Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6: (230 commits)
  [SPARC64]: Update defconfig.
  [SPARC64]: Fix 2 bugs in huge page support.
  [SPARC64]: CONFIG_BLK_DEV_RAM fix
  [SPARC64]: Optimized TSB table initialization.
  [SPARC64]: Allow CONFIG_MEMORY_HOTPLUG to build.
  [SPARC64]: Use SLAB caches for TSB tables.
  [SPARC64]: Don't kill the page allocator when growing a TSB.
  [SPARC64]: Randomize mm->mmap_base when PF_RANDOMIZE is set.
  [SPARC64]: Increase top of 32-bit process stack.
  [SPARC64]: Top-down address space allocation for 32-bit tasks.
  [SPARC64] bbc_i2c: Fix cpu check and add missing module license.
  [SPARC64]: Fix and re-enable dynamic TSB sizing.
  [SUNSU]: Fix missing spinlock initialization.
  [TG3]: Do not try to access NIC_SRAM_DATA_SIG on Sun parts.
  [SPARC64]: First cut at VIS simulator for Niagara.
  [SPARC64]: Fix system type in /proc/cpuinfo and remove bogus OBP check.
  [SPARC64]: Add SMT scheduling support for Niagara.
  [SPARC64]: Fix 32-bit truncation which broke sparsemem.
  [SPARC64]: Move over to sparsemem.
  [SPARC64]: Fix new context version SMP handling.
  ...

arch/sparc/kernel/ioport.c

@@ -217,7 +217,7 @@ static void _sparc_free_io(struct resource *res)
 	unsigned long plen;
 
 	plen = res->end - res->start + 1;
-	if ((plen & (PAGE_SIZE-1)) != 0) BUG();
+	BUG_ON((plen & (PAGE_SIZE-1)) != 0);
 	sparc_unmapiorange(res->start, plen);
 	release_resource(res);
 }
@@ -512,8 +512,7 @@ void pci_free_consistent(struct pci_dev *pdev, size_t n, void *p, dma_addr_t ba)
 dma_addr_t pci_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
     int direction)
 {
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	/* IIep is write-through, not flushing. */
 	return virt_to_phys(ptr);
 }
@@ -528,8 +527,7 @@ dma_addr_t pci_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
 void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t ba, size_t size,
     int direction)
 {
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	if (direction != PCI_DMA_TODEVICE) {
 		mmu_inval_dma_area((unsigned long)phys_to_virt(ba),
 		    (size + PAGE_SIZE-1) & PAGE_MASK);
@@ -542,8 +540,7 @@ void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t ba, size_t size,
 dma_addr_t pci_map_page(struct pci_dev *hwdev, struct page *page,
 			unsigned long offset, size_t size, int direction)
 {
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	/* IIep is write-through, not flushing. */
 	return page_to_phys(page) + offset;
 }
@@ -551,8 +548,7 @@ dma_addr_t pci_map_page(struct pci_dev *hwdev, struct page *page,
 void pci_unmap_page(struct pci_dev *hwdev,
 			dma_addr_t dma_address, size_t size, int direction)
 {
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	/* mmu_inval_dma_area XXX */
 }
 
@@ -576,11 +572,10 @@ int pci_map_sg(struct pci_dev *hwdev, struct scatterlist *sg, int nents,
 {
 	int n;
 
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	/* IIep is write-through, not flushing. */
 	for (n = 0; n < nents; n++) {
-		if (page_address(sg->page) == NULL) BUG();
+		BUG_ON(page_address(sg->page) == NULL);
 		sg->dvma_address = virt_to_phys(page_address(sg->page));
 		sg->dvma_length = sg->length;
 		sg++;
@@ -597,11 +592,10 @@ void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sg, int nents,
 {
 	int n;
 
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	if (direction != PCI_DMA_TODEVICE) {
 		for (n = 0; n < nents; n++) {
-			if (page_address(sg->page) == NULL) BUG();
+			BUG_ON(page_address(sg->page) == NULL);
 			mmu_inval_dma_area(
 			    (unsigned long) page_address(sg->page),
 			    (sg->length + PAGE_SIZE-1) & PAGE_MASK);
@@ -622,8 +616,7 @@ void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sg, int nents,
  */
 void pci_dma_sync_single_for_cpu(struct pci_dev *hwdev, dma_addr_t ba, size_t size, int direction)
 {
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	if (direction != PCI_DMA_TODEVICE) {
 		mmu_inval_dma_area((unsigned long)phys_to_virt(ba),
 		    (size + PAGE_SIZE-1) & PAGE_MASK);
@@ -632,8 +625,7 @@ void pci_dma_sync_single_for_cpu(struct pci_dev *hwdev, dma_addr_t ba, size_t si
 
 void pci_dma_sync_single_for_device(struct pci_dev *hwdev, dma_addr_t ba, size_t size, int direction)
 {
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	if (direction != PCI_DMA_TODEVICE) {
 		mmu_inval_dma_area((unsigned long)phys_to_virt(ba),
 		    (size + PAGE_SIZE-1) & PAGE_MASK);
@@ -650,11 +642,10 @@ void pci_dma_sync_sg_for_cpu(struct pci_dev *hwdev, struct scatterlist *sg, int
 {
 	int n;
 
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	if (direction != PCI_DMA_TODEVICE) {
 		for (n = 0; n < nents; n++) {
-			if (page_address(sg->page) == NULL) BUG();
+			BUG_ON(page_address(sg->page) == NULL);
 			mmu_inval_dma_area(
 			    (unsigned long) page_address(sg->page),
 			    (sg->length + PAGE_SIZE-1) & PAGE_MASK);
@@ -667,11 +658,10 @@ void pci_dma_sync_sg_for_device(struct pci_dev *hwdev, struct scatterlist *sg, i
 {
 	int n;
 
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	BUG_ON(direction == PCI_DMA_NONE);
 	if (direction != PCI_DMA_TODEVICE) {
 		for (n = 0; n < nents; n++) {
-			if (page_address(sg->page) == NULL) BUG();
+			BUG_ON(page_address(sg->page) == NULL);
 			mmu_inval_dma_area(
 			    (unsigned long) page_address(sg->page),
 			    (sg->length + PAGE_SIZE-1) & PAGE_MASK);

arch/sparc64/Kconfig

@@ -186,6 +186,15 @@ endchoice
 
 endmenu
 
+config ARCH_SPARSEMEM_ENABLE
+	def_bool y
+
+config ARCH_SPARSEMEM_DEFAULT
+	def_bool y
+
+config LARGE_ALLOCS
+	def_bool y
+
 source "mm/Kconfig"
 
 config GENERIC_ISA_DMA
@@ -350,6 +359,15 @@ config SOLARIS_EMUL
 
 endmenu
 
+config SCHED_SMT
+	bool "SMT (Hyperthreading) scheduler support"
+	depends on SMP
+	default y
+	help
+	  SMT scheduler support improves the CPU scheduler's decision making
+	  when dealing with UltraSPARC cpus at a cost of slightly increased
+	  overhead in some places. If unsure say N here.
+
 config CMDLINE_BOOL
 	bool "Default bootloader kernel arguments"
 

arch/sparc64/defconfig

 #
 # Automatically generated make config: don't edit
-# Linux kernel version: 2.6.16-rc2
-# Tue Feb  7 17:47:18 2006
+# Linux kernel version: 2.6.16
+# Mon Mar 20 01:23:21 2006
 #
 CONFIG_SPARC=y
 CONFIG_SPARC64=y
@@ -115,14 +115,20 @@ CONFIG_GENERIC_CALIBRATE_DELAY=y
 CONFIG_HUGETLB_PAGE_SIZE_4MB=y
 # CONFIG_HUGETLB_PAGE_SIZE_512K is not set
 # CONFIG_HUGETLB_PAGE_SIZE_64K is not set
+CONFIG_ARCH_SPARSEMEM_ENABLE=y
+CONFIG_ARCH_SPARSEMEM_DEFAULT=y
+CONFIG_LARGE_ALLOCS=y
 CONFIG_SELECT_MEMORY_MODEL=y
-CONFIG_FLATMEM_MANUAL=y
+# CONFIG_FLATMEM_MANUAL is not set
 # CONFIG_DISCONTIGMEM_MANUAL is not set
-# CONFIG_SPARSEMEM_MANUAL is not set
-CONFIG_FLATMEM=y
-CONFIG_FLAT_NODE_MEM_MAP=y
+CONFIG_SPARSEMEM_MANUAL=y
+CONFIG_SPARSEMEM=y
+CONFIG_HAVE_MEMORY_PRESENT=y
 # CONFIG_SPARSEMEM_STATIC is not set
+CONFIG_SPARSEMEM_EXTREME=y
+CONFIG_MEMORY_HOTPLUG=y
 CONFIG_SPLIT_PTLOCK_CPUS=4
+CONFIG_MIGRATION=y
 CONFIG_GENERIC_ISA_DMA=y
 CONFIG_SBUS=y
 CONFIG_SBUSCHAR=y
@@ -655,6 +661,7 @@ CONFIG_SERIAL_SUNCORE=y
 CONFIG_SERIAL_SUNSU=y
 CONFIG_SERIAL_SUNSU_CONSOLE=y
 CONFIG_SERIAL_SUNSAB=m
+CONFIG_SERIAL_SUNHV=y
 CONFIG_SERIAL_CORE=y
 CONFIG_SERIAL_CORE_CONSOLE=y
 # CONFIG_SERIAL_JSM is not set
@@ -1116,11 +1123,7 @@ CONFIG_USB_HIDDEV=y
 # CONFIG_INFINIBAND is not set
 
 #
-# SN Devices
-#
-
-#
-# EDAC - error detection and reporting (RAS)
+# EDAC - error detection and reporting (RAS) (EXPERIMENTAL)
 #
 
 #

arch/sparc64/kernel/Makefile

@@ -11,10 +11,12 @@ obj-y		:= process.o setup.o cpu.o idprom.o \
 		   traps.o devices.o auxio.o una_asm.o \
 		   irq.o ptrace.o time.o sys_sparc.o signal.o \
 		   unaligned.o central.o pci.o starfire.o semaphore.o \
-		   power.o sbus.o iommu_common.o sparc64_ksyms.o chmc.o
+		   power.o sbus.o iommu_common.o sparc64_ksyms.o chmc.o \
+		   visemul.o
 
 obj-$(CONFIG_PCI)	 += ebus.o isa.o pci_common.o pci_iommu.o \
-			    pci_psycho.o pci_sabre.o pci_schizo.o
+			    pci_psycho.o pci_sabre.o pci_schizo.o \
+			    pci_sun4v.o pci_sun4v_asm.o
 obj-$(CONFIG_SMP)	 += smp.o trampoline.o
 obj-$(CONFIG_SPARC32_COMPAT) += sys32.o sys_sparc32.o signal32.o
 obj-$(CONFIG_BINFMT_ELF32) += binfmt_elf32.o
@@ -38,5 +40,5 @@ else
   CMODEL_CFLAG := -m64 -mcmodel=medlow
 endif
 
-head.o: head.S ttable.S itlb_base.S dtlb_base.S dtlb_backend.S dtlb_prot.S \
+head.o: head.S ttable.S itlb_miss.S dtlb_miss.S ktlb.S tsb.S \
 	etrap.S rtrap.S winfixup.S entry.S

arch/sparc64/kernel/binfmt_aout32.c

@@ -31,6 +31,7 @@
 #include <asm/system.h>
 #include <asm/uaccess.h>
 #include <asm/pgalloc.h>
+#include <asm/mmu_context.h>
 
 static int load_aout32_binary(struct linux_binprm *, struct pt_regs * regs);
 static int load_aout32_library(struct file*);
@@ -238,6 +239,8 @@ static int load_aout32_binary(struct linux_binprm * bprm, struct pt_regs * regs)
 		(current->mm->start_data = N_DATADDR(ex));
 	current->mm->brk = ex.a_bss +
 		(current->mm->start_brk = N_BSSADDR(ex));
+	current->mm->free_area_cache = current->mm->mmap_base;
+	current->mm->cached_hole_size = 0;
 
 	current->mm->mmap = NULL;
 	compute_creds(bprm);
@@ -329,15 +332,8 @@ beyond_if:
 
 	current->mm->start_stack =
 		(unsigned long) create_aout32_tables((char __user *)bprm->p, bprm);
-	if (!(orig_thr_flags & _TIF_32BIT)) {
-		unsigned long pgd_cache = get_pgd_cache(current->mm->pgd);
-
-		__asm__ __volatile__("stxa\t%0, [%1] %2\n\t"
-				     "membar #Sync"
-				     : /* no outputs */
-				     : "r" (pgd_cache),
-				       "r" (TSB_REG), "i" (ASI_DMMU));
-	}
+	tsb_context_switch(current->mm);
+
 	start_thread32(regs, ex.a_entry, current->mm->start_stack);
 	if (current->ptrace & PT_PTRACED)
 		send_sig(SIGTRAP, current, 0);

arch/sparc64/kernel/binfmt_elf32.c

@@ -153,7 +153,9 @@ MODULE_AUTHOR("Eric Youngdale, David S. Miller, Jakub Jelinek");
 #undef MODULE_DESCRIPTION
 #undef MODULE_AUTHOR
 
+#include <asm/a.out.h>
+
 #undef TASK_SIZE
-#define TASK_SIZE 0xf0000000
+#define TASK_SIZE STACK_TOP32
 
 #include "../../../fs/binfmt_elf.c"

arch/sparc64/kernel/cpu.c

@@ -13,6 +13,7 @@
 #include <asm/system.h>
 #include <asm/fpumacro.h>
 #include <asm/cpudata.h>
+#include <asm/spitfire.h>
 
 DEFINE_PER_CPU(cpuinfo_sparc, __cpu_data) = { 0 };
 
@@ -71,6 +72,12 @@ void __init cpu_probe(void)
 	unsigned long ver, fpu_vers, manuf, impl, fprs;
 	int i;
 	
+	if (tlb_type == hypervisor) {
+		sparc_cpu_type = "UltraSparc T1 (Niagara)";
+		sparc_fpu_type = "UltraSparc T1 integrated FPU";
+		return;
+	}
+
 	fprs = fprs_read();
 	fprs_write(FPRS_FEF);
 	__asm__ __volatile__ ("rdpr %%ver, %0; stx %%fsr, [%1]"

arch/sparc64/kernel/devices.c

@@ -12,6 +12,7 @@
 #include <linux/string.h>
 #include <linux/spinlock.h>
 #include <linux/errno.h>
+#include <linux/bootmem.h>
 
 #include <asm/page.h>
 #include <asm/oplib.h>
@@ -20,6 +21,8 @@
 #include <asm/spitfire.h>
 #include <asm/timer.h>
 #include <asm/cpudata.h>
+#include <asm/vdev.h>
+#include <asm/irq.h>
 
 /* Used to synchronize acceses to NatSemi SUPER I/O chip configure
  * operations in asm/ns87303.h
@@ -29,13 +32,158 @@ DEFINE_SPINLOCK(ns87303_lock);
 extern void cpu_probe(void);
 extern void central_probe(void);
 
-static char *cpu_mid_prop(void)
+u32 sun4v_vdev_devhandle;
+int sun4v_vdev_root;
+
+struct vdev_intmap {
+	unsigned int phys;
+	unsigned int irq;
+	unsigned int cnode;
+	unsigned int cinterrupt;
+};
+
+struct vdev_intmask {
+	unsigned int phys;
+	unsigned int interrupt;
+	unsigned int __unused;
+};
+
+static struct vdev_intmap *vdev_intmap;
+static int vdev_num_intmap;
+static struct vdev_intmask vdev_intmask;
+
+static void __init sun4v_virtual_device_probe(void)
+{
+	struct linux_prom64_registers regs;
+	struct vdev_intmap *ip;
+	int node, sz, err;
+
+	if (tlb_type != hypervisor)
+		return;
+
+	node = prom_getchild(prom_root_node);
+	node = prom_searchsiblings(node, "virtual-devices");
+	if (!node) {
+		prom_printf("SUN4V: Fatal error, no virtual-devices node.\n");
+		prom_halt();
+	}
+
+	sun4v_vdev_root = node;
+
+	prom_getproperty(node, "reg", (char *)&regs, sizeof(regs));
+	sun4v_vdev_devhandle = (regs.phys_addr >> 32UL) & 0x0fffffff;
+
+	sz = prom_getproplen(node, "interrupt-map");
+	if (sz <= 0) {
+		prom_printf("SUN4V: Error, no vdev interrupt-map.\n");
+		prom_halt();
+	}
+
+	if ((sz % sizeof(*ip)) != 0) {
+		prom_printf("SUN4V: Bogus interrupt-map property size %d\n",
+			    sz);
+		prom_halt();
+	}
+
+	vdev_intmap = ip = alloc_bootmem_low_pages(sz);
+	if (!vdev_intmap) {
+		prom_printf("SUN4V: Error, cannot allocate vdev_intmap.\n");
+		prom_halt();
+	}
+
+	err = prom_getproperty(node, "interrupt-map", (char *) ip, sz);
+	if (err == -1) {
+		prom_printf("SUN4V: Fatal error, no vdev interrupt-map.\n");
+		prom_halt();
+	}
+	if (err != sz) {
+		prom_printf("SUN4V: Inconsistent interrupt-map size, "
+			    "proplen(%d) vs getprop(%d).\n", sz,err);
+		prom_halt();
+	}
+
+	vdev_num_intmap = err / sizeof(*ip);
+
+	err = prom_getproperty(node, "interrupt-map-mask",
+			       (char *) &vdev_intmask,
+			       sizeof(vdev_intmask));
+	if (err <= 0) {
+		prom_printf("SUN4V: Fatal error, no vdev "
+			    "interrupt-map-mask.\n");
+		prom_halt();
+	}
+	if (err % sizeof(vdev_intmask)) {
+		prom_printf("SUN4V: Bogus interrupt-map-mask "
+			    "property size %d\n", err);
+		prom_halt();
+	}
+
+	printk("SUN4V: virtual-devices devhandle[%x]\n",
+	       sun4v_vdev_devhandle);
+}
+
+unsigned int sun4v_vdev_device_interrupt(unsigned int dev_node)
+{
+	unsigned int irq, reg;
+	int err, i;
+
+	err = prom_getproperty(dev_node, "interrupts",
+			       (char *) &irq, sizeof(irq));
+	if (err <= 0) {
+		printk("VDEV: Cannot get \"interrupts\" "
+		       "property for OBP node %x\n", dev_node);
+		return 0;
+	}
+
+	err = prom_getproperty(dev_node, "reg",
+			       (char *) &reg, sizeof(reg));
+	if (err <= 0) {
+		printk("VDEV: Cannot get \"reg\" "
+		       "property for OBP node %x\n", dev_node);
+		return 0;
+	}
+
+	for (i = 0; i < vdev_num_intmap; i++) {
+		if (vdev_intmap[i].phys == (reg & vdev_intmask.phys) &&
+		    vdev_intmap[i].irq == (irq & vdev_intmask.interrupt)) {
+			irq = vdev_intmap[i].cinterrupt;
+			break;
+		}
+	}
+
+	if (i == vdev_num_intmap) {
+		printk("VDEV: No matching interrupt map entry "
+		       "for OBP node %x\n", dev_node);
+		return 0;
+	}
+
+	return sun4v_build_irq(sun4v_vdev_devhandle, irq, 5, 0);
+}
+
+static const char *cpu_mid_prop(void)
 {
 	if (tlb_type == spitfire)
 		return "upa-portid";
 	return "portid";
 }
 
+static int get_cpu_mid(int prom_node)
+{
+	if (tlb_type == hypervisor) {
+		struct linux_prom64_registers reg;
+
+		if (prom_getproplen(prom_node, "cpuid") == 4)
+			return prom_getintdefault(prom_node, "cpuid", 0);
+
+		prom_getproperty(prom_node, "reg", (char *) &reg, sizeof(reg));
+		return (reg.phys_addr >> 32) & 0x0fffffffUL;
+	} else {
+		const char *prop_name = cpu_mid_prop();
+
+		return prom_getintdefault(prom_node, prop_name, 0);
+	}
+}
+
 static int check_cpu_node(int nd, int *cur_inst,
 			  int (*compare)(int, int, void *), void *compare_arg,
 			  int *prom_node, int *mid)
@@ -50,7 +198,7 @@ static int check_cpu_node(int nd, int *cur_inst,
 		if (prom_node)
 			*prom_node = nd;
 		if (mid)
-			*mid = prom_getintdefault(nd, cpu_mid_prop(), 0);
+			*mid = get_cpu_mid(nd);
 		return 0;
 	}
 
@@ -105,7 +253,7 @@ static int cpu_mid_compare(int nd, int instance, void *_arg)
 	int desired_mid = (int) (long) _arg;
 	int this_mid;
 
-	this_mid = prom_getintdefault(nd, cpu_mid_prop(), 0);
+	this_mid = get_cpu_mid(nd);
 	if (this_mid == desired_mid)
 		return 0;
 	return -ENODEV;
@@ -126,7 +274,8 @@ void __init device_scan(void)
 
 #ifndef CONFIG_SMP
 	{
-		int err, cpu_node;
+		int err, cpu_node, def;
+
 		err = cpu_find_by_instance(0, &cpu_node, NULL);
 		if (err) {
 			prom_printf("No cpu nodes, cannot continue\n");
@@ -135,21 +284,40 @@ void __init device_scan(void)
 		cpu_data(0).clock_tick = prom_getintdefault(cpu_node,
 							    "clock-frequency",
 							    0);
+
+		def = ((tlb_type == hypervisor) ?
+		       (8 * 1024) :
+		       (16 * 1024));
 		cpu_data(0).dcache_size = prom_getintdefault(cpu_node,
 							     "dcache-size",
-							     16 * 1024);
+							     def);
+
+		def = 32;
 		cpu_data(0).dcache_line_size =
-			prom_getintdefault(cpu_node, "dcache-line-size", 32);
+			prom_getintdefault(cpu_node, "dcache-line-size",
+					   def);
+
+		def = 16 * 1024;
 		cpu_data(0).icache_size = prom_getintdefault(cpu_node,
 							     "icache-size",
-							     16 * 1024);
+							     def);
+
+		def = 32;
 		cpu_data(0).icache_line_size =
-			prom_getintdefault(cpu_node, "icache-line-size", 32);
+			prom_getintdefault(cpu_node, "icache-line-size",
+					   def);
+
+		def = ((tlb_type == hypervisor) ?
+		       (3 * 1024 * 1024) :
+		       (4 * 1024 * 1024));
 		cpu_data(0).ecache_size = prom_getintdefault(cpu_node,
 							     "ecache-size",
-							     4 * 1024 * 1024);
+							     def);
+
+		def = 64;
 		cpu_data(0).ecache_line_size =
-			prom_getintdefault(cpu_node, "ecache-line-size", 64);
+			prom_getintdefault(cpu_node, "ecache-line-size",
+					   def);
 		printk("CPU[0]: Caches "
 		       "D[sz(%d):line_sz(%d)] "
 		       "I[sz(%d):line_sz(%d)] "
@@ -160,6 +328,7 @@ void __init device_scan(void)
 	}
 #endif
 
+	sun4v_virtual_device_probe();
 	central_probe();
 
 	cpu_probe();

arch/sparc64/kernel/dtlb_backend.S

-/* $Id: dtlb_backend.S,v 1.16 2001/10/09 04:02:11 davem Exp $
- * dtlb_backend.S: Back end to DTLB miss replacement strategy.
- *                 This is included directly into the trap table.
- *
- * Copyright (C) 1996,1998 David S. Miller (davem@redhat.com)
- * Copyright (C) 1997,1998 Jakub Jelinek   (jj@ultra.linux.cz)
- */
-
-#include <asm/pgtable.h>
-#include <asm/mmu.h>
-
-#define VALID_SZ_BITS	(_PAGE_VALID | _PAGE_SZBITS)
-
-#define VPTE_BITS		(_PAGE_CP | _PAGE_CV | _PAGE_P )
-#define VPTE_SHIFT		(PAGE_SHIFT - 3)
-
-/* Ways we can get here:
- *
- * 1) Nucleus loads and stores to/from PA-->VA direct mappings at tl>1.
- * 2) Nucleus loads and stores to/from user/kernel window save areas.
- * 3) VPTE misses from dtlb_base and itlb_base.
- *
- * We need to extract out the PMD and PGDIR indexes from the
- * linear virtual page table access address.  The PTE index
- * is at the bottom, but we are not concerned with it.  Bits
- * 0 to 2 are clear since each PTE is 8 bytes in size.  Each
- * PMD and PGDIR entry are 4 bytes in size.   Thus, this
- * address looks something like:
- *
- * |---------------------------------------------------------------|
- * |  ...   |    PGDIR index    |    PMD index    | PTE index  |   |
- * |---------------------------------------------------------------|
- *   63   F   E               D   C             B   A         3 2 0  <- bit nr
- *
- *  The variable bits above are defined as:
- *  A --> 3 + (PAGE_SHIFT - log2(8))
- *    --> 3 + (PAGE_SHIFT - 3) - 1
- *        (ie. this is "bit 3" + PAGE_SIZE - size of PTE entry in bits - 1)
- *  B --> A + 1
- *  C --> B + (PAGE_SHIFT - log2(4))
- *    -->  B + (PAGE_SHIFT - 2) - 1
- *        (ie. this is "bit B" + PAGE_SIZE - size of PMD entry in bits - 1)
- *  D --> C + 1
- *  E --> D + (PAGE_SHIFT - log2(4))
- *    --> D + (PAGE_SHIFT - 2) - 1
- *        (ie. this is "bit D" + PAGE_SIZE - size of PGDIR entry in bits - 1)
- *  F --> E + 1
- *
- * (Note how "B" always evalutes to PAGE_SHIFT, all the other constants
- *  cancel out.)
- *
- * For 8K PAGE_SIZE (thus, PAGE_SHIFT of 13) the bit numbers are:
- * A --> 12
- * B --> 13
- * C --> 23
- * D --> 24
- * E --> 34
- * F --> 35
- *
- * For 64K PAGE_SIZE (thus, PAGE_SHIFT of 16) the bit numbers are:
- * A --> 15
- * B --> 16
- * C --> 29
- * D --> 30
- * E --> 43
- * F --> 44
- *
- * Because bits both above and below each PGDIR and PMD index need to
- * be masked out, and the index can be as long as 14 bits (when using a
- * 64K PAGE_SIZE, and thus a PAGE_SHIFT of 16), we need 3 instructions
- * to extract each index out.
- *
- * Shifts do not pair very well on UltraSPARC-I, II, IIi, and IIe, so
- * we try to avoid using them for the entire operation.  We could setup
- * a mask anywhere from bit 31 down to bit 10 using the sethi instruction.
- *
- * We need a mask covering bits B --> C and one covering D --> E.
- * For 8K PAGE_SIZE these masks are 0x00ffe000 and 0x7ff000000.
- * For 64K PAGE_SIZE these masks are 0x3fff0000 and 0xfffc0000000.
- * The second in each set cannot be loaded with a single sethi
- * instruction, because the upper bits are past bit 32.  We would
- * need to use a sethi + a shift.
- *
- * For the time being, we use 2 shifts and a simple "and" mask.
- * We shift left to clear the bits above the index, we shift down
- * to clear the bits below the index (sans the log2(4 or 8) bits)
- * and a mask to clear the log2(4 or 8) bits.  We need therefore
- * define 4 shift counts, all of which are relative to PAGE_SHIFT.
- *
- * Although unsupportable for other reasons, this does mean that
- * 512K and 4MB page sizes would be generaally supported by the
- * kernel.  (ELF binaries would break with > 64K PAGE_SIZE since
- * the sections are only aligned that strongly).
- *
- * The operations performed for extraction are thus:
- *
- *      ((X << FOO_SHIFT_LEFT) >> FOO_SHIFT_RIGHT) & ~0x3
- *
- */
-
-#define A (3 + (PAGE_SHIFT - 3) - 1)
-#define B (A + 1)
-#define C (B + (PAGE_SHIFT - 2) - 1)
-#define D (C + 1)
-#define E (D + (PAGE_SHIFT - 2) - 1)
-#define F (E + 1)
-
-#define PMD_SHIFT_LEFT		(64 - D)
-#define PMD_SHIFT_RIGHT		(64 - (D - B) - 2)
-#define PGDIR_SHIFT_LEFT 	(64 - F)
-#define PGDIR_SHIFT_RIGHT	(64 - (F - D) - 2)
-#define LOW_MASK_BITS		0x3
-
-/* TLB1 ** ICACHE line 1: tl1 DTLB and quick VPTE miss	*/
-	ldxa		[%g1 + %g1] ASI_DMMU, %g4	! Get TAG_ACCESS
-	add		%g3, %g3, %g5			! Compute VPTE base
-	cmp		%g4, %g5			! VPTE miss?
-	bgeu,pt		%xcc, 1f			! Continue here
-	 andcc		%g4, TAG_CONTEXT_BITS, %g5	! tl0 miss Nucleus test
-	ba,a,pt		%xcc, from_tl1_trap		! Fall to tl0 miss
-1:	sllx		%g6, VPTE_SHIFT, %g4		! Position TAG_ACCESS
-	or		%g4, %g5, %g4			! Prepare TAG_ACCESS
-
-/* TLB1 ** ICACHE line 2: Quick VPTE miss	  	*/
-	mov		TSB_REG, %g1			! Grab TSB reg
-	ldxa		[%g1] ASI_DMMU, %g5		! Doing PGD caching?
-	sllx		%g6, PMD_SHIFT_LEFT, %g1	! Position PMD offset
-	be,pn		%xcc, sparc64_vpte_nucleus	! Is it from Nucleus?
-	 srlx		%g1, PMD_SHIFT_RIGHT, %g1	! Mask PMD offset bits
-	brnz,pt		%g5, sparc64_vpte_continue	! Yep, go like smoke
-	 andn		%g1, LOW_MASK_BITS, %g1		! Final PMD mask
-	sllx		%g6, PGDIR_SHIFT_LEFT, %g5	! Position PGD offset
-
-/* TLB1 ** ICACHE line 3: Quick VPTE miss	  	*/
-	srlx		%g5, PGDIR_SHIFT_RIGHT, %g5	! Mask PGD offset bits
-	andn		%g5, LOW_MASK_BITS, %g5		! Final PGD mask
-	lduwa		[%g7 + %g5] ASI_PHYS_USE_EC, %g5! Load PGD
-	brz,pn		%g5, vpte_noent			! Valid?
-sparc64_kpte_continue:
-	 sllx		%g5, 11, %g5			! Shift into place
-sparc64_vpte_continue:
-	lduwa		[%g5 + %g1] ASI_PHYS_USE_EC, %g5! Load PMD
-	sllx		%g5, 11, %g5			! Shift into place
-	brz,pn		%g5, vpte_noent			! Valid?
-
-/* TLB1 ** ICACHE line 4: Quick VPTE miss	  	*/
-	 mov		(VALID_SZ_BITS >> 61), %g1	! upper vpte into %g1
-	sllx		%g1, 61, %g1			! finish calc
-	or		%g5, VPTE_BITS, %g5		! Prepare VPTE data
-	or		%g5, %g1, %g5			! ...
-	mov		TLB_SFSR, %g1			! Restore %g1 value
-	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Load VPTE into TLB
-	stxa		%g4, [%g1 + %g1] ASI_DMMU	! Restore previous TAG_ACCESS
-	retry						! Load PTE once again
-
-#undef VALID_SZ_BITS
-#undef VPTE_SHIFT
-#undef VPTE_BITS
-#undef A
-#undef B
-#undef C
-#undef D
-#undef E
-#undef F
-#undef PMD_SHIFT_LEFT
-#undef PMD_SHIFT_RIGHT
-#undef PGDIR_SHIFT_LEFT
-#undef PGDIR_SHIFT_RIGHT
-#undef LOW_MASK_BITS
-

arch/sparc64/kernel/dtlb_base.S

-/* $Id: dtlb_base.S,v 1.17 2001/10/11 22:33:52 davem Exp $
- * dtlb_base.S:	Front end to DTLB miss replacement strategy.
- *              This is included directly into the trap table.
- *
- * Copyright (C) 1996,1998 David S. Miller (davem@redhat.com)
- * Copyright (C) 1997,1998 Jakub Jelinek   (jj@ultra.linux.cz)
- */
-
-#include <asm/pgtable.h>
-#include <asm/mmu.h>
-
-/* %g1	TLB_SFSR	(%g1 + %g1 == TLB_TAG_ACCESS)
- * %g2	(KERN_HIGHBITS | KERN_LOWBITS)
- * %g3  VPTE base	(0xfffffffe00000000)	Spitfire/Blackbird (44-bit VA space)
- *			(0xffe0000000000000)	Cheetah		   (64-bit VA space)
- * %g7	__pa(current->mm->pgd)
- *
- * The VPTE base value is completely magic, but note that
- * few places in the kernel other than these TLB miss
- * handlers know anything about the VPTE mechanism or
- * how it works (see VPTE_SIZE, TASK_SIZE and PTRS_PER_PGD).
- * Consider the 44-bit VADDR Ultra-I/II case as an example:
- *
- * VA[0 :  (1<<43)] produce VPTE index [%g3                        :   0]
- * VA[0 : -(1<<43)] produce VPTE index [%g3-(1<<(43-PAGE_SHIFT+3)) : %g3]
- *
- * For Cheetah's 64-bit VADDR space this is:
- *
- * VA[0 :  (1<<63)] produce VPTE index [%g3                        :   0]
- * VA[0 : -(1<<63)] produce VPTE index [%g3-(1<<(63-PAGE_SHIFT+3)) : %g3]
- *
- * If you're paying attention you'll notice that this means half of
- * the VPTE table is above %g3 and half is below, low VA addresses
- * map progressively upwards from %g3, and high VA addresses map
- * progressively upwards towards %g3.  This trick was needed to make
- * the same 8 instruction handler work both for Spitfire/Blackbird's
- * peculiar VA space hole configuration and the full 64-bit VA space
- * one of Cheetah at the same time.
- */
-
-/* Ways we can get here:
- *
- * 1) Nucleus loads and stores to/from PA-->VA direct mappings.
- * 2) Nucleus loads and stores to/from vmalloc() areas.
- * 3) User loads and stores.
- * 4) User space accesses by nucleus at tl0
- */
-
-#if PAGE_SHIFT == 13
-/*
- * To compute vpte offset, we need to do ((addr >> 13) << 3),
- * which can be optimized to (addr >> 10) if bits 10/11/12 can
- * be guaranteed to be 0 ... mmu_context.h does guarantee this
- * by only using 10 bits in the hwcontext value.
- */
-#define CREATE_VPTE_OFFSET1(r1, r2) nop
-#define CREATE_VPTE_OFFSET2(r1, r2) \
-				srax	r1, 10, r2
-#else
-#define CREATE_VPTE_OFFSET1(r1, r2) \
-				srax	r1, PAGE_SHIFT, r2
-#define CREATE_VPTE_OFFSET2(r1, r2) \
-				sllx	r2, 3, r2
-#endif
-
-/* DTLB ** ICACHE line 1: Quick user TLB misses		*/
-	mov		TLB_SFSR, %g1
-	ldxa		[%g1 + %g1] ASI_DMMU, %g4	! Get TAG_ACCESS
-	andcc		%g4, TAG_CONTEXT_BITS, %g0	! From Nucleus?
-from_tl1_trap:
-	rdpr		%tl, %g5			! For TL==3 test
-	CREATE_VPTE_OFFSET1(%g4, %g6)			! Create VPTE offset
-	be,pn		%xcc, kvmap			! Yep, special processing
-	 CREATE_VPTE_OFFSET2(%g4, %g6)			! Create VPTE offset
-	cmp		%g5, 4				! Last trap level?
-
-/* DTLB ** ICACHE line 2: User finish + quick kernel TLB misses	*/
-	be,pn		%xcc, longpath			! Yep, cannot risk VPTE miss
-	 nop						! delay slot
-	ldxa		[%g3 + %g6] ASI_S, %g5		! Load VPTE
-1:	brgez,pn	%g5, longpath			! Invalid, branch out
-	 nop						! Delay-slot
-9:	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Reload TLB
-	retry						! Trap return
-	nop
-
-/* DTLB ** ICACHE line 3: winfixups+real_faults		*/
-longpath:
-	rdpr		%pstate, %g5			! Move into alternate globals
-	wrpr		%g5, PSTATE_AG|PSTATE_MG, %pstate
-	rdpr		%tl, %g4			! See where we came from.
-	cmp		%g4, 1				! Is etrap/rtrap window fault?
-	mov		TLB_TAG_ACCESS, %g4		! Prepare for fault processing
-	ldxa		[%g4] ASI_DMMU, %g5		! Load faulting VA page
-	be,pt		%xcc, sparc64_realfault_common	! Jump to normal fault handling
-	 mov		FAULT_CODE_DTLB, %g4		! It was read from DTLB
-
-/* DTLB ** ICACHE line 4: Unused...	*/
-	ba,a,pt		%xcc, winfix_trampoline		! Call window fixup code
-	nop
-	nop
-	nop
-	nop
-	nop
-	nop
-	nop
-
-#undef CREATE_VPTE_OFFSET1
-#undef CREATE_VPTE_OFFSET2

arch/sparc64/kernel/dtlb_miss.S

+/* DTLB ** ICACHE line 1: Context 0 check and TSB load	*/
+	ldxa	[%g0] ASI_DMMU_TSB_8KB_PTR, %g1	! Get TSB 8K pointer
+	ldxa	[%g0] ASI_DMMU, %g6		! Get TAG TARGET
+	srlx	%g6, 48, %g5			! Get context
+	sllx	%g6, 22, %g6			! Zero out context
+	brz,pn	%g5, kvmap_dtlb			! Context 0 processing
+	 srlx	%g6, 22, %g6			! Delay slot
+	TSB_LOAD_QUAD(%g1, %g4)			! Load TSB entry
+	cmp	%g4, %g6			! Compare TAG
+
+/* DTLB ** ICACHE line 2: TSB compare and TLB load	*/
+	bne,pn	%xcc, tsb_miss_dtlb		! Miss
+	 mov	FAULT_CODE_DTLB, %g3
+	stxa	%g5, [%g0] ASI_DTLB_DATA_IN	! Load TLB
+	retry					! Trap done
+	nop
+	nop
+	nop
+	nop
+
+/* DTLB ** ICACHE line 3:				*/
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+
+/* DTLB ** ICACHE line 4: 				*/
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop

arch/sparc64/kernel/ebus.c

@@ -277,10 +277,9 @@ static inline void *ebus_alloc(size_t size)
 {
 	void *mem;
 
-	mem = kmalloc(size, GFP_ATOMIC);
+	mem = kzalloc(size, GFP_ATOMIC);
 	if (!mem)
 		panic("ebus_alloc: out of memory");
-	memset((char *)mem, 0, size);
 	return mem;
 }
 

arch/sparc64/kernel/entry.S

@@ -50,7 +50,8 @@ do_fpdis:
 	add		%g0, %g0, %g0
 	ba,a,pt		%xcc, rtrap_clr_l6
 
-1:	ldub		[%g6 + TI_FPSAVED], %g5
+1:	TRAP_LOAD_THREAD_REG(%g6, %g1)
+	ldub		[%g6 + TI_FPSAVED], %g5
 	wr		%g0, FPRS_FEF, %fprs
 	andcc		%g5, FPRS_FEF, %g0
 	be,a,pt		%icc, 1f
@@ -96,10 +97,22 @@ do_fpdis:
 	add		%g6, TI_FPREGS + 0x80, %g1
 	faddd		%f0, %f2, %f4
 	fmuld		%f0, %f2, %f6
-	ldxa		[%g3] ASI_DMMU, %g5
+
+661:	ldxa		[%g3] ASI_DMMU, %g5
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	ldxa		[%g3] ASI_MMU, %g5
+	.previous
+
 	sethi		%hi(sparc64_kern_sec_context), %g2
 	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
-	stxa		%g2, [%g3] ASI_DMMU
+
+661:	stxa		%g2, [%g3] ASI_DMMU
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	stxa		%g2, [%g3] ASI_MMU
+	.previous
+
 	membar		#Sync
 	add		%g6, TI_FPREGS + 0xc0, %g2
 	faddd		%f0, %f2, %f8
@@ -125,11 +138,23 @@ do_fpdis:
 	 fzero		%f32
 	mov		SECONDARY_CONTEXT, %g3
 	fzero		%f34
-	ldxa		[%g3] ASI_DMMU, %g5
+
+661:	ldxa		[%g3] ASI_DMMU, %g5
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	ldxa		[%g3] ASI_MMU, %g5
+	.previous
+
 	add		%g6, TI_FPREGS, %g1
 	sethi		%hi(sparc64_kern_sec_context), %g2
 	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
-	stxa		%g2, [%g3] ASI_DMMU
+
+661:	stxa		%g2, [%g3] ASI_DMMU
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	stxa		%g2, [%g3] ASI_MMU
+	.previous
+
 	membar		#Sync
 	add		%g6, TI_FPREGS + 0x40, %g2
 	faddd		%f32, %f34, %f36
@@ -154,10 +179,22 @@ do_fpdis:
 	 nop
 3:	mov		SECONDARY_CONTEXT, %g3
 	add		%g6, TI_FPREGS, %g1
-	ldxa		[%g3] ASI_DMMU, %g5
+
+661:	ldxa		[%g3] ASI_DMMU, %g5
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	ldxa		[%g3] ASI_MMU, %g5
+	.previous
+
 	sethi		%hi(sparc64_kern_sec_context), %g2
 	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
-	stxa		%g2, [%g3] ASI_DMMU
+
+661:	stxa		%g2, [%g3] ASI_DMMU
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	stxa		%g2, [%g3] ASI_MMU
+	.previous
+
 	membar		#Sync
 	mov		0x40, %g2
 	membar		#Sync
@@ -168,7 +205,13 @@ do_fpdis:
 	ldda		[%g1 + %g2] ASI_BLK_S, %f48
 	membar		#Sync
 fpdis_exit:
-	stxa		%g5, [%g3] ASI_DMMU
+
+661:	stxa		%g5, [%g3] ASI_DMMU
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	stxa		%g5, [%g3] ASI_MMU
+	.previous
+
 	membar		#Sync
 fpdis_exit2:
 	wr		%g7, 0, %gsr
@@ -189,6 +232,7 @@ fp_other_bounce:
 	.globl		do_fpother_check_fitos
 	.align		32
 do_fpother_check_fitos:
+	TRAP_LOAD_THREAD_REG(%g6, %g1)
 	sethi		%hi(fp_other_bounce - 4), %g7
 	or		%g7, %lo(fp_other_bounce - 4), %g7
 
@@ -312,6 +356,7 @@ fitos_emul_fini:
 	.globl		do_fptrap
 	.align		32
 do_fptrap:
+	TRAP_LOAD_THREAD_REG(%g6, %g1)
 	stx		%fsr, [%g6 + TI_XFSR]
 do_fptrap_after_fsr:
 	ldub		[%g6 + TI_FPSAVED], %g3
@@ -321,10 +366,22 @@ do_fptrap_after_fsr:
 	rd		%gsr, %g3
 	stx		%g3, [%g6 + TI_GSR]
 	mov		SECONDARY_CONTEXT, %g3
-	ldxa		[%g3] ASI_DMMU, %g5
+
+661:	ldxa		[%g3] ASI_DMMU, %g5
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	ldxa		[%g3] ASI_MMU, %g5
+	.previous
+
 	sethi		%hi(sparc64_kern_sec_context), %g2
 	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
-	stxa		%g2, [%g3] ASI_DMMU
+
+661:	stxa		%g2, [%g3] ASI_DMMU
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	stxa		%g2, [%g3] ASI_MMU
+	.previous
+
 	membar		#Sync
 	add		%g6, TI_FPREGS, %g2
 	andcc		%g1, FPRS_DL, %g0
@@ -339,7 +396,13 @@ do_fptrap_after_fsr:
 	stda		%f48, [%g2 + %g3] ASI_BLK_S
 5:	mov		SECONDARY_CONTEXT, %g1
 	membar		#Sync
-	stxa		%g5, [%g1] ASI_DMMU
+
+661:	stxa		%g5, [%g1] ASI_DMMU
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	stxa		%g5, [%g1] ASI_MMU
+	.previous
+
 	membar		#Sync
 	ba,pt		%xcc, etrap
 	 wr		%g0, 0, %fprs
@@ -353,8 +416,6 @@ do_fptrap_after_fsr:
 	 *
 	 * With this method we can do most of the cross-call tlb/cache
 	 * flushing very quickly.
-	 *
-	 * Current CPU's IRQ worklist table is locked into %g6, don't touch.
 	 */
 	.text
 	.align		32
@@ -378,6 +439,8 @@ do_ivec:
 	sllx		%g2, %g4, %g2
 	sllx		%g4, 2, %g4
 
+	TRAP_LOAD_IRQ_WORK(%g6, %g1)
+
 	lduw		[%g6 + %g4], %g5	/* g5 = irq_work(cpu, pil) */
 	stw		%g5, [%g3 + 0x00]	/* bucket->irq_chain = g5 */
 	stw		%g3, [%g6 + %g4]	/* irq_work(cpu, pil) = bucket */
@@ -399,76 +462,6 @@ do_ivec_xcall:
 1:	jmpl		%g3, %g0
 	 nop
 
-	.globl		save_alternate_globals
-save_alternate_globals: /* %o0 = save_area */
-	rdpr		%pstate, %o5
-	andn		%o5, PSTATE_IE, %o1
-	wrpr		%o1, PSTATE_AG, %pstate
-	stx		%g0, [%o0 + 0x00]
-	stx		%g1, [%o0 + 0x08]
-	stx		%g2, [%o0 + 0x10]
-	stx		%g3, [%o0 + 0x18]
-	stx		%g4, [%o0 + 0x20]
-	stx		%g5, [%o0 + 0x28]
-	stx		%g6, [%o0 + 0x30]
-	stx		%g7, [%o0 + 0x38]
-	wrpr		%o1, PSTATE_IG, %pstate
-	stx		%g0, [%o0 + 0x40]
-	stx		%g1, [%o0 + 0x48]
-	stx		%g2, [%o0 + 0x50]
-	stx		%g3, [%o0 + 0x58]
-	stx		%g4, [%o0 + 0x60]
-	stx		%g5, [%o0 + 0x68]
-	stx		%g6, [%o0 + 0x70]
-	stx		%g7, [%o0 + 0x78]
-	wrpr		%o1, PSTATE_MG, %pstate
-	stx		%g0, [%o0 + 0x80]
-	stx		%g1, [%o0 + 0x88]
-	stx		%g2, [%o0 + 0x90]
-	stx		%g3, [%o0 + 0x98]
-	stx		%g4, [%o0 + 0xa0]
-	stx		%g5, [%o0 + 0xa8]
-	stx		%g6, [%o0 + 0xb0]
-	stx		%g7, [%o0 + 0xb8]
-	wrpr		%o5, 0x0, %pstate
-	retl
-	 nop
-
-	.globl		restore_alternate_globals
-restore_alternate_globals: /* %o0 = save_area */
-	rdpr		%pstate, %o5
-	andn		%o5, PSTATE_IE, %o1
-	wrpr		%o1, PSTATE_AG, %pstate
-	ldx		[%o0 + 0x00], %g0
-	ldx		[%o0 + 0x08], %g1
-	ldx		[%o0 + 0x10], %g2
-	ldx		[%o0 + 0x18], %g3
-	ldx		[%o0 + 0x20], %g4
-	ldx		[%o0 + 0x28], %g5
-	ldx		[%o0 + 0x30], %g6
-	ldx		[%o0 + 0x38], %g7
-	wrpr		%o1, PSTATE_IG, %pstate
-	ldx		[%o0 + 0x40], %g0
-	ldx		[%o0 + 0x48], %g1
-	ldx		[%o0 + 0x50], %g2
-	ldx		[%o0 + 0x58], %g3
-	ldx		[%o0 + 0x60], %g4
-	ldx		[%o0 + 0x68], %g5
-	ldx		[%o0 + 0x70], %g6
-	ldx		[%o0 + 0x78], %g7
-	wrpr		%o1, PSTATE_MG, %pstate
-	ldx		[%o0 + 0x80], %g0
-	ldx		[%o0 + 0x88], %g1
-	ldx		[%o0 + 0x90], %g2
-	ldx		[%o0 + 0x98], %g3
-	ldx		[%o0 + 0xa0], %g4
-	ldx		[%o0 + 0xa8], %g5
-	ldx		[%o0 + 0xb0], %g6
-	ldx		[%o0 + 0xb8], %g7
-	wrpr		%o5, 0x0, %pstate
-	retl
-	 nop
-
 	.globl		getcc, setcc
 getcc:
 	ldx		[%o0 + PT_V9_TSTATE], %o1
@@ -488,9 +481,24 @@ setcc:
 	retl
 	 stx		%o1, [%o0 + PT_V9_TSTATE]
 
-	.globl		utrap, utrap_ill
-utrap:	brz,pn		%g1, etrap
+	.globl		utrap_trap
+utrap_trap:		/* %g3=handler,%g4=level */
+	TRAP_LOAD_THREAD_REG(%g6, %g1)
+	ldx		[%g6 + TI_UTRAPS], %g1
+	brnz,pt		%g1, invoke_utrap
 	 nop
+
+	ba,pt		%xcc, etrap
+	 rd		%pc, %g7
+	mov		%l4, %o1
+        call		bad_trap
+	 add		%sp, PTREGS_OFF, %o0
+	ba,pt		%xcc, rtrap
+	 clr		%l6
+
+invoke_utrap:
+	sllx		%g3, 3, %g3
+	ldx		[%g1 + %g3], %g1
 	save		%sp, -128, %sp
 	rdpr		%tstate, %l6
 	rdpr		%cwp, %l7
@@ -500,17 +508,6 @@ utrap:	brz,pn		%g1, etrap
 	rdpr		%tnpc, %l7
 	wrpr		%g1, 0, %tnpc
 	done
-utrap_ill:
-        call		bad_trap
-	 add		%sp, PTREGS_OFF, %o0
-	ba,pt		%xcc, rtrap
-	 clr		%l6
-
-	/* XXX Here is stuff we still need to write... -DaveM XXX */
-	.globl		netbsd_syscall
-netbsd_syscall:
-	retl
-	 nop
 
 	/* We need to carefully read the error status, ACK
 	 * the errors, prevent recursive traps, and pass the
@@ -1001,7 +998,7 @@ dcpe_icpe_tl1_common:
 	 * %g3:		scratch
 	 * %g4:		AFSR
 	 * %g5:		AFAR
-	 * %g6:		current thread ptr
+	 * %g6:		unused, will have current thread ptr after etrap
 	 * %g7:		scratch
 	 */
 __cheetah_log_error:
@@ -1539,13 +1536,14 @@ ret_from_syscall:
 
 1:		b,pt		%xcc, ret_sys_call
 		 ldx		[%sp + PTREGS_OFF + PT_V9_I0], %o0
-sparc_exit:	wrpr		%g0, (PSTATE_RMO | PSTATE_PEF | PSTATE_PRIV), %pstate
+sparc_exit:	rdpr		%pstate, %g2
+		wrpr		%g2, PSTATE_IE, %pstate
 		rdpr		%otherwin, %g1
 		rdpr		%cansave, %g3
 		add		%g3, %g1, %g3
 		wrpr		%g3, 0x0, %cansave
 		wrpr		%g0, 0x0, %otherwin
-		wrpr		%g0, (PSTATE_RMO | PSTATE_PEF | PSTATE_PRIV | PSTATE_IE), %pstate
+		wrpr		%g2, 0x0, %pstate
 		ba,pt		%xcc, sys_exit
 		 stb		%g0, [%g6 + TI_WSAVED]
 
@@ -1690,3 +1688,138 @@ __flushw_user:
 	 restore	%g0, %g0, %g0
 2:	retl
 	 nop
+
+#ifdef CONFIG_SMP
+	.globl		hard_smp_processor_id
+hard_smp_processor_id:
+#endif
+	.globl		real_hard_smp_processor_id
+real_hard_smp_processor_id:
+	__GET_CPUID(%o0)
+	retl
+	 nop
+
+	/* %o0: devhandle
+	 * %o1:	devino
+	 *
+	 * returns %o0: sysino
+	 */
+	.globl	sun4v_devino_to_sysino
+sun4v_devino_to_sysino:
+	mov	HV_FAST_INTR_DEVINO2SYSINO, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 mov	%o1, %o0
+
+	/* %o0: sysino
+	 *
+	 * returns %o0: intr_enabled (HV_INTR_{DISABLED,ENABLED})
+	 */
+	.globl	sun4v_intr_getenabled
+sun4v_intr_getenabled:
+	mov	HV_FAST_INTR_GETENABLED, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 mov	%o1, %o0
+
+	/* %o0: sysino
+	 * %o1: intr_enabled (HV_INTR_{DISABLED,ENABLED})
+	 */
+	.globl	sun4v_intr_setenabled
+sun4v_intr_setenabled:
+	mov	HV_FAST_INTR_SETENABLED, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 nop
+
+	/* %o0: sysino
+	 *
+	 * returns %o0: intr_state (HV_INTR_STATE_*)
+	 */
+	.globl	sun4v_intr_getstate
+sun4v_intr_getstate:
+	mov	HV_FAST_INTR_GETSTATE, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 mov	%o1, %o0
+
+	/* %o0: sysino
+	 * %o1: intr_state (HV_INTR_STATE_*)
+	 */
+	.globl	sun4v_intr_setstate
+sun4v_intr_setstate:
+	mov	HV_FAST_INTR_SETSTATE, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 nop
+
+	/* %o0: sysino
+	 *
+	 * returns %o0: cpuid
+	 */
+	.globl	sun4v_intr_gettarget
+sun4v_intr_gettarget:
+	mov	HV_FAST_INTR_GETTARGET, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 mov	%o1, %o0
+
+	/* %o0: sysino
+	 * %o1: cpuid
+	 */
+	.globl	sun4v_intr_settarget
+sun4v_intr_settarget:
+	mov	HV_FAST_INTR_SETTARGET, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 nop
+
+	/* %o0:	type
+	 * %o1:	queue paddr
+	 * %o2:	num queue entries
+	 *
+	 * returns %o0:	status
+	 */
+	.globl	sun4v_cpu_qconf
+sun4v_cpu_qconf:
+	mov	HV_FAST_CPU_QCONF, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 nop
+
+	/* returns %o0:	status
+	 */
+	.globl	sun4v_cpu_yield
+sun4v_cpu_yield:
+	mov	HV_FAST_CPU_YIELD, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 nop
+
+	/* %o0:	num cpus in cpu list
+	 * %o1:	cpu list paddr
+	 * %o2:	mondo block paddr
+	 *
+	 * returns %o0: status
+	 */
+	.globl	sun4v_cpu_mondo_send
+sun4v_cpu_mondo_send:
+	mov	HV_FAST_CPU_MONDO_SEND, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 nop
+
+	/* %o0:	CPU ID
+	 *
+	 * returns %o0:	-status if status non-zero, else
+	 *         %o0:	cpu state as HV_CPU_STATE_*
+	 */
+	.globl	sun4v_cpu_state
+sun4v_cpu_state:
+	mov	HV_FAST_CPU_STATE, %o5
+	ta	HV_FAST_TRAP
+	brnz,pn	%o0, 1f
+	 sub	%g0, %o0, %o0
+	mov	%o1, %o0
+1:	retl
+	 nop

arch/sparc64/kernel/etrap.S

@@ -31,6 +31,7 @@
 		.globl	etrap, etrap_irq, etraptl1
 etrap:		rdpr	%pil, %g2
 etrap_irq:
+		TRAP_LOAD_THREAD_REG(%g6, %g1)
 		rdpr	%tstate, %g1
 		sllx	%g2, 20, %g3
 		andcc	%g1, TSTATE_PRIV, %g0
@@ -54,7 +55,31 @@ etrap_irq:
 		rd	%y, %g3
 		stx	%g1, [%g2 + STACKFRAME_SZ + PT_V9_TNPC]
 		st	%g3, [%g2 + STACKFRAME_SZ + PT_V9_Y]
-		save	%g2, -STACK_BIAS, %sp	! Ordering here is critical
+
+		rdpr	%cansave, %g1
+		brnz,pt %g1, etrap_save
+		 nop
+
+		rdpr	%cwp, %g1
+		add	%g1, 2, %g1
+		wrpr	%g1, %cwp
+		be,pt	%xcc, etrap_user_spill
+		 mov	ASI_AIUP, %g3
+
+		rdpr	%otherwin, %g3
+		brz	%g3, etrap_kernel_spill
+		 mov	ASI_AIUS, %g3
+
+etrap_user_spill:
+
+		wr	%g3, 0x0, %asi
+		ldx	[%g6 + TI_FLAGS], %g3
+		and	%g3, _TIF_32BIT, %g3
+		brnz,pt	%g3, etrap_user_spill_32bit
+		 nop
+		ba,a,pt	%xcc, etrap_user_spill_64bit
+
+etrap_save:	save	%g2, -STACK_BIAS, %sp
 		mov	%g6, %l6
 
 		bne,pn	%xcc, 3f
@@ -70,42 +95,56 @@ etrap_irq:
 		wrpr	%g2, 0, %wstate
 		sethi	%hi(sparc64_kern_pri_context), %g2
 		ldx	[%g2 + %lo(sparc64_kern_pri_context)], %g3
-		stxa	%g3, [%l4] ASI_DMMU
-		flush	%l6
-		wr	%g0, ASI_AIUS, %asi
-2:		wrpr	%g0, 0x0, %tl
-		mov	%g4, %l4
+
+661:		stxa	%g3, [%l4] ASI_DMMU
+		.section .sun4v_1insn_patch, "ax"
+		.word	661b
+		stxa	%g3, [%l4] ASI_MMU
+		.previous
+
+		sethi	%hi(KERNBASE), %l4
+		flush	%l4
+		mov	ASI_AIUS, %l7
+2:		mov	%g4, %l4
 		mov	%g5, %l5
+		add	%g7, 4, %l2
+
+		/* Go to trap time globals so we can save them.  */
+661:		wrpr	%g0, ETRAP_PSTATE1, %pstate
+		.section .sun4v_1insn_patch, "ax"
+		.word	661b
+		SET_GL(0)
+		.previous
 
-		mov	%g7, %l2
-		wrpr	%g0, ETRAP_PSTATE1, %pstate
 		stx	%g1, [%sp + PTREGS_OFF + PT_V9_G1]
 		stx	%g2, [%sp + PTREGS_OFF + PT_V9_G2]
+		sllx	%l7, 24, %l7
 		stx	%g3, [%sp + PTREGS_OFF + PT_V9_G3]
+		rdpr	%cwp, %l0
 		stx	%g4, [%sp + PTREGS_OFF + PT_V9_G4]
 		stx	%g5, [%sp + PTREGS_OFF + PT_V9_G5]
 		stx	%g6, [%sp + PTREGS_OFF + PT_V9_G6]
-
 		stx	%g7, [%sp + PTREGS_OFF + PT_V9_G7]
+		or	%l7, %l0, %l7
+		sethi	%hi(TSTATE_RMO | TSTATE_PEF), %l0
+		or	%l7, %l0, %l7
+		wrpr	%l2, %tnpc
+		wrpr	%l7, (TSTATE_PRIV | TSTATE_IE), %tstate
 		stx	%i0, [%sp + PTREGS_OFF + PT_V9_I0]
 		stx	%i1, [%sp + PTREGS_OFF + PT_V9_I1]
 		stx	%i2, [%sp + PTREGS_OFF + PT_V9_I2]
 		stx	%i3, [%sp + PTREGS_OFF + PT_V9_I3]
 		stx	%i4, [%sp + PTREGS_OFF + PT_V9_I4]
 		stx	%i5, [%sp + PTREGS_OFF + PT_V9_I5]
-
 		stx	%i6, [%sp + PTREGS_OFF + PT_V9_I6]
-		stx	%i7, [%sp + PTREGS_OFF + PT_V9_I7]
-		wrpr	%g0, ETRAP_PSTATE2, %pstate
 		mov	%l6, %g6
-#ifdef CONFIG_SMP
-		mov	TSB_REG, %g3
-		ldxa	[%g3] ASI_IMMU, %g5
-#endif
-		jmpl	%l2 + 0x4, %g0
-		 ldx	[%g6 + TI_TASK], %g4
+		stx	%i7, [%sp + PTREGS_OFF + PT_V9_I7]
+		LOAD_PER_CPU_BASE(%g5, %g6, %g4, %g3, %l1)
+		ldx	[%g6 + TI_TASK], %g4
+		done
 
-3:		ldub	[%l6 + TI_FPDEPTH], %l5
+3:		mov	ASI_P, %l7
+		ldub	[%l6 + TI_FPDEPTH], %l5
 		add	%l6, TI_FPSAVED + 1, %l4
 		srl	%l5, 1, %l3
 		add	%l5, 2, %l5
@@ -125,6 +164,7 @@ etraptl1:	/* Save tstate/tpc/tnpc of TL 1-->4 and the tl register itself.
 		 *	0x58	TL4's TT
 		 *	0x60	TL
 		 */
+		TRAP_LOAD_THREAD_REG(%g6, %g1)
 		sub	%sp, ((4 * 8) * 4) + 8, %g2
 		rdpr	%tl, %g1
 
@@ -148,6 +188,11 @@ etraptl1:	/* Save tstate/tpc/tnpc of TL 1-->4 and the tl register itself.
 		rdpr	%tt, %g3
 		stx	%g3, [%g2 + STACK_BIAS + 0x38]
 
+		sethi	%hi(is_sun4v), %g3
+		lduw	[%g3 + %lo(is_sun4v)], %g3
+		brnz,pn	%g3, finish_tl1_capture
+		 nop
+
 		wrpr	%g0, 3, %tl
 		rdpr	%tstate, %g3
 		stx	%g3, [%g2 + STACK_BIAS + 0x40]
@@ -168,91 +213,20 @@ etraptl1:	/* Save tstate/tpc/tnpc of TL 1-->4 and the tl register itself.
 		rdpr	%tt, %g3
 		stx	%g3, [%g2 + STACK_BIAS + 0x78]
 
-		wrpr	%g1, %tl
 		stx	%g1, [%g2 + STACK_BIAS + 0x80]
 
+finish_tl1_capture:
+		wrpr	%g0, 1, %tl
+661:		nop
+		.section .sun4v_1insn_patch, "ax"
+		.word	661b
+		SET_GL(1)
+		.previous
+
 		rdpr	%tstate, %g1
 		sub	%g2, STACKFRAME_SZ + TRACEREG_SZ - STACK_BIAS, %g2
 		ba,pt	%xcc, 1b
 		 andcc	%g1, TSTATE_PRIV, %g0
 
-		.align	64
-		.globl	scetrap
-scetrap:	rdpr	%pil, %g2
-		rdpr	%tstate, %g1
-		sllx	%g2, 20, %g3
-		andcc	%g1, TSTATE_PRIV, %g0
-		or	%g1, %g3, %g1
-		bne,pn	%xcc, 1f
-		 sub	%sp, (STACKFRAME_SZ+TRACEREG_SZ-STACK_BIAS), %g2
-		wrpr	%g0, 7, %cleanwin
-
-		sllx	%g1, 51, %g3
-		sethi	%hi(TASK_REGOFF), %g2
-		or	%g2, %lo(TASK_REGOFF), %g2
-		brlz,pn	%g3, 1f
-		 add	%g6, %g2, %g2
-		wr	%g0, 0, %fprs
-1:		rdpr	%tpc, %g3
-		stx	%g1, [%g2 + STACKFRAME_SZ + PT_V9_TSTATE]
-
-		rdpr	%tnpc, %g1
-		stx	%g3, [%g2 + STACKFRAME_SZ + PT_V9_TPC]
-		stx	%g1, [%g2 + STACKFRAME_SZ + PT_V9_TNPC]
-		save	%g2, -STACK_BIAS, %sp	! Ordering here is critical
-		mov	%g6, %l6
-		bne,pn	%xcc, 2f
-		 mov	ASI_P, %l7
-		rdpr	%canrestore, %g3
-
-		rdpr	%wstate, %g2
-		wrpr	%g0, 0, %canrestore
-		sll	%g2, 3, %g2
-		mov	PRIMARY_CONTEXT, %l4
-		wrpr	%g3, 0, %otherwin
-		wrpr	%g2, 0, %wstate
-		sethi	%hi(sparc64_kern_pri_context), %g2
-		ldx	[%g2 + %lo(sparc64_kern_pri_context)], %g3
-		stxa	%g3, [%l4] ASI_DMMU
-		flush	%l6
-
-		mov	ASI_AIUS, %l7
-2:		mov	%g4, %l4
-		mov	%g5, %l5
-		add	%g7, 0x4, %l2
-		wrpr	%g0, ETRAP_PSTATE1, %pstate
-		stx	%g1, [%sp + PTREGS_OFF + PT_V9_G1]
-		stx	%g2, [%sp + PTREGS_OFF + PT_V9_G2]
-		sllx	%l7, 24, %l7
-
-		stx	%g3, [%sp + PTREGS_OFF + PT_V9_G3]
-		rdpr	%cwp, %l0
-		stx	%g4, [%sp + PTREGS_OFF + PT_V9_G4]
-		stx	%g5, [%sp + PTREGS_OFF + PT_V9_G5]
-		stx	%g6, [%sp + PTREGS_OFF + PT_V9_G6]
-		stx	%g7, [%sp + PTREGS_OFF + PT_V9_G7]
-		or	%l7, %l0, %l7
-		sethi	%hi(TSTATE_RMO | TSTATE_PEF), %l0
-
-		or	%l7, %l0, %l7
-		wrpr	%l2, %tnpc
-		wrpr	%l7, (TSTATE_PRIV | TSTATE_IE), %tstate
-		stx	%i0, [%sp + PTREGS_OFF + PT_V9_I0]
-		stx	%i1, [%sp + PTREGS_OFF + PT_V9_I1]
-		stx	%i2, [%sp + PTREGS_OFF + PT_V9_I2]
-		stx	%i3, [%sp + PTREGS_OFF + PT_V9_I3]
-		stx	%i4, [%sp + PTREGS_OFF + PT_V9_I4]
-
-		stx	%i5, [%sp + PTREGS_OFF + PT_V9_I5]
-		stx	%i6, [%sp + PTREGS_OFF + PT_V9_I6]
-		mov	%l6, %g6
-		stx	%i7, [%sp + PTREGS_OFF + PT_V9_I7]
-#ifdef CONFIG_SMP
-		mov	TSB_REG, %g3
-		ldxa	[%g3] ASI_IMMU, %g5
-#endif
-		ldx	[%g6 + TI_TASK], %g4
-		done
-
 #undef TASK_REGOFF
 #undef ETRAP_PSTATE1

arch/sparc64/kernel/head.S

@@ -26,6 +26,7 @@
 #include <asm/head.h>
 #include <asm/ttable.h>
 #include <asm/mmu.h>
+#include <asm/cpudata.h>
 	
 /* This section from from _start to sparc64_boot_end should fit into
  * 0x0000000000404000 to 0x0000000000408000.
@@ -94,12 +95,17 @@ sparc64_boot:
 	wrpr	%g1, 0x0, %pstate
 	ba,a,pt	%xcc, 1f
 
-	.globl	prom_finddev_name, prom_chosen_path
-	.globl	prom_getprop_name, prom_mmu_name
-	.globl	prom_callmethod_name, prom_translate_name
+	.globl	prom_finddev_name, prom_chosen_path, prom_root_node
+	.globl	prom_getprop_name, prom_mmu_name, prom_peer_name
+	.globl	prom_callmethod_name, prom_translate_name, prom_root_compatible
 	.globl	prom_map_name, prom_unmap_name, prom_mmu_ihandle_cache
 	.globl	prom_boot_mapped_pc, prom_boot_mapping_mode
 	.globl	prom_boot_mapping_phys_high, prom_boot_mapping_phys_low
+	.globl	is_sun4v
+prom_peer_name:
+	.asciz	"peer"
+prom_compatible_name:
+	.asciz	"compatible"
 prom_finddev_name:
 	.asciz	"finddevice"
 prom_chosen_path:
@@ -116,7 +122,13 @@ prom_map_name:
 	.asciz	"map"
 prom_unmap_name:
 	.asciz	"unmap"
+prom_sun4v_name:
+	.asciz	"sun4v"
 	.align	4
+prom_root_compatible:
+	.skip	64
+prom_root_node:
+	.word	0
 prom_mmu_ihandle_cache:
 	.word	0
 prom_boot_mapped_pc:
@@ -128,8 +140,54 @@ prom_boot_mapping_phys_high:
 	.xword	0
 prom_boot_mapping_phys_low:
 	.xword	0
+is_sun4v:
+	.word	0
 1:
 	rd	%pc, %l0
+
+	mov	(1b - prom_peer_name), %l1
+	sub	%l0, %l1, %l1
+	mov	0, %l2
+
+	/* prom_root_node = prom_peer(0) */
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "peer"
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 1
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	stx	%l2, [%sp + 2047 + 128 + 0x18]	! arg1, 0
+	stx	%g0, [%sp + 2047 + 128 + 0x20]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	ldx	[%sp + 2047 + 128 + 0x20], %l4	! prom root node
+	mov	(1b - prom_root_node), %l1
+	sub	%l0, %l1, %l1
+	stw	%l4, [%l1]
+
+	mov	(1b - prom_getprop_name), %l1
+	mov	(1b - prom_compatible_name), %l2
+	mov	(1b - prom_root_compatible), %l5
+	sub	%l0, %l1, %l1
+	sub	%l0, %l2, %l2
+	sub	%l0, %l5, %l5
+
+	/* prom_getproperty(prom_root_node, "compatible",
+	 *                  &prom_root_compatible, 64)
+	 */
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "getprop"
+	mov	4, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 4
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	stx	%l4, [%sp + 2047 + 128 + 0x18]	! arg1, prom_root_node
+	stx	%l2, [%sp + 2047 + 128 + 0x20]	! arg2, "compatible"
+	stx	%l5, [%sp + 2047 + 128 + 0x28]	! arg3, &prom_root_compatible
+	mov	64, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x30]	! arg4, size
+	stx	%g0, [%sp + 2047 + 128 + 0x38]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
 	mov	(1b - prom_finddev_name), %l1
 	mov	(1b - prom_chosen_path), %l2
 	mov	(1b - prom_boot_mapped_pc), %l3
@@ -238,6 +296,27 @@ prom_boot_mapping_phys_low:
 	add	%sp, (192 + 128), %sp
 
 sparc64_boot_after_remap:
+	sethi	%hi(prom_root_compatible), %g1
+	or	%g1, %lo(prom_root_compatible), %g1
+	sethi	%hi(prom_sun4v_name), %g7
+	or	%g7, %lo(prom_sun4v_name), %g7
+	mov	5, %g3
+1:	ldub	[%g7], %g2
+	ldub	[%g1], %g4
+	cmp	%g2, %g4
+	bne,pn	%icc, 2f
+	 add	%g7, 1, %g7
+	subcc	%g3, 1, %g3
+	bne,pt	%xcc, 1b
+	 add	%g1, 1, %g1
+
+	sethi	%hi(is_sun4v), %g1
+	or	%g1, %lo(is_sun4v), %g1
+	mov	1, %g7
+	stw	%g7, [%g1]
+
+2:
+	BRANCH_IF_SUN4V(g1, jump_to_sun4u_init)
 	BRANCH_IF_CHEETAH_BASE(g1,g7,cheetah_boot)
 	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,cheetah_plus_boot)
 	ba,pt	%xcc, spitfire_boot
@@ -301,20 +380,58 @@ jump_to_sun4u_init:
 	 nop
 
 sun4u_init:
+	BRANCH_IF_SUN4V(g1, sun4v_init)
+
 	/* Set ctx 0 */
-	mov	PRIMARY_CONTEXT, %g7
-	stxa	%g0, [%g7] ASI_DMMU
-	membar	#Sync
+	mov		PRIMARY_CONTEXT, %g7
+	stxa		%g0, [%g7] ASI_DMMU
+	membar		#Sync
 
-	mov	SECONDARY_CONTEXT, %g7
-	stxa	%g0, [%g7] ASI_DMMU
+	mov		SECONDARY_CONTEXT, %g7
+	stxa		%g0, [%g7] ASI_DMMU
 	membar	#Sync
 
-	BRANCH_IF_ANY_CHEETAH(g1,g7,cheetah_tlb_fixup)
+	ba,pt		%xcc, sun4u_continue
+	 nop
+
+sun4v_init:
+	/* Set ctx 0 */
+	mov		PRIMARY_CONTEXT, %g7
+	stxa		%g0, [%g7] ASI_MMU
+	membar		#Sync
+
+	mov		SECONDARY_CONTEXT, %g7
+	stxa		%g0, [%g7] ASI_MMU
+	membar		#Sync
+	ba,pt		%xcc, niagara_tlb_fixup
+	 nop
+
+sun4u_continue:
+	BRANCH_IF_ANY_CHEETAH(g1, g7, cheetah_tlb_fixup)
 
 	ba,pt	%xcc, spitfire_tlb_fixup
 	 nop
 
+niagara_tlb_fixup:
+	mov	3, %g2		/* Set TLB type to hypervisor. */
+	sethi	%hi(tlb_type), %g1
+	stw	%g2, [%g1 + %lo(tlb_type)]
+
+	/* Patch copy/clear ops.  */
+	call	niagara_patch_copyops
+	 nop
+	call	niagara_patch_bzero
+	 nop
+	call	niagara_patch_pageops
+	 nop
+
+	/* Patch TLB/cache ops.  */
+	call	hypervisor_patch_cachetlbops
+	 nop
+
+	ba,pt	%xcc, tlb_fixup_done
+	 nop
+
 cheetah_tlb_fixup:
 	mov	2, %g2		/* Set TLB type to cheetah+. */
 	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f)
@@ -411,85 +528,55 @@ setup_trap_table:
 	wrpr	%g0, 15, %pil
 
 	/* Make the firmware call to jump over to the Linux trap table.  */
-	call	prom_set_trap_table
-	 sethi	%hi(sparc64_ttable_tl0), %o0
+	sethi	%hi(is_sun4v), %o0
+	lduw	[%o0 + %lo(is_sun4v)], %o0
+	brz,pt	%o0, 1f
+	 nop
 
-	/* Start using proper page size encodings in ctx register.  */
-	sethi	%hi(sparc64_kern_pri_context), %g3
-	ldx	[%g3 + %lo(sparc64_kern_pri_context)], %g2
-	mov	PRIMARY_CONTEXT, %g1
-	stxa	%g2, [%g1] ASI_DMMU
-	membar	#Sync
+	TRAP_LOAD_TRAP_BLOCK(%g2, %g3)
+	add	%g2, TRAP_PER_CPU_FAULT_INFO, %g2
+	stxa	%g2, [%g0] ASI_SCRATCHPAD
 
-	/* The Linux trap handlers expect various trap global registers
-	 * to be setup with some fixed values.  So here we set these
-	 * up very carefully.  These globals are:
-	 *
-	 * Alternate Globals (PSTATE_AG):
-	 *
-	 * %g6			--> current_thread_info()
-	 *
-	 * MMU Globals (PSTATE_MG):
-	 *
-	 * %g1			--> TLB_SFSR
-	 * %g2			--> ((_PAGE_VALID | _PAGE_SZ4MB |
-	 *			      _PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W)
-	 *			     ^ 0xfffff80000000000)
-	 * (this %g2 value is used for computing the PAGE_OFFSET kernel
-	 *  TLB entries quickly, the virtual address of the fault XOR'd
-	 *  with this %g2 value is the PTE to load into the TLB)
-	 * %g3			--> VPTE_BASE_CHEETAH or VPTE_BASE_SPITFIRE
+	/* Compute physical address:
 	 *
-	 * Interrupt Globals (PSTATE_IG, setup by init_irqwork_curcpu()):
-	 *
-	 * %g6			--> __irq_work[smp_processor_id()]
+	 * paddr = kern_base + (mmfsa_vaddr - KERNBASE)
 	 */
+	sethi	%hi(KERNBASE), %g3
+	sub	%g2, %g3, %g2
+	sethi	%hi(kern_base), %g3
+	ldx	[%g3 + %lo(kern_base)], %g3
+	add	%g2, %g3, %o1
 
-	rdpr	%pstate, %o1
-	mov	%g6, %o2
-	wrpr	%o1, PSTATE_AG, %pstate
-	mov	%o2, %g6
-
-#define KERN_HIGHBITS		((_PAGE_VALID|_PAGE_SZ4MB)^0xfffff80000000000)
-#define KERN_LOWBITS		(_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W)
-	wrpr	%o1, PSTATE_MG, %pstate
-	mov	TSB_REG, %g1
-	stxa	%g0, [%g1] ASI_DMMU
-	membar	#Sync
-	stxa	%g0, [%g1] ASI_IMMU
-	membar	#Sync
-	mov	TLB_SFSR, %g1
-	sethi	%uhi(KERN_HIGHBITS), %g2
-	or	%g2, %ulo(KERN_HIGHBITS), %g2
-	sllx	%g2, 32, %g2
-	or	%g2, KERN_LOWBITS, %g2
-
-	BRANCH_IF_ANY_CHEETAH(g3,g7,8f)
-	ba,pt	%xcc, 9f
+	call	prom_set_trap_table_sun4v
+	 sethi	%hi(sparc64_ttable_tl0), %o0
+
+	ba,pt	%xcc, 2f
 	 nop
 
-8:
-	sethi		%uhi(VPTE_BASE_CHEETAH), %g3
-	or		%g3, %ulo(VPTE_BASE_CHEETAH), %g3
-	ba,pt		%xcc, 2f
-	 sllx		%g3, 32, %g3
+1:	call	prom_set_trap_table
+	 sethi	%hi(sparc64_ttable_tl0), %o0
 
-9:
-	sethi		%uhi(VPTE_BASE_SPITFIRE), %g3
-	or		%g3, %ulo(VPTE_BASE_SPITFIRE), %g3
-	sllx		%g3, 32, %g3
+	/* Start using proper page size encodings in ctx register.  */
+2:	sethi	%hi(sparc64_kern_pri_context), %g3
+	ldx	[%g3 + %lo(sparc64_kern_pri_context)], %g2
 
-2:
-	clr	%g7
-#undef KERN_HIGHBITS
-#undef KERN_LOWBITS
+	mov		PRIMARY_CONTEXT, %g1
+
+661:	stxa		%g2, [%g1] ASI_DMMU
+	.section	.sun4v_1insn_patch, "ax"
+	.word		661b
+	stxa		%g2, [%g1] ASI_MMU
+	.previous
+
+	membar	#Sync
 
 	/* Kill PROM timer */
 	sethi	%hi(0x80000000), %o2
 	sllx	%o2, 32, %o2
 	wr	%o2, 0, %tick_cmpr
 
-	BRANCH_IF_ANY_CHEETAH(o2,o3,1f)
+	BRANCH_IF_SUN4V(o2, 1f)
+	BRANCH_IF_ANY_CHEETAH(o2, o3, 1f)
 
 	ba,pt	%xcc, 2f
 	 nop
@@ -502,7 +589,6 @@ setup_trap_table:
 
 2:
 	wrpr	%g0, %g0, %wstate
-	wrpr	%o1, 0x0, %pstate
 
 	call	init_irqwork_curcpu
 	 nop
@@ -517,7 +603,7 @@ setup_trap_table:
 	 restore
 
 	.globl	setup_tba
-setup_tba:	/* i0 = is_starfire */
+setup_tba:
 	save	%sp, -192, %sp
 
 	/* The boot processor is the only cpu which invokes this
@@ -536,31 +622,35 @@ setup_tba:	/* i0 = is_starfire */
 	 restore
 sparc64_boot_end:
 
-#include "systbls.S"
 #include "ktlb.S"
+#include "tsb.S"
 #include "etrap.S"
 #include "rtrap.S"
 #include "winfixup.S"
 #include "entry.S"
+#include "sun4v_tlb_miss.S"
+#include "sun4v_ivec.S"
 
 /*
  * The following skip makes sure the trap table in ttable.S is aligned
  * on a 32K boundary as required by the v9 specs for TBA register.
+ *
+ * We align to a 32K boundary, then we have the 32K kernel TSB,
+ * then the 32K aligned trap table.
  */
 1:
 	.skip	0x4000 + _start - 1b
 
-#ifdef CONFIG_SBUS
-/* This is just a hack to fool make depend config.h discovering
-   strategy: As the .S files below need config.h, but
-   make depend does not find it for them, we include config.h
-   in head.S */
-#endif
+	.globl	swapper_tsb
+swapper_tsb:
+	.skip	(32 * 1024)
 
 ! 0x0000000000408000
 
 #include "ttable.S"
 
+#include "systbls.S"
+
 	.data
 	.align	8
 	.globl	prom_tba, tlb_type

arch/sparc64/kernel/irq.c

@@ -21,6 +21,7 @@
 #include <linux/delay.h>
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
+#include <linux/bootmem.h>
 
 #include <asm/ptrace.h>
 #include <asm/processor.h>
@@ -39,6 +40,7 @@
 #include <asm/cache.h>
 #include <asm/cpudata.h>
 #include <asm/auxio.h>
+#include <asm/head.h>
 
 #ifdef CONFIG_SMP
 static void distribute_irqs(void);
@@ -136,12 +138,48 @@ out_unlock:
 	return 0;
 }
 
+extern unsigned long real_hard_smp_processor_id(void);
+
+static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
+{
+	unsigned int tid;
+
+	if (this_is_starfire) {
+		tid = starfire_translate(imap, cpuid);
+		tid <<= IMAP_TID_SHIFT;
+		tid &= IMAP_TID_UPA;
+	} else {
+		if (tlb_type == cheetah || tlb_type == cheetah_plus) {
+			unsigned long ver;
+
+			__asm__ ("rdpr %%ver, %0" : "=r" (ver));
+			if ((ver >> 32UL) == __JALAPENO_ID ||
+			    (ver >> 32UL) == __SERRANO_ID) {
+				tid = cpuid << IMAP_TID_SHIFT;
+				tid &= IMAP_TID_JBUS;
+			} else {
+				unsigned int a = cpuid & 0x1f;
+				unsigned int n = (cpuid >> 5) & 0x1f;
+
+				tid = ((a << IMAP_AID_SHIFT) |
+				       (n << IMAP_NID_SHIFT));
+				tid &= (IMAP_AID_SAFARI |
+					IMAP_NID_SAFARI);;
+			}
+		} else {
+			tid = cpuid << IMAP_TID_SHIFT;
+			tid &= IMAP_TID_UPA;
+		}
+	}
+
+	return tid;
+}
+
 /* Now these are always passed a true fully specified sun4u INO. */
 void enable_irq(unsigned int irq)
 {
 	struct ino_bucket *bucket = __bucket(irq);
-	unsigned long imap;
-	unsigned long tid;
+	unsigned long imap, cpuid;
 
 	imap = bucket->imap;
 	if (imap == 0UL)
@@ -149,47 +187,38 @@ void enable_irq(unsigned int irq)
 
 	preempt_disable();
 
-	if (tlb_type == cheetah || tlb_type == cheetah_plus) {
-		unsigned long ver;
-
-		__asm__ ("rdpr %%ver, %0" : "=r" (ver));
-		if ((ver >> 32) == 0x003e0016) {
-			/* We set it to our JBUS ID. */
-			__asm__ __volatile__("ldxa [%%g0] %1, %0"
-					     : "=r" (tid)
-					     : "i" (ASI_JBUS_CONFIG));
-			tid = ((tid & (0x1fUL<<17)) << 9);
-			tid &= IMAP_TID_JBUS;
-		} else {
-			/* We set it to our Safari AID. */
-			__asm__ __volatile__("ldxa [%%g0] %1, %0"
-					     : "=r" (tid)
-					     : "i" (ASI_SAFARI_CONFIG));
-			tid = ((tid & (0x3ffUL<<17)) << 9);
-			tid &= IMAP_AID_SAFARI;
-		}
-	} else if (this_is_starfire == 0) {
-		/* We set it to our UPA MID. */
-		__asm__ __volatile__("ldxa [%%g0] %1, %0"
-				     : "=r" (tid)
-				     : "i" (ASI_UPA_CONFIG));
-		tid = ((tid & UPA_CONFIG_MID) << 9);
-		tid &= IMAP_TID_UPA;
+	/* This gets the physical processor ID, even on uniprocessor,
+	 * so we can always program the interrupt target correctly.
+	 */
+	cpuid = real_hard_smp_processor_id();
+
+	if (tlb_type == hypervisor) {
+		unsigned int ino = __irq_ino(irq);
+		int err;
+
+		err = sun4v_intr_settarget(ino, cpuid);
+		if (err != HV_EOK)
+			printk("sun4v_intr_settarget(%x,%lu): err(%d)\n",
+			       ino, cpuid, err);
+		err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
+		if (err != HV_EOK)
+			printk("sun4v_intr_setenabled(%x): err(%d)\n",
+			       ino, err);
 	} else {
-		tid = (starfire_translate(imap, smp_processor_id()) << 26);
-		tid &= IMAP_TID_UPA;
+		unsigned int tid = sun4u_compute_tid(imap, cpuid);
+
+		/* NOTE NOTE NOTE, IGN and INO are read-only, IGN is a product
+		 * of this SYSIO's preconfigured IGN in the SYSIO Control
+		 * Register, the hardware just mirrors that value here.
+		 * However for Graphics and UPA Slave devices the full
+		 * IMAP_INR field can be set by the programmer here.
+		 *
+		 * Things like FFB can now be handled via the new IRQ
+		 * mechanism.
+		 */
+		upa_writel(tid | IMAP_VALID, imap);
 	}
 
-	/* NOTE NOTE NOTE, IGN and INO are read-only, IGN is a product
-	 * of this SYSIO's preconfigured IGN in the SYSIO Control
-	 * Register, the hardware just mirrors that value here.
-	 * However for Graphics and UPA Slave devices the full
-	 * IMAP_INR field can be set by the programmer here.
-	 *
-	 * Things like FFB can now be handled via the new IRQ mechanism.
-	 */
-	upa_writel(tid | IMAP_VALID, imap);
-
 	preempt_enable();
 }
 
@@ -201,16 +230,26 @@ void disable_irq(unsigned int irq)
 
 	imap = bucket->imap;
 	if (imap != 0UL) {
-		u32 tmp;
+		if (tlb_type == hypervisor) {
+			unsigned int ino = __irq_ino(irq);
+			int err;
+
+			err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
+			if (err != HV_EOK)
+				printk("sun4v_intr_setenabled(%x): "
+				       "err(%d)\n", ino, err);
+		} else {
+			u32 tmp;
 
-		/* NOTE: We do not want to futz with the IRQ clear registers
-		 *       and move the state to IDLE, the SCSI code does call
-		 *       disable_irq() to assure atomicity in the queue cmd
-		 *       SCSI adapter driver code.  Thus we'd lose interrupts.
-		 */
-		tmp = upa_readl(imap);
-		tmp &= ~IMAP_VALID;
-		upa_writel(tmp, imap);
+			/* NOTE: We do not want to futz with the IRQ clear registers
+			 *       and move the state to IDLE, the SCSI code does call
+			 *       disable_irq() to assure atomicity in the queue cmd
+			 *       SCSI adapter driver code.  Thus we'd lose interrupts.
+			 */
+			tmp = upa_readl(imap);
+			tmp &= ~IMAP_VALID;
+			upa_writel(tmp, imap);
+		}
 	}
 }
 
@@ -248,6 +287,8 @@ unsigned int build_irq(int pil, int inofixup, unsigned long iclr, unsigned long
 		return __irq(&pil0_dummy_bucket);
 	}
 
+	BUG_ON(tlb_type == hypervisor);
+
 	/* RULE: Both must be specified in all other cases. */
 	if (iclr == 0UL || imap == 0UL) {
 		prom_printf("Invalid build_irq %d %d %016lx %016lx\n",
@@ -275,12 +316,11 @@ unsigned int build_irq(int pil, int inofixup, unsigned long iclr, unsigned long
 		goto out;
 	}
 
-	bucket->irq_info = kmalloc(sizeof(struct irq_desc), GFP_ATOMIC);
+	bucket->irq_info = kzalloc(sizeof(struct irq_desc), GFP_ATOMIC);
 	if (!bucket->irq_info) {
 		prom_printf("IRQ: Error, kmalloc(irq_desc) failed.\n");
 		prom_halt();
 	}
-	memset(bucket->irq_info, 0, sizeof(struct irq_desc));
 
 	/* Ok, looks good, set it up.  Don't touch the irq_chain or
 	 * the pending flag.
@@ -294,6 +334,37 @@ out:
 	return __irq(bucket);
 }
 
+unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino, int pil, unsigned char flags)
+{
+	struct ino_bucket *bucket;
+	unsigned long sysino;
+
+	sysino = sun4v_devino_to_sysino(devhandle, devino);
+
+	bucket = &ivector_table[sysino];
+
+	/* Catch accidental accesses to these things.  IMAP/ICLR handling
+	 * is done by hypervisor calls on sun4v platforms, not by direct
+	 * register accesses.
+	 *
+	 * But we need to make them look unique for the disable_irq() logic
+	 * in free_irq().
+	 */
+	bucket->imap = ~0UL - sysino;
+	bucket->iclr = ~0UL - sysino;
+
+	bucket->pil = pil;
+	bucket->flags = flags;
+
+	bucket->irq_info = kzalloc(sizeof(struct irq_desc), GFP_ATOMIC);
+	if (!bucket->irq_info) {
+		prom_printf("IRQ: Error, kmalloc(irq_desc) failed.\n");
+		prom_halt();
+	}
+
+	return __irq(bucket);
+}
+
 static void atomic_bucket_insert(struct ino_bucket *bucket)
 {
 	unsigned long pstate;
@@ -482,7 +553,6 @@ void free_irq(unsigned int irq, void *dev_id)
 	bucket = __bucket(irq);
 	if (bucket != &pil0_dummy_bucket) {
 		struct irq_desc *desc = bucket->irq_info;
-		unsigned long imap = bucket->imap;
 		int ent, i;
 
 		for (i = 0; i < MAX_IRQ_DESC_ACTION; i++) {
@@ -495,6 +565,8 @@ void free_irq(unsigned int irq, void *dev_id)
 		}
 
 		if (!desc->action_active_mask) {
+			unsigned long imap = bucket->imap;
+
 			/* This unique interrupt source is now inactive. */
 			bucket->flags &= ~IBF_ACTIVE;
 
@@ -592,7 +664,18 @@ static void process_bucket(int irq, struct ino_bucket *bp, struct pt_regs *regs)
 			break;
 	}
 	if (bp->pil != 0) {
-		upa_writel(ICLR_IDLE, bp->iclr);
+		if (tlb_type == hypervisor) {
+			unsigned int ino = __irq_ino(bp);
+			int err;
+
+			err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
+			if (err != HV_EOK)
+				printk("sun4v_intr_setstate(%x): "
+				       "err(%d)\n", ino, err);
+		} else {
+			upa_writel(ICLR_IDLE, bp->iclr);
+		}
+
 		/* Test and add entropy */
 		if (random & SA_SAMPLE_RANDOM)
 			add_interrupt_randomness(irq);
@@ -694,7 +777,7 @@ irqreturn_t sparc_floppy_irq(int irq, void *dev_cookie, struct pt_regs *regs)
 		val = readb(auxio_register);
 		val |= AUXIO_AUX1_FTCNT;
 		writeb(val, auxio_register);
-		val &= AUXIO_AUX1_FTCNT;
+		val &= ~AUXIO_AUX1_FTCNT;
 		writeb(val, auxio_register);
 
 		doing_pdma = 0;
@@ -727,25 +810,23 @@ EXPORT_SYMBOL(probe_irq_off);
 static int retarget_one_irq(struct irqaction *p, int goal_cpu)
 {
 	struct ino_bucket *bucket = get_ino_in_irqaction(p) + ivector_table;
-	unsigned long imap = bucket->imap;
-	unsigned int tid;
 
 	while (!cpu_online(goal_cpu)) {
 		if (++goal_cpu >= NR_CPUS)
 			goal_cpu = 0;
 	}
 
-	if (tlb_type == cheetah || tlb_type == cheetah_plus) {
-		tid = goal_cpu << 26;
-		tid &= IMAP_AID_SAFARI;
-	} else if (this_is_starfire == 0) {
-		tid = goal_cpu << 26;
-		tid &= IMAP_TID_UPA;
+	if (tlb_type == hypervisor) {
+		unsigned int ino = __irq_ino(bucket);
+
+		sun4v_intr_settarget(ino, goal_cpu);
+		sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
 	} else {
-		tid = (starfire_translate(imap, goal_cpu) << 26);
-		tid &= IMAP_TID_UPA;
+		unsigned long imap = bucket->imap;
+		unsigned int tid = sun4u_compute_tid(imap, goal_cpu);
+
+		upa_writel(tid | IMAP_VALID, imap);
 	}
-	upa_writel(tid | IMAP_VALID, imap);
 
 	do {
 		if (++goal_cpu >= NR_CPUS)
@@ -848,33 +929,114 @@ static void kill_prom_timer(void)
 
 void init_irqwork_curcpu(void)
 {
-	register struct irq_work_struct *workp asm("o2");
-	register unsigned long tmp asm("o3");
 	int cpu = hard_smp_processor_id();
 
-	memset(__irq_work + cpu, 0, sizeof(*workp));
-
-	/* Make sure we are called with PSTATE_IE disabled.  */
-	__asm__ __volatile__("rdpr	%%pstate, %0\n\t"
-			     : "=r" (tmp));
-	if (tmp & PSTATE_IE) {
-		prom_printf("BUG: init_irqwork_curcpu() called with "
-			    "PSTATE_IE enabled, bailing.\n");
-		__asm__ __volatile__("mov	%%i7, %0\n\t"
-				     : "=r" (tmp));
-		prom_printf("BUG: Called from %lx\n", tmp);
+	memset(__irq_work + cpu, 0, sizeof(struct irq_work_struct));
+}
+
+static void __cpuinit register_one_mondo(unsigned long paddr, unsigned long type)
+{
+	unsigned long num_entries = 128;
+	unsigned long status;
+
+	status = sun4v_cpu_qconf(type, paddr, num_entries);
+	if (status != HV_EOK) {
+		prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
+			    "err %lu\n", type, paddr, num_entries, status);
 		prom_halt();
 	}
+}
 
-	/* Set interrupt globals.  */
-	workp = &__irq_work[cpu];
-	__asm__ __volatile__(
-	"rdpr	%%pstate, %0\n\t"
-	"wrpr	%0, %1, %%pstate\n\t"
-	"mov	%2, %%g6\n\t"
-	"wrpr	%0, 0x0, %%pstate\n\t"
-	: "=&r" (tmp)
-	: "i" (PSTATE_IG), "r" (workp));
+static void __cpuinit sun4v_register_mondo_queues(int this_cpu)
+{
+	struct trap_per_cpu *tb = &trap_block[this_cpu];
+
+	register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO);
+	register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO);
+	register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR);
+	register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR);
+}
+
+static void __cpuinit alloc_one_mondo(unsigned long *pa_ptr, int use_bootmem)
+{
+	void *page;
+
+	if (use_bootmem)
+		page = alloc_bootmem_low_pages(PAGE_SIZE);
+	else
+		page = (void *) get_zeroed_page(GFP_ATOMIC);
+
+	if (!page) {
+		prom_printf("SUN4V: Error, cannot allocate mondo queue.\n");
+		prom_halt();
+	}
+
+	*pa_ptr = __pa(page);
+}
+
+static void __cpuinit alloc_one_kbuf(unsigned long *pa_ptr, int use_bootmem)
+{
+	void *page;
+
+	if (use_bootmem)
+		page = alloc_bootmem_low_pages(PAGE_SIZE);
+	else
+		page = (void *) get_zeroed_page(GFP_ATOMIC);
+
+	if (!page) {
+		prom_printf("SUN4V: Error, cannot allocate kbuf page.\n");
+		prom_halt();
+	}
+
+	*pa_ptr = __pa(page);
+}
+
+static void __cpuinit init_cpu_send_mondo_info(struct trap_per_cpu *tb, int use_bootmem)
+{
+#ifdef CONFIG_SMP
+	void *page;
+
+	BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
+
+	if (use_bootmem)
+		page = alloc_bootmem_low_pages(PAGE_SIZE);
+	else
+		page = (void *) get_zeroed_page(GFP_ATOMIC);
+
+	if (!page) {
+		prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
+		prom_halt();
+	}
+
+	tb->cpu_mondo_block_pa = __pa(page);
+	tb->cpu_list_pa = __pa(page + 64);
+#endif
+}
+
+/* Allocate and register the mondo and error queues for this cpu.  */
+void __cpuinit sun4v_init_mondo_queues(int use_bootmem, int cpu, int alloc, int load)
+{
+	struct trap_per_cpu *tb = &trap_block[cpu];
+
+	if (alloc) {
+		alloc_one_mondo(&tb->cpu_mondo_pa, use_bootmem);
+		alloc_one_mondo(&tb->dev_mondo_pa, use_bootmem);
+		alloc_one_mondo(&tb->resum_mondo_pa, use_bootmem);
+		alloc_one_kbuf(&tb->resum_kernel_buf_pa, use_bootmem);
+		alloc_one_mondo(&tb->nonresum_mondo_pa, use_bootmem);
+		alloc_one_kbuf(&tb->nonresum_kernel_buf_pa, use_bootmem);
+
+		init_cpu_send_mondo_info(tb, use_bootmem);
+	}
+
+	if (load) {
+		if (cpu != hard_smp_processor_id()) {
+			prom_printf("SUN4V: init mondo on cpu %d not %d\n",
+				    cpu, hard_smp_processor_id());
+			prom_halt();
+		}
+		sun4v_register_mondo_queues(cpu);
+	}
 }
 
 /* Only invoked on boot processor. */
@@ -884,6 +1046,9 @@ void __init init_IRQ(void)
 	kill_prom_timer();
 	memset(&ivector_table[0], 0, sizeof(ivector_table));
 
+	if (tlb_type == hypervisor)
+		sun4v_init_mondo_queues(1, hard_smp_processor_id(), 1, 1);
+
 	/* We need to clear any IRQ's pending in the soft interrupt
 	 * registers, a spurious one could be left around from the
 	 * PROM timer which we just disabled.

arch/sparc64/kernel/itlb_base.S

-/* $Id: itlb_base.S,v 1.12 2002/02/09 19:49:30 davem Exp $
- * itlb_base.S:	Front end to ITLB miss replacement strategy.
- *              This is included directly into the trap table.
- *
- * Copyright (C) 1996,1998 David S. Miller (davem@redhat.com)
- * Copyright (C) 1997,1998 Jakub Jelinek   (jj@ultra.linux.cz)
- */
-
-#if PAGE_SHIFT == 13
-/*
- * To compute vpte offset, we need to do ((addr >> 13) << 3),
- * which can be optimized to (addr >> 10) if bits 10/11/12 can
- * be guaranteed to be 0 ... mmu_context.h does guarantee this
- * by only using 10 bits in the hwcontext value.
- */
-#define CREATE_VPTE_OFFSET1(r1, r2) \
-				srax	r1, 10, r2
-#define CREATE_VPTE_OFFSET2(r1, r2) nop
-#else /* PAGE_SHIFT */
-#define CREATE_VPTE_OFFSET1(r1, r2) \
-				srax	r1, PAGE_SHIFT, r2
-#define CREATE_VPTE_OFFSET2(r1, r2) \
-				sllx	r2, 3, r2
-#endif /* PAGE_SHIFT */
-
-
-/* Ways we can get here:
- *
- * 1) Nucleus instruction misses from module code.
- * 2) All user instruction misses.
- *
- * All real page faults merge their code paths to the
- * sparc64_realfault_common label below.
- */
-
-/* ITLB ** ICACHE line 1: Quick user TLB misses		*/
-	mov		TLB_SFSR, %g1
-	ldxa		[%g1 + %g1] ASI_IMMU, %g4	! Get TAG_ACCESS
-	CREATE_VPTE_OFFSET1(%g4, %g6)			! Create VPTE offset
-	CREATE_VPTE_OFFSET2(%g4, %g6)			! Create VPTE offset
-	ldxa		[%g3 + %g6] ASI_P, %g5		! Load VPTE
-1:	brgez,pn	%g5, 3f				! Not valid, branch out
-	 sethi		%hi(_PAGE_EXEC), %g4		! Delay-slot
-	andcc		%g5, %g4, %g0			! Executable?
-
-/* ITLB ** ICACHE line 2: Real faults			*/
-	be,pn		%xcc, 3f			! Nope, branch.
-	 nop						! Delay-slot
-2:	stxa		%g5, [%g0] ASI_ITLB_DATA_IN	! Load PTE into TLB
-	retry						! Trap return
-3:	rdpr		%pstate, %g4			! Move into alt-globals
-	wrpr		%g4, PSTATE_AG|PSTATE_MG, %pstate
-	rdpr		%tpc, %g5			! And load faulting VA
-	mov		FAULT_CODE_ITLB, %g4		! It was read from ITLB
-
-/* ITLB ** ICACHE line 3: Finish faults	*/
-sparc64_realfault_common:				! Called by dtlb_miss
-	stb		%g4, [%g6 + TI_FAULT_CODE]
-	stx		%g5, [%g6 + TI_FAULT_ADDR]
-	ba,pt		%xcc, etrap			! Save state
-1:	 rd		%pc, %g7			! ...
-	call		do_sparc64_fault		! Call fault handler
-	 add		%sp, PTREGS_OFF, %o0! Compute pt_regs arg
-	ba,pt		%xcc, rtrap_clr_l6		! Restore cpu state
-	 nop
-
-/* ITLB ** ICACHE line 4: Window fixups */
-winfix_trampoline:
-	rdpr		%tpc, %g3			! Prepare winfixup TNPC
-	or		%g3, 0x7c, %g3			! Compute branch offset
-	wrpr		%g3, %tnpc			! Write it into TNPC
-	done						! Do it to it
-	nop
-	nop
-	nop
-	nop
-
-#undef CREATE_VPTE_OFFSET1
-#undef CREATE_VPTE_OFFSET2

arch/sparc64/kernel/itlb_miss.S

+/* ITLB ** ICACHE line 1: Context 0 check and TSB load	*/
+	ldxa	[%g0] ASI_IMMU_TSB_8KB_PTR, %g1	! Get TSB 8K pointer
+	ldxa	[%g0] ASI_IMMU, %g6		! Get TAG TARGET
+	srlx	%g6, 48, %g5			! Get context
+	sllx	%g6, 22, %g6			! Zero out context
+	brz,pn	%g5, kvmap_itlb			! Context 0 processing
+	 srlx	%g6, 22, %g6			! Delay slot
+	TSB_LOAD_QUAD(%g1, %g4)			! Load TSB entry
+	cmp	%g4, %g6			! Compare TAG
+
+/* ITLB ** ICACHE line 2: TSB compare and TLB load	*/
+	bne,pn	%xcc, tsb_miss_itlb		! Miss
+	 mov	FAULT_CODE_ITLB, %g3
+	andcc	%g5, _PAGE_EXEC_4U, %g0		! Executable?
+	be,pn	%xcc, tsb_do_fault
+	 nop					! Delay slot, fill me
+	stxa	%g5, [%g0] ASI_ITLB_DATA_IN	! Load TLB
+	retry					! Trap done
+	nop
+
+/* ITLB ** ICACHE line 3: 				*/
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+
+/* ITLB ** ICACHE line 4: 				*/
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop
+	nop

arch/sparc64/kernel/ktlb.S

@@ -4,191 +4,276 @@
  * Copyright (C) 1996 Eddie C. Dost        (ecd@brainaid.de)
  * Copyright (C) 1996 Miguel de Icaza      (miguel@nuclecu.unam.mx)
  * Copyright (C) 1996,98,99 Jakub Jelinek  (jj@sunsite.mff.cuni.cz)
-*/
+ */
 
 #include <linux/config.h>
 #include <asm/head.h>
 #include <asm/asi.h>
 #include <asm/page.h>
 #include <asm/pgtable.h>
+#include <asm/tsb.h>
 
 	.text
 	.align		32
 
-/*
- * On a second level vpte miss, check whether the original fault is to the OBP 
- * range (note that this is only possible for instruction miss, data misses to
- * obp range do not use vpte). If so, go back directly to the faulting address.
- * This is because we want to read the tpc, otherwise we have no way of knowing
- * the 8k aligned faulting address if we are using >8k kernel pagesize. This
- * also ensures no vpte range addresses are dropped into tlb while obp is
- * executing (see inherit_locked_prom_mappings() rant).
- */
-sparc64_vpte_nucleus:
-	/* Note that kvmap below has verified that the address is
-	 * in the range MODULES_VADDR --> VMALLOC_END already.  So
-	 * here we need only check if it is an OBP address or not.
+kvmap_itlb:
+	/* g6: TAG TARGET */
+	mov		TLB_TAG_ACCESS, %g4
+	ldxa		[%g4] ASI_IMMU, %g4
+
+	/* sun4v_itlb_miss branches here with the missing virtual
+	 * address already loaded into %g4
 	 */
+kvmap_itlb_4v:
+
+kvmap_itlb_nonlinear:
+	/* Catch kernel NULL pointer calls.  */
+	sethi		%hi(PAGE_SIZE), %g5
+	cmp		%g4, %g5
+	bleu,pn		%xcc, kvmap_dtlb_longpath
+	 nop
+
+	KERN_TSB_LOOKUP_TL1(%g4, %g6, %g5, %g1, %g2, %g3, kvmap_itlb_load)
+
+kvmap_itlb_tsb_miss:
 	sethi		%hi(LOW_OBP_ADDRESS), %g5
 	cmp		%g4, %g5
-	blu,pn		%xcc, kern_vpte
+	blu,pn		%xcc, kvmap_itlb_vmalloc_addr
 	 mov		0x1, %g5
 	sllx		%g5, 32, %g5
 	cmp		%g4, %g5
-	blu,pn		%xcc, vpte_insn_obp
+	blu,pn		%xcc, kvmap_itlb_obp
 	 nop
 
-	/* These two instructions are patched by paginig_init().  */
-kern_vpte:
-	sethi		%hi(swapper_pgd_zero), %g5
-	lduw		[%g5 + %lo(swapper_pgd_zero)], %g5
+kvmap_itlb_vmalloc_addr:
+	KERN_PGTABLE_WALK(%g4, %g5, %g2, kvmap_itlb_longpath)
 
-	/* With kernel PGD in %g5, branch back into dtlb_backend.  */
-	ba,pt		%xcc, sparc64_kpte_continue
-	 andn		%g1, 0x3, %g1	/* Finish PMD offset adjustment.  */
+	KTSB_LOCK_TAG(%g1, %g2, %g7)
 
-vpte_noent:
-	/* Restore previous TAG_ACCESS, %g5 is zero, and we will
-	 * skip over the trap instruction so that the top level
-	 * TLB miss handler will thing this %g5 value is just an
-	 * invalid PTE, thus branching to full fault processing.
-	 */
-	mov		TLB_SFSR, %g1
-	stxa		%g4, [%g1 + %g1] ASI_DMMU
-	done
-
-vpte_insn_obp:
-	/* Behave as if we are at TL0.  */
-	wrpr		%g0, 1, %tl
-	rdpr		%tpc, %g4	/* Find original faulting iaddr */
-	srlx		%g4, 13, %g4	/* Throw out context bits */
-	sllx		%g4, 13, %g4	/* g4 has vpn + ctx0 now */
-
-	/* Restore previous TAG_ACCESS.  */
-	mov		TLB_SFSR, %g1
-	stxa		%g4, [%g1 + %g1] ASI_IMMU
-
-	sethi		%hi(prom_trans), %g5
-	or		%g5, %lo(prom_trans), %g5
-
-1:	ldx		[%g5 + 0x00], %g6	! base
-	brz,a,pn	%g6, longpath		! no more entries, fail
-	 mov		TLB_SFSR, %g1		! and restore %g1
-	ldx		[%g5 + 0x08], %g1	! len
-	add		%g6, %g1, %g1		! end
-	cmp		%g6, %g4
-	bgu,pt		%xcc, 2f
-	 cmp		%g4, %g1
-	bgeu,pt		%xcc, 2f
-	 ldx		[%g5 + 0x10], %g1	! PTE
-
-	/* TLB load, restore %g1, and return from trap.  */
-	sub		%g4, %g6, %g6
-	add		%g1, %g6, %g5
-	mov		TLB_SFSR, %g1
-	stxa		%g5, [%g0] ASI_ITLB_DATA_IN
-	retry
+	/* Load and check PTE.  */
+	ldxa		[%g5] ASI_PHYS_USE_EC, %g5
+	mov		1, %g7
+	sllx		%g7, TSB_TAG_INVALID_BIT, %g7
+	brgez,a,pn	%g5, kvmap_itlb_longpath
+	 KTSB_STORE(%g1, %g7)
+
+	KTSB_WRITE(%g1, %g5, %g6)
+
+	/* fallthrough to TLB load */
 
-2:	ba,pt		%xcc, 1b
-	 add		%g5, (3 * 8), %g5	! next entry
-
-kvmap_do_obp:
-	sethi		%hi(prom_trans), %g5
-	or		%g5, %lo(prom_trans), %g5
-	srlx		%g4, 13, %g4
-	sllx		%g4, 13, %g4
-
-1:	ldx		[%g5 + 0x00], %g6	! base
-	brz,a,pn	%g6, longpath		! no more entries, fail
-	 mov		TLB_SFSR, %g1		! and restore %g1
-	ldx		[%g5 + 0x08], %g1	! len
-	add		%g6, %g1, %g1		! end
-	cmp		%g6, %g4
-	bgu,pt		%xcc, 2f
-	 cmp		%g4, %g1
-	bgeu,pt		%xcc, 2f
-	 ldx		[%g5 + 0x10], %g1	! PTE
-
-	/* TLB load, restore %g1, and return from trap.  */
-	sub		%g4, %g6, %g6
-	add		%g1, %g6, %g5
-	mov		TLB_SFSR, %g1
-	stxa		%g5, [%g0] ASI_DTLB_DATA_IN
+kvmap_itlb_load:
+
+661:	stxa		%g5, [%g0] ASI_ITLB_DATA_IN
 	retry
+	.section	.sun4v_2insn_patch, "ax"
+	.word		661b
+	nop
+	nop
+	.previous
+
+	/* For sun4v the ASI_ITLB_DATA_IN store and the retry
+	 * instruction get nop'd out and we get here to branch
+	 * to the sun4v tlb load code.  The registers are setup
+	 * as follows:
+	 *
+	 * %g4: vaddr
+	 * %g5: PTE
+	 * %g6:	TAG
+	 *
+	 * The sun4v TLB load wants the PTE in %g3 so we fix that
+	 * up here.
+	 */
+	ba,pt		%xcc, sun4v_itlb_load
+	 mov		%g5, %g3
 
-2:	ba,pt		%xcc, 1b
-	 add		%g5, (3 * 8), %g5	! next entry
+kvmap_itlb_longpath:
+
+661:	rdpr	%pstate, %g5
+	wrpr	%g5, PSTATE_AG | PSTATE_MG, %pstate
+	.section .sun4v_2insn_patch, "ax"
+	.word	661b
+	SET_GL(1)
+	nop
+	.previous
+
+	rdpr	%tpc, %g5
+	ba,pt	%xcc, sparc64_realfault_common
+	 mov	FAULT_CODE_ITLB, %g4
+
+kvmap_itlb_obp:
+	OBP_TRANS_LOOKUP(%g4, %g5, %g2, %g3, kvmap_itlb_longpath)
+
+	KTSB_LOCK_TAG(%g1, %g2, %g7)
+
+	KTSB_WRITE(%g1, %g5, %g6)
+
+	ba,pt		%xcc, kvmap_itlb_load
+	 nop
+
+kvmap_dtlb_obp:
+	OBP_TRANS_LOOKUP(%g4, %g5, %g2, %g3, kvmap_dtlb_longpath)
+
+	KTSB_LOCK_TAG(%g1, %g2, %g7)
+
+	KTSB_WRITE(%g1, %g5, %g6)
+
+	ba,pt		%xcc, kvmap_dtlb_load
+	 nop
 
-/*
- * On a first level data miss, check whether this is to the OBP range (note
- * that such accesses can be made by prom, as well as by kernel using
- * prom_getproperty on "address"), and if so, do not use vpte access ...
- * rather, use information saved during inherit_prom_mappings() using 8k
- * pagesize.
- */
 	.align		32
-kvmap:
-	brgez,pn	%g4, kvmap_nonlinear
+kvmap_dtlb_tsb4m_load:
+	KTSB_LOCK_TAG(%g1, %g2, %g7)
+	KTSB_WRITE(%g1, %g5, %g6)
+	ba,pt		%xcc, kvmap_dtlb_load
 	 nop
 
-#ifdef CONFIG_DEBUG_PAGEALLOC
+kvmap_dtlb:
+	/* %g6: TAG TARGET */
+	mov		TLB_TAG_ACCESS, %g4
+	ldxa		[%g4] ASI_DMMU, %g4
+
+	/* sun4v_dtlb_miss branches here with the missing virtual
+	 * address already loaded into %g4
+	 */
+kvmap_dtlb_4v:
+	brgez,pn	%g4, kvmap_dtlb_nonlinear
+	 nop
+
+	/* Correct TAG_TARGET is already in %g6, check 4mb TSB.  */
+	KERN_TSB4M_LOOKUP_TL1(%g6, %g5, %g1, %g2, %g3, kvmap_dtlb_load)
+
+	/* TSB entry address left in %g1, lookup linear PTE.
+	 * Must preserve %g1 and %g6 (TAG).
+	 */
+kvmap_dtlb_tsb4m_miss:
+	sethi		%hi(kpte_linear_bitmap), %g2
+	or		%g2, %lo(kpte_linear_bitmap), %g2
+
+	/* Clear the PAGE_OFFSET top virtual bits, then shift
+	 * down to get a 256MB physical address index.
+	 */
+	sllx		%g4, 21, %g5
+	mov		1, %g7
+	srlx		%g5, 21 + 28, %g5
+
+	/* Don't try this at home kids... this depends upon srlx
+	 * only taking the low 6 bits of the shift count in %g5.
+	 */
+	sllx		%g7, %g5, %g7
+
+	/* Divide by 64 to get the offset into the bitmask.  */
+	srlx		%g5, 6, %g5
+	sllx		%g5, 3, %g5
+
+	/* kern_linear_pte_xor[((mask & bit) ? 1 : 0)] */
+	ldx		[%g2 + %g5], %g2
+	andcc		%g2, %g7, %g0
+	sethi		%hi(kern_linear_pte_xor), %g5
+	or		%g5, %lo(kern_linear_pte_xor), %g5
+	bne,a,pt	%xcc, 1f
+	 add		%g5, 8, %g5
+
+1:	ldx		[%g5], %g2
+
 	.globl		kvmap_linear_patch
 kvmap_linear_patch:
-#endif
-	ba,pt		%xcc, kvmap_load
+	ba,pt		%xcc, kvmap_dtlb_tsb4m_load
 	 xor		%g2, %g4, %g5
 
-#ifdef CONFIG_DEBUG_PAGEALLOC
-	sethi		%hi(swapper_pg_dir), %g5
-	or		%g5, %lo(swapper_pg_dir), %g5
-	sllx		%g4, 64 - (PGDIR_SHIFT + PGDIR_BITS), %g6
-	srlx		%g6, 64 - PAGE_SHIFT, %g6
-	andn		%g6, 0x3, %g6
-	lduw		[%g5 + %g6], %g5
-	brz,pn		%g5, longpath
-	 sllx		%g4, 64 - (PMD_SHIFT + PMD_BITS), %g6
-	srlx		%g6, 64 - PAGE_SHIFT, %g6
-	sllx		%g5, 11, %g5
-	andn		%g6, 0x3, %g6
-	lduwa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brz,pn		%g5, longpath
-	 sllx		%g4, 64 - PMD_SHIFT, %g6
-	srlx		%g6, 64 - PAGE_SHIFT, %g6
-	sllx		%g5, 11, %g5
-	andn		%g6, 0x7, %g6
-	ldxa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brz,pn		%g5, longpath
+kvmap_dtlb_vmalloc_addr:
+	KERN_PGTABLE_WALK(%g4, %g5, %g2, kvmap_dtlb_longpath)
+
+	KTSB_LOCK_TAG(%g1, %g2, %g7)
+
+	/* Load and check PTE.  */
+	ldxa		[%g5] ASI_PHYS_USE_EC, %g5
+	mov		1, %g7
+	sllx		%g7, TSB_TAG_INVALID_BIT, %g7
+	brgez,a,pn	%g5, kvmap_dtlb_longpath
+	 KTSB_STORE(%g1, %g7)
+
+	KTSB_WRITE(%g1, %g5, %g6)
+
+	/* fallthrough to TLB load */
+
+kvmap_dtlb_load:
+
+661:	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Reload TLB
+	retry
+	.section	.sun4v_2insn_patch, "ax"
+	.word		661b
+	nop
+	nop
+	.previous
+
+	/* For sun4v the ASI_DTLB_DATA_IN store and the retry
+	 * instruction get nop'd out and we get here to branch
+	 * to the sun4v tlb load code.  The registers are setup
+	 * as follows:
+	 *
+	 * %g4: vaddr
+	 * %g5: PTE
+	 * %g6:	TAG
+	 *
+	 * The sun4v TLB load wants the PTE in %g3 so we fix that
+	 * up here.
+	 */
+	ba,pt		%xcc, sun4v_dtlb_load
+	 mov		%g5, %g3
+
+kvmap_dtlb_nonlinear:
+	/* Catch kernel NULL pointer derefs.  */
+	sethi		%hi(PAGE_SIZE), %g5
+	cmp		%g4, %g5
+	bleu,pn		%xcc, kvmap_dtlb_longpath
 	 nop
-	ba,a,pt		%xcc, kvmap_load
-#endif
 
-kvmap_nonlinear:
+	KERN_TSB_LOOKUP_TL1(%g4, %g6, %g5, %g1, %g2, %g3, kvmap_dtlb_load)
+
+kvmap_dtlb_tsbmiss:
 	sethi		%hi(MODULES_VADDR), %g5
 	cmp		%g4, %g5
-	blu,pn		%xcc, longpath
+	blu,pn		%xcc, kvmap_dtlb_longpath
 	 mov		(VMALLOC_END >> 24), %g5
 	sllx		%g5, 24, %g5
 	cmp		%g4, %g5
-	bgeu,pn		%xcc, longpath
+	bgeu,pn		%xcc, kvmap_dtlb_longpath
 	 nop
 
 kvmap_check_obp:
 	sethi		%hi(LOW_OBP_ADDRESS), %g5
 	cmp		%g4, %g5
-	blu,pn		%xcc, kvmap_vmalloc_addr
+	blu,pn		%xcc, kvmap_dtlb_vmalloc_addr
 	 mov		0x1, %g5
 	sllx		%g5, 32, %g5
 	cmp		%g4, %g5
-	blu,pn		%xcc, kvmap_do_obp
+	blu,pn		%xcc, kvmap_dtlb_obp
 	 nop
-
-kvmap_vmalloc_addr:
-	/* If we get here, a vmalloc addr was accessed, load kernel VPTE.  */
-	ldxa		[%g3 + %g6] ASI_N, %g5
-	brgez,pn	%g5, longpath
+	ba,pt		%xcc, kvmap_dtlb_vmalloc_addr
 	 nop
 
-kvmap_load:
-	/* PTE is valid, load into TLB and return from trap.  */
-	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Reload TLB
-	retry
+kvmap_dtlb_longpath:
+
+661:	rdpr	%pstate, %g5
+	wrpr	%g5, PSTATE_AG | PSTATE_MG, %pstate
+	.section .sun4v_2insn_patch, "ax"
+	.word	661b
+	SET_GL(1)
+	ldxa		[%g0] ASI_SCRATCHPAD, %g5
+	.previous
+
+	rdpr	%tl, %g3
+	cmp	%g3, 1
+
+661:	mov	TLB_TAG_ACCESS, %g4
+	ldxa	[%g4] ASI_DMMU, %g5
+	.section .sun4v_2insn_patch, "ax"
+	.word	661b
+	ldx	[%g5 + HV_FAULT_D_ADDR_OFFSET], %g5
+	nop
+	.previous
+
+	be,pt	%xcc, sparc64_realfault_common
+	 mov	FAULT_CODE_DTLB, %g4
+	ba,pt	%xcc, winfix_trampoline
+	 nop

arch/sparc64/kernel/pci.c

@@ -188,6 +188,7 @@ extern void psycho_init(int, char *);
 extern void schizo_init(int, char *);
 extern void schizo_plus_init(int, char *);
 extern void tomatillo_init(int, char *);
+extern void sun4v_pci_init(int, char *);
 
 static struct {
 	char *model_name;
@@ -204,6 +205,7 @@ static struct {
 	{ "pci108e,8002", schizo_plus_init },
 	{ "SUNW,tomatillo", tomatillo_init },
 	{ "pci108e,a801", tomatillo_init },
+	{ "SUNW,sun4v-pci", sun4v_pci_init },
 };
 #define PCI_NUM_CONTROLLER_TYPES (sizeof(pci_controller_table) / \
 				  sizeof(pci_controller_table[0]))
@@ -283,6 +285,12 @@ int __init pcic_present(void)
 	return pci_controller_scan(pci_is_controller);
 }
 
+struct pci_iommu_ops *pci_iommu_ops;
+EXPORT_SYMBOL(pci_iommu_ops);
+
+extern struct pci_iommu_ops pci_sun4u_iommu_ops,
+	pci_sun4v_iommu_ops;
+
 /* Find each controller in the system, attach and initialize
  * software state structure for each and link into the
  * pci_controller_root.  Setup the controller enough such
@@ -290,6 +298,11 @@ int __init pcic_present(void)
  */
 static void __init pci_controller_probe(void)
 {
+	if (tlb_type == hypervisor)
+		pci_iommu_ops = &pci_sun4v_iommu_ops;
+	else
+		pci_iommu_ops = &pci_sun4u_iommu_ops;
+
 	printk("PCI: Probing for controllers.\n");
 
 	pci_controller_scan(pci_controller_init);

arch/sparc64/kernel/pci_common.c

@@ -39,6 +39,8 @@ static int __init find_device_prom_node(struct pci_pbm_info *pbm,
 {
 	int node;
 
+	*nregs = 0;
+
 	/*
 	 * Return the PBM's PROM node in case we are it's PCI device,
 	 * as the PBM's reg property is different to standard PCI reg
@@ -51,10 +53,8 @@ static int __init find_device_prom_node(struct pci_pbm_info *pbm,
 	     pdev->device == PCI_DEVICE_ID_SUN_SCHIZO ||
 	     pdev->device == PCI_DEVICE_ID_SUN_TOMATILLO ||
 	     pdev->device == PCI_DEVICE_ID_SUN_SABRE ||
-	     pdev->device == PCI_DEVICE_ID_SUN_HUMMINGBIRD)) {
-		*nregs = 0;
+	     pdev->device == PCI_DEVICE_ID_SUN_HUMMINGBIRD))
 		return bus_prom_node;
-	}
 
 	node = prom_getchild(bus_prom_node);
 	while (node != 0) {
@@ -541,135 +541,183 @@ void __init pci_assign_unassigned(struct pci_pbm_info *pbm,
 		pci_assign_unassigned(pbm, bus);
 }
 
-static int __init pci_intmap_match(struct pci_dev *pdev, unsigned int *interrupt)
+static inline unsigned int pci_slot_swivel(struct pci_pbm_info *pbm,
+					   struct pci_dev *toplevel_pdev,
+					   struct pci_dev *pdev,
+					   unsigned int interrupt)
 {
-	struct linux_prom_pci_intmap bridge_local_intmap[PROM_PCIIMAP_MAX], *intmap;
-	struct linux_prom_pci_intmask bridge_local_intmask, *intmask;
-	struct pcidev_cookie *dev_pcp = pdev->sysdata;
-	struct pci_pbm_info *pbm = dev_pcp->pbm;
-	struct linux_prom_pci_registers *pregs = dev_pcp->prom_regs;
-	unsigned int hi, mid, lo, irq;
-	int i, num_intmap, map_slot;
+	unsigned int ret;
 
-	intmap = &pbm->pbm_intmap[0];
-	intmask = &pbm->pbm_intmask;
-	num_intmap = pbm->num_pbm_intmap;
-	map_slot = 0;
+	if (unlikely(interrupt < 1 || interrupt > 4)) {
+		printk("%s: Device %s interrupt value of %u is strange.\n",
+		       pbm->name, pci_name(pdev), interrupt);
+		return interrupt;
+	}
 
-	/* If we are underneath a PCI bridge, use PROM register
-	 * property of the parent bridge which is closest to
-	 * the PBM.
-	 *
-	 * However if that parent bridge has interrupt map/mask
-	 * properties of its own we use the PROM register property
-	 * of the next child device on the path to PDEV.
-	 *
-	 * In detail the two cases are (note that the 'X' below is the
-	 * 'next child on the path to PDEV' mentioned above):
-	 *
-	 * 1) PBM --> PCI bus lacking int{map,mask} --> X ... PDEV
-	 *
-	 *    Here we use regs of 'PCI bus' device.
-	 *
-	 * 2) PBM --> PCI bus with int{map,mask} --> X ... PDEV
-	 *
-	 *    Here we use regs of 'X'.  Note that X can be PDEV.
-	 */
-	if (pdev->bus->number != pbm->pci_first_busno) {
-		struct pcidev_cookie *bus_pcp, *regs_pcp;
-		struct pci_dev *bus_dev, *regs_dev;
-		int plen;
+	ret = ((interrupt - 1 + (PCI_SLOT(pdev->devfn) & 3)) & 3) + 1;
+
+	printk("%s: %s IRQ Swivel %s [%x:%x] -> [%x]\n",
+	       pbm->name, pci_name(toplevel_pdev), pci_name(pdev),
+	       interrupt, PCI_SLOT(pdev->devfn), ret);
+
+	return ret;
+}
+
+static inline unsigned int pci_apply_intmap(struct pci_pbm_info *pbm,
+					    struct pci_dev *toplevel_pdev,
+					    struct pci_dev *pbus,
+					    struct pci_dev *pdev,
+					    unsigned int interrupt,
+					    unsigned int *cnode)
+{
+	struct linux_prom_pci_intmap imap[PROM_PCIIMAP_MAX];
+	struct linux_prom_pci_intmask imask;
+	struct pcidev_cookie *pbus_pcp = pbus->sysdata;
+	struct pcidev_cookie *pdev_pcp = pdev->sysdata;
+	struct linux_prom_pci_registers *pregs = pdev_pcp->prom_regs;
+	int plen, num_imap, i;
+	unsigned int hi, mid, lo, irq, orig_interrupt;
+
+	*cnode = pbus_pcp->prom_node;
+
+	plen = prom_getproperty(pbus_pcp->prom_node, "interrupt-map",
+				(char *) &imap[0], sizeof(imap));
+	if (plen <= 0 ||
+	    (plen % sizeof(struct linux_prom_pci_intmap)) != 0) {
+		printk("%s: Device %s interrupt-map has bad len %d\n",
+		       pbm->name, pci_name(pbus), plen);
+		goto no_intmap;
+	}
+	num_imap = plen / sizeof(struct linux_prom_pci_intmap);
+
+	plen = prom_getproperty(pbus_pcp->prom_node, "interrupt-map-mask",
+				(char *) &imask, sizeof(imask));
+	if (plen <= 0 ||
+	    (plen % sizeof(struct linux_prom_pci_intmask)) != 0) {
+		printk("%s: Device %s interrupt-map-mask has bad len %d\n",
+		       pbm->name, pci_name(pbus), plen);
+		goto no_intmap;
+	}
+
+	orig_interrupt = interrupt;
 
-		bus_dev = pdev->bus->self;
-		regs_dev = pdev;
+	hi   = pregs->phys_hi & imask.phys_hi;
+	mid  = pregs->phys_mid & imask.phys_mid;
+	lo   = pregs->phys_lo & imask.phys_lo;
+	irq  = interrupt & imask.interrupt;
 
-		while (bus_dev->bus &&
-		       bus_dev->bus->number != pbm->pci_first_busno) {
-			regs_dev = bus_dev;
-			bus_dev = bus_dev->bus->self;
+	for (i = 0; i < num_imap; i++) {
+		if (imap[i].phys_hi  == hi   &&
+		    imap[i].phys_mid == mid  &&
+		    imap[i].phys_lo  == lo   &&
+		    imap[i].interrupt == irq) {
+			*cnode = imap[i].cnode;
+			interrupt = imap[i].cinterrupt;
 		}
+	}
 
-		regs_pcp = regs_dev->sysdata;
-		pregs = regs_pcp->prom_regs;
+	printk("%s: %s MAP BUS %s DEV %s [%x] -> [%x]\n",
+	       pbm->name, pci_name(toplevel_pdev),
+	       pci_name(pbus), pci_name(pdev),
+	       orig_interrupt, interrupt);
 
-		bus_pcp = bus_dev->sysdata;
+no_intmap:
+	return interrupt;
+}
 
-		/* But if the PCI bridge has it's own interrupt map
-		 * and mask properties, use that and the regs of the
-		 * PCI entity at the next level down on the path to the
-		 * device.
-		 */
-		plen = prom_getproperty(bus_pcp->prom_node, "interrupt-map",
-					(char *) &bridge_local_intmap[0],
-					sizeof(bridge_local_intmap));
-		if (plen != -1) {
-			intmap = &bridge_local_intmap[0];
-			num_intmap = plen / sizeof(struct linux_prom_pci_intmap);
-			plen = prom_getproperty(bus_pcp->prom_node,
-						"interrupt-map-mask",
-						(char *) &bridge_local_intmask,
-						sizeof(bridge_local_intmask));
-			if (plen == -1) {
-				printk("pci_intmap_match: Warning! Bridge has intmap "
-				       "but no intmask.\n");
-				printk("pci_intmap_match: Trying to recover.\n");
-				return 0;
-			}
+/* For each PCI bus on the way to the root:
+ * 1) If it has an interrupt-map property, apply it.
+ * 2) Else, swivel the interrupt number based upon the PCI device number.
+ *
+ * Return the "IRQ controller" node.  If this is the PBM's device node,
+ * all interrupt translations are complete, else we should use that node's
+ * "reg" property to apply the PBM's "interrupt-{map,mask}" to the interrupt.
+ */
+static unsigned int __init pci_intmap_match_to_root(struct pci_pbm_info *pbm,
+						    struct pci_dev *pdev,
+						    unsigned int *interrupt)
+{
+	struct pci_dev *toplevel_pdev = pdev;
+	struct pcidev_cookie *toplevel_pcp = toplevel_pdev->sysdata;
+	unsigned int cnode = toplevel_pcp->prom_node;
+
+	while (pdev->bus->number != pbm->pci_first_busno) {
+		struct pci_dev *pbus = pdev->bus->self;
+		struct pcidev_cookie *pcp = pbus->sysdata;
+		int plen;
 
-			if (pdev->bus->self != bus_dev)
-				map_slot = 1;
+		plen = prom_getproplen(pcp->prom_node, "interrupt-map");
+		if (plen <= 0) {
+			*interrupt = pci_slot_swivel(pbm, toplevel_pdev,
+						     pdev, *interrupt);
+			cnode = pcp->prom_node;
 		} else {
-			pregs = bus_pcp->prom_regs;
-			map_slot = 1;
+			*interrupt = pci_apply_intmap(pbm, toplevel_pdev,
+						      pbus, pdev,
+						      *interrupt, &cnode);
+
+			while (pcp->prom_node != cnode &&
+			       pbus->bus->number != pbm->pci_first_busno) {
+				pbus = pbus->bus->self;
+				pcp = pbus->sysdata;
+			}
 		}
-	}
+		pdev = pbus;
 
-	if (map_slot) {
-		*interrupt = ((*interrupt
-			       - 1
-			       + PCI_SLOT(pdev->devfn)) & 0x3) + 1;
+		if (cnode == pbm->prom_node)
+			break;
 	}
 
-	hi   = pregs->phys_hi & intmask->phys_hi;
-	mid  = pregs->phys_mid & intmask->phys_mid;
-	lo   = pregs->phys_lo & intmask->phys_lo;
-	irq  = *interrupt & intmask->interrupt;
-
-	for (i = 0; i < num_intmap; i++) {
-		if (intmap[i].phys_hi  == hi	&&
-		    intmap[i].phys_mid == mid	&&
-		    intmap[i].phys_lo  == lo	&&
-		    intmap[i].interrupt == irq) {
-			*interrupt = intmap[i].cinterrupt;
-			printk("PCI-IRQ: Routing bus[%2x] slot[%2x] map[%d] to INO[%02x]\n",
-			       pdev->bus->number, PCI_SLOT(pdev->devfn),
-			       map_slot, *interrupt);
-			return 1;
-		}
+	return cnode;
+}
+
+static int __init pci_intmap_match(struct pci_dev *pdev, unsigned int *interrupt)
+{
+	struct pcidev_cookie *dev_pcp = pdev->sysdata;
+	struct pci_pbm_info *pbm = dev_pcp->pbm;
+	struct linux_prom_pci_registers reg[PROMREG_MAX];
+	unsigned int hi, mid, lo, irq;
+	int i, cnode, plen;
+
+	cnode = pci_intmap_match_to_root(pbm, pdev, interrupt);
+	if (cnode == pbm->prom_node)
+		goto success;
+
+	plen = prom_getproperty(cnode, "reg", (char *) reg, sizeof(reg));
+	if (plen <= 0 ||
+	    (plen % sizeof(struct linux_prom_pci_registers)) != 0) {
+		printk("%s: OBP node %x reg property has bad len %d\n",
+		       pbm->name, cnode, plen);
+		goto fail;
 	}
 
-	/* We will run this code even if pbm->num_pbm_intmap is zero, just so
-	 * we can apply the slot mapping to the PROM interrupt property value.
-	 * So do not spit out these warnings in that case.
-	 */
-	if (num_intmap != 0) {
-		/* Print it both to OBP console and kernel one so that if bootup
-		 * hangs here the user has the information to report.
-		 */
-		prom_printf("pci_intmap_match: bus %02x, devfn %02x: ",
-			    pdev->bus->number, pdev->devfn);
-		prom_printf("IRQ [%08x.%08x.%08x.%08x] not found in interrupt-map\n",
-			    pregs->phys_hi, pregs->phys_mid, pregs->phys_lo, *interrupt);
-		prom_printf("Please email this information to davem@redhat.com\n");
-
-		printk("pci_intmap_match: bus %02x, devfn %02x: ",
-		       pdev->bus->number, pdev->devfn);
-		printk("IRQ [%08x.%08x.%08x.%08x] not found in interrupt-map\n",
-		       pregs->phys_hi, pregs->phys_mid, pregs->phys_lo, *interrupt);
-		printk("Please email this information to davem@redhat.com\n");
+	hi   = reg[0].phys_hi & pbm->pbm_intmask.phys_hi;
+	mid  = reg[0].phys_mid & pbm->pbm_intmask.phys_mid;
+	lo   = reg[0].phys_lo & pbm->pbm_intmask.phys_lo;
+	irq  = *interrupt & pbm->pbm_intmask.interrupt;
+
+	for (i = 0; i < pbm->num_pbm_intmap; i++) {
+		struct linux_prom_pci_intmap *intmap;
+
+		intmap = &pbm->pbm_intmap[i];
+
+		if (intmap->phys_hi  == hi  &&
+		    intmap->phys_mid == mid &&
+		    intmap->phys_lo  == lo  &&
+		    intmap->interrupt == irq) {
+			*interrupt = intmap->cinterrupt;
+			goto success;
+		}
 	}
 
+fail:
 	return 0;
+
+success:
+	printk("PCI-IRQ: Routing bus[%2x] slot[%2x] to INO[%02x]\n",
+	       pdev->bus->number, PCI_SLOT(pdev->devfn),
+	       *interrupt);
+	return 1;
 }
 
 static void __init pdev_fixup_irq(struct pci_dev *pdev)
@@ -703,16 +751,18 @@ static void __init pdev_fixup_irq(struct pci_dev *pdev)
 		return;
 	}
 
-	/* Fully specified already? */
-	if (((prom_irq & PCI_IRQ_IGN) >> 6) == portid) {
-		pdev->irq = p->irq_build(pbm, pdev, prom_irq);
-		goto have_irq;
-	}
+	if (tlb_type != hypervisor) {
+		/* Fully specified already? */
+		if (((prom_irq & PCI_IRQ_IGN) >> 6) == portid) {
+			pdev->irq = p->irq_build(pbm, pdev, prom_irq);
+			goto have_irq;
+		}
 
-	/* An onboard device? (bit 5 set) */
-	if ((prom_irq & PCI_IRQ_INO) & 0x20) {
-		pdev->irq = p->irq_build(pbm, pdev, (portid << 6 | prom_irq));
-		goto have_irq;
+		/* An onboard device? (bit 5 set) */
+		if ((prom_irq & PCI_IRQ_INO) & 0x20) {
+			pdev->irq = p->irq_build(pbm, pdev, (portid << 6 | prom_irq));
+			goto have_irq;
+		}
 	}
 
 	/* Can we find a matching entry in the interrupt-map? */
@@ -927,33 +977,30 @@ void pci_register_legacy_regions(struct resource *io_res,
 	struct resource *p;
 
 	/* VGA Video RAM. */
-	p = kmalloc(sizeof(*p), GFP_KERNEL);
+	p = kzalloc(sizeof(*p), GFP_KERNEL);
 	if (!p)
 		return;
 
-	memset(p, 0, sizeof(*p));
 	p->name = "Video RAM area";
 	p->start = mem_res->start + 0xa0000UL;
 	p->end = p->start + 0x1ffffUL;
 	p->flags = IORESOURCE_BUSY;
 	request_resource(mem_res, p);
 
-	p = kmalloc(sizeof(*p), GFP_KERNEL);
+	p = kzalloc(sizeof(*p), GFP_KERNEL);
 	if (!p)
 		return;
 
-	memset(p, 0, sizeof(*p));
 	p->name = "System ROM";
 	p->start = mem_res->start + 0xf0000UL;
 	p->end = p->start + 0xffffUL;
 	p->flags = IORESOURCE_BUSY;
 	request_resource(mem_res, p);
 
-	p = kmalloc(sizeof(*p), GFP_KERNEL);
+	p = kzalloc(sizeof(*p), GFP_KERNEL);
 	if (!p)
 		return;
 
-	memset(p, 0, sizeof(*p));
 	p->name = "Video ROM";
 	p->start = mem_res->start + 0xc0000UL;
 	p->end = p->start + 0x7fffUL;

arch/sparc64/kernel/pci_iommu.c

@@ -139,12 +139,11 @@ void pci_iommu_table_init(struct pci_iommu *iommu, int tsbsize, u32 dma_offset,
 	/* Allocate and initialize the free area map.  */
 	sz = num_tsb_entries / 8;
 	sz = (sz + 7UL) & ~7UL;
-	iommu->arena.map = kmalloc(sz, GFP_KERNEL);
+	iommu->arena.map = kzalloc(sz, GFP_KERNEL);
 	if (!iommu->arena.map) {
 		prom_printf("PCI_IOMMU: Error, kmalloc(arena.map) failed.\n");
 		prom_halt();
 	}
-	memset(iommu->arena.map, 0, sz);
 	iommu->arena.limit = num_tsb_entries;
 
 	/* Allocate and initialize the dummy page which we
@@ -219,7 +218,7 @@ static inline void iommu_free_ctx(struct pci_iommu *iommu, int ctx)
  * DMA for PCI device PDEV.  Return non-NULL cpu-side address if
  * successful and set *DMA_ADDRP to the PCI side dma address.
  */
-void *pci_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_addrp)
+static void *pci_4u_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_addrp)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -267,7 +266,7 @@ void *pci_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_ad
 }
 
 /* Free and unmap a consistent DMA translation. */
-void pci_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_t dvma)
+static void pci_4u_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_t dvma)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -294,7 +293,7 @@ void pci_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_
 /* Map a single buffer at PTR of SZ bytes for PCI DMA
  * in streaming mode.
  */
-dma_addr_t pci_map_single(struct pci_dev *pdev, void *ptr, size_t sz, int direction)
+static dma_addr_t pci_4u_map_single(struct pci_dev *pdev, void *ptr, size_t sz, int direction)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -415,7 +414,7 @@ do_flush_sync:
 }
 
 /* Unmap a single streaming mode DMA translation. */
-void pci_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
+static void pci_4u_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -548,7 +547,7 @@ static inline void fill_sg(iopte_t *iopte, struct scatterlist *sg,
  * When making changes here, inspect the assembly output. I was having
  * hard time to kepp this routine out of using stack slots for holding variables.
  */
-int pci_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
+static int pci_4u_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -562,9 +561,9 @@ int pci_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int
 	/* Fast path single entry scatterlists. */
 	if (nelems == 1) {
 		sglist->dma_address =
-			pci_map_single(pdev,
-				       (page_address(sglist->page) + sglist->offset),
-				       sglist->length, direction);
+			pci_4u_map_single(pdev,
+					  (page_address(sglist->page) + sglist->offset),
+					  sglist->length, direction);
 		if (unlikely(sglist->dma_address == PCI_DMA_ERROR_CODE))
 			return 0;
 		sglist->dma_length = sglist->length;
@@ -635,7 +634,7 @@ bad_no_ctx:
 }
 
 /* Unmap a set of streaming mode DMA translations. */
-void pci_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
+static void pci_4u_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -695,7 +694,7 @@ void pci_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems,
 /* Make physical memory consistent for a single
  * streaming mode DMA translation after a transfer.
  */
-void pci_dma_sync_single_for_cpu(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
+static void pci_4u_dma_sync_single_for_cpu(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -735,7 +734,7 @@ void pci_dma_sync_single_for_cpu(struct pci_dev *pdev, dma_addr_t bus_addr, size
 /* Make physical memory consistent for a set of streaming
  * mode DMA translations after a transfer.
  */
-void pci_dma_sync_sg_for_cpu(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
+static void pci_4u_dma_sync_sg_for_cpu(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
 {
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
@@ -776,6 +775,17 @@ void pci_dma_sync_sg_for_cpu(struct pci_dev *pdev, struct scatterlist *sglist, i
 	spin_unlock_irqrestore(&iommu->lock, flags);
 }
 
+struct pci_iommu_ops pci_sun4u_iommu_ops = {
+	.alloc_consistent		= pci_4u_alloc_consistent,
+	.free_consistent		= pci_4u_free_consistent,
+	.map_single			= pci_4u_map_single,
+	.unmap_single			= pci_4u_unmap_single,
+	.map_sg				= pci_4u_map_sg,
+	.unmap_sg			= pci_4u_unmap_sg,
+	.dma_sync_single_for_cpu	= pci_4u_dma_sync_single_for_cpu,
+	.dma_sync_sg_for_cpu		= pci_4u_dma_sync_sg_for_cpu,
+};
+
 static void ali_sound_dma_hack(struct pci_dev *pdev, int set_bit)
 {
 	struct pci_dev *ali_isa_bridge;

arch/sparc64/kernel/pci_psycho.c

@@ -286,17 +286,17 @@ static unsigned char psycho_pil_table[] = {
 /*0x14*/0, 0, 0, 0,	/* PCI B slot 1  Int A, B, C, D */
 /*0x18*/0, 0, 0, 0,	/* PCI B slot 2  Int A, B, C, D */
 /*0x1c*/0, 0, 0, 0,	/* PCI B slot 3  Int A, B, C, D */
-/*0x20*/4,		/* SCSI				*/
+/*0x20*/5,		/* SCSI				*/
 /*0x21*/5,		/* Ethernet			*/
 /*0x22*/8,		/* Parallel Port		*/
 /*0x23*/13,		/* Audio Record			*/
 /*0x24*/14,		/* Audio Playback		*/
 /*0x25*/15,		/* PowerFail			*/
-/*0x26*/4,		/* second SCSI			*/
+/*0x26*/5,		/* second SCSI			*/
 /*0x27*/11,		/* Floppy			*/
-/*0x28*/4,		/* Spare Hardware		*/
+/*0x28*/5,		/* Spare Hardware		*/
 /*0x29*/9,		/* Keyboard			*/
-/*0x2a*/4,		/* Mouse			*/
+/*0x2a*/5,		/* Mouse			*/
 /*0x2b*/12,		/* Serial			*/
 /*0x2c*/10,		/* Timer 0			*/
 /*0x2d*/11,		/* Timer 1			*/
@@ -313,11 +313,11 @@ static int psycho_ino_to_pil(struct pci_dev *pdev, unsigned int ino)
 
 	ret = psycho_pil_table[ino];
 	if (ret == 0 && pdev == NULL) {
-		ret = 4;
+		ret = 5;
 	} else if (ret == 0) {
 		switch ((pdev->class >> 16) & 0xff) {
 		case PCI_BASE_CLASS_STORAGE:
-			ret = 4;
+			ret = 5;
 			break;
 
 		case PCI_BASE_CLASS_NETWORK:
@@ -336,7 +336,7 @@ static int psycho_ino_to_pil(struct pci_dev *pdev, unsigned int ino)
 			break;
 
 		default:
-			ret = 4;
+			ret = 5;
 			break;
 		};
 	}
@@ -1164,7 +1164,7 @@ static void pbm_config_busmastering(struct pci_pbm_info *pbm)
 static void pbm_scan_bus(struct pci_controller_info *p,
 			 struct pci_pbm_info *pbm)
 {
-	struct pcidev_cookie *cookie = kmalloc(sizeof(*cookie), GFP_KERNEL);
+	struct pcidev_cookie *cookie = kzalloc(sizeof(*cookie), GFP_KERNEL);
 
 	if (!cookie) {
 		prom_printf("PSYCHO: Critical allocation failure.\n");
@@ -1172,7 +1172,6 @@ static void pbm_scan_bus(struct pci_controller_info *p,
 	}
 
 	/* All we care about is the PBM. */
-	memset(cookie, 0, sizeof(*cookie));
 	cookie->pbm = pbm;
 
 	pbm->pci_bus = pci_scan_bus(pbm->pci_first_busno,
@@ -1465,18 +1464,16 @@ void psycho_init(int node, char *model_name)
 		}
 	}
 
-	p = kmalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
+	p = kzalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
 	if (!p) {
 		prom_printf("PSYCHO: Fatal memory allocation error.\n");
 		prom_halt();
 	}
-	memset(p, 0, sizeof(*p));
-	iommu = kmalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
+	iommu = kzalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
 	if (!iommu) {
 		prom_printf("PSYCHO: Fatal memory allocation error.\n");
 		prom_halt();
 	}
-	memset(iommu, 0, sizeof(*iommu));
 	p->pbm_A.iommu = p->pbm_B.iommu = iommu;
 
 	p->next = pci_controller_root;

arch/sparc64/kernel/pci_sabre.c

@@ -533,17 +533,17 @@ static unsigned char sabre_pil_table[] = {
 /*0x14*/0, 0, 0, 0,	/* PCI B slot 1  Int A, B, C, D */
 /*0x18*/0, 0, 0, 0,	/* PCI B slot 2  Int A, B, C, D */
 /*0x1c*/0, 0, 0, 0,	/* PCI B slot 3  Int A, B, C, D */
-/*0x20*/4,		/* SCSI				*/
+/*0x20*/5,		/* SCSI				*/
 /*0x21*/5,		/* Ethernet			*/
 /*0x22*/8,		/* Parallel Port		*/
 /*0x23*/13,		/* Audio Record			*/
 /*0x24*/14,		/* Audio Playback		*/
 /*0x25*/15,		/* PowerFail			*/
-/*0x26*/4,		/* second SCSI			*/
+/*0x26*/5,		/* second SCSI			*/
 /*0x27*/11,		/* Floppy			*/
-/*0x28*/4,		/* Spare Hardware		*/
+/*0x28*/5,		/* Spare Hardware		*/
 /*0x29*/9,		/* Keyboard			*/
-/*0x2a*/4,		/* Mouse			*/
+/*0x2a*/5,		/* Mouse			*/
 /*0x2b*/12,		/* Serial			*/
 /*0x2c*/10,		/* Timer 0			*/
 /*0x2d*/11,		/* Timer 1			*/
@@ -565,11 +565,11 @@ static int sabre_ino_to_pil(struct pci_dev *pdev, unsigned int ino)
 
 	ret = sabre_pil_table[ino];
 	if (ret == 0 && pdev == NULL) {
-		ret = 4;
+		ret = 5;
 	} else if (ret == 0) {
 		switch ((pdev->class >> 16) & 0xff) {
 		case PCI_BASE_CLASS_STORAGE:
-			ret = 4;
+			ret = 5;
 			break;
 
 		case PCI_BASE_CLASS_NETWORK:
@@ -588,7 +588,7 @@ static int sabre_ino_to_pil(struct pci_dev *pdev, unsigned int ino)
 			break;
 
 		default:
-			ret = 4;
+			ret = 5;
 			break;
 		};
 	}
@@ -1167,7 +1167,7 @@ static void apb_init(struct pci_controller_info *p, struct pci_bus *sabre_bus)
 
 static struct pcidev_cookie *alloc_bridge_cookie(struct pci_pbm_info *pbm)
 {
-	struct pcidev_cookie *cookie = kmalloc(sizeof(*cookie), GFP_KERNEL);
+	struct pcidev_cookie *cookie = kzalloc(sizeof(*cookie), GFP_KERNEL);
 
 	if (!cookie) {
 		prom_printf("SABRE: Critical allocation failure.\n");
@@ -1175,7 +1175,6 @@ static struct pcidev_cookie *alloc_bridge_cookie(struct pci_pbm_info *pbm)
 	}
 
 	/* All we care about is the PBM. */
-	memset(cookie, 0, sizeof(*cookie));
 	cookie->pbm = pbm;
 
 	return cookie;
@@ -1556,19 +1555,17 @@ void sabre_init(int pnode, char *model_name)
 		}
 	}
 
-	p = kmalloc(sizeof(*p), GFP_ATOMIC);
+	p = kzalloc(sizeof(*p), GFP_ATOMIC);
 	if (!p) {
 		prom_printf("SABRE: Error, kmalloc(pci_controller_info) failed.\n");
 		prom_halt();
 	}
-	memset(p, 0, sizeof(*p));
 
-	iommu = kmalloc(sizeof(*iommu), GFP_ATOMIC);
+	iommu = kzalloc(sizeof(*iommu), GFP_ATOMIC);
 	if (!iommu) {
 		prom_printf("SABRE: Error, kmalloc(pci_iommu) failed.\n");
 		prom_halt();
 	}
-	memset(iommu, 0, sizeof(*iommu));
 	p->pbm_A.iommu = p->pbm_B.iommu = iommu;
 
 	upa_portid = prom_getintdefault(pnode, "upa-portid", 0xff);

arch/sparc64/kernel/pci_schizo.c

@@ -243,8 +243,8 @@ static unsigned char schizo_pil_table[] = {
 /*0x0c*/0, 0, 0, 0,	/* PCI slot 3  Int A, B, C, D	*/
 /*0x10*/0, 0, 0, 0,	/* PCI slot 4  Int A, B, C, D	*/
 /*0x14*/0, 0, 0, 0,	/* PCI slot 5  Int A, B, C, D	*/
-/*0x18*/4,		/* SCSI				*/
-/*0x19*/4,		/* second SCSI			*/
+/*0x18*/5,		/* SCSI				*/
+/*0x19*/5,		/* second SCSI			*/
 /*0x1a*/0,		/* UNKNOWN			*/
 /*0x1b*/0,		/* UNKNOWN			*/
 /*0x1c*/8,		/* Parallel			*/
@@ -254,7 +254,7 @@ static unsigned char schizo_pil_table[] = {
 /*0x20*/13,		/* Audio Record			*/
 /*0x21*/14,		/* Audio Playback		*/
 /*0x22*/12,		/* Serial			*/
-/*0x23*/4,		/* EBUS I2C 			*/
+/*0x23*/5,		/* EBUS I2C 			*/
 /*0x24*/10,		/* RTC Clock			*/
 /*0x25*/11,		/* Floppy			*/
 /*0x26*/0,		/* UNKNOWN			*/
@@ -296,11 +296,11 @@ static int schizo_ino_to_pil(struct pci_dev *pdev, unsigned int ino)
 
 	ret = schizo_pil_table[ino];
 	if (ret == 0 && pdev == NULL) {
-		ret = 4;
+		ret = 5;
 	} else if (ret == 0) {
 		switch ((pdev->class >> 16) & 0xff) {
 		case PCI_BASE_CLASS_STORAGE:
-			ret = 4;
+			ret = 5;
 			break;
 
 		case PCI_BASE_CLASS_NETWORK:
@@ -319,7 +319,7 @@ static int schizo_ino_to_pil(struct pci_dev *pdev, unsigned int ino)
 			break;
 
 		default:
-			ret = 4;
+			ret = 5;
 			break;
 		};
 	}
@@ -1525,7 +1525,7 @@ static void pbm_config_busmastering(struct pci_pbm_info *pbm)
 static void pbm_scan_bus(struct pci_controller_info *p,
 			 struct pci_pbm_info *pbm)
 {
-	struct pcidev_cookie *cookie = kmalloc(sizeof(*cookie), GFP_KERNEL);
+	struct pcidev_cookie *cookie = kzalloc(sizeof(*cookie), GFP_KERNEL);
 
 	if (!cookie) {
 		prom_printf("%s: Critical allocation failure.\n", pbm->name);
@@ -1533,7 +1533,6 @@ static void pbm_scan_bus(struct pci_controller_info *p,
 	}
 
 	/* All we care about is the PBM. */
-	memset(cookie, 0, sizeof(*cookie));
 	cookie->pbm = pbm;
 
 	pbm->pci_bus = pci_scan_bus(pbm->pci_first_busno,
@@ -2120,27 +2119,24 @@ static void __schizo_init(int node, char *model_name, int chip_type)
 		}
 	}
 
-	p = kmalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
+	p = kzalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
 	if (!p) {
 		prom_printf("SCHIZO: Fatal memory allocation error.\n");
 		prom_halt();
 	}
-	memset(p, 0, sizeof(*p));
 
-	iommu = kmalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
+	iommu = kzalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
 	if (!iommu) {
 		prom_printf("SCHIZO: Fatal memory allocation error.\n");
 		prom_halt();
 	}
-	memset(iommu, 0, sizeof(*iommu));
 	p->pbm_A.iommu = iommu;
 
-	iommu = kmalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
+	iommu = kzalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
 	if (!iommu) {
 		prom_printf("SCHIZO: Fatal memory allocation error.\n");
 		prom_halt();
 	}
-	memset(iommu, 0, sizeof(*iommu));
 	p->pbm_B.iommu = iommu;
 
 	p->next = pci_controller_root;

arch/sparc64/kernel/pci_sun4v.h

+/* pci_sun4v.h: SUN4V specific PCI controller support.
+ *
+ * Copyright (C) 2006 David S. Miller (davem@davemloft.net)
+ */
+
+#ifndef _PCI_SUN4V_H
+#define _PCI_SUN4V_H
+
+extern long pci_sun4v_iommu_map(unsigned long devhandle,
+				unsigned long tsbid,
+				unsigned long num_ttes,
+				unsigned long io_attributes,
+				unsigned long io_page_list_pa);
+extern unsigned long pci_sun4v_iommu_demap(unsigned long devhandle,
+					   unsigned long tsbid,
+					   unsigned long num_ttes);
+extern unsigned long pci_sun4v_iommu_getmap(unsigned long devhandle,
+					    unsigned long tsbid,
+					    unsigned long *io_attributes,
+					    unsigned long *real_address);
+extern unsigned long pci_sun4v_config_get(unsigned long devhandle,
+					  unsigned long pci_device,
+					  unsigned long config_offset,
+					  unsigned long size);
+extern int pci_sun4v_config_put(unsigned long devhandle,
+				unsigned long pci_device,
+				unsigned long config_offset,
+				unsigned long size,
+				unsigned long data);
+
+#endif /* !(_PCI_SUN4V_H) */

arch/sparc64/kernel/pci_sun4v_asm.S

+/* pci_sun4v_asm: Hypervisor calls for PCI support.
+ *
+ * Copyright (C) 2006 David S. Miller <davem@davemloft.net>
+ */
+
+#include <asm/hypervisor.h>
+
+	/* %o0: devhandle
+	 * %o1:	tsbid
+	 * %o2:	num ttes
+	 * %o3:	io_attributes
+	 * %o4:	io_page_list phys address
+	 *
+	 * returns %o0:	-status if status was non-zero, else
+	 *         %o0:	num pages mapped
+	 */
+	.globl	pci_sun4v_iommu_map
+pci_sun4v_iommu_map:
+	mov	%o5, %g1
+	mov	HV_FAST_PCI_IOMMU_MAP, %o5
+	ta	HV_FAST_TRAP
+	brnz,pn %o0, 1f
+	 sub	%g0, %o0, %o0
+	mov	%o1, %o0
+1:	retl
+	 nop
+
+	/* %o0: devhandle
+	 * %o1:	tsbid
+	 * %o2:	num ttes
+	 *
+	 * returns %o0:	num ttes demapped
+	 */
+	.globl	pci_sun4v_iommu_demap
+pci_sun4v_iommu_demap:
+	mov	HV_FAST_PCI_IOMMU_DEMAP, %o5
+	ta	HV_FAST_TRAP
+	retl
+	 mov	%o1, %o0
+
+	/* %o0: devhandle
+	 * %o1:	tsbid
+	 * %o2:	&io_attributes
+	 * %o3:	&real_address
+	 *
+	 * returns %o0:	status
+	 */
+	.globl	pci_sun4v_iommu_getmap
+pci_sun4v_iommu_getmap:
+	mov	%o2, %o4
+	mov	HV_FAST_PCI_IOMMU_GETMAP, %o5
+	ta	HV_FAST_TRAP
+	stx	%o1, [%o4]
+	stx	%o2, [%o3]
+	retl
+	 mov	%o0, %o0
+
+	/* %o0: devhandle
+	 * %o1:	pci_device
+	 * %o2:	pci_config_offset
+	 * %o3:	size
+	 *
+	 * returns %o0:	data
+	 *
+	 * If there is an error, the data will be returned
+	 * as all 1's.
+	 */
+	.globl	pci_sun4v_config_get
+pci_sun4v_config_get:
+	mov	HV_FAST_PCI_CONFIG_GET, %o5
+	ta	HV_FAST_TRAP
+	brnz,a,pn %o1, 1f
+	 mov	-1, %o2
+1:	retl
+	 mov	%o2, %o0
+
+	/* %o0: devhandle
+	 * %o1:	pci_device
+	 * %o2:	pci_config_offset
+	 * %o3:	size
+	 * %o4:	data
+	 *
+	 * returns %o0:	status
+	 *
+	 * status will be zero if the operation completed
+	 * successfully, else -1 if not
+	 */
+	.globl	pci_sun4v_config_put
+pci_sun4v_config_put:
+	mov	HV_FAST_PCI_CONFIG_PUT, %o5
+	ta	HV_FAST_TRAP
+	brnz,a,pn %o1, 1f
+	 mov	-1, %o1
+1:	retl
+	 mov	%o1, %o0

arch/sparc64/kernel/ptrace.c

@@ -124,6 +124,9 @@ void flush_ptrace_access(struct vm_area_struct *vma, struct page *page,
 {
 	BUG_ON(len > PAGE_SIZE);
 
+	if (tlb_type == hypervisor)
+		return;
+
 #ifdef DCACHE_ALIASING_POSSIBLE
 	/* If bit 13 of the kernel address we used to access the
 	 * user page is the same as the virtual address that page

arch/sparc64/kernel/sbus.c

@@ -693,11 +693,11 @@ void sbus_set_sbus64(struct sbus_dev *sdev, int bursts)
 
 /* SBUS SYSIO INO number to Sparc PIL level. */
 static unsigned char sysio_ino_to_pil[] = {
-	0, 4, 4, 7, 5, 7, 8, 9,		/* SBUS slot 0 */
-	0, 4, 4, 7, 5, 7, 8, 9,		/* SBUS slot 1 */
-	0, 4, 4, 7, 5, 7, 8, 9,		/* SBUS slot 2 */
-	0, 4, 4, 7, 5, 7, 8, 9,		/* SBUS slot 3 */
-	4, /* Onboard SCSI */
+	0, 5, 5, 7, 5, 7, 8, 9,		/* SBUS slot 0 */
+	0, 5, 5, 7, 5, 7, 8, 9,		/* SBUS slot 1 */
+	0, 5, 5, 7, 5, 7, 8, 9,		/* SBUS slot 2 */
+	0, 5, 5, 7, 5, 7, 8, 9,		/* SBUS slot 3 */
+	5, /* Onboard SCSI */
 	5, /* Onboard Ethernet */
 /*XXX*/	8, /* Onboard BPP */
 	0, /* Bogon */

arch/sparc64/kernel/sparc64_ksyms.c

@@ -95,9 +95,6 @@ extern int __ashrdi3(int, int);
 
 extern int dump_fpu (struct pt_regs * regs, elf_fpregset_t * fpregs);
 
-extern unsigned long phys_base;
-extern unsigned long pfn_base;
-
 extern unsigned int sys_call_table[];
 
 extern void xor_vis_2(unsigned long, unsigned long *, unsigned long *);
@@ -108,6 +105,14 @@ extern void xor_vis_4(unsigned long, unsigned long *, unsigned long *,
 extern void xor_vis_5(unsigned long, unsigned long *, unsigned long *,
 		      unsigned long *, unsigned long *, unsigned long *);
 
+extern void xor_niagara_2(unsigned long, unsigned long *, unsigned long *);
+extern void xor_niagara_3(unsigned long, unsigned long *, unsigned long *,
+			  unsigned long *);
+extern void xor_niagara_4(unsigned long, unsigned long *, unsigned long *,
+			  unsigned long *, unsigned long *);
+extern void xor_niagara_5(unsigned long, unsigned long *, unsigned long *,
+			  unsigned long *, unsigned long *, unsigned long *);
+
 /* Per-CPU information table */
 EXPORT_PER_CPU_SYMBOL(__cpu_data);
 
@@ -241,10 +246,6 @@ EXPORT_SYMBOL(verify_compat_iovec);
 #endif
 
 EXPORT_SYMBOL(dump_fpu);
-EXPORT_SYMBOL(pte_alloc_one_kernel);
-#ifndef CONFIG_SMP
-EXPORT_SYMBOL(pgt_quicklists);
-#endif
 EXPORT_SYMBOL(put_fs_struct);
 
 /* math-emu wants this */
@@ -339,14 +340,10 @@ EXPORT_SYMBOL(copy_to_user_fixup);
 EXPORT_SYMBOL(copy_from_user_fixup);
 EXPORT_SYMBOL(copy_in_user_fixup);
 EXPORT_SYMBOL(__strncpy_from_user);
-EXPORT_SYMBOL(__bzero_noasi);
+EXPORT_SYMBOL(__clear_user);
 
 /* Various address conversion macros use this. */
-EXPORT_SYMBOL(phys_base);
-EXPORT_SYMBOL(pfn_base);
 EXPORT_SYMBOL(sparc64_valid_addr_bitmap);
-EXPORT_SYMBOL(page_to_pfn);
-EXPORT_SYMBOL(pfn_to_page);
 
 /* No version information on this, heavily used in inline asm,
  * and will always be 'void __ret_efault(void)'.
@@ -392,4 +389,9 @@ EXPORT_SYMBOL(xor_vis_3);
 EXPORT_SYMBOL(xor_vis_4);
 EXPORT_SYMBOL(xor_vis_5);
 
+EXPORT_SYMBOL(xor_niagara_2);
+EXPORT_SYMBOL(xor_niagara_3);
+EXPORT_SYMBOL(xor_niagara_4);
+EXPORT_SYMBOL(xor_niagara_5);
+
 EXPORT_SYMBOL(prom_palette);

arch/sparc64/lib/Makefile

@@ -11,6 +11,8 @@ lib-y := PeeCeeI.o copy_page.o clear_page.o strlen.o strncmp.o \
 	 VISsave.o atomic.o bitops.o \
 	 U1memcpy.o U1copy_from_user.o U1copy_to_user.o \
 	 U3memcpy.o U3copy_from_user.o U3copy_to_user.o U3patch.o \
+	 NGmemcpy.o NGcopy_from_user.o NGcopy_to_user.o NGpatch.o \
+	 NGpage.o NGbzero.o \
 	 copy_in_user.o user_fixup.o memmove.o \
 	 mcount.o ipcsum.o rwsem.o xor.o find_bit.o delay.o