(np, 0);
+	if (jr->irq == NO_IRQ) {
+		dev_err(dev, "%s: No IRQ defined for JR %d\n", __func__, q);
+		ret = -ENODEV;
+		goto err_free_2;
+	}
+
+	tasklet_init(&jr->irqtask, re_jr_dequeue,
+			(unsigned long)&jr_ofdev->dev);
+	ret = request_irq(jr->irq, re_jr_interrupt, 0, "re-jr", &jr_ofdev->dev);
+	if (ret) {
+		dev_err(dev, "%s: Unable to register JR interrupt for JR %d\n",
+			__func__, q);
+		ret = -EINVAL;
+		goto err_free_2;
+	}
+
+	repriv->re_jrs[q] = jr;
+	jr->chan.device = dma_dev;
+	jr->dev = &ofdev->dev;
+	jr->chan.private = jr;
+
+	INIT_LIST_HEAD(&jr->submit_q);
+	INIT_LIST_HEAD(&jr->ack_q);
+	spin_lock_init(&jr->desc_lock);
+
+	init_timer(&jr->timer);
+	jr->timer.expires = jiffies + 10*HZ;
+	jr->timer.function = raide_timer_handler;
+
+	list_add_tail(&jr->chan.device_node, &dma_dev->channels);
+	dma_dev->chancnt++;
+
+	jr->inb_desc_pool = dma_pool_create("re_jr_inb_desc_pool", jr->dev,
+			sizeof(struct jr_hw_desc) * RING_SIZE,
+			FRAME_DESC_ALIGNMENT, 0);
+	if (!jr->inb_desc_pool) {
+		dev_err(dev, "%s:No memory for re_jr_inb_desc_pool\n",
+			__func__);
+		ret = -ENOMEM;
+		goto err_free_2;
+	}
+
+	jr->inb_ring_virt_addr = dma_pool_alloc(jr->inb_desc_pool, GFP_ATOMIC,
+			&jr->inb_phys_addr);
+	if (!jr->inb_ring_virt_addr) {
+		dev_err(dev, "%s:No dma_pool mem for inb_ring_virt_addr\n",
+			__func__);
+		ret = -ENOMEM;
+		goto pool_destroy;
+	}
+
+	jr->oub_desc_pool = dma_pool_create("re_jr_oub_desc_pool", jr->dev,
+			sizeof(struct jr_hw_desc) * RING_SIZE,
+			FRAME_DESC_ALIGNMENT, 0);
+	if (!jr->oub_desc_pool) {
+		dev_err(dev, "%s:No memory for re_jr_oub_desc_pool\n",
+			__func__);
+		ret = -ENOMEM;
+		goto pool_free;
+	}
+
+	jr->oub_ring_virt_addr = dma_pool_alloc(jr->oub_desc_pool, GFP_ATOMIC,
+			&jr->oub_phys_addr);
+	if (!jr->inb_ring_virt_addr) {
+		dev_err(dev, "%s:No dma_pool mem for oub_ring_virt_addr\n",
+			__func__);
+		ret = -ENOMEM;
+		goto pool_destroy_2;
+	}
+
+	jr->desc_pool = dma_pool_create("re_jr_desc_pool", jr->dev,
+			RE_CF_CDB_SIZE, RE_CF_CDB_ALIGN, 0);
+
+	if (!jr->desc_pool) {
+		dev_err(dev, "%s:No memory for re_jr_desc_pool\n",
+			__func__);
+		ret = -ENOMEM;
+		goto err_free_2;
+	}
+
+	jr->inb_ring_index = 0;
+	jr->inb_count = 0;
+	jr->oub_count = 0;
+
+	status = in_be32(&jr->jrregs->jr_status);
+
+	if (status & RE_JR_PAUSE) {
+		dev_info(dev, "%s: JR is in paused state...enable it\n",
+			__func__);
+	} else {
+		dev_err(dev, "%s: Error:- JR shud be in paused state\n",
+			__func__);
+		ret = -EINVAL;
+		goto pool_free_2;
+	}
+
+	/* Program the Inbound/Outbound ring base addresses and size */
+	out_be32(&jr->jrregs->inbring_base_h,
+			jr->inb_phys_addr & RE_JR_ADDRESS_BIT_MASK);
+	out_be32(&jr->jrregs->oubring_base_h,
+			jr->oub_phys_addr & RE_JR_ADDRESS_BIT_MASK);
+	out_be32(&jr->jrregs->inbring_base_l,
+			jr->inb_phys_addr >> RE_JR_ADDRESS_BIT_SHIFT);
+	out_be32(&jr->jrregs->oubring_base_l,
+			jr->oub_phys_addr >> RE_JR_ADDRESS_BIT_SHIFT);
+	out_be32(&jr->jrregs->inbring_size, RING_SIZE << RING_SIZE_SHIFT);
+	out_be32(&jr->jrregs->oubring_size, RING_SIZE << RING_SIZE_SHIFT);
+
+	/* Read LIODN value from u-boot */
+	status = in_be32(&jr->jrregs->jr_config_1) & RE_JR_REG_LIODN_MASK;
+
+	/* Program the CFG reg */
+	out_be32(&jr->jrregs->jr_config_1,
+			RE_JR_CFG1_CBSI | RE_JR_CFG1_CBS0 | status);
+
+	/* Enable RE/JR */
+	out_be32(&jr->jrregs->jr_command, RE_JR_ENABLE);
+
+	/* Zero'ing the array */
+	for (k = 0; k < RING_SIZE; k++) {
+		jr->virt_arry[k].virt_addr = 0;
+		jr->virt_arry[k].phys_addr = 0;
+	}
+
+	return 0;
+
+pool_free_2:
+	dma_pool_free(jr->oub_desc_pool, jr->oub_ring_virt_addr,
+			jr->oub_phys_addr);
+pool_destroy_2:
+	dma_pool_destroy(jr->oub_desc_pool);
+pool_free:
+	dma_pool_free(jr->inb_desc_pool, jr->inb_ring_virt_addr,
+			jr->inb_phys_addr);
+pool_destroy:
+	dma_pool_destroy(jr->inb_desc_pool);
+err_free_2:
+	kfree(jr);
+	return ret;
+}
+
+/* Probe function for RAID Engine */
+static int __devinit raide_probe(struct platform_device *ofdev)
+{
+	struct re_drv_private *repriv = NULL;
+	struct device *dev = NULL;
+	struct device_node *np = NULL;
+	struct device_node *child = NULL;
+	u32 *off = NULL;
+	u8 ridx = 0;
+	struct dma_device *dma_dev = NULL;
+
+	dev_info(&ofdev->dev, "Freescale RAID Engine driver\n");
+
+	repriv = kzalloc(sizeof(struct re_drv_private), GFP_KERNEL);
+	if (!repriv) {
+		dev_err(dev, "%s: No memory for repriv\n", __func__);
+		return -ENOMEM;
+	}
+
+	dev = &ofdev->dev;
+	dev_set_drvdata(dev, repriv);
+
+	/* IOMAP the entire RAID Engine region */
+	repriv->re_regs = of_iomap(ofdev->dev.of_node, 0);
+	if (repriv->re_regs == NULL) {
+		dev_err(dev, "%s: of_iomap failed\n", __func__);
+		kfree(repriv);
+		return -ENOMEM;
+	}
+
+
+	re_sw_desc_cache = kmem_cache_create("re_desc_cache",
+				sizeof(struct fsl_re_dma_async_tx_desc), 0,
+				SLAB_HWCACHE_ALIGN, NULL);
+	if (re_sw_desc_cache == NULL) {
+		dev_err(dev, "%s: No memory for re_sw_desc_cache\n", __func__);
+		iounmap(repriv->re_regs);
+		kfree(repriv);
+		return -ENOMEM;
+	}
+
+	/* Print the RE version to make sure RE is alive */
+	dev_info(dev, "Ver = %x\n", in_be32(&repriv->re_regs->re_version_id));
+
+	/* Program the RE mode */
+	out_be32(&repriv->re_regs->global_config, RE_NON_DPAA_MODE);
+	dev_info(dev, "%s:RE mode is %x\n", __func__,
+			in_be32(&repriv->re_regs->global_config));
+
+	/* Program Galois Field polymomial */
+	out_be32(&repriv->re_regs->galois_field_config, RE_GFM_POLY);
+	dev_info(dev, "%s:Galois Field Polynomial is %x\n", __func__,
+			in_be32(&repriv->re_regs->galois_field_config));
+
+	dma_dev = &repriv->dma_dev;
+	dma_dev->dev = dev;
+	INIT_LIST_HEAD(&dma_dev->channels);
+
+	dma_dev->device_alloc_chan_resources = re_jr_alloc_chan_resources;
+	dma_dev->device_tx_status = re_jr_tx_status;
+	dma_dev->device_issue_pending = re_jr_issue_pending;
+
+	dma_dev->max_xor = MAX_XOR_SRCS;
+	dma_dev->device_prep_dma_xor = re_jr_prep_dma_xor;
+	dma_cap_set(DMA_XOR, dma_dev->cap_mask);
+
+	dma_dev->max_pq = MAX_PQ_SRCS;
+	dma_dev->device_prep_dma_pq = re_jr_prep_pq;
+	dma_cap_set(DMA_PQ, dma_dev->cap_mask);
+
+	dma_dev->device_prep_dma_memcpy = re_jr_prep_memcpy;
+	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
+
+	dma_dev->device_free_chan_resources = re_jr_free_chan_resources;
+
+	repriv->total_jrs = 0;
+
+	/* Parse Device tree to find out the total number of JQs present */
+	for_each_compatible_node(np, NULL, "fsl,raideng-v1.0-job-queue") {
+		off = (u32 *)of_get_property(np, "reg", NULL);
+		if (!off) {
+			dev_err(dev, "%s: Reg property not found in JQ node\n",
+				__func__);
+			return -ENODEV;
+		}
+
+		/* Find out the Job Rings present under each JQ */
+		for_each_child_of_node(np, child) {
+			if (of_device_is_compatible(child,
+				"fsl,raideng-v1.0-job-ring")) {
+				re_jr_probe(ofdev, child, ridx++, off);
+				repriv->total_jrs++;
+			}
+		}
+	}
+
+	dma_async_device_register(dma_dev);
+
+	return 0;
+}
+
+static void release_jr(struct re_jr *jr)
+{
+	/* Free the memory allocated from DMA pools and destroy them */
+	dma_pool_free(jr->oub_desc_pool, jr->oub_ring_virt_addr,
+			jr->oub_phys_addr);
+	dma_pool_destroy(jr->oub_desc_pool);
+	dma_pool_free(jr->inb_desc_pool, jr->inb_ring_virt_addr,
+			jr->inb_phys_addr);
+	dma_pool_destroy(jr->inb_desc_pool);
+
+	kfree(jr);
+}
+
+static int raide_remove(struct platform_device *ofdev)
+{
+	struct re_drv_private *repriv = NULL;
+	struct device *dev = NULL;
+	int i;
+
+	dev = &ofdev->dev;
+	repriv = dev_get_drvdata(dev);
+
+	/* Cleanup JR related memory areas */
+	for (i = 0; i < repriv->total_jrs; i++)
+		release_jr(repriv->re_jrs[i]);
+
+	kmem_cache_destroy(re_sw_desc_cache);
+
+	/* Unregister the driver */
+	dma_async_device_unregister(&repriv->dma_dev);
+
+	/* Unmap the RAID Engine region */
+	iounmap(repriv->re_regs);
+
+	kfree(repriv);
+
+	return 0;
+}
+
+static struct of_device_id raide_ids[] = {
+	{ .compatible = "fsl,raideng-v1.0", },
+	{}
+};
+
+static struct platform_driver raide_driver = {
+	.driver = {
+		.name = "fsl-raideng",
+		.owner = THIS_MODULE,
+		.of_match_table = raide_ids,
+	},
+	.probe = raide_probe,
+	.remove = raide_remove,
+};
+
+static __init int raide_init(void)
+{
+	int ret = 0;
+
+	ret = platform_driver_register(&raide_driver);
+	if (ret)
+		pr_err("fsl-raid: Failed to register platform driver\n");
+
+	return ret;
+}
+
+static void __exit raide_exit(void)
+{
+	platform_driver_unregister(&raide_driver);
+}
+
+subsys_initcall(raide_init);
+module_exit(raide_exit);
+
+MODULE_AUTHOR("Harninder Rai <harninder.rai@freescale.com>");
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("Freescale RAID Engine Device Driver");
diff --git a/drivers/dma/fsl_raid.h b/drivers/dma/fsl_raid.h
new file mode 100644
index 0000000..0f39b8d
--- /dev/null
+++ b/drivers/dma/fsl_raid.h
@@ -0,0 +1,294 @@ 
+/*
+ * drivers/dma/fsl_raid.h
+ *
+ * Freescale RAID Engine device driver
+ *
+ * Author:
+ *	Harninder Rai <harninder.rai@freescale.com>
+ *	Naveen Burmi <naveenburmi@freescale.com>
+ *	Xuelin Shi <b29237@freescale.com>
+ *
+ * Copyright (c) 2010-2012 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.
+ *
+ */
+
+#define RE_DPAA_MODE		(1 << 30)
+#define RE_NON_DPAA_MODE	(1 << 31)
+#define RE_GFM_POLY		(0x1d000000)
+#define RE_JR_INB_JOB_ADD	(1 << 16)
+#define RE_JR_OUB_JOB_REMOVE	(1 << 16)
+#define RE_JR_CFG1_CBSI		0x08000000
+#define RE_JR_CFG1_CBS0		0x00080000
+#define RE_JR_OUB_SLOT_FULL_SHIFT	8
+#define RE_JR_OUB_SLOT_FULL(x)	((x) >> RE_JR_OUB_SLOT_FULL_SHIFT)
+#define RE_PQ_OPCODE		0x1B
+#define RE_XOR_OPCODE		0x1A
+#define RE_MOVE_OPCODE		0x8
+#define FRAME_DESC_ALIGNMENT	16
+#define RE_BLOCK_SIZE		0x3 /* 4096 bytes */
+#define CACHEABLE_INPUT_OUTPUT	0x0
+#define BUFFERABLE_OUTPUT	0x0
+#define INTERRUPT_ON_ERROR	0x1
+#define DATA_DEPENDENCY		0x1
+#define ENABLE_DPI		0x0
+#define RING_SIZE		0x1000
+#define RING_SIZE_SHIFT		8
+#define RE_JR_ADDRESS_BIT_SHIFT	4
+#define RE_JR_ADDRESS_BIT_MASK	((1 << RE_JR_ADDRESS_BIT_SHIFT) - 1)
+#define RE_JR_ERROR		0x40000000
+#define RE_JR_INTERRUPT		0x80000000
+#define RE_JR_CLEAR_INT		0x80000000
+#define RE_JR_PAUSE		0x80000000
+#define RE_JR_ENABLE		0x80000000
+
+#define RE_JR_REG_LIODN_MASK	0x00000fff
+#define RE_CF_CDB_ALIGN		64
+
+/*
+ * max size is 19*sizeof(struct cmpnd_frame), 64 bytes align to 320
+ * here 19 = 1(cdb)+2(dest)+16(src)
+ */
+#define RE_CF_CDB_SIZE		320
+
+struct re_ctrl {
+	/* General Configuration Registers */
+	__be32 global_config;	/* Global Configuration Register */
+	u8     rsvd1[4];
+	__be32 galois_field_config; /* Galois Field Configuration Register */
+	u8     rsvd2[4];
+	__be32 jq_wrr_config;   /* WRR Configuration register */
+	u8     rsvd3[4];
+	__be32 crc_config;	/* CRC Configuration register */
+	u8     rsvd4[228];
+	__be32 system_reset;	/* System Reset Register */
+	u8     rsvd5[252];
+	__be32 global_status;	/* Global Status Register */
+	u8     rsvd6[832];
+	__be32 re_liodn_base;	/* LIODN Base Register */
+	u8     rsvd7[1712];
+	__be32 re_version_id;	/* Version ID register of RE */
+	__be32 re_version_id_2; /* Version ID 2 register of RE */
+	u8     rsvd8[512];
+	__be32 host_config;	/* Host I/F Configuration Register */
+};
+
+struct jr_config_regs {
+	/* Registers for JR interface */
+	__be32 jr_config_0;	/* Job Queue Configuration 0 Register */
+	__be32 jr_config_1;	/* Job Queue Configuration 1 Register */
+	__be32 jr_interrupt_status; /* Job Queue Interrupt Status Register */
+	u8     rsvd1[4];
+	__be32 jr_command;	/* Job Queue Command Register */
+	u8     rsvd2[4];
+	__be32 jr_status;	/* Job Queue Status Register */
+	u8     rsvd3[228];
+
+	/* Input Ring */
+	__be32 inbring_base_h;	/* Inbound Ring Base Address Register - High */
+	__be32 inbring_base_l;	/* Inbound Ring Base Address Register - Low */
+	__be32 inbring_size;	/* Inbound Ring Size Register */
+	u8     rsvd4[4];
+	__be32 inbring_slot_avail; /* Inbound Ring Slot Available Register */
+	u8     rsvd5[4];
+	__be32 inbring_add_job;	/* Inbound Ring Add Job Register */
+	u8     rsvd6[4];
+	__be32 inbring_cnsmr_indx; /* Inbound Ring Consumer Index Register */
+	u8     rsvd7[220];
+
+	/* Output Ring */
+	__be32 oubring_base_h;	/* Outbound Ring Base Address Register - High */
+	__be32 oubring_base_l;	/* Outbound Ring Base Address Register - Low */
+	__be32 oubring_size;	/* Outbound Ring Size Register */
+	u8     rsvd8[4];
+	__be32 oubring_job_rmvd; /* Outbound Ring Job Removed Register */
+	u8     rsvd9[4];
+	__be32 oubring_slot_full; /* Outbound Ring Slot Full Register */
+	u8     rsvd10[4];
+	__be32 oubring_prdcr_indx; /* Outbound Ring Producer Index */
+};
+
+/*
+ * Command Descriptor Block (CDB) for unicast move command.
+ * In RAID Engine terms, memcpy is done through move command
+ */
+struct move_cdb {
+	u32 opcode:5;
+	u32 rsvd1:11;
+	u32 blk_size:2;
+	u32 cache_attrib:2;
+	u32 buffer_attrib:1;
+	u32 error_attrib:1;
+	u32 rsvd2:6;
+	u32 data_depend:1;
+	u32 dpi:1;
+	u32 rsvd3:2;
+} __packed;
+
+/* Data protection/integrity related fields */
+struct dpi_related {
+	u32 apps_mthd:2;
+	u32 ref_mthd:2;
+	u32 guard_mthd:2;
+	u32 dpi_attr:2;
+	u32 rsvd1:8;
+	u32 meta_tag:16;
+	u32 ref_tag:32;
+} __packed;
+
+/*
+ * CDB for GenQ command. In RAID Engine terminology, XOR is
+ * done through this command
+ */
+struct xor_cdb {
+	u32 opcode:5;
+	u32 rsvd1:11;
+	u32 blk_size:2;
+	u32 cache_attrib:2;
+	u32 buffer_attrib:1;
+	u32 error_attrib:1;
+	u32 nrcs:4;
+	u32 rsvd2:2;
+	u32 data_depend:1;
+	u32 dpi:1;
+	u32 rsvd3:2;
+	u8 gfm[16];
+	struct dpi_related dpi_dest_spec;
+	struct dpi_related dpi_src_spec[16];
+} __packed;
+
+/* CDB for no-op command */
+struct noop_cdb {
+	u32 opcode:5;
+	u32 rsvd1:23;
+	u32 dependency:1;
+	u32 rsvd2:3;
+} __packed;
+
+/*
+ * CDB for GenQQ command. In RAID Engine terminology, P/Q is
+ * done through this command
+ */
+struct pq_cdb {
+	u32 opcode:5;
+	u32 rsvd1:1;
+	u32 excl_enable:2;
+	u32 excl_q1:4;
+	u32 excl_q2:4;
+	u32 blk_size:2;
+	u32 cache_attrib:2;
+	u32 buffer_attrib:1;
+	u32 error_attrib:1;
+	u32 nrcs:4;
+	u32 rsvd2:2;
+	u32 data_depend:1;
+	u32 dpi:1;
+	u32 rsvd3:2;
+	u8 gfm_q1[16];
+	u8 gfm_q2[16];
+	struct dpi_related dpi_dest_spec[2];
+	struct dpi_related dpi_src_spec[16];
+} __packed;
+
+/* Compound frame */
+struct cmpnd_frame {
+	u64 rsvd1:24;
+	u64 address:40;
+	u32 extension:1;
+	u32 final:1;
+	u32 rsvd3:10;
+	u32 length:20;
+	u32 rsvd4:8;
+	u32 bpid:8;
+	u32 rsvd5:3;
+	u32 offset:13;
+} __packed;
+
+/* Frame descriptor */
+struct jr_hw_desc {
+	u64 debug:2;
+	u64 liodn_off:6;
+	u64 bpid:8;
+	u64 eliodn_off:4;
+	u64 rsvd1:4;
+	u64 address:40;
+	u64 format:3;
+	u64 rsvd2:29;
+	u64 status:32;
+} __packed;
+
+/* Array to store the virtual/physical address of descriptors */
+struct virt_struct {
+	unsigned long virt_addr;
+	phys_addr_t phys_addr;
+};
+
+/* Per job ring data structure */
+struct re_jr {
+	dma_cookie_t completed_cookie;
+	spinlock_t desc_lock;
+	struct list_head submit_q;
+	struct list_head ack_q;
+	struct device *dev;
+	struct dma_chan chan;
+	struct jr_config_regs *jrregs;
+	int irq;
+	struct tasklet_struct irqtask;
+	dma_addr_t inb_phys_addr;
+	dma_addr_t oub_phys_addr;
+	struct dma_pool *inb_desc_pool;
+	struct dma_pool *oub_desc_pool;
+	struct jr_hw_desc *inb_ring_virt_addr;
+	struct jr_hw_desc *oub_ring_virt_addr;
+	u32 inb_ring_index;
+	u32 inb_count;
+	u32 oub_count;
+	struct virt_struct virt_arry[RING_SIZE];
+	struct timer_list timer;
+	struct dma_pool *desc_pool;
+};
+
+/* Async transaction descriptor */
+struct fsl_re_dma_async_tx_desc {
+	struct dma_async_tx_descriptor async_tx;
+	struct list_head node;
+	struct list_head tx_list;
+	struct jr_hw_desc hwdesc;
+	struct re_jr *jr;
+	void *cf_addr;
+	int dma_len;
+	int dest_cnt;
+	int src_cnt;
+	dma_addr_t cf_paddr;
+	int cf_len;
+	void *cdb_addr;
+	u32 cdb_opcode;
+	dma_addr_t cdb_paddr;
+	int cdb_len;
+};