Merge commit 'v3.3-rc1' into stable/for-linus-fixes-3.3

* commit 'v3.3-rc1': (9775 commits)
  Linux 3.3-rc1
  x86, syscall: Need __ARCH_WANT_SYS_IPC for 32 bits
  qnx4: don't leak ->BitMap on late failure exits
  qnx4: reduce the insane nesting in qnx4_checkroot()
  qnx4: di_fname is an array, for crying out loud...
  KEYS: Permit key_serial() to be called with a const key pointer
  keys: fix user_defined key sparse messages
  ima: fix cred sparse warning
  uml: fix compile for x86-64
  MPILIB: Add a missing ENOMEM check
  tpm: fix (ACPI S3) suspend regression
  nvme: fix merge error due to change of 'make_request_fn' fn type
  xen: using EXPORT_SYMBOL requires including export.h
  gpio: tps65910: Use correct offset for gpio initialization
  acpi/apei/einj: Add extensions to EINJ from rev 5.0 of acpi spec
  intel_idle: Split up and provide per CPU initialization func
  ACPI processor: Remove unneeded variable passed by acpi_processor_hotadd_init V2
  tg3: Fix single-vector MSI-X code
  openvswitch: Fix multipart datapath dumps.
  ipv6: fix per device IP snmp counters
  ...

14 files changed, 1446 insertions, 445 deletions

diff --git a/arch/mips/pci/Makefile b/arch/mips/pci/Makefile
index bb82cbdbc62..c3ac4b086eb 100644
--- a/arch/mips/pci/Makefile
+++ b/arch/mips/pci/Makefile
@@ -19,6 +19,7 @@ obj-$(CONFIG_BCM47XX)		+= pci-bcm47xx.o
 obj-$(CONFIG_BCM63XX)		+= pci-bcm63xx.o fixup-bcm63xx.o \
 					ops-bcm63xx.o
 obj-$(CONFIG_MIPS_ALCHEMY)	+= pci-alchemy.o
+obj-$(CONFIG_SOC_AR724X)	+= pci-ath724x.o
 
 #
 # These are still pretty much in the old state, watch, go blind.
@@ -55,7 +56,7 @@ obj-$(CONFIG_ZAO_CAPCELLA)	+= fixup-capcella.o
 obj-$(CONFIG_WR_PPMC)		+= fixup-wrppmc.o
 obj-$(CONFIG_MIKROTIK_RB532)	+= pci-rc32434.o ops-rc32434.o fixup-rc32434.o
 obj-$(CONFIG_CPU_CAVIUM_OCTEON)	+= pci-octeon.o pcie-octeon.o
-obj-$(CONFIG_NLM_XLR)		+= pci-xlr.o
+obj-$(CONFIG_CPU_XLR)		+= pci-xlr.o
 
 ifdef CONFIG_PCI_MSI
 obj-$(CONFIG_CPU_CAVIUM_OCTEON)	+= msi-octeon.o
diff --git a/arch/mips/pci/msi-octeon.c b/arch/mips/pci/msi-octeon.c
index 5d530f89d87..d37be36dc65 100644
--- a/arch/mips/pci/msi-octeon.c
+++ b/arch/mips/pci/msi-octeon.c
@@ -162,7 +162,7 @@ msi_irq_allocated:
 		msg.address_hi = (0 + CVMX_NPEI_PCIE_MSI_RCV) >> 32;
 		break;
 	default:
-		panic("arch_setup_msi_irq: Invalid octeon_dma_bar_type\n");
+		panic("arch_setup_msi_irq: Invalid octeon_dma_bar_type");
 	}
 	msg.data = irq - OCTEON_IRQ_MSI_BIT0;
 
diff --git a/arch/mips/pci/ops-pmcmsp.c b/arch/mips/pci/ops-pmcmsp.c
index 8fbfbf2b931..389bf669d56 100644
--- a/arch/mips/pci/ops-pmcmsp.c
+++ b/arch/mips/pci/ops-pmcmsp.c
@@ -405,7 +405,7 @@ int msp_pcibios_config_access(unsigned char access_type,
 	if (pciirqflag == 0) {
 		ret = request_irq(MSP_INT_PCI,/* Hardcoded internal MSP7120 wiring */
 				bpci_interrupt,
-				IRQF_SHARED | IRQF_DISABLED,
+				IRQF_SHARED,
 				"PMC MSP PCI Host",
 				preg);
 		if (ret != 0)
diff --git a/arch/mips/pci/ops-tx3927.c b/arch/mips/pci/ops-tx3927.c
index 6a3bdb5ffa8..02d64f77e96 100644
--- a/arch/mips/pci/ops-tx3927.c
+++ b/arch/mips/pci/ops-tx3927.c
@@ -225,7 +225,7 @@ void __init tx3927_setup_pcierr_irq(void)
 {
 	if (request_irq(TXX9_IRQ_BASE + TX3927_IR_PCI,
 			tx3927_pcierr_interrupt,
-			IRQF_DISABLED, "PCI error",
+			0, "PCI error",
 			(void *)TX3927_PCIC_REG))
 		printk(KERN_WARNING "Failed to request irq for PCIERR\n");
 }
diff --git a/arch/mips/pci/pci-alchemy.c b/arch/mips/pci/pci-alchemy.c
index b5ce041cdaf..ec125bed721 100644
--- a/arch/mips/pci/pci-alchemy.c
+++ b/arch/mips/pci/pci-alchemy.c
@@ -13,9 +13,11 @@
 #include <linux/platform_device.h>
 #include <linux/kernel.h>
 #include <linux/init.h>
+#include <linux/syscore_ops.h>
 #include <linux/vmalloc.h>
 
 #include <asm/mach-au1x00/au1000.h>
+#include <asm/tlbmisc.h>
 
 #ifdef CONFIG_DEBUG_PCI
 #define DBG(x...) printk(KERN_DEBUG x)
@@ -41,6 +43,12 @@ struct alchemy_pci_context {
 	int (*board_pci_idsel)(unsigned int devsel, int assert);
 };
 
+/* for syscore_ops. There's only one PCI controller on Alchemy chips, so this
+ * should suffice for now.
+ */
+static struct alchemy_pci_context *__alchemy_pci_ctx;
+
+
 /* IO/MEM resources for PCI. Keep the memres in sync with __fixup_bigphys_addr
  * in arch/mips/alchemy/common/setup.c
  */
@@ -99,18 +107,6 @@ static int config_access(unsigned char access_type, struct pci_bus *bus,
 		return -1;
 	}
 
-	/* YAMON on all db1xxx boards wipes the TLB and writes zero to C0_wired
-	 * on resume, clearing our wired entry.  Unfortunately the ->resume()
-	 * callback is called way way way too late (and ->suspend() too early)
-	 * to have them destroy and recreate it.  Instead just test if c0_wired
-	 * is now lower than the index we retrieved before suspending and then
-	 * recreate the entry if necessary.  Of course this is totally bonkers
-	 * and breaks as soon as someone else adds another wired entry somewhere
-	 * else.  Anyone have any ideas how to handle this better?
-	 */
-	if (unlikely(read_c0_wired() < ctx->wired_entry))
-		alchemy_pci_wired_entry(ctx);
-
 	local_irq_save(flags);
 	r = __raw_readl(ctx->regs + PCI_REG_STATCMD) & 0x0000ffff;
 	r |= PCI_STATCMD_STATUS(0x2000);
@@ -304,6 +300,62 @@ static int alchemy_pci_def_idsel(unsigned int devsel, int assert)
 	return 1;	/* success */
 }
 
+/* save PCI controller register contents. */
+static int alchemy_pci_suspend(void)
+{
+	struct alchemy_pci_context *ctx = __alchemy_pci_ctx;
+	if (!ctx)
+		return 0;
+
+	ctx->pm[0]  = __raw_readl(ctx->regs + PCI_REG_CMEM);
+	ctx->pm[1]  = __raw_readl(ctx->regs + PCI_REG_CONFIG) & 0x0009ffff;
+	ctx->pm[2]  = __raw_readl(ctx->regs + PCI_REG_B2BMASK_CCH);
+	ctx->pm[3]  = __raw_readl(ctx->regs + PCI_REG_B2BBASE0_VID);
+	ctx->pm[4]  = __raw_readl(ctx->regs + PCI_REG_B2BBASE1_SID);
+	ctx->pm[5]  = __raw_readl(ctx->regs + PCI_REG_MWMASK_DEV);
+	ctx->pm[6]  = __raw_readl(ctx->regs + PCI_REG_MWBASE_REV_CCL);
+	ctx->pm[7]  = __raw_readl(ctx->regs + PCI_REG_ID);
+	ctx->pm[8]  = __raw_readl(ctx->regs + PCI_REG_CLASSREV);
+	ctx->pm[9]  = __raw_readl(ctx->regs + PCI_REG_PARAM);
+	ctx->pm[10] = __raw_readl(ctx->regs + PCI_REG_MBAR);
+	ctx->pm[11] = __raw_readl(ctx->regs + PCI_REG_TIMEOUT);
+
+	return 0;
+}
+
+static void alchemy_pci_resume(void)
+{
+	struct alchemy_pci_context *ctx = __alchemy_pci_ctx;
+	if (!ctx)
+		return;
+
+	__raw_writel(ctx->pm[0],  ctx->regs + PCI_REG_CMEM);
+	__raw_writel(ctx->pm[2],  ctx->regs + PCI_REG_B2BMASK_CCH);
+	__raw_writel(ctx->pm[3],  ctx->regs + PCI_REG_B2BBASE0_VID);
+	__raw_writel(ctx->pm[4],  ctx->regs + PCI_REG_B2BBASE1_SID);
+	__raw_writel(ctx->pm[5],  ctx->regs + PCI_REG_MWMASK_DEV);
+	__raw_writel(ctx->pm[6],  ctx->regs + PCI_REG_MWBASE_REV_CCL);
+	__raw_writel(ctx->pm[7],  ctx->regs + PCI_REG_ID);
+	__raw_writel(ctx->pm[8],  ctx->regs + PCI_REG_CLASSREV);
+	__raw_writel(ctx->pm[9],  ctx->regs + PCI_REG_PARAM);
+	__raw_writel(ctx->pm[10], ctx->regs + PCI_REG_MBAR);
+	__raw_writel(ctx->pm[11], ctx->regs + PCI_REG_TIMEOUT);
+	wmb();
+	__raw_writel(ctx->pm[1],  ctx->regs + PCI_REG_CONFIG);
+	wmb();
+
+	/* YAMON on all db1xxx boards wipes the TLB and writes zero to C0_wired
+	 * on resume, making it necessary to recreate it as soon as possible.
+	 */
+	ctx->wired_entry = 8191;	/* impossibly high value */
+	alchemy_pci_wired_entry(ctx);	/* install it */
+}
+
+static struct syscore_ops alchemy_pci_pmops = {
+	.suspend	= alchemy_pci_suspend,
+	.resume		= alchemy_pci_resume,
+};
+
 static int __devinit alchemy_pci_probe(struct platform_device *pdev)
 {
 	struct alchemy_pci_platdata *pd = pdev->dev.platform_data;
@@ -396,7 +448,8 @@ static int __devinit alchemy_pci_probe(struct platform_device *pdev)
 		ret = -ENOMEM;
 		goto out4;
 	}
-	ctx->wired_entry = 8192;	/* impossibly high value */
+	ctx->wired_entry = 8191;	/* impossibly high value */
+	alchemy_pci_wired_entry(ctx);	/* install it */
 
 	set_io_port_base((unsigned long)ctx->alchemy_pci_ctrl.io_map_base);
 
@@ -408,7 +461,9 @@ static int __devinit alchemy_pci_probe(struct platform_device *pdev)
 	__raw_writel(val, ctx->regs + PCI_REG_CONFIG);
 	wmb();
 
+	__alchemy_pci_ctx = ctx;
 	platform_set_drvdata(pdev, ctx);
+	register_syscore_ops(&alchemy_pci_pmops);
 	register_pci_controller(&ctx->alchemy_pci_ctrl);
 
 	return 0;
@@ -425,68 +480,11 @@ out:
 	return ret;
 }
 
-
-#ifdef CONFIG_PM
-/* save PCI controller register contents. */
-static int alchemy_pci_suspend(struct device *dev)
-{
-	struct alchemy_pci_context *ctx = dev_get_drvdata(dev);
-
-	ctx->pm[0]  = __raw_readl(ctx->regs + PCI_REG_CMEM);
-	ctx->pm[1]  = __raw_readl(ctx->regs + PCI_REG_CONFIG) & 0x0009ffff;
-	ctx->pm[2]  = __raw_readl(ctx->regs + PCI_REG_B2BMASK_CCH);
-	ctx->pm[3]  = __raw_readl(ctx->regs + PCI_REG_B2BBASE0_VID);
-	ctx->pm[4]  = __raw_readl(ctx->regs + PCI_REG_B2BBASE1_SID);
-	ctx->pm[5]  = __raw_readl(ctx->regs + PCI_REG_MWMASK_DEV);
-	ctx->pm[6]  = __raw_readl(ctx->regs + PCI_REG_MWBASE_REV_CCL);
-	ctx->pm[7]  = __raw_readl(ctx->regs + PCI_REG_ID);
-	ctx->pm[8]  = __raw_readl(ctx->regs + PCI_REG_CLASSREV);
-	ctx->pm[9]  = __raw_readl(ctx->regs + PCI_REG_PARAM);
-	ctx->pm[10] = __raw_readl(ctx->regs + PCI_REG_MBAR);
-	ctx->pm[11] = __raw_readl(ctx->regs + PCI_REG_TIMEOUT);
-
-	return 0;
-}
-
-static int alchemy_pci_resume(struct device *dev)
-{
-	struct alchemy_pci_context *ctx = dev_get_drvdata(dev);
-
-	__raw_writel(ctx->pm[0],  ctx->regs + PCI_REG_CMEM);
-	__raw_writel(ctx->pm[2],  ctx->regs + PCI_REG_B2BMASK_CCH);
-	__raw_writel(ctx->pm[3],  ctx->regs + PCI_REG_B2BBASE0_VID);
-	__raw_writel(ctx->pm[4],  ctx->regs + PCI_REG_B2BBASE1_SID);
-	__raw_writel(ctx->pm[5],  ctx->regs + PCI_REG_MWMASK_DEV);
-	__raw_writel(ctx->pm[6],  ctx->regs + PCI_REG_MWBASE_REV_CCL);
-	__raw_writel(ctx->pm[7],  ctx->regs + PCI_REG_ID);
-	__raw_writel(ctx->pm[8],  ctx->regs + PCI_REG_CLASSREV);
-	__raw_writel(ctx->pm[9],  ctx->regs + PCI_REG_PARAM);
-	__raw_writel(ctx->pm[10], ctx->regs + PCI_REG_MBAR);
-	__raw_writel(ctx->pm[11], ctx->regs + PCI_REG_TIMEOUT);
-	wmb();
-	__raw_writel(ctx->pm[1],  ctx->regs + PCI_REG_CONFIG);
-	wmb();
-
-	return 0;
-}
-
-static const struct dev_pm_ops alchemy_pci_pmops = {
-	.suspend	= alchemy_pci_suspend,
-	.resume		= alchemy_pci_resume,
-};
-
-#define ALCHEMY_PCICTL_PM	(&alchemy_pci_pmops)
-
-#else
-#define ALCHEMY_PCICTL_PM	NULL
-#endif
-
 static struct platform_driver alchemy_pcictl_driver = {
 	.probe		= alchemy_pci_probe,
 	.driver	= {
 		.name	= "alchemy-pci",
 		.owner	= THIS_MODULE,
-		.pm	= ALCHEMY_PCICTL_PM,
 	},
 };
 
diff --git a/arch/mips/pci/pci-ath724x.c b/arch/mips/pci/pci-ath724x.c
new file mode 100644
index 00000000000..a4dd24a4130
--- /dev/null
+++ b/arch/mips/pci/pci-ath724x.c
@@ -0,0 +1,174 @@
+/*
+ *  Atheros 724x PCI support
+ *
+ *  Copyright (C) 2011 René Bolldorf <xsecute@googlemail.com>
+ *
+ *  This program is free software; you can redistribute it and/or modify it
+ *  under the terms of the GNU General Public License version 2 as published
+ *  by the Free Software Foundation.
+ */
+
+#include <linux/pci.h>
+#include <asm/mach-ath79/pci-ath724x.h>
+
+#define reg_read(_phys)		(*(unsigned int *) KSEG1ADDR(_phys))
+#define reg_write(_phys, _val)	((*(unsigned int *) KSEG1ADDR(_phys)) = (_val))
+
+#define ATH724X_PCI_DEV_BASE	0x14000000
+#define ATH724X_PCI_MEM_BASE	0x10000000
+#define ATH724X_PCI_MEM_SIZE	0x08000000
+
+static DEFINE_SPINLOCK(ath724x_pci_lock);
+static struct ath724x_pci_data *pci_data;
+static int pci_data_size;
+
+static int ath724x_pci_read(struct pci_bus *bus, unsigned int devfn, int where,
+			    int size, uint32_t *value)
+{
+	unsigned long flags, addr, tval, mask;
+
+	if (devfn)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+
+	if (where & (size - 1))
+		return PCIBIOS_BAD_REGISTER_NUMBER;
+
+	spin_lock_irqsave(&ath724x_pci_lock, flags);
+
+	switch (size) {
+	case 1:
+		addr = where & ~3;
+		mask = 0xff000000 >> ((where % 4) * 8);
+		tval = reg_read(ATH724X_PCI_DEV_BASE + addr);
+		tval = tval & ~mask;
+		*value = (tval >> ((4 - (where % 4))*8));
+		break;
+	case 2:
+		addr = where & ~3;
+		mask = 0xffff0000 >> ((where % 4)*8);
+		tval = reg_read(ATH724X_PCI_DEV_BASE + addr);
+		tval = tval & ~mask;
+		*value = (tval >> ((4 - (where % 4))*8));
+		break;
+	case 4:
+		*value = reg_read(ATH724X_PCI_DEV_BASE + where);
+		break;
+	default:
+		spin_unlock_irqrestore(&ath724x_pci_lock, flags);
+
+		return PCIBIOS_BAD_REGISTER_NUMBER;
+	}
+
+	spin_unlock_irqrestore(&ath724x_pci_lock, flags);
+
+	return PCIBIOS_SUCCESSFUL;
+}
+
+static int ath724x_pci_write(struct pci_bus *bus, unsigned int devfn, int where,
+			     int size, uint32_t value)
+{
+	unsigned long flags, tval, addr, mask;
+
+	if (devfn)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+
+	if (where & (size - 1))
+		return PCIBIOS_BAD_REGISTER_NUMBER;
+
+	spin_lock_irqsave(&ath724x_pci_lock, flags);
+
+	switch (size) {
+	case 1:
+		addr = (ATH724X_PCI_DEV_BASE + where) & ~3;
+		mask = 0xff000000 >> ((where % 4)*8);
+		tval = reg_read(addr);
+		tval = tval & ~mask;
+		tval |= (value << ((4 - (where % 4))*8)) & mask;
+		reg_write(addr, tval);
+		break;
+	case 2:
+		addr = (ATH724X_PCI_DEV_BASE + where) & ~3;
+		mask = 0xffff0000 >> ((where % 4)*8);
+		tval = reg_read(addr);
+		tval = tval & ~mask;
+		tval |= (value << ((4 - (where % 4))*8)) & mask;
+		reg_write(addr, tval);
+		break;
+	case 4:
+		reg_write((ATH724X_PCI_DEV_BASE + where), value);
+		break;
+	default:
+		spin_unlock_irqrestore(&ath724x_pci_lock, flags);
+
+		return PCIBIOS_BAD_REGISTER_NUMBER;
+	}
+
+	spin_unlock_irqrestore(&ath724x_pci_lock, flags);
+
+	return PCIBIOS_SUCCESSFUL;
+}
+
+static struct pci_ops ath724x_pci_ops = {
+	.read	= ath724x_pci_read,
+	.write	= ath724x_pci_write,
+};
+
+static struct resource ath724x_io_resource = {
+	.name   = "PCI IO space",
+	.start  = 0,
+	.end    = 0,
+	.flags  = IORESOURCE_IO,
+};
+
+static struct resource ath724x_mem_resource = {
+	.name   = "PCI memory space",
+	.start  = ATH724X_PCI_MEM_BASE,
+	.end    = ATH724X_PCI_MEM_BASE + ATH724X_PCI_MEM_SIZE - 1,
+	.flags  = IORESOURCE_MEM,
+};
+
+static struct pci_controller ath724x_pci_controller = {
+	.pci_ops        = &ath724x_pci_ops,
+	.io_resource    = &ath724x_io_resource,
+	.mem_resource	= &ath724x_mem_resource,
+};
+
+void ath724x_pci_add_data(struct ath724x_pci_data *data, int size)
+{
+	pci_data	= data;
+	pci_data_size	= size;
+}
+
+int __init pcibios_map_irq(const struct pci_dev *dev, uint8_t slot, uint8_t pin)
+{
+	unsigned int devfn = dev->devfn;
+	int irq = -1;
+
+	if (devfn > pci_data_size - 1)
+		return irq;
+
+	irq = pci_data[devfn].irq;
+
+	return irq;
+}
+
+int pcibios_plat_dev_init(struct pci_dev *dev)
+{
+	unsigned int devfn = dev->devfn;
+
+	if (devfn > pci_data_size - 1)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+
+	dev->dev.platform_data = pci_data[devfn].pdata;
+
+	return PCIBIOS_SUCCESSFUL;
+}
+
+static int __init ath724x_pcibios_init(void)
+{
+	register_pci_controller(&ath724x_pci_controller);
+
+	return PCIBIOS_SUCCESSFUL;
+}
+
+arch_initcall(ath724x_pcibios_init);
diff --git a/arch/mips/pci/pci-bcm63xx.c b/arch/mips/pci/pci-bcm63xx.c
index 82e0fde1dba..39eb7c417e2 100644
--- a/arch/mips/pci/pci-bcm63xx.c
+++ b/arch/mips/pci/pci-bcm63xx.c
@@ -99,7 +99,7 @@ static int __init bcm63xx_pci_init(void)
 	unsigned int mem_size;
 	u32 val;
 
-	if (!BCMCPU_IS_6348() && !BCMCPU_IS_6358())
+	if (!BCMCPU_IS_6348() && !BCMCPU_IS_6358() && !BCMCPU_IS_6368())
 		return -ENODEV;
 
 	if (!bcm63xx_pci_enabled)
@@ -159,7 +159,7 @@ static int __init bcm63xx_pci_init(void)
 	/* setup PCI to local bus access, used by PCI device to target
 	 * local RAM while bus mastering */
 	bcm63xx_int_cfg_writel(0, PCI_BASE_ADDRESS_3);
-	if (BCMCPU_IS_6358())
+	if (BCMCPU_IS_6358() || BCMCPU_IS_6368())
 		val = MPI_SP0_REMAP_ENABLE_MASK;
 	else
 		val = 0;
diff --git a/arch/mips/pci/pci-octeon.c b/arch/mips/pci/pci-octeon.c
index ed1c54284b8..52a1ba70b3b 100644
--- a/arch/mips/pci/pci-octeon.c
+++ b/arch/mips/pci/pci-octeon.c
@@ -99,7 +99,7 @@ int pcibios_plat_dev_init(struct pci_dev *dev)
 	 */
 	pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, 64 / 4);
 	/* Set latency timers for all devices */
-	pci_write_config_byte(dev, PCI_LATENCY_TIMER, 48);
+	pci_write_config_byte(dev, PCI_LATENCY_TIMER, 64);
 
 	/* Enable reporting System errors and parity errors on all devices */
 	/* Enable parity checking and error reporting */
@@ -109,7 +109,7 @@ int pcibios_plat_dev_init(struct pci_dev *dev)
 
 	if (dev->subordinate) {
 		/* Set latency timers on sub bridges */
-		pci_write_config_byte(dev, PCI_SEC_LATENCY_TIMER, 48);
+		pci_write_config_byte(dev, PCI_SEC_LATENCY_TIMER, 64);
 		/* More bridge error detection */
 		pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &config);
 		config |= PCI_BRIDGE_CTL_PARITY | PCI_BRIDGE_CTL_SERR;
@@ -121,14 +121,10 @@ int pcibios_plat_dev_init(struct pci_dev *dev)
 	if (pos) {
 		/* Update Device Control */
 		pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &config);
-		/* Correctable Error Reporting */
-		config |= PCI_EXP_DEVCTL_CERE;
-		/* Non-Fatal Error Reporting */
-		config |= PCI_EXP_DEVCTL_NFERE;
-		/* Fatal Error Reporting */
-		config |= PCI_EXP_DEVCTL_FERE;
-		/* Unsupported Request */
-		config |= PCI_EXP_DEVCTL_URRE;
+		config |= PCI_EXP_DEVCTL_CERE; /* Correctable Error Reporting */
+		config |= PCI_EXP_DEVCTL_NFERE; /* Non-Fatal Error Reporting */
+		config |= PCI_EXP_DEVCTL_FERE;  /* Fatal Error Reporting */
+		config |= PCI_EXP_DEVCTL_URRE;  /* Unsupported Request */
 		pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, config);
 	}
 
diff --git a/arch/mips/pci/pci-tx4927.c b/arch/mips/pci/pci-tx4927.c
index a5807406a7f..a032ae0a533 100644
--- a/arch/mips/pci/pci-tx4927.c
+++ b/arch/mips/pci/pci-tx4927.c
@@ -85,7 +85,7 @@ void __init tx4927_setup_pcierr_irq(void)
 {
 	if (request_irq(TXX9_IRQ_BASE + TX4927_IR_PCIERR,
 			tx4927_pcierr_interrupt,
-			IRQF_DISABLED, "PCI error",
+			0, "PCI error",
 			(void *)TX4927_PCIC_REG))
 		printk(KERN_WARNING "Failed to request irq for PCIERR\n");
 }
diff --git a/arch/mips/pci/pci-tx4938.c b/arch/mips/pci/pci-tx4938.c
index 20e45f30b2e..141bba56248 100644
--- a/arch/mips/pci/pci-tx4938.c
+++ b/arch/mips/pci/pci-tx4938.c
@@ -136,7 +136,7 @@ void __init tx4938_setup_pcierr_irq(void)
 {
 	if (request_irq(TXX9_IRQ_BASE + TX4938_IR_PCIERR,
 			tx4927_pcierr_interrupt,
-			IRQF_DISABLED, "PCI error",
+			0, "PCI error",
 			(void *)TX4927_PCIC_REG))
 		printk(KERN_WARNING "Failed to request irq for PCIERR\n");
 }
diff --git a/arch/mips/pci/pci-tx4939.c b/arch/mips/pci/pci-tx4939.c
index 9ef840693ba..c10fbf2a19d 100644
--- a/arch/mips/pci/pci-tx4939.c
+++ b/arch/mips/pci/pci-tx4939.c
@@ -101,7 +101,7 @@ void __init tx4939_setup_pcierr_irq(void)
 {
 	if (request_irq(TXX9_IRQ_BASE + TX4939_IR_PCIERR,
 			tx4927_pcierr_interrupt,
-			IRQF_DISABLED, "PCI error",
+			0, "PCI error",
 			(void *)TX4939_PCIC_REG))
 		pr_warning("Failed to request irq for PCIERR\n");
 }
diff --git a/arch/mips/pci/pci-xlr.c b/arch/mips/pci/pci-xlr.c
index 38fece16c43..3d701a962ef 100644
--- a/arch/mips/pci/pci-xlr.c
+++ b/arch/mips/pci/pci-xlr.c
@@ -36,12 +36,18 @@
 #include <linux/pci.h>
 #include <linux/kernel.h>
 #include <linux/init.h>
+#include <linux/msi.h>
 #include <linux/mm.h>
+#include <linux/irq.h>
+#include <linux/irqdesc.h>
 #include <linux/console.h>
 
 #include <asm/io.h>
 
 #include <asm/netlogic/interrupt.h>
+#include <asm/netlogic/haldefs.h>
+
+#include <asm/netlogic/xlr/msidef.h>
 #include <asm/netlogic/xlr/iomap.h>
 #include <asm/netlogic/xlr/pic.h>
 #include <asm/netlogic/xlr/xlr.h>
@@ -150,7 +156,7 @@ struct pci_controller nlm_pci_controller = {
 	.io_offset      = 0x00000000UL,
 };
 
-int __init pcibios_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
+static int get_irq_vector(const struct pci_dev *dev)
 {
 	if (!nlm_chip_is_xls())
 		return	PIC_PCIX_IRQ;	/* for XLR just one IRQ*/
@@ -182,6 +188,101 @@ int __init pcibios_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
 	return 0;
 }
 
+#ifdef CONFIG_PCI_MSI
+void destroy_irq(unsigned int irq)
+{
+	    /* nothing to do yet */
+}
+
+void arch_teardown_msi_irq(unsigned int irq)
+{
+	destroy_irq(irq);
+}
+
+int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
+{
+	struct msi_msg msg;
+	int irq, ret;
+
+	irq = get_irq_vector(dev);
+	if (irq <= 0)
+		return 1;
+
+	msg.address_hi = MSI_ADDR_BASE_HI;
+	msg.address_lo = MSI_ADDR_BASE_LO   |
+		MSI_ADDR_DEST_MODE_PHYSICAL |
+		MSI_ADDR_REDIRECTION_CPU;
+
+	msg.data = MSI_DATA_TRIGGER_EDGE |
+		MSI_DATA_LEVEL_ASSERT    |
+		MSI_DATA_DELIVERY_FIXED;
+
+	ret = irq_set_msi_desc(irq, desc);
+	if (ret < 0) {
+		destroy_irq(irq);
+		return ret;
+	}
+
+	write_msi_msg(irq, &msg);
+	return 0;
+}
+#endif
+
+/* Extra ACK needed for XLR on chip PCI controller */
+static void xlr_pci_ack(struct irq_data *d)
+{
+	uint64_t pcibase = nlm_mmio_base(NETLOGIC_IO_PCIX_OFFSET);
+
+	nlm_read_reg(pcibase, (0x140 >> 2));
+}
+
+/* Extra ACK needed for XLS on chip PCIe controller */
+static void xls_pcie_ack(struct irq_data *d)
+{
+	uint64_t pciebase_le = nlm_mmio_base(NETLOGIC_IO_PCIE_1_OFFSET);
+
+	switch (d->irq) {
+	case PIC_PCIE_LINK0_IRQ:
+		nlm_write_reg(pciebase_le, (0x90 >> 2), 0xffffffff);
+		break;
+	case PIC_PCIE_LINK1_IRQ:
+		nlm_write_reg(pciebase_le, (0x94 >> 2), 0xffffffff);
+		break;
+	case PIC_PCIE_LINK2_IRQ:
+		nlm_write_reg(pciebase_le, (0x190 >> 2), 0xffffffff);
+		break;
+	case PIC_PCIE_LINK3_IRQ:
+		nlm_write_reg(pciebase_le, (0x194 >> 2), 0xffffffff);
+		break;
+	}
+}
+
+/* For XLS B silicon, the 3,4 PCI interrupts are different */
+static void xls_pcie_ack_b(struct irq_data *d)
+{
+	uint64_t pciebase_le = nlm_mmio_base(NETLOGIC_IO_PCIE_1_OFFSET);
+
+	switch (d->irq) {
+	case PIC_PCIE_LINK0_IRQ:
+		nlm_write_reg(pciebase_le, (0x90 >> 2), 0xffffffff);
+		break;
+	case PIC_PCIE_LINK1_IRQ:
+		nlm_write_reg(pciebase_le, (0x94 >> 2), 0xffffffff);
+		break;
+	case PIC_PCIE_XLSB0_LINK2_IRQ:
+		nlm_write_reg(pciebase_le, (0x190 >> 2), 0xffffffff);
+		break;
+	case PIC_PCIE_XLSB0_LINK3_IRQ:
+		nlm_write_reg(pciebase_le, (0x194 >> 2), 0xffffffff);
+		break;
+	}
+}
+
+int __init pcibios_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
+{
+	return get_irq_vector(dev);
+}
+
 /* Do platform specific device initialization at pci_enable_device() time */
 int pcibios_plat_dev_init(struct pci_dev *dev)
 {
@@ -204,6 +305,31 @@ static int __init pcibios_init(void)
 	pr_info("Registering XLR/XLS PCIX/PCIE Controller.\n");
 	register_pci_controller(&nlm_pci_controller);
 
+	/*
+	 * For PCI interrupts, we need to ack the PCI controller too, overload
+	 * irq handler data to do this
+	 */
+	if (nlm_chip_is_xls()) {
+		if (nlm_chip_is_xls_b()) {
+			irq_set_handler_data(PIC_PCIE_LINK0_IRQ,
+							xls_pcie_ack_b);
+			irq_set_handler_data(PIC_PCIE_LINK1_IRQ,
+							xls_pcie_ack_b);
+			irq_set_handler_data(PIC_PCIE_XLSB0_LINK2_IRQ,
+							xls_pcie_ack_b);
+			irq_set_handler_data(PIC_PCIE_XLSB0_LINK3_IRQ,
+							xls_pcie_ack_b);
+		} else {
+			irq_set_handler_data(PIC_PCIE_LINK0_IRQ, xls_pcie_ack);
+			irq_set_handler_data(PIC_PCIE_LINK1_IRQ, xls_pcie_ack);
+			irq_set_handler_data(PIC_PCIE_LINK2_IRQ, xls_pcie_ack);
+			irq_set_handler_data(PIC_PCIE_LINK3_IRQ, xls_pcie_ack);
+		}
+	} else {
+		/* XLR PCI controller ACK */
+		irq_set_handler_data(PIC_PCIE_XLSB0_LINK3_IRQ, xlr_pci_ack);
+	}
+
 	return 0;
 }
 
diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
index 41af7fa2887..aec2b111d35 100644
--- a/arch/mips/pci/pci.c
+++ b/arch/mips/pci/pci.c
@@ -4,8 +4,11 @@
  * Free Software Foundation;  either version 2 of the  License, or (at your
  * option) any later version.
  *
- * Copyright (C) 2003, 04 Ralf Baechle (ralf@linux-mips.org)
+ * Copyright (C) 2003, 04, 11 Ralf Baechle (ralf@linux-mips.org)
+ * Copyright (C) 2011 Wind River Systems,
+ *   written by Ralf Baechle (ralf@linux-mips.org)
  */
+#include <linux/bug.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
 #include <linux/bootmem.h>
@@ -14,6 +17,8 @@
 #include <linux/types.h>
 #include <linux/pci.h>
 
+#include <asm/cpu-info.h>
+
 /*
  * Indicate whether we respect the PCI setup left by the firmware.
  *
@@ -81,6 +86,7 @@ static void __devinit pcibios_scanbus(struct pci_controller *hose)
 {
 	static int next_busno;
 	static int need_domain_info;
+	LIST_HEAD(resources);
 	struct pci_bus *bus;
 
 	if (!hose->iommu)
@@ -89,7 +95,13 @@ static void __devinit pcibios_scanbus(struct pci_controller *hose)
 	if (hose->get_busno && pci_probe_only)
 		next_busno = (*hose->get_busno)();
 
-	bus = pci_scan_bus(next_busno, hose->pci_ops, hose);
+	pci_add_resource(&resources, hose->mem_resource);
+	pci_add_resource(&resources, hose->io_resource);
+	bus = pci_scan_root_bus(NULL, next_busno, hose->pci_ops, hose,
+				&resources);
+	if (!bus)
+		pci_free_resource_list(&resources);
+
 	hose->bus = bus;
 
 	need_domain_info = need_domain_info || hose->index;
@@ -150,10 +162,32 @@ out:
 	       "Skipping PCI bus scan due to resource conflict\n");
 }
 
+static void __init pcibios_set_cache_line_size(void)
+{
+	struct cpuinfo_mips *c = &current_cpu_data;
+	unsigned int lsize;
+
+	/*
+	 * Set PCI cacheline size to that of the highest level in the
+	 * cache hierarchy.
+	 */
+	lsize = c->dcache.linesz;
+	lsize = c->scache.linesz ? : lsize;
+	lsize = c->tcache.linesz ? : lsize;
+
+	BUG_ON(!lsize);
+
+	pci_dfl_cache_line_size = lsize >> 2;
+
+	pr_debug("PCI: pci_cache_line_size set to %d bytes\n", lsize);
+}
+
 static int __init pcibios_init(void)
 {
 	struct pci_controller *hose;
 
+	pcibios_set_cache_line_size();
+
 	/* Scan all of the recorded PCI controllers.  */
 	for (hose = hose_head; hose; hose = hose->next)
 		pcibios_scanbus(hose);
@@ -205,27 +239,6 @@ static int pcibios_enable_resources(struct pci_dev *dev, int mask)
 	return 0;
 }
 
-/*
- *  If we set up a device for bus mastering, we need to check the latency
- *  timer as certain crappy BIOSes forget to set it properly.
- */
-static unsigned int pcibios_max_latency = 255;
-
-void pcibios_set_master(struct pci_dev *dev)
-{
-	u8 lat;
-	pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
-	if (lat < 16)
-		lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
-	else if (lat > pcibios_max_latency)
-		lat = pcibios_max_latency;
-	else
-		return;
-	printk(KERN_DEBUG "PCI: Setting latency timer of device %s to %d\n",
-	       pci_name(dev), lat);
-	pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
-}
-
 unsigned int pcibios_assign_all_busses(void)
 {
 	return (pci_probe & PCI_ASSIGN_ALL_BUSSES) ? 1 : 0;
@@ -266,15 +279,11 @@ void __devinit pcibios_fixup_bus(struct pci_bus *bus)
 {
 	/* Propagate hose info into the subordinate devices.  */
 
-	struct pci_controller *hose = bus->sysdata;
 	struct list_head *ln;
 	struct pci_dev *dev = bus->self;
 
-	if (!dev) {
-		bus->resource[0] = hose->io_resource;
-		bus->resource[1] = hose->mem_resource;
-	} else if (pci_probe_only &&
-		   (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
+	if (pci_probe_only && dev &&
+	    (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
 		pci_read_bridge_bases(bus);
 		pcibios_fixup_device_resources(dev, bus);
 	}
diff --git a/arch/mips/pci/pcie-octeon.c b/arch/mips/pci/pcie-octeon.c
index 0583c463e5f..fdb4d558c0c 100644
--- a/arch/mips/pci/pcie-octeon.c
+++ b/arch/mips/pci/pcie-octeon.c
@@ -3,7 +3,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 2007, 2008 Cavium Networks
+ * Copyright (C) 2007, 2008, 2009, 2010, 2011 Cavium Networks
  */
 #include <linux/kernel.h>
 #include <linux/init.h>
@@ -11,15 +11,32 @@
 #include <linux/interrupt.h>
 #include <linux/time.h>
 #include <linux/delay.h>
+#include <linux/module.h>
 
 #include <asm/octeon/octeon.h>
 #include <asm/octeon/cvmx-npei-defs.h>
 #include <asm/octeon/cvmx-pciercx-defs.h>
 #include <asm/octeon/cvmx-pescx-defs.h>
 #include <asm/octeon/cvmx-pexp-defs.h>
+#include <asm/octeon/cvmx-pemx-defs.h>
+#include <asm/octeon/cvmx-dpi-defs.h>
+#include <asm/octeon/cvmx-sli-defs.h>
+#include <asm/octeon/cvmx-sriox-defs.h>
 #include <asm/octeon/cvmx-helper-errata.h>
 #include <asm/octeon/pci-octeon.h>
 
+#define MRRS_CN5XXX 0 /* 128 byte Max Read Request Size */
+#define MPS_CN5XXX  0 /* 128 byte Max Packet Size (Limit of most PCs) */
+#define MRRS_CN6XXX 3 /* 1024 byte Max Read Request Size */
+#define MPS_CN6XXX  0 /* 128 byte Max Packet Size (Limit of most PCs) */
+
+/* Module parameter to disable PCI probing */
+static int pcie_disable;
+module_param(pcie_disable, int, S_IRUGO);
+
+static int enable_pcie_14459_war;
+static int enable_pcie_bus_num_war[2];
+
 union cvmx_pcie_address {
 	uint64_t u64;
 	struct {
@@ -75,6 +92,8 @@ union cvmx_pcie_address {
 	} mem;
 };
 
+static int cvmx_pcie_rc_initialize(int pcie_port);
+
 #include <dma-coherence.h>
 
 /**
@@ -154,12 +173,21 @@ static inline uint64_t cvmx_pcie_get_mem_size(int pcie_port)
  */
 static uint32_t cvmx_pcie_cfgx_read(int pcie_port, uint32_t cfg_offset)
 {
-	union cvmx_pescx_cfg_rd pescx_cfg_rd;
-	pescx_cfg_rd.u64 = 0;
-	pescx_cfg_rd.s.addr = cfg_offset;
-	cvmx_write_csr(CVMX_PESCX_CFG_RD(pcie_port), pescx_cfg_rd.u64);
-	pescx_cfg_rd.u64 = cvmx_read_csr(CVMX_PESCX_CFG_RD(pcie_port));
-	return pescx_cfg_rd.s.data;
+	if (octeon_has_feature(OCTEON_FEATURE_NPEI)) {
+		union cvmx_pescx_cfg_rd pescx_cfg_rd;
+		pescx_cfg_rd.u64 = 0;
+		pescx_cfg_rd.s.addr = cfg_offset;
+		cvmx_write_csr(CVMX_PESCX_CFG_RD(pcie_port), pescx_cfg_rd.u64);
+		pescx_cfg_rd.u64 = cvmx_read_csr(CVMX_PESCX_CFG_RD(pcie_port));
+		return pescx_cfg_rd.s.data;
+	} else {
+		union cvmx_pemx_cfg_rd pemx_cfg_rd;
+		pemx_cfg_rd.u64 = 0;
+		pemx_cfg_rd.s.addr = cfg_offset;
+		cvmx_write_csr(CVMX_PEMX_CFG_RD(pcie_port), pemx_cfg_rd.u64);
+		pemx_cfg_rd.u64 = cvmx_read_csr(CVMX_PEMX_CFG_RD(pcie_port));
+		return pemx_cfg_rd.s.data;
+	}
 }
 
 /**
@@ -173,11 +201,19 @@ static uint32_t cvmx_pcie_cfgx_read(int pcie_port, uint32_t cfg_offset)
 static void cvmx_pcie_cfgx_write(int pcie_port, uint32_t cfg_offset,
 				 uint32_t val)
 {
-	union cvmx_pescx_cfg_wr pescx_cfg_wr;
-	pescx_cfg_wr.u64 = 0;
-	pescx_cfg_wr.s.addr = cfg_offset;
-	pescx_cfg_wr.s.data = val;
-	cvmx_write_csr(CVMX_PESCX_CFG_WR(pcie_port), pescx_cfg_wr.u64);
+	if (octeon_has_feature(OCTEON_FEATURE_NPEI)) {
+		union cvmx_pescx_cfg_wr pescx_cfg_wr;
+		pescx_cfg_wr.u64 = 0;
+		pescx_cfg_wr.s.addr = cfg_offset;
+		pescx_cfg_wr.s.data = val;
+		cvmx_write_csr(CVMX_PESCX_CFG_WR(pcie_port), pescx_cfg_wr.u64);
+	} else {
+		union cvmx_pemx_cfg_wr pemx_cfg_wr;
+		pemx_cfg_wr.u64 = 0;
+		pemx_cfg_wr.s.addr = cfg_offset;
+		pemx_cfg_wr.s.data = val;
+		cvmx_write_csr(CVMX_PEMX_CFG_WR(pcie_port), pemx_cfg_wr.u64);
+	}
 }
 
 /**
@@ -348,7 +384,6 @@ static void cvmx_pcie_config_write32(int pcie_port, int bus, int dev, int fn,
 static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
 {
 	union cvmx_pciercx_cfg030 pciercx_cfg030;
-	union cvmx_npei_ctl_status2 npei_ctl_status2;
 	union cvmx_pciercx_cfg070 pciercx_cfg070;
 	union cvmx_pciercx_cfg001 pciercx_cfg001;
 	union cvmx_pciercx_cfg032 pciercx_cfg032;
@@ -365,21 +400,21 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
 	/* Max Read Request Size (PCIE*_CFG030[MRRS]) */
 	/* Relaxed-order, no-snoop enables (PCIE*_CFG030[RO_EN,NS_EN] */
 	/* Error Message Enables (PCIE*_CFG030[CE_EN,NFE_EN,FE_EN,UR_EN]) */
-	pciercx_cfg030.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG030(pcie_port));
-	/*
-	 * Max payload size = 128 bytes for best Octeon DMA
-	 * performance.
-	 */
-	pciercx_cfg030.s.mps = 0;
+
+	pciercx_cfg030.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG030(pcie_port));
+	if (OCTEON_IS_MODEL(OCTEON_CN5XXX)) {
+		pciercx_cfg030.s.mps = MPS_CN5XXX;
+		pciercx_cfg030.s.mrrs = MRRS_CN5XXX;
+	} else {
+		pciercx_cfg030.s.mps = MPS_CN6XXX;
+		pciercx_cfg030.s.mrrs = MRRS_CN6XXX;
+	}
 	/*
-	 * Max read request size = 128 bytes for best Octeon DMA
-	 * performance.
+	 * Enable relaxed order processing. This will allow devices to
+	 * affect read response ordering.
 	 */
-	pciercx_cfg030.s.mrrs = 0;
-	/* Enable relaxed ordering. */
 	pciercx_cfg030.s.ro_en = 1;
-	/* Enable no snoop. */
+	/* Enable no snoop processing. Not used by Octeon */
 	pciercx_cfg030.s.ns_en = 1;
 	/* Correctable error reporting enable. */
 	pciercx_cfg030.s.ce_en = 1;
@@ -389,50 +424,67 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
 	pciercx_cfg030.s.fe_en = 1;
 	/* Unsupported request reporting enable. */
 	pciercx_cfg030.s.ur_en = 1;
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG030(pcie_port),
-			     pciercx_cfg030.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG030(pcie_port), pciercx_cfg030.u32);
 
-	/*
-	 * Max Payload Size (NPEI_CTL_STATUS2[MPS]) must match
-	 * PCIE*_CFG030[MPS]
-	 *
-	 * Max Read Request Size (NPEI_CTL_STATUS2[MRRS]) must not
-	 * exceed PCIE*_CFG030[MRRS].
-	 */
-	npei_ctl_status2.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS2);
-	/* Max payload size = 128 bytes for best Octeon DMA performance */
-	npei_ctl_status2.s.mps = 0;
-	/* Max read request size = 128 bytes for best Octeon DMA performance */
-	npei_ctl_status2.s.mrrs = 0;
-	if (pcie_port)
-		npei_ctl_status2.s.c1_b1_s = 3; /* Port1 BAR1 Size 256MB */
-	else
-		npei_ctl_status2.s.c0_b1_s = 3; /* Port0 BAR1 Size 256MB */
-	cvmx_write_csr(CVMX_PEXP_NPEI_CTL_STATUS2, npei_ctl_status2.u64);
+
+	if (octeon_has_feature(OCTEON_FEATURE_NPEI)) {
+		union cvmx_npei_ctl_status2 npei_ctl_status2;
+		/*
+		 * Max Payload Size (NPEI_CTL_STATUS2[MPS]) must match
+		 * PCIE*_CFG030[MPS].  Max Read Request Size
+		 * (NPEI_CTL_STATUS2[MRRS]) must not exceed
+		 * PCIE*_CFG030[MRRS]
+		 */
+		npei_ctl_status2.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS2);
+		/* Max payload size = 128 bytes for best Octeon DMA performance */
+		npei_ctl_status2.s.mps = MPS_CN5XXX;
+		/* Max read request size = 128 bytes for best Octeon DMA performance */
+		npei_ctl_status2.s.mrrs = MRRS_CN5XXX;
+		if (pcie_port)
+			npei_ctl_status2.s.c1_b1_s = 3; /* Port1 BAR1 Size 256MB */
+		else
+			npei_ctl_status2.s.c0_b1_s = 3; /* Port0 BAR1 Size 256MB */
+
+		cvmx_write_csr(CVMX_PEXP_NPEI_CTL_STATUS2, npei_ctl_status2.u64);
+	} else {
+		/*
+		 * Max Payload Size (DPI_SLI_PRTX_CFG[MPS]) must match
+		 * PCIE*_CFG030[MPS].  Max Read Request Size
+		 * (DPI_SLI_PRTX_CFG[MRRS]) must not exceed
+		 * PCIE*_CFG030[MRRS].
+		 */
+		union cvmx_dpi_sli_prtx_cfg prt_cfg;
+		union cvmx_sli_s2m_portx_ctl sli_s2m_portx_ctl;
+		prt_cfg.u64 = cvmx_read_csr(CVMX_DPI_SLI_PRTX_CFG(pcie_port));
+		prt_cfg.s.mps = MPS_CN6XXX;
+		prt_cfg.s.mrrs = MRRS_CN6XXX;
+		/* Max outstanding load request. */
+		prt_cfg.s.molr = 32;
+		cvmx_write_csr(CVMX_DPI_SLI_PRTX_CFG(pcie_port), prt_cfg.u64);
+
+		sli_s2m_portx_ctl.u64 = cvmx_read_csr(CVMX_PEXP_SLI_S2M_PORTX_CTL(pcie_port));
+		sli_s2m_portx_ctl.s.mrrs = MRRS_CN6XXX;
+		cvmx_write_csr(CVMX_PEXP_SLI_S2M_PORTX_CTL(pcie_port), sli_s2m_portx_ctl.u64);
+	}
 
 	/* ECRC Generation (PCIE*_CFG070[GE,CE]) */
-	pciercx_cfg070.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG070(pcie_port));
+	pciercx_cfg070.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG070(pcie_port));
 	pciercx_cfg070.s.ge = 1;	/* ECRC generation enable. */
 	pciercx_cfg070.s.ce = 1;	/* ECRC check enable. */
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG070(pcie_port),
-			     pciercx_cfg070.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG070(pcie_port), pciercx_cfg070.u32);
 
 	/*
-	 * Access Enables (PCIE*_CFG001[MSAE,ME]) ME and MSAE should
-	 * always be set.
-	 *
-	 * Interrupt Disable (PCIE*_CFG001[I_DIS]) System Error
-	 * Message Enable (PCIE*_CFG001[SEE])
+	 * Access Enables (PCIE*_CFG001[MSAE,ME])
+	 * ME and MSAE should always be set.
+	 * Interrupt Disable (PCIE*_CFG001[I_DIS])
+	 * System Error Message Enable (PCIE*_CFG001[SEE])
 	 */
-	pciercx_cfg001.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG001(pcie_port));
+	pciercx_cfg001.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG001(pcie_port));
 	pciercx_cfg001.s.msae = 1;	/* Memory space enable. */
 	pciercx_cfg001.s.me = 1;	/* Bus master enable. */
 	pciercx_cfg001.s.i_dis = 1;	/* INTx assertion disable. */
 	pciercx_cfg001.s.see = 1;	/* SERR# enable */
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG001(pcie_port),
-			pciercx_cfg001.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG001(pcie_port), pciercx_cfg001.u32);
 
 	/* Advanced Error Recovery Message Enables */
 	/* (PCIE*_CFG066,PCIE*_CFG067,PCIE*_CFG069) */
@@ -440,14 +492,11 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
 	/* Use CVMX_PCIERCX_CFG067 hardware default */
 	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG069(pcie_port), 0);
 
-	/* Active State Power Management (PCIE*_CFG032[ASLPC]) */
-	pciercx_cfg032.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG032(pcie_port));
-	pciercx_cfg032.s.aslpc = 0;	/* Active state Link PM control. */
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG032(pcie_port),
-			     pciercx_cfg032.u32);
 
-	/* Entrance Latencies (PCIE*_CFG451[L0EL,L1EL]) */
+	/* Active State Power Management (PCIE*_CFG032[ASLPC]) */
+	pciercx_cfg032.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG032(pcie_port));
+	pciercx_cfg032.s.aslpc = 0; /* Active state Link PM control. */
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG032(pcie_port), pciercx_cfg032.u32);
 
 	/*
 	 * Link Width Mode (PCIERCn_CFG452[LME]) - Set during
@@ -462,8 +511,8 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
 	pciercx_cfg006.s.pbnum = 1;
 	pciercx_cfg006.s.sbnum = 1;
 	pciercx_cfg006.s.subbnum = 1;
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG006(pcie_port),
-			     pciercx_cfg006.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG006(pcie_port), pciercx_cfg006.u32);
+
 
 	/*
 	 * Memory-mapped I/O BAR (PCIERCn_CFG008)
@@ -473,8 +522,8 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
 	pciercx_cfg008.u32 = 0;
 	pciercx_cfg008.s.mb_addr = 0x100;
 	pciercx_cfg008.s.ml_addr = 0;
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG008(pcie_port),
-			     pciercx_cfg008.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG008(pcie_port), pciercx_cfg008.u32);
+
 
 	/*
 	 * Prefetchable BAR (PCIERCn_CFG009,PCIERCn_CFG010,PCIERCn_CFG011)
@@ -482,72 +531,51 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
 	 * PCIERCn_CFG011[UMEM_LIMIT],PCIERCn_CFG009[LMEM_LIMIT] <
 	 * PCIERCn_CFG010[UMEM_BASE],PCIERCn_CFG009[LMEM_BASE]
 	 */
-	pciercx_cfg009.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG009(pcie_port));
-	pciercx_cfg010.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG010(pcie_port));
-	pciercx_cfg011.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG011(pcie_port));
+	pciercx_cfg009.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG009(pcie_port));
+	pciercx_cfg010.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG010(pcie_port));
+	pciercx_cfg011.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG011(pcie_port));
 	pciercx_cfg009.s.lmem_base = 0x100;
 	pciercx_cfg009.s.lmem_limit = 0;
 	pciercx_cfg010.s.umem_base = 0x100;
 	pciercx_cfg011.s.umem_limit = 0;
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG009(pcie_port),
-			     pciercx_cfg009.u32);
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG010(pcie_port),
-			     pciercx_cfg010.u32);
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG011(pcie_port),
-			     pciercx_cfg011.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG009(pcie_port), pciercx_cfg009.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG010(pcie_port), pciercx_cfg010.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG011(pcie_port), pciercx_cfg011.u32);
 
 	/*
 	 * System Error Interrupt Enables (PCIERCn_CFG035[SECEE,SEFEE,SENFEE])
 	 * PME Interrupt Enables (PCIERCn_CFG035[PMEIE])
-	 */
-	pciercx_cfg035.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG035(pcie_port));
-	/* System error on correctable error enable. */
-	pciercx_cfg035.s.secee = 1;
-	/* System error on fatal error enable. */
-	pciercx_cfg035.s.sefee = 1;
-	/* System error on non-fatal error enable. */
-	pciercx_cfg035.s.senfee = 1;
-	/* PME interrupt enable. */
-	pciercx_cfg035.s.pmeie = 1;
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG035(pcie_port),
-			     pciercx_cfg035.u32);
+	*/
+	pciercx_cfg035.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG035(pcie_port));
+	pciercx_cfg035.s.secee = 1; /* System error on correctable error enable. */
+	pciercx_cfg035.s.sefee = 1; /* System error on fatal error enable. */
+	pciercx_cfg035.s.senfee = 1; /* System error on non-fatal error enable. */
+	pciercx_cfg035.s.pmeie = 1; /* PME interrupt enable. */
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG035(pcie_port), pciercx_cfg035.u32);
 
 	/*
 	 * Advanced Error Recovery Interrupt Enables
 	 * (PCIERCn_CFG075[CERE,NFERE,FERE])
 	 */
-	pciercx_cfg075.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG075(pcie_port));
-	/* Correctable error reporting enable. */
-	pciercx_cfg075.s.cere = 1;
-	/* Non-fatal error reporting enable. */
-	pciercx_cfg075.s.nfere = 1;
-	/* Fatal error reporting enable. */
-	pciercx_cfg075.s.fere = 1;
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG075(pcie_port),
-			     pciercx_cfg075.u32);
+	pciercx_cfg075.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG075(pcie_port));
+	pciercx_cfg075.s.cere = 1; /* Correctable error reporting enable. */
+	pciercx_cfg075.s.nfere = 1; /* Non-fatal error reporting enable. */
+	pciercx_cfg075.s.fere = 1; /* Fatal error reporting enable. */
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG075(pcie_port), pciercx_cfg075.u32);
 
-	/* HP Interrupt Enables (PCIERCn_CFG034[HPINT_EN],
+	/*
+	 * HP Interrupt Enables (PCIERCn_CFG034[HPINT_EN],
 	 * PCIERCn_CFG034[DLLS_EN,CCINT_EN])
 	 */
-	pciercx_cfg034.u32 =
-		cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG034(pcie_port));
-	/* Hot-plug interrupt enable. */
-	pciercx_cfg034.s.hpint_en = 1;
-	/* Data Link Layer state changed enable */
-	pciercx_cfg034.s.dlls_en = 1;
-	/* Command completed interrupt enable. */
-	pciercx_cfg034.s.ccint_en = 1;
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG034(pcie_port),
-			     pciercx_cfg034.u32);
+	pciercx_cfg034.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG034(pcie_port));
+	pciercx_cfg034.s.hpint_en = 1; /* Hot-plug interrupt enable. */
+	pciercx_cfg034.s.dlls_en = 1; /* Data Link Layer state changed enable */
+	pciercx_cfg034.s.ccint_en = 1; /* Command completed interrupt enable. */
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG034(pcie_port), pciercx_cfg034.u32);
 }
 
 /**
- * Initialize a host mode PCIe link. This function takes a PCIe
+ * Initialize a host mode PCIe gen 1 link. This function takes a PCIe
  * port from reset to a link up state. Software can then begin
  * configuring the rest of the link.
  *
@@ -555,7 +583,7 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
  *
  * Returns Zero on success
  */
-static int __cvmx_pcie_rc_initialize_link(int pcie_port)
+static int __cvmx_pcie_rc_initialize_link_gen1(int pcie_port)
 {
 	uint64_t start_cycle;
 	union cvmx_pescx_ctl_status pescx_ctl_status;
@@ -564,18 +592,15 @@ static int __cvmx_pcie_rc_initialize_link(int pcie_port)
 	union cvmx_pciercx_cfg448 pciercx_cfg448;
 
 	/* Set the lane width */
-	pciercx_cfg452.u32 =
-	    cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG452(pcie_port));
+	pciercx_cfg452.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG452(pcie_port));
 	pescx_ctl_status.u64 = cvmx_read_csr(CVMX_PESCX_CTL_STATUS(pcie_port));
-	if (pescx_ctl_status.s.qlm_cfg == 0) {
+	if (pescx_ctl_status.s.qlm_cfg == 0)
 		/* We're in 8 lane (56XX) or 4 lane (54XX) mode */
 		pciercx_cfg452.s.lme = 0xf;
-	} else {
+	else
 		/* We're in 4 lane (56XX) or 2 lane (52XX) mode */
 		pciercx_cfg452.s.lme = 0x7;
-	}
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG452(pcie_port),
-			     pciercx_cfg452.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG452(pcie_port), pciercx_cfg452.u32);
 
 	/*
 	 * CN52XX pass 1.x has an errata where length mismatches on UR
@@ -584,19 +609,15 @@ static int __cvmx_pcie_rc_initialize_link(int pcie_port)
 	 */
 	if (OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X)) {
 		union cvmx_pciercx_cfg455 pciercx_cfg455;
-		pciercx_cfg455.u32 =
-		    cvmx_pcie_cfgx_read(pcie_port,
-					CVMX_PCIERCX_CFG455(pcie_port));
+		pciercx_cfg455.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG455(pcie_port));
 		pciercx_cfg455.s.m_cpl_len_err = 1;
-		cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG455(pcie_port),
-				     pciercx_cfg455.u32);
+		cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG455(pcie_port), pciercx_cfg455.u32);
 	}
 
 	/* Lane swap needs to be manually enabled for CN52XX */
 	if (OCTEON_IS_MODEL(OCTEON_CN52XX) && (pcie_port == 1)) {
 		pescx_ctl_status.s.lane_swp = 1;
-		cvmx_write_csr(CVMX_PESCX_CTL_STATUS(pcie_port),
-			       pescx_ctl_status.u64);
+		cvmx_write_csr(CVMX_PESCX_CTL_STATUS(pcie_port), pescx_ctl_status.u64);
 	}
 
 	/* Bring up the link */
@@ -612,24 +633,18 @@ static int __cvmx_pcie_rc_initialize_link(int pcie_port)
 		__cvmx_helper_errata_qlm_disable_2nd_order_cdr(0);
 
 	/* Wait for the link to come up */
-	cvmx_dprintf("PCIe: Waiting for port %d link\n", pcie_port);
 	start_cycle = cvmx_get_cycle();
 	do {
-		if (cvmx_get_cycle() - start_cycle >
-		    2 * cvmx_sysinfo_get()->cpu_clock_hz) {
-			cvmx_dprintf("PCIe: Port %d link timeout\n",
-				     pcie_port);
+		if (cvmx_get_cycle() - start_cycle > 2 * octeon_get_clock_rate()) {
+			cvmx_dprintf("PCIe: Port %d link timeout\n", pcie_port);
 			return -1;
 		}
 		cvmx_wait(10000);
-		pciercx_cfg032.u32 =
-		    cvmx_pcie_cfgx_read(pcie_port,
-					CVMX_PCIERCX_CFG032(pcie_port));
+		pciercx_cfg032.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG032(pcie_port));
 	} while (pciercx_cfg032.s.dlla == 0);
 
-	/* Display the link status */
-	cvmx_dprintf("PCIe: Port %d link active, %d lanes\n", pcie_port,
-		     pciercx_cfg032.s.nlw);
+	/* Clear all pending errors */
+	cvmx_write_csr(CVMX_PEXP_NPEI_INT_SUM, cvmx_read_csr(CVMX_PEXP_NPEI_INT_SUM));
 
 	/*
 	 * Update the Replay Time Limit. Empirically, some PCIe
@@ -639,8 +654,7 @@ static int __cvmx_pcie_rc_initialize_link(int pcie_port)
 	 * our actual 256 byte MPS. The numbers below are directly
 	 * from the PCIe spec table 3-4.
 	 */
-	pciercx_cfg448.u32 =
-	    cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG448(pcie_port));
+	pciercx_cfg448.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG448(pcie_port));
 	switch (pciercx_cfg032.s.nlw) {
 	case 1:		/* 1 lane */
 		pciercx_cfg448.s.rtl = 1677;
@@ -655,21 +669,28 @@ static int __cvmx_pcie_rc_initialize_link(int pcie_port)
 		pciercx_cfg448.s.rtl = 258;
 		break;
 	}
-	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG448(pcie_port),
-			     pciercx_cfg448.u32);
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG448(pcie_port), pciercx_cfg448.u32);
 
 	return 0;
 }
 
+static void __cvmx_increment_ba(union cvmx_sli_mem_access_subidx *pmas)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+		pmas->cn68xx.ba++;
+	else
+		pmas->cn63xx.ba++;
+}
+
 /**
- * Initialize a PCIe port for use in host(RC) mode. It doesn't
+ * Initialize a PCIe gen 1 port for use in host(RC) mode. It doesn't
  * enumerate the bus.
  *
  * @pcie_port: PCIe port to initialize
  *
  * Returns Zero on success
  */
-static int cvmx_pcie_rc_initialize(int pcie_port)
+static int __cvmx_pcie_rc_initialize_gen1(int pcie_port)
 {
 	int i;
 	int base;
@@ -682,16 +703,17 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 	union cvmx_npei_mem_access_subidx mem_access_subid;
 	union cvmx_npei_dbg_data npei_dbg_data;
 	union cvmx_pescx_ctl_status2 pescx_ctl_status2;
+	union cvmx_pciercx_cfg032 pciercx_cfg032;
 	union cvmx_npei_bar1_indexx bar1_index;
 
+retry:
 	/*
 	 * Make sure we aren't trying to setup a target mode interface
 	 * in host mode.
 	 */
 	npei_ctl_status.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS);
 	if ((pcie_port == 0) && !npei_ctl_status.s.host_mode) {
-		cvmx_dprintf("PCIe: ERROR: cvmx_pcie_rc_initialize() called "
-			     "on port0, but port0 is not in host mode\n");
+		cvmx_dprintf("PCIe: Port %d in endpoint mode\n", pcie_port);
 		return -1;
 	}
 
@@ -702,9 +724,7 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 	if (OCTEON_IS_MODEL(OCTEON_CN52XX)) {
 		npei_dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA);
 		if ((pcie_port == 1) && npei_dbg_data.cn52xx.qlm0_link_width) {
-			cvmx_dprintf("PCIe: ERROR: cvmx_pcie_rc_initialize() "
-				     "called on port1, but port1 is "
-				     "disabled\n");
+			cvmx_dprintf("PCIe: ERROR: cvmx_pcie_rc_initialize() called on port1, but port1 is disabled\n");
 			return -1;
 		}
 	}
@@ -733,7 +753,7 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 		 * the board. As a workaround for this bug, we bring
 		 * both PCIe ports out of reset at the same time
 		 * instead of on separate calls. So for port 0, we
-		 * bring both out of reset and do nothing on port 1.
+		 * bring both out of reset and do nothing on port 1
 		 */
 		if (pcie_port == 0) {
 			ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST);
@@ -746,13 +766,10 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 			if (ciu_soft_prst.s.soft_prst == 0) {
 				/* Reset the ports */
 				ciu_soft_prst.s.soft_prst = 1;
-				cvmx_write_csr(CVMX_CIU_SOFT_PRST,
-					       ciu_soft_prst.u64);
-				ciu_soft_prst.u64 =
-				    cvmx_read_csr(CVMX_CIU_SOFT_PRST1);
+				cvmx_write_csr(CVMX_CIU_SOFT_PRST, ciu_soft_prst.u64);
+				ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST1);
 				ciu_soft_prst.s.soft_prst = 1;
-				cvmx_write_csr(CVMX_CIU_SOFT_PRST1,
-					       ciu_soft_prst.u64);
+				cvmx_write_csr(CVMX_CIU_SOFT_PRST1, ciu_soft_prst.u64);
 				/* Wait until pcie resets the ports. */
 				udelay(2000);
 			}
@@ -782,11 +799,9 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 			/* Reset the port */
 			ciu_soft_prst.s.soft_prst = 1;
 			if (pcie_port)
-				cvmx_write_csr(CVMX_CIU_SOFT_PRST1,
-					       ciu_soft_prst.u64);
+				cvmx_write_csr(CVMX_CIU_SOFT_PRST1, ciu_soft_prst.u64);
 			else
-				cvmx_write_csr(CVMX_CIU_SOFT_PRST,
-					       ciu_soft_prst.u64);
+				cvmx_write_csr(CVMX_CIU_SOFT_PRST, ciu_soft_prst.u64);
 			/* Wait until pcie resets the ports. */
 			udelay(2000);
 		}
@@ -808,25 +823,21 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 	 */
 	cvmx_wait(400000);
 
-	/* PESCX_BIST_STATUS2[PCLK_RUN] was missing on pass 1 of CN56XX and
-	   CN52XX, so we only probe it on newer chips */
-	if (!OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_X)
-	    && !OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X)) {
+	/*
+	 * PESCX_BIST_STATUS2[PCLK_RUN] was missing on pass 1 of
+	 * CN56XX and CN52XX, so we only probe it on newer chips
+	 */
+	if (!OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_X) && !OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X)) {
 		/* Clear PCLK_RUN so we can check if the clock is running */
-		pescx_ctl_status2.u64 =
-		    cvmx_read_csr(CVMX_PESCX_CTL_STATUS2(pcie_port));
+		pescx_ctl_status2.u64 = cvmx_read_csr(CVMX_PESCX_CTL_STATUS2(pcie_port));
 		pescx_ctl_status2.s.pclk_run = 1;
-		cvmx_write_csr(CVMX_PESCX_CTL_STATUS2(pcie_port),
-			       pescx_ctl_status2.u64);
-		/*
-		 * Now that we cleared PCLK_RUN, wait for it to be set
-		 * again telling us the clock is running.
+		cvmx_write_csr(CVMX_PESCX_CTL_STATUS2(pcie_port), pescx_ctl_status2.u64);
+		/* Now that we cleared PCLK_RUN, wait for it to be set
+		 * again telling us the clock is running
 		 */
 		if (CVMX_WAIT_FOR_FIELD64(CVMX_PESCX_CTL_STATUS2(pcie_port),
-					  union cvmx_pescx_ctl_status2,
-					  pclk_run, ==, 1, 10000)) {
-			cvmx_dprintf("PCIe: Port %d isn't clocked, skipping.\n",
-				     pcie_port);
+					  union cvmx_pescx_ctl_status2, pclk_run, ==, 1, 10000)) {
+			cvmx_dprintf("PCIe: Port %d isn't clocked, skipping.\n", pcie_port);
 			return -1;
 		}
 	}
@@ -836,30 +847,26 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 	 * the board probably hasn't wired the clocks up and the
 	 * interface should be skipped.
 	 */
-	pescx_ctl_status2.u64 =
-	    cvmx_read_csr(CVMX_PESCX_CTL_STATUS2(pcie_port));
+	pescx_ctl_status2.u64 = cvmx_read_csr(CVMX_PESCX_CTL_STATUS2(pcie_port));
 	if (pescx_ctl_status2.s.pcierst) {
-		cvmx_dprintf("PCIe: Port %d stuck in reset, skipping.\n",
-			     pcie_port);
+		cvmx_dprintf("PCIe: Port %d stuck in reset, skipping.\n", pcie_port);
 		return -1;
 	}
 
 	/*
-	 * Check BIST2 status. If any bits are set skip this interface. This
-	 * is an attempt to catch PCIE-813 on pass 1 parts.
+	 * Check BIST2 status. If any bits are set skip this
+	 * interface. This is an attempt to catch PCIE-813 on pass 1
+	 * parts.
 	 */
-	pescx_bist_status2.u64 =
-	    cvmx_read_csr(CVMX_PESCX_BIST_STATUS2(pcie_port));
+	pescx_bist_status2.u64 = cvmx_read_csr(CVMX_PESCX_BIST_STATUS2(pcie_port));
 	if (pescx_bist_status2.u64) {
-		cvmx_dprintf("PCIe: Port %d BIST2 failed. Most likely this "
-			     "port isn't hooked up, skipping.\n",
+		cvmx_dprintf("PCIe: Port %d BIST2 failed. Most likely this port isn't hooked up, skipping.\n",
 			     pcie_port);
 		return -1;
 	}
 
 	/* Check BIST status */
-	pescx_bist_status.u64 =
-	    cvmx_read_csr(CVMX_PESCX_BIST_STATUS(pcie_port));
+	pescx_bist_status.u64 = cvmx_read_csr(CVMX_PESCX_BIST_STATUS(pcie_port));
 	if (pescx_bist_status.u64)
 		cvmx_dprintf("PCIe: BIST FAILED for port %d (0x%016llx)\n",
 			     pcie_port, CAST64(pescx_bist_status.u64));
@@ -868,50 +875,37 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 	__cvmx_pcie_rc_initialize_config_space(pcie_port);
 
 	/* Bring the link up */
-	if (__cvmx_pcie_rc_initialize_link(pcie_port)) {
-		cvmx_dprintf
-		    ("PCIe: ERROR: cvmx_pcie_rc_initialize_link() failed\n");
+	if (__cvmx_pcie_rc_initialize_link_gen1(pcie_port)) {
+		cvmx_dprintf("PCIe: Failed to initialize port %d, probably the slot is empty\n",
+			     pcie_port);
 		return -1;
 	}
 
 	/* Store merge control (NPEI_MEM_ACCESS_CTL[TIMER,MAX_WORD]) */
 	npei_mem_access_ctl.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_MEM_ACCESS_CTL);
-	/* Allow 16 words to combine */
-	npei_mem_access_ctl.s.max_word = 0;
-	/* Wait up to 127 cycles for more data */
-	npei_mem_access_ctl.s.timer = 127;
+	npei_mem_access_ctl.s.max_word = 0;     /* Allow 16 words to combine */
+	npei_mem_access_ctl.s.timer = 127;      /* Wait up to 127 cycles for more data */
 	cvmx_write_csr(CVMX_PEXP_NPEI_MEM_ACCESS_CTL, npei_mem_access_ctl.u64);
 
 	/* Setup Mem access SubDIDs */
 	mem_access_subid.u64 = 0;
-	/* Port the request is sent to. */
-	mem_access_subid.s.port = pcie_port;
-	/* Due to an errata on pass 1 chips, no merging is allowed. */
-	mem_access_subid.s.nmerge = 1;
-	/* Endian-swap for Reads. */
-	mem_access_subid.s.esr = 1;
-	/* Endian-swap for Writes. */
-	mem_access_subid.s.esw = 1;
-	/* No Snoop for Reads. */
-	mem_access_subid.s.nsr = 1;
-	/* No Snoop for Writes. */
-	mem_access_subid.s.nsw = 1;
-	/* Disable Relaxed Ordering for Reads. */
-	mem_access_subid.s.ror = 0;
-	/* Disable Relaxed Ordering for Writes. */
-	mem_access_subid.s.row = 0;
-	/* PCIe Address Bits <63:34>. */
-	mem_access_subid.s.ba = 0;
+	mem_access_subid.s.port = pcie_port; /* Port the request is sent to. */
+	mem_access_subid.s.nmerge = 1;  /* Due to an errata on pass 1 chips, no merging is allowed. */
+	mem_access_subid.s.esr = 1;	/* Endian-swap for Reads. */
+	mem_access_subid.s.esw = 1;	/* Endian-swap for Writes. */
+	mem_access_subid.s.nsr = 0;	/* Enable Snooping for Reads. Octeon doesn't care, but devices might want this more conservative setting */
+	mem_access_subid.s.nsw = 0;	/* Enable Snoop for Writes. */
+	mem_access_subid.s.ror = 0;	/* Disable Relaxed Ordering for Reads. */
+	mem_access_subid.s.row = 0;	/* Disable Relaxed Ordering for Writes. */
+	mem_access_subid.s.ba = 0;	/* PCIe Adddress Bits <63:34>. */
 
 	/*
 	 * Setup mem access 12-15 for port 0, 16-19 for port 1,
 	 * supplying 36 bits of address space.
 	 */
 	for (i = 12 + pcie_port * 4; i < 16 + pcie_port * 4; i++) {
-		cvmx_write_csr(CVMX_PEXP_NPEI_MEM_ACCESS_SUBIDX(i),
-			       mem_access_subid.u64);
-		/* Set each SUBID to extend the addressable range */
-		mem_access_subid.s.ba += 1;
+		cvmx_write_csr(CVMX_PEXP_NPEI_MEM_ACCESS_SUBIDX(i), mem_access_subid.u64);
+		mem_access_subid.s.ba += 1; /* Set each SUBID to extend the addressable range */
 	}
 
 	/*
@@ -927,7 +921,7 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 	/* Set Octeon's BAR0 to decode 0-16KB. It overlaps with Bar2 */
 	cvmx_write_csr(CVMX_PESCX_P2N_BAR0_START(pcie_port), 0);
 
-	/* BAR1 follows BAR2 with a gap. */
+	/* BAR1 follows BAR2 with a gap so it has the same address as for gen2. */
 	cvmx_write_csr(CVMX_PESCX_P2N_BAR1_START(pcie_port), CVMX_PCIE_BAR1_RC_BASE);
 
 	bar1_index.u32 = 0;
@@ -992,14 +986,474 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
 		npei_ctl_port.s.waitl_com = 0;
 		cvmx_write_csr(CVMX_PEXP_NPEI_CTL_PORT0, npei_ctl_port.u64);
 	}
+
+	/*
+	 * Both pass 1 and pass 2 of CN52XX and CN56XX have an errata
+	 * that causes TLP ordering to not be preserved after multiple
+	 * PCIe port resets. This code detects this fault and corrects
+	 * it by aligning the TLP counters properly. Another link
+	 * reset is then performed. See PCIE-13340
+	 */
+	if (OCTEON_IS_MODEL(OCTEON_CN56XX_PASS2_X) ||
+	    OCTEON_IS_MODEL(OCTEON_CN52XX_PASS2_X) ||
+	    OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_X) ||
+	    OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X)) {
+		union cvmx_npei_dbg_data dbg_data;
+		int old_in_fif_p_count;
+		int in_fif_p_count;
+		int out_p_count;
+		int in_p_offset = (OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X) || OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_X)) ? 4 : 1;
+		int i;
+
+		/*
+		 * Choose a write address of 1MB. It should be
+		 * harmless as all bars haven't been setup.
+		 */
+		uint64_t write_address = (cvmx_pcie_get_mem_base_address(pcie_port) + 0x100000) | (1ull<<63);
+
+		/*
+		 * Make sure at least in_p_offset have been executed before we try and
+		 * read in_fif_p_count
+		 */
+		i = in_p_offset;
+		while (i--) {
+			cvmx_write64_uint32(write_address, 0);
+			cvmx_wait(10000);
+		}
+
+		/*
+		 * Read the IN_FIF_P_COUNT from the debug
+		 * select. IN_FIF_P_COUNT can be unstable sometimes so
+		 * read it twice with a write between the reads.  This
+		 * way we can tell the value is good as it will
+		 * increment by one due to the write
+		 */
+		cvmx_write_csr(CVMX_PEXP_NPEI_DBG_SELECT, (pcie_port) ? 0xd7fc : 0xcffc);
+		cvmx_read_csr(CVMX_PEXP_NPEI_DBG_SELECT);
+		do {
+			dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA);
+			old_in_fif_p_count = dbg_data.s.data & 0xff;
+			cvmx_write64_uint32(write_address, 0);
+			cvmx_wait(10000);
+			dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA);
+			in_fif_p_count = dbg_data.s.data & 0xff;
+		} while (in_fif_p_count != ((old_in_fif_p_count+1) & 0xff));
+
+		/* Update in_fif_p_count for it's offset with respect to out_p_count */
+		in_fif_p_count = (in_fif_p_count + in_p_offset) & 0xff;
+
+		/* Read the OUT_P_COUNT from the debug select */
+		cvmx_write_csr(CVMX_PEXP_NPEI_DBG_SELECT, (pcie_port) ? 0xd00f : 0xc80f);
+		cvmx_read_csr(CVMX_PEXP_NPEI_DBG_SELECT);
+		dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA);
+		out_p_count = (dbg_data.s.data>>1) & 0xff;
+
+		/* Check that the two counters are aligned */
+		if (out_p_count != in_fif_p_count) {
+			cvmx_dprintf("PCIe: Port %d aligning TLP counters as workaround to maintain ordering\n", pcie_port);
+			while (in_fif_p_count != 0) {
+				cvmx_write64_uint32(write_address, 0);
+				cvmx_wait(10000);
+				in_fif_p_count = (in_fif_p_count + 1) & 0xff;
+			}
+			/*
+			 * The EBH5200 board swapped the PCIe reset
+			 * lines on the board. This means we must
+			 * bring both links down and up, which will
+			 * cause the PCIe0 to need alignment
+			 * again. Lots of messages will be displayed,
+			 * but everything should work
+			 */
+			if ((cvmx_sysinfo_get()->board_type == CVMX_BOARD_TYPE_EBH5200) &&
+				(pcie_port == 1))
+				cvmx_pcie_rc_initialize(0);
+			/* Rety bringing this port up */
+			goto retry;
+		}
+	}
+
+	/* Display the link status */
+	pciercx_cfg032.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG032(pcie_port));
+	cvmx_dprintf("PCIe: Port %d link active, %d lanes\n", pcie_port, pciercx_cfg032.s.nlw);
+
 	return 0;
 }
 
+/**
+  * Initialize a host mode PCIe gen 2 link. This function takes a PCIe
+ * port from reset to a link up state. Software can then begin
+ * configuring the rest of the link.
+ *
+ * @pcie_port: PCIe port to initialize
+ *
+ * Return Zero on success.
+ */
+static int __cvmx_pcie_rc_initialize_link_gen2(int pcie_port)
+{
+	uint64_t start_cycle;
+	union cvmx_pemx_ctl_status pem_ctl_status;
+	union cvmx_pciercx_cfg032 pciercx_cfg032;
+	union cvmx_pciercx_cfg448 pciercx_cfg448;
 
-/* Above was cvmx-pcie.c, below original pcie.c */
+	/* Bring up the link */
+	pem_ctl_status.u64 = cvmx_read_csr(CVMX_PEMX_CTL_STATUS(pcie_port));
+	pem_ctl_status.s.lnk_enb = 1;
+	cvmx_write_csr(CVMX_PEMX_CTL_STATUS(pcie_port), pem_ctl_status.u64);
+
+	/* Wait for the link to come up */
+	start_cycle = cvmx_get_cycle();
+	do {
+		if (cvmx_get_cycle() - start_cycle >  octeon_get_clock_rate())
+			return -1;
+		cvmx_wait(10000);
+		pciercx_cfg032.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG032(pcie_port));
+	} while ((pciercx_cfg032.s.dlla == 0) || (pciercx_cfg032.s.lt == 1));
+
+	/*
+	 * Update the Replay Time Limit. Empirically, some PCIe
+	 * devices take a little longer to respond than expected under
+	 * load. As a workaround for this we configure the Replay Time
+	 * Limit to the value expected for a 512 byte MPS instead of
+	 * our actual 256 byte MPS. The numbers below are directly
+	 * from the PCIe spec table 3-4
+	 */
+	pciercx_cfg448.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG448(pcie_port));
+	switch (pciercx_cfg032.s.nlw) {
+	case 1: /* 1 lane */
+		pciercx_cfg448.s.rtl = 1677;
+		break;
+	case 2: /* 2 lanes */
+		pciercx_cfg448.s.rtl = 867;
+		break;
+	case 4: /* 4 lanes */
+		pciercx_cfg448.s.rtl = 462;
+		break;
+	case 8: /* 8 lanes */
+		pciercx_cfg448.s.rtl = 258;
+		break;
+	}
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG448(pcie_port), pciercx_cfg448.u32);
+
+	return 0;
+}
 
 
 /**
+ * Initialize a PCIe gen 2 port for use in host(RC) mode. It doesn't enumerate
+ * the bus.
+ *
+ * @pcie_port: PCIe port to initialize
+ *
+ * Returns Zero on success.
+ */
+static int __cvmx_pcie_rc_initialize_gen2(int pcie_port)
+{
+	int i;
+	union cvmx_ciu_soft_prst ciu_soft_prst;
+	union cvmx_mio_rst_ctlx mio_rst_ctl;
+	union cvmx_pemx_bar_ctl pemx_bar_ctl;
+	union cvmx_pemx_ctl_status pemx_ctl_status;
+	union cvmx_pemx_bist_status pemx_bist_status;
+	union cvmx_pemx_bist_status2 pemx_bist_status2;
+	union cvmx_pciercx_cfg032 pciercx_cfg032;
+	union cvmx_pciercx_cfg515 pciercx_cfg515;
+	union cvmx_sli_ctl_portx sli_ctl_portx;
+	union cvmx_sli_mem_access_ctl sli_mem_access_ctl;
+	union cvmx_sli_mem_access_subidx mem_access_subid;
+	union cvmx_sriox_status_reg sriox_status_reg;
+	union cvmx_pemx_bar1_indexx bar1_index;
+
+	if (octeon_has_feature(OCTEON_FEATURE_SRIO)) {
+		/* Make sure this interface isn't SRIO */
+		if (OCTEON_IS_MODEL(OCTEON_CN66XX)) {
+			/*
+			 * The CN66XX requires reading the
+			 * MIO_QLMX_CFG register to figure out the
+			 * port type.
+			 */
+			union cvmx_mio_qlmx_cfg qlmx_cfg;
+			qlmx_cfg.u64 = cvmx_read_csr(CVMX_MIO_QLMX_CFG(pcie_port));
+
+			if (qlmx_cfg.s.qlm_spd == 15) {
+				pr_notice("PCIe: Port %d is disabled, skipping.\n", pcie_port);
+				return -1;
+			}
+
+			switch (qlmx_cfg.s.qlm_spd) {
+			case 0x1: /* SRIO 1x4 short */
+			case 0x3: /* SRIO 1x4 long */
+			case 0x4: /* SRIO 2x2 short */
+			case 0x6: /* SRIO 2x2 long */
+				pr_notice("PCIe: Port %d is SRIO, skipping.\n", pcie_port);
+				return -1;
+			case 0x9: /* SGMII */
+				pr_notice("PCIe: Port %d is SGMII, skipping.\n", pcie_port);
+				return -1;
+			case 0xb: /* XAUI */
+				pr_notice("PCIe: Port %d is XAUI, skipping.\n", pcie_port);
+				return -1;
+			case 0x0: /* PCIE gen2 */
+			case 0x8: /* PCIE gen2 (alias) */
+			case 0x2: /* PCIE gen1 */
+			case 0xa: /* PCIE gen1 (alias) */
+				break;
+			default:
+				pr_notice("PCIe: Port %d is unknown, skipping.\n", pcie_port);
+				return -1;
+			}
+		} else {
+			sriox_status_reg.u64 = cvmx_read_csr(CVMX_SRIOX_STATUS_REG(pcie_port));
+			if (sriox_status_reg.s.srio) {
+				pr_notice("PCIe: Port %d is SRIO, skipping.\n", pcie_port);
+				return -1;
+			}
+		}
+	}
+
+#if 0
+    /* This code is so that the PCIe analyzer is able to see 63XX traffic */
+	pr_notice("PCIE : init for pcie analyzer.\n");
+	cvmx_helper_qlm_jtag_init();
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 85);
+	cvmx_helper_qlm_jtag_shift(pcie_port, 1, 1);
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 300-86);
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 85);
+	cvmx_helper_qlm_jtag_shift(pcie_port, 1, 1);
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 300-86);
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 85);
+	cvmx_helper_qlm_jtag_shift(pcie_port, 1, 1);
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 300-86);
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 85);
+	cvmx_helper_qlm_jtag_shift(pcie_port, 1, 1);
+	cvmx_helper_qlm_jtag_shift_zeros(pcie_port, 300-86);
+	cvmx_helper_qlm_jtag_update(pcie_port);
+#endif
+
+	/* Make sure we aren't trying to setup a target mode interface in host mode */
+	mio_rst_ctl.u64 = cvmx_read_csr(CVMX_MIO_RST_CTLX(pcie_port));
+	if (!mio_rst_ctl.s.host_mode) {
+		pr_notice("PCIe: Port %d in endpoint mode.\n", pcie_port);
+		return -1;
+	}
+
+	/* CN63XX Pass 1.0 errata G-14395 requires the QLM De-emphasis be programmed */
+	if (OCTEON_IS_MODEL(OCTEON_CN63XX_PASS1_0)) {
+		if (pcie_port) {
+			union cvmx_ciu_qlm1 ciu_qlm;
+			ciu_qlm.u64 = cvmx_read_csr(CVMX_CIU_QLM1);
+			ciu_qlm.s.txbypass = 1;
+			ciu_qlm.s.txdeemph = 5;
+			ciu_qlm.s.txmargin = 0x17;
+			cvmx_write_csr(CVMX_CIU_QLM1, ciu_qlm.u64);
+		} else {
+			union cvmx_ciu_qlm0 ciu_qlm;
+			ciu_qlm.u64 = cvmx_read_csr(CVMX_CIU_QLM0);
+			ciu_qlm.s.txbypass = 1;
+			ciu_qlm.s.txdeemph = 5;
+			ciu_qlm.s.txmargin = 0x17;
+			cvmx_write_csr(CVMX_CIU_QLM0, ciu_qlm.u64);
+		}
+	}
+	/* Bring the PCIe out of reset */
+	if (pcie_port)
+		ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST1);
+	else
+		ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST);
+	/*
+	 * After a chip reset the PCIe will also be in reset. If it
+	 * isn't, most likely someone is trying to init it again
+	 * without a proper PCIe reset
+	 */
+	if (ciu_soft_prst.s.soft_prst == 0) {
+		/* Reset the port */
+		ciu_soft_prst.s.soft_prst = 1;
+		if (pcie_port)
+			cvmx_write_csr(CVMX_CIU_SOFT_PRST1, ciu_soft_prst.u64);
+		else
+			cvmx_write_csr(CVMX_CIU_SOFT_PRST, ciu_soft_prst.u64);
+		/* Wait until pcie resets the ports. */
+		udelay(2000);
+	}
+	if (pcie_port) {
+		ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST1);
+		ciu_soft_prst.s.soft_prst = 0;
+		cvmx_write_csr(CVMX_CIU_SOFT_PRST1, ciu_soft_prst.u64);
+	} else {
+		ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST);
+		ciu_soft_prst.s.soft_prst = 0;
+		cvmx_write_csr(CVMX_CIU_SOFT_PRST, ciu_soft_prst.u64);
+	}
+
+	/* Wait for PCIe reset to complete */
+	udelay(1000);
+
+	/*
+	 * Check and make sure PCIe came out of reset. If it doesn't
+	 * the board probably hasn't wired the clocks up and the
+	 * interface should be skipped.
+	 */
+	if (CVMX_WAIT_FOR_FIELD64(CVMX_MIO_RST_CTLX(pcie_port), union cvmx_mio_rst_ctlx, rst_done, ==, 1, 10000)) {
+		pr_notice("PCIe: Port %d stuck in reset, skipping.\n", pcie_port);
+		return -1;
+	}
+
+	/* Check BIST status */
+	pemx_bist_status.u64 = cvmx_read_csr(CVMX_PEMX_BIST_STATUS(pcie_port));
+	if (pemx_bist_status.u64)
+		pr_notice("PCIe: BIST FAILED for port %d (0x%016llx)\n", pcie_port, CAST64(pemx_bist_status.u64));
+	pemx_bist_status2.u64 = cvmx_read_csr(CVMX_PEMX_BIST_STATUS2(pcie_port));
+	/* Errata PCIE-14766 may cause the lower 6 bits to be randomly set on CN63XXp1 */
+	if (OCTEON_IS_MODEL(OCTEON_CN63XX_PASS1_X))
+		pemx_bist_status2.u64 &= ~0x3full;
+	if (pemx_bist_status2.u64)
+		pr_notice("PCIe: BIST2 FAILED for port %d (0x%016llx)\n", pcie_port, CAST64(pemx_bist_status2.u64));
+
+	/* Initialize the config space CSRs */
+	__cvmx_pcie_rc_initialize_config_space(pcie_port);
+
+	/* Enable gen2 speed selection */
+	pciercx_cfg515.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG515(pcie_port));
+	pciercx_cfg515.s.dsc = 1;
+	cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG515(pcie_port), pciercx_cfg515.u32);
+
+	/* Bring the link up */
+	if (__cvmx_pcie_rc_initialize_link_gen2(pcie_port)) {
+		/*
+		 * Some gen1 devices don't handle the gen 2 training
+		 * correctly. Disable gen2 and try again with only
+		 * gen1
+		 */
+		union cvmx_pciercx_cfg031 pciercx_cfg031;
+		pciercx_cfg031.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG031(pcie_port));
+		pciercx_cfg031.s.mls = 1;
+		cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG031(pcie_port), pciercx_cfg031.u32);
+		if (__cvmx_pcie_rc_initialize_link_gen2(pcie_port)) {
+			pr_notice("PCIe: Link timeout on port %d, probably the slot is empty\n", pcie_port);
+			return -1;
+		}
+	}
+
+	/* Store merge control (SLI_MEM_ACCESS_CTL[TIMER,MAX_WORD]) */
+	sli_mem_access_ctl.u64 = cvmx_read_csr(CVMX_PEXP_SLI_MEM_ACCESS_CTL);
+	sli_mem_access_ctl.s.max_word = 0;	/* Allow 16 words to combine */
+	sli_mem_access_ctl.s.timer = 127;	/* Wait up to 127 cycles for more data */
+	cvmx_write_csr(CVMX_PEXP_SLI_MEM_ACCESS_CTL, sli_mem_access_ctl.u64);
+
+	/* Setup Mem access SubDIDs */
+	mem_access_subid.u64 = 0;
+	mem_access_subid.s.port = pcie_port; /* Port the request is sent to. */
+	mem_access_subid.s.nmerge = 0;  /* Allow merging as it works on CN6XXX. */
+	mem_access_subid.s.esr = 1;     /* Endian-swap for Reads. */
+	mem_access_subid.s.esw = 1;     /* Endian-swap for Writes. */
+	mem_access_subid.s.wtype = 0;   /* "No snoop" and "Relaxed ordering" are not set */
+	mem_access_subid.s.rtype = 0;   /* "No snoop" and "Relaxed ordering" are not set */
+	/* PCIe Adddress Bits <63:34>. */
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+		mem_access_subid.cn68xx.ba = 0;
+	else
+		mem_access_subid.cn63xx.ba = 0;
+
+	/*
+	 * Setup mem access 12-15 for port 0, 16-19 for port 1,
+	 * supplying 36 bits of address space.
+	 */
+	for (i = 12 + pcie_port * 4; i < 16 + pcie_port * 4; i++) {
+		cvmx_write_csr(CVMX_PEXP_SLI_MEM_ACCESS_SUBIDX(i), mem_access_subid.u64);
+		/* Set each SUBID to extend the addressable range */
+		__cvmx_increment_ba(&mem_access_subid);
+	}
+
+	/*
+	 * Disable the peer to peer forwarding register. This must be
+	 * setup by the OS after it enumerates the bus and assigns
+	 * addresses to the PCIe busses.
+	 */
+	for (i = 0; i < 4; i++) {
+		cvmx_write_csr(CVMX_PEMX_P2P_BARX_START(i, pcie_port), -1);
+		cvmx_write_csr(CVMX_PEMX_P2P_BARX_END(i, pcie_port), -1);
+	}
+
+	/* Set Octeon's BAR0 to decode 0-16KB. It overlaps with Bar2 */
+	cvmx_write_csr(CVMX_PEMX_P2N_BAR0_START(pcie_port), 0);
+
+	/*
+	 * Set Octeon's BAR2 to decode 0-2^41. Bar0 and Bar1 take
+	 * precedence where they overlap. It also overlaps with the
+	 * device addresses, so make sure the peer to peer forwarding
+	 * is set right.
+	 */
+	cvmx_write_csr(CVMX_PEMX_P2N_BAR2_START(pcie_port), 0);
+
+	/*
+	 * Setup BAR2 attributes
+	 * Relaxed Ordering (NPEI_CTL_PORTn[PTLP_RO,CTLP_RO, WAIT_COM])
+	 * - PTLP_RO,CTLP_RO should normally be set (except for debug).
+	 * - WAIT_COM=0 will likely work for all applications.
+	 * Load completion relaxed ordering (NPEI_CTL_PORTn[WAITL_COM])
+	 */
+	pemx_bar_ctl.u64 = cvmx_read_csr(CVMX_PEMX_BAR_CTL(pcie_port));
+	pemx_bar_ctl.s.bar1_siz = 3;  /* 256MB BAR1*/
+	pemx_bar_ctl.s.bar2_enb = 1;
+	pemx_bar_ctl.s.bar2_esx = 1;
+	pemx_bar_ctl.s.bar2_cax = 0;
+	cvmx_write_csr(CVMX_PEMX_BAR_CTL(pcie_port), pemx_bar_ctl.u64);
+	sli_ctl_portx.u64 = cvmx_read_csr(CVMX_PEXP_SLI_CTL_PORTX(pcie_port));
+	sli_ctl_portx.s.ptlp_ro = 1;
+	sli_ctl_portx.s.ctlp_ro = 1;
+	sli_ctl_portx.s.wait_com = 0;
+	sli_ctl_portx.s.waitl_com = 0;
+	cvmx_write_csr(CVMX_PEXP_SLI_CTL_PORTX(pcie_port), sli_ctl_portx.u64);
+
+	/* BAR1 follows BAR2 */
+	cvmx_write_csr(CVMX_PEMX_P2N_BAR1_START(pcie_port), CVMX_PCIE_BAR1_RC_BASE);
+
+	bar1_index.u64 = 0;
+	bar1_index.s.addr_idx = (CVMX_PCIE_BAR1_PHYS_BASE >> 22);
+	bar1_index.s.ca = 1;       /* Not Cached */
+	bar1_index.s.end_swp = 1;  /* Endian Swap mode */
+	bar1_index.s.addr_v = 1;   /* Valid entry */
+
+	for (i = 0; i < 16; i++) {
+		cvmx_write_csr(CVMX_PEMX_BAR1_INDEXX(i, pcie_port), bar1_index.u64);
+		/* 256MB / 16 >> 22 == 4 */
+		bar1_index.s.addr_idx += (((1ull << 28) / 16ull) >> 22);
+	}
+
+	/*
+	 * Allow config retries for 250ms. Count is based off the 5Ghz
+	 * SERDES clock.
+	 */
+	pemx_ctl_status.u64 = cvmx_read_csr(CVMX_PEMX_CTL_STATUS(pcie_port));
+	pemx_ctl_status.s.cfg_rtry = 250 * 5000000 / 0x10000;
+	cvmx_write_csr(CVMX_PEMX_CTL_STATUS(pcie_port), pemx_ctl_status.u64);
+
+	/* Display the link status */
+	pciercx_cfg032.u32 = cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG032(pcie_port));
+	pr_notice("PCIe: Port %d link active, %d lanes, speed gen%d\n", pcie_port, pciercx_cfg032.s.nlw, pciercx_cfg032.s.ls);
+
+	return 0;
+}
+
+/**
+ * Initialize a PCIe port for use in host(RC) mode. It doesn't enumerate the bus.
+ *
+ * @pcie_port: PCIe port to initialize
+ *
+ * Returns Zero on success
+ */
+static int cvmx_pcie_rc_initialize(int pcie_port)
+{
+	int result;
+	if (octeon_has_feature(OCTEON_FEATURE_NPEI))
+		result = __cvmx_pcie_rc_initialize_gen1(pcie_port);
+	else
+		result = __cvmx_pcie_rc_initialize_gen2(pcie_port);
+	return result;
+}
+
+/* Above was cvmx-pcie.c, below original pcie.c */
+
+/**
  * Map a PCI device to the appropriate interrupt line
  *
  * @dev:    The Linux PCI device structure for the device to map
@@ -1027,11 +1481,12 @@ int __init octeon_pcie_pcibios_map_irq(const struct pci_dev *dev,
 		 */
 		while (dev->bus && dev->bus->parent)
 			dev = to_pci_dev(dev->bus->bridge);
-		/* If the root bus is number 0 and the PEX 8114 is the
+		/*
+		 * If the root bus is number 0 and the PEX 8114 is the
 		 * root, assume we are behind the miswired bus. We
 		 * need to correct the swizzle level by two. Yuck.
 		 */
-		if ((dev->bus->number == 0) &&
+		if ((dev->bus->number == 1) &&
 		    (dev->vendor == 0x10b5) && (dev->device == 0x8114)) {
 			/*
 			 * The pin field is one based, not zero. We
@@ -1048,39 +1503,73 @@ int __init octeon_pcie_pcibios_map_irq(const struct pci_dev *dev,
 	return pin - 1 + OCTEON_IRQ_PCI_INT0;
 }
 
-/**
+static  void set_cfg_read_retry(u32 retry_cnt)
+{
+	union cvmx_pemx_ctl_status pemx_ctl;
+	pemx_ctl.u64 = cvmx_read_csr(CVMX_PEMX_CTL_STATUS(1));
+	pemx_ctl.s.cfg_rtry = retry_cnt;
+	cvmx_write_csr(CVMX_PEMX_CTL_STATUS(1), pemx_ctl.u64);
+}
+
+
+static u32 disable_cfg_read_retry(void)
+{
+	u32 retry_cnt;
+
+	union cvmx_pemx_ctl_status pemx_ctl;
+	pemx_ctl.u64 = cvmx_read_csr(CVMX_PEMX_CTL_STATUS(1));
+	retry_cnt =  pemx_ctl.s.cfg_rtry;
+	pemx_ctl.s.cfg_rtry = 0;
+	cvmx_write_csr(CVMX_PEMX_CTL_STATUS(1), pemx_ctl.u64);
+	return retry_cnt;
+}
+
+static int is_cfg_retry(void)
+{
+	union cvmx_pemx_int_sum pemx_int_sum;
+	pemx_int_sum.u64 = cvmx_read_csr(CVMX_PEMX_INT_SUM(1));
+	if (pemx_int_sum.s.crs_dr)
+		return 1;
+	return 0;
+}
+
+/*
  * Read a value from configuration space
  *
- * @bus:
- * @devfn:
- * @reg:
- * @size:
- * @val:
- * Returns
  */
-static inline int octeon_pcie_read_config(int pcie_port, struct pci_bus *bus,
-					  unsigned int devfn, int reg, int size,
-					  u32 *val)
+static int octeon_pcie_read_config(unsigned int pcie_port, struct pci_bus *bus,
+				   unsigned int devfn, int reg, int size,
+				   u32 *val)
 {
 	union octeon_cvmemctl cvmmemctl;
 	union octeon_cvmemctl cvmmemctl_save;
 	int bus_number = bus->number;
+	int cfg_retry = 0;
+	int retry_cnt = 0;
+	int max_retry_cnt = 10;
+	u32 cfg_retry_cnt = 0;
 
+	cvmmemctl_save.u64 = 0;
+	BUG_ON(pcie_port >= ARRAY_SIZE(enable_pcie_bus_num_war));
 	/*
 	 * For the top level bus make sure our hardware bus number
-	 * matches the software one.
+	 * matches the software one
 	 */
 	if (bus->parent == NULL) {
-		union cvmx_pciercx_cfg006 pciercx_cfg006;
-		pciercx_cfg006.u32 = cvmx_pcie_cfgx_read(pcie_port,
-			CVMX_PCIERCX_CFG006(pcie_port));
-		if (pciercx_cfg006.s.pbnum != bus_number) {
-			pciercx_cfg006.s.pbnum = bus_number;
-			pciercx_cfg006.s.sbnum = bus_number;
-			pciercx_cfg006.s.subbnum = bus_number;
-			cvmx_pcie_cfgx_write(pcie_port,
-				CVMX_PCIERCX_CFG006(pcie_port),
-				pciercx_cfg006.u32);
+		if (enable_pcie_bus_num_war[pcie_port])
+			bus_number = 0;
+		else {
+			union cvmx_pciercx_cfg006 pciercx_cfg006;
+			pciercx_cfg006.u32 = cvmx_pcie_cfgx_read(pcie_port,
+					     CVMX_PCIERCX_CFG006(pcie_port));
+			if (pciercx_cfg006.s.pbnum != bus_number) {
+				pciercx_cfg006.s.pbnum = bus_number;
+				pciercx_cfg006.s.sbnum = bus_number;
+				pciercx_cfg006.s.subbnum = bus_number;
+				cvmx_pcie_cfgx_write(pcie_port,
+					    CVMX_PCIERCX_CFG006(pcie_port),
+					    pciercx_cfg006.u32);
+			}
 		}
 	}
 
@@ -1116,29 +1605,52 @@ static inline int octeon_pcie_read_config(int pcie_port, struct pci_bus *bus,
 		 */
 #if 1
 		/* Use this option if you aren't using either slot */
-		if (bus_number == 1)
+		if (bus_number == 2)
 			return PCIBIOS_FUNC_NOT_SUPPORTED;
 #elif 0
 		/*
 		 * Use this option if you are using the first slot but
 		 * not the second.
 		 */
-		if ((bus_number == 1) && (devfn >> 3 != 2))
+		if ((bus_number == 2) && (devfn >> 3 != 2))
 			return PCIBIOS_FUNC_NOT_SUPPORTED;
 #elif 0
 		/*
 		 * Use this option if you are using the second slot
 		 * but not the first.
 		 */
-		if ((bus_number == 1) && (devfn >> 3 != 3))
+		if ((bus_number == 2) && (devfn >> 3 != 3))
 			return PCIBIOS_FUNC_NOT_SUPPORTED;
 #elif 0
 		/* Use this opion if you are using both slots */
-		if ((bus_number == 1) &&
+		if ((bus_number == 2) &&
 		    !((devfn == (2 << 3)) || (devfn == (3 << 3))))
 			return PCIBIOS_FUNC_NOT_SUPPORTED;
 #endif
 
+		/* The following #if gives a more complicated example. This is
+		   the required checks for running a Nitrox CN16XX-NHBX in the
+		   slot of the EBH5600. This card has a PLX PCIe bridge with
+		   four Nitrox PLX parts behind it */
+#if 0
+		/* PLX bridge with 4 ports */
+		if ((bus_number == 4) &&
+		    !((devfn >> 3 >= 1) && (devfn >> 3 <= 4)))
+			return PCIBIOS_FUNC_NOT_SUPPORTED;
+		/* Nitrox behind PLX 1 */
+		if ((bus_number == 5) && (devfn >> 3 != 0))
+			return PCIBIOS_FUNC_NOT_SUPPORTED;
+		/* Nitrox behind PLX 2 */
+		if ((bus_number == 6) && (devfn >> 3 != 0))
+			return PCIBIOS_FUNC_NOT_SUPPORTED;
+		/* Nitrox behind PLX 3 */
+		if ((bus_number == 7) && (devfn >> 3 != 0))
+			return PCIBIOS_FUNC_NOT_SUPPORTED;
+		/* Nitrox behind PLX 4 */
+		if ((bus_number == 8) && (devfn >> 3 != 0))
+			return PCIBIOS_FUNC_NOT_SUPPORTED;
+#endif
+
 		/*
 		 * Shorten the DID timeout so bus errors for PCIe
 		 * config reads from non existent devices happen
@@ -1152,26 +1664,48 @@ static inline int octeon_pcie_read_config(int pcie_port, struct pci_bus *bus,
 		__write_64bit_c0_register($11, 7, cvmmemctl.u64);
 	}
 
-	switch (size) {
-	case 4:
-		*val = cvmx_pcie_config_read32(pcie_port, bus_number,
-					       devfn >> 3, devfn & 0x7, reg);
+	if ((OCTEON_IS_MODEL(OCTEON_CN63XX)) && (enable_pcie_14459_war))
+		cfg_retry_cnt = disable_cfg_read_retry();
+
+	pr_debug("pcie_cfg_rd port=%d b=%d devfn=0x%03x reg=0x%03x"
+		 " size=%d ", pcie_port, bus_number, devfn, reg, size);
+	do {
+		switch (size) {
+		case 4:
+			*val = cvmx_pcie_config_read32(pcie_port, bus_number,
+				devfn >> 3, devfn & 0x7, reg);
 		break;
-	case 2:
-		*val = cvmx_pcie_config_read16(pcie_port, bus_number,
-					       devfn >> 3, devfn & 0x7, reg);
+		case 2:
+			*val = cvmx_pcie_config_read16(pcie_port, bus_number,
+				devfn >> 3, devfn & 0x7, reg);
 		break;
-	case 1:
-		*val = cvmx_pcie_config_read8(pcie_port, bus_number, devfn >> 3,
-					      devfn & 0x7, reg);
+		case 1:
+			*val = cvmx_pcie_config_read8(pcie_port, bus_number,
+				devfn >> 3, devfn & 0x7, reg);
 		break;
-	default:
-		return PCIBIOS_FUNC_NOT_SUPPORTED;
-	}
+		default:
+			if (OCTEON_IS_MODEL(OCTEON_CN63XX))
+				set_cfg_read_retry(cfg_retry_cnt);
+			return PCIBIOS_FUNC_NOT_SUPPORTED;
+		}
+		if ((OCTEON_IS_MODEL(OCTEON_CN63XX)) &&
+			(enable_pcie_14459_war)) {
+			cfg_retry = is_cfg_retry();
+			retry_cnt++;
+			if (retry_cnt > max_retry_cnt) {
+				pr_err(" pcie cfg_read retries failed. retry_cnt=%d\n",
+				       retry_cnt);
+				cfg_retry = 0;
+			}
+		}
+	} while (cfg_retry);
 
+	if ((OCTEON_IS_MODEL(OCTEON_CN63XX)) && (enable_pcie_14459_war))
+		set_cfg_read_retry(cfg_retry_cnt);
+	pr_debug("val=%08x  : tries=%02d\n", *val, retry_cnt);
 	if (OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1) ||
 	    OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_1))
-		__write_64bit_c0_register($11, 7, cvmmemctl_save.u64);
+		write_c0_cvmmemctl(cvmmemctl_save.u64);
 	return PCIBIOS_SUCCESSFUL;
 }
 
@@ -1187,42 +1721,56 @@ static int octeon_pcie1_read_config(struct pci_bus *bus, unsigned int devfn,
 	return octeon_pcie_read_config(1, bus, devfn, reg, size, val);
 }
 
+static int octeon_dummy_read_config(struct pci_bus *bus, unsigned int devfn,
+				    int reg, int size, u32 *val)
+{
+	return PCIBIOS_FUNC_NOT_SUPPORTED;
+}
 
-
-/**
+/*
  * Write a value to PCI configuration space
- *
- * @bus:
- * @devfn:
- * @reg:
- * @size:
- * @val:
- * Returns
  */
-static inline int octeon_pcie_write_config(int pcie_port, struct pci_bus *bus,
-					   unsigned int devfn, int reg,
-					   int size, u32 val)
+static int octeon_pcie_write_config(unsigned int pcie_port, struct pci_bus *bus,
+				    unsigned int devfn, int reg,
+				    int size, u32 val)
 {
 	int bus_number = bus->number;
 
+	BUG_ON(pcie_port >= ARRAY_SIZE(enable_pcie_bus_num_war));
+
+	if ((bus->parent == NULL) && (enable_pcie_bus_num_war[pcie_port]))
+		bus_number = 0;
+
+	pr_debug("pcie_cfg_wr port=%d b=%d devfn=0x%03x"
+		 " reg=0x%03x size=%d val=%08x\n", pcie_port, bus_number, devfn,
+		 reg, size, val);
+
+
 	switch (size) {
 	case 4:
 		cvmx_pcie_config_write32(pcie_port, bus_number, devfn >> 3,
 					 devfn & 0x7, reg, val);
-		return PCIBIOS_SUCCESSFUL;
+		break;
 	case 2:
 		cvmx_pcie_config_write16(pcie_port, bus_number, devfn >> 3,
 					 devfn & 0x7, reg, val);
-		return PCIBIOS_SUCCESSFUL;
+		break;
 	case 1:
 		cvmx_pcie_config_write8(pcie_port, bus_number, devfn >> 3,
 					devfn & 0x7, reg, val);
-		return PCIBIOS_SUCCESSFUL;
+		break;
+	default:
+		return PCIBIOS_FUNC_NOT_SUPPORTED;
 	}
 #if PCI_CONFIG_SPACE_DELAY
+	/*
+	 * Delay on writes so that devices have time to come up. Some
+	 * bridges need this to allow time for the secondary busses to
+	 * work
+	 */
 	udelay(PCI_CONFIG_SPACE_DELAY);
 #endif
-	return PCIBIOS_FUNC_NOT_SUPPORTED;
+	return PCIBIOS_SUCCESSFUL;
 }
 
 static int octeon_pcie0_write_config(struct pci_bus *bus, unsigned int devfn,
@@ -1237,6 +1785,12 @@ static int octeon_pcie1_write_config(struct pci_bus *bus, unsigned int devfn,
 	return octeon_pcie_write_config(1, bus, devfn, reg, size, val);
 }
 
+static int octeon_dummy_write_config(struct pci_bus *bus, unsigned int devfn,
+				     int reg, int size, u32 val)
+{
+	return PCIBIOS_FUNC_NOT_SUPPORTED;
+}
+
 static struct pci_ops octeon_pcie0_ops = {
 	octeon_pcie0_read_config,
 	octeon_pcie0_write_config,
@@ -1279,6 +1833,35 @@ static struct pci_controller octeon_pcie1_controller = {
 	.io_resource = &octeon_pcie1_io_resource,
 };
 
+static struct pci_ops octeon_dummy_ops = {
+	octeon_dummy_read_config,
+	octeon_dummy_write_config,
+};
+
+static struct resource octeon_dummy_mem_resource = {
+	.name = "Virtual PCIe MEM",
+	.flags = IORESOURCE_MEM,
+};
+
+static struct resource octeon_dummy_io_resource = {
+	.name = "Virtual PCIe IO",
+	.flags = IORESOURCE_IO,
+};
+
+static struct pci_controller octeon_dummy_controller = {
+	.pci_ops = &octeon_dummy_ops,
+	.mem_resource = &octeon_dummy_mem_resource,
+	.io_resource = &octeon_dummy_io_resource,
+};
+
+static int device_needs_bus_num_war(uint32_t deviceid)
+{
+#define IDT_VENDOR_ID 0x111d
+
+	if ((deviceid  & 0xffff) == IDT_VENDOR_ID)
+		return 1;
+	return 0;
+}
 
 /**
  * Initialize the Octeon PCIe controllers
@@ -1287,19 +1870,27 @@ static struct pci_controller octeon_pcie1_controller = {
  */
 static int __init octeon_pcie_setup(void)
 {
-	union cvmx_npei_ctl_status npei_ctl_status;
 	int result;
+	int host_mode;
+	int srio_war15205 = 0, port;
+	union cvmx_sli_ctl_portx sli_ctl_portx;
+	union cvmx_sriox_status_reg sriox_status_reg;
 
 	/* These chips don't have PCIe */
 	if (!octeon_has_feature(OCTEON_FEATURE_PCIE))
 		return 0;
 
+	/* No PCIe simulation */
+	if (octeon_is_simulation())
+		return 0;
+
+	/* Disable PCI if instructed on the command line */
+	if (pcie_disable)
+		return 0;
+
 	/* Point pcibios_map_irq() to the PCIe version of it */
 	octeon_pcibios_map_irq = octeon_pcie_pcibios_map_irq;
 
-	/* Use the PCIe based DMA mappings */
-	octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_PCIE;
-
 	/*
 	 * PCIe I/O range. It is based on port 0 but includes up until
 	 * port 1's end.
@@ -1310,11 +1901,43 @@ static int __init octeon_pcie_setup(void)
 		cvmx_pcie_get_io_base_address(1) -
 		cvmx_pcie_get_io_base_address(0) + cvmx_pcie_get_io_size(1) - 1;
 
-	npei_ctl_status.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS);
-	if (npei_ctl_status.s.host_mode) {
+	/*
+	 * Create a dummy PCIe controller to swallow up bus 0. IDT bridges
+	 * don't work if the primary bus number is zero. Here we add a fake
+	 * PCIe controller that the kernel will give bus 0. This allows
+	 * us to not change the normal kernel bus enumeration
+	 */
+	octeon_dummy_controller.io_map_base = -1;
+	octeon_dummy_controller.mem_resource->start = (1ull<<48);
+	octeon_dummy_controller.mem_resource->end = (1ull<<48);
+	register_pci_controller(&octeon_dummy_controller);
+
+	if (octeon_has_feature(OCTEON_FEATURE_NPEI)) {
+		union cvmx_npei_ctl_status npei_ctl_status;
+		npei_ctl_status.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS);
+		host_mode = npei_ctl_status.s.host_mode;
+		octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_PCIE;
+	} else {
+		union cvmx_mio_rst_ctlx mio_rst_ctl;
+		mio_rst_ctl.u64 = cvmx_read_csr(CVMX_MIO_RST_CTLX(0));
+		host_mode = mio_rst_ctl.s.host_mode;
+		octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_PCIE2;
+	}
+
+	if (host_mode) {
 		pr_notice("PCIe: Initializing port 0\n");
+		/* CN63XX pass 1_x/2.0 errata PCIe-15205 */
+		if (OCTEON_IS_MODEL(OCTEON_CN63XX_PASS1_X) ||
+			OCTEON_IS_MODEL(OCTEON_CN63XX_PASS2_0)) {
+			sriox_status_reg.u64 = cvmx_read_csr(CVMX_SRIOX_STATUS_REG(0));
+			if (sriox_status_reg.s.srio) {
+				srio_war15205 += 1;      /* Port is SRIO */
+				port = 0;
+			}
+		}
 		result = cvmx_pcie_rc_initialize(0);
 		if (result == 0) {
+			uint32_t device0;
 			/* Memory offsets are physical addresses */
 			octeon_pcie0_controller.mem_offset =
 				cvmx_pcie_get_mem_base_address(0);
@@ -1343,60 +1966,134 @@ static int __init octeon_pcie_setup(void)
 			octeon_pcie0_controller.io_resource->start = 4 << 10;
 			octeon_pcie0_controller.io_resource->end =
 				cvmx_pcie_get_io_size(0) - 1;
+			msleep(100); /* Some devices need extra time */
 			register_pci_controller(&octeon_pcie0_controller);
+			device0 = cvmx_pcie_config_read32(0, 0, 0, 0, 0);
+			enable_pcie_bus_num_war[0] =
+				device_needs_bus_num_war(device0);
 		}
 	} else {
 		pr_notice("PCIe: Port 0 in endpoint mode, skipping.\n");
+		/* CN63XX pass 1_x/2.0 errata PCIe-15205 */
+		if (OCTEON_IS_MODEL(OCTEON_CN63XX_PASS1_X) ||
+			OCTEON_IS_MODEL(OCTEON_CN63XX_PASS2_0)) {
+			srio_war15205 += 1;
+			port = 0;
+		}
 	}
 
-	/* Skip the 2nd port on CN52XX if port 0 is in 4 lane mode */
-	if (OCTEON_IS_MODEL(OCTEON_CN52XX)) {
-		union cvmx_npei_dbg_data npei_dbg_data;
-		npei_dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA);
-		if (npei_dbg_data.cn52xx.qlm0_link_width)
-			return 0;
+	if (octeon_has_feature(OCTEON_FEATURE_NPEI)) {
+		host_mode = 1;
+		/* Skip the 2nd port on CN52XX if port 0 is in 4 lane mode */
+		if (OCTEON_IS_MODEL(OCTEON_CN52XX)) {
+			union cvmx_npei_dbg_data dbg_data;
+			dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA);
+			if (dbg_data.cn52xx.qlm0_link_width)
+				host_mode = 0;
+		}
+	} else {
+		union cvmx_mio_rst_ctlx mio_rst_ctl;
+		mio_rst_ctl.u64 = cvmx_read_csr(CVMX_MIO_RST_CTLX(1));
+		host_mode = mio_rst_ctl.s.host_mode;
 	}
 
-	pr_notice("PCIe: Initializing port 1\n");
-	result = cvmx_pcie_rc_initialize(1);
-	if (result == 0) {
-		/* Memory offsets are physical addresses */
-		octeon_pcie1_controller.mem_offset =
-			cvmx_pcie_get_mem_base_address(1);
-		/* IO offsets are Mips virtual addresses */
-		octeon_pcie1_controller.io_map_base =
-			CVMX_ADD_IO_SEG(cvmx_pcie_get_io_base_address(1));
-		octeon_pcie1_controller.io_offset =
-			cvmx_pcie_get_io_base_address(1) -
-			cvmx_pcie_get_io_base_address(0);
-		/*
-		 * To keep things similar to PCI, we start device
-		 * addresses at the same place as PCI uisng big bar
-		 * support. This normally translates to 4GB-256MB,
-		 * which is the same as most x86 PCs.
-		 */
-		octeon_pcie1_controller.mem_resource->start =
-			cvmx_pcie_get_mem_base_address(1) + (4ul << 30) -
-			(OCTEON_PCI_BAR1_HOLE_SIZE << 20);
-		octeon_pcie1_controller.mem_resource->end =
-			cvmx_pcie_get_mem_base_address(1) +
-			cvmx_pcie_get_mem_size(1) - 1;
-		/*
-		 * Ports must be above 16KB for the ISA bus filtering
-		 * in the PCI-X to PCI bridge.
-		 */
-		octeon_pcie1_controller.io_resource->start =
-			cvmx_pcie_get_io_base_address(1) -
-			cvmx_pcie_get_io_base_address(0);
-		octeon_pcie1_controller.io_resource->end =
-			octeon_pcie1_controller.io_resource->start +
-			cvmx_pcie_get_io_size(1) - 1;
-		register_pci_controller(&octeon_pcie1_controller);
+	if (host_mode) {
+		pr_notice("PCIe: Initializing port 1\n");
+		/* CN63XX pass 1_x/2.0 errata PCIe-15205 */
+		if (OCTEON_IS_MODEL(OCTEON_CN63XX_PASS1_X) ||
+			OCTEON_IS_MODEL(OCTEON_CN63XX_PASS2_0)) {
+			sriox_status_reg.u64 = cvmx_read_csr(CVMX_SRIOX_STATUS_REG(1));
+			if (sriox_status_reg.s.srio) {
+				srio_war15205 += 1;      /* Port is SRIO */
+				port = 1;
+			}
+		}
+		result = cvmx_pcie_rc_initialize(1);
+		if (result == 0) {
+			uint32_t device0;
+			/* Memory offsets are physical addresses */
+			octeon_pcie1_controller.mem_offset =
+				cvmx_pcie_get_mem_base_address(1);
+			/*
+			 * To calculate the address for accessing the 2nd PCIe device,
+			 * either 'io_map_base' (pci_iomap()), or 'mips_io_port_base'
+			 * (ioport_map()) value is added to
+			 * pci_resource_start(dev,bar)). The 'mips_io_port_base' is set
+			 * only once based on first PCIe. Also changing 'io_map_base'
+			 * based on first slot's value so that both the routines will
+			 * work properly.
+			 */
+			octeon_pcie1_controller.io_map_base =
+				CVMX_ADD_IO_SEG(cvmx_pcie_get_io_base_address(0));
+			/* IO offsets are Mips virtual addresses */
+			octeon_pcie1_controller.io_offset =
+				cvmx_pcie_get_io_base_address(1) -
+				cvmx_pcie_get_io_base_address(0);
+			/*
+			 * To keep things similar to PCI, we start device
+			 * addresses at the same place as PCI uisng big bar
+			 * support. This normally translates to 4GB-256MB,
+			 * which is the same as most x86 PCs.
+			 */
+			octeon_pcie1_controller.mem_resource->start =
+				cvmx_pcie_get_mem_base_address(1) + (4ul << 30) -
+				(OCTEON_PCI_BAR1_HOLE_SIZE << 20);
+			octeon_pcie1_controller.mem_resource->end =
+				cvmx_pcie_get_mem_base_address(1) +
+				cvmx_pcie_get_mem_size(1) - 1;
+			/*
+			 * Ports must be above 16KB for the ISA bus filtering
+			 * in the PCI-X to PCI bridge.
+			 */
+			octeon_pcie1_controller.io_resource->start =
+				cvmx_pcie_get_io_base_address(1) -
+				cvmx_pcie_get_io_base_address(0);
+			octeon_pcie1_controller.io_resource->end =
+				octeon_pcie1_controller.io_resource->start +
+				cvmx_pcie_get_io_size(1) - 1;
+			msleep(100); /* Some devices need extra time */
+			register_pci_controller(&octeon_pcie1_controller);
+			device0 = cvmx_pcie_config_read32(1, 0, 0, 0, 0);
+			enable_pcie_bus_num_war[1] =
+				device_needs_bus_num_war(device0);
+		}
+	} else {
+		pr_notice("PCIe: Port 1 not in root complex mode, skipping.\n");
+		/* CN63XX pass 1_x/2.0 errata PCIe-15205  */
+		if (OCTEON_IS_MODEL(OCTEON_CN63XX_PASS1_X) ||
+			OCTEON_IS_MODEL(OCTEON_CN63XX_PASS2_0)) {
+			srio_war15205 += 1;
+			port = 1;
+		}
+	}
+
+	/*
+	 * CN63XX pass 1_x/2.0 errata PCIe-15205 requires setting all
+	 * of SRIO MACs SLI_CTL_PORT*[INT*_MAP] to similar value and
+	 * all of PCIe Macs SLI_CTL_PORT*[INT*_MAP] to different value
+	 * from the previous set values
+	 */
+	if (OCTEON_IS_MODEL(OCTEON_CN63XX_PASS1_X) ||
+		OCTEON_IS_MODEL(OCTEON_CN63XX_PASS2_0)) {
+		if (srio_war15205 == 1) {
+			sli_ctl_portx.u64 = cvmx_read_csr(CVMX_PEXP_SLI_CTL_PORTX(port));
+			sli_ctl_portx.s.inta_map = 1;
+			sli_ctl_portx.s.intb_map = 1;
+			sli_ctl_portx.s.intc_map = 1;
+			sli_ctl_portx.s.intd_map = 1;
+			cvmx_write_csr(CVMX_PEXP_SLI_CTL_PORTX(port), sli_ctl_portx.u64);
+
+			sli_ctl_portx.u64 = cvmx_read_csr(CVMX_PEXP_SLI_CTL_PORTX(!port));
+			sli_ctl_portx.s.inta_map = 0;
+			sli_ctl_portx.s.intb_map = 0;
+			sli_ctl_portx.s.intc_map = 0;
+			sli_ctl_portx.s.intd_map = 0;
+			cvmx_write_csr(CVMX_PEXP_SLI_CTL_PORTX(!port), sli_ctl_portx.u64);
+		}
 	}
 
 	octeon_pci_dma_init();
 
 	return 0;
 }
-
 arch_initcall(octeon_pcie_setup);