diff mbox

[U-Boot,15/23] pci: tegra: Add Tegra PCIe driver

Message ID 1408346196-30419-16-git-send-email-thierry.reding@gmail.com
State Superseded
Headers show

Commit Message

Thierry Reding Aug. 18, 2014, 7:16 a.m. UTC
From: Thierry Reding <treding@nvidia.com>

Add support for the PCIe controller found on some generations of Tegra.
Tegra20 has 2 root ports with a total of 4 lanes, Tegra30 has 3 root
ports with a total of 6 lanes and Tegra124 has 2 root ports with a total
of 5 lanes.

Signed-off-by: Thierry Reding <treding@nvidia.com>
---
 drivers/pci/Makefile    |    1 +
 drivers/pci/pci_tegra.c | 1133 +++++++++++++++++++++++++++++++++++++++++++++++
 include/fdtdec.h        |    3 +
 lib/fdtdec.c            |    3 +
 4 files changed, 1140 insertions(+)
 create mode 100644 drivers/pci/pci_tegra.c

Comments

Stephen Warren Aug. 20, 2014, 7:04 p.m. UTC | #1
On 08/18/2014 01:16 AM, Thierry Reding wrote:
> From: Thierry Reding <treding@nvidia.com>
>
> Add support for the PCIe controller found on some generations of Tegra.
> Tegra20 has 2 root ports with a total of 4 lanes, Tegra30 has 3 root
> ports with a total of 6 lanes and Tegra124 has 2 root ports with a total
> of 5 lanes.

> diff --git a/drivers/pci/pci_tegra.c b/drivers/pci/pci_tegra.c

> + * Copyright (c) 2010, CompuLab, Ltd.
> + * Author: Mike Rapoport <mike@compulab.co.il>
> + *
> + * Based on NVIDIA PCIe driver
> + * Copyright (c) 2008-2009, NVIDIA Corporation.

Presumably also at least (c) 2013-2014 NVIDIA too?

Do you have Mike's S-o-b for the initial work?

> +static int tegra_pcie_read_conf(struct pci_controller *hose, pci_dev_t bdf,
> +				int where, u32 *value)
> +{
> +	struct tegra_pcie *pcie = to_tegra_pcie(hose);
> +	unsigned long address;
> +	int err;
> +
> +	err = tegra_pcie_conf_address(pcie, bdf, where, &address);
> +	if (err < 0) {
> +		*value = 0xffffffff;
> +		return 1;
> +	}

In the kernel, don't we have to play games to dynamically map/unmap 
memory regions and/or map pages in a non-linear fashion since the config 
space is so large, and bus/device ID -> address mapping is odd? I don't 
see that here.

> +static bool tegra_pcie_port_check_link(struct tegra_pcie_port *port)
> +{
> +	unsigned int retries = 3;
> +	unsigned long value;
> +
> +	value = rp_readl(port, RP_PRIV_MISC);
> +	value &= ~RP_PRIV_MISC_PRSNT_MAP_EP_ABSNT;
> +	value |= RP_PRIV_MISC_PRSNT_MAP_EP_PRSNT;
> +	rp_writel(port, value, RP_PRIV_MISC);
> +
> +	do {
> +		unsigned int timeout = 200;
> +
> +		do {
> +			value = rp_readl(port, RP_VEND_XP);
> +			if (value & RP_VEND_XP_DL_UP)
> +				break;

On my board, this almost never succeeds for Jetson TK1's built-in NIC, 
although it works fine for the mini-PCIe slot. On the other two boards I 
borrowed, this succeeds almost always. I'd prefer if we could get to the 
bottom of this before we actually apply this, although the bug is 
proving quite perplexing. The mainline kernel PCIe patches have the same 
issue.

> +static int process_nodes(const void *fdt, int nodes[], unsigned int count)

> +#ifdef CONFIG_PCI_SCAN_SHOW
> +		printf("PCI: Enumerating devices...\n");
> +		printf("---------------------------------------\n");
> +		printf("  Device        ID          Description\n");
> +		printf("  ------        --          -----------\n");
> +#endif
> +
> +		pcie->hose.last_busno = pci_hose_scan(&pcie->hose);

Shouldn't that ifdef'd block be part of pci_hose_scan()?
Thierry Reding Aug. 22, 2014, 3:24 p.m. UTC | #2
On Wed, Aug 20, 2014 at 01:04:22PM -0600, Stephen Warren wrote:
> On 08/18/2014 01:16 AM, Thierry Reding wrote:
> >From: Thierry Reding <treding@nvidia.com>
> >
> >Add support for the PCIe controller found on some generations of Tegra.
> >Tegra20 has 2 root ports with a total of 4 lanes, Tegra30 has 3 root
> >ports with a total of 6 lanes and Tegra124 has 2 root ports with a total
> >of 5 lanes.
> 
> >diff --git a/drivers/pci/pci_tegra.c b/drivers/pci/pci_tegra.c
> 
> >+ * Copyright (c) 2010, CompuLab, Ltd.
> >+ * Author: Mike Rapoport <mike@compulab.co.il>
> >+ *
> >+ * Based on NVIDIA PCIe driver
> >+ * Copyright (c) 2008-2009, NVIDIA Corporation.
> 
> Presumably also at least (c) 2013-2014 NVIDIA too?

Yes.

> Do you have Mike's S-o-b for the initial work?

Yes, I can take it from the kernel's git log.

> >+static int tegra_pcie_read_conf(struct pci_controller *hose, pci_dev_t bdf,
> >+				int where, u32 *value)
> >+{
> >+	struct tegra_pcie *pcie = to_tegra_pcie(hose);
> >+	unsigned long address;
> >+	int err;
> >+
> >+	err = tegra_pcie_conf_address(pcie, bdf, where, &address);
> >+	if (err < 0) {
> >+		*value = 0xffffffff;
> >+		return 1;
> >+	}
> 
> In the kernel, don't we have to play games to dynamically map/unmap memory
> regions and/or map pages in a non-linear fashion since the config space is
> so large, and bus/device ID -> address mapping is odd? I don't see that
> here.

We do that in the kernel to avoid having to I/O remap a full 256 MiB of
memory even if only a single PCI bus is needed. Since memory is mapped
1:1 in U-Boot and we access physical memory "directly", we don't have to
worry about being wasteful. The tegra_pcie_conf_offset() function
computes the proper offset:

	static unsigned long tegra_pcie_conf_offset(pci_dev_t bdf, int where)
	{
		return ((where & 0xf00) << 16) | (PCI_BUS(bdf) << 16) |
		       (PCI_DEV(bdf) << 11) | (PCI_FUNC(bdf) << 8) |
		       (where & 0xfc);
	}

> >+static bool tegra_pcie_port_check_link(struct tegra_pcie_port *port)
> >+{
> >+	unsigned int retries = 3;
> >+	unsigned long value;
> >+
> >+	value = rp_readl(port, RP_PRIV_MISC);
> >+	value &= ~RP_PRIV_MISC_PRSNT_MAP_EP_ABSNT;
> >+	value |= RP_PRIV_MISC_PRSNT_MAP_EP_PRSNT;
> >+	rp_writel(port, value, RP_PRIV_MISC);
> >+
> >+	do {
> >+		unsigned int timeout = 200;
> >+
> >+		do {
> >+			value = rp_readl(port, RP_VEND_XP);
> >+			if (value & RP_VEND_XP_DL_UP)
> >+				break;
> 
> On my board, this almost never succeeds for Jetson TK1's built-in NIC,
> although it works fine for the mini-PCIe slot. On the other two boards I
> borrowed, this succeeds almost always. I'd prefer if we could get to the
> bottom of this before we actually apply this, although the bug is proving
> quite perplexing. The mainline kernel PCIe patches have the same issue.

That's indeed odd. It's possible that the powergate fix that I've
applied (to wait for the partition to become powered) may improve
things. I'll be pushing out my latest work shortly (I don't think
I'll manage to send out the whole series for review, though), so
in case you get the chance you may want to test it.

> >+static int process_nodes(const void *fdt, int nodes[], unsigned int count)
> 
> >+#ifdef CONFIG_PCI_SCAN_SHOW
> >+		printf("PCI: Enumerating devices...\n");
> >+		printf("---------------------------------------\n");
> >+		printf("  Device        ID          Description\n");
> >+		printf("  ------        --          -----------\n");
> >+#endif
> >+
> >+		pcie->hose.last_busno = pci_hose_scan(&pcie->hose);
> 
> Shouldn't that ifdef'd block be part of pci_hose_scan()?

Every driver does it this way, but I agree that it should presumably go
in the core somewhere. I'll see if I can find a good place to do it.

Thierry
Stephen Warren Aug. 22, 2014, 5:33 p.m. UTC | #3
On 08/20/2014 01:04 PM, Stephen Warren wrote:
> On 08/18/2014 01:16 AM, Thierry Reding wrote:
>> From: Thierry Reding <treding@nvidia.com>
>>
>> Add support for the PCIe controller found on some generations of Tegra.
>> Tegra20 has 2 root ports with a total of 4 lanes, Tegra30 has 3 root
>> ports with a total of 6 lanes and Tegra124 has 2 root ports with a total
>> of 5 lanes.
>
>> diff --git a/drivers/pci/pci_tegra.c b/drivers/pci/pci_tegra.c

>> +static bool tegra_pcie_port_check_link(struct tegra_pcie_port *port)
>> +{
>> +    unsigned int retries = 3;
>> +    unsigned long value;
>> +
>> +    value = rp_readl(port, RP_PRIV_MISC);
>> +    value &= ~RP_PRIV_MISC_PRSNT_MAP_EP_ABSNT;
>> +    value |= RP_PRIV_MISC_PRSNT_MAP_EP_PRSNT;
>> +    rp_writel(port, value, RP_PRIV_MISC);
>> +
>> +    do {
>> +        unsigned int timeout = 200;
>> +
>> +        do {
>> +            value = rp_readl(port, RP_VEND_XP);
>> +            if (value & RP_VEND_XP_DL_UP)
>> +                break;
>
> On my board, this almost never succeeds for Jetson TK1's built-in NIC,
> although it works fine for the mini-PCIe slot. On the other two boards I
> borrowed, this succeeds almost always. I'd prefer if we could get to the
> bottom of this before we actually apply this, although the bug is
> proving quite perplexing. The mainline kernel PCIe patches have the same
> issue.

I tracked the issue down to some missing pinmux configuration for the 
PCIe reset lane. With that in place, both U-Boot and the mainline Linux 
kernel PCIe support appear to work reliably. I'll work on patches to 
resolve that. So, I think we can go ahead and merge these patches 
without issue ASAP, as soon as any review comments are addressed.
Thierry Reding Aug. 22, 2014, 7:41 p.m. UTC | #4
On Fri, Aug 22, 2014 at 11:33:39AM -0600, Stephen Warren wrote:
> On 08/20/2014 01:04 PM, Stephen Warren wrote:
> >On 08/18/2014 01:16 AM, Thierry Reding wrote:
> >>From: Thierry Reding <treding@nvidia.com>
> >>
> >>Add support for the PCIe controller found on some generations of Tegra.
> >>Tegra20 has 2 root ports with a total of 4 lanes, Tegra30 has 3 root
> >>ports with a total of 6 lanes and Tegra124 has 2 root ports with a total
> >>of 5 lanes.
> >
> >>diff --git a/drivers/pci/pci_tegra.c b/drivers/pci/pci_tegra.c
> 
> >>+static bool tegra_pcie_port_check_link(struct tegra_pcie_port *port)
> >>+{
> >>+    unsigned int retries = 3;
> >>+    unsigned long value;
> >>+
> >>+    value = rp_readl(port, RP_PRIV_MISC);
> >>+    value &= ~RP_PRIV_MISC_PRSNT_MAP_EP_ABSNT;
> >>+    value |= RP_PRIV_MISC_PRSNT_MAP_EP_PRSNT;
> >>+    rp_writel(port, value, RP_PRIV_MISC);
> >>+
> >>+    do {
> >>+        unsigned int timeout = 200;
> >>+
> >>+        do {
> >>+            value = rp_readl(port, RP_VEND_XP);
> >>+            if (value & RP_VEND_XP_DL_UP)
> >>+                break;
> >
> >On my board, this almost never succeeds for Jetson TK1's built-in NIC,
> >although it works fine for the mini-PCIe slot. On the other two boards I
> >borrowed, this succeeds almost always. I'd prefer if we could get to the
> >bottom of this before we actually apply this, although the bug is
> >proving quite perplexing. The mainline kernel PCIe patches have the same
> >issue.
> 
> I tracked the issue down to some missing pinmux configuration for the PCIe
> reset lane. With that in place, both U-Boot and the mainline Linux kernel
> PCIe support appear to work reliably. I'll work on patches to resolve that.

Hey, that's great news.

> So, I think we can go ahead and merge these patches without issue ASAP, as
> soon as any review comments are addressed.

Okay, I have v2 pretty much ready with all the review comments addressed
and just need some time to write up a cover letter and send the series
out.

Thierry
diff mbox

Patch

diff --git a/drivers/pci/Makefile b/drivers/pci/Makefile
index e73a49861983..8eb2320539f9 100644
--- a/drivers/pci/Makefile
+++ b/drivers/pci/Makefile
@@ -15,5 +15,6 @@  obj-$(CONFIG_FTPCI100) += pci_ftpci100.o
 obj-$(CONFIG_SH4_PCI) += pci_sh4.o
 obj-$(CONFIG_SH7751_PCI) +=pci_sh7751.o
 obj-$(CONFIG_SH7780_PCI) +=pci_sh7780.o
+obj-$(CONFIG_PCI_TEGRA) += pci_tegra.o
 obj-$(CONFIG_TSI108_PCI) += tsi108_pci.o
 obj-$(CONFIG_WINBOND_83C553) += w83c553f.o
diff --git a/drivers/pci/pci_tegra.c b/drivers/pci/pci_tegra.c
new file mode 100644
index 000000000000..3316ef34e092
--- /dev/null
+++ b/drivers/pci/pci_tegra.c
@@ -0,0 +1,1133 @@ 
+/*
+ * Copyright (c) 2010, CompuLab, Ltd.
+ * Author: Mike Rapoport <mike@compulab.co.il>
+ *
+ * Based on NVIDIA PCIe driver
+ * Copyright (c) 2008-2009, NVIDIA Corporation.
+ *
+ * SPDX-License-Identifier:	GPL-2.0
+ */
+
+#define DEBUG
+#define pr_fmt(fmt) "tegra-pcie: " fmt
+
+#include <common.h>
+#include <errno.h>
+#include <fdtdec.h>
+#include <malloc.h>
+#include <pci.h>
+
+#include <asm/io.h>
+#include <asm/gpio.h>
+
+#include <asm/arch/clock.h>
+#include <asm/arch/powergate.h>
+#include <asm/arch/xusb-padctl.h>
+
+#include <linux/list.h>
+
+#include <dt-bindings/pinctrl/pinctrl-tegra-xusb.h>
+
+DECLARE_GLOBAL_DATA_PTR;
+
+#define AFI_AXI_BAR0_SZ	0x00
+#define AFI_AXI_BAR1_SZ	0x04
+#define AFI_AXI_BAR2_SZ	0x08
+#define AFI_AXI_BAR3_SZ	0x0c
+#define AFI_AXI_BAR4_SZ	0x10
+#define AFI_AXI_BAR5_SZ	0x14
+
+#define AFI_AXI_BAR0_START	0x18
+#define AFI_AXI_BAR1_START	0x1c
+#define AFI_AXI_BAR2_START	0x20
+#define AFI_AXI_BAR3_START	0x24
+#define AFI_AXI_BAR4_START	0x28
+#define AFI_AXI_BAR5_START	0x2c
+
+#define AFI_FPCI_BAR0	0x30
+#define AFI_FPCI_BAR1	0x34
+#define AFI_FPCI_BAR2	0x38
+#define AFI_FPCI_BAR3	0x3c
+#define AFI_FPCI_BAR4	0x40
+#define AFI_FPCI_BAR5	0x44
+
+#define AFI_CACHE_BAR0_SZ	0x48
+#define AFI_CACHE_BAR0_ST	0x4c
+#define AFI_CACHE_BAR1_SZ	0x50
+#define AFI_CACHE_BAR1_ST	0x54
+
+#define AFI_MSI_BAR_SZ		0x60
+#define AFI_MSI_FPCI_BAR_ST	0x64
+#define AFI_MSI_AXI_BAR_ST	0x68
+
+#define AFI_CONFIGURATION		0xac
+#define  AFI_CONFIGURATION_EN_FPCI	(1 << 0)
+
+#define AFI_FPCI_ERROR_MASKS	0xb0
+
+#define AFI_INTR_MASK		0xb4
+#define  AFI_INTR_MASK_INT_MASK	(1 << 0)
+#define  AFI_INTR_MASK_MSI_MASK	(1 << 8)
+
+#define AFI_SM_INTR_ENABLE	0xc4
+#define  AFI_SM_INTR_INTA_ASSERT	(1 << 0)
+#define  AFI_SM_INTR_INTB_ASSERT	(1 << 1)
+#define  AFI_SM_INTR_INTC_ASSERT	(1 << 2)
+#define  AFI_SM_INTR_INTD_ASSERT	(1 << 3)
+#define  AFI_SM_INTR_INTA_DEASSERT	(1 << 4)
+#define  AFI_SM_INTR_INTB_DEASSERT	(1 << 5)
+#define  AFI_SM_INTR_INTC_DEASSERT	(1 << 6)
+#define  AFI_SM_INTR_INTD_DEASSERT	(1 << 7)
+
+#define AFI_AFI_INTR_ENABLE		0xc8
+#define  AFI_INTR_EN_INI_SLVERR		(1 << 0)
+#define  AFI_INTR_EN_INI_DECERR		(1 << 1)
+#define  AFI_INTR_EN_TGT_SLVERR		(1 << 2)
+#define  AFI_INTR_EN_TGT_DECERR		(1 << 3)
+#define  AFI_INTR_EN_TGT_WRERR		(1 << 4)
+#define  AFI_INTR_EN_DFPCI_DECERR	(1 << 5)
+#define  AFI_INTR_EN_AXI_DECERR		(1 << 6)
+#define  AFI_INTR_EN_FPCI_TIMEOUT	(1 << 7)
+#define  AFI_INTR_EN_PRSNT_SENSE	(1 << 8)
+
+#define AFI_PCIE_CONFIG					0x0f8
+#define  AFI_PCIE_CONFIG_PCIE_DISABLE(x)		(1 << ((x) + 1))
+#define  AFI_PCIE_CONFIG_PCIE_DISABLE_ALL		0xe
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_MASK	(0xf << 20)
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_SINGLE	(0x0 << 20)
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_420	(0x0 << 20)
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_X2_X1	(0x0 << 20)
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_DUAL	(0x1 << 20)
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_222	(0x1 << 20)
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_X4_X1	(0x1 << 20)
+#define  AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_411	(0x2 << 20)
+
+#define AFI_FUSE			0x104
+#define  AFI_FUSE_PCIE_T0_GEN2_DIS	(1 << 2)
+
+#define AFI_PEX0_CTRL			0x110
+#define AFI_PEX1_CTRL			0x118
+#define AFI_PEX2_CTRL			0x128
+#define  AFI_PEX_CTRL_RST		(1 << 0)
+#define  AFI_PEX_CTRL_CLKREQ_EN		(1 << 1)
+#define  AFI_PEX_CTRL_REFCLK_EN		(1 << 3)
+#define  AFI_PEX_CTRL_OVERRIDE_EN	(1 << 4)
+
+#define AFI_PLLE_CONTROL		0x160
+#define  AFI_PLLE_CONTROL_BYPASS_PADS2PLLE_CONTROL (1 << 9)
+#define  AFI_PLLE_CONTROL_PADS2PLLE_CONTROL_EN (1 << 1)
+
+#define AFI_PEXBIAS_CTRL_0		0x168
+
+#define PADS_CTL_SEL		0x0000009C
+
+#define PADS_CTL		0x000000A0
+#define  PADS_CTL_IDDQ_1L	(1 <<  0)
+#define  PADS_CTL_TX_DATA_EN_1L	(1 <<  6)
+#define  PADS_CTL_RX_DATA_EN_1L	(1 << 10)
+
+#define PADS_PLL_CTL_TEGRA20			0x000000B8
+#define PADS_PLL_CTL_TEGRA30			0x000000B4
+#define  PADS_PLL_CTL_RST_B4SM			(0x1 <<  1)
+#define  PADS_PLL_CTL_LOCKDET			(0x1 <<  8)
+#define  PADS_PLL_CTL_REFCLK_MASK		(0x3 << 16)
+#define  PADS_PLL_CTL_REFCLK_INTERNAL_CML	(0x0 << 16)
+#define  PADS_PLL_CTL_REFCLK_INTERNAL_CMOS	(0x1 << 16)
+#define  PADS_PLL_CTL_REFCLK_EXTERNAL		(0x2 << 16)
+#define  PADS_PLL_CTL_TXCLKREF_MASK		(0x1 << 20)
+#define  PADS_PLL_CTL_TXCLKREF_DIV10		(0x0 << 20)
+#define  PADS_PLL_CTL_TXCLKREF_DIV5		(0x1 << 20)
+#define  PADS_PLL_CTL_TXCLKREF_BUF_EN		(0x1 << 22)
+
+#define PADS_REFCLK_CFG0			0x000000C8
+#define PADS_REFCLK_CFG1			0x000000CC
+
+/*
+ * Fields in PADS_REFCLK_CFG*. Those registers form an array of 16-bit
+ * entries, one entry per PCIe port. These field definitions and desired
+ * values aren't in the TRM, but do come from NVIDIA.
+ */
+#define PADS_REFCLK_CFG_TERM_SHIFT		2  /* 6:2 */
+#define PADS_REFCLK_CFG_E_TERM_SHIFT		7
+#define PADS_REFCLK_CFG_PREDI_SHIFT		8  /* 11:8 */
+#define PADS_REFCLK_CFG_DRVI_SHIFT		12 /* 15:12 */
+
+/* Default value provided by HW engineering is 0xfa5c */
+#define PADS_REFCLK_CFG_VALUE \
+	( \
+		(0x17 << PADS_REFCLK_CFG_TERM_SHIFT)   | \
+		(0    << PADS_REFCLK_CFG_E_TERM_SHIFT) | \
+		(0xa  << PADS_REFCLK_CFG_PREDI_SHIFT)  | \
+		(0xf  << PADS_REFCLK_CFG_DRVI_SHIFT)     \
+	)
+
+#define RP_VEND_XP	0x00000F00
+#define  RP_VEND_XP_DL_UP	(1 << 30)
+
+#define RP_PRIV_MISC	0x00000FE0
+#define  RP_PRIV_MISC_PRSNT_MAP_EP_PRSNT (0xE << 0)
+#define  RP_PRIV_MISC_PRSNT_MAP_EP_ABSNT (0xF << 0)
+
+#define RP_LINK_CONTROL_STATUS			0x00000090
+#define  RP_LINK_CONTROL_STATUS_DL_LINK_ACTIVE	0x20000000
+#define  RP_LINK_CONTROL_STATUS_LINKSTAT_MASK	0x3fff0000
+
+struct tegra_pcie;
+
+struct tegra_pcie_port {
+	struct tegra_pcie *pcie;
+
+	struct fdt_resource regs;
+	unsigned int num_lanes;
+	unsigned int index;
+
+	struct list_head list;
+};
+
+struct tegra_pcie_soc {
+	unsigned int num_ports;
+	unsigned long pads_pll_ctl;
+	unsigned long tx_ref_sel;
+	bool has_pex_clkreq_en;
+	bool has_pex_bias_ctrl;
+	bool has_cml_clk;
+	bool has_gen2;
+};
+
+struct tegra_pcie {
+	struct pci_controller hose;
+
+	struct fdt_resource pads;
+	struct fdt_resource afi;
+	struct fdt_resource cs;
+
+	struct fdt_resource prefetch;
+	struct fdt_resource mem;
+	struct fdt_resource io;
+
+	struct list_head ports;
+	unsigned long xbar;
+
+	const struct tegra_pcie_soc *soc;
+	struct tegra_xusb_phy *phy;
+};
+
+static inline struct tegra_pcie *to_tegra_pcie(struct pci_controller *hose)
+{
+	return container_of(hose, struct tegra_pcie, hose);
+}
+
+static void afi_writel(struct tegra_pcie *pcie, unsigned long value,
+		       unsigned long offset)
+{
+	writel(value, pcie->afi.start + offset);
+}
+
+static unsigned long afi_readl(struct tegra_pcie *pcie, unsigned long offset)
+{
+	return readl(pcie->afi.start + offset);
+}
+
+static void pads_writel(struct tegra_pcie *pcie, unsigned long value,
+			unsigned long offset)
+{
+	writel(value, pcie->pads.start + offset);
+}
+
+static unsigned long pads_readl(struct tegra_pcie *pcie, unsigned long offset)
+{
+	return readl(pcie->pads.start + offset);
+}
+
+static unsigned long rp_readl(struct tegra_pcie_port *port,
+			      unsigned long offset)
+{
+	return readl(port->regs.start + offset);
+}
+
+static void rp_writel(struct tegra_pcie_port *port, unsigned long value,
+		      unsigned long offset)
+{
+	writel(value, port->regs.start + offset);
+}
+
+static unsigned long tegra_pcie_conf_offset(pci_dev_t bdf, int where)
+{
+	return ((where & 0xf00) << 16) | (PCI_BUS(bdf) << 16) |
+	       (PCI_DEV(bdf) << 11) | (PCI_FUNC(bdf) << 8) |
+	       (where & 0xfc);
+}
+
+static int tegra_pcie_conf_address(struct tegra_pcie *pcie, pci_dev_t bdf,
+				   int where, unsigned long *address)
+{
+	unsigned int bus = PCI_BUS(bdf);
+
+	if (bus == 0) {
+		unsigned int dev = PCI_DEV(bdf);
+		struct tegra_pcie_port *port;
+
+		list_for_each_entry(port, &pcie->ports, list) {
+			if (port->index + 1 == dev) {
+				*address = port->regs.start + (where & ~3);
+				return 0;
+			}
+		}
+	} else {
+		*address = pcie->cs.start + tegra_pcie_conf_offset(bdf, where);
+		return 0;
+	}
+
+	return -1;
+}
+
+static int tegra_pcie_read_conf(struct pci_controller *hose, pci_dev_t bdf,
+				int where, u32 *value)
+{
+	struct tegra_pcie *pcie = to_tegra_pcie(hose);
+	unsigned long address;
+	int err;
+
+	err = tegra_pcie_conf_address(pcie, bdf, where, &address);
+	if (err < 0) {
+		*value = 0xffffffff;
+		return 1;
+	}
+
+	*value = readl(address);
+
+	/* fixup root port class */
+	if (PCI_BUS(bdf) == 0) {
+		if (where == PCI_CLASS_REVISION) {
+			*value &= ~0x00ff0000;
+			*value |= PCI_CLASS_BRIDGE_PCI << 16;
+		}
+	}
+
+	return 0;
+}
+
+static int tegra_pcie_write_conf(struct pci_controller *hose, pci_dev_t bdf,
+				 int where, u32 value)
+{
+	struct tegra_pcie *pcie = to_tegra_pcie(hose);
+	unsigned long address;
+	int err;
+
+	err = tegra_pcie_conf_address(pcie, bdf, where, &address);
+	if (err < 0)
+		return 1;
+
+	writel(value, address);
+
+	return 0;
+}
+
+static int tegra_pcie_port_parse_dt(const void *fdt, int node,
+				    struct tegra_pcie_port *port)
+{
+	const u32 *addr;
+	int len;
+
+	addr = fdt_getprop(fdt, node, "assigned-addresses", &len);
+	if (!addr) {
+		error("property \"assigned-addresses\" not found");
+		return -FDT_ERR_NOTFOUND;
+	}
+
+	port->regs.start = fdt32_to_cpu(addr[2]);
+	port->regs.end = port->regs.start + fdt32_to_cpu(addr[4]);
+
+	return 0;
+}
+
+static int tegra_pcie_get_xbar_config(const void *fdt, int node, u32 lanes,
+				      unsigned long *xbar)
+{
+	enum fdt_compat_id id = fdtdec_lookup(fdt, node);
+
+	switch (id) {
+	case COMPAT_NVIDIA_TEGRA20_PCIE:
+		switch (lanes) {
+		case 0x00000004:
+			debug("single-mode configuration\n");
+			*xbar = AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_SINGLE;
+			return 0;
+
+		case 0x00000202:
+			debug("dual-mode configuration\n");
+			*xbar = AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_DUAL;
+			return 0;
+		}
+		break;
+
+	case COMPAT_NVIDIA_TEGRA30_PCIE:
+		switch (lanes) {
+		case 0x00000204:
+			debug("4x1, 2x1 configuration\n");
+			*xbar = AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_420;
+			return 0;
+
+		case 0x00020202:
+			debug("2x3 configuration\n");
+			*xbar = AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_222;
+			return 0;
+
+		case 0x00010104:
+			debug("4x1, 1x2 configuration\n");
+			*xbar = AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_411;
+			return 0;
+		}
+		break;
+
+	case COMPAT_NVIDIA_TEGRA124_PCIE:
+		switch (lanes) {
+		case 0x0000104:
+			debug("4x1, 1x1 configuration\n");
+			*xbar = AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_X4_X1;
+			return 0;
+
+		case 0x0000102:
+			debug("2x1, 1x1 configuration\n");
+			*xbar = AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_X2_X1;
+			return 0;
+		}
+		break;
+
+	default:
+		break;
+	}
+
+	return -FDT_ERR_NOTFOUND;
+}
+
+static int tegra_pcie_parse_dt_ranges(const void *fdt, int node,
+				      struct tegra_pcie *pcie)
+{
+	const u32 *ptr, *end;
+	int len;
+
+	ptr = fdt_getprop(fdt, node, "ranges", &len);
+	if (!ptr) {
+		error("missing \"ranges\" property");
+		return -FDT_ERR_NOTFOUND;
+	}
+
+	end = ptr + len / 4;
+
+	while (ptr < end) {
+		struct fdt_resource *res = NULL;
+		u32 space = fdt32_to_cpu(*ptr);
+
+		switch ((space >> 24) & 0x3) {
+		case 0x01:
+			res = &pcie->io;
+			break;
+
+		case 0x02: /* 32 bit */
+		case 0x03: /* 64 bit */
+			if (space & (1 << 30))
+				res = &pcie->prefetch;
+			else
+				res = &pcie->mem;
+
+			break;
+		}
+
+		if (res) {
+			res->start = fdt32_to_cpu(ptr[3]);
+			res->end = res->start + fdt32_to_cpu(ptr[5]);
+		}
+
+		ptr += 3 + 1 + 2;
+	}
+
+	debug("PCI regions:\n");
+	debug("  I/O: %#x-%#x\n", pcie->io.start, pcie->io.end);
+	debug("  non-prefetchable memory: %#x-%#x\n", pcie->mem.start,
+	      pcie->mem.end);
+	debug("  prefetchable memory: %#x-%#x\n", pcie->prefetch.start,
+	      pcie->prefetch.end);
+
+	return 0;
+}
+
+static int tegra_pcie_parse_port_info(const void *fdt, int node,
+				      unsigned int *index,
+				      unsigned int *lanes)
+{
+	pci_dev_t bdf;
+	int err;
+
+	err = fdtdec_get_int(fdt, node, "nvidia,num-lanes", 0);
+	if (err < 0) {
+		error("failed to parse \"nvidia,num-lanes\" property");
+		return err;
+	}
+
+	*lanes = err;
+
+	err = fdtdec_pci_get_bdf(fdt, node, &bdf);
+	if (err < 0) {
+		error("failed to parse \"reg\" property");
+		return err;
+	}
+
+	*index = PCI_DEV(bdf) - 1;
+
+	return 0;
+}
+
+static int tegra_pcie_parse_dt(const void *fdt, int node,
+			       struct tegra_pcie *pcie)
+{
+	int err, subnode;
+	u32 lanes = 0;
+
+	err = fdt_get_named_resource(fdt, node, "reg", "reg-names", "pads",
+				     &pcie->pads);
+	if (err < 0) {
+		error("resource \"pads\" not found");
+		return err;
+	}
+
+	err = fdt_get_named_resource(fdt, node, "reg", "reg-names", "afi",
+				     &pcie->afi);
+	if (err < 0) {
+		error("resource \"afi\" not found");
+		return err;
+	}
+
+	err = fdt_get_named_resource(fdt, node, "reg", "reg-names", "cs",
+				     &pcie->cs);
+	if (err < 0) {
+		error("resource \"cs\" not found");
+		return err;
+	}
+
+	pcie->phy = tegra_xusb_phy_get(TEGRA_XUSB_PADCTL_PCIE);
+	if (pcie->phy) {
+		err = tegra_xusb_phy_prepare(pcie->phy);
+		if (err < 0) {
+			error("failed to prepare PHY: %d", err);
+			return err;
+		}
+	}
+
+	err = tegra_pcie_parse_dt_ranges(fdt, node, pcie);
+	if (err < 0) {
+		error("failed to parse \"ranges\" property");
+		return err;
+	}
+
+	fdt_for_each_subnode(fdt, subnode, node) {
+		unsigned int index = 0, num_lanes = 0;
+		struct tegra_pcie_port *port;
+
+		err = tegra_pcie_parse_port_info(fdt, subnode, &index,
+						 &num_lanes);
+		if (err < 0) {
+			error("failed to obtain root port info");
+			continue;
+		}
+
+		lanes |= num_lanes << (index << 3);
+
+		if (!fdtdec_get_is_enabled(fdt, subnode))
+			continue;
+
+		port = malloc(sizeof(*port));
+		if (!port)
+			continue;
+
+		memset(port, 0, sizeof(*port));
+		port->num_lanes = num_lanes;
+		port->index = index;
+
+		err = tegra_pcie_port_parse_dt(fdt, subnode, port);
+		if (err < 0) {
+			free(port);
+			continue;
+		}
+
+		list_add_tail(&port->list, &pcie->ports);
+		port->pcie = pcie;
+	}
+
+	err = tegra_pcie_get_xbar_config(fdt, node, lanes, &pcie->xbar);
+	if (err < 0) {
+		error("invalid lane configuration");
+		return err;
+	}
+
+	return 0;
+}
+
+int __weak tegra_pcie_board_init(void)
+{
+	return 0;
+}
+
+static int tegra_pcie_power_on(struct tegra_pcie *pcie)
+{
+	const struct tegra_pcie_soc *soc = pcie->soc;
+	unsigned long value;
+	int err;
+
+	/* reset PCIEXCLK logic, AFI controller and PCIe controller */
+	reset_set_enable(PERIPH_ID_PCIEXCLK, 1);
+	reset_set_enable(PERIPH_ID_AFI, 1);
+	reset_set_enable(PERIPH_ID_PCIE, 1);
+
+	tegra_powergate_power_off(TEGRA_POWERGATE_PCIE);
+
+	tegra_pcie_board_init();
+
+	tegra_powergate_sequence_power_up(TEGRA_POWERGATE_PCIE,
+					  PERIPH_ID_PCIE);
+
+	/* take AFI controller out of reset */
+	reset_set_enable(PERIPH_ID_AFI, 0);
+
+	/* enable AFI clock */
+	clock_enable(PERIPH_ID_AFI);
+
+	if (soc->has_cml_clk) {
+		/* enable CML clock */
+		value = readl(NV_PA_CLK_RST_BASE + 0x48c);
+		value |= (1 << 0);
+		value &= ~(1 << 1);
+		writel(value, NV_PA_CLK_RST_BASE + 0x48c);
+	}
+
+	err = tegra_plle_enable();
+	if (err < 0) {
+		error("failed to enable PLLE: %d\n", err);
+		return err;
+	}
+
+	return 0;
+}
+
+static int tegra_pcie_pll_wait(struct tegra_pcie *pcie, unsigned long timeout)
+{
+	const struct tegra_pcie_soc *soc = pcie->soc;
+	unsigned long start = get_timer(0);
+	u32 value;
+
+	while (get_timer(start) < timeout) {
+		value = pads_readl(pcie, soc->pads_pll_ctl);
+		if (value & PADS_PLL_CTL_LOCKDET)
+			return 0;
+	}
+
+	return -ETIMEDOUT;
+}
+
+static int tegra_pcie_phy_enable(struct tegra_pcie *pcie)
+{
+	const struct tegra_pcie_soc *soc = pcie->soc;
+	u32 value;
+	int err;
+
+	/* initialize internal PHY, enable up to 16 PCIe lanes */
+	pads_writel(pcie, 0, PADS_CTL_SEL);
+
+	/* override IDDQ to 1 on all 4 lanes */
+	value = pads_readl(pcie, PADS_CTL);
+	value |= PADS_CTL_IDDQ_1L;
+	pads_writel(pcie, value, PADS_CTL);
+
+	/*
+	 * Set up PHY PLL inputs select PLLE output as refclock, set TX
+	 * ref sel to div10 (not div5).
+	 */
+	value = pads_readl(pcie, soc->pads_pll_ctl);
+	value &= ~(PADS_PLL_CTL_REFCLK_MASK | PADS_PLL_CTL_TXCLKREF_MASK);
+	value |= PADS_PLL_CTL_REFCLK_INTERNAL_CML | soc->tx_ref_sel;
+	pads_writel(pcie, value, soc->pads_pll_ctl);
+
+	/* reset PLL */
+	value = pads_readl(pcie, soc->pads_pll_ctl);
+	value &= ~PADS_PLL_CTL_RST_B4SM;
+	pads_writel(pcie, value, soc->pads_pll_ctl);
+
+	udelay(20);
+
+	/* take PLL out of reset */
+	value = pads_readl(pcie, soc->pads_pll_ctl);
+	value |= PADS_PLL_CTL_RST_B4SM;
+	pads_writel(pcie, value, soc->pads_pll_ctl);
+
+	/* configure the reference clock driver */
+	value = PADS_REFCLK_CFG_VALUE | (PADS_REFCLK_CFG_VALUE << 16);
+	pads_writel(pcie, value, PADS_REFCLK_CFG0);
+
+	if (soc->num_ports > 2)
+		pads_writel(pcie, PADS_REFCLK_CFG_VALUE, PADS_REFCLK_CFG1);
+
+	/* wait for the PLL to lock */
+	err = tegra_pcie_pll_wait(pcie, 500);
+	if (err < 0) {
+		error("PLL failed to lock: %d", err);
+		return err;
+	}
+
+	/* turn off IDDQ override */
+	value = pads_readl(pcie, PADS_CTL);
+	value &= ~PADS_CTL_IDDQ_1L;
+	pads_writel(pcie, value, PADS_CTL);
+
+	/* enable TX/RX data */
+	value = pads_readl(pcie, PADS_CTL);
+	value |= PADS_CTL_TX_DATA_EN_1L | PADS_CTL_RX_DATA_EN_1L;
+	pads_writel(pcie, value, PADS_CTL);
+
+	return 0;
+}
+
+static int tegra_pcie_enable_controller(struct tegra_pcie *pcie)
+{
+	const struct tegra_pcie_soc *soc = pcie->soc;
+	struct tegra_pcie_port *port;
+	u32 value;
+	int err;
+
+	if (pcie->phy) {
+		value = afi_readl(pcie, AFI_PLLE_CONTROL);
+		value &= ~AFI_PLLE_CONTROL_BYPASS_PADS2PLLE_CONTROL;
+		value |= AFI_PLLE_CONTROL_PADS2PLLE_CONTROL_EN;
+		afi_writel(pcie, value, AFI_PLLE_CONTROL);
+	}
+
+	if (soc->has_pex_bias_ctrl)
+		afi_writel(pcie, 0, AFI_PEXBIAS_CTRL_0);
+
+	value = afi_readl(pcie, AFI_PCIE_CONFIG);
+	value &= ~AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_MASK;
+	value |= AFI_PCIE_CONFIG_PCIE_DISABLE_ALL | pcie->xbar;
+
+	list_for_each_entry(port, &pcie->ports, list)
+		value &= ~AFI_PCIE_CONFIG_PCIE_DISABLE(port->index);
+
+	afi_writel(pcie, value, AFI_PCIE_CONFIG);
+
+	value = afi_readl(pcie, AFI_FUSE);
+
+	if (soc->has_gen2)
+		value &= ~AFI_FUSE_PCIE_T0_GEN2_DIS;
+	else
+		value |= AFI_FUSE_PCIE_T0_GEN2_DIS;
+
+	afi_writel(pcie, value, AFI_FUSE);
+
+	if (pcie->phy)
+		err = tegra_xusb_phy_enable(pcie->phy);
+	else
+		err = tegra_pcie_phy_enable(pcie);
+
+	if (err < 0) {
+		error("failed to power on PHY: %d\n", err);
+		return err;
+	}
+
+	/* take the PCIEXCLK logic out of reset */
+	reset_set_enable(PERIPH_ID_PCIEXCLK, 0);
+
+	/* finally enable PCIe */
+	value = afi_readl(pcie, AFI_CONFIGURATION);
+	value |= AFI_CONFIGURATION_EN_FPCI;
+	afi_writel(pcie, value, AFI_CONFIGURATION);
+
+	/* disable all interrupts */
+	afi_writel(pcie, 0, AFI_AFI_INTR_ENABLE);
+	afi_writel(pcie, 0, AFI_SM_INTR_ENABLE);
+	afi_writel(pcie, 0, AFI_INTR_MASK);
+	afi_writel(pcie, 0, AFI_FPCI_ERROR_MASKS);
+
+	return 0;
+}
+
+static void tegra_pcie_setup_translations(struct tegra_pcie *pcie)
+{
+	unsigned long fpci, axi, size;
+
+	/* BAR 0: type 1 extended configuration space */
+	fpci = 0xfe100000;
+	size = fdt_resource_size(&pcie->cs);
+	axi = pcie->cs.start;
+
+	afi_writel(pcie, axi, AFI_AXI_BAR0_START);
+	afi_writel(pcie, size >> 12, AFI_AXI_BAR0_SZ);
+	afi_writel(pcie, fpci, AFI_FPCI_BAR0);
+
+	/* BAR 1: downstream I/O */
+	fpci = 0xfdfc0000;
+	size = fdt_resource_size(&pcie->io);
+	axi = pcie->io.start;
+
+	afi_writel(pcie, axi, AFI_AXI_BAR1_START);
+	afi_writel(pcie, size >> 12, AFI_AXI_BAR1_SZ);
+	afi_writel(pcie, fpci, AFI_FPCI_BAR1);
+
+	/* BAR 2: prefetchable memory */
+	fpci = (((pcie->prefetch.start >> 12) & 0x0fffffff) << 4) | 0x1;
+	size = fdt_resource_size(&pcie->prefetch);
+	axi = pcie->prefetch.start;
+
+	afi_writel(pcie, axi, AFI_AXI_BAR2_START);
+	afi_writel(pcie, size >> 12, AFI_AXI_BAR2_SZ);
+	afi_writel(pcie, fpci, AFI_FPCI_BAR2);
+
+	/* BAR 3: non-prefetchable memory */
+	fpci = (((pcie->mem.start >> 12) & 0x0fffffff) << 4) | 0x1;
+	size = fdt_resource_size(&pcie->mem);
+	axi = pcie->mem.start;
+
+	afi_writel(pcie, axi, AFI_AXI_BAR3_START);
+	afi_writel(pcie, size >> 12, AFI_AXI_BAR3_SZ);
+	afi_writel(pcie, fpci, AFI_FPCI_BAR3);
+
+	/* NULL out the remaining BARs as they are not used */
+	afi_writel(pcie, 0, AFI_AXI_BAR4_START);
+	afi_writel(pcie, 0, AFI_AXI_BAR4_SZ);
+	afi_writel(pcie, 0, AFI_FPCI_BAR4);
+
+	afi_writel(pcie, 0, AFI_AXI_BAR5_START);
+	afi_writel(pcie, 0, AFI_AXI_BAR5_SZ);
+	afi_writel(pcie, 0, AFI_FPCI_BAR5);
+
+	/* map all upstream transactions as uncached */
+	afi_writel(pcie, NV_PA_SDRAM_BASE, AFI_CACHE_BAR0_ST);
+	afi_writel(pcie, 0, AFI_CACHE_BAR0_SZ);
+	afi_writel(pcie, 0, AFI_CACHE_BAR1_ST);
+	afi_writel(pcie, 0, AFI_CACHE_BAR1_SZ);
+
+	/* MSI translations are setup only when needed */
+	afi_writel(pcie, 0, AFI_MSI_FPCI_BAR_ST);
+	afi_writel(pcie, 0, AFI_MSI_BAR_SZ);
+	afi_writel(pcie, 0, AFI_MSI_AXI_BAR_ST);
+	afi_writel(pcie, 0, AFI_MSI_BAR_SZ);
+}
+
+static unsigned long tegra_pcie_port_get_pex_ctrl(struct tegra_pcie_port *port)
+{
+	unsigned long ret = 0;
+
+	switch (port->index) {
+	case 0:
+		ret = AFI_PEX0_CTRL;
+		break;
+
+	case 1:
+		ret = AFI_PEX1_CTRL;
+		break;
+
+	case 2:
+		ret = AFI_PEX2_CTRL;
+		break;
+	}
+
+	return ret;
+}
+
+static void tegra_pcie_port_reset(struct tegra_pcie_port *port)
+{
+	unsigned long ctrl = tegra_pcie_port_get_pex_ctrl(port);
+	unsigned long value;
+
+	/* pulse reset signel */
+	value = afi_readl(port->pcie, ctrl);
+	value &= ~AFI_PEX_CTRL_RST;
+	afi_writel(port->pcie, value, ctrl);
+
+	udelay(2000);
+
+	value = afi_readl(port->pcie, ctrl);
+	value |= AFI_PEX_CTRL_RST;
+	afi_writel(port->pcie, value, ctrl);
+}
+
+static void tegra_pcie_port_enable(struct tegra_pcie_port *port)
+{
+	unsigned long ctrl = tegra_pcie_port_get_pex_ctrl(port);
+	unsigned long value;
+
+	/* enable reference clock */
+	value = afi_readl(port->pcie, ctrl);
+	value |= AFI_PEX_CTRL_REFCLK_EN;
+
+	if (port->pcie->soc->has_pex_clkreq_en)
+		value |= AFI_PEX_CTRL_CLKREQ_EN;
+
+	value |= AFI_PEX_CTRL_OVERRIDE_EN;
+
+	afi_writel(port->pcie, value, ctrl);
+
+	tegra_pcie_port_reset(port);
+}
+
+static bool tegra_pcie_port_check_link(struct tegra_pcie_port *port)
+{
+	unsigned int retries = 3;
+	unsigned long value;
+
+	value = rp_readl(port, RP_PRIV_MISC);
+	value &= ~RP_PRIV_MISC_PRSNT_MAP_EP_ABSNT;
+	value |= RP_PRIV_MISC_PRSNT_MAP_EP_PRSNT;
+	rp_writel(port, value, RP_PRIV_MISC);
+
+	do {
+		unsigned int timeout = 200;
+
+		do {
+			value = rp_readl(port, RP_VEND_XP);
+			if (value & RP_VEND_XP_DL_UP)
+				break;
+
+			udelay(2000);
+		} while (--timeout);
+
+		if (!timeout) {
+			debug("link %u down, retrying\n", port->index);
+			goto retry;
+		}
+
+		timeout = 200;
+
+		do {
+			value = rp_readl(port, RP_LINK_CONTROL_STATUS);
+			if (value & RP_LINK_CONTROL_STATUS_DL_LINK_ACTIVE)
+				return true;
+
+			udelay(2000);
+		} while (--timeout);
+
+retry:
+		tegra_pcie_port_reset(port);
+	} while (--retries);
+
+	return false;
+}
+
+static void tegra_pcie_port_disable(struct tegra_pcie_port *port)
+{
+	unsigned long ctrl = tegra_pcie_port_get_pex_ctrl(port);
+	unsigned long value;
+
+	/* assert port reset */
+	value = afi_readl(port->pcie, ctrl);
+	value &= ~AFI_PEX_CTRL_RST;
+	afi_writel(port->pcie, value, ctrl);
+
+	/* disable reference clock */
+	value = afi_readl(port->pcie, ctrl);
+	value &= ~AFI_PEX_CTRL_REFCLK_EN;
+	afi_writel(port->pcie, value, ctrl);
+}
+
+static void tegra_pcie_port_free(struct tegra_pcie_port *port)
+{
+	list_del(&port->list);
+	free(port);
+}
+
+static int tegra_pcie_enable(struct tegra_pcie *pcie)
+{
+	struct tegra_pcie_port *port, *tmp;
+
+	list_for_each_entry_safe(port, tmp, &pcie->ports, list) {
+		debug("probing port %u, using %u lanes\n", port->index,
+		      port->num_lanes);
+
+		tegra_pcie_port_enable(port);
+
+		if (tegra_pcie_port_check_link(port))
+			continue;
+
+		debug("link %u down, ignoring\n", port->index);
+
+		tegra_pcie_port_disable(port);
+		tegra_pcie_port_free(port);
+	}
+
+	return 0;
+}
+
+static const struct tegra_pcie_soc tegra20_pcie_soc = {
+	.num_ports = 2,
+	.pads_pll_ctl = PADS_PLL_CTL_TEGRA20,
+	.tx_ref_sel = PADS_PLL_CTL_TXCLKREF_DIV10,
+	.has_pex_clkreq_en = false,
+	.has_pex_bias_ctrl = false,
+	.has_cml_clk = false,
+	.has_gen2 = false,
+};
+
+static const struct tegra_pcie_soc tegra30_pcie_soc = {
+	.num_ports = 3,
+	.pads_pll_ctl = PADS_PLL_CTL_TEGRA30,
+	.tx_ref_sel = PADS_PLL_CTL_TXCLKREF_BUF_EN,
+	.has_pex_clkreq_en = true,
+	.has_pex_bias_ctrl = true,
+	.has_cml_clk = true,
+	.has_gen2 = false,
+};
+
+static const struct tegra_pcie_soc tegra124_pcie_soc = {
+	.num_ports = 2,
+	.pads_pll_ctl = PADS_PLL_CTL_TEGRA30,
+	.tx_ref_sel = PADS_PLL_CTL_TXCLKREF_BUF_EN,
+	.has_pex_clkreq_en = true,
+	.has_pex_bias_ctrl = true,
+	.has_cml_clk = true,
+	.has_gen2 = true,
+};
+
+static int process_nodes(const void *fdt, int nodes[], unsigned int count)
+{
+	unsigned int i;
+
+	for (i = 0; i < count; i++) {
+		const struct tegra_pcie_soc *soc;
+		struct tegra_pcie *pcie;
+		enum fdt_compat_id id;
+		int err;
+
+		if (!fdtdec_get_is_enabled(fdt, nodes[i]))
+			continue;
+
+		id = fdtdec_lookup(fdt, nodes[i]);
+		switch (id) {
+		case COMPAT_NVIDIA_TEGRA20_PCIE:
+			soc = &tegra20_pcie_soc;
+			break;
+
+		case COMPAT_NVIDIA_TEGRA30_PCIE:
+			soc = &tegra30_pcie_soc;
+			break;
+
+		case COMPAT_NVIDIA_TEGRA124_PCIE:
+			soc = &tegra124_pcie_soc;
+			break;
+
+		default:
+			error("unsupported compatible: %s",
+			      fdtdec_get_compatible(id));
+			continue;
+		}
+
+		pcie = malloc(sizeof(*pcie));
+		if (!pcie) {
+			error("failed to allocate controller");
+			continue;
+		}
+
+		memset(pcie, 0, sizeof(*pcie));
+		pcie->soc = soc;
+
+		INIT_LIST_HEAD(&pcie->ports);
+
+		err = tegra_pcie_parse_dt(fdt, nodes[i], pcie);
+		if (err < 0) {
+			free(pcie);
+			continue;
+		}
+
+		err = tegra_pcie_power_on(pcie);
+		if (err < 0) {
+			error("failed to power on");
+			continue;
+		}
+
+		err = tegra_pcie_enable_controller(pcie);
+		if (err < 0) {
+			error("failed to enable controller");
+			continue;
+		}
+
+		tegra_pcie_setup_translations(pcie);
+
+		err = tegra_pcie_enable(pcie);
+		if (err < 0) {
+			error("failed to enable PCIe");
+			continue;
+		}
+
+		pcie->hose.first_busno = 0;
+		pcie->hose.current_busno = 0;
+		pcie->hose.last_busno = 0;
+
+		pci_set_region(&pcie->hose.regions[0], NV_PA_SDRAM_BASE,
+			       NV_PA_SDRAM_BASE, gd->ram_size,
+			       PCI_REGION_MEM | PCI_REGION_SYS_MEMORY);
+
+		pci_set_region(&pcie->hose.regions[1], pcie->io.start,
+			       pcie->io.start, fdt_resource_size(&pcie->io),
+			       PCI_REGION_IO);
+
+		pci_set_region(&pcie->hose.regions[2], pcie->mem.start,
+			       pcie->mem.start, fdt_resource_size(&pcie->mem),
+			       PCI_REGION_MEM);
+
+		pci_set_region(&pcie->hose.regions[3], pcie->prefetch.start,
+			       pcie->prefetch.start,
+			       fdt_resource_size(&pcie->prefetch),
+			       PCI_REGION_MEM | PCI_REGION_PREFETCH);
+
+		pcie->hose.region_count = 4;
+
+		pci_set_ops(&pcie->hose,
+			    pci_hose_read_config_byte_via_dword,
+			    pci_hose_read_config_word_via_dword,
+			    tegra_pcie_read_conf,
+			    pci_hose_write_config_byte_via_dword,
+			    pci_hose_write_config_word_via_dword,
+			    tegra_pcie_write_conf);
+
+		pci_register_hose(&pcie->hose);
+
+#ifdef CONFIG_PCI_SCAN_SHOW
+		printf("PCI: Enumerating devices...\n");
+		printf("---------------------------------------\n");
+		printf("  Device        ID          Description\n");
+		printf("  ------        --          -----------\n");
+#endif
+
+		pcie->hose.last_busno = pci_hose_scan(&pcie->hose);
+	}
+
+	return 0;
+}
+
+void pci_init_board(void)
+{
+	const void *fdt = gd->fdt_blob;
+	int count, nodes[1];
+
+	count = fdtdec_find_aliases_for_id(fdt, "pcie-controller",
+					   COMPAT_NVIDIA_TEGRA124_PCIE,
+					   nodes, ARRAY_SIZE(nodes));
+	if (process_nodes(fdt, nodes, count))
+		return;
+
+	count = fdtdec_find_aliases_for_id(fdt, "pcie-controller",
+					   COMPAT_NVIDIA_TEGRA30_PCIE,
+					   nodes, ARRAY_SIZE(nodes));
+	if (process_nodes(fdt, nodes, count))
+		return;
+
+	count = fdtdec_find_aliases_for_id(fdt, "pcie-controller",
+					   COMPAT_NVIDIA_TEGRA20_PCIE,
+					   nodes, ARRAY_SIZE(nodes));
+	if (process_nodes(fdt, nodes, count))
+		return;
+}
+
+int pci_skip_dev(struct pci_controller *hose, pci_dev_t dev)
+{
+	if (PCI_BUS(dev) != 0 && PCI_DEV(dev) > 0)
+		return 1;
+
+	return 0;
+}
diff --git a/include/fdtdec.h b/include/fdtdec.h
index c0464f07961d..2869cd0ae206 100644
--- a/include/fdtdec.h
+++ b/include/fdtdec.h
@@ -82,6 +82,9 @@  enum fdt_compat_id {
 	COMPAT_NVIDIA_TEGRA20_SFLASH,	/* Tegra 2 SPI flash controller */
 	COMPAT_NVIDIA_TEGRA20_SLINK,	/* Tegra 2 SPI SLINK controller */
 	COMPAT_NVIDIA_TEGRA114_SPI,	/* Tegra 114 SPI controller */
+	COMPAT_NVIDIA_TEGRA124_PCIE,	/* Tegra 124 PCIe controller */
+	COMPAT_NVIDIA_TEGRA30_PCIE,	/* Tegra 30 PCIe controller */
+	COMPAT_NVIDIA_TEGRA20_PCIE,	/* Tegra 20 PCIe controller */
 	COMPAT_NVIDIA_TEGRA124_XUSB_PADCTL,
 					/* Tegra124 XUSB pad controller */
 	COMPAT_SMSC_LAN9215,		/* SMSC 10/100 Ethernet LAN9215 */
diff --git a/lib/fdtdec.c b/lib/fdtdec.c
index 16e0714b0fed..3e41b0c99683 100644
--- a/lib/fdtdec.c
+++ b/lib/fdtdec.c
@@ -41,6 +41,9 @@  static const char * const compat_names[COMPAT_COUNT] = {
 	COMPAT(NVIDIA_TEGRA20_SFLASH, "nvidia,tegra20-sflash"),
 	COMPAT(NVIDIA_TEGRA20_SLINK, "nvidia,tegra20-slink"),
 	COMPAT(NVIDIA_TEGRA114_SPI, "nvidia,tegra114-spi"),
+	COMPAT(NVIDIA_TEGRA124_PCIE, "nvidia,tegra124-pcie"),
+	COMPAT(NVIDIA_TEGRA30_PCIE, "nvidia,tegra30-pcie"),
+	COMPAT(NVIDIA_TEGRA20_PCIE, "nvidia,tegra20-pcie"),
 	COMPAT(NVIDIA_TEGRA124_XUSB_PADCTL, "nvidia,tegra124-xusb-padctl"),
 	COMPAT(SMSC_LAN9215, "smsc,lan9215"),
 	COMPAT(SAMSUNG_EXYNOS5_SROMC, "samsung,exynos-sromc"),