Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-next

Pull sparc updates from David Miller:

 1) Hibernation support, as well as removal of excess interrupt
    twiddling in MMU context allocation on sparc64 from Kirill Tkhai.

 2) Kill references to __ARCH_WANT_UNLOCKED_CTXSW.

 3) Sparc32 LEON bug fixes from Daniel Hellstrom and Andreas Larsson.

 4) Provide cmpxchg64(), from Geert Uytterhoeven.

 5) Device refcount and registry bug fixes from Federico Vaga and Wei
    Yongjun.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-next:
  serial: sunsu: add missing platform_driver_unregister() when module exit
  sparc32, leon: Do not overwrite previously set irq flow handlers
  sparc/kernel/vio.c: add put_device() after device_find_child()
  sparc64: Do not save/restore interrupts in get_new_mmu_context()
  sparc: Consistently use 'wr' and 'rd' instructions for ASRs.
  sparc64: Kill __ARCH_WANT_UNLOCKED_CTXSW
  sparc64: Provide cmpxchg64()
  sparc64: Do not change num_physpages during initmem freeing
  sparc64: Hibernation support
  sparc,leon: updated GRPCI2 config name
  sparc,leon: support for GRPCI1 PCI host bridge controller
  sparc32,leon: add support for PCI busn resource for GRPCI2

20 files changed, 985 insertions, 19 deletions

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index f5041d741dea..a639c0d07b8b 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -99,6 +99,9 @@ config HAVE_LATENCYTOP_SUPPORT
 	bool
 	default y if SPARC64
 
+config ARCH_HIBERNATION_POSSIBLE
+	def_bool y if SPARC64
+
 config AUDIT_ARCH
 	bool
 	default y
@@ -303,6 +306,10 @@ config ARCH_SPARSEMEM_DEFAULT
 
 source "mm/Kconfig"
 
+if SPARC64
+source "kernel/power/Kconfig"
+endif
+
 config SCHED_SMT
 	bool "SMT (Hyperthreading) scheduler support"
 	depends on SPARC64 && SMP
@@ -472,7 +479,18 @@ config LEON_PCI
 	depends on PCI && SPARC_LEON
 	default y
 
-config GRPCI2
+config SPARC_GRPCI1
+	bool "GRPCI Host Bridge Support"
+	depends on LEON_PCI
+	default y
+	help
+	  Say Y here to include the GRPCI Host Bridge Driver. The GRPCI
+	  PCI host controller is typically found in GRLIB SPARC32/LEON
+	  systems. The driver has one property (all_pci_errors) controlled
+	  from the bootloader that makes the GRPCI to generate interrupts
+	  on detected PCI Parity and System errors.
+
+config SPARC_GRPCI2
 	bool "GRPCI2 Host Bridge Support"
 	depends on LEON_PCI
 	default y
diff --git a/arch/sparc/Makefile b/arch/sparc/Makefile
index 541b8b075c7d..9ff423678cbc 100644
--- a/arch/sparc/Makefile
+++ b/arch/sparc/Makefile
@@ -57,6 +57,7 @@ core-y                 += arch/sparc/
 libs-y                 += arch/sparc/prom/
 libs-y                 += arch/sparc/lib/
 
+drivers-$(CONFIG_PM) += arch/sparc/power/
 drivers-$(CONFIG_OPROFILE)	+= arch/sparc/oprofile/
 
 boot := arch/sparc/boot
diff --git a/arch/sparc/include/asm/cmpxchg_64.h b/arch/sparc/include/asm/cmpxchg_64.h
index b30eb37294c5..4adefe8e2885 100644
--- a/arch/sparc/include/asm/cmpxchg_64.h
+++ b/arch/sparc/include/asm/cmpxchg_64.h
@@ -141,5 +141,6 @@ static inline unsigned long __cmpxchg_local(volatile void *ptr,
 	BUILD_BUG_ON(sizeof(*(ptr)) != 8);				\
 	cmpxchg_local((ptr), (o), (n));					\
   })
+#define cmpxchg64(ptr, o, n)	cmpxchg64_local((ptr), (o), (n))
 
 #endif /* __ARCH_SPARC64_CMPXCHG__ */
diff --git a/arch/sparc/include/asm/head_32.h b/arch/sparc/include/asm/head_32.h
index a76874838f61..5f1dbe315bc8 100644
--- a/arch/sparc/include/asm/head_32.h
+++ b/arch/sparc/include/asm/head_32.h
@@ -55,15 +55,15 @@
 
 /* The Get Condition Codes software trap for userland. */
 #define GETCC_TRAP \
-        b getcc_trap_handler; mov %psr, %l0; nop; nop;
+        b getcc_trap_handler; rd %psr, %l0; nop; nop;
 
 /* The Set Condition Codes software trap for userland. */
 #define SETCC_TRAP \
-        b setcc_trap_handler; mov %psr, %l0; nop; nop;
+        b setcc_trap_handler; rd %psr, %l0; nop; nop;
 
 /* The Get PSR software trap for userland. */
 #define GETPSR_TRAP \
-	mov %psr, %i0; jmp %l2; rett %l2 + 4; nop;
+	rd %psr, %i0; jmp %l2; rett %l2 + 4; nop;
 
 /* This is for hard interrupts from level 1-14, 15 is non-maskable (nmi) and
  * gets handled with another macro.
diff --git a/arch/sparc/include/asm/hibernate.h b/arch/sparc/include/asm/hibernate.h
new file mode 100644
index 000000000000..2ec34f842249
--- /dev/null
+++ b/arch/sparc/include/asm/hibernate.h
@@ -0,0 +1,23 @@
+/*
+ * hibernate.h:  Hibernaton support specific for sparc64.
+ *
+ * Copyright (C) 2013 Kirill V Tkhai (tkhai@yandex.ru)
+ */
+
+#ifndef ___SPARC_HIBERNATE_H
+#define ___SPARC_HIBERNATE_H
+
+struct saved_context {
+	unsigned long fp;
+	unsigned long cwp;
+	unsigned long wstate;
+
+	unsigned long tick;
+	unsigned long pstate;
+
+	unsigned long g4;
+	unsigned long g5;
+	unsigned long g6;
+};
+
+#endif
diff --git a/arch/sparc/include/asm/leon_pci.h b/arch/sparc/include/asm/leon_pci.h
index f48527ebdd8f..bfd3ab3092b5 100644
--- a/arch/sparc/include/asm/leon_pci.h
+++ b/arch/sparc/include/asm/leon_pci.h
@@ -12,6 +12,7 @@ struct leon_pci_info {
 	struct pci_ops *ops;
 	struct resource	io_space;
 	struct resource	mem_space;
+	struct resource	busn;
 	int (*map_irq)(const struct pci_dev *dev, u8 slot, u8 pin);
 };
 
diff --git a/arch/sparc/include/asm/mmu_context_64.h b/arch/sparc/include/asm/mmu_context_64.h
index 9191ca62ed9c..3d528f06e4b0 100644
--- a/arch/sparc/include/asm/mmu_context_64.h
+++ b/arch/sparc/include/asm/mmu_context_64.h
@@ -68,7 +68,7 @@ extern void smp_tsb_sync(struct mm_struct *mm);
 
 extern void __flush_tlb_mm(unsigned long, unsigned long);
 
-/* Switch the current MM context.  Interrupts are disabled.  */
+/* Switch the current MM context. */
 static inline void switch_mm(struct mm_struct *old_mm, struct mm_struct *mm, struct task_struct *tsk)
 {
 	unsigned long ctx_valid, flags;
diff --git a/arch/sparc/include/asm/processor_64.h b/arch/sparc/include/asm/processor_64.h
index cce72ce4c334..4c3f7f01c709 100644
--- a/arch/sparc/include/asm/processor_64.h
+++ b/arch/sparc/include/asm/processor_64.h
@@ -18,9 +18,6 @@
 #include <asm/ptrace.h>
 #include <asm/page.h>
 
-/* Don't hold the runqueue lock over context switch */
-#define __ARCH_WANT_UNLOCKED_CTXSW
-
 /* The sparc has no problems with write protection */
 #define wp_works_ok 1
 #define wp_works_ok__is_a_macro /* for versions in ksyms.c */
diff --git a/arch/sparc/kernel/Makefile b/arch/sparc/kernel/Makefile
index 5276fd4e9d03..d432fb20358e 100644
--- a/arch/sparc/kernel/Makefile
+++ b/arch/sparc/kernel/Makefile
@@ -74,7 +74,8 @@ obj-y                     += dma.o
 
 obj-$(CONFIG_PCIC_PCI)    += pcic.o
 obj-$(CONFIG_LEON_PCI)    += leon_pci.o
-obj-$(CONFIG_GRPCI2)      += leon_pci_grpci2.o
+obj-$(CONFIG_SPARC_GRPCI2)+= leon_pci_grpci2.o
+obj-$(CONFIG_SPARC_GRPCI1)+= leon_pci_grpci1.o
 
 obj-$(CONFIG_SMP)         += trampoline_$(BITS).o smp_$(BITS).o
 obj-$(CONFIG_SPARC32_SMP) += sun4m_smp.o sun4d_smp.o leon_smp.o
diff --git a/arch/sparc/kernel/asm-offsets.c b/arch/sparc/kernel/asm-offsets.c
index 68f7e1118e9b..961b87f99e69 100644
--- a/arch/sparc/kernel/asm-offsets.c
+++ b/arch/sparc/kernel/asm-offsets.c
@@ -14,6 +14,8 @@
 // #include <linux/mm.h>
 #include <linux/kbuild.h>
 
+#include <asm/hibernate.h>
+
 #ifdef CONFIG_SPARC32
 int sparc32_foo(void)
 {
@@ -24,6 +26,19 @@ int sparc32_foo(void)
 #else
 int sparc64_foo(void)
 {
+#ifdef CONFIG_HIBERNATION
+	BLANK();
+	OFFSET(SC_REG_FP, saved_context, fp);
+	OFFSET(SC_REG_CWP, saved_context, cwp);
+	OFFSET(SC_REG_WSTATE, saved_context, wstate);
+
+	OFFSET(SC_REG_TICK, saved_context, tick);
+	OFFSET(SC_REG_PSTATE, saved_context, pstate);
+
+	OFFSET(SC_REG_G4, saved_context, g4);
+	OFFSET(SC_REG_G5, saved_context, g5);
+	OFFSET(SC_REG_G6, saved_context, g6);
+#endif
 	return 0;
 }
 #endif
diff --git a/arch/sparc/kernel/leon_kernel.c b/arch/sparc/kernel/leon_kernel.c
index 87f60ee65433..7c0231dabe44 100644
--- a/arch/sparc/kernel/leon_kernel.c
+++ b/arch/sparc/kernel/leon_kernel.c
@@ -213,6 +213,7 @@ unsigned int leon_build_device_irq(unsigned int real_irq,
 {
 	unsigned int irq;
 	unsigned long mask;
+	struct irq_desc *desc;
 
 	irq = 0;
 	mask = leon_get_irqmask(real_irq);
@@ -226,9 +227,12 @@ unsigned int leon_build_device_irq(unsigned int real_irq,
 	if (do_ack)
 		mask |= LEON_DO_ACK_HW;
 
-	irq_set_chip_and_handler_name(irq, &leon_irq,
-				      flow_handler, name);
-	irq_set_chip_data(irq, (void *)mask);
+	desc = irq_to_desc(irq);
+	if (!desc || !desc->handle_irq || desc->handle_irq == handle_bad_irq) {
+		irq_set_chip_and_handler_name(irq, &leon_irq,
+					      flow_handler, name);
+		irq_set_chip_data(irq, (void *)mask);
+	}
 
 out:
 	return irq;
diff --git a/arch/sparc/kernel/leon_pci.c b/arch/sparc/kernel/leon_pci.c
index 852dc8430528..88aaaa57bb64 100644
--- a/arch/sparc/kernel/leon_pci.c
+++ b/arch/sparc/kernel/leon_pci.c
@@ -29,6 +29,8 @@ void leon_pci_init(struct platform_device *ofdev, struct leon_pci_info *info)
 	pci_add_resource_offset(&resources, &info->io_space,
 				info->io_space.start - 0x1000);
 	pci_add_resource(&resources, &info->mem_space);
+	info->busn.flags = IORESOURCE_BUS;
+	pci_add_resource(&resources, &info->busn);
 
 	root_bus = pci_scan_root_bus(&ofdev->dev, 0, info->ops, info,
 				     &resources);
diff --git a/arch/sparc/kernel/leon_pci_grpci1.c b/arch/sparc/kernel/leon_pci_grpci1.c
new file mode 100644
index 000000000000..7739a54315e2
--- /dev/null
+++ b/arch/sparc/kernel/leon_pci_grpci1.c
@@ -0,0 +1,724 @@
+/*
+ * leon_pci_grpci1.c: GRPCI1 Host PCI driver
+ *
+ * Copyright (C) 2013 Aeroflex Gaisler AB
+ *
+ * This GRPCI1 driver does not support PCI interrupts taken from
+ * GPIO pins. Interrupt generation at PCI parity and system error
+ * detection is by default turned off since some GRPCI1 cores does
+ * not support detection. It can be turned on from the bootloader
+ * using the all_pci_errors property.
+ *
+ * Contributors: Daniel Hellstrom <daniel@gaisler.com>
+ */
+
+#include <linux/of_device.h>
+#include <linux/export.h>
+#include <linux/kernel.h>
+#include <linux/of_irq.h>
+#include <linux/delay.h>
+#include <linux/pci.h>
+
+#include <asm/leon_pci.h>
+#include <asm/sections.h>
+#include <asm/vaddrs.h>
+#include <asm/leon.h>
+#include <asm/io.h>
+
+#include "irq.h"
+
+/* Enable/Disable Debugging Configuration Space Access */
+#undef GRPCI1_DEBUG_CFGACCESS
+
+/*
+ * GRPCI1 APB Register MAP
+ */
+struct grpci1_regs {
+	unsigned int cfg_stat;		/* 0x00 Configuration / Status */
+	unsigned int bar0;		/* 0x04 BAR0 (RO) */
+	unsigned int page0;		/* 0x08 PAGE0 (RO) */
+	unsigned int bar1;		/* 0x0C BAR1 (RO) */
+	unsigned int page1;		/* 0x10 PAGE1 */
+	unsigned int iomap;		/* 0x14 IO Map */
+	unsigned int stat_cmd;		/* 0x18 PCI Status & Command (RO) */
+	unsigned int irq;		/* 0x1C Interrupt register */
+};
+
+#define REGLOAD(a)	(be32_to_cpu(__raw_readl(&(a))))
+#define REGSTORE(a, v)	(__raw_writel(cpu_to_be32(v), &(a)))
+
+#define PAGE0_BTEN_BIT    0
+#define PAGE0_BTEN        (1 << PAGE0_BTEN_BIT)
+
+#define CFGSTAT_HOST_BIT  13
+#define CFGSTAT_CTO_BIT   8
+#define CFGSTAT_HOST      (1 << CFGSTAT_HOST_BIT)
+#define CFGSTAT_CTO       (1 << CFGSTAT_CTO_BIT)
+
+#define IRQ_DPE (1 << 9)
+#define IRQ_SSE (1 << 8)
+#define IRQ_RMA (1 << 7)
+#define IRQ_RTA (1 << 6)
+#define IRQ_STA (1 << 5)
+#define IRQ_DPED (1 << 4)
+#define IRQ_INTD (1 << 3)
+#define IRQ_INTC (1 << 2)
+#define IRQ_INTB (1 << 1)
+#define IRQ_INTA (1 << 0)
+#define IRQ_DEF_ERRORS (IRQ_RMA | IRQ_RTA | IRQ_STA)
+#define IRQ_ALL_ERRORS (IRQ_DPED | IRQ_DEF_ERRORS | IRQ_SSE | IRQ_DPE)
+#define IRQ_INTX (IRQ_INTA | IRQ_INTB | IRQ_INTC | IRQ_INTD)
+#define IRQ_MASK_BIT 16
+
+#define DEF_PCI_ERRORS (PCI_STATUS_SIG_TARGET_ABORT | \
+			PCI_STATUS_REC_TARGET_ABORT | \
+			PCI_STATUS_REC_MASTER_ABORT)
+#define ALL_PCI_ERRORS (PCI_STATUS_PARITY | PCI_STATUS_DETECTED_PARITY | \
+			PCI_STATUS_SIG_SYSTEM_ERROR | DEF_PCI_ERRORS)
+
+#define TGT 256
+
+struct grpci1_priv {
+	struct leon_pci_info	info; /* must be on top of this structure */
+	struct grpci1_regs	*regs;		/* GRPCI register map */
+	struct device		*dev;
+	int			pci_err_mask;	/* STATUS register error mask */
+	int			irq;		/* LEON irqctrl GRPCI IRQ */
+	unsigned char		irq_map[4];	/* GRPCI nexus PCI INTX# IRQs */
+	unsigned int		irq_err;	/* GRPCI nexus Virt Error IRQ */
+
+	/* AHB PCI Windows */
+	unsigned long		pci_area;	/* MEMORY */
+	unsigned long		pci_area_end;
+	unsigned long		pci_io;		/* I/O */
+	unsigned long		pci_conf;	/* CONFIGURATION */
+	unsigned long		pci_conf_end;
+	unsigned long		pci_io_va;
+};
+
+static struct grpci1_priv *grpci1priv;
+
+static int grpci1_cfg_w32(struct grpci1_priv *priv, unsigned int bus,
+				unsigned int devfn, int where, u32 val);
+
+int grpci1_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
+{
+	struct grpci1_priv *priv = dev->bus->sysdata;
+	int irq_group;
+
+	/* Use default IRQ decoding on PCI BUS0 according slot numbering */
+	irq_group = slot & 0x3;
+	pin = ((pin - 1) + irq_group) & 0x3;
+
+	return priv->irq_map[pin];
+}
+
+static int grpci1_cfg_r32(struct grpci1_priv *priv, unsigned int bus,
+				unsigned int devfn, int where, u32 *val)
+{
+	u32 *pci_conf, tmp, cfg;
+
+	if (where & 0x3)
+		return -EINVAL;
+
+	if (bus == 0) {
+		devfn += (0x8 * 6); /* start at AD16=Device0 */
+	} else if (bus == TGT) {
+		bus = 0;
+		devfn = 0; /* special case: bridge controller itself */
+	}
+
+	/* Select bus */
+	cfg = REGLOAD(priv->regs->cfg_stat);
+	REGSTORE(priv->regs->cfg_stat, (cfg & ~(0xf << 23)) | (bus << 23));
+
+	/* do read access */
+	pci_conf = (u32 *) (priv->pci_conf | (devfn << 8) | (where & 0xfc));
+	tmp = LEON3_BYPASS_LOAD_PA(pci_conf);
+
+	/* check if master abort was received */
+	if (REGLOAD(priv->regs->cfg_stat) & CFGSTAT_CTO) {
+		*val = 0xffffffff;
+		/* Clear Master abort bit in PCI cfg space (is set) */
+		tmp = REGLOAD(priv->regs->stat_cmd);
+		grpci1_cfg_w32(priv, TGT, 0, PCI_COMMAND, tmp);
+	} else {
+		/* Bus always little endian (unaffected by byte-swapping) */
+		*val = flip_dword(tmp);
+	}
+
+	return 0;
+}
+
+static int grpci1_cfg_r16(struct grpci1_priv *priv, unsigned int bus,
+				unsigned int devfn, int where, u32 *val)
+{
+	u32 v;
+	int ret;
+
+	if (where & 0x1)
+		return -EINVAL;
+	ret = grpci1_cfg_r32(priv, bus, devfn, where & ~0x3, &v);
+	*val = 0xffff & (v >> (8 * (where & 0x3)));
+	return ret;
+}
+
+static int grpci1_cfg_r8(struct grpci1_priv *priv, unsigned int bus,
+				unsigned int devfn, int where, u32 *val)
+{
+	u32 v;
+	int ret;
+
+	ret = grpci1_cfg_r32(priv, bus, devfn, where & ~0x3, &v);
+	*val = 0xff & (v >> (8 * (where & 3)));
+
+	return ret;
+}
+
+static int grpci1_cfg_w32(struct grpci1_priv *priv, unsigned int bus,
+				unsigned int devfn, int where, u32 val)
+{
+	unsigned int *pci_conf;
+	u32 cfg;
+
+	if (where & 0x3)
+		return -EINVAL;
+
+	if (bus == 0) {
+		devfn += (0x8 * 6); /* start at AD16=Device0 */
+	} else if (bus == TGT) {
+		bus = 0;
+		devfn = 0; /* special case: bridge controller itself */
+	}
+
+	/* Select bus */
+	cfg = REGLOAD(priv->regs->cfg_stat);
+	REGSTORE(priv->regs->cfg_stat, (cfg & ~(0xf << 23)) | (bus << 23));
+
+	pci_conf = (unsigned int *) (priv->pci_conf |
+						(devfn << 8) | (where & 0xfc));
+	LEON3_BYPASS_STORE_PA(pci_conf, flip_dword(val));
+
+	return 0;
+}
+
+static int grpci1_cfg_w16(struct grpci1_priv *priv, unsigned int bus,
+				unsigned int devfn, int where, u32 val)
+{
+	int ret;
+	u32 v;
+
+	if (where & 0x1)
+		return -EINVAL;
+	ret = grpci1_cfg_r32(priv, bus, devfn, where&~3, &v);
+	if (ret)
+		return ret;
+	v = (v & ~(0xffff << (8 * (where & 0x3)))) |
+	    ((0xffff & val) << (8 * (where & 0x3)));
+	return grpci1_cfg_w32(priv, bus, devfn, where & ~0x3, v);
+}
+
+static int grpci1_cfg_w8(struct grpci1_priv *priv, unsigned int bus,
+				unsigned int devfn, int where, u32 val)
+{
+	int ret;
+	u32 v;
+
+	ret = grpci1_cfg_r32(priv, bus, devfn, where & ~0x3, &v);
+	if (ret != 0)
+		return ret;
+	v = (v & ~(0xff << (8 * (where & 0x3)))) |
+	    ((0xff & val) << (8 * (where & 0x3)));
+	return grpci1_cfg_w32(priv, bus, devfn, where & ~0x3, v);
+}
+
+/* Read from Configuration Space. When entering here the PCI layer has taken
+ * the pci_lock spinlock and IRQ is off.
+ */
+static int grpci1_read_config(struct pci_bus *bus, unsigned int devfn,
+			      int where, int size, u32 *val)
+{
+	struct grpci1_priv *priv = grpci1priv;
+	unsigned int busno = bus->number;
+	int ret;
+
+	if (PCI_SLOT(devfn) > 15 || busno > 15) {
+		*val = ~0;
+		return 0;
+	}
+
+	switch (size) {
+	case 1:
+		ret = grpci1_cfg_r8(priv, busno, devfn, where, val);
+		break;
+	case 2:
+		ret = grpci1_cfg_r16(priv, busno, devfn, where, val);
+		break;
+	case 4:
+		ret = grpci1_cfg_r32(priv, busno, devfn, where, val);
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+#ifdef GRPCI1_DEBUG_CFGACCESS
+	printk(KERN_INFO
+		"grpci1_read_config: [%02x:%02x:%x] ofs=%d val=%x size=%d\n",
+		busno, PCI_SLOT(devfn), PCI_FUNC(devfn), where, *val, size);
+#endif
+
+	return ret;
+}
+
+/* Write to Configuration Space. When entering here the PCI layer has taken
+ * the pci_lock spinlock and IRQ is off.
+ */
+static int grpci1_write_config(struct pci_bus *bus, unsigned int devfn,
+			       int where, int size, u32 val)
+{
+	struct grpci1_priv *priv = grpci1priv;
+	unsigned int busno = bus->number;
+
+	if (PCI_SLOT(devfn) > 15 || busno > 15)
+		return 0;
+
+#ifdef GRPCI1_DEBUG_CFGACCESS
+	printk(KERN_INFO
+		"grpci1_write_config: [%02x:%02x:%x] ofs=%d size=%d val=%x\n",
+		busno, PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
+#endif
+
+	switch (size) {
+	default:
+		return -EINVAL;
+	case 1:
+		return grpci1_cfg_w8(priv, busno, devfn, where, val);
+	case 2:
+		return grpci1_cfg_w16(priv, busno, devfn, where, val);
+	case 4:
+		return grpci1_cfg_w32(priv, busno, devfn, where, val);
+	}
+}
+
+static struct pci_ops grpci1_ops = {
+	.read =		grpci1_read_config,
+	.write =	grpci1_write_config,
+};
+
+/* GENIRQ IRQ chip implementation for grpci1 irqmode=0..2. In configuration
+ * 3 where all PCI Interrupts has a separate IRQ on the system IRQ controller
+ * this is not needed and the standard IRQ controller can be used.
+ */
+
+static void grpci1_mask_irq(struct irq_data *data)
+{
+	u32 irqidx;
+	struct grpci1_priv *priv = grpci1priv;
+
+	irqidx = (u32)data->chip_data - 1;
+	if (irqidx > 3) /* only mask PCI interrupts here */
+		return;
+	irqidx += IRQ_MASK_BIT;
+
+	REGSTORE(priv->regs->irq, REGLOAD(priv->regs->irq) & ~(1 << irqidx));
+}
+
+static void grpci1_unmask_irq(struct irq_data *data)
+{
+	u32 irqidx;
+	struct grpci1_priv *priv = grpci1priv;
+
+	irqidx = (u32)data->chip_data - 1;
+	if (irqidx > 3) /* only unmask PCI interrupts here */
+		return;
+	irqidx += IRQ_MASK_BIT;
+
+	REGSTORE(priv->regs->irq, REGLOAD(priv->regs->irq) | (1 << irqidx));
+}
+
+static unsigned int grpci1_startup_irq(struct irq_data *data)
+{
+	grpci1_unmask_irq(data);
+	return 0;
+}
+
+static void grpci1_shutdown_irq(struct irq_data *data)
+{
+	grpci1_mask_irq(data);
+}
+
+static struct irq_chip grpci1_irq = {
+	.name		= "grpci1",
+	.irq_startup	= grpci1_startup_irq,
+	.irq_shutdown	= grpci1_shutdown_irq,
+	.irq_mask	= grpci1_mask_irq,
+	.irq_unmask	= grpci1_unmask_irq,
+};
+
+/* Handle one or multiple IRQs from the PCI core */
+static void grpci1_pci_flow_irq(unsigned int irq, struct irq_desc *desc)
+{
+	struct grpci1_priv *priv = grpci1priv;
+	int i, ack = 0;
+	unsigned int irqreg;
+
+	irqreg = REGLOAD(priv->regs->irq);
+	irqreg = (irqreg >> IRQ_MASK_BIT) & irqreg;
+
+	/* Error Interrupt? */
+	if (irqreg & IRQ_ALL_ERRORS) {
+		generic_handle_irq(priv->irq_err);
+		ack = 1;
+	}
+
+	/* PCI Interrupt? */
+	if (irqreg & IRQ_INTX) {
+		/* Call respective PCI Interrupt handler */
+		for (i = 0; i < 4; i++) {
+			if (irqreg & (1 << i))
+				generic_handle_irq(priv->irq_map[i]);
+		}
+		ack = 1;
+	}
+
+	/*
+	 * Call "first level" IRQ chip end-of-irq handler. It will ACK LEON IRQ
+	 * Controller, this must be done after IRQ sources have been handled to
+	 * avoid double IRQ generation
+	 */
+	if (ack)
+		desc->irq_data.chip->irq_eoi(&desc->irq_data);
+}
+
+/* Create a virtual IRQ */
+static unsigned int grpci1_build_device_irq(unsigned int irq)
+{
+	unsigned int virq = 0, pil;
+
+	pil = 1 << 8;
+	virq = irq_alloc(irq, pil);
+	if (virq == 0)
+		goto out;
+
+	irq_set_chip_and_handler_name(virq, &grpci1_irq, handle_simple_irq,
+				      "pcilvl");
+	irq_set_chip_data(virq, (void *)irq);
+
+out:
+	return virq;
+}
+
+/*
+ * Initialize mappings AMBA<->PCI, clear IRQ state, setup PCI interface
+ *
+ * Target BARs:
+ *  BAR0: unused in this implementation
+ *  BAR1: peripheral DMA to host's memory (size at least 256MByte)
+ *  BAR2..BAR5: not implemented in hardware
+ */
+void grpci1_hw_init(struct grpci1_priv *priv)
+{
+	u32 ahbadr, bar_sz, data, pciadr;
+	struct grpci1_regs *regs = priv->regs;
+
+	/* set 1:1 mapping between AHB -> PCI memory space */
+	REGSTORE(regs->cfg_stat, priv->pci_area & 0xf0000000);
+
+	/* map PCI accesses to target BAR1 to Linux kernel memory 1:1 */
+	ahbadr = 0xf0000000 & (u32)__pa(PAGE_ALIGN((unsigned long) &_end));
+	REGSTORE(regs->page1, ahbadr);
+
+	/* translate I/O accesses to 0, I/O Space always @ PCI low 64Kbytes */
+	REGSTORE(regs->iomap, REGLOAD(regs->iomap) & 0x0000ffff);
+
+	/* disable and clear pending interrupts */
+	REGSTORE(regs->irq, 0);
+
+	/* Setup BAR0 outside access range so that it does not conflict with
+	 * peripheral DMA. There is no need to set up the PAGE0 register.
+	 */
+	grpci1_cfg_w32(priv, TGT, 0, PCI_BASE_ADDRESS_0, 0xffffffff);
+	grpci1_cfg_r32(priv, TGT, 0, PCI_BASE_ADDRESS_0, &bar_sz);
+	bar_sz = ~bar_sz + 1;
+	pciadr = priv->pci_area - bar_sz;
+	grpci1_cfg_w32(priv, TGT, 0, PCI_BASE_ADDRESS_0, pciadr);
+
+	/*
+	 * Setup the Host's PCI Target BAR1 for other peripherals to access,
+	 * and do DMA to the host's memory.
+	 */
+	grpci1_cfg_w32(priv, TGT, 0, PCI_BASE_ADDRESS_1, ahbadr);
+
+	/*
+	 * Setup Latency Timer and cache line size. Default cache line
+	 * size will result in poor performance (256 word fetches), 0xff
+	 * will set it according to the max size of the PCI FIFO.
+	 */
+	grpci1_cfg_w8(priv, TGT, 0, PCI_CACHE_LINE_SIZE, 0xff);
+	grpci1_cfg_w8(priv, TGT, 0, PCI_LATENCY_TIMER, 0x40);
+
+	/* set as bus master, enable pci memory responses, clear status bits */
+	grpci1_cfg_r32(priv, TGT, 0, PCI_COMMAND, &data);
+	data |= (PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
+	grpci1_cfg_w32(priv, TGT, 0, PCI_COMMAND, data);
+}
+
+static irqreturn_t grpci1_jump_interrupt(int irq, void *arg)
+{
+	struct grpci1_priv *priv = arg;
+	dev_err(priv->dev, "Jump IRQ happened\n");
+	return IRQ_NONE;
+}
+
+/* Handle GRPCI1 Error Interrupt */
+static irqreturn_t grpci1_err_interrupt(int irq, void *arg)
+{
+	struct grpci1_priv *priv = arg;
+	u32 status;
+
+	grpci1_cfg_r16(priv, TGT, 0, PCI_STATUS, &status);
+	status &= priv->pci_err_mask;
+
+	if (status == 0)
+		return IRQ_NONE;
+
+	if (status & PCI_STATUS_PARITY)
+		dev_err(priv->dev, "Data Parity Error\n");
+
+	if (status & PCI_STATUS_SIG_TARGET_ABORT)
+		dev_err(priv->dev, "Signalled Target Abort\n");
+
+	if (status & PCI_STATUS_REC_TARGET_ABORT)
+		dev_err(priv->dev, "Received Target Abort\n");
+
+	if (status & PCI_STATUS_REC_MASTER_ABORT)
+		dev_err(priv->dev, "Received Master Abort\n");
+
+	if (status & PCI_STATUS_SIG_SYSTEM_ERROR)
+		dev_err(priv->dev, "Signalled System Error\n");
+
+	if (status & PCI_STATUS_DETECTED_PARITY)
+		dev_err(priv->dev, "Parity Error\n");
+
+	/* Clear handled INT TYPE IRQs */
+	grpci1_cfg_w16(priv, TGT, 0, PCI_STATUS, status);
+
+	return IRQ_HANDLED;
+}
+
+static int grpci1_of_probe(struct platform_device *ofdev)
+{
+	struct grpci1_regs *regs;
+	struct grpci1_priv *priv;
+	int err, len;
+	const int *tmp;
+	u32 cfg, size, err_mask;
+	struct resource *res;
+
+	if (grpci1priv) {
+		dev_err(&ofdev->dev, "only one GRPCI1 supported\n");
+		return -ENODEV;
+	}
+
+	if (ofdev->num_resources < 3) {
+		dev_err(&ofdev->dev, "not enough APB/AHB resources\n");
+		return -EIO;
+	}
+
+	priv = devm_kzalloc(&ofdev->dev, sizeof(*priv), GFP_KERNEL);
+	if (!priv) {
+		dev_err(&ofdev->dev, "memory allocation failed\n");
+		return -ENOMEM;
+	}
+	platform_set_drvdata(ofdev, priv);
+	priv->dev = &ofdev->dev;
+
+	/* find device register base address */
+	res = platform_get_resource(ofdev, IORESOURCE_MEM, 0);
+	regs = devm_request_and_ioremap(&ofdev->dev, res);
+	if (!regs) {
+		dev_err(&ofdev->dev, "io-regs mapping failed\n");
+		return -EADDRNOTAVAIL;
+	}
+
+	/*
+	 * check that we're in Host Slot and that we can act as a Host Bridge
+	 * and not only as target/peripheral.
+	 */
+	cfg = REGLOAD(regs->cfg_stat);
+	if ((cfg & CFGSTAT_HOST) == 0) {
+		dev_err(&ofdev->dev, "not in host system slot\n");
+		return -EIO;
+	}
+
+	/* check that BAR1 support 256 MByte so that we can map kernel space */
+	REGSTORE(regs->page1, 0xffffffff);
+	size = ~REGLOAD(regs->page1) + 1;
+	if (size < 0x10000000) {
+		dev_err(&ofdev->dev, "BAR1 must be at least 256MByte\n");
+		return -EIO;
+	}
+
+	/* hardware must support little-endian PCI (byte-twisting) */
+	if ((REGLOAD(regs->page0) & PAGE0_BTEN) == 0) {
+		dev_err(&ofdev->dev, "byte-twisting is required\n");
+		return -EIO;
+	}
+
+	priv->regs = regs;
+	priv->irq = irq_of_parse_and_map(ofdev->dev.of_node, 0);
+	dev_info(&ofdev->dev, "host found at 0x%p, irq%d\n", regs, priv->irq);
+
+	/* Find PCI Memory, I/O and Configuration Space Windows */
+	priv->pci_area = ofdev->resource[1].start;
+	priv->pci_area_end = ofdev->resource[1].end+1;
+	priv->pci_io = ofdev->resource[2].start;
+	priv->pci_conf = ofdev->resource[2].start + 0x10000;
+	priv->pci_conf_end = priv->pci_conf + 0x10000;
+	priv->pci_io_va = (unsigned long)ioremap(priv->pci_io, 0x10000);
+	if (!priv->pci_io_va) {
+		dev_err(&ofdev->dev, "unable to map PCI I/O area\n");
+		return -EIO;
+	}
+
+	printk(KERN_INFO
+		"GRPCI1: MEMORY SPACE [0x%08lx - 0x%08lx]\n"
+		"        I/O    SPACE [0x%08lx - 0x%08lx]\n"
+		"        CONFIG SPACE [0x%08lx - 0x%08lx]\n",
+		priv->pci_area, priv->pci_area_end-1,
+		priv->pci_io, priv->pci_conf-1,
+		priv->pci_conf, priv->pci_conf_end-1);
+
+	/*
+	 * I/O Space resources in I/O Window mapped into Virtual Adr Space
+	 * We never use low 4KB because some devices seem have problems using
+	 * address 0.
+	 */
+	priv->info.io_space.name = "GRPCI1 PCI I/O Space";
+	priv->info.io_space.start = priv->pci_io_va + 0x1000;
+	priv->info.io_space.end = priv->pci_io_va + 0x10000 - 1;
+	priv->info.io_space.flags = IORESOURCE_IO;
+
+	/*
+	 * grpci1 has no prefetchable memory, map everything as
+	 * non-prefetchable memory
+	 */
+	priv->info.mem_space.name = "GRPCI1 PCI MEM Space";
+	priv->info.mem_space.start = priv->pci_area;
+	priv->info.mem_space.end = priv->pci_area_end - 1;
+	priv->info.mem_space.flags = IORESOURCE_MEM;
+
+	if (request_resource(&iomem_resource, &priv->info.mem_space) < 0) {
+		dev_err(&ofdev->dev, "unable to request PCI memory area\n");
+		err = -ENOMEM;
+		goto err1;
+	}
+
+	if (request_resource(&ioport_resource, &priv->info.io_space) < 0) {
+		dev_err(&ofdev->dev, "unable to request PCI I/O area\n");
+		err = -ENOMEM;
+		goto err2;
+	}
+
+	/* setup maximum supported PCI buses */
+	priv->info.busn.name = "GRPCI1 busn";
+	priv->info.busn.start = 0;
+	priv->info.busn.end = 15;
+
+	grpci1priv = priv;
+
+	/* Initialize hardware */
+	grpci1_hw_init(priv);
+
+	/*
+	 * Get PCI Interrupt to System IRQ mapping and setup IRQ handling
+	 * Error IRQ. All PCI and PCI-Error interrupts are shared using the
+	 * same system IRQ.
+	 */
+	leon_update_virq_handling(priv->irq, grpci1_pci_flow_irq, "pcilvl", 0);
+
+	priv->irq_map[0] = grpci1_build_device_irq(1);
+	priv->irq_map[1] = grpci1_build_device_irq(2);
+	priv->irq_map[2] = grpci1_build_device_irq(3);
+	priv->irq_map[3] = grpci1_build_device_irq(4);
+	priv->irq_err = grpci1_build_device_irq(5);
+
+	printk(KERN_INFO "        PCI INTA..D#: IRQ%d, IRQ%d, IRQ%d, IRQ%d\n",
+		priv->irq_map[0], priv->irq_map[1], priv->irq_map[2],
+		priv->irq_map[3]);
+
+	/* Enable IRQs on LEON IRQ controller */
+	err = devm_request_irq(&ofdev->dev, priv->irq, grpci1_jump_interrupt, 0,
+				"GRPCI1_JUMP", priv);
+	if (err) {
+		dev_err(&ofdev->dev, "ERR IRQ request failed: %d\n", err);
+		goto err3;
+	}
+
+	/* Setup IRQ handler for access errors */
+	err = devm_request_irq(&ofdev->dev, priv->irq_err,
+				grpci1_err_interrupt, IRQF_SHARED, "GRPCI1_ERR",
+				priv);
+	if (err) {
+		dev_err(&ofdev->dev, "ERR VIRQ request failed: %d\n", err);
+		goto err3;
+	}
+
+	tmp = of_get_property(ofdev->dev.of_node, "all_pci_errors", &len);
+	if (tmp && (len == 4)) {
+		priv->pci_err_mask = ALL_PCI_ERRORS;
+		err_mask = IRQ_ALL_ERRORS << IRQ_MASK_BIT;
+	} else {
+		priv->pci_err_mask = DEF_PCI_ERRORS;
+		err_mask = IRQ_DEF_ERRORS << IRQ_MASK_BIT;
+	}
+
+	/*
+	 * Enable Error Interrupts. PCI interrupts are unmasked once request_irq
+	 * is called by the PCI Device drivers
+	 */
+	REGSTORE(regs->irq, err_mask);
+
+	/* Init common layer and scan buses */
+	priv->info.ops = &grpci1_ops;
+	priv->info.map_irq = grpci1_map_irq;
+	leon_pci_init(ofdev, &priv->info);
+
+	return 0;
+
+err3:
+	release_resource(&priv->info.io_space);
+err2:
+	release_resource(&priv->info.mem_space);
+err1:
+	iounmap((void *)priv->pci_io_va);
+	grpci1priv = NULL;
+	return err;
+}
+
+static struct of_device_id grpci1_of_match[] = {
+	{
+	 .name = "GAISLER_PCIFBRG",
+	 },
+	{
+	 .name = "01_014",
+	 },
+	{},
+};
+
+static struct platform_driver grpci1_of_driver = {
+	.driver = {
+		.name = "grpci1",
+		.owner = THIS_MODULE,
+		.of_match_table = grpci1_of_match,
+	},
+	.probe = grpci1_of_probe,
+};
+
+static int __init grpci1_init(void)
+{
+	return platform_driver_register(&grpci1_of_driver);
+}
+
+subsys_initcall(grpci1_init);
diff --git a/arch/sparc/kernel/leon_pci_grpci2.c b/arch/sparc/kernel/leon_pci_grpci2.c
index 4d1487138d26..5f0402aab7fb 100644
--- a/arch/sparc/kernel/leon_pci_grpci2.c
+++ b/arch/sparc/kernel/leon_pci_grpci2.c
@@ -799,6 +799,11 @@ static int grpci2_of_probe(struct platform_device *ofdev)
 	if (request_resource(&ioport_resource, &priv->info.io_space) < 0)
 		goto err4;
 
+	/* setup maximum supported PCI buses */
+	priv->info.busn.name = "GRPCI2 busn";
+	priv->info.busn.start = 0;
+	priv->info.busn.end = 255;
+
 	grpci2_hw_init(priv);
 
 	/*
diff --git a/arch/sparc/kernel/leon_pmc.c b/arch/sparc/kernel/leon_pmc.c
index 708bca435219..bdf53d9a8d46 100644
--- a/arch/sparc/kernel/leon_pmc.c
+++ b/arch/sparc/kernel/leon_pmc.c
@@ -48,7 +48,7 @@ void pmc_leon_idle_fixup(void)
 	 */
 	register unsigned int address = (unsigned int)leon3_irqctrl_regs;
 	__asm__ __volatile__ (
-		"mov	%%g0, %%asr19\n"
+		"wr	%%g0, %%asr19\n"
 		"lda	[%0] %1, %%g0\n"
 		:
 		: "r"(address), "i"(ASI_LEON_BYPASS));
@@ -61,7 +61,7 @@ void pmc_leon_idle_fixup(void)
 void pmc_leon_idle(void)
 {
 	/* For systems without power-down, this will be no-op */
-	__asm__ __volatile__ ("mov	%g0, %asr19\n\t");
+	__asm__ __volatile__ ("wr	%g0, %asr19\n\t");
 }
 
 /* Install LEON Power Down function */
diff --git a/arch/sparc/kernel/vio.c b/arch/sparc/kernel/vio.c
index 3e244f31e56b..8647fcc5ca6c 100644
--- a/arch/sparc/kernel/vio.c
+++ b/arch/sparc/kernel/vio.c
@@ -342,6 +342,7 @@ static void vio_remove(struct mdesc_handle *hp, u64 node)
 		printk(KERN_INFO "VIO: Removing device %s\n", dev_name(dev));
 
 		device_unregister(dev);
+		put_device(dev);
 	}
 }
 
diff --git a/arch/sparc/mm/init_64.c b/arch/sparc/mm/init_64.c
index 6ac99d64a13c..cf72a8a5b3aa 100644
--- a/arch/sparc/mm/init_64.c
+++ b/arch/sparc/mm/init_64.c
@@ -681,10 +681,9 @@ void get_new_mmu_context(struct mm_struct *mm)
 {
 	unsigned long ctx, new_ctx;
 	unsigned long orig_pgsz_bits;
-	unsigned long flags;
 	int new_version;
 
-	spin_lock_irqsave(&ctx_alloc_lock, flags);
+	spin_lock(&ctx_alloc_lock);
 	orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
 	ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
 	new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
@@ -720,7 +719,7 @@ void get_new_mmu_context(struct mm_struct *mm)
 out:
 	tlb_context_cache = new_ctx;
 	mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
-	spin_unlock_irqrestore(&ctx_alloc_lock, flags);
+	spin_unlock(&ctx_alloc_lock);
 
 	if (unlikely(new_version))
 		smp_new_mmu_context_version();
@@ -2125,7 +2124,6 @@ void free_initmem(void)
 			ClearPageReserved(p);
 			init_page_count(p);
 			__free_page(p);
-			num_physpages++;
 			totalram_pages++;
 		}
 	}
@@ -2142,7 +2140,6 @@ void free_initrd_mem(unsigned long start, unsigned long end)
 		ClearPageReserved(p);
 		init_page_count(p);
 		__free_page(p);
-		num_physpages++;
 		totalram_pages++;
 	}
 }
diff --git a/arch/sparc/power/Makefile b/arch/sparc/power/Makefile
new file mode 100644
index 000000000000..3201ace0ddbd
--- /dev/null
+++ b/arch/sparc/power/Makefile
@@ -0,0 +1,3 @@
+# Makefile for Sparc-specific hibernate files.
+
+obj-$(CONFIG_HIBERNATION)	+= hibernate.o hibernate_asm.o
diff --git a/arch/sparc/power/hibernate.c b/arch/sparc/power/hibernate.c
new file mode 100644
index 000000000000..42b0b8ce699a
--- /dev/null
+++ b/arch/sparc/power/hibernate.c
@@ -0,0 +1,42 @@
+/*
+ * hibernate.c:  Hibernaton support specific for sparc64.
+ *
+ * Copyright (C) 2013 Kirill V Tkhai (tkhai@yandex.ru)
+ */
+
+#include <linux/mm.h>
+
+#include <asm/hibernate.h>
+#include <asm/visasm.h>
+#include <asm/page.h>
+#include <asm/tlb.h>
+
+/* References to section boundaries */
+extern const void __nosave_begin, __nosave_end;
+
+struct saved_context saved_context;
+
+/*
+ *	pfn_is_nosave - check if given pfn is in the 'nosave' section
+ */
+
+int pfn_is_nosave(unsigned long pfn)
+{
+	unsigned long nosave_begin_pfn = PFN_DOWN((unsigned long)&__nosave_begin);
+	unsigned long nosave_end_pfn = PFN_DOWN((unsigned long)&__nosave_end);
+
+	return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
+}
+
+void save_processor_state(void)
+{
+	save_and_clear_fpu();
+}
+
+void restore_processor_state(void)
+{
+	struct mm_struct *mm = current->active_mm;
+
+	load_secondary_context(mm);
+	tsb_context_switch(mm);
+}
diff --git a/arch/sparc/power/hibernate_asm.S b/arch/sparc/power/hibernate_asm.S
new file mode 100644
index 000000000000..79942166df84
--- /dev/null
+++ b/arch/sparc/power/hibernate_asm.S
@@ -0,0 +1,131 @@
+/*
+ * hibernate_asm.S:  Hibernaton support specific for sparc64.
+ *
+ * Copyright (C) 2013 Kirill V Tkhai (tkhai@yandex.ru)
+ */
+
+#include <linux/linkage.h>
+
+#include <asm/asm-offsets.h>
+#include <asm/cpudata.h>
+#include <asm/page.h>
+
+ENTRY(swsusp_arch_suspend)
+	save	%sp, -128, %sp
+	save	%sp, -128, %sp
+	flushw
+
+	setuw	saved_context, %g3
+
+	/* Save window regs */
+	rdpr	%cwp, %g2
+	stx	%g2, [%g3 + SC_REG_CWP]
+	rdpr	%wstate, %g2
+	stx	%g2, [%g3 + SC_REG_WSTATE]
+	stx	%fp, [%g3 + SC_REG_FP]
+
+	/* Save state regs */
+	rdpr	%tick, %g2
+	stx	%g2, [%g3 + SC_REG_TICK]
+	rdpr	%pstate, %g2
+	stx	%g2, [%g3 + SC_REG_PSTATE]
+
+	/* Save global regs */
+	stx	%g4, [%g3 + SC_REG_G4]
+	stx	%g5, [%g3 + SC_REG_G5]
+	stx	%g6, [%g3 + SC_REG_G6]
+
+	call	swsusp_save
+	 nop
+
+	mov	%o0, %i0
+	restore
+
+	mov	%o0, %i0
+	ret
+	 restore
+
+ENTRY(swsusp_arch_resume)
+	/* Write restore_pblist to %l0 */
+	sethi	%hi(restore_pblist), %l0
+	ldx	[%l0 + %lo(restore_pblist)], %l0
+
+	call	__flush_tlb_all
+	 nop
+
+	/* Write PAGE_OFFSET to %g7 */
+	sethi	%uhi(PAGE_OFFSET), %g7
+	sllx	%g7, 32, %g7
+
+	setuw	(PAGE_SIZE-8), %g3
+
+	/* Use MMU Bypass */
+	rd	%asi, %g1
+	wr	%g0, ASI_PHYS_USE_EC, %asi
+
+	ba	fill_itlb
+	 nop
+
+pbe_loop:
+	cmp	%l0, %g0
+	be	restore_ctx
+	 sub	%l0, %g7, %l0
+
+	ldxa	[%l0    ] %asi, %l1 /* address */
+	ldxa	[%l0 + 8] %asi, %l2 /* orig_address */
+
+	/* phys addr */
+	sub	%l1, %g7, %l1
+	sub	%l2, %g7, %l2
+
+	mov	%g3, %l3 /* PAGE_SIZE-8 */
+copy_loop:
+	ldxa	[%l1 + %l3] ASI_PHYS_USE_EC, %g2
+	stxa	%g2, [%l2 + %l3] ASI_PHYS_USE_EC
+	cmp	%l3, %g0
+	bne	copy_loop
+	 sub	%l3, 8, %l3
+
+	/* next pbe */
+	ba	pbe_loop
+	 ldxa	[%l0 + 16] %asi, %l0
+
+restore_ctx:
+	setuw	saved_context, %g3
+
+	/* Restore window regs */
+	wrpr    %g0, 0, %canrestore
+	wrpr    %g0, 0, %otherwin
+	wrpr	%g0, 6, %cansave
+	wrpr    %g0, 0, %cleanwin
+
+	ldxa	[%g3 + SC_REG_CWP] %asi, %g2
+	wrpr	%g2, %cwp
+	ldxa	[%g3 + SC_REG_WSTATE] %asi, %g2
+	wrpr	%g2, %wstate
+	ldxa	[%g3 + SC_REG_FP] %asi, %fp
+
+	/* Restore state regs */
+	ldxa	[%g3 + SC_REG_PSTATE] %asi, %g2
+	wrpr	%g2, %pstate
+	ldxa	[%g3 + SC_REG_TICK] %asi, %g2
+	wrpr	%g2, %tick
+
+	/* Restore global regs */
+	ldxa	[%g3 + SC_REG_G4] %asi, %g4
+	ldxa	[%g3 + SC_REG_G5] %asi, %g5
+	ldxa	[%g3 + SC_REG_G6] %asi, %g6
+
+	wr	%g1, %g0, %asi
+
+	restore
+	restore
+
+	wrpr	%g0, 14, %pil
+
+	retl
+	 mov	%g0, %o0
+
+fill_itlb:
+	ba	pbe_loop
+	 wrpr	%g0, 15, %pil