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.
  ...

38 files changed, 3797 insertions, 682 deletions

diff --git a/include/asm-sparc/idprom.h b/include/asm-sparc/idprom.h
index d856e64..59083ed 100644
--- a/include/asm-sparc/idprom.h
+++ b/include/asm-sparc/idprom.h
@@ -7,27 +7,19 @@
 #ifndef _SPARC_IDPROM_H
 #define _SPARC_IDPROM_H
 
-/* Offset into the EEPROM where the id PROM is located on the 4c */
-#define IDPROM_OFFSET  0x7d8
+#include <linux/types.h>
 
-/* On sun4m; physical. */
-/* MicroSPARC(-II) does not decode 31rd bit, but it works. */
-#define IDPROM_OFFSET_M  0xfd8
-
-struct idprom
-{
-	unsigned char	id_format;	/* Format identifier (always 0x01) */
-	unsigned char	id_machtype;	/* Machine type */
-	unsigned char	id_ethaddr[6];	/* Hardware ethernet address */
-	long		id_date;	/* Date of manufacture */
-	unsigned int	id_sernum:24;	/* Unique serial number */
-	unsigned char	id_cksum;	/* Checksum - xor of the data bytes */
-	unsigned char	reserved[16];
+struct idprom {
+	u8		id_format;	/* Format identifier (always 0x01) */
+	u8		id_machtype;	/* Machine type */
+	u8		id_ethaddr[6];	/* Hardware ethernet address */
+	s32		id_date;	/* Date of manufacture */
+	u32		id_sernum:24;	/* Unique serial number */
+	u8		id_cksum;	/* Checksum - xor of the data bytes */
+	u8		reserved[16];
 };
 
 extern struct idprom *idprom;
 extern void idprom_init(void);
 
-#define IDPROM_SIZE  (sizeof(struct idprom))
-
 #endif /* !(_SPARC_IDPROM_H) */
diff --git a/include/asm-sparc/oplib.h b/include/asm-sparc/oplib.h
index d0d76b3..f283f8a 100644
--- a/include/asm-sparc/oplib.h
+++ b/include/asm-sparc/oplib.h
@@ -165,6 +165,7 @@ enum prom_input_device {
 	PROMDEV_ITTYA,			/* input from ttya */
 	PROMDEV_ITTYB,			/* input from ttyb */
 	PROMDEV_IRSC,			/* input from rsc */
+	PROMDEV_IVCONS,			/* input from virtual-console */
 	PROMDEV_I_UNK,
 };
 
@@ -177,6 +178,7 @@ enum prom_output_device {
 	PROMDEV_OTTYA,			/* to ttya */
 	PROMDEV_OTTYB,			/* to ttyb */
 	PROMDEV_ORSC,			/* to rsc */
+	PROMDEV_OVCONS,			/* to virtual-console */
 	PROMDEV_O_UNK,
 };
 
diff --git a/include/asm-sparc/uaccess.h b/include/asm-sparc/uaccess.h
index f8f1ec1..3cf132e 100644
--- a/include/asm-sparc/uaccess.h
+++ b/include/asm-sparc/uaccess.h
@@ -120,17 +120,6 @@ case 8: __put_user_asm(x,d,addr,__pu_ret); break; \
 default: __pu_ret = __put_user_bad(); break; \
 } } else { __pu_ret = -EFAULT; } __pu_ret; })
 
-#define __put_user_check_ret(x,addr,size,retval) ({ \
-register int __foo __asm__ ("l1"); \
-if (__access_ok(addr,size)) { \
-switch (size) { \
-case 1: __put_user_asm_ret(x,b,addr,retval,__foo); break; \
-case 2: __put_user_asm_ret(x,h,addr,retval,__foo); break; \
-case 4: __put_user_asm_ret(x,,addr,retval,__foo); break; \
-case 8: __put_user_asm_ret(x,d,addr,retval,__foo); break; \
-default: if (__put_user_bad()) return retval; break; \
-} } else return retval; })
-
 #define __put_user_nocheck(x,addr,size) ({ \
 register int __pu_ret; \
 switch (size) { \
@@ -141,16 +130,6 @@ case 8: __put_user_asm(x,d,addr,__pu_ret); break; \
 default: __pu_ret = __put_user_bad(); break; \
 } __pu_ret; })
 
-#define __put_user_nocheck_ret(x,addr,size,retval) ({ \
-register int __foo __asm__ ("l1"); \
-switch (size) { \
-case 1: __put_user_asm_ret(x,b,addr,retval,__foo); break; \
-case 2: __put_user_asm_ret(x,h,addr,retval,__foo); break; \
-case 4: __put_user_asm_ret(x,,addr,retval,__foo); break; \
-case 8: __put_user_asm_ret(x,d,addr,retval,__foo); break; \
-default: if (__put_user_bad()) return retval; break; \
-} })
-
 #define __put_user_asm(x,size,addr,ret)					\
 __asm__ __volatile__(							\
 	"/* Put user asm, inline. */\n"					\
@@ -170,32 +149,6 @@ __asm__ __volatile__(							\
        : "=&r" (ret) : "r" (x), "m" (*__m(addr)),			\
 	 "i" (-EFAULT))
 
-#define __put_user_asm_ret(x,size,addr,ret,foo)				\
-if (__builtin_constant_p(ret) && ret == -EFAULT)			\
-__asm__ __volatile__(							\
-	"/* Put user asm ret, inline. */\n"				\
-"1:\t"	"st"#size " %1, %2\n\n\t"					\
-	".section __ex_table,#alloc\n\t"				\
-	".align	4\n\t"							\
-	".word	1b, __ret_efault\n\n\t"					\
-	".previous\n\n\t"						\
-       : "=r" (foo) : "r" (x), "m" (*__m(addr)));			\
-else									\
-__asm__ __volatile(							\
-	"/* Put user asm ret, inline. */\n"				\
-"1:\t"	"st"#size " %1, %2\n\n\t"					\
-	".section .fixup,#alloc,#execinstr\n\t"				\
-	".align	4\n"							\
-"3:\n\t"								\
-	"ret\n\t"							\
-	" restore %%g0, %3, %%o0\n\t"					\
-	".previous\n\n\t"						\
-	".section __ex_table,#alloc\n\t"				\
-	".align	4\n\t"							\
-	".word	1b, 3b\n\n\t"						\
-	".previous\n\n\t"						\
-       : "=r" (foo) : "r" (x), "m" (*__m(addr)), "i" (ret))
-
 extern int __put_user_bad(void);
 
 #define __get_user_check(x,addr,size,type) ({ \
diff --git a/include/asm-sparc64/a.out.h b/include/asm-sparc64/a.out.h
index 02af289..35cb5c9 100644
--- a/include/asm-sparc64/a.out.h
+++ b/include/asm-sparc64/a.out.h
@@ -95,7 +95,11 @@ struct relocation_info /* used when header.a_machtype == M_SPARC */
 
 #ifdef __KERNEL__
 
-#define STACK_TOP (test_thread_flag(TIF_32BIT) ? 0xf0000000 : 0x80000000000L)
+#define STACK_TOP32	((1UL << 32UL) - PAGE_SIZE)
+#define STACK_TOP64	(0x0000080000000000UL - (1UL << 32UL))
+
+#define STACK_TOP (test_thread_flag(TIF_32BIT) ? \
+		   STACK_TOP32 : STACK_TOP64)
 
 #endif
 
diff --git a/include/asm-sparc64/asi.h b/include/asm-sparc64/asi.h
index 5348556..662a211 100644
--- a/include/asm-sparc64/asi.h
+++ b/include/asm-sparc64/asi.h
@@ -25,14 +25,27 @@
 
 /* SpitFire and later extended ASIs.  The "(III)" marker designates
  * UltraSparc-III and later specific ASIs.  The "(CMT)" marker designates
- * Chip Multi Threading specific ASIs.
+ * Chip Multi Threading specific ASIs.  "(NG)" designates Niagara specific
+ * ASIs, "(4V)" designates SUN4V specific ASIs.
  */
 #define ASI_PHYS_USE_EC		0x14 /* PADDR, E-cachable		*/
 #define ASI_PHYS_BYPASS_EC_E	0x15 /* PADDR, E-bit			*/
+#define ASI_BLK_AIUP_4V		0x16 /* (4V) Prim, user, block ld/st	*/
+#define ASI_BLK_AIUS_4V		0x17 /* (4V) Sec, user, block ld/st	*/
 #define ASI_PHYS_USE_EC_L	0x1c /* PADDR, E-cachable, little endian*/
 #define ASI_PHYS_BYPASS_EC_E_L	0x1d /* PADDR, E-bit, little endian	*/
+#define ASI_BLK_AIUP_L_4V	0x1e /* (4V) Prim, user, block, l-endian*/
+#define ASI_BLK_AIUS_L_4V	0x1f /* (4V) Sec, user, block, l-endian	*/
+#define ASI_SCRATCHPAD		0x20 /* (4V) Scratch Pad Registers	*/
+#define ASI_MMU			0x21 /* (4V) MMU Context Registers	*/
+#define ASI_BLK_INIT_QUAD_LDD_AIUS 0x23 /* (NG) init-store, twin load,
+					 * secondary, user
+					 */
 #define ASI_NUCLEUS_QUAD_LDD	0x24 /* Cachable, qword load		*/
+#define ASI_QUEUE		0x25 /* (4V) Interrupt Queue Registers	*/
+#define ASI_QUAD_LDD_PHYS_4V	0x26 /* (4V) Physical, qword load	*/
 #define ASI_NUCLEUS_QUAD_LDD_L	0x2c /* Cachable, qword load, l-endian 	*/
+#define ASI_QUAD_LDD_PHYS_L_4V	0x2e /* (4V) Phys, qword load, l-endian	*/
 #define ASI_PCACHE_DATA_STATUS	0x30 /* (III) PCache data stat RAM diag	*/
 #define ASI_PCACHE_DATA		0x31 /* (III) PCache data RAM diag	*/
 #define ASI_PCACHE_TAG		0x32 /* (III) PCache tag RAM diag	*/
@@ -137,6 +150,9 @@
 #define ASI_FL16_SL		0xdb /* Secondary, 1 16-bit, fpu ld/st,L*/
 #define ASI_BLK_COMMIT_P	0xe0 /* Primary, blk store commit	*/
 #define ASI_BLK_COMMIT_S	0xe1 /* Secondary, blk store commit	*/
+#define ASI_BLK_INIT_QUAD_LDD_P	0xe2 /* (NG) init-store, twin load,
+				      * primary, implicit
+				      */
 #define ASI_BLK_P		0xf0 /* Primary, blk ld/st		*/
 #define ASI_BLK_S		0xf1 /* Secondary, blk ld/st		*/
 #define ASI_BLK_PL		0xf8 /* Primary, blk ld/st, little	*/
diff --git a/include/asm-sparc64/cpudata.h b/include/asm-sparc64/cpudata.h
index 74de79d..c66a81b 100644
--- a/include/asm-sparc64/cpudata.h
+++ b/include/asm-sparc64/cpudata.h
@@ -1,41 +1,224 @@
 /* cpudata.h: Per-cpu parameters.
  *
- * Copyright (C) 2003, 2005 David S. Miller (davem@redhat.com)
+ * Copyright (C) 2003, 2005, 2006 David S. Miller (davem@davemloft.net)
  */
 
 #ifndef _SPARC64_CPUDATA_H
 #define _SPARC64_CPUDATA_H
 
+#include <asm/hypervisor.h>
+#include <asm/asi.h>
+
+#ifndef __ASSEMBLY__
+
 #include <linux/percpu.h>
+#include <linux/threads.h>
 
 typedef struct {
 	/* Dcache line 1 */
 	unsigned int	__softirq_pending; /* must be 1st, see rtrap.S */
 	unsigned int	multiplier;
 	unsigned int	counter;
-	unsigned int	idle_volume;
+	unsigned int	__pad1;
 	unsigned long	clock_tick;	/* %tick's per second */
 	unsigned long	udelay_val;
 
-	/* Dcache line 2 */
-	unsigned int	pgcache_size;
-	unsigned int	__pad1;
-	unsigned long	*pte_cache[2];
-	unsigned long	*pgd_cache;
-
-	/* Dcache line 3, rarely used */
+	/* Dcache line 2, rarely used */
 	unsigned int	dcache_size;
 	unsigned int	dcache_line_size;
 	unsigned int	icache_size;
 	unsigned int	icache_line_size;
 	unsigned int	ecache_size;
 	unsigned int	ecache_line_size;
-	unsigned int	__pad2;
 	unsigned int	__pad3;
+	unsigned int	__pad4;
 } cpuinfo_sparc;
 
 DECLARE_PER_CPU(cpuinfo_sparc, __cpu_data);
 #define cpu_data(__cpu)		per_cpu(__cpu_data, (__cpu))
 #define local_cpu_data()	__get_cpu_var(__cpu_data)
 
+/* Trap handling code needs to get at a few critical values upon
+ * trap entry and to process TSB misses.  These cannot be in the
+ * per_cpu() area as we really need to lock them into the TLB and
+ * thus make them part of the main kernel image.  As a result we
+ * try to make this as small as possible.
+ *
+ * This is padded out and aligned to 64-bytes to avoid false sharing
+ * on SMP.
+ */
+
+/* If you modify the size of this structure, please update
+ * TRAP_BLOCK_SZ_SHIFT below.
+ */
+struct thread_info;
+struct trap_per_cpu {
+/* D-cache line 1: Basic thread information, cpu and device mondo queues */
+	struct thread_info	*thread;
+	unsigned long		pgd_paddr;
+	unsigned long		cpu_mondo_pa;
+	unsigned long		dev_mondo_pa;
+
+/* D-cache line 2: Error Mondo Queue and kernel buffer pointers */
+	unsigned long		resum_mondo_pa;
+	unsigned long		resum_kernel_buf_pa;
+	unsigned long		nonresum_mondo_pa;
+	unsigned long		nonresum_kernel_buf_pa;
+
+/* Dcache lines 3, 4, 5, and 6: Hypervisor Fault Status */
+	struct hv_fault_status	fault_info;
+
+/* Dcache line 7: Physical addresses of CPU send mondo block and CPU list.  */
+	unsigned long		cpu_mondo_block_pa;
+	unsigned long		cpu_list_pa;
+	unsigned long		__pad1[2];
+
+/* Dcache line 8: Unused, needed to keep trap_block a power-of-2 in size.  */
+	unsigned long		__pad2[4];
+} __attribute__((aligned(64)));
+extern struct trap_per_cpu trap_block[NR_CPUS];
+extern void init_cur_cpu_trap(struct thread_info *);
+extern void setup_tba(void);
+
+struct cpuid_patch_entry {
+	unsigned int	addr;
+	unsigned int	cheetah_safari[4];
+	unsigned int	cheetah_jbus[4];
+	unsigned int	starfire[4];
+	unsigned int	sun4v[4];
+};
+extern struct cpuid_patch_entry __cpuid_patch, __cpuid_patch_end;
+
+struct sun4v_1insn_patch_entry {
+	unsigned int	addr;
+	unsigned int	insn;
+};
+extern struct sun4v_1insn_patch_entry __sun4v_1insn_patch,
+	__sun4v_1insn_patch_end;
+
+struct sun4v_2insn_patch_entry {
+	unsigned int	addr;
+	unsigned int	insns[2];
+};
+extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
+	__sun4v_2insn_patch_end;
+
+#endif /* !(__ASSEMBLY__) */
+
+#define TRAP_PER_CPU_THREAD		0x00
+#define TRAP_PER_CPU_PGD_PADDR		0x08
+#define TRAP_PER_CPU_CPU_MONDO_PA	0x10
+#define TRAP_PER_CPU_DEV_MONDO_PA	0x18
+#define TRAP_PER_CPU_RESUM_MONDO_PA	0x20
+#define TRAP_PER_CPU_RESUM_KBUF_PA	0x28
+#define TRAP_PER_CPU_NONRESUM_MONDO_PA	0x30
+#define TRAP_PER_CPU_NONRESUM_KBUF_PA	0x38
+#define TRAP_PER_CPU_FAULT_INFO		0x40
+#define TRAP_PER_CPU_CPU_MONDO_BLOCK_PA	0xc0
+#define TRAP_PER_CPU_CPU_LIST_PA	0xc8
+
+#define TRAP_BLOCK_SZ_SHIFT		8
+
+#include <asm/scratchpad.h>
+
+#define __GET_CPUID(REG)				\
+	/* Spitfire implementation (default). */	\
+661:	ldxa		[%g0] ASI_UPA_CONFIG, REG;	\
+	srlx		REG, 17, REG;			\
+	 and		REG, 0x1f, REG;			\
+	nop;						\
+	.section	.cpuid_patch, "ax";		\
+	/* Instruction location. */			\
+	.word		661b;				\
+	/* Cheetah Safari implementation. */		\
+	ldxa		[%g0] ASI_SAFARI_CONFIG, REG;	\
+	srlx		REG, 17, REG;			\
+	and		REG, 0x3ff, REG;		\
+	nop;						\
+	/* Cheetah JBUS implementation. */		\
+	ldxa		[%g0] ASI_JBUS_CONFIG, REG;	\
+	srlx		REG, 17, REG;			\
+	and		REG, 0x1f, REG;			\
+	nop;						\
+	/* Starfire implementation. */			\
+	sethi		%hi(0x1fff40000d0 >> 9), REG;	\
+	sllx		REG, 9, REG;			\
+	or		REG, 0xd0, REG;			\
+	lduwa		[REG] ASI_PHYS_BYPASS_EC_E, REG;\
+	/* sun4v implementation. */			\
+	mov		SCRATCHPAD_CPUID, REG;		\
+	ldxa		[REG] ASI_SCRATCHPAD, REG;	\
+	nop;						\
+	nop;						\
+	.previous;
+
+#ifdef CONFIG_SMP
+
+#define TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	__GET_CPUID(TMP)			\
+	sethi	%hi(trap_block), DEST;		\
+	sllx	TMP, TRAP_BLOCK_SZ_SHIFT, TMP;	\
+	or	DEST, %lo(trap_block), DEST;	\
+	add	DEST, TMP, DEST;		\
+
+/* Clobbers TMP, current address space PGD phys address into DEST.  */
+#define TRAP_LOAD_PGD_PHYS(DEST, TMP)		\
+	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	ldx	[DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
+
+/* Clobbers TMP, loads local processor's IRQ work area into DEST.  */
+#define TRAP_LOAD_IRQ_WORK(DEST, TMP)		\
+	__GET_CPUID(TMP)			\
+	sethi	%hi(__irq_work), DEST;		\
+	sllx	TMP, 6, TMP;			\
+	or	DEST, %lo(__irq_work), DEST;	\
+	add	DEST, TMP, DEST;
+
+/* Clobbers TMP, loads DEST with current thread info pointer.  */
+#define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
+	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	ldx	[DEST + TRAP_PER_CPU_THREAD], DEST;
+
+/* Given the current thread info pointer in THR, load the per-cpu
+ * area base of the current processor into DEST.  REG1, REG2, and REG3 are
+ * clobbered.
+ *
+ * You absolutely cannot use DEST as a temporary in this code.  The
+ * reason is that traps can happen during execution, and return from
+ * trap will load the fully resolved DEST per-cpu base.  This can corrupt
+ * the calculations done by the macro mid-stream.
+ */
+#define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3)	\
+	ldub	[THR + TI_CPU], REG1;			\
+	sethi	%hi(__per_cpu_shift), REG3;		\
+	sethi	%hi(__per_cpu_base), REG2;		\
+	ldx	[REG3 + %lo(__per_cpu_shift)], REG3;	\
+	ldx	[REG2 + %lo(__per_cpu_base)], REG2;	\
+	sllx	REG1, REG3, REG3;			\
+	add	REG3, REG2, DEST;
+
+#else
+
+#define TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	sethi	%hi(trap_block), DEST;		\
+	or	DEST, %lo(trap_block), DEST;	\
+
+/* Uniprocessor versions, we know the cpuid is zero.  */
+#define TRAP_LOAD_PGD_PHYS(DEST, TMP)		\
+	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	ldx	[DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
+
+#define TRAP_LOAD_IRQ_WORK(DEST, TMP)		\
+	sethi	%hi(__irq_work), DEST;		\
+	or	DEST, %lo(__irq_work), DEST;
+
+#define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
+	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	ldx	[DEST + TRAP_PER_CPU_THREAD], DEST;
+
+/* No per-cpu areas on uniprocessor, so no need to load DEST.  */
+#define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3)
+
+#endif /* !(CONFIG_SMP) */
+
 #endif /* _SPARC64_CPUDATA_H */
diff --git a/include/asm-sparc64/elf.h b/include/asm-sparc64/elf.h
index 69539a8..303d85e 100644
--- a/include/asm-sparc64/elf.h
+++ b/include/asm-sparc64/elf.h
@@ -10,6 +10,7 @@
 #ifdef __KERNEL__
 #include <asm/processor.h>
 #include <asm/uaccess.h>
+#include <asm/spitfire.h>
 #endif
 
 /*
@@ -68,6 +69,7 @@
 #define HWCAP_SPARC_MULDIV      8
 #define HWCAP_SPARC_V9		16
 #define HWCAP_SPARC_ULTRA3	32
+#define HWCAP_SPARC_BLKINIT	64
 
 /*
  * These are used to set parameters in the core dumps.
@@ -145,11 +147,21 @@ typedef struct {
    instruction set this cpu supports.  */
 
 /* On Ultra, we support all of the v8 capabilities. */
-#define ELF_HWCAP	((HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | \
-			  HWCAP_SPARC_SWAP | HWCAP_SPARC_MULDIV | \
-			  HWCAP_SPARC_V9) | \
-			 ((tlb_type == cheetah || tlb_type == cheetah_plus) ? \
-			  HWCAP_SPARC_ULTRA3 : 0))
+static inline unsigned int sparc64_elf_hwcap(void)
+{
+	unsigned int cap = (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR |
+			    HWCAP_SPARC_SWAP | HWCAP_SPARC_MULDIV |
+			    HWCAP_SPARC_V9);
+
+	if (tlb_type == cheetah || tlb_type == cheetah_plus)
+		cap |= HWCAP_SPARC_ULTRA3;
+	else if (tlb_type == hypervisor)
+		cap |= HWCAP_SPARC_BLKINIT;
+
+	return cap;
+}
+
+#define ELF_HWCAP	sparc64_elf_hwcap();
 
 /* This yields a string that ld.so will use to load implementation
    specific libraries for optimization.  This is more specific in
diff --git a/include/asm-sparc64/head.h b/include/asm-sparc64/head.h
index 0abd3a6..67960a7 100644
--- a/include/asm-sparc64/head.h
+++ b/include/asm-sparc64/head.h
@@ -4,12 +4,21 @@
 
 #include <asm/pstate.h>
 
+	/* wrpr	%g0, val, %gl */
+#define SET_GL(val)	\
+	.word	0xa1902000 | val
+
+	/* rdpr %gl, %gN */
+#define GET_GL_GLOBAL(N)	\
+	.word	0x81540000 | (N << 25)
+
 #define KERNBASE	0x400000
 
 #define	PTREGS_OFF	(STACK_BIAS + STACKFRAME_SZ)
 
 #define __CHEETAH_ID	0x003e0014
 #define __JALAPENO_ID	0x003e0016
+#define __SERRANO_ID	0x003e0022
 
 #define CHEETAH_MANUF		0x003e
 #define CHEETAH_IMPL		0x0014 /* Ultra-III   */
@@ -19,6 +28,12 @@
 #define PANTHER_IMPL		0x0019 /* Ultra-IV+   */
 #define SERRANO_IMPL		0x0022 /* Ultra-IIIi+ */
 
+#define BRANCH_IF_SUN4V(tmp1,label)		\
+	sethi	%hi(is_sun4v), %tmp1;		\
+	lduw	[%tmp1 + %lo(is_sun4v)], %tmp1; \
+	brnz,pn	%tmp1, label;			\
+	 nop
+
 #define BRANCH_IF_CHEETAH_BASE(tmp1,tmp2,label)	\
 	rdpr	%ver, %tmp1;			\
 	sethi	%hi(__CHEETAH_ID), %tmp2;	\
diff --git a/include/asm-sparc64/hypervisor.h b/include/asm-sparc64/hypervisor.h
new file mode 100644
index 0000000..612bf31
--- /dev/null
+++ b/include/asm-sparc64/hypervisor.h
@@ -0,0 +1,2128 @@
+#ifndef _SPARC64_HYPERVISOR_H
+#define _SPARC64_HYPERVISOR_H
+
+/* Sun4v hypervisor interfaces and defines.
+ *
+ * Hypervisor calls are made via traps to software traps number 0x80
+ * and above.  Registers %o0 to %o5 serve as argument, status, and
+ * return value registers.
+ *
+ * There are two kinds of these traps.  First there are the normal
+ * "fast traps" which use software trap 0x80 and encode the function
+ * to invoke by number in register %o5.  Argument and return value
+ * handling is as follows:
+ *
+ * -----------------------------------------------
+ * |  %o5  | function number |     undefined     |
+ * |  %o0  |   argument 0    |   return status   |
+ * |  %o1  |   argument 1    |   return value 1  |
+ * |  %o2  |   argument 2    |   return value 2  |
+ * |  %o3  |   argument 3    |   return value 3  |
+ * |  %o4  |   argument 4    |   return value 4  |
+ * -----------------------------------------------
+ *
+ * The second type are "hyper-fast traps" which encode the function
+ * number in the software trap number itself.  So these use trap
+ * numbers > 0x80.  The register usage for hyper-fast traps is as
+ * follows:
+ *
+ * -----------------------------------------------
+ * |  %o0  |   argument 0    |   return status   |
+ * |  %o1  |   argument 1    |   return value 1  |
+ * |  %o2  |   argument 2    |   return value 2  |
+ * |  %o3  |   argument 3    |   return value 3  |
+ * |  %o4  |   argument 4    |   return value 4  |
+ * -----------------------------------------------
+ *
+ * Registers providing explicit arguments to the hypervisor calls
+ * are volatile across the call.  Upon return their values are
+ * undefined unless explicitly specified as containing a particular
+ * return value by the specific call.  The return status is always
+ * returned in register %o0, zero indicates a successful execution of
+ * the hypervisor call and other values indicate an error status as
+ * defined below.  So, for example, if a hyper-fast trap takes
+ * arguments 0, 1, and 2, then %o0, %o1, and %o2 are volatile across
+ * the call and %o3, %o4, and %o5 would be preserved.
+ *
+ * If the hypervisor trap is invalid, or the fast trap function number
+ * is invalid, HV_EBADTRAP will be returned in %o0.  Also, all 64-bits
+ * of the argument and return values are significant.
+ */
+
+/* Trap numbers.  */
+#define HV_FAST_TRAP		0x80
+#define HV_MMU_MAP_ADDR_TRAP	0x83
+#define HV_MMU_UNMAP_ADDR_TRAP	0x84
+#define HV_TTRACE_ADDENTRY_TRAP	0x85
+#define HV_CORE_TRAP		0xff
+
+/* Error codes.  */
+#define HV_EOK				0  /* Successful return            */
+#define HV_ENOCPU			1  /* Invalid CPU id               */
+#define HV_ENORADDR			2  /* Invalid real address         */
+#define HV_ENOINTR			3  /* Invalid interrupt id         */
+#define HV_EBADPGSZ			4  /* Invalid pagesize encoding    */
+#define HV_EBADTSB			5  /* Invalid TSB description      */
+#define HV_EINVAL			6  /* Invalid argument             */
+#define HV_EBADTRAP			7  /* Invalid function number      */
+#define HV_EBADALIGN			8  /* Invalid address alignment    */
+#define HV_EWOULDBLOCK			9  /* Cannot complete w/o blocking */
+#define HV_ENOACCESS			10 /* No access to resource        */
+#define HV_EIO				11 /* I/O error                    */
+#define HV_ECPUERROR			12 /* CPU in error state           */
+#define HV_ENOTSUPPORTED		13 /* Function not supported       */
+#define HV_ENOMAP			14 /* No mapping found             */
+#define HV_ETOOMANY			15 /* Too many items specified     */
+
+/* mach_exit()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MACH_EXIT
+ * ARG0:	exit code
+ * ERRORS:	This service does not return.
+ *
+ * Stop all CPUs in the virtual domain and place them into the stopped
+ * state.  The 64-bit exit code may be passed to a service entity as
+ * the domain's exit status.  On systems without a service entity, the
+ * domain will undergo a reset, and the boot firmware will be
+ * reloaded.
+ *
+ * This function will never return to the guest that invokes it.
+ *
+ * Note: By convention an exit code of zero denotes a successful exit by
+ *       the guest code.  A non-zero exit code denotes a guest specific
+ *       error indication.
+ *
+ */
+#define HV_FAST_MACH_EXIT		0x00
+
+/* Domain services.  */
+
+/* mach_desc()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MACH_DESC
+ * ARG0:	buffer
+ * ARG1:	length
+ * RET0:	status
+ * RET1:	length
+ * ERRORS:	HV_EBADALIGN	Buffer is badly aligned
+ *		HV_ENORADDR	Buffer is to an illegal real address.
+ *		HV_EINVAL	Buffer length is too small for complete
+ *				machine description.
+ *
+ * Copy the most current machine description into the buffer indicated
+ * by the real address in ARG0.  The buffer provided must be 16 byte
+ * aligned.  Upon success or HV_EINVAL, this service returns the
+ * actual size of the machine description in the RET1 return value.
+ *
+ * Note: A method of determining the appropriate buffer size for the
+ *       machine description is to first call this service with a buffer
+ *       length of 0 bytes.
+ */
+#define HV_FAST_MACH_DESC		0x01
+
+/* mach_exit()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MACH_SIR
+ * ERRORS:	This service does not return.
+ *
+ * Perform a software initiated reset of the virtual machine domain.
+ * All CPUs are captured as soon as possible, all hardware devices are
+ * returned to the entry default state, and the domain is restarted at
+ * the SIR (trap type 0x04) real trap table (RTBA) entry point on one
+ * of the CPUs.  The single CPU restarted is selected as determined by
+ * platform specific policy.  Memory is preserved across this
+ * operation.
+ */
+#define HV_FAST_MACH_SIR		0x02
+
+/* mach_set_soft_state()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MACH_SET_SOFT_STATE
+ * ARG0:	software state
+ * ARG1:	software state description pointer
+ * RET0:	status
+ * ERRORS:	EINVAL		software state not valid or software state
+ *				description is not NULL terminated
+ *		ENORADDR	software state description pointer is not a
+ *				valid real address
+ *		EBADALIGNED	software state description is not correctly
+ *				aligned
+ *
+ * This allows the guest to report it's soft state to the hypervisor.  There
+ * are two primary components to this state.  The first part states whether
+ * the guest software is running or not.  The second containts optional
+ * details specific to the software.
+ *
+ * The software state argument is defined below in HV_SOFT_STATE_*, and
+ * indicates whether the guest is operating normally or in a transitional
+ * state.
+ *
+ * The software state description argument is a real address of a data buffer
+ * of size 32-bytes aligned on a 32-byte boundary.  It is treated as a NULL
+ * terminated 7-bit ASCII string of up to 31 characters not including the
+ * NULL termination.
+ */
+#define HV_FAST_MACH_SET_SOFT_STATE	0x03
+#define  HV_SOFT_STATE_NORMAL		 0x01
+#define  HV_SOFT_STATE_TRANSITION	 0x02
+
+/* mach_get_soft_state()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MACH_GET_SOFT_STATE
+ * ARG0:	software state description pointer
+ * RET0:	status
+ * RET1:	software state
+ * ERRORS:	ENORADDR	software state description pointer is not a
+ *				valid real address
+ *		EBADALIGNED	software state description is not correctly
+ *				aligned
+ *
+ * Retrieve the current value of the guest's software state.  The rules
+ * for the software state pointer are the same as for mach_set_soft_state()
+ * above.
+ */
+#define HV_FAST_MACH_GET_SOFT_STATE	0x04
+
+/* CPU services.
+ *
+ * CPUs represent devices that can execute software threads.  A single
+ * chip that contains multiple cores or strands is represented as
+ * multiple CPUs with unique CPU identifiers.  CPUs are exported to
+ * OBP via the machine description (and to the OS via the OBP device
+ * tree).  CPUs are always in one of three states: stopped, running,
+ * or error.
+ *
+ * A CPU ID is a pre-assigned 16-bit value that uniquely identifies a
+ * CPU within a logical domain.  Operations that are to be performed
+ * on multiple CPUs specify them via a CPU list.  A CPU list is an
+ * array in real memory, of which each 16-bit word is a CPU ID.  CPU
+ * lists are passed through the API as two arguments.  The first is
+ * the number of entries (16-bit words) in the CPU list, and the
+ * second is the (real address) pointer to the CPU ID list.
+ */
+
+/* cpu_start()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_START
+ * ARG0:	CPU ID
+ * ARG1:	PC
+ * ARG1:	RTBA
+ * ARG1:	target ARG0
+ * RET0:	status
+ * ERRORS:	ENOCPU		Invalid CPU ID
+ *		EINVAL		Target CPU ID is not in the stopped state
+ *		ENORADDR	Invalid PC or RTBA real address
+ *		EBADALIGN	Unaligned PC or unaligned RTBA
+ *		EWOULDBLOCK	Starting resources are not available
+ *
+ * Start CPU with given CPU ID with PC in %pc and with a real trap
+ * base address value of RTBA.  The indicated CPU must be in the
+ * stopped state.  The supplied RTBA must be aligned on a 256 byte
+ * boundary.  On successful completion, the specified CPU will be in
+ * the running state and will be supplied with "target ARG0" in %o0
+ * and RTBA in %tba.
+ */
+#define HV_FAST_CPU_START		0x10
+
+/* cpu_stop()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_STOP
+ * ARG0:	CPU ID
+ * RET0:	status
+ * ERRORS:	ENOCPU		Invalid CPU ID
+ *		EINVAL		Target CPU ID is the current cpu
+ *		EINVAL		Target CPU ID is not in the running state
+ *		EWOULDBLOCK	Stopping resources are not available
+ *		ENOTSUPPORTED	Not supported on this platform
+ *
+ * The specified CPU is stopped.  The indicated CPU must be in the
+ * running state.  On completion, it will be in the stopped state.  It
+ * is not legal to stop the current CPU.
+ *
+ * Note: As this service cannot be used to stop the current cpu, this service
+ *       may not be used to stop the last running CPU in a domain.  To stop
+ *       and exit a running domain, a guest must use the mach_exit() service.
+ */
+#define HV_FAST_CPU_STOP		0x11
+
+/* cpu_yield()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_YIELD
+ * RET0:	status
+ * ERRORS:	No possible error.
+ *
+ * Suspend execution on the current CPU.  Execution will resume when
+ * an interrupt (device, %stick_compare, or cross-call) is targeted to
+ * the CPU.  On some CPUs, this API may be used by the hypervisor to
+ * save power by disabling hardware strands.
+ */
+#define HV_FAST_CPU_YIELD		0x12
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_cpu_yield(void);
+#endif
+
+/* cpu_qconf()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_QCONF
+ * ARG0:	queue
+ * ARG1:	base real address
+ * ARG2:	number of entries
+ * RET0:	status
+ * ERRORS:	ENORADDR	Invalid base real address
+ *		EINVAL		Invalid queue or number of entries is less
+ *				than 2 or too large.
+ *		EBADALIGN	Base real address is not correctly aligned
+ *				for size.
+ *
+ * Configure the given queue to be placed at the given base real
+ * address, with the given number of entries.  The number of entries
+ * must be a power of 2.  The base real address must be aligned
+ * exactly to match the queue size.  Each queue entry is 64 bytes
+ * long, so for example a 32 entry queue must be aligned on a 2048
+ * byte real address boundary.
+ *
+ * The specified queue is unconfigured if the number of entries is given
+ * as zero.
+ *
+ * For the current version of this API service, the argument queue is defined
+ * as follows:
+ *
+ *	queue		description
+ *	-----		-------------------------
+ *	0x3c		cpu mondo queue
+ *	0x3d		device mondo queue
+ *	0x3e		resumable error queue
+ *	0x3f		non-resumable error queue
+ *
+ * Note: The maximum number of entries for each queue for a specific cpu may
+ *       be determined from the machine description.
+ */
+#define HV_FAST_CPU_QCONF		0x14
+#define  HV_CPU_QUEUE_CPU_MONDO		 0x3c
+#define  HV_CPU_QUEUE_DEVICE_MONDO	 0x3d
+#define  HV_CPU_QUEUE_RES_ERROR		 0x3e
+#define  HV_CPU_QUEUE_NONRES_ERROR	 0x3f
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_cpu_qconf(unsigned long type,
+				     unsigned long queue_paddr,
+				     unsigned long num_queue_entries);
+#endif
+
+/* cpu_qinfo()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_QINFO
+ * ARG0:	queue
+ * RET0:	status
+ * RET1:	base real address
+ * RET1:	number of entries
+ * ERRORS:	EINVAL		Invalid queue
+ *
+ * Return the configuration info for the given queue.  The base real
+ * address and number of entries of the defined queue are returned.
+ * The queue argument values are the same as for cpu_qconf() above.
+ *
+ * If the specified queue is a valid queue number, but no queue has
+ * been defined, the number of entries will be set to zero and the
+ * base real address returned is undefined.
+ */
+#define HV_FAST_CPU_QINFO		0x15
+
+/* cpu_mondo_send()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_MONDO_SEND
+ * ARG0-1:	CPU list
+ * ARG2:	data real address
+ * RET0:	status
+ * ERRORS:	EBADALIGN	Mondo data is not 64-byte aligned or CPU list
+ *				is not 2-byte aligned.
+ *		ENORADDR	Invalid data mondo address, or invalid cpu list
+ *				address.
+ *		ENOCPU		Invalid cpu in CPU list
+ *		EWOULDBLOCK	Some or all of the listed CPUs did not receive
+ *				the mondo
+ *		ECPUERROR	One or more of the listed CPUs are in error
+ *				state, use HV_FAST_CPU_STATE to see which ones
+ *		EINVAL		CPU list includes caller's CPU ID
+ *
+ * Send a mondo interrupt to the CPUs in the given CPU list with the
+ * 64-bytes at the given data real address.  The data must be 64-byte
+ * aligned.  The mondo data will be delivered to the cpu_mondo queues
+ * of the recipient CPUs.
+ *
+ * In all cases, error or not, the CPUs in the CPU list to which the
+ * mondo has been successfully delivered will be indicated by having
+ * their entry in CPU list updated with the value 0xffff.
+ */
+#define HV_FAST_CPU_MONDO_SEND		0x42
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_cpu_mondo_send(unsigned long cpu_count, unsigned long cpu_list_pa, unsigned long mondo_block_pa);
+#endif
+
+/* cpu_myid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_MYID
+ * RET0:	status
+ * RET1:	CPU ID
+ * ERRORS:	No errors defined.
+ *
+ * Return the hypervisor ID handle for the current CPU.  Use by a
+ * virtual CPU to discover it's own identity.
+ */
+#define HV_FAST_CPU_MYID		0x16
+
+/* cpu_state()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_STATE
+ * ARG0:	CPU ID
+ * RET0:	status
+ * RET1:	state
+ * ERRORS:	ENOCPU		Invalid CPU ID
+ *
+ * Retrieve the current state of the CPU with the given CPU ID.
+ */
+#define HV_FAST_CPU_STATE		0x17
+#define  HV_CPU_STATE_STOPPED		 0x01
+#define  HV_CPU_STATE_RUNNING		 0x02
+#define  HV_CPU_STATE_ERROR		 0x03
+
+#ifndef __ASSEMBLY__
+extern long sun4v_cpu_state(unsigned long cpuid);
+#endif
+
+/* cpu_set_rtba()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_SET_RTBA
+ * ARG0:	RTBA
+ * RET0:	status
+ * RET1:	previous RTBA
+ * ERRORS:	ENORADDR	Invalid RTBA real address
+ *		EBADALIGN	RTBA is incorrectly aligned for a trap table
+ *
+ * Set the real trap base address of the local cpu to the given RTBA.
+ * The supplied RTBA must be aligned on a 256 byte boundary.  Upon
+ * success the previous value of the RTBA is returned in RET1.
+ *
+ * Note: This service does not affect %tba
+ */
+#define HV_FAST_CPU_SET_RTBA		0x18
+
+/* cpu_set_rtba()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_GET_RTBA
+ * RET0:	status
+ * RET1:	previous RTBA
+ * ERRORS:	No possible error.
+ *
+ * Returns the current value of RTBA in RET1.
+ */
+#define HV_FAST_CPU_GET_RTBA		0x19
+
+/* MMU services.
+ *
+ * Layout of a TSB description for mmu_tsb_ctx{,non}0() calls.
+ */
+#ifndef __ASSEMBLY__
+struct hv_tsb_descr {
+	unsigned short		pgsz_idx;
+	unsigned short		assoc;
+	unsigned int		num_ttes;	/* in TTEs */
+	unsigned int		ctx_idx;
+	unsigned int		pgsz_mask;
+	unsigned long		tsb_base;
+	unsigned long		resv;
+};
+#endif
+#define HV_TSB_DESCR_PGSZ_IDX_OFFSET	0x00
+#define HV_TSB_DESCR_ASSOC_OFFSET	0x02
+#define HV_TSB_DESCR_NUM_TTES_OFFSET	0x04
+#define HV_TSB_DESCR_CTX_IDX_OFFSET	0x08
+#define HV_TSB_DESCR_PGSZ_MASK_OFFSET	0x0c
+#define HV_TSB_DESCR_TSB_BASE_OFFSET	0x10
+#define HV_TSB_DESCR_RESV_OFFSET	0x18
+
+/* Page size bitmask.  */
+#define HV_PGSZ_MASK_8K			(1 << 0)
+#define HV_PGSZ_MASK_64K		(1 << 1)
+#define HV_PGSZ_MASK_512K		(1 << 2)
+#define HV_PGSZ_MASK_4MB		(1 << 3)
+#define HV_PGSZ_MASK_32MB		(1 << 4)
+#define HV_PGSZ_MASK_256MB		(1 << 5)
+#define HV_PGSZ_MASK_2GB		(1 << 6)
+#define HV_PGSZ_MASK_16GB		(1 << 7)
+
+/* Page size index.  The value given in the TSB descriptor must correspond
+ * to the smallest page size specified in the pgsz_mask page size bitmask.
+ */
+#define HV_PGSZ_IDX_8K			0
+#define HV_PGSZ_IDX_64K			1
+#define HV_PGSZ_IDX_512K		2
+#define HV_PGSZ_IDX_4MB			3
+#define HV_PGSZ_IDX_32MB		4
+#define HV_PGSZ_IDX_256MB		5
+#define HV_PGSZ_IDX_2GB			6
+#define HV_PGSZ_IDX_16GB		7
+
+/* MMU fault status area.
+ *
+ * MMU related faults have their status and fault address information
+ * placed into a memory region made available by privileged code.  Each
+ * virtual processor must make a mmu_fault_area_conf() call to tell the
+ * hypervisor where that processor's fault status should be stored.
+ *
+ * The fault status block is a multiple of 64-bytes and must be aligned
+ * on a 64-byte boundary.
+ */
+#ifndef __ASSEMBLY__
+struct hv_fault_status {
+	unsigned long		i_fault_type;
+	unsigned long		i_fault_addr;
+	unsigned long		i_fault_ctx;
+	unsigned long		i_reserved[5];
+	unsigned long		d_fault_type;
+	unsigned long		d_fault_addr;
+	unsigned long		d_fault_ctx;
+	unsigned long		d_reserved[5];
+};
+#endif
+#define HV_FAULT_I_TYPE_OFFSET	0x00
+#define HV_FAULT_I_ADDR_OFFSET	0x08
+#define HV_FAULT_I_CTX_OFFSET	0x10
+#define HV_FAULT_D_TYPE_OFFSET	0x40
+#define HV_FAULT_D_ADDR_OFFSET	0x48
+#define HV_FAULT_D_CTX_OFFSET	0x50
+
+#define HV_FAULT_TYPE_FAST_MISS	1
+#define HV_FAULT_TYPE_FAST_PROT	2
+#define HV_FAULT_TYPE_MMU_MISS	3
+#define HV_FAULT_TYPE_INV_RA	4
+#define HV_FAULT_TYPE_PRIV_VIOL	5
+#define HV_FAULT_TYPE_PROT_VIOL	6
+#define HV_FAULT_TYPE_NFO	7
+#define HV_FAULT_TYPE_NFO_SEFF	8
+#define HV_FAULT_TYPE_INV_VA	9
+#define HV_FAULT_TYPE_INV_ASI	10
+#define HV_FAULT_TYPE_NC_ATOMIC	11
+#define HV_FAULT_TYPE_PRIV_ACT	12
+#define HV_FAULT_TYPE_RESV1	13
+#define HV_FAULT_TYPE_UNALIGNED	14
+#define HV_FAULT_TYPE_INV_PGSZ	15
+/* Values 16 --> -2 are reserved.  */
+#define HV_FAULT_TYPE_MULTIPLE	-1
+
+/* Flags argument for mmu_{map,unmap}_addr(), mmu_demap_{page,context,all}(),
+ * and mmu_{map,unmap}_perm_addr().
+ */
+#define HV_MMU_DMMU			0x01
+#define HV_MMU_IMMU			0x02
+#define HV_MMU_ALL			(HV_MMU_DMMU | HV_MMU_IMMU)
+
+/* mmu_map_addr()
+ * TRAP:	HV_MMU_MAP_ADDR_TRAP
+ * ARG0:	virtual address
+ * ARG1:	mmu context
+ * ARG2:	TTE
+ * ARG3:	flags (HV_MMU_{IMMU,DMMU})
+ * ERRORS:	EINVAL		Invalid virtual address, mmu context, or flags
+ *		EBADPGSZ	Invalid page size value
+ *		ENORADDR	Invalid real address in TTE
+ *
+ * Create a non-permanent mapping using the given TTE, virtual
+ * address, and mmu context.  The flags argument determines which
+ * (data, or instruction, or both) TLB the mapping gets loaded into.
+ *
+ * The behavior is undefined if the valid bit is clear in the TTE.
+ *
+ * Note: This API call is for privileged code to specify temporary translation
+ *       mappings without the need to create and manage a TSB.
+ */
+
+/* mmu_unmap_addr()
+ * TRAP:	HV_MMU_UNMAP_ADDR_TRAP
+ * ARG0:	virtual address
+ * ARG1:	mmu context
+ * ARG2:	flags (HV_MMU_{IMMU,DMMU})
+ * ERRORS:	EINVAL		Invalid virtual address, mmu context, or flags
+ *
+ * Demaps the given virtual address in the given mmu context on this
+ * CPU.  This function is intended to be used to demap pages mapped
+ * with mmu_map_addr.  This service is equivalent to invoking
+ * mmu_demap_page() with only the current CPU in the CPU list. The
+ * flags argument determines which (data, or instruction, or both) TLB
+ * the mapping gets unmapped from.
+ *
+ * Attempting to perform an unmap operation for a previously defined
+ * permanent mapping will have undefined results.
+ */
+
+/* mmu_tsb_ctx0()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_TSB_CTX0
+ * ARG0:	number of TSB descriptions
+ * ARG1:	TSB descriptions pointer
+ * RET0:	status
+ * ERRORS:	ENORADDR		Invalid TSB descriptions pointer or
+ *					TSB base within a descriptor
+ *		EBADALIGN		TSB descriptions pointer is not aligned
+ *					to an 8-byte boundary, or TSB base
+ *					within a descriptor is not aligned for
+ *					the given TSB size
+ *		EBADPGSZ		Invalid page size in a TSB descriptor
+ *		EBADTSB			Invalid associativity or size in a TSB
+ *					descriptor
+ *		EINVAL			Invalid number of TSB descriptions, or
+ *					invalid context index in a TSB
+ *					descriptor, or index page size not
+ *					equal to smallest page size in page
+ *					size bitmask field.
+ *
+ * Configures the TSBs for the current CPU for virtual addresses with
+ * context zero.  The TSB descriptions pointer is a pointer to an
+ * array of the given number of TSB descriptions.
+ *
+ * Note: The maximum number of TSBs available to a virtual CPU is given by the
+ *       mmu-max-#tsbs property of the cpu's corresponding "cpu" node in the
+ *       machine description.
+ */
+#define HV_FAST_MMU_TSB_CTX0		0x20
+
+/* mmu_tsb_ctxnon0()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_TSB_CTXNON0
+ * ARG0:	number of TSB descriptions
+ * ARG1:	TSB descriptions pointer
+ * RET0:	status
+ * ERRORS:	Same as for mmu_tsb_ctx0() above.
+ *
+ * Configures the TSBs for the current CPU for virtual addresses with
+ * non-zero contexts.  The TSB descriptions pointer is a pointer to an
+ * array of the given number of TSB descriptions.
+ *
+ * Note: A maximum of 16 TSBs may be specified in the TSB description list.
+ */
+#define HV_FAST_MMU_TSB_CTXNON0		0x21
+
+/* mmu_demap_page()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_DEMAP_PAGE
+ * ARG0:	reserved, must be zero
+ * ARG1:	reserved, must be zero
+ * ARG2:	virtual address
+ * ARG3:	mmu context
+ * ARG4:	flags (HV_MMU_{IMMU,DMMU})
+ * RET0:	status
+ * ERRORS:	EINVAL			Invalid virutal address, context, or
+ *					flags value
+ *		ENOTSUPPORTED		ARG0 or ARG1 is non-zero
+ *
+ * Demaps any page mapping of the given virtual address in the given
+ * mmu context for the current virtual CPU.  Any virtually tagged
+ * caches are guaranteed to be kept consistent.  The flags argument
+ * determines which TLB (instruction, or data, or both) participate in
+ * the operation.
+ *
+ * ARG0 and ARG1 are both reserved and must be set to zero.
+ */
+#define HV_FAST_MMU_DEMAP_PAGE		0x22
+
+/* mmu_demap_ctx()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_DEMAP_CTX
+ * ARG0:	reserved, must be zero
+ * ARG1:	reserved, must be zero
+ * ARG2:	mmu context
+ * ARG3:	flags (HV_MMU_{IMMU,DMMU})
+ * RET0:	status
+ * ERRORS:	EINVAL			Invalid context or flags value
+ *		ENOTSUPPORTED		ARG0 or ARG1 is non-zero
+ *
+ * Demaps all non-permanent virtual page mappings previously specified
+ * for the given context for the current virtual CPU.  Any virtual
+ * tagged caches are guaranteed to be kept consistent.  The flags
+ * argument determines which TLB (instruction, or data, or both)
+ * participate in the operation.
+ *
+ * ARG0 and ARG1 are both reserved and must be set to zero.
+ */
+#define HV_FAST_MMU_DEMAP_CTX		0x23
+
+/* mmu_demap_all()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_DEMAP_ALL
+ * ARG0:	reserved, must be zero
+ * ARG1:	reserved, must be zero
+ * ARG2:	flags (HV_MMU_{IMMU,DMMU})
+ * RET0:	status
+ * ERRORS:	EINVAL			Invalid flags value
+ *		ENOTSUPPORTED		ARG0 or ARG1 is non-zero
+ *
+ * Demaps all non-permanent virtual page mappings previously specified
+ * for the current virtual CPU.  Any virtual tagged caches are
+ * guaranteed to be kept consistent.  The flags argument determines
+ * which TLB (instruction, or data, or both) participate in the
+ * operation.
+ *
+ * ARG0 and ARG1 are both reserved and must be set to zero.
+ */
+#define HV_FAST_MMU_DEMAP_ALL		0x24
+
+/* mmu_map_perm_addr()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_MAP_PERM_ADDR
+ * ARG0:	virtual address
+ * ARG1:	reserved, must be zero
+ * ARG2:	TTE
+ * ARG3:	flags (HV_MMU_{IMMU,DMMU})
+ * RET0:	status
+ * ERRORS:	EINVAL			Invalid virutal address or flags value
+ *		EBADPGSZ		Invalid page size value
+ *		ENORADDR		Invalid real address in TTE
+ *		ETOOMANY		Too many mappings (max of 8 reached)
+ *
+ * Create a permanent mapping using the given TTE and virtual address
+ * for context 0 on the calling virtual CPU.  A maximum of 8 such
+ * permanent mappings may be specified by privileged code.  Mappings
+ * may be removed with mmu_unmap_perm_addr().
+ *
+ * The behavior is undefined if a TTE with the valid bit clear is given.
+ *
+ * Note: This call is used to specify address space mappings for which
+ *       privileged code does not expect to receive misses.  For example,
+ *       this mechanism can be used to map kernel nucleus code and data.
+ */
+#define HV_FAST_MMU_MAP_PERM_ADDR	0x25
+
+/* mmu_fault_area_conf()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_FAULT_AREA_CONF
+ * ARG0:	real address
+ * RET0:	status
+ * RET1:	previous mmu fault area real address
+ * ERRORS:	ENORADDR		Invalid real address
+ *		EBADALIGN		Invalid alignment for fault area
+ *
+ * Configure the MMU fault status area for the calling CPU.  A 64-byte
+ * aligned real address specifies where MMU fault status information
+ * is placed.  The return value is the previously specified area, or 0
+ * for the first invocation.  Specifying a fault area at real address
+ * 0 is not allowed.
+ */
+#define HV_FAST_MMU_FAULT_AREA_CONF	0x26
+
+/* mmu_enable()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_ENABLE
+ * ARG0:	enable flag
+ * ARG1:	return target address
+ * RET0:	status
+ * ERRORS:	ENORADDR		Invalid real address when disabling
+ *					translation.
+ *		EBADALIGN		The return target address is not
+ *					aligned to an instruction.
+ *		EINVAL			The enable flag request the current
+ *					operating mode (e.g. disable if already
+ *					disabled)
+ *
+ * Enable or disable virtual address translation for the calling CPU
+ * within the virtual machine domain.  If the enable flag is zero,
+ * translation is disabled, any non-zero value will enable
+ * translation.
+ *
+ * When this function returns, the newly selected translation mode
+ * will be active.  If the mmu is being enabled, then the return
+ * target address is a virtual address else it is a real address.
+ *
+ * Upon successful completion, control will be returned to the given
+ * return target address (ie. the cpu will jump to that address).  On
+ * failure, the previous mmu mode remains and the trap simply returns
+ * as normal with the appropriate error code in RET0.
+ */
+#define HV_FAST_MMU_ENABLE		0x27
+
+/* mmu_unmap_perm_addr()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_UNMAP_PERM_ADDR
+ * ARG0:	virtual address
+ * ARG1:	reserved, must be zero
+ * ARG2:	flags (HV_MMU_{IMMU,DMMU})
+ * RET0:	status
+ * ERRORS:	EINVAL			Invalid virutal address or flags value
+ *		ENOMAP			Specified mapping was not found
+ *
+ * Demaps any permanent page mapping (established via
+ * mmu_map_perm_addr()) at the given virtual address for context 0 on
+ * the current virtual CPU.  Any virtual tagged caches are guaranteed
+ * to be kept consistent.
+ */
+#define HV_FAST_MMU_UNMAP_PERM_ADDR	0x28
+
+/* mmu_tsb_ctx0_info()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_TSB_CTX0_INFO
+ * ARG0:	max TSBs
+ * ARG1:	buffer pointer
+ * RET0:	status
+ * RET1:	number of TSBs
+ * ERRORS:	EINVAL			Supplied buffer is too small
+ *		EBADALIGN		The buffer pointer is badly aligned
+ *		ENORADDR		Invalid real address for buffer pointer
+ *
+ * Return the TSB configuration as previous defined by mmu_tsb_ctx0()
+ * into the provided buffer.  The size of the buffer is given in ARG1
+ * in terms of the number of TSB description entries.
+ *
+ * Upon return, RET1 always contains the number of TSB descriptions
+ * previously configured.  If zero TSBs were configured, EOK is
+ * returned with RET1 containing 0.
+ */
+#define HV_FAST_MMU_TSB_CTX0_INFO	0x29
+
+/* mmu_tsb_ctxnon0_info()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_TSB_CTXNON0_INFO
+ * ARG0:	max TSBs
+ * ARG1:	buffer pointer
+ * RET0:	status
+ * RET1:	number of TSBs
+ * ERRORS:	EINVAL			Supplied buffer is too small
+ *		EBADALIGN		The buffer pointer is badly aligned
+ *		ENORADDR		Invalid real address for buffer pointer
+ *
+ * Return the TSB configuration as previous defined by
+ * mmu_tsb_ctxnon0() into the provided buffer.  The size of the buffer
+ * is given in ARG1 in terms of the number of TSB description entries.
+ *
+ * Upon return, RET1 always contains the number of TSB descriptions
+ * previously configured.  If zero TSBs were configured, EOK is
+ * returned with RET1 containing 0.
+ */
+#define HV_FAST_MMU_TSB_CTXNON0_INFO	0x2a
+
+/* mmu_fault_area_info()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMU_FAULT_AREA_INFO
+ * RET0:	status
+ * RET1:	fault area real address
+ * ERRORS:	No errors defined.
+ *
+ * Return the currently defined MMU fault status area for the current
+ * CPU.  The real address of the fault status area is returned in
+ * RET1, or 0 is returned in RET1 if no fault status area is defined.
+ *
+ * Note: mmu_fault_area_conf() may be called with the return value (RET1)
+ *       from this service if there is a need to save and restore the fault
+ *	 area for a cpu.
+ */
+#define HV_FAST_MMU_FAULT_AREA_INFO	0x2b
+
+/* Cache and Memory services. */
+
+/* mem_scrub()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MEM_SCRUB
+ * ARG0:	real address
+ * ARG1:	length
+ * RET0:	status
+ * RET1:	length scrubbed
+ * ERRORS:	ENORADDR	Invalid real address
+ *		EBADALIGN	Start address or length are not correctly
+ *				aligned
+ *		EINVAL		Length is zero
+ *
+ * Zero the memory contents in the range real address to real address
+ * plus length minus 1.  Also, valid ECC will be generated for that
+ * memory address range.  Scrubbing is started at the given real
+ * address, but may not scrub the entire given length.  The actual
+ * length scrubbed will be returned in RET1.
+ *
+ * The real address and length must be aligned on an 8K boundary, or
+ * contain the start address and length from a sun4v error report.
+ *
+ * Note: There are two uses for this function.  The first use is to block clear
+ *       and initialize memory and the second is to scrub an u ncorrectable
+ *       error reported via a resumable or non-resumable trap.  The second
+ *       use requires the arguments to be equal to the real address and length
+ *       provided in a sun4v memory error report.
+ */
+#define HV_FAST_MEM_SCRUB		0x31
+
+/* mem_sync()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MEM_SYNC
+ * ARG0:	real address
+ * ARG1:	length
+ * RET0:	status
+ * RET1:	length synced
+ * ERRORS:	ENORADDR	Invalid real address
+ *		EBADALIGN	Start address or length are not correctly
+ *				aligned
+ *		EINVAL		Length is zero
+ *
+ * Force the next access within the real address to real address plus
+ * length minus 1 to be fetches from main system memory.  Less than
+ * the given length may be synced, the actual amount synced is
+ * returned in RET1.  The real address and length must be aligned on
+ * an 8K boundary.
+ */
+#define HV_FAST_MEM_SYNC		0x32
+
+/* Time of day services.
+ *
+ * The hypervisor maintains the time of day on a per-domain basis.
+ * Changing the time of day in one domain does not affect the time of
+ * day on any other domain.
+ *
+ * Time is described by a single unsigned 64-bit word which is the
+ * number of seconds since the UNIX Epoch (00:00:00 UTC, January 1,
+ * 1970).
+ */
+
+/* tod_get()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_TOD_GET
+ * RET0:	status
+ * RET1:	TOD
+ * ERRORS:	EWOULDBLOCK	TOD resource is temporarily unavailable
+ *		ENOTSUPPORTED	If TOD not supported on this platform
+ *
+ * Return the current time of day.  May block if TOD access is
+ * temporarily not possible.
+ */
+#define HV_FAST_TOD_GET			0x50
+
+/* tod_set()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_TOD_SET
+ * ARG0:	TOD
+ * RET0:	status
+ * ERRORS:	EWOULDBLOCK	TOD resource is temporarily unavailable
+ *		ENOTSUPPORTED	If TOD not supported on this platform
+ *
+ * The current time of day is set to the value specified in ARG0.  May
+ * block if TOD access is temporarily not possible.
+ */
+#define HV_FAST_TOD_SET			0x51
+
+/* Console services */
+
+/* con_getchar()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CONS_GETCHAR
+ * RET0:	status
+ * RET1:	character
+ * ERRORS:	EWOULDBLOCK	No character available.
+ *
+ * Returns a character from the console device.  If no character is
+ * available then an EWOULDBLOCK error is returned.  If a character is
+ * available, then the returned status is EOK and the character value
+ * is in RET1.
+ *
+ * A virtual BREAK is represented by the 64-bit value -1.
+ *
+ * A virtual HUP signal is represented by the 64-bit value -2.
+ */
+#define HV_FAST_CONS_GETCHAR		0x60
+
+/* con_putchar()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CONS_PUTCHAR
+ * ARG0:	character
+ * RET0:	status
+ * ERRORS:	EINVAL		Illegal character
+ *		EWOULDBLOCK	Output buffer currently full, would block
+ *
+ * Send a character to the console device.  Only character values
+ * between 0 and 255 may be used.  Values outside this range are
+ * invalid except for the 64-bit value -1 which is used to send a
+ * virtual BREAK.
+ */
+#define HV_FAST_CONS_PUTCHAR		0x61
+
+/* Trap trace services.
+ *
+ * The hypervisor provides a trap tracing capability for privileged
+ * code running on each virtual CPU.  Privileged code provides a
+ * round-robin trap trace queue within which the hypervisor writes
+ * 64-byte entries detailing hyperprivileged traps taken n behalf of
+ * privileged code.  This is provided as a debugging capability for
+ * privileged code.
+ *
+ * The trap trace control structure is 64-bytes long and placed at the
+ * start (offset 0) of the trap trace buffer, and is described as
+ * follows:
+ */
+#ifndef __ASSEMBLY__
+struct hv_trap_trace_control {
+	unsigned long		head_offset;
+	unsigned long		tail_offset;
+	unsigned long		__reserved[0x30 / sizeof(unsigned long)];
+};
+#endif
+#define HV_TRAP_TRACE_CTRL_HEAD_OFFSET	0x00
+#define HV_TRAP_TRACE_CTRL_TAIL_OFFSET	0x08
+
+/* The head offset is the offset of the most recently completed entry
+ * in the trap-trace buffer.  The tail offset is the offset of the
+ * next entry to be written.  The control structure is owned and
+ * modified by the hypervisor.  A guest may not modify the control
+ * structure contents.  Attempts to do so will result in undefined
+ * behavior for the guest.
+ *
+ * Each trap trace buffer entry is layed out as follows:
+ */
+#ifndef __ASSEMBLY__
+struct hv_trap_trace_entry {
+	unsigned char	type;		/* Hypervisor or guest entry?	*/
+	unsigned char	hpstate;	/* Hyper-privileged state	*/
+	unsigned char	tl;		/* Trap level			*/
+	unsigned char	gl;		/* Global register level	*/
+	unsigned short	tt;		/* Trap type			*/
+	unsigned short	tag;		/* Extended trap identifier	*/
+	unsigned long	tstate;		/* Trap state			*/
+	unsigned long	tick;		/* Tick				*/
+	unsigned long	tpc;		/* Trap PC			*/
+	unsigned long	f1;		/* Entry specific		*/
+	unsigned long	f2;		/* Entry specific		*/
+	unsigned long	f3;		/* Entry specific		*/
+	unsigned long	f4;		/* Entry specific		*/
+};
+#endif
+#define HV_TRAP_TRACE_ENTRY_TYPE	0x00
+#define HV_TRAP_TRACE_ENTRY_HPSTATE	0x01
+#define HV_TRAP_TRACE_ENTRY_TL		0x02
+#define HV_TRAP_TRACE_ENTRY_GL		0x03
+#define HV_TRAP_TRACE_ENTRY_TT		0x04
+#define HV_TRAP_TRACE_ENTRY_TAG		0x06
+#define HV_TRAP_TRACE_ENTRY_TSTATE	0x08
+#define HV_TRAP_TRACE_ENTRY_TICK	0x10
+#define HV_TRAP_TRACE_ENTRY_TPC		0x18
+#define HV_TRAP_TRACE_ENTRY_F1		0x20
+#define HV_TRAP_TRACE_ENTRY_F2		0x28
+#define HV_TRAP_TRACE_ENTRY_F3		0x30
+#define HV_TRAP_TRACE_ENTRY_F4		0x38
+
+/* The type field is encoded as follows.  */
+#define HV_TRAP_TYPE_UNDEF		0x00 /* Entry content undefined     */
+#define HV_TRAP_TYPE_HV			0x01 /* Hypervisor trap entry       */
+#define HV_TRAP_TYPE_GUEST		0xff /* Added via ttrace_addentry() */
+
+/* ttrace_buf_conf()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_TTRACE_BUF_CONF
+ * ARG0:	real address
+ * ARG1:	number of entries
+ * RET0:	status
+ * RET1:	number of entries
+ * ERRORS:	ENORADDR	Invalid real address
+ *		EINVAL		Size is too small
+ *		EBADALIGN	Real address not aligned on 64-byte boundary
+ *
+ * Requests hypervisor trap tracing and declares a virtual CPU's trap
+ * trace buffer to the hypervisor.  The real address supplies the real
+ * base address of the trap trace queue and must be 64-byte aligned.
+ * Specifying a value of 0 for the number of entries disables trap
+ * tracing for the calling virtual CPU.  The buffer allocated must be
+ * sized for a power of two number of 64-byte trap trace entries plus
+ * an initial 64-byte control structure.
+ * 
+ * This may be invoked any number of times so that a virtual CPU may
+ * relocate a trap trace buffer or create "snapshots" of information.
+ *
+ * If the real address is illegal or badly aligned, then trap tracing
+ * is disabled and an error is returned.
+ *
+ * Upon failure with EINVAL, this service call returns in RET1 the
+ * minimum number of buffer entries required.  Upon other failures
+ * RET1 is undefined.
+ */
+#define HV_FAST_TTRACE_BUF_CONF		0x90
+
+/* ttrace_buf_info()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_TTRACE_BUF_INFO
+ * RET0:	status
+ * RET1:	real address
+ * RET2:	size
+ * ERRORS:	None defined.
+ *
+ * Returns the size and location of the previously declared trap-trace
+ * buffer.  In the event that no buffer was previously defined, or the
+ * buffer is disabled, this call will return a size of zero bytes.
+ */
+#define HV_FAST_TTRACE_BUF_INFO		0x91
+
+/* ttrace_enable()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_TTRACE_ENABLE
+ * ARG0:	enable
+ * RET0:	status
+ * RET1:	previous enable state
+ * ERRORS:	EINVAL		No trap trace buffer currently defined
+ *
+ * Enable or disable trap tracing, and return the previous enabled
+ * state in RET1.  Future systems may define various flags for the
+ * enable argument (ARG0), for the moment a guest should pass
+ * "(uint64_t) -1" to enable, and "(uint64_t) 0" to disable all
+ * tracing - which will ensure future compatability.
+ */
+#define HV_FAST_TTRACE_ENABLE		0x92
+
+/* ttrace_freeze()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_TTRACE_FREEZE
+ * ARG0:	freeze
+ * RET0:	status
+ * RET1:	previous freeze state
+ * ERRORS:	EINVAL		No trap trace buffer currently defined
+ *
+ * Freeze or unfreeze trap tracing, returning the previous freeze
+ * state in RET1.  A guest should pass a non-zero value to freeze and
+ * a zero value to unfreeze all tracing.  The returned previous state
+ * is 0 for not frozen and 1 for frozen.
+ */
+#define HV_FAST_TTRACE_FREEZE		0x93
+
+/* ttrace_addentry()
+ * TRAP:	HV_TTRACE_ADDENTRY_TRAP
+ * ARG0:	tag (16-bits)
+ * ARG1:	data word 0
+ * ARG2:	data word 1
+ * ARG3:	data word 2
+ * ARG4:	data word 3
+ * RET0:	status
+ * ERRORS:	EINVAL		No trap trace buffer currently defined
+ *
+ * Add an entry to the trap trace buffer.  Upon return only ARG0/RET0
+ * is modified - none of the other registers holding arguments are
+ * volatile across this hypervisor service.
+ */
+
+/* Core dump services.
+ *
+ * Since the hypervisor viraulizes and thus obscures a lot of the
+ * physical machine layout and state, traditional OS crash dumps can
+ * be difficult to diagnose especially when the problem is a
+ * configuration error of some sort.
+ *
+ * The dump services provide an opaque buffer into which the
+ * hypervisor can place it's internal state in order to assist in
+ * debugging such situations.  The contents are opaque and extremely
+ * platform and hypervisor implementation specific.  The guest, during
+ * a core dump, requests that the hypervisor update any information in
+ * the dump buffer in preparation to being dumped as part of the
+ * domain's memory image.
+ */
+
+/* dump_buf_update()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_DUMP_BUF_UPDATE
+ * ARG0:	real address
+ * ARG1:	size
+ * RET0:	status
+ * RET1:	required size of dump buffer
+ * ERRORS:	ENORADDR	Invalid real address
+ *		EBADALIGN	Real address is not aligned on a 64-byte
+ *				boundary
+ *		EINVAL		Size is non-zero but less than minimum size
+ *				required
+ *		ENOTSUPPORTED	Operation not supported on current logical
+ *				domain
+ *
+ * Declare a domain dump buffer to the hypervisor.  The real address
+ * provided for the domain dump buffer must be 64-byte aligned.  The
+ * size specifies the size of the dump buffer and may be larger than
+ * the minimum size specified in the machine description.  The
+ * hypervisor will fill the dump buffer with opaque data.
+ *
+ * Note: A guest may elect to include dump buffer contents as part of a crash
+ *       dump to assist with debugging.  This function may be called any number
+ *       of times so that a guest may relocate a dump buffer, or create
+ *       "snapshots" of any dump-buffer information.  Each call to
+ *       dump_buf_update() atomically declares the new dump buffer to the
+ *       hypervisor.
+ *
+ * A specified size of 0 unconfigures the dump buffer.  If the real
+ * address is illegal or badly aligned, then any currently active dump
+ * buffer is disabled and an error is returned.
+ *
+ * In the event that the call fails with EINVAL, RET1 contains the
+ * minimum size requires by the hypervisor for a valid dump buffer.
+ */
+#define HV_FAST_DUMP_BUF_UPDATE		0x94
+
+/* dump_buf_info()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_DUMP_BUF_INFO
+ * RET0:	status
+ * RET1:	real address of current dump buffer
+ * RET2:	size of current dump buffer
+ * ERRORS:	No errors defined.
+ *
+ * Return the currently configures dump buffer description.  A
+ * returned size of 0 bytes indicates an undefined dump buffer.  In
+ * this case the return address in RET1 is undefined.
+ */
+#define HV_FAST_DUMP_BUF_INFO		0x95
+
+/* Device interrupt services.
+ *
+ * Device interrupts are allocated to system bus bridges by the hypervisor,
+ * and described to OBP in the machine description.  OBP then describes
+ * these interrupts to the OS via properties in the device tree.
+ *
+ * Terminology:
+ *
+ *	cpuid		Unique opaque value which represents a target cpu.
+ *
+ *	devhandle	Device handle.  It uniquely identifies a device, and
+ *			consistes of the lower 28-bits of the hi-cell of the
+ *			first entry of the device's "reg" property in the
+ *			OBP device tree.
+ *
+ *	devino		Device interrupt number.  Specifies the relative
+ *			interrupt number within the device.  The unique
+ *			combination of devhandle and devino are used to
+ *			identify a specific device interrupt.
+ *
+ *			Note: The devino value is the same as the values in the
+ *			      "interrupts" property or "interrupt-map" property
+ *			      in the OBP device tree for that device.
+ *
+ *	sysino		System interrupt number.  A 64-bit unsigned interger
+ *			representing a unique interrupt within a virtual
+ *			machine.
+ *
+ *	intr_state	A flag representing the interrupt state for a given
+ *			sysino.  The state values are defined below.
+ *
+ *	intr_enabled	A flag representing the 'enabled' state for a given
+ *			sysino.  The enable values are defined below.
+ */
+
+#define HV_INTR_STATE_IDLE		0 /* Nothing pending */
+#define HV_INTR_STATE_RECEIVED		1 /* Interrupt received by hardware */
+#define HV_INTR_STATE_DELIVERED		2 /* Interrupt delivered to queue */
+
+#define HV_INTR_DISABLED		0 /* sysino not enabled */
+#define HV_INTR_ENABLED			1 /* sysino enabled */
+
+/* intr_devino_to_sysino()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_INTR_DEVINO2SYSINO
+ * ARG0:	devhandle
+ * ARG1:	devino
+ * RET0:	status
+ * RET1:	sysino
+ * ERRORS:	EINVAL		Invalid devhandle/devino
+ *
+ * Converts a device specific interrupt number of the given
+ * devhandle/devino into a system specific ino (sysino).
+ */
+#define HV_FAST_INTR_DEVINO2SYSINO	0xa0
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_devino_to_sysino(unsigned long devhandle,
+					    unsigned long devino);
+#endif
+
+/* intr_getenabled()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_INTR_GETENABLED
+ * ARG0:	sysino
+ * RET0:	status
+ * RET1:	intr_enabled (HV_INTR_{DISABLED,ENABLED})
+ * ERRORS:	EINVAL		Invalid sysino
+ *
+ * Returns interrupt enabled state in RET1 for the interrupt defined
+ * by the given sysino.
+ */
+#define HV_FAST_INTR_GETENABLED		0xa1
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_intr_getenabled(unsigned long sysino);
+#endif
+
+/* intr_setenabled()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_INTR_SETENABLED
+ * ARG0:	sysino
+ * ARG1:	intr_enabled (HV_INTR_{DISABLED,ENABLED})
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid sysino or intr_enabled value
+ *
+ * Set the 'enabled' state of the interrupt sysino.
+ */
+#define HV_FAST_INTR_SETENABLED		0xa2
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_intr_setenabled(unsigned long sysino, unsigned long intr_enabled);
+#endif
+
+/* intr_getstate()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_INTR_GETSTATE
+ * ARG0:	sysino
+ * RET0:	status
+ * RET1:	intr_state (HV_INTR_STATE_*)
+ * ERRORS:	EINVAL		Invalid sysino
+ *
+ * Returns current state of the interrupt defined by the given sysino.
+ */
+#define HV_FAST_INTR_GETSTATE		0xa3
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_intr_getstate(unsigned long sysino);
+#endif
+
+/* intr_setstate()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_INTR_SETSTATE
+ * ARG0:	sysino
+ * ARG1:	intr_state (HV_INTR_STATE_*)
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid sysino or intr_state value
+ *
+ * Sets the current state of the interrupt described by the given sysino
+ * value.
+ *
+ * Note: Setting the state to HV_INTR_STATE_IDLE clears any pending
+ *       interrupt for sysino.
+ */
+#define HV_FAST_INTR_SETSTATE		0xa4
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_intr_setstate(unsigned long sysino, unsigned long intr_state);
+#endif
+
+/* intr_gettarget()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_INTR_GETTARGET
+ * ARG0:	sysino
+ * RET0:	status
+ * RET1:	cpuid
+ * ERRORS:	EINVAL		Invalid sysino
+ *
+ * Returns CPU that is the current target of the interrupt defined by
+ * the given sysino.  The CPU value returned is undefined if the target
+ * has not been set via intr_settarget().
+ */
+#define HV_FAST_INTR_GETTARGET		0xa5
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_intr_gettarget(unsigned long sysino);
+#endif
+
+/* intr_settarget()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_INTR_SETTARGET
+ * ARG0:	sysino
+ * ARG1:	cpuid
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid sysino
+ *		ENOCPU		Invalid cpuid
+ *
+ * Set the target CPU for the interrupt defined by the given sysino.
+ */
+#define HV_FAST_INTR_SETTARGET		0xa6
+
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_intr_settarget(unsigned long sysino, unsigned long cpuid);
+#endif
+
+/* PCI IO services.
+ *
+ * See the terminology descriptions in the device interrupt services
+ * section above as those apply here too.  Here are terminology
+ * definitions specific to these PCI IO services:
+ *
+ *	tsbnum		TSB number.  Indentifies which io-tsb is used.
+ *			For this version of the specification, tsbnum
+ *			must be zero.
+ *
+ *	tsbindex	TSB index.  Identifies which entry in the TSB
+ *			is used.  The first entry is zero.
+ *
+ *	tsbid		A 64-bit aligned data structure which contains
+ *			a tsbnum and a tsbindex.  Bits 63:32 contain the
+ *			tsbnum and bits 31:00 contain the tsbindex.
+ *
+ *			Use the HV_PCI_TSBID() macro to construct such
+ * 			values.
+ *
+ *	io_attributes	IO attributes for IOMMU mappings.  One of more
+ *			of the attritbute bits are stores in a 64-bit
+ *			value.  The values are defined below.
+ *
+ *	r_addr		64-bit real address
+ *
+ *	pci_device	PCI device address.  A PCI device address identifies
+ *			a specific device on a specific PCI bus segment.
+ *			A PCI device address ia a 32-bit unsigned integer
+ *			with the following format:
+ *
+ *				00000000.bbbbbbbb.dddddfff.00000000
+ *
+ *			Use the HV_PCI_DEVICE_BUILD() macro to construct
+ *			such values.
+ *
+ *	pci_config_offset
+ *			PCI configureation space offset.  For conventional
+ *			PCI a value between 0 and 255.  For extended
+ *			configuration space, a value between 0 and 4095.
+ *
+ *			Note: For PCI configuration space accesses, the offset
+ *			      must be aligned to the access size.
+ *
+ *	error_flag	A return value which specifies if the action succeeded
+ *			or failed.  0 means no error, non-0 means some error
+ *			occurred while performing the service.
+ *
+ *	io_sync_direction
+ *			Direction definition for pci_dma_sync(), defined
+ *			below in HV_PCI_SYNC_*.
+ *
+ *	io_page_list	A list of io_page_addresses, an io_page_address is
+ *			a real address.
+ *
+ *	io_page_list_p	A pointer to an io_page_list.
+ *
+ *	"size based byte swap" - Some functions do size based byte swapping
+ *				 which allows sw to access pointers and
+ *				 counters in native form when the processor
+ *				 operates in a different endianness than the
+ *				 IO bus.  Size-based byte swapping converts a
+ *				 multi-byte field between big-endian and
+ *				 little-endian format.
+ */
+
+#define HV_PCI_MAP_ATTR_READ		0x01
+#define HV_PCI_MAP_ATTR_WRITE		0x02
+
+#define HV_PCI_DEVICE_BUILD(b,d,f)	\
+	((((b) & 0xff) << 16) | \
+	 (((d) & 0x1f) << 11) | \
+	 (((f) & 0x07) <<  8))
+
+#define HV_PCI_TSBID(__tsb_num, __tsb_index) \
+	((((u64)(__tsb_num)) << 32UL) | ((u64)(__tsb_index)))
+
+#define HV_PCI_SYNC_FOR_DEVICE		0x01
+#define HV_PCI_SYNC_FOR_CPU		0x02
+
+/* pci_iommu_map()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_IOMMU_MAP
+ * ARG0:	devhandle
+ * ARG1:	tsbid
+ * ARG2:	#ttes
+ * ARG3:	io_attributes
+ * ARG4:	io_page_list_p
+ * RET0:	status
+ * RET1:	#ttes mapped
+ * ERRORS:	EINVAL		Invalid devhandle/tsbnum/tsbindex/io_attributes
+ *		EBADALIGN	Improperly aligned real address
+ *		ENORADDR	Invalid real address
+ *
+ * Create IOMMU mappings in the sun4v device defined by the given
+ * devhandle.  The mappings are created in the TSB defined by the
+ * tsbnum component of the given tsbid.  The first mapping is created
+ * in the TSB i ndex defined by the tsbindex component of the given tsbid.
+ * The call creates up to #ttes mappings, the first one at tsbnum, tsbindex,
+ * the second at tsbnum, tsbindex + 1, etc.
+ *
+ * All mappings are created with the attributes defined by the io_attributes
+ * argument.  The page mapping addresses are described in the io_page_list
+ * defined by the given io_page_list_p, which is a pointer to the io_page_list.
+ * The first entry in the io_page_list is the address for the first iotte, the
+ * 2nd for the 2nd iotte, and so on.
+ *
+ * Each io_page_address in the io_page_list must be appropriately aligned.
+ * #ttes must be greater than zero.  For this version of the spec, the tsbnum
+ * component of the given tsbid must be zero.
+ *
+ * Returns the actual number of mappings creates, which may be less than
+ * or equal to the argument #ttes.  If the function returns a value which
+ * is less than the #ttes, the caller may continus to call the function with
+ * an updated tsbid, #ttes, io_page_list_p arguments until all pages are
+ * mapped.
+ *
+ * Note: This function does not imply an iotte cache flush.  The guest must
+ *       demap an entry before re-mapping it.
+ */
+#define HV_FAST_PCI_IOMMU_MAP		0xb0
+
+/* pci_iommu_demap()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_IOMMU_DEMAP
+ * ARG0:	devhandle
+ * ARG1:	tsbid
+ * ARG2:	#ttes
+ * RET0:	status
+ * RET1:	#ttes demapped
+ * ERRORS:	EINVAL		Invalid devhandle/tsbnum/tsbindex
+ *
+ * Demap and flush IOMMU mappings in the device defined by the given
+ * devhandle.  Demaps up to #ttes entries in the TSB defined by the tsbnum
+ * component of the given tsbid, starting at the TSB index defined by the
+ * tsbindex component of the given tsbid.
+ *
+ * For this version of the spec, the tsbnum of the given tsbid must be zero.
+ * #ttes must be greater than zero.
+ *
+ * Returns the actual number of ttes demapped, which may be less than or equal
+ * to the argument #ttes.  If #ttes demapped is less than #ttes, the caller
+ * may continue to call this function with updated tsbid and #ttes arguments
+ * until all pages are demapped.
+ *
+ * Note: Entries do not have to be mapped to be demapped.  A demap of an
+ *       unmapped page will flush the entry from the tte cache.
+ */
+#define HV_FAST_PCI_IOMMU_DEMAP		0xb1
+
+/* pci_iommu_getmap()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_IOMMU_GETMAP
+ * ARG0:	devhandle
+ * ARG1:	tsbid
+ * RET0:	status
+ * RET1:	io_attributes
+ * RET2:	real address
+ * ERRORS:	EINVAL		Invalid devhandle/tsbnum/tsbindex
+ *		ENOMAP		Mapping is not valid, no translation exists
+ *
+ * Read and return the mapping in the device described by the given devhandle
+ * and tsbid.  If successful, the io_attributes shall be returned in RET1
+ * and the page address of the mapping shall be returned in RET2.
+ *
+ * For this version of the spec, the tsbnum component of the given tsbid
+ * must be zero.
+ */
+#define HV_FAST_PCI_IOMMU_GETMAP	0xb2
+
+/* pci_iommu_getbypass()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_IOMMU_GETBYPASS
+ * ARG0:	devhandle
+ * ARG1:	real address
+ * ARG2:	io_attributes
+ * RET0:	status
+ * RET1:	io_addr
+ * ERRORS:	EINVAL		Invalid devhandle/io_attributes
+ *		ENORADDR	Invalid real address
+ *		ENOTSUPPORTED	Function not supported in this implementation.
+ *
+ * Create a "special" mapping in the device described by the given devhandle,
+ * for the given real address and attributes.  Return the IO address in RET1
+ * if successful.
+ */
+#define HV_FAST_PCI_IOMMU_GETBYPASS	0xb3
+
+/* pci_config_get()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_CONFIG_GET
+ * ARG0:	devhandle
+ * ARG1:	pci_device
+ * ARG2:	pci_config_offset
+ * ARG3:	size
+ * RET0:	status
+ * RET1:	error_flag
+ * RET2:	data
+ * ERRORS:	EINVAL		Invalid devhandle/pci_device/offset/size
+ *		EBADALIGN	pci_config_offset not size aligned
+ *		ENOACCESS	Access to this offset is not permitted
+ *
+ * Read PCI configuration space for the adapter described by the given
+ * devhandle.  Read size (1, 2, or 4) bytes of data from the given
+ * pci_device, at pci_config_offset from the beginning of the device's
+ * configuration space.  If there was no error, RET1 is set to zero and
+ * RET2 is set to the data read.  Insignificant bits in RET2 are not
+ * guarenteed to have any specific value and therefore must be ignored.
+ *
+ * The data returned in RET2 is size based byte swapped.
+ *
+ * If an error occurs during the read, set RET1 to a non-zero value.  The
+ * given pci_config_offset must be 'size' aligned.
+ */
+#define HV_FAST_PCI_CONFIG_GET		0xb4
+
+/* pci_config_put()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_CONFIG_PUT
+ * ARG0:	devhandle
+ * ARG1:	pci_device
+ * ARG2:	pci_config_offset
+ * ARG3:	size
+ * ARG4:	data
+ * RET0:	status
+ * RET1:	error_flag
+ * ERRORS:	EINVAL		Invalid devhandle/pci_device/offset/size
+ *		EBADALIGN	pci_config_offset not size aligned
+ *		ENOACCESS	Access to this offset is not permitted
+ *
+ * Write PCI configuration space for the adapter described by the given
+ * devhandle.  Write size (1, 2, or 4) bytes of data in a single operation,
+ * at pci_config_offset from the beginning of the device's configuration
+ * space.  The data argument contains the data to be written to configuration
+ * space.  Prior to writing, the data is size based byte swapped.
+ *
+ * If an error occurs during the write access, do not generate an error
+ * report, do set RET1 to a non-zero value.  Otherwise RET1 is zero.
+ * The given pci_config_offset must be 'size' aligned.
+ *
+ * This function is permitted to read from offset zero in the configuration
+ * space described by the given pci_device if necessary to ensure that the
+ * write access to config space completes.
+ */
+#define HV_FAST_PCI_CONFIG_PUT		0xb5
+
+/* pci_peek()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_PEEK
+ * ARG0:	devhandle
+ * ARG1:	real address
+ * ARG2:	size
+ * RET0:	status
+ * RET1:	error_flag
+ * RET2:	data
+ * ERRORS:	EINVAL		Invalid devhandle or size
+ *		EBADALIGN	Improperly aligned real address
+ *		ENORADDR	Bad real address
+ *		ENOACCESS	Guest access prohibited
+ *
+ * Attempt to read the IO address given by the given devhandle, real address,
+ * and size.  Size must be 1, 2, 4, or 8.  The read is performed as a single
+ * access operation using the given size.  If an error occurs when reading
+ * from the given location, do not generate an error report, but return a
+ * non-zero value in RET1.  If the read was successful, return zero in RET1
+ * and return the actual data read in RET2.  The data returned is size based
+ * byte swapped.
+ *
+ * Non-significant bits in RET2 are not guarenteed to have any specific value
+ * and therefore must be ignored.  If RET1 is returned as non-zero, the data 
+ * value is not guarenteed to have any specific value and should be ignored.
+ *
+ * The caller must have permission to read from the given devhandle, real
+ * address, which must be an IO address.  The argument real address must be a
+ * size aligned address.
+ *
+ * The hypervisor implementation of this function must block access to any
+ * IO address that the guest does not have explicit permission to access.
+ */
+#define HV_FAST_PCI_PEEK		0xb6
+
+/* pci_poke()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_POKE
+ * ARG0:	devhandle
+ * ARG1:	real address
+ * ARG2:	size
+ * ARG3:	data
+ * ARG4:	pci_device
+ * RET0:	status
+ * RET1:	error_flag
+ * ERRORS:	EINVAL		Invalid devhandle, size, or pci_device
+ *		EBADALIGN	Improperly aligned real address
+ *		ENORADDR	Bad real address
+ *		ENOACCESS	Guest access prohibited
+ *		ENOTSUPPORTED	Function is not supported by implementation
+ *
+ * Attempt to write data to the IO address given by the given devhandle,
+ * real address, and size.  Size must be 1, 2, 4, or 8.  The write is
+ * performed as a single access operation using the given size. Prior to
+ * writing the data is size based swapped.
+ *
+ * If an error occurs when writing to the given location, do not generate an
+ * error report, but return a non-zero value in RET1.  If the write was
+ * successful, return zero in RET1.
+ *
+ * pci_device describes the configuration address of the device being
+ * written to.  The implementation may safely read from offset 0 with
+ * the configuration space of the device described by devhandle and
+ * pci_device in order to guarantee that the write portion of the operation
+ * completes
+ *
+ * Any error that occurs due to the read shall be reported using the normal
+ * error reporting mechanisms .. the read error is not suppressed.
+ *
+ * The caller must have permission to write to the given devhandle, real
+ * address, which must be an IO address.  The argument real address must be a
+ * size aligned address.  The caller must have permission to read from
+ * the given devhandle, pci_device cofiguration space offset 0.
+ *
+ * The hypervisor implementation of this function must block access to any
+ * IO address that the guest does not have explicit permission to access.
+ */
+#define HV_FAST_PCI_POKE		0xb7
+
+/* pci_dma_sync()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_DMA_SYNC
+ * ARG0:	devhandle
+ * ARG1:	real address
+ * ARG2:	size
+ * ARG3:	io_sync_direction
+ * RET0:	status
+ * RET1:	#synced
+ * ERRORS:	EINVAL		Invalid devhandle or io_sync_direction
+ *		ENORADDR	Bad real address
+ *
+ * Synchronize a memory region described by the given real address and size,
+ * for the device defined by the given devhandle using the direction(s)
+ * defined by the given io_sync_direction.  The argument size is the size of
+ * the memory region in bytes.
+ *
+ * Return the actual number of bytes synchronized in the return value #synced,
+ * which may be less than or equal to the argument size.  If the return
+ * value #synced is less than size, the caller must continue to call this
+ * function with updated real address and size arguments until the entire
+ * memory region is synchronized.
+ */
+#define HV_FAST_PCI_DMA_SYNC		0xb8
+
+/* PCI MSI services.  */
+
+#define HV_MSITYPE_MSI32		0x00
+#define HV_MSITYPE_MSI64		0x01
+
+#define HV_MSIQSTATE_IDLE		0x00
+#define HV_MSIQSTATE_ERROR		0x01
+
+#define HV_MSIQ_INVALID			0x00
+#define HV_MSIQ_VALID			0x01
+
+#define HV_MSISTATE_IDLE		0x00
+#define HV_MSISTATE_DELIVERED		0x01
+
+#define HV_MSIVALID_INVALID		0x00
+#define HV_MSIVALID_VALID		0x01
+
+#define HV_PCIE_MSGTYPE_PME_MSG		0x18
+#define HV_PCIE_MSGTYPE_PME_ACK_MSG	0x1b
+#define HV_PCIE_MSGTYPE_CORR_MSG	0x30
+#define HV_PCIE_MSGTYPE_NONFATAL_MSG	0x31
+#define HV_PCIE_MSGTYPE_FATAL_MSG	0x33
+
+#define HV_MSG_INVALID			0x00
+#define HV_MSG_VALID			0x01
+
+/* pci_msiq_conf()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_CONF
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * ARG2:	real address
+ * ARG3:	number of entries
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle, msiqid or nentries
+ *		EBADALIGN	Improperly aligned real address
+ *		ENORADDR	Bad real address
+ *
+ * Configure the MSI queue given by the devhandle and msiqid arguments,
+ * and to be placed at the given real address and be of the given
+ * number of entries.  The real address must be aligned exactly to match
+ * the queue size.  Each queue entry is 64-bytes long, so f.e. a 32 entry
+ * queue must be aligned on a 2048 byte real address boundary.  The MSI-EQ
+ * Head and Tail are initialized so that the MSI-EQ is 'empty'.
+ *
+ * Implementation Note: Certain implementations have fixed sized queues.  In
+ *                      that case, number of entries must contain the correct
+ *                      value.
+ */
+#define HV_FAST_PCI_MSIQ_CONF		0xc0
+
+/* pci_msiq_info()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_INFO
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * RET0:	status
+ * RET1:	real address
+ * RET2:	number of entries
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid
+ *
+ * Return the configuration information for the MSI queue described
+ * by the given devhandle and msiqid.  The base address of the queue
+ * is returned in ARG1 and the number of entries is returned in ARG2.
+ * If the queue is unconfigured, the real address is undefined and the
+ * number of entries will be returned as zero.
+ */
+#define HV_FAST_PCI_MSIQ_INFO		0xc1
+
+/* pci_msiq_getvalid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_GETVALID
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * RET0:	status
+ * RET1:	msiqvalid	(HV_MSIQ_VALID or HV_MSIQ_INVALID)
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid
+ *
+ * Get the valid state of the MSI-EQ described by the given devhandle and
+ * msiqid.
+ */
+#define HV_FAST_PCI_MSIQ_GETVALID	0xc2
+
+/* pci_msiq_setvalid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_SETVALID
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * ARG2:	msiqvalid	(HV_MSIQ_VALID or HV_MSIQ_INVALID)
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid or msiqvalid
+ *				value or MSI EQ is uninitialized
+ *
+ * Set the valid state of the MSI-EQ described by the given devhandle and
+ * msiqid to the given msiqvalid.
+ */
+#define HV_FAST_PCI_MSIQ_SETVALID	0xc3
+
+/* pci_msiq_getstate()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_GETSTATE
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * RET0:	status
+ * RET1:	msiqstate	(HV_MSIQSTATE_IDLE or HV_MSIQSTATE_ERROR)
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid
+ *
+ * Get the state of the MSI-EQ described by the given devhandle and
+ * msiqid.
+ */
+#define HV_FAST_PCI_MSIQ_GETSTATE	0xc4
+
+/* pci_msiq_getvalid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_GETVALID
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * ARG2:	msiqstate	(HV_MSIQSTATE_IDLE or HV_MSIQSTATE_ERROR)
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid or msiqstate
+ *				value or MSI EQ is uninitialized
+ *
+ * Set the state of the MSI-EQ described by the given devhandle and
+ * msiqid to the given msiqvalid.
+ */
+#define HV_FAST_PCI_MSIQ_SETSTATE	0xc5
+
+/* pci_msiq_gethead()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_GETHEAD
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * RET0:	status
+ * RET1:	msiqhead
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid
+ *
+ * Get the current MSI EQ queue head for the MSI-EQ described by the
+ * given devhandle and msiqid.
+ */
+#define HV_FAST_PCI_MSIQ_GETHEAD	0xc6
+
+/* pci_msiq_sethead()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_SETHEAD
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * ARG2:	msiqhead
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid or msiqhead,
+ *				or MSI EQ is uninitialized
+ *
+ * Set the current MSI EQ queue head for the MSI-EQ described by the
+ * given devhandle and msiqid.
+ */
+#define HV_FAST_PCI_MSIQ_SETHEAD	0xc7
+
+/* pci_msiq_gettail()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSIQ_GETTAIL
+ * ARG0:	devhandle
+ * ARG1:	msiqid
+ * RET0:	status
+ * RET1:	msiqtail
+ * ERRORS:	EINVAL		Invalid devhandle or msiqid
+ *
+ * Get the current MSI EQ queue tail for the MSI-EQ described by the
+ * given devhandle and msiqid.
+ */
+#define HV_FAST_PCI_MSIQ_GETTAIL	0xc8
+
+/* pci_msi_getvalid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSI_GETVALID
+ * ARG0:	devhandle
+ * ARG1:	msinum
+ * RET0:	status
+ * RET1:	msivalidstate
+ * ERRORS:	EINVAL		Invalid devhandle or msinum
+ *
+ * Get the current valid/enabled state for the MSI defined by the
+ * given devhandle and msinum.
+ */
+#define HV_FAST_PCI_MSI_GETVALID	0xc9
+
+/* pci_msi_setvalid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSI_SETVALID
+ * ARG0:	devhandle
+ * ARG1:	msinum
+ * ARG2:	msivalidstate
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle or msinum or msivalidstate
+ *
+ * Set the current valid/enabled state for the MSI defined by the
+ * given devhandle and msinum.
+ */
+#define HV_FAST_PCI_MSI_SETVALID	0xca
+
+/* pci_msi_getmsiq()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSI_GETMSIQ
+ * ARG0:	devhandle
+ * ARG1:	msinum
+ * RET0:	status
+ * RET1:	msiqid
+ * ERRORS:	EINVAL		Invalid devhandle or msinum or MSI is unbound
+ *
+ * Get the MSI EQ that the MSI defined by the given devhandle and
+ * msinum is bound to.
+ */
+#define HV_FAST_PCI_MSI_GETMSIQ		0xcb
+
+/* pci_msi_setmsiq()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSI_SETMSIQ
+ * ARG0:	devhandle
+ * ARG1:	msinum
+ * ARG2:	msitype
+ * ARG3:	msiqid
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle or msinum or msiqid
+ *
+ * Set the MSI EQ that the MSI defined by the given devhandle and
+ * msinum is bound to.
+ */
+#define HV_FAST_PCI_MSI_SETMSIQ		0xcc
+
+/* pci_msi_getstate()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSI_GETSTATE
+ * ARG0:	devhandle
+ * ARG1:	msinum
+ * RET0:	status
+ * RET1:	msistate
+ * ERRORS:	EINVAL		Invalid devhandle or msinum
+ *
+ * Get the state of the MSI defined by the given devhandle and msinum.
+ * If not initialized, return HV_MSISTATE_IDLE.
+ */
+#define HV_FAST_PCI_MSI_GETSTATE	0xcd
+
+/* pci_msi_setstate()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSI_SETSTATE
+ * ARG0:	devhandle
+ * ARG1:	msinum
+ * ARG2:	msistate
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle or msinum or msistate
+ *
+ * Set the state of the MSI defined by the given devhandle and msinum.
+ */
+#define HV_FAST_PCI_MSI_SETSTATE	0xce
+
+/* pci_msg_getmsiq()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSG_GETMSIQ
+ * ARG0:	devhandle
+ * ARG1:	msgtype
+ * RET0:	status
+ * RET1:	msiqid
+ * ERRORS:	EINVAL		Invalid devhandle or msgtype
+ *
+ * Get the MSI EQ of the MSG defined by the given devhandle and msgtype.
+ */
+#define HV_FAST_PCI_MSG_GETMSIQ		0xd0
+
+/* pci_msg_setmsiq()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSG_SETMSIQ
+ * ARG0:	devhandle
+ * ARG1:	msgtype
+ * ARG2:	msiqid
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle, msgtype, or msiqid
+ *
+ * Set the MSI EQ of the MSG defined by the given devhandle and msgtype.
+ */
+#define HV_FAST_PCI_MSG_SETMSIQ		0xd1
+
+/* pci_msg_getvalid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSG_GETVALID
+ * ARG0:	devhandle
+ * ARG1:	msgtype
+ * RET0:	status
+ * RET1:	msgvalidstate
+ * ERRORS:	EINVAL		Invalid devhandle or msgtype
+ *
+ * Get the valid/enabled state of the MSG defined by the given
+ * devhandle and msgtype.
+ */
+#define HV_FAST_PCI_MSG_GETVALID	0xd2
+
+/* pci_msg_setvalid()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_PCI_MSG_SETVALID
+ * ARG0:	devhandle
+ * ARG1:	msgtype
+ * ARG2:	msgvalidstate
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid devhandle or msgtype or msgvalidstate
+ *
+ * Set the valid/enabled state of the MSG defined by the given
+ * devhandle and msgtype.
+ */
+#define HV_FAST_PCI_MSG_SETVALID	0xd3
+
+/* Performance counter services.  */
+
+#define HV_PERF_JBUS_PERF_CTRL_REG	0x00
+#define HV_PERF_JBUS_PERF_CNT_REG	0x01
+#define HV_PERF_DRAM_PERF_CTRL_REG_0	0x02
+#define HV_PERF_DRAM_PERF_CNT_REG_0	0x03
+#define HV_PERF_DRAM_PERF_CTRL_REG_1	0x04
+#define HV_PERF_DRAM_PERF_CNT_REG_1	0x05
+#define HV_PERF_DRAM_PERF_CTRL_REG_2	0x06
+#define HV_PERF_DRAM_PERF_CNT_REG_2	0x07
+#define HV_PERF_DRAM_PERF_CTRL_REG_3	0x08
+#define HV_PERF_DRAM_PERF_CNT_REG_3	0x09
+
+/* get_perfreg()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_GET_PERFREG
+ * ARG0:	performance reg number
+ * RET0:	status
+ * RET1:	performance reg value
+ * ERRORS:	EINVAL		Invalid performance register number
+ *		ENOACCESS	No access allowed to performance counters
+ *
+ * Read the value of the given DRAM/JBUS performance counter/control register.
+ */
+#define HV_FAST_GET_PERFREG		0x100
+
+/* set_perfreg()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_SET_PERFREG
+ * ARG0:	performance reg number
+ * ARG1:	performance reg value
+ * RET0:	status
+ * ERRORS:	EINVAL		Invalid performance register number
+ *		ENOACCESS	No access allowed to performance counters
+ *
+ * Write the given performance reg value to the given DRAM/JBUS
+ * performance counter/control register.
+ */
+#define HV_FAST_SET_PERFREG		0x101
+
+/* MMU statistics services.
+ *
+ * The hypervisor maintains MMU statistics and privileged code provides
+ * a buffer where these statistics can be collected.  It is continually
+ * updated once configured.  The layout is as follows:
+ */
+#ifndef __ASSEMBLY__
+struct hv_mmu_statistics {
+	unsigned long immu_tsb_hits_ctx0_8k_tte;
+	unsigned long immu_tsb_ticks_ctx0_8k_tte;
+	unsigned long immu_tsb_hits_ctx0_64k_tte;
+	unsigned long immu_tsb_ticks_ctx0_64k_tte;
+	unsigned long __reserved1[2];
+	unsigned long immu_tsb_hits_ctx0_4mb_tte;
+	unsigned long immu_tsb_ticks_ctx0_4mb_tte;
+	unsigned long __reserved2[2];
+	unsigned long immu_tsb_hits_ctx0_256mb_tte;
+	unsigned long immu_tsb_ticks_ctx0_256mb_tte;
+	unsigned long __reserved3[4];
+	unsigned long immu_tsb_hits_ctxnon0_8k_tte;
+	unsigned long immu_tsb_ticks_ctxnon0_8k_tte;
+	unsigned long immu_tsb_hits_ctxnon0_64k_tte;
+	unsigned long immu_tsb_ticks_ctxnon0_64k_tte;
+	unsigned long __reserved4[2];
+	unsigned long immu_tsb_hits_ctxnon0_4mb_tte;
+	unsigned long immu_tsb_ticks_ctxnon0_4mb_tte;
+	unsigned long __reserved5[2];
+	unsigned long immu_tsb_hits_ctxnon0_256mb_tte;
+	unsigned long immu_tsb_ticks_ctxnon0_256mb_tte;
+	unsigned long __reserved6[4];
+	unsigned long dmmu_tsb_hits_ctx0_8k_tte;
+	unsigned long dmmu_tsb_ticks_ctx0_8k_tte;
+	unsigned long dmmu_tsb_hits_ctx0_64k_tte;
+	unsigned long dmmu_tsb_ticks_ctx0_64k_tte;
+	unsigned long __reserved7[2];
+	unsigned long dmmu_tsb_hits_ctx0_4mb_tte;
+	unsigned long dmmu_tsb_ticks_ctx0_4mb_tte;
+	unsigned long __reserved8[2];
+	unsigned long dmmu_tsb_hits_ctx0_256mb_tte;
+	unsigned long dmmu_tsb_ticks_ctx0_256mb_tte;
+	unsigned long __reserved9[4];
+	unsigned long dmmu_tsb_hits_ctxnon0_8k_tte;
+	unsigned long dmmu_tsb_ticks_ctxnon0_8k_tte;
+	unsigned long dmmu_tsb_hits_ctxnon0_64k_tte;
+	unsigned long dmmu_tsb_ticks_ctxnon0_64k_tte;
+	unsigned long __reserved10[2];
+	unsigned long dmmu_tsb_hits_ctxnon0_4mb_tte;
+	unsigned long dmmu_tsb_ticks_ctxnon0_4mb_tte;
+	unsigned long __reserved11[2];
+	unsigned long dmmu_tsb_hits_ctxnon0_256mb_tte;
+	unsigned long dmmu_tsb_ticks_ctxnon0_256mb_tte;
+	unsigned long __reserved12[4];
+};
+#endif
+
+/* mmustat_conf()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMUSTAT_CONF
+ * ARG0:	real address
+ * RET0:	status
+ * RET1:	real address
+ * ERRORS:	ENORADDR	Invalid real address
+ *		EBADALIGN	Real address not aligned on 64-byte boundary
+ *		EBADTRAP	API not supported on this processor
+ *
+ * Enable MMU statistic gathering using the buffer at the given real
+ * address on the current virtual CPU.  The new buffer real address
+ * is given in ARG1, and the previously specified buffer real address
+ * is returned in RET1, or is returned as zero for the first invocation.
+ *
+ * If the passed in real address argument is zero, this will disable
+ * MMU statistic collection on the current virtual CPU.  If an error is
+ * returned then no statistics are collected.
+ *
+ * The buffer contents should be initialized to all zeros before being
+ * given to the hypervisor or else the statistics will be meaningless.
+ */
+#define HV_FAST_MMUSTAT_CONF		0x102
+
+/* mmustat_info()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_MMUSTAT_INFO
+ * RET0:	status
+ * RET1:	real address
+ * ERRORS:	EBADTRAP	API not supported on this processor
+ *
+ * Return the current state and real address of the currently configured
+ * MMU statistics buffer on the current virtual CPU.
+ */
+#define HV_FAST_MMUSTAT_INFO		0x103
+
+/* Function numbers for HV_CORE_TRAP.  */
+#define HV_CORE_VER			0x00
+#define HV_CORE_PUTCHAR			0x01
+#define HV_CORE_EXIT			0x02
+
+#endif /* !(_SPARC64_HYPERVISOR_H) */
diff --git a/include/asm-sparc64/idprom.h b/include/asm-sparc64/idprom.h
index 701483c..77fbf98 100644
--- a/include/asm-sparc64/idprom.h
+++ b/include/asm-sparc64/idprom.h
@@ -9,15 +9,7 @@
 
 #include <linux/types.h>
 
-/* Offset into the EEPROM where the id PROM is located on the 4c */
-#define IDPROM_OFFSET  0x7d8
-
-/* On sun4m; physical. */
-/* MicroSPARC(-II) does not decode 31rd bit, but it works. */
-#define IDPROM_OFFSET_M  0xfd8
-
-struct idprom
-{
+struct idprom {
 	u8		id_format;	/* Format identifier (always 0x01) */
 	u8		id_machtype;	/* Machine type */
 	u8		id_ethaddr[6];	/* Hardware ethernet address */
@@ -30,6 +22,4 @@ struct idprom
 extern struct idprom *idprom;
 extern void idprom_init(void);
 
-#define IDPROM_SIZE  (sizeof(struct idprom))
-
 #endif /* !(_SPARC_IDPROM_H) */
diff --git a/include/asm-sparc64/intr_queue.h b/include/asm-sparc64/intr_queue.h
new file mode 100644
index 0000000..206077d
--- /dev/null
+++ b/include/asm-sparc64/intr_queue.h
@@ -0,0 +1,15 @@
+#ifndef _SPARC64_INTR_QUEUE_H
+#define _SPARC64_INTR_QUEUE_H
+
+/* Sun4v interrupt queue registers, accessed via ASI_QUEUE.  */
+
+#define INTRQ_CPU_MONDO_HEAD	  0x3c0 /* CPU mondo head	          */
+#define INTRQ_CPU_MONDO_TAIL	  0x3c8 /* CPU mondo tail	          */
+#define INTRQ_DEVICE_MONDO_HEAD	  0x3d0 /* Device mondo head	          */
+#define INTRQ_DEVICE_MONDO_TAIL	  0x3d8 /* Device mondo tail	          */
+#define INTRQ_RESUM_MONDO_HEAD	  0x3e0 /* Resumable error mondo head     */
+#define INTRQ_RESUM_MONDO_TAIL	  0x3e8 /* Resumable error mondo tail     */
+#define INTRQ_NONRESUM_MONDO_HEAD 0x3f0 /* Non-resumable error mondo head */
+#define INTRQ_NONRESUM_MONDO_TAIL 0x3f8 /* Non-resumable error mondo head */
+
+#endif /* !(_SPARC64_INTR_QUEUE_H) */
diff --git a/include/asm-sparc64/irq.h b/include/asm-sparc64/irq.h
index 8b70edc..de33d6e 100644
--- a/include/asm-sparc64/irq.h
+++ b/include/asm-sparc64/irq.h
@@ -72,8 +72,11 @@ struct ino_bucket {
 #define IMAP_VALID		0x80000000	/* IRQ Enabled		*/
 #define IMAP_TID_UPA		0x7c000000	/* UPA TargetID		*/
 #define IMAP_TID_JBUS		0x7c000000	/* JBUS TargetID	*/
+#define IMAP_TID_SHIFT		26
 #define IMAP_AID_SAFARI		0x7c000000	/* Safari AgentID	*/
+#define IMAP_AID_SHIFT		26
 #define IMAP_NID_SAFARI		0x03e00000	/* Safari NodeID	*/
+#define IMAP_NID_SHIFT		21
 #define IMAP_IGN		0x000007c0	/* IRQ Group Number	*/
 #define IMAP_INO		0x0000003f	/* IRQ Number		*/
 #define IMAP_INR		0x000007ff	/* Full interrupt number*/
@@ -111,6 +114,7 @@ extern void disable_irq(unsigned int);
 #define disable_irq_nosync disable_irq
 extern void enable_irq(unsigned int);
 extern unsigned int build_irq(int pil, int inofixup, unsigned long iclr, unsigned long imap);
+extern unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino, int pil, unsigned char flags);
 extern unsigned int sbus_build_irq(void *sbus, unsigned int ino);
 
 static __inline__ void set_softint(unsigned long bits)
diff --git a/include/asm-sparc64/mmu.h b/include/asm-sparc64/mmu.h
index 8627eed..230ba67 100644
--- a/include/asm-sparc64/mmu.h
+++ b/include/asm-sparc64/mmu.h
@@ -4,20 +4,9 @@
 #include <linux/config.h>
 #include <asm/page.h>
 #include <asm/const.h>
+#include <asm/hypervisor.h>
 
-/*
- * For the 8k pagesize kernel, use only 10 hw context bits to optimize some
- * shifts in the fast tlbmiss handlers, instead of all 13 bits (specifically
- * for vpte offset calculation). For other pagesizes, this optimization in
- * the tlbhandlers can not be done; but still, all 13 bits can not be used
- * because the tlb handlers use "andcc" instruction which sign extends 13
- * bit arguments.
- */
-#if PAGE_SHIFT == 13
-#define CTX_NR_BITS		10
-#else
-#define CTX_NR_BITS		12
-#endif
+#define CTX_NR_BITS		13
 
 #define TAG_CONTEXT_BITS	((_AC(1,UL) << CTX_NR_BITS) - _AC(1,UL))
 
@@ -90,8 +79,27 @@
 
 #ifndef __ASSEMBLY__
 
+#define TSB_ENTRY_ALIGNMENT	16
+
+struct tsb {
+	unsigned long tag;
+	unsigned long pte;
+} __attribute__((aligned(TSB_ENTRY_ALIGNMENT)));
+
+extern void __tsb_insert(unsigned long ent, unsigned long tag, unsigned long pte);
+extern void tsb_flush(unsigned long ent, unsigned long tag);
+extern void tsb_init(struct tsb *tsb, unsigned long size);
+
 typedef struct {
-	unsigned long	sparc64_ctx_val;
+	spinlock_t		lock;
+	unsigned long		sparc64_ctx_val;
+	struct tsb		*tsb;
+	unsigned long		tsb_rss_limit;
+	unsigned long		tsb_nentries;
+	unsigned long		tsb_reg_val;
+	unsigned long		tsb_map_vaddr;
+	unsigned long		tsb_map_pte;
+	struct hv_tsb_descr	tsb_descr;
 } mm_context_t;
 
 #endif /* !__ASSEMBLY__ */
diff --git a/include/asm-sparc64/mmu_context.h b/include/asm-sparc64/mmu_context.h
index 57ee7b3..e797432 100644
--- a/include/asm-sparc64/mmu_context.h
+++ b/include/asm-sparc64/mmu_context.h
@@ -19,96 +19,98 @@ extern unsigned long tlb_context_cache;
 extern unsigned long mmu_context_bmap[];
 
 extern void get_new_mmu_context(struct mm_struct *mm);
+#ifdef CONFIG_SMP
+extern void smp_new_mmu_context_version(void);
+#else
+#define smp_new_mmu_context_version() do { } while (0)
+#endif
+
+extern int init_new_context(struct task_struct *tsk, struct mm_struct *mm);
+extern void destroy_context(struct mm_struct *mm);
+
+extern void __tsb_context_switch(unsigned long pgd_pa,
+				 unsigned long tsb_reg,
+				 unsigned long tsb_vaddr,
+				 unsigned long tsb_pte,
+				 unsigned long tsb_descr_pa);
+
+static inline void tsb_context_switch(struct mm_struct *mm)
+{
+	__tsb_context_switch(__pa(mm->pgd), mm->context.tsb_reg_val,
+			     mm->context.tsb_map_vaddr,
+			     mm->context.tsb_map_pte,
+			     __pa(&mm->context.tsb_descr));
+}
 
-/* Initialize a new mmu context.  This is invoked when a new
- * address space instance (unique or shared) is instantiated.
- * This just needs to set mm->context to an invalid context.
- */
-#define init_new_context(__tsk, __mm)	\
-	(((__mm)->context.sparc64_ctx_val = 0UL), 0)
-
-/* Destroy a dead context.  This occurs when mmput drops the
- * mm_users count to zero, the mmaps have been released, and
- * all the page tables have been flushed.  Our job is to destroy
- * any remaining processor-specific state, and in the sparc64
- * case this just means freeing up the mmu context ID held by
- * this task if valid.
- */
-#define destroy_context(__mm)					\
-do {	spin_lock(&ctx_alloc_lock);				\
-	if (CTX_VALID((__mm)->context)) {			\
-		unsigned long nr = CTX_NRBITS((__mm)->context);	\
-		mmu_context_bmap[nr>>6] &= ~(1UL << (nr & 63));	\
-	}							\
-	spin_unlock(&ctx_alloc_lock);				\
-} while(0)
-
-/* Reload the two core values used by TLB miss handler
- * processing on sparc64.  They are:
- * 1) The physical address of mm->pgd, when full page
- *    table walks are necessary, this is where the
- *    search begins.
- * 2) A "PGD cache".  For 32-bit tasks only pgd[0] is
- *    ever used since that maps the entire low 4GB
- *    completely.  To speed up TLB miss processing we
- *    make this value available to the handlers.  This
- *    decreases the amount of memory traffic incurred.
- */
-#define reload_tlbmiss_state(__tsk, __mm) \
-do { \
-	register unsigned long paddr asm("o5"); \
-	register unsigned long pgd_cache asm("o4"); \
-	paddr = __pa((__mm)->pgd); \
-	pgd_cache = 0UL; \
-	if (task_thread_info(__tsk)->flags & _TIF_32BIT) \
-		pgd_cache = get_pgd_cache((__mm)->pgd); \
-	__asm__ __volatile__("wrpr	%%g0, 0x494, %%pstate\n\t" \
-			     "mov	%3, %%g4\n\t" \
-			     "mov	%0, %%g7\n\t" \
-			     "stxa	%1, [%%g4] %2\n\t" \
-			     "membar	#Sync\n\t" \
-			     "wrpr	%%g0, 0x096, %%pstate" \
-			     : /* no outputs */ \
-			     : "r" (paddr), "r" (pgd_cache),\
-			       "i" (ASI_DMMU), "i" (TSB_REG)); \
-} while(0)
+extern void tsb_grow(struct mm_struct *mm, unsigned long mm_rss);
+#ifdef CONFIG_SMP
+extern void smp_tsb_sync(struct mm_struct *mm);
+#else
+#define smp_tsb_sync(__mm) do { } while (0)
+#endif
 
 /* Set MMU context in the actual hardware. */
 #define load_secondary_context(__mm) \
-	__asm__ __volatile__("stxa	%0, [%1] %2\n\t" \
-			     "flush	%%g6" \
-			     : /* No outputs */ \
-			     : "r" (CTX_HWBITS((__mm)->context)), \
-			       "r" (SECONDARY_CONTEXT), "i" (ASI_DMMU))
+	__asm__ __volatile__( \
+	"\n661:	stxa		%0, [%1] %2\n" \
+	"	.section	.sun4v_1insn_patch, \"ax\"\n" \
+	"	.word		661b\n" \
+	"	stxa		%0, [%1] %3\n" \
+	"	.previous\n" \
+	"	flush		%%g6\n" \
+	: /* No outputs */ \
+	: "r" (CTX_HWBITS((__mm)->context)), \
+	  "r" (SECONDARY_CONTEXT), "i" (ASI_DMMU), "i" (ASI_MMU))
 
 extern void __flush_tlb_mm(unsigned long, unsigned long);
 
-/* Switch the current MM context. */
+/* Switch the current MM context.  Interrupts are disabled.  */
 static inline void switch_mm(struct mm_struct *old_mm, struct mm_struct *mm, struct task_struct *tsk)
 {
-	unsigned long ctx_valid;
+	unsigned long ctx_valid, flags;
 	int cpu;
 
-	/* Note: page_table_lock is used here to serialize switch_mm
-	 * and activate_mm, and their calls to get_new_mmu_context.
-	 * This use of page_table_lock is unrelated to its other uses.
-	 */ 
-	spin_lock(&mm->page_table_lock);
+	spin_lock_irqsave(&mm->context.lock, flags);
 	ctx_valid = CTX_VALID(mm->context);
 	if (!ctx_valid)
 		get_new_mmu_context(mm);
-	spin_unlock(&mm->page_table_lock);
 
-	if (!ctx_valid || (old_mm != mm)) {
-		load_secondary_context(mm);
-		reload_tlbmiss_state(tsk, mm);
-	}
+	/* We have to be extremely careful here or else we will miss
+	 * a TSB grow if we switch back and forth between a kernel
+	 * thread and an address space which has it's TSB size increased
+	 * on another processor.
+	 *
+	 * It is possible to play some games in order to optimize the
+	 * switch, but the safest thing to do is to unconditionally
+	 * perform the secondary context load and the TSB context switch.
+	 *
+	 * For reference the bad case is, for address space "A":
+	 *
+	 *		CPU 0			CPU 1
+	 *	run address space A
+	 *	set cpu0's bits in cpu_vm_mask
+	 *	switch to kernel thread, borrow
+	 *	address space A via entry_lazy_tlb
+	 *					run address space A
+	 *					set cpu1's bit in cpu_vm_mask
+	 *					flush_tlb_pending()
+	 *					reset cpu_vm_mask to just cpu1
+	 *					TSB grow
+	 *	run address space A
+	 *	context was valid, so skip
+	 *	TSB context switch
+	 *
+	 * At that point cpu0 continues to use a stale TSB, the one from
+	 * before the TSB grow performed on cpu1.  cpu1 did not cross-call
+	 * cpu0 to update it's TSB because at that point the cpu_vm_mask
+	 * only had cpu1 set in it.
+	 */
+	load_secondary_context(mm);
+	tsb_context_switch(mm);
 
-	/* Even if (mm == old_mm) we _must_ check
-	 * the cpu_vm_mask.  If we do not we could
-	 * corrupt the TLB state because of how
-	 * smp_flush_tlb_{page,range,mm} on sparc64
-	 * and lazy tlb switches work. -DaveM
+	/* Any time a processor runs a context on an address space
+	 * for the first time, we must flush that context out of the
+	 * local TLB.
 	 */
 	cpu = smp_processor_id();
 	if (!ctx_valid || !cpu_isset(cpu, mm->cpu_vm_mask)) {
@@ -116,6 +118,7 @@ static inline void switch_mm(struct mm_struct *old_mm, struct mm_struct *mm, str
 		__flush_tlb_mm(CTX_HWBITS(mm->context),
 			       SECONDARY_CONTEXT);
 	}
+	spin_unlock_irqrestore(&mm->context.lock, flags);
 }
 
 #define deactivate_mm(tsk,mm)	do { } while (0)
@@ -123,23 +126,20 @@ static inline void switch_mm(struct mm_struct *old_mm, struct mm_struct *mm, str
 /* Activate a new MM instance for the current task. */
 static inline void activate_mm(struct mm_struct *active_mm, struct mm_struct *mm)
 {
+	unsigned long flags;
 	int cpu;
 
-	/* Note: page_table_lock is used here to serialize switch_mm
-	 * and activate_mm, and their calls to get_new_mmu_context.
-	 * This use of page_table_lock is unrelated to its other uses.
-	 */ 
-	spin_lock(&mm->page_table_lock);
+	spin_lock_irqsave(&mm->context.lock, flags);
 	if (!CTX_VALID(mm->context))
 		get_new_mmu_context(mm);
 	cpu = smp_processor_id();
 	if (!cpu_isset(cpu, mm->cpu_vm_mask))
 		cpu_set(cpu, mm->cpu_vm_mask);
-	spin_unlock(&mm->page_table_lock);
 
 	load_secondary_context(mm);
 	__flush_tlb_mm(CTX_HWBITS(mm->context), SECONDARY_CONTEXT);
-	reload_tlbmiss_state(current, mm);
+	tsb_context_switch(mm);
+	spin_unlock_irqrestore(&mm->context.lock, flags);
 }
 
 #endif /* !(__ASSEMBLY__) */
diff --git a/include/asm-sparc64/numnodes.h b/include/asm-sparc64/numnodes.h
new file mode 100644
index 0000000..017e7e7
--- /dev/null
+++ b/include/asm-sparc64/numnodes.h
@@ -0,0 +1,6 @@
+#ifndef _SPARC64_NUMNODES_H
+#define _SPARC64_NUMNODES_H
+
+#define NODES_SHIFT	0
+
+#endif /* !(_SPARC64_NUMNODES_H) */
diff --git a/include/asm-sparc64/oplib.h b/include/asm-sparc64/oplib.h
index 3c59b26..c754676 100644
--- a/include/asm-sparc64/oplib.h
+++ b/include/asm-sparc64/oplib.h
@@ -12,18 +12,8 @@
 #include <linux/config.h>
 #include <asm/openprom.h>
 
-/* Enumeration to describe the prom major version we have detected. */
-enum prom_major_version {
-	PROM_V0,      /* Original sun4c V0 prom */
-	PROM_V2,      /* sun4c and early sun4m V2 prom */
-	PROM_V3,      /* sun4m and later, up to sun4d/sun4e machines V3 */
-	PROM_P1275,   /* IEEE compliant ISA based Sun PROM, only sun4u */
-        PROM_AP1000,  /* actually no prom at all */
-};
-
-extern enum prom_major_version prom_vers;
-/* Revision, and firmware revision. */
-extern unsigned int prom_rev, prom_prev;
+/* OBP version string. */
+extern char prom_version[];
 
 /* Root node of the prom device tree, this stays constant after
  * initialization is complete.
@@ -39,6 +29,9 @@ extern int prom_stdin, prom_stdout;
 extern int prom_chosen_node;
 
 /* Helper values and strings in arch/sparc64/kernel/head.S */
+extern const char prom_peer_name[];
+extern const char prom_compatible_name[];
+extern const char prom_root_compatible[];
 extern const char prom_finddev_name[];
 extern const char prom_chosen_path[];
 extern const char prom_getprop_name[];
@@ -130,15 +123,6 @@ extern void prom_setcallback(callback_func_t func_ptr);
  */
 extern unsigned char prom_get_idprom(char *idp_buffer, int idpbuf_size);
 
-/* Get the prom major version. */
-extern int prom_version(void);
-
-/* Get the prom plugin revision. */
-extern int prom_getrev(void);
-
-/* Get the prom firmware revision. */
-extern int prom_getprev(void);
-
 /* Character operations to/from the console.... */
 
 /* Non-blocking get character from console. */
@@ -164,6 +148,7 @@ enum prom_input_device {
 	PROMDEV_ITTYA,			/* input from ttya */
 	PROMDEV_ITTYB,			/* input from ttyb */
 	PROMDEV_IRSC,			/* input from rsc */
+	PROMDEV_IVCONS,			/* input from virtual-console */
 	PROMDEV_I_UNK,
 };
 
@@ -176,6 +161,7 @@ enum prom_output_device {
 	PROMDEV_OTTYA,			/* to ttya */
 	PROMDEV_OTTYB,			/* to ttyb */
 	PROMDEV_ORSC,			/* to rsc */
+	PROMDEV_OVCONS,			/* to virtual-console */
 	PROMDEV_O_UNK,
 };
 
@@ -183,10 +169,18 @@ extern enum prom_output_device prom_query_output_device(void);
 
 /* Multiprocessor operations... */
 #ifdef CONFIG_SMP
-/* Start the CPU with the given device tree node, context table, and context
- * at the passed program counter.
+/* Start the CPU with the given device tree node at the passed program
+ * counter with the given arg passed in via register %o0.
+ */
+extern void prom_startcpu(int cpunode, unsigned long pc, unsigned long arg);
+
+/* Start the CPU with the given cpu ID at the passed program
+ * counter with the given arg passed in via register %o0.
  */
-extern void prom_startcpu(int cpunode, unsigned long pc, unsigned long o0);
+extern void prom_startcpu_cpuid(int cpuid, unsigned long pc, unsigned long arg);
+
+/* Stop the CPU with the given cpu ID.  */
+extern void prom_stopcpu_cpuid(int cpuid);
 
 /* Stop the current CPU. */
 extern void prom_stopself(void);
@@ -335,6 +329,7 @@ int cpu_find_by_mid(int mid, int *prom_node);
 
 /* Client interface level routines. */
 extern void prom_set_trap_table(unsigned long tba);
+extern void prom_set_trap_table_sun4v(unsigned long tba, unsigned long mmfsa);
 
 extern long p1275_cmd(const char *, long, ...);
 				   
diff --git a/include/asm-sparc64/page.h b/include/asm-sparc64/page.h
index 5426bb2..fcb2812 100644
--- a/include/asm-sparc64/page.h
+++ b/include/asm-sparc64/page.h
@@ -104,10 +104,12 @@ typedef unsigned long pgprot_t;
 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
 #define ARCH_HAS_SETCLEAR_HUGE_PTE
 #define ARCH_HAS_HUGETLB_PREFAULT_HOOK
+#define HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 #endif
 
 #define TASK_UNMAPPED_BASE	(test_thread_flag(TIF_32BIT) ? \
-				 (_AC(0x0000000070000000,UL)) : (PAGE_OFFSET))
+				 (_AC(0x0000000070000000,UL)) : \
+				 (_AC(0xfffff80000000000,UL) + (1UL << 32UL)))
 
 #endif /* !(__ASSEMBLY__) */
 
@@ -124,17 +126,10 @@ typedef unsigned long pgprot_t;
 #define __pa(x)			((unsigned long)(x) - PAGE_OFFSET)
 #define __va(x)			((void *)((unsigned long) (x) + PAGE_OFFSET))
 
-/* PFNs are real physical page numbers.  However, mem_map only begins to record
- * per-page information starting at pfn_base.  This is to handle systems where
- * the first physical page in the machine is at some huge physical address,
- * such as 4GB.   This is common on a partitioned E10000, for example.
- */
-extern struct page *pfn_to_page(unsigned long pfn);
-extern unsigned long page_to_pfn(struct page *);
+#define pfn_to_kaddr(pfn)	__va((pfn) << PAGE_SHIFT)
 
 #define virt_to_page(kaddr)	pfn_to_page(__pa(kaddr)>>PAGE_SHIFT)
 
-#define pfn_valid(pfn)		(((pfn)-(pfn_base)) < max_mapnr)
 #define virt_addr_valid(kaddr)	pfn_valid(__pa(kaddr) >> PAGE_SHIFT)
 
 #define virt_to_phys __pa
diff --git a/include/asm-sparc64/pbm.h b/include/asm-sparc64/pbm.h
index dd35a2c..1396f11 100644
--- a/include/asm-sparc64/pbm.h
+++ b/include/asm-sparc64/pbm.h
@@ -139,6 +139,9 @@ struct pci_pbm_info {
 	/* Opaque 32-bit system bus Port ID. */
 	u32				portid;
 
+	/* Opaque 32-bit handle used for hypervisor calls.  */
+	u32				devhandle;
+
 	/* Chipset version information. */
 	int				chip_type;
 #define PBM_CHIP_TYPE_SABRE		1
diff --git a/include/asm-sparc64/pci.h b/include/asm-sparc64/pci.h
index 89bd71b..7c5a589 100644
--- a/include/asm-sparc64/pci.h
+++ b/include/asm-sparc64/pci.h
@@ -41,10 +41,26 @@ static inline void pcibios_penalize_isa_irq(int irq, int active)
 
 struct pci_dev;
 
+struct pci_iommu_ops {
+	void *(*alloc_consistent)(struct pci_dev *, size_t, dma_addr_t *);
+	void (*free_consistent)(struct pci_dev *, size_t, void *, dma_addr_t);
+	dma_addr_t (*map_single)(struct pci_dev *, void *, size_t, int);
+	void (*unmap_single)(struct pci_dev *, dma_addr_t, size_t, int);
+	int (*map_sg)(struct pci_dev *, struct scatterlist *, int, int);
+	void (*unmap_sg)(struct pci_dev *, struct scatterlist *, int, int);
+	void (*dma_sync_single_for_cpu)(struct pci_dev *, dma_addr_t, size_t, int);
+	void (*dma_sync_sg_for_cpu)(struct pci_dev *, struct scatterlist *, int, int);
+};
+
+extern struct pci_iommu_ops *pci_iommu_ops;
+
 /* Allocate and map kernel buffer using consistent mode DMA for a device.
  * hwdev should be valid struct pci_dev pointer for PCI devices.
  */
-extern void *pci_alloc_consistent(struct pci_dev *hwdev, size_t size, dma_addr_t *dma_handle);
+static inline void *pci_alloc_consistent(struct pci_dev *hwdev, size_t size, dma_addr_t *dma_handle)
+{
+	return pci_iommu_ops->alloc_consistent(hwdev, size, dma_handle);
+}
 
 /* Free and unmap a consistent DMA buffer.
  * cpu_addr is what was returned from pci_alloc_consistent,
@@ -54,7 +70,10 @@ extern void *pci_alloc_consistent(struct pci_dev *hwdev, size_t size, dma_addr_t
  * References to the memory and mappings associated with cpu_addr/dma_addr
  * past this call are illegal.
  */
-extern void pci_free_consistent(struct pci_dev *hwdev, size_t size, void *vaddr, dma_addr_t dma_handle);
+static inline void pci_free_consistent(struct pci_dev *hwdev, size_t size, void *vaddr, dma_addr_t dma_handle)
+{
+	return pci_iommu_ops->free_consistent(hwdev, size, vaddr, dma_handle);
+}
 
 /* Map a single buffer of the indicated size for DMA in streaming mode.
  * The 32-bit bus address to use is returned.
@@ -62,7 +81,10 @@ extern void pci_free_consistent(struct pci_dev *hwdev, size_t size, void *vaddr,
  * Once the device is given the dma address, the device owns this memory
  * until either pci_unmap_single or pci_dma_sync_single_for_cpu is performed.
  */
-extern dma_addr_t pci_map_single(struct pci_dev *hwdev, void *ptr, size_t size, int direction);
+static inline dma_addr_t pci_map_single(struct pci_dev *hwdev, void *ptr, size_t size, int direction)
+{
+	return pci_iommu_ops->map_single(hwdev, ptr, size, direction);
+}
 
 /* Unmap a single streaming mode DMA translation.  The dma_addr and size
  * must match what was provided for in a previous pci_map_single call.  All
@@ -71,7 +93,10 @@ extern dma_addr_t pci_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
  * After this call, reads by the cpu to the buffer are guaranteed to see
  * whatever the device wrote there.
  */
-extern void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t dma_addr, size_t size, int direction);
+static inline void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t dma_addr, size_t size, int direction)
+{
+	pci_iommu_ops->unmap_single(hwdev, dma_addr, size, direction);
+}
 
 /* No highmem on sparc64, plus we have an IOMMU, so mapping pages is easy. */
 #define pci_map_page(dev, page, off, size, dir) \
@@ -107,15 +132,19 @@ extern void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t dma_addr, size_t
  * Device ownership issues as mentioned above for pci_map_single are
  * the same here.
  */
-extern int pci_map_sg(struct pci_dev *hwdev, struct scatterlist *sg,
-		      int nents, int direction);
+static inline int pci_map_sg(struct pci_dev *hwdev, struct scatterlist *sg, int nents, int direction)
+{
+	return pci_iommu_ops->map_sg(hwdev, sg, nents, direction);
+}
 
 /* Unmap a set of streaming mode DMA translations.
  * Again, cpu read rules concerning calls here are the same as for
  * pci_unmap_single() above.
  */
-extern void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sg,
-			 int nhwents, int direction);
+static inline void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sg, int nhwents, int direction)
+{
+	pci_iommu_ops->unmap_sg(hwdev, sg, nhwents, direction);
+}
 
 /* Make physical memory consistent for a single
  * streaming mode DMA translation after a transfer.
@@ -127,8 +156,10 @@ extern void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sg,
  * must first perform a pci_dma_sync_for_device, and then the
  * device again owns the buffer.
  */
-extern void pci_dma_sync_single_for_cpu(struct pci_dev *hwdev, dma_addr_t dma_handle,
-					size_t size, int direction);
+static inline void pci_dma_sync_single_for_cpu(struct pci_dev *hwdev, dma_addr_t dma_handle, size_t size, int direction)
+{
+	pci_iommu_ops->dma_sync_single_for_cpu(hwdev, dma_handle, size, direction);
+}
 
 static inline void
 pci_dma_sync_single_for_device(struct pci_dev *hwdev, dma_addr_t dma_handle,
@@ -144,7 +175,10 @@ pci_dma_sync_single_for_device(struct pci_dev *hwdev, dma_addr_t dma_handle,
  * The same as pci_dma_sync_single_* but for a scatter-gather list,
  * same rules and usage.
  */
-extern void pci_dma_sync_sg_for_cpu(struct pci_dev *hwdev, struct scatterlist *sg, int nelems, int direction);
+static inline void pci_dma_sync_sg_for_cpu(struct pci_dev *hwdev, struct scatterlist *sg, int nelems, int direction)
+{
+	pci_iommu_ops->dma_sync_sg_for_cpu(hwdev, sg, nelems, direction);
+}
 
 static inline void
 pci_dma_sync_sg_for_device(struct pci_dev *hwdev, struct scatterlist *sg,
diff --git a/include/asm-sparc64/pgalloc.h b/include/asm-sparc64/pgalloc.h
index a96067c..12e4a27 100644
--- a/include/asm-sparc64/pgalloc.h
+++ b/include/asm-sparc64/pgalloc.h
@@ -6,6 +6,7 @@
 #include <linux/kernel.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
+#include <linux/slab.h>
 
 #include <asm/spitfire.h>
 #include <asm/cpudata.h>
@@ -13,172 +14,59 @@
 #include <asm/page.h>
 
 /* Page table allocation/freeing. */
-#ifdef CONFIG_SMP
-/* Sliiiicck */
-#define pgt_quicklists	local_cpu_data()
-#else
-extern struct pgtable_cache_struct {
-	unsigned long *pgd_cache;
-	unsigned long *pte_cache[2];
-	unsigned int pgcache_size;
-} pgt_quicklists;
-#endif
-#define pgd_quicklist		(pgt_quicklists.pgd_cache)
-#define pmd_quicklist		((unsigned long *)0)
-#define pte_quicklist		(pgt_quicklists.pte_cache)
-#define pgtable_cache_size	(pgt_quicklists.pgcache_size)
+extern kmem_cache_t *pgtable_cache;
 
-static __inline__ void free_pgd_fast(pgd_t *pgd)
+static inline pgd_t *pgd_alloc(struct mm_struct *mm)
 {
-	preempt_disable();
-	*(unsigned long *)pgd = (unsigned long) pgd_quicklist;
-	pgd_quicklist = (unsigned long *) pgd;
-	pgtable_cache_size++;
-	preempt_enable();
+	return kmem_cache_alloc(pgtable_cache, GFP_KERNEL);
 }
 
-static __inline__ pgd_t *get_pgd_fast(void)
+static inline void pgd_free(pgd_t *pgd)
 {
-	unsigned long *ret;
-
-	preempt_disable();
-	if((ret = pgd_quicklist) != NULL) {
-		pgd_quicklist = (unsigned long *)(*ret);
-		ret[0] = 0;
-		pgtable_cache_size--;
-		preempt_enable();
-	} else {
-		preempt_enable();
-		ret = (unsigned long *) __get_free_page(GFP_KERNEL|__GFP_REPEAT);
-		if(ret)
-			memset(ret, 0, PAGE_SIZE);
-	}
-	return (pgd_t *)ret;
+	kmem_cache_free(pgtable_cache, pgd);
 }
 
-static __inline__ void free_pgd_slow(pgd_t *pgd)
-{
-	free_page((unsigned long)pgd);
-}
-
-#ifdef DCACHE_ALIASING_POSSIBLE
-#define VPTE_COLOR(address)		(((address) >> (PAGE_SHIFT + 10)) & 1UL)
-#define DCACHE_COLOR(address)		(((address) >> PAGE_SHIFT) & 1UL)
-#else
-#define VPTE_COLOR(address)		0
-#define DCACHE_COLOR(address)		0
-#endif
-
 #define pud_populate(MM, PUD, PMD)	pud_set(PUD, PMD)
 
-static __inline__ pmd_t *pmd_alloc_one_fast(struct mm_struct *mm, unsigned long address)
+static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long addr)
 {
-	unsigned long *ret;
-	int color = 0;
-
-	preempt_disable();
-	if (pte_quicklist[color] == NULL)
-		color = 1;
-
-	if((ret = (unsigned long *)pte_quicklist[color]) != NULL) {
-		pte_quicklist[color] = (unsigned long *)(*ret);
-		ret[0] = 0;
-		pgtable_cache_size--;
-	}
-	preempt_enable();
-
-	return (pmd_t *)ret;
+	return kmem_cache_alloc(pgtable_cache,
+				GFP_KERNEL|__GFP_REPEAT);
 }
 
-static __inline__ pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long address)
+static inline void pmd_free(pmd_t *pmd)
 {
-	pmd_t *pmd;
-
-	pmd = pmd_alloc_one_fast(mm, address);
-	if (!pmd) {
-		pmd = (pmd_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT);
-		if (pmd)
-			memset(pmd, 0, PAGE_SIZE);
-	}
-	return pmd;
+	kmem_cache_free(pgtable_cache, pmd);
 }
 
-static __inline__ void free_pmd_fast(pmd_t *pmd)
+static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm,
+					  unsigned long address)
 {
-	unsigned long color = DCACHE_COLOR((unsigned long)pmd);
-
-	preempt_disable();
-	*(unsigned long *)pmd = (unsigned long) pte_quicklist[color];
-	pte_quicklist[color] = (unsigned long *) pmd;
-	pgtable_cache_size++;
-	preempt_enable();
+	return kmem_cache_alloc(pgtable_cache,
+				GFP_KERNEL|__GFP_REPEAT);
 }
 
-static __inline__ void free_pmd_slow(pmd_t *pmd)
+static inline struct page *pte_alloc_one(struct mm_struct *mm,
+					 unsigned long address)
 {
-	free_page((unsigned long)pmd);
+	return virt_to_page(pte_alloc_one_kernel(mm, address));
 }
-
-#define pmd_populate_kernel(MM, PMD, PTE)	pmd_set(PMD, PTE)
-#define pmd_populate(MM,PMD,PTE_PAGE)		\
-	pmd_populate_kernel(MM,PMD,page_address(PTE_PAGE))
-
-extern pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address);
-
-static inline struct page *
-pte_alloc_one(struct mm_struct *mm, unsigned long addr)
-{
-	pte_t *pte = pte_alloc_one_kernel(mm, addr);
-
-	if (pte)
-		return virt_to_page(pte);
-
-	return NULL;
-}
-
-static __inline__ pte_t *pte_alloc_one_fast(struct mm_struct *mm, unsigned long address)
-{
-	unsigned long color = VPTE_COLOR(address);
-	unsigned long *ret;
-
-	preempt_disable();
-	if((ret = (unsigned long *)pte_quicklist[color]) != NULL) {
-		pte_quicklist[color] = (unsigned long *)(*ret);
-		ret[0] = 0;
-		pgtable_cache_size--;
-	}
-	preempt_enable();
-	return (pte_t *)ret;
-}
-
-static __inline__ void free_pte_fast(pte_t *pte)
-{
-	unsigned long color = DCACHE_COLOR((unsigned long)pte);
-
-	preempt_disable();
-	*(unsigned long *)pte = (unsigned long) pte_quicklist[color];
-	pte_quicklist[color] = (unsigned long *) pte;
-	pgtable_cache_size++;
-	preempt_enable();
-}
-
-static __inline__ void free_pte_slow(pte_t *pte)
-{
-	free_page((unsigned long)pte);
-}
-
+		
 static inline void pte_free_kernel(pte_t *pte)
 {
-	free_pte_fast(pte);
+	kmem_cache_free(pgtable_cache, pte);
 }
 
 static inline void pte_free(struct page *ptepage)
 {
-	free_pte_fast(page_address(ptepage));
+	pte_free_kernel(page_address(ptepage));
 }
 
-#define pmd_free(pmd)		free_pmd_fast(pmd)
-#define pgd_free(pgd)		free_pgd_fast(pgd)
-#define pgd_alloc(mm)		get_pgd_fast()
+
+#define pmd_populate_kernel(MM, PMD, PTE)	pmd_set(PMD, PTE)
+#define pmd_populate(MM,PMD,PTE_PAGE)		\
+	pmd_populate_kernel(MM,PMD,page_address(PTE_PAGE))
+
+#define check_pgt_cache()	do { } while (0)
 
 #endif /* _SPARC64_PGALLOC_H */
diff --git a/include/asm-sparc64/pgtable.h b/include/asm-sparc64/pgtable.h
index f0a9b44..ed4124e 100644
--- a/include/asm-sparc64/pgtable.h
+++ b/include/asm-sparc64/pgtable.h
@@ -25,7 +25,8 @@
 #include <asm/const.h>
 
 /* The kernel image occupies 0x4000000 to 0x1000000 (4MB --> 32MB).
- * The page copy blockops can use 0x2000000 to 0x10000000.
+ * The page copy blockops can use 0x2000000 to 0x4000000.
+ * The TSB is mapped in the 0x4000000 to 0x6000000 range.
  * The PROM resides in an area spanning 0xf0000000 to 0x100000000.
  * The vmalloc area spans 0x100000000 to 0x200000000.
  * Since modules need to be in the lowest 32-bits of the address space,
@@ -34,6 +35,7 @@
  * 0x400000000.
  */
 #define	TLBTEMP_BASE		_AC(0x0000000002000000,UL)
+#define	TSBMAP_BASE		_AC(0x0000000004000000,UL)
 #define MODULES_VADDR		_AC(0x0000000010000000,UL)
 #define MODULES_LEN		_AC(0x00000000e0000000,UL)
 #define MODULES_END		_AC(0x00000000f0000000,UL)
@@ -88,162 +90,538 @@
 
 #endif /* !(__ASSEMBLY__) */
 
-/* Spitfire/Cheetah TTE bits. */
-#define _PAGE_VALID	_AC(0x8000000000000000,UL) /* Valid TTE              */
-#define _PAGE_R		_AC(0x8000000000000000,UL) /* Keep ref bit up to date*/
-#define _PAGE_SZ4MB	_AC(0x6000000000000000,UL) /* 4MB Page               */
-#define _PAGE_SZ512K	_AC(0x4000000000000000,UL) /* 512K Page              */
-#define _PAGE_SZ64K	_AC(0x2000000000000000,UL) /* 64K Page               */
-#define _PAGE_SZ8K	_AC(0x0000000000000000,UL) /* 8K Page                */
-#define _PAGE_NFO	_AC(0x1000000000000000,UL) /* No Fault Only          */
-#define _PAGE_IE	_AC(0x0800000000000000,UL) /* Invert Endianness      */
-#define _PAGE_SOFT2	_AC(0x07FC000000000000,UL) /* Software bits, set 2   */
-#define _PAGE_RES1	_AC(0x0002000000000000,UL) /* Reserved               */
-#define _PAGE_SZ32MB	_AC(0x0001000000000000,UL) /* (Panther) 32MB page    */
-#define _PAGE_SZ256MB	_AC(0x2001000000000000,UL) /* (Panther) 256MB page   */
-#define _PAGE_SN	_AC(0x0000800000000000,UL) /* (Cheetah) Snoop        */
-#define _PAGE_RES2	_AC(0x0000780000000000,UL) /* Reserved               */
-#define _PAGE_PADDR_SF	_AC(0x000001FFFFFFE000,UL) /* (Spitfire) paddr[40:13]*/
-#define _PAGE_PADDR	_AC(0x000007FFFFFFE000,UL) /* (Cheetah) paddr[42:13] */
-#define _PAGE_SOFT	_AC(0x0000000000001F80,UL) /* Software bits          */
-#define _PAGE_L		_AC(0x0000000000000040,UL) /* Locked TTE             */
-#define _PAGE_CP	_AC(0x0000000000000020,UL) /* Cacheable in P-Cache   */
-#define _PAGE_CV	_AC(0x0000000000000010,UL) /* Cacheable in V-Cache   */
-#define _PAGE_E		_AC(0x0000000000000008,UL) /* side-Effect            */
-#define _PAGE_P		_AC(0x0000000000000004,UL) /* Privileged Page        */
-#define _PAGE_W		_AC(0x0000000000000002,UL) /* Writable               */
-#define _PAGE_G		_AC(0x0000000000000001,UL) /* Global                 */
-
-/* Here are the SpitFire software bits we use in the TTE's.
- *
- * WARNING: If you are going to try and start using some
- *          of the soft2 bits, you will need to make
- *          modifications to the swap entry implementation.
- *	    For example, one thing that could happen is that
- *          swp_entry_to_pte() would BUG_ON() if you tried
- *          to use one of the soft2 bits for _PAGE_FILE.
- *
- * Like other architectures, I have aliased _PAGE_FILE with
- * _PAGE_MODIFIED.  This works because _PAGE_FILE is never
- * interpreted that way unless _PAGE_PRESENT is clear.
- */
-#define _PAGE_EXEC	_AC(0x0000000000001000,UL)	/* Executable SW bit */
-#define _PAGE_MODIFIED	_AC(0x0000000000000800,UL)	/* Modified (dirty)  */
-#define _PAGE_FILE	_AC(0x0000000000000800,UL)	/* Pagecache page    */
-#define _PAGE_ACCESSED	_AC(0x0000000000000400,UL)	/* Accessed (ref'd)  */
-#define _PAGE_READ	_AC(0x0000000000000200,UL)	/* Readable SW Bit   */
-#define _PAGE_WRITE	_AC(0x0000000000000100,UL)	/* Writable SW Bit   */
-#define _PAGE_PRESENT	_AC(0x0000000000000080,UL)	/* Present           */
+/* PTE bits which are the same in SUN4U and SUN4V format.  */
+#define _PAGE_VALID	  _AC(0x8000000000000000,UL) /* Valid TTE            */
+#define _PAGE_R	  	  _AC(0x8000000000000000,UL) /* Keep ref bit uptodate*/
+
+/* SUN4U pte bits... */
+#define _PAGE_SZ4MB_4U	  _AC(0x6000000000000000,UL) /* 4MB Page             */
+#define _PAGE_SZ512K_4U	  _AC(0x4000000000000000,UL) /* 512K Page            */
+#define _PAGE_SZ64K_4U	  _AC(0x2000000000000000,UL) /* 64K Page             */
+#define _PAGE_SZ8K_4U	  _AC(0x0000000000000000,UL) /* 8K Page              */
+#define _PAGE_NFO_4U	  _AC(0x1000000000000000,UL) /* No Fault Only        */
+#define _PAGE_IE_4U	  _AC(0x0800000000000000,UL) /* Invert Endianness    */
+#define _PAGE_SOFT2_4U	  _AC(0x07FC000000000000,UL) /* Software bits, set 2 */
+#define _PAGE_RES1_4U	  _AC(0x0002000000000000,UL) /* Reserved             */
+#define _PAGE_SZ32MB_4U	  _AC(0x0001000000000000,UL) /* (Panther) 32MB page  */
+#define _PAGE_SZ256MB_4U  _AC(0x2001000000000000,UL) /* (Panther) 256MB page */
+#define _PAGE_SN_4U	  _AC(0x0000800000000000,UL) /* (Cheetah) Snoop      */
+#define _PAGE_RES2_4U	  _AC(0x0000780000000000,UL) /* Reserved             */
+#define _PAGE_PADDR_4U	  _AC(0x000007FFFFFFE000,UL) /* (Cheetah) pa[42:13]  */
+#define _PAGE_SOFT_4U	  _AC(0x0000000000001F80,UL) /* Software bits:       */
+#define _PAGE_EXEC_4U	  _AC(0x0000000000001000,UL) /* Executable SW bit    */
+#define _PAGE_MODIFIED_4U _AC(0x0000000000000800,UL) /* Modified (dirty)     */
+#define _PAGE_FILE_4U	  _AC(0x0000000000000800,UL) /* Pagecache page       */
+#define _PAGE_ACCESSED_4U _AC(0x0000000000000400,UL) /* Accessed (ref'd)     */
+#define _PAGE_READ_4U	  _AC(0x0000000000000200,UL) /* Readable SW Bit      */
+#define _PAGE_WRITE_4U	  _AC(0x0000000000000100,UL) /* Writable SW Bit      */
+#define _PAGE_PRESENT_4U  _AC(0x0000000000000080,UL) /* Present              */
+#define _PAGE_L_4U	  _AC(0x0000000000000040,UL) /* Locked TTE           */
+#define _PAGE_CP_4U	  _AC(0x0000000000000020,UL) /* Cacheable in P-Cache */
+#define _PAGE_CV_4U	  _AC(0x0000000000000010,UL) /* Cacheable in V-Cache */
+#define _PAGE_E_4U	  _AC(0x0000000000000008,UL) /* side-Effect          */
+#define _PAGE_P_4U	  _AC(0x0000000000000004,UL) /* Privileged Page      */
+#define _PAGE_W_4U	  _AC(0x0000000000000002,UL) /* Writable             */
+
+/* SUN4V pte bits... */
+#define _PAGE_NFO_4V	  _AC(0x4000000000000000,UL) /* No Fault Only        */
+#define _PAGE_SOFT2_4V	  _AC(0x3F00000000000000,UL) /* Software bits, set 2 */
+#define _PAGE_MODIFIED_4V _AC(0x2000000000000000,UL) /* Modified (dirty)     */
+#define _PAGE_ACCESSED_4V _AC(0x1000000000000000,UL) /* Accessed (ref'd)     */
+#define _PAGE_READ_4V	  _AC(0x0800000000000000,UL) /* Readable SW Bit      */
+#define _PAGE_WRITE_4V	  _AC(0x0400000000000000,UL) /* Writable SW Bit      */
+#define _PAGE_PADDR_4V	  _AC(0x00FFFFFFFFFFE000,UL) /* paddr[55:13]         */
+#define _PAGE_IE_4V	  _AC(0x0000000000001000,UL) /* Invert Endianness    */
+#define _PAGE_E_4V	  _AC(0x0000000000000800,UL) /* side-Effect          */
+#define _PAGE_CP_4V	  _AC(0x0000000000000400,UL) /* Cacheable in P-Cache */
+#define _PAGE_CV_4V	  _AC(0x0000000000000200,UL) /* Cacheable in V-Cache */
+#define _PAGE_P_4V	  _AC(0x0000000000000100,UL) /* Privileged Page      */
+#define _PAGE_EXEC_4V	  _AC(0x0000000000000080,UL) /* Executable Page      */
+#define _PAGE_W_4V	  _AC(0x0000000000000040,UL) /* Writable             */
+#define _PAGE_SOFT_4V	  _AC(0x0000000000000030,UL) /* Software bits        */
+#define _PAGE_FILE_4V	  _AC(0x0000000000000020,UL) /* Pagecache page       */
+#define _PAGE_PRESENT_4V  _AC(0x0000000000000010,UL) /* Present              */
+#define _PAGE_RESV_4V	  _AC(0x0000000000000008,UL) /* Reserved             */
+#define _PAGE_SZ16GB_4V	  _AC(0x0000000000000007,UL) /* 16GB Page            */
+#define _PAGE_SZ2GB_4V	  _AC(0x0000000000000006,UL) /* 2GB Page             */
+#define _PAGE_SZ256MB_4V  _AC(0x0000000000000005,UL) /* 256MB Page           */
+#define _PAGE_SZ32MB_4V	  _AC(0x0000000000000004,UL) /* 32MB Page            */
+#define _PAGE_SZ4MB_4V	  _AC(0x0000000000000003,UL) /* 4MB Page             */
+#define _PAGE_SZ512K_4V	  _AC(0x0000000000000002,UL) /* 512K Page            */
+#define _PAGE_SZ64K_4V	  _AC(0x0000000000000001,UL) /* 64K Page             */
+#define _PAGE_SZ8K_4V	  _AC(0x0000000000000000,UL) /* 8K Page              */
 
 #if PAGE_SHIFT == 13
-#define _PAGE_SZBITS	_PAGE_SZ8K
+#define _PAGE_SZBITS_4U	_PAGE_SZ8K_4U
+#define _PAGE_SZBITS_4V	_PAGE_SZ8K_4V
 #elif PAGE_SHIFT == 16
-#define _PAGE_SZBITS	_PAGE_SZ64K
+#define _PAGE_SZBITS_4U	_PAGE_SZ64K_4U
+#define _PAGE_SZBITS_4V	_PAGE_SZ64K_4V
 #elif PAGE_SHIFT == 19
-#define _PAGE_SZBITS	_PAGE_SZ512K
+#define _PAGE_SZBITS_4U	_PAGE_SZ512K_4U
+#define _PAGE_SZBITS_4V	_PAGE_SZ512K_4V
 #elif PAGE_SHIFT == 22
-#define _PAGE_SZBITS	_PAGE_SZ4MB
+#define _PAGE_SZBITS_4U	_PAGE_SZ4MB_4U
+#define _PAGE_SZBITS_4V	_PAGE_SZ4MB_4V
 #else
 #error Wrong PAGE_SHIFT specified
 #endif
 
 #if defined(CONFIG_HUGETLB_PAGE_SIZE_4MB)
-#define _PAGE_SZHUGE	_PAGE_SZ4MB
+#define _PAGE_SZHUGE_4U	_PAGE_SZ4MB_4U
+#define _PAGE_SZHUGE_4V	_PAGE_SZ4MB_4V
 #elif defined(CONFIG_HUGETLB_PAGE_SIZE_512K)
-#define _PAGE_SZHUGE	_PAGE_SZ512K
+#define _PAGE_SZHUGE_4U	_PAGE_SZ512K_4U
+#define _PAGE_SZHUGE_4V	_PAGE_SZ512K_4V
 #elif defined(CONFIG_HUGETLB_PAGE_SIZE_64K)
-#define _PAGE_SZHUGE	_PAGE_SZ64K
+#define _PAGE_SZHUGE_4U	_PAGE_SZ64K_4U
+#define _PAGE_SZHUGE_4V	_PAGE_SZ64K_4V
 #endif
 
-#define _PAGE_CACHE	(_PAGE_CP | _PAGE_CV)
+/* These are actually filled in at boot time by sun4{u,v}_pgprot_init() */
+#define __P000	__pgprot(0)
+#define __P001	__pgprot(0)
+#define __P010	__pgprot(0)
+#define __P011	__pgprot(0)
+#define __P100	__pgprot(0)
+#define __P101	__pgprot(0)
+#define __P110	__pgprot(0)
+#define __P111	__pgprot(0)
+
+#define __S000	__pgprot(0)
+#define __S001	__pgprot(0)
+#define __S010	__pgprot(0)
+#define __S011	__pgprot(0)
+#define __S100	__pgprot(0)
+#define __S101	__pgprot(0)
+#define __S110	__pgprot(0)
+#define __S111	__pgprot(0)
 
-#define __DIRTY_BITS	(_PAGE_MODIFIED | _PAGE_WRITE | _PAGE_W)
-#define __ACCESS_BITS	(_PAGE_ACCESSED | _PAGE_READ | _PAGE_R)
-#define __PRIV_BITS	_PAGE_P
+#ifndef __ASSEMBLY__
 
-#define PAGE_NONE	__pgprot (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_CACHE)
+extern pte_t mk_pte_io(unsigned long, pgprot_t, int, unsigned long);
 
-/* Don't set the TTE _PAGE_W bit here, else the dirty bit never gets set. */
-#define PAGE_SHARED	__pgprot (_PAGE_PRESENT | _PAGE_VALID | _PAGE_CACHE | \
-				  __ACCESS_BITS | _PAGE_WRITE | _PAGE_EXEC)
+extern unsigned long pte_sz_bits(unsigned long size);
 
-#define PAGE_COPY	__pgprot (_PAGE_PRESENT | _PAGE_VALID | _PAGE_CACHE | \
-				  __ACCESS_BITS | _PAGE_EXEC)
+extern pgprot_t PAGE_KERNEL;
+extern pgprot_t PAGE_KERNEL_LOCKED;
+extern pgprot_t PAGE_COPY;
+extern pgprot_t PAGE_SHARED;
 
-#define PAGE_READONLY	__pgprot (_PAGE_PRESENT | _PAGE_VALID | _PAGE_CACHE | \
-				  __ACCESS_BITS | _PAGE_EXEC)
+/* XXX This uglyness is for the atyfb driver's sparc mmap() support. XXX */
+extern unsigned long _PAGE_IE;
+extern unsigned long _PAGE_E;
+extern unsigned long _PAGE_CACHE;
 
-#define PAGE_KERNEL	__pgprot (_PAGE_PRESENT | _PAGE_VALID | _PAGE_CACHE | \
-				  __PRIV_BITS | \
-				  __ACCESS_BITS | __DIRTY_BITS | _PAGE_EXEC)
+extern unsigned long pg_iobits;
+extern unsigned long _PAGE_ALL_SZ_BITS;
+extern unsigned long _PAGE_SZBITS;
 
-#define PAGE_SHARED_NOEXEC	__pgprot (_PAGE_PRESENT | _PAGE_VALID | \
-					  _PAGE_CACHE | \
-					  __ACCESS_BITS | _PAGE_WRITE)
+extern struct page *mem_map_zero;
+#define ZERO_PAGE(vaddr)	(mem_map_zero)
 
-#define PAGE_COPY_NOEXEC	__pgprot (_PAGE_PRESENT | _PAGE_VALID | \
-					  _PAGE_CACHE | __ACCESS_BITS)
+/* PFNs are real physical page numbers.  However, mem_map only begins to record
+ * per-page information starting at pfn_base.  This is to handle systems where
+ * the first physical page in the machine is at some huge physical address,
+ * such as 4GB.   This is common on a partitioned E10000, for example.
+ */
+static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
+{
+	unsigned long paddr = pfn << PAGE_SHIFT;
+	unsigned long sz_bits;
+
+	sz_bits = 0UL;
+	if (_PAGE_SZBITS_4U != 0UL || _PAGE_SZBITS_4V != 0UL) {
+		__asm__ __volatile__(
+		"\n661:	sethi		%uhi(%1), %0\n"
+		"	sllx		%0, 32, %0\n"
+		"	.section	.sun4v_2insn_patch, \"ax\"\n"
+		"	.word		661b\n"
+		"	mov		%2, %0\n"
+		"	nop\n"
+		"	.previous\n"
+		: "=r" (sz_bits)
+		: "i" (_PAGE_SZBITS_4U), "i" (_PAGE_SZBITS_4V));
+	}
+	return __pte(paddr | sz_bits | pgprot_val(prot));
+}
+#define mk_pte(page, pgprot)	pfn_pte(page_to_pfn(page), (pgprot))
 
-#define PAGE_READONLY_NOEXEC	__pgprot (_PAGE_PRESENT | _PAGE_VALID | \
-					  _PAGE_CACHE | __ACCESS_BITS)
+/* This one can be done with two shifts.  */
+static inline unsigned long pte_pfn(pte_t pte)
+{
+	unsigned long ret;
+
+	__asm__ __volatile__(
+	"\n661:	sllx		%1, %2, %0\n"
+	"	srlx		%0, %3, %0\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sllx		%1, %4, %0\n"
+	"	srlx		%0, %5, %0\n"
+	"	.previous\n"
+	: "=r" (ret)
+	: "r" (pte_val(pte)),
+	  "i" (21), "i" (21 + PAGE_SHIFT),
+	  "i" (8), "i" (8 + PAGE_SHIFT));
+
+	return ret;
+}
+#define pte_page(x) pfn_to_page(pte_pfn(x))
 
-#define _PFN_MASK	_PAGE_PADDR
+static inline pte_t pte_modify(pte_t pte, pgprot_t prot)
+{
+	unsigned long mask, tmp;
+
+	/* SUN4U: 0x600307ffffffecb8 (negated == 0x9ffcf80000001347)
+	 * SUN4V: 0x30ffffffffffee17 (negated == 0xcf000000000011e8)
+	 *
+	 * Even if we use negation tricks the result is still a 6
+	 * instruction sequence, so don't try to play fancy and just
+	 * do the most straightforward implementation.
+	 *
+	 * Note: We encode this into 3 sun4v 2-insn patch sequences.
+	 */
 
-#define pg_iobits (_PAGE_VALID | _PAGE_PRESENT | __DIRTY_BITS | \
-		   __ACCESS_BITS | _PAGE_E)
+	__asm__ __volatile__(
+	"\n661:	sethi		%%uhi(%2), %1\n"
+	"	sethi		%%hi(%2), %0\n"
+	"\n662:	or		%1, %%ulo(%2), %1\n"
+	"	or		%0, %%lo(%2), %0\n"
+	"\n663:	sllx		%1, 32, %1\n"
+	"	or		%0, %1, %0\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%3), %1\n"
+	"	sethi		%%hi(%3), %0\n"
+	"	.word		662b\n"
+	"	or		%1, %%ulo(%3), %1\n"
+	"	or		%0, %%lo(%3), %0\n"
+	"	.word		663b\n"
+	"	sllx		%1, 32, %1\n"
+	"	or		%0, %1, %0\n"
+	"	.previous\n"
+	: "=r" (mask), "=r" (tmp)
+	: "i" (_PAGE_PADDR_4U | _PAGE_MODIFIED_4U | _PAGE_ACCESSED_4U |
+	       _PAGE_CP_4U | _PAGE_CV_4U | _PAGE_E_4U | _PAGE_PRESENT_4U |
+	       _PAGE_SZBITS_4U),
+	  "i" (_PAGE_PADDR_4V | _PAGE_MODIFIED_4V | _PAGE_ACCESSED_4V |
+	       _PAGE_CP_4V | _PAGE_CV_4V | _PAGE_E_4V | _PAGE_PRESENT_4V |
+	       _PAGE_SZBITS_4V));
+
+	return __pte((pte_val(pte) & mask) | (pgprot_val(prot) & ~mask));
+}
 
-#define __P000	PAGE_NONE
-#define __P001	PAGE_READONLY_NOEXEC
-#define __P010	PAGE_COPY_NOEXEC
-#define __P011	PAGE_COPY_NOEXEC
-#define __P100	PAGE_READONLY
-#define __P101	PAGE_READONLY
-#define __P110	PAGE_COPY
-#define __P111	PAGE_COPY
+static inline pte_t pgoff_to_pte(unsigned long off)
+{
+	off <<= PAGE_SHIFT;
+
+	__asm__ __volatile__(
+	"\n661:	or		%0, %2, %0\n"
+	"	.section	.sun4v_1insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	or		%0, %3, %0\n"
+	"	.previous\n"
+	: "=r" (off)
+	: "0" (off), "i" (_PAGE_FILE_4U), "i" (_PAGE_FILE_4V));
+
+	return __pte(off);
+}
 
-#define __S000	PAGE_NONE
-#define __S001	PAGE_READONLY_NOEXEC
-#define __S010	PAGE_SHARED_NOEXEC
-#define __S011	PAGE_SHARED_NOEXEC
-#define __S100	PAGE_READONLY
-#define __S101	PAGE_READONLY
-#define __S110	PAGE_SHARED
-#define __S111	PAGE_SHARED
+static inline pgprot_t pgprot_noncached(pgprot_t prot)
+{
+	unsigned long val = pgprot_val(prot);
+
+	__asm__ __volatile__(
+	"\n661:	andn		%0, %2, %0\n"
+	"	or		%0, %3, %0\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	andn		%0, %4, %0\n"
+	"	or		%0, %3, %0\n"
+	"	.previous\n"
+	: "=r" (val)
+	: "0" (val), "i" (_PAGE_CP_4U | _PAGE_CV_4U), "i" (_PAGE_E_4U),
+	             "i" (_PAGE_CP_4V | _PAGE_CV_4V), "i" (_PAGE_E_4V));
+
+	return __pgprot(val);
+}
+/* Various pieces of code check for platform support by ifdef testing
+ * on "pgprot_noncached".  That's broken and should be fixed, but for
+ * now...
+ */
+#define pgprot_noncached pgprot_noncached
 
-#ifndef __ASSEMBLY__
+#ifdef CONFIG_HUGETLB_PAGE
+static inline pte_t pte_mkhuge(pte_t pte)
+{
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	sethi		%%uhi(%1), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	mov		%2, %0\n"
+	"	nop\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_SZHUGE_4U), "i" (_PAGE_SZHUGE_4V));
+
+	return __pte(pte_val(pte) | mask);
+}
+#endif
 
-extern unsigned long phys_base;
-extern unsigned long pfn_base;
+static inline pte_t pte_mkdirty(pte_t pte)
+{
+	unsigned long val = pte_val(pte), tmp;
+
+	__asm__ __volatile__(
+	"\n661:	or		%0, %3, %0\n"
+	"	nop\n"
+	"\n662:	nop\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%4), %1\n"
+	"	sllx		%1, 32, %1\n"
+	"	.word		662b\n"
+	"	or		%1, %%lo(%4), %1\n"
+	"	or		%0, %1, %0\n"
+	"	.previous\n"
+	: "=r" (val), "=r" (tmp)
+	: "0" (val), "i" (_PAGE_MODIFIED_4U | _PAGE_W_4U),
+	  "i" (_PAGE_MODIFIED_4V | _PAGE_W_4V));
+
+	return __pte(val);
+}
 
-extern struct page *mem_map_zero;
-#define ZERO_PAGE(vaddr)	(mem_map_zero)
+static inline pte_t pte_mkclean(pte_t pte)
+{
+	unsigned long val = pte_val(pte), tmp;
+
+	__asm__ __volatile__(
+	"\n661:	andn		%0, %3, %0\n"
+	"	nop\n"
+	"\n662:	nop\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%4), %1\n"
+	"	sllx		%1, 32, %1\n"
+	"	.word		662b\n"
+	"	or		%1, %%lo(%4), %1\n"
+	"	andn		%0, %1, %0\n"
+	"	.previous\n"
+	: "=r" (val), "=r" (tmp)
+	: "0" (val), "i" (_PAGE_MODIFIED_4U | _PAGE_W_4U),
+	  "i" (_PAGE_MODIFIED_4V | _PAGE_W_4V));
+
+	return __pte(val);
+}
 
-/* PFNs are real physical page numbers.  However, mem_map only begins to record
- * per-page information starting at pfn_base.  This is to handle systems where
- * the first physical page in the machine is at some huge physical address,
- * such as 4GB.   This is common on a partitioned E10000, for example.
- */
+static inline pte_t pte_mkwrite(pte_t pte)
+{
+	unsigned long val = pte_val(pte), mask;
+
+	__asm__ __volatile__(
+	"\n661:	mov		%1, %0\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%2), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_WRITE_4U), "i" (_PAGE_WRITE_4V));
+
+	return __pte(val | mask);
+}
 
-#define pfn_pte(pfn, prot)	\
-	__pte(((pfn) << PAGE_SHIFT) | pgprot_val(prot) | _PAGE_SZBITS)
-#define mk_pte(page, pgprot)	pfn_pte(page_to_pfn(page), (pgprot))
+static inline pte_t pte_wrprotect(pte_t pte)
+{
+	unsigned long val = pte_val(pte), tmp;
+
+	__asm__ __volatile__(
+	"\n661:	andn		%0, %3, %0\n"
+	"	nop\n"
+	"\n662:	nop\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%4), %1\n"
+	"	sllx		%1, 32, %1\n"
+	"	.word		662b\n"
+	"	or		%1, %%lo(%4), %1\n"
+	"	andn		%0, %1, %0\n"
+	"	.previous\n"
+	: "=r" (val), "=r" (tmp)
+	: "0" (val), "i" (_PAGE_WRITE_4U | _PAGE_W_4U),
+	  "i" (_PAGE_WRITE_4V | _PAGE_W_4V));
+
+	return __pte(val);
+}
+
+static inline pte_t pte_mkold(pte_t pte)
+{
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	mov		%1, %0\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%2), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_ACCESSED_4U), "i" (_PAGE_ACCESSED_4V));
+
+	mask |= _PAGE_R;
+
+	return __pte(pte_val(pte) & ~mask);
+}
+
+static inline pte_t pte_mkyoung(pte_t pte)
+{
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	mov		%1, %0\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%2), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_ACCESSED_4U), "i" (_PAGE_ACCESSED_4V));
+
+	mask |= _PAGE_R;
+
+	return __pte(pte_val(pte) | mask);
+}
 
-#define pte_pfn(x)		((pte_val(x) & _PAGE_PADDR)>>PAGE_SHIFT)
-#define pte_page(x)		pfn_to_page(pte_pfn(x))
+static inline unsigned long pte_young(pte_t pte)
+{
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	mov		%1, %0\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%2), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_ACCESSED_4U), "i" (_PAGE_ACCESSED_4V));
+
+	return (pte_val(pte) & mask);
+}
+
+static inline unsigned long pte_dirty(pte_t pte)
+{
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	mov		%1, %0\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%2), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_MODIFIED_4U), "i" (_PAGE_MODIFIED_4V));
+
+	return (pte_val(pte) & mask);
+}
 
-static inline pte_t pte_modify(pte_t orig_pte, pgprot_t new_prot)
+static inline unsigned long pte_write(pte_t pte)
 {
-	pte_t __pte;
-	const unsigned long preserve_mask = (_PFN_MASK |
-					     _PAGE_MODIFIED | _PAGE_ACCESSED |
-					     _PAGE_CACHE | _PAGE_E |
-					     _PAGE_PRESENT | _PAGE_SZBITS);
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	mov		%1, %0\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%2), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_WRITE_4U), "i" (_PAGE_WRITE_4V));
+
+	return (pte_val(pte) & mask);
+}
 
-	pte_val(__pte) = (pte_val(orig_pte) & preserve_mask) |
-		(pgprot_val(new_prot) & ~preserve_mask);
+static inline unsigned long pte_exec(pte_t pte)
+{
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	sethi		%%hi(%1), %0\n"
+	"	.section	.sun4v_1insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	mov		%2, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_EXEC_4U), "i" (_PAGE_EXEC_4V));
+
+	return (pte_val(pte) & mask);
+}
 
-	return __pte;
+static inline unsigned long pte_read(pte_t pte)
+{
+	unsigned long mask;
+
+	__asm__ __volatile__(
+	"\n661:	mov		%1, %0\n"
+	"	nop\n"
+	"	.section	.sun4v_2insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	sethi		%%uhi(%2), %0\n"
+	"	sllx		%0, 32, %0\n"
+	"	.previous\n"
+	: "=r" (mask)
+	: "i" (_PAGE_READ_4U), "i" (_PAGE_READ_4V));
+
+	return (pte_val(pte) & mask);
 }
+
+static inline unsigned long pte_file(pte_t pte)
+{
+	unsigned long val = pte_val(pte);
+
+	__asm__ __volatile__(
+	"\n661:	and		%0, %2, %0\n"
+	"	.section	.sun4v_1insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	and		%0, %3, %0\n"
+	"	.previous\n"
+	: "=r" (val)
+	: "0" (val), "i" (_PAGE_FILE_4U), "i" (_PAGE_FILE_4V));
+
+	return val;
+}
+
+static inline unsigned long pte_present(pte_t pte)
+{
+	unsigned long val = pte_val(pte);
+
+	__asm__ __volatile__(
+	"\n661:	and		%0, %2, %0\n"
+	"	.section	.sun4v_1insn_patch, \"ax\"\n"
+	"	.word		661b\n"
+	"	and		%0, %3, %0\n"
+	"	.previous\n"
+	: "=r" (val)
+	: "0" (val), "i" (_PAGE_PRESENT_4U), "i" (_PAGE_PRESENT_4V));
+
+	return val;
+}
+
 #define pmd_set(pmdp, ptep)	\
 	(pmd_val(*(pmdp)) = (__pa((unsigned long) (ptep)) >> 11UL))
 #define pud_set(pudp, pmdp)	\
@@ -253,8 +631,6 @@ static inline pte_t pte_modify(pte_t orig_pte, pgprot_t new_prot)
 #define pmd_page(pmd) 			virt_to_page((void *)__pmd_page(pmd))
 #define pud_page(pud)		\
 	((unsigned long) __va((((unsigned long)pud_val(pud))<<11UL)))
-#define pte_none(pte) 			(!pte_val(pte))
-#define pte_present(pte)		(pte_val(pte) & _PAGE_PRESENT)
 #define pmd_none(pmd)			(!pmd_val(pmd))
 #define pmd_bad(pmd)			(0)
 #define pmd_present(pmd)		(pmd_val(pmd) != 0U)
@@ -264,30 +640,8 @@ static inline pte_t pte_modify(pte_t orig_pte, pgprot_t new_prot)
 #define pud_present(pud)		(pud_val(pud) != 0U)
 #define pud_clear(pudp)			(pud_val(*(pudp)) = 0U)
 
-/* The following only work if pte_present() is true.
- * Undefined behaviour if not..
- */
-#define pte_read(pte)		(pte_val(pte) & _PAGE_READ)
-#define pte_exec(pte)		(pte_val(pte) & _PAGE_EXEC)
-#define pte_write(pte)		(pte_val(pte) & _PAGE_WRITE)
-#define pte_dirty(pte)		(pte_val(pte) & _PAGE_MODIFIED)
-#define pte_young(pte)		(pte_val(pte) & _PAGE_ACCESSED)
-#define pte_wrprotect(pte)	(__pte(pte_val(pte) & ~(_PAGE_WRITE|_PAGE_W)))
-#define pte_rdprotect(pte)	\
-	(__pte(((pte_val(pte)<<1UL)>>1UL) & ~_PAGE_READ))
-#define pte_mkclean(pte)	\
-	(__pte(pte_val(pte) & ~(_PAGE_MODIFIED|_PAGE_W)))
-#define pte_mkold(pte)		\
-	(__pte(((pte_val(pte)<<1UL)>>1UL) & ~_PAGE_ACCESSED))
-
-/* Permanent address of a page. */
-#define __page_address(page)	page_address(page)
-
-/* Be very careful when you change these three, they are delicate. */
-#define pte_mkyoung(pte)	(__pte(pte_val(pte) | _PAGE_ACCESSED | _PAGE_R))
-#define pte_mkwrite(pte)	(__pte(pte_val(pte) | _PAGE_WRITE))
-#define pte_mkdirty(pte)	(__pte(pte_val(pte) | _PAGE_MODIFIED | _PAGE_W))
-#define pte_mkhuge(pte)		(__pte(pte_val(pte) | _PAGE_SZHUGE))
+/* Same in both SUN4V and SUN4U.  */
+#define pte_none(pte) 			(!pte_val(pte))
 
 /* to find an entry in a page-table-directory. */
 #define pgd_index(address)	(((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
@@ -296,11 +650,6 @@ static inline pte_t pte_modify(pte_t orig_pte, pgprot_t new_prot)
 /* to find an entry in a kernel page-table-directory */
 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
 
-/* extract the pgd cache used for optimizing the tlb miss
- * slow path when executing 32-bit compat processes
- */
-#define get_pgd_cache(pgd)	((unsigned long) pgd_val(*pgd) << 11)
-
 /* Find an entry in the second-level page table.. */
 #define pmd_offset(pudp, address)	\
 	((pmd_t *) pud_page(*(pudp)) + \
@@ -327,6 +676,9 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *p
 
 	/* It is more efficient to let flush_tlb_kernel_range()
 	 * handle init_mm tlb flushes.
+	 *
+	 * SUN4V NOTE: _PAGE_VALID is the same value in both the SUN4U
+	 *             and SUN4V pte layout, so this inline test is fine.
 	 */
 	if (likely(mm != &init_mm) && (pte_val(orig) & _PAGE_VALID))
 		tlb_batch_add(mm, addr, ptep, orig);
@@ -361,42 +713,23 @@ extern void update_mmu_cache(struct vm_area_struct *, unsigned long, pte_t);
 #define __swp_entry_to_pte(x)		((pte_t) { (x).val })
 
 /* File offset in PTE support. */
-#define pte_file(pte)		(pte_val(pte) & _PAGE_FILE)
+extern unsigned long pte_file(pte_t);
 #define pte_to_pgoff(pte)	(pte_val(pte) >> PAGE_SHIFT)
-#define pgoff_to_pte(off)	(__pte(((off) << PAGE_SHIFT) | _PAGE_FILE))
+extern pte_t pgoff_to_pte(unsigned long);
 #define PTE_FILE_MAX_BITS	(64UL - PAGE_SHIFT - 1UL)
 
 extern unsigned long prom_virt_to_phys(unsigned long, int *);
 
-static __inline__ unsigned long
-sun4u_get_pte (unsigned long addr)
-{
-	pgd_t *pgdp;
-	pud_t *pudp;
-	pmd_t *pmdp;
-	pte_t *ptep;
-
-	if (addr >= PAGE_OFFSET)
-		return addr & _PAGE_PADDR;
-	if ((addr >= LOW_OBP_ADDRESS) && (addr < HI_OBP_ADDRESS))
-		return prom_virt_to_phys(addr, NULL);
-	pgdp = pgd_offset_k(addr);
-	pudp = pud_offset(pgdp, addr);
-	pmdp = pmd_offset(pudp, addr);
-	ptep = pte_offset_kernel(pmdp, addr);
-	return pte_val(*ptep) & _PAGE_PADDR;
-}
+extern unsigned long sun4u_get_pte(unsigned long);
 
-static __inline__ unsigned long
-__get_phys (unsigned long addr)
+static inline unsigned long __get_phys(unsigned long addr)
 {
-	return sun4u_get_pte (addr);
+	return sun4u_get_pte(addr);
 }
 
-static __inline__ int
-__get_iospace (unsigned long addr)
+static inline int __get_iospace(unsigned long addr)
 {
-	return ((sun4u_get_pte (addr) & 0xf0000000) >> 28);
+	return ((sun4u_get_pte(addr) & 0xf0000000) >> 28);
 }
 
 extern unsigned long *sparc64_valid_addr_bitmap;
@@ -409,11 +742,6 @@ extern int io_remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
 			       unsigned long pfn,
 			       unsigned long size, pgprot_t prot);
 
-/* Clear virtual and physical cachability, set side-effect bit.  */
-#define pgprot_noncached(prot) \
-	(__pgprot((pgprot_val(prot) & ~(_PAGE_CP | _PAGE_CV)) | \
-	 _PAGE_E))
-
 /*
  * For sparc32&64, the pfn in io_remap_pfn_range() carries <iospace> in
  * its high 4 bits.  These macros/functions put it there or get it from there.
@@ -424,8 +752,11 @@ extern int io_remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
 
 #include <asm-generic/pgtable.h>
 
-/* We provide our own get_unmapped_area to cope with VA holes for userland */
+/* We provide our own get_unmapped_area to cope with VA holes and
+ * SHM area cache aliasing for userland.
+ */
 #define HAVE_ARCH_UNMAPPED_AREA
+#define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
 
 /* We provide a special get_unmapped_area for framebuffer mmaps to try and use
  * the largest alignment possible such that larget PTEs can be used.
@@ -435,12 +766,9 @@ extern unsigned long get_fb_unmapped_area(struct file *filp, unsigned long,
 					  unsigned long);
 #define HAVE_ARCH_FB_UNMAPPED_AREA
 
-/*
- * No page table caches to initialise
- */
-#define pgtable_cache_init()	do { } while (0)
-
-extern void check_pgt_cache(void);
+extern void pgtable_cache_init(void);
+extern void sun4v_register_fault_status(void);
+extern void sun4v_ktsb_register(void);
 
 #endif /* !(__ASSEMBLY__) */
 
diff --git a/include/asm-sparc64/pil.h b/include/asm-sparc64/pil.h
index 8f87750..79f827e 100644
--- a/include/asm-sparc64/pil.h
+++ b/include/asm-sparc64/pil.h
@@ -16,11 +16,13 @@
 #define PIL_SMP_CALL_FUNC	1
 #define PIL_SMP_RECEIVE_SIGNAL	2
 #define PIL_SMP_CAPTURE		3
+#define PIL_SMP_CTX_NEW_VERSION	4
 
 #ifndef __ASSEMBLY__
 #define PIL_RESERVED(PIL)	((PIL) == PIL_SMP_CALL_FUNC || \
 				 (PIL) == PIL_SMP_RECEIVE_SIGNAL || \
-				 (PIL) == PIL_SMP_CAPTURE)
+				 (PIL) == PIL_SMP_CAPTURE || \
+				 (PIL) == PIL_SMP_CTX_NEW_VERSION)
 #endif
 
 #endif /* !(_SPARC64_PIL_H) */
diff --git a/include/asm-sparc64/processor.h b/include/asm-sparc64/processor.h
index cd8d9b4..c6896b8 100644
--- a/include/asm-sparc64/processor.h
+++ b/include/asm-sparc64/processor.h
@@ -28,6 +28,8 @@
  * User lives in his very own context, and cannot reference us. Note
  * that TASK_SIZE is a misnomer, it really gives maximum user virtual 
  * address that the kernel will allocate out.
+ *
+ * XXX No longer using virtual page tables, kill this upper limit...
  */
 #define VA_BITS		44
 #ifndef __ASSEMBLY__
@@ -37,18 +39,6 @@
 #endif
 #define TASK_SIZE	((unsigned long)-VPTE_SIZE)
 
-/*
- * The vpte base must be able to hold the entire vpte, half
- * of which lives above, and half below, the base. And it
- * is placed as close to the highest address range as possible.
- */
-#define VPTE_BASE_SPITFIRE	(-(VPTE_SIZE/2))
-#if 1
-#define VPTE_BASE_CHEETAH	VPTE_BASE_SPITFIRE
-#else
-#define VPTE_BASE_CHEETAH	0xffe0000000000000
-#endif
-
 #ifndef __ASSEMBLY__
 
 typedef struct {
@@ -101,7 +91,8 @@ extern unsigned long thread_saved_pc(struct task_struct *);
 /* Do necessary setup to start up a newly executed thread. */
 #define start_thread(regs, pc, sp) \
 do { \
-	regs->tstate = (regs->tstate & (TSTATE_CWP)) | (TSTATE_INITIAL_MM|TSTATE_IE) | (ASI_PNF << 24); \
+	unsigned long __asi = ASI_PNF; \
+	regs->tstate = (regs->tstate & (TSTATE_CWP)) | (TSTATE_INITIAL_MM|TSTATE_IE) | (__asi << 24UL); \
 	regs->tpc = ((pc & (~3)) - 4); \
 	regs->tnpc = regs->tpc + 4; \
 	regs->y = 0; \
@@ -138,10 +129,10 @@ do { \
 
 #define start_thread32(regs, pc, sp) \
 do { \
+	unsigned long __asi = ASI_PNF; \
 	pc &= 0x00000000ffffffffUL; \
 	sp &= 0x00000000ffffffffUL; \
-\
-	regs->tstate = (regs->tstate & (TSTATE_CWP))|(TSTATE_INITIAL_MM|TSTATE_IE|TSTATE_AM); \
+	regs->tstate = (regs->tstate & (TSTATE_CWP))|(TSTATE_INITIAL_MM|TSTATE_IE|TSTATE_AM) | (__asi << 24UL); \
 	regs->tpc = ((pc & (~3)) - 4); \
 	regs->tnpc = regs->tpc + 4; \
 	regs->y = 0; \
@@ -226,6 +217,8 @@ static inline void prefetchw(const void *x)
 
 #define spin_lock_prefetch(x)	prefetchw(x)
 
+#define HAVE_ARCH_PICK_MMAP_LAYOUT
+
 #endif /* !(__ASSEMBLY__) */
 
 #endif /* !(__ASM_SPARC64_PROCESSOR_H) */
diff --git a/include/asm-sparc64/pstate.h b/include/asm-sparc64/pstate.h
index 29fb74a..49a7924 100644
--- a/include/asm-sparc64/pstate.h
+++ b/include/asm-sparc64/pstate.h
@@ -28,11 +28,12 @@
 
 /* The V9 TSTATE Register (with SpitFire and Linux extensions).
  *
- * ---------------------------------------------------------------
- * |  Resv  |  CCR  |  ASI  |  %pil  |  PSTATE  |  Resv  |  CWP  |
- * ---------------------------------------------------------------
- *  63    40 39   32 31   24 23    20 19       8 7      5 4     0
+ * ---------------------------------------------------------------------
+ * |  Resv |  GL  |  CCR  |  ASI  |  %pil  |  PSTATE  |  Resv  |  CWP  |
+ * ---------------------------------------------------------------------
+ *  63   43 42  40 39   32 31   24 23    20 19       8 7      5 4     0
  */
+#define TSTATE_GL	_AC(0x0000070000000000,UL) /* Global reg level  */
 #define TSTATE_CCR	_AC(0x000000ff00000000,UL) /* Condition Codes.	*/
 #define TSTATE_XCC	_AC(0x000000f000000000,UL) /* Condition Codes.	*/
 #define TSTATE_XNEG	_AC(0x0000008000000000,UL) /* %xcc Negative.	*/
diff --git a/include/asm-sparc64/scratchpad.h b/include/asm-sparc64/scratchpad.h
new file mode 100644
index 0000000..5e8b01f
--- /dev/null
+++ b/include/asm-sparc64/scratchpad.h
@@ -0,0 +1,14 @@
+#ifndef _SPARC64_SCRATCHPAD_H
+#define _SPARC64_SCRATCHPAD_H
+
+/* Sun4v scratchpad registers, accessed via ASI_SCRATCHPAD.  */
+
+#define SCRATCHPAD_MMU_MISS	0x00 /* Shared with OBP - set by OBP	    */
+#define SCRATCHPAD_CPUID	0x08 /* Shared with OBP - set by hypervisor */
+#define SCRATCHPAD_UTSBREG1	0x10
+#define SCRATCHPAD_UTSBREG2	0x18
+	/* 0x20 and 0x28, hypervisor only... */
+#define SCRATCHPAD_UNUSED1	0x30
+#define SCRATCHPAD_UNUSED2	0x38 /* Reserved for OBP		    */
+
+#endif /* !(_SPARC64_SCRATCHPAD_H) */
diff --git a/include/asm-sparc64/smp.h b/include/asm-sparc64/smp.h
index 473edb2..89d86ec 100644
--- a/include/asm-sparc64/smp.h
+++ b/include/asm-sparc64/smp.h
@@ -33,37 +33,13 @@
 extern cpumask_t phys_cpu_present_map;
 #define cpu_possible_map phys_cpu_present_map
 
+extern cpumask_t cpu_sibling_map[NR_CPUS];
+
 /*
  *	General functions that each host system must provide.
  */
 
-static __inline__ int hard_smp_processor_id(void)
-{
-	if (tlb_type == cheetah || tlb_type == cheetah_plus) {
-		unsigned long cfg, ver;
-		__asm__ __volatile__("rdpr %%ver, %0" : "=r" (ver));
-		if ((ver >> 32) == 0x003e0016) {
-			__asm__ __volatile__("ldxa [%%g0] %1, %0"
-					     : "=r" (cfg)
-					     : "i" (ASI_JBUS_CONFIG));
-			return ((cfg >> 17) & 0x1f);
-		} else {
-			__asm__ __volatile__("ldxa [%%g0] %1, %0"
-					     : "=r" (cfg)
-					     : "i" (ASI_SAFARI_CONFIG));
-			return ((cfg >> 17) & 0x3ff);
-		}
-	} else if (this_is_starfire != 0) {
-		return starfire_hard_smp_processor_id();
-	} else {
-		unsigned long upaconfig;
-		__asm__ __volatile__("ldxa	[%%g0] %1, %0"
-				     : "=r" (upaconfig)
-				     : "i" (ASI_UPA_CONFIG));
-		return ((upaconfig >> 17) & 0x1f);
-	}
-}
-
+extern int hard_smp_processor_id(void);
 #define raw_smp_processor_id() (current_thread_info()->cpu)
 
 extern void smp_setup_cpu_possible_map(void);
diff --git a/include/asm-sparc64/sparsemem.h b/include/asm-sparc64/sparsemem.h
new file mode 100644
index 0000000..ed5c9d8
--- /dev/null
+++ b/include/asm-sparc64/sparsemem.h
@@ -0,0 +1,12 @@
+#ifndef _SPARC64_SPARSEMEM_H
+#define _SPARC64_SPARSEMEM_H
+
+#ifdef __KERNEL__
+
+#define SECTION_SIZE_BITS       26
+#define MAX_PHYSADDR_BITS       42
+#define MAX_PHYSMEM_BITS        42
+
+#endif /* !(__KERNEL__) */
+
+#endif /* !(_SPARC64_SPARSEMEM_H) */
diff --git a/include/asm-sparc64/spitfire.h b/include/asm-sparc64/spitfire.h
index 962638c..23ad8a7 100644
--- a/include/asm-sparc64/spitfire.h
+++ b/include/asm-sparc64/spitfire.h
@@ -44,6 +44,7 @@ enum ultra_tlb_layout {
 	spitfire = 0,
 	cheetah = 1,
 	cheetah_plus = 2,
+	hypervisor = 3,
 };
 
 extern enum ultra_tlb_layout tlb_type;
diff --git a/include/asm-sparc64/system.h b/include/asm-sparc64/system.h
index af254e5..a18ec87 100644
--- a/include/asm-sparc64/system.h
+++ b/include/asm-sparc64/system.h
@@ -209,9 +209,10 @@ do {	if (test_thread_flag(TIF_PERFCTR)) {				\
 	/* so that ASI is only written if it changes, think again. */	\
 	__asm__ __volatile__("wr %%g0, %0, %%asi"			\
 	: : "r" (__thread_flag_byte_ptr(task_thread_info(next))[TI_FLAG_BYTE_CURRENT_DS]));\
+	trap_block[current_thread_info()->cpu].thread =			\
+		task_thread_info(next);					\
 	__asm__ __volatile__(						\
 	"mov	%%g4, %%g7\n\t"						\
-	"wrpr	%%g0, 0x95, %%pstate\n\t"				\
 	"stx	%%i6, [%%sp + 2047 + 0x70]\n\t"				\
 	"stx	%%i7, [%%sp + 2047 + 0x78]\n\t"				\
 	"rdpr	%%wstate, %%o5\n\t"					\
@@ -225,14 +226,10 @@ do {	if (test_thread_flag(TIF_PERFCTR)) {				\
 	"ldx	[%%g6 + %3], %%o6\n\t"					\
 	"ldub	[%%g6 + %2], %%o5\n\t"					\
 	"ldub	[%%g6 + %4], %%o7\n\t"					\
-	"mov	%%g6, %%l2\n\t"						\
 	"wrpr	%%o5, 0x0, %%wstate\n\t"				\
 	"ldx	[%%sp + 2047 + 0x70], %%i6\n\t"				\
 	"ldx	[%%sp + 2047 + 0x78], %%i7\n\t"				\
-	"wrpr	%%g0, 0x94, %%pstate\n\t"				\
-	"mov	%%l2, %%g6\n\t"						\
 	"ldx	[%%g6 + %6], %%g4\n\t"					\
-	"wrpr	%%g0, 0x96, %%pstate\n\t"				\
 	"brz,pt %%o7, 1f\n\t"						\
 	" mov	%%g7, %0\n\t"						\
 	"b,a ret_from_syscall\n\t"					\
diff --git a/include/asm-sparc64/thread_info.h b/include/asm-sparc64/thread_info.h
index ac9d068..2ebf7f2 100644
--- a/include/asm-sparc64/thread_info.h
+++ b/include/asm-sparc64/thread_info.h
@@ -64,8 +64,6 @@ struct thread_info {
 	__u64			kernel_cntd0, kernel_cntd1;
 	__u64			pcr_reg;
 
-	__u64			cee_stuff;
-
 	struct restart_block	restart_block;
 
 	struct pt_regs		*kern_una_regs;
@@ -104,10 +102,9 @@ struct thread_info {
 #define TI_KERN_CNTD0	0x00000480
 #define TI_KERN_CNTD1	0x00000488
 #define TI_PCR		0x00000490
-#define TI_CEE_STUFF	0x00000498
-#define TI_RESTART_BLOCK 0x000004a0
-#define TI_KUNA_REGS	0x000004c8
-#define TI_KUNA_INSN	0x000004d0
+#define TI_RESTART_BLOCK 0x00000498
+#define TI_KUNA_REGS	0x000004c0
+#define TI_KUNA_INSN	0x000004c8
 #define TI_FPREGS	0x00000500
 
 /* We embed this in the uppermost byte of thread_info->flags */
diff --git a/include/asm-sparc64/timex.h b/include/asm-sparc64/timex.h
index 9e8d417..2a5e4eb 100644
--- a/include/asm-sparc64/timex.h
+++ b/include/asm-sparc64/timex.h
@@ -14,4 +14,10 @@
 typedef unsigned long cycles_t;
 #define get_cycles()	tick_ops->get_tick()
 
+#define ARCH_HAS_READ_CURRENT_TIMER	1
+#define read_current_timer(timer_val_p) 	\
+({	*timer_val_p = tick_ops->get_tick();	\
+	0;					\
+})
+
 #endif
diff --git a/include/asm-sparc64/tlbflush.h b/include/asm-sparc64/tlbflush.h
index 3ef9909..9ad5d9c 100644
--- a/include/asm-sparc64/tlbflush.h
+++ b/include/asm-sparc64/tlbflush.h
@@ -5,6 +5,11 @@
 #include <linux/mm.h>
 #include <asm/mmu_context.h>
 
+/* TSB flush operations. */
+struct mmu_gather;
+extern void flush_tsb_kernel_range(unsigned long start, unsigned long end);
+extern void flush_tsb_user(struct mmu_gather *mp);
+
 /* TLB flush operations. */
 
 extern void flush_tlb_pending(void);
@@ -14,28 +19,36 @@ extern void flush_tlb_pending(void);
 #define flush_tlb_page(vma,addr)	flush_tlb_pending()
 #define flush_tlb_mm(mm)		flush_tlb_pending()
 
+/* Local cpu only.  */
 extern void __flush_tlb_all(void);
+
 extern void __flush_tlb_page(unsigned long context, unsigned long page, unsigned long r);
 
 extern void __flush_tlb_kernel_range(unsigned long start, unsigned long end);
 
 #ifndef CONFIG_SMP
 
-#define flush_tlb_all()		__flush_tlb_all()
 #define flush_tlb_kernel_range(start,end) \
-	__flush_tlb_kernel_range(start,end)
+do {	flush_tsb_kernel_range(start,end); \
+	__flush_tlb_kernel_range(start,end); \
+} while (0)
 
 #else /* CONFIG_SMP */
 
-extern void smp_flush_tlb_all(void);
 extern void smp_flush_tlb_kernel_range(unsigned long start, unsigned long end);
 
-#define flush_tlb_all()		smp_flush_tlb_all()
 #define flush_tlb_kernel_range(start, end) \
-	smp_flush_tlb_kernel_range(start, end)
+do {	flush_tsb_kernel_range(start,end); \
+	smp_flush_tlb_kernel_range(start, end); \
+} while (0)
 
 #endif /* ! CONFIG_SMP */
 
-extern void flush_tlb_pgtables(struct mm_struct *, unsigned long, unsigned long);
+static inline void flush_tlb_pgtables(struct mm_struct *mm, unsigned long start, unsigned long end)
+{
+	/* We don't use virtual page tables for TLB miss processing
+	 * any more.  Nowadays we use the TSB.
+	 */
+}
 
 #endif /* _SPARC64_TLBFLUSH_H */
diff --git a/include/asm-sparc64/tsb.h b/include/asm-sparc64/tsb.h
new file mode 100644
index 0000000..e82612c
--- /dev/null
+++ b/include/asm-sparc64/tsb.h
@@ -0,0 +1,281 @@
+#ifndef _SPARC64_TSB_H
+#define _SPARC64_TSB_H
+
+/* The sparc64 TSB is similar to the powerpc hashtables.  It's a
+ * power-of-2 sized table of TAG/PTE pairs.  The cpu precomputes
+ * pointers into this table for 8K and 64K page sizes, and also a
+ * comparison TAG based upon the virtual address and context which
+ * faults.
+ *
+ * TLB miss trap handler software does the actual lookup via something
+ * of the form:
+ *
+ * 	ldxa		[%g0] ASI_{D,I}MMU_TSB_8KB_PTR, %g1
+ * 	ldxa		[%g0] ASI_{D,I}MMU, %g6
+ *	sllx		%g6, 22, %g6
+ *	srlx		%g6, 22, %g6
+ * 	ldda		[%g1] ASI_NUCLEUS_QUAD_LDD, %g4
+ * 	cmp		%g4, %g6
+ * 	bne,pn	%xcc, tsb_miss_{d,i}tlb
+ * 	 mov		FAULT_CODE_{D,I}TLB, %g3
+ * 	stxa		%g5, [%g0] ASI_{D,I}TLB_DATA_IN
+ * 	retry
+ *
+ *
+ * Each 16-byte slot of the TSB is the 8-byte tag and then the 8-byte
+ * PTE.  The TAG is of the same layout as the TLB TAG TARGET mmu
+ * register which is:
+ *
+ * -------------------------------------------------
+ * |  -  |  CONTEXT |  -  |    VADDR bits 63:22    |
+ * -------------------------------------------------
+ *  63 61 60      48 47 42 41                     0
+ *
+ * But actually, since we use per-mm TSB's, we zero out the CONTEXT
+ * field.
+ *
+ * Like the powerpc hashtables we need to use locking in order to
+ * synchronize while we update the entries.  PTE updates need locking
+ * as well.
+ *
+ * We need to carefully choose a lock bits for the TSB entry.  We
+ * choose to use bit 47 in the tag.  Also, since we never map anything
+ * at page zero in context zero, we use zero as an invalid tag entry.
+ * When the lock bit is set, this forces a tag comparison failure.
+ */
+
+#define TSB_TAG_LOCK_BIT	47
+#define TSB_TAG_LOCK_HIGH	(1 << (TSB_TAG_LOCK_BIT - 32))
+
+#define TSB_TAG_INVALID_BIT	46
+#define TSB_TAG_INVALID_HIGH	(1 << (TSB_TAG_INVALID_BIT - 32))
+
+#define TSB_MEMBAR	membar	#StoreStore
+
+/* Some cpus support physical address quad loads.  We want to use
+ * those if possible so we don't need to hard-lock the TSB mapping
+ * into the TLB.  We encode some instruction patching in order to
+ * support this.
+ *
+ * The kernel TSB is locked into the TLB by virtue of being in the
+ * kernel image, so we don't play these games for swapper_tsb access.
+ */
+#ifndef __ASSEMBLY__
+struct tsb_ldquad_phys_patch_entry {
+	unsigned int	addr;
+	unsigned int	sun4u_insn;
+	unsigned int	sun4v_insn;
+};
+extern struct tsb_ldquad_phys_patch_entry __tsb_ldquad_phys_patch,
+	__tsb_ldquad_phys_patch_end;
+
+struct tsb_phys_patch_entry {
+	unsigned int	addr;
+	unsigned int	insn;
+};
+extern struct tsb_phys_patch_entry __tsb_phys_patch, __tsb_phys_patch_end;
+#endif
+#define TSB_LOAD_QUAD(TSB, REG)	\
+661:	ldda		[TSB] ASI_NUCLEUS_QUAD_LDD, REG; \
+	.section	.tsb_ldquad_phys_patch, "ax"; \
+	.word		661b; \
+	ldda		[TSB] ASI_QUAD_LDD_PHYS, REG; \
+	ldda		[TSB] ASI_QUAD_LDD_PHYS_4V, REG; \
+	.previous
+
+#define TSB_LOAD_TAG_HIGH(TSB, REG) \
+661:	lduwa		[TSB] ASI_N, REG; \
+	.section	.tsb_phys_patch, "ax"; \
+	.word		661b; \
+	lduwa		[TSB] ASI_PHYS_USE_EC, REG; \
+	.previous
+
+#define TSB_LOAD_TAG(TSB, REG) \
+661:	ldxa		[TSB] ASI_N, REG; \
+	.section	.tsb_phys_patch, "ax"; \
+	.word		661b; \
+	ldxa		[TSB] ASI_PHYS_USE_EC, REG; \
+	.previous
+
+#define TSB_CAS_TAG_HIGH(TSB, REG1, REG2) \
+661:	casa		[TSB] ASI_N, REG1, REG2; \
+	.section	.tsb_phys_patch, "ax"; \
+	.word		661b; \
+	casa		[TSB] ASI_PHYS_USE_EC, REG1, REG2; \
+	.previous
+
+#define TSB_CAS_TAG(TSB, REG1, REG2) \
+661:	casxa		[TSB] ASI_N, REG1, REG2; \
+	.section	.tsb_phys_patch, "ax"; \
+	.word		661b; \
+	casxa		[TSB] ASI_PHYS_USE_EC, REG1, REG2; \
+	.previous
+
+#define TSB_STORE(ADDR, VAL) \
+661:	stxa		VAL, [ADDR] ASI_N; \
+	.section	.tsb_phys_patch, "ax"; \
+	.word		661b; \
+	stxa		VAL, [ADDR] ASI_PHYS_USE_EC; \
+	.previous
+
+#define TSB_LOCK_TAG(TSB, REG1, REG2)	\
+99:	TSB_LOAD_TAG_HIGH(TSB, REG1);	\
+	sethi	%hi(TSB_TAG_LOCK_HIGH), REG2;\
+	andcc	REG1, REG2, %g0;	\
+	bne,pn	%icc, 99b;		\
+	 nop;				\
+	TSB_CAS_TAG_HIGH(TSB, REG1, REG2);	\
+	cmp	REG1, REG2;		\
+	bne,pn	%icc, 99b;		\
+	 nop;				\
+	TSB_MEMBAR
+
+#define TSB_WRITE(TSB, TTE, TAG) \
+	add	TSB, 0x8, TSB;   \
+	TSB_STORE(TSB, TTE);     \
+	sub	TSB, 0x8, TSB;   \
+	TSB_MEMBAR;              \
+	TSB_STORE(TSB, TAG);
+
+#define KTSB_LOAD_QUAD(TSB, REG) \
+	ldda		[TSB] ASI_NUCLEUS_QUAD_LDD, REG;
+
+#define KTSB_STORE(ADDR, VAL) \
+	stxa		VAL, [ADDR] ASI_N;
+
+#define KTSB_LOCK_TAG(TSB, REG1, REG2)	\
+99:	lduwa	[TSB] ASI_N, REG1;	\
+	sethi	%hi(TSB_TAG_LOCK_HIGH), REG2;\
+	andcc	REG1, REG2, %g0;	\
+	bne,pn	%icc, 99b;		\
+	 nop;				\
+	casa	[TSB] ASI_N, REG1, REG2;\
+	cmp	REG1, REG2;		\
+	bne,pn	%icc, 99b;		\
+	 nop;				\
+	TSB_MEMBAR
+
+#define KTSB_WRITE(TSB, TTE, TAG) \
+	add	TSB, 0x8, TSB;   \
+	stxa	TTE, [TSB] ASI_N;     \
+	sub	TSB, 0x8, TSB;   \
+	TSB_MEMBAR;              \
+	stxa	TAG, [TSB] ASI_N;
+
+	/* Do a kernel page table walk.  Leaves physical PTE pointer in
+	 * REG1.  Jumps to FAIL_LABEL on early page table walk termination.
+	 * VADDR will not be clobbered, but REG2 will.
+	 */
+#define KERN_PGTABLE_WALK(VADDR, REG1, REG2, FAIL_LABEL)	\
+	sethi		%hi(swapper_pg_dir), REG1; \
+	or		REG1, %lo(swapper_pg_dir), REG1; \
+	sllx		VADDR, 64 - (PGDIR_SHIFT + PGDIR_BITS), REG2; \
+	srlx		REG2, 64 - PAGE_SHIFT, REG2; \
+	andn		REG2, 0x3, REG2; \
+	lduw		[REG1 + REG2], REG1; \
+	brz,pn		REG1, FAIL_LABEL; \
+	 sllx		VADDR, 64 - (PMD_SHIFT + PMD_BITS), REG2; \
+	srlx		REG2, 64 - PAGE_SHIFT, REG2; \
+	sllx		REG1, 11, REG1; \
+	andn		REG2, 0x3, REG2; \
+	lduwa		[REG1 + REG2] ASI_PHYS_USE_EC, REG1; \
+	brz,pn		REG1, FAIL_LABEL; \
+	 sllx		VADDR, 64 - PMD_SHIFT, REG2; \
+	srlx		REG2, 64 - PAGE_SHIFT, REG2; \
+	sllx		REG1, 11, REG1; \
+	andn		REG2, 0x7, REG2; \
+	add		REG1, REG2, REG1;
+
+	/* Do a user page table walk in MMU globals.  Leaves physical PTE
+	 * pointer in REG1.  Jumps to FAIL_LABEL on early page table walk
+	 * termination.  Physical base of page tables is in PHYS_PGD which
+	 * will not be modified.
+	 *
+	 * VADDR will not be clobbered, but REG1 and REG2 will.
+	 */
+#define USER_PGTABLE_WALK_TL1(VADDR, PHYS_PGD, REG1, REG2, FAIL_LABEL)	\
+	sllx		VADDR, 64 - (PGDIR_SHIFT + PGDIR_BITS), REG2; \
+	srlx		REG2, 64 - PAGE_SHIFT, REG2; \
+	andn		REG2, 0x3, REG2; \
+	lduwa		[PHYS_PGD + REG2] ASI_PHYS_USE_EC, REG1; \
+	brz,pn		REG1, FAIL_LABEL; \
+	 sllx		VADDR, 64 - (PMD_SHIFT + PMD_BITS), REG2; \
+	srlx		REG2, 64 - PAGE_SHIFT, REG2; \
+	sllx		REG1, 11, REG1; \
+	andn		REG2, 0x3, REG2; \
+	lduwa		[REG1 + REG2] ASI_PHYS_USE_EC, REG1; \
+	brz,pn		REG1, FAIL_LABEL; \
+	 sllx		VADDR, 64 - PMD_SHIFT, REG2; \
+	srlx		REG2, 64 - PAGE_SHIFT, REG2; \
+	sllx		REG1, 11, REG1; \
+	andn		REG2, 0x7, REG2; \
+	add		REG1, REG2, REG1;
+
+/* Lookup a OBP mapping on VADDR in the prom_trans[] table at TL>0.
+ * If no entry is found, FAIL_LABEL will be branched to.  On success
+ * the resulting PTE value will be left in REG1.  VADDR is preserved
+ * by this routine.
+ */
+#define OBP_TRANS_LOOKUP(VADDR, REG1, REG2, REG3, FAIL_LABEL) \
+	sethi		%hi(prom_trans), REG1; \
+	or		REG1, %lo(prom_trans), REG1; \
+97:	ldx		[REG1 + 0x00], REG2; \
+	brz,pn		REG2, FAIL_LABEL; \
+	 nop; \
+	ldx		[REG1 + 0x08], REG3; \
+	add		REG2, REG3, REG3; \
+	cmp		REG2, VADDR; \
+	bgu,pt		%xcc, 98f; \
+	 cmp		VADDR, REG3; \
+	bgeu,pt		%xcc, 98f; \
+	 ldx		[REG1 + 0x10], REG3; \
+	sub		VADDR, REG2, REG2; \
+	ba,pt		%xcc, 99f; \
+	 add		REG3, REG2, REG1; \
+98:	ba,pt		%xcc, 97b; \
+	 add		REG1, (3 * 8), REG1; \
+99:
+
+	/* We use a 32K TSB for the whole kernel, this allows to
+	 * handle about 16MB of modules and vmalloc mappings without
+	 * incurring many hash conflicts.
+	 */
+#define KERNEL_TSB_SIZE_BYTES	(32 * 1024)
+#define KERNEL_TSB_NENTRIES	\
+	(KERNEL_TSB_SIZE_BYTES / 16)
+#define KERNEL_TSB4M_NENTRIES	4096
+
+	/* Do a kernel TSB lookup at tl>0 on VADDR+TAG, branch to OK_LABEL
+	 * on TSB hit.  REG1, REG2, REG3, and REG4 are used as temporaries
+	 * and the found TTE will be left in REG1.  REG3 and REG4 must
+	 * be an even/odd pair of registers.
+	 *
+	 * VADDR and TAG will be preserved and not clobbered by this macro.
+	 */
+#define KERN_TSB_LOOKUP_TL1(VADDR, TAG, REG1, REG2, REG3, REG4, OK_LABEL) \
+	sethi		%hi(swapper_tsb), REG1; \
+	or		REG1, %lo(swapper_tsb), REG1; \
+	srlx		VADDR, PAGE_SHIFT, REG2; \
+	and		REG2, (KERNEL_TSB_NENTRIES - 1), REG2; \
+	sllx		REG2, 4, REG2; \
+	add		REG1, REG2, REG2; \
+	KTSB_LOAD_QUAD(REG2, REG3); \
+	cmp		REG3, TAG; \
+	be,a,pt		%xcc, OK_LABEL; \
+	 mov		REG4, REG1;
+
+	/* This version uses a trick, the TAG is already (VADDR >> 22) so
+	 * we can make use of that for the index computation.
+	 */
+#define KERN_TSB4M_LOOKUP_TL1(TAG, REG1, REG2, REG3, REG4, OK_LABEL) \
+	sethi		%hi(swapper_4m_tsb), REG1; \
+	or		REG1, %lo(swapper_4m_tsb), REG1; \
+	and		TAG, (KERNEL_TSB_NENTRIES - 1), REG2; \
+	sllx		REG2, 4, REG2; \
+	add		REG1, REG2, REG2; \
+	KTSB_LOAD_QUAD(REG2, REG3); \
+	cmp		REG3, TAG; \
+	be,a,pt		%xcc, OK_LABEL; \
+	 mov		REG4, REG1;
+
+#endif /* !(_SPARC64_TSB_H) */
diff --git a/include/asm-sparc64/ttable.h b/include/asm-sparc64/ttable.h
index 2784f80..2d5e3c4 100644
--- a/include/asm-sparc64/ttable.h
+++ b/include/asm-sparc64/ttable.h
@@ -93,7 +93,7 @@
 	
 #define SYSCALL_TRAP(routine, systbl)			\
 	sethi	%hi(109f), %g7;				\
-	ba,pt	%xcc, scetrap;				\
+	ba,pt	%xcc, etrap;				\
 109:	 or	%g7, %lo(109b), %g7;			\
 	sethi	%hi(systbl), %l7;			\
 	ba,pt	%xcc, routine;				\
@@ -109,14 +109,14 @@
 	nop;nop;nop;
 	
 #define TRAP_UTRAP(handler,lvl)				\
-	ldx	[%g6 + TI_UTRAPS], %g1;			\
-	sethi	%hi(109f), %g7;				\
-	brz,pn	%g1, utrap;				\
-	 or	%g7, %lo(109f), %g7;			\
-	ba,pt	%xcc, utrap;				\
-109:	 ldx	[%g1 + handler*8], %g1;			\
-	ba,pt	%xcc, utrap_ill;			\
-	 mov	lvl, %o1;
+	mov	handler, %g3;				\
+	ba,pt	%xcc, utrap_trap;			\
+	 mov	lvl, %g4;				\
+	nop;						\
+	nop;						\
+	nop;						\
+	nop;						\
+	nop;
 
 #ifdef CONFIG_SUNOS_EMUL
 #define SUNOS_SYSCALL_TRAP SYSCALL_TRAP(linux_sparc_syscall32, sunos_sys_table)
@@ -136,8 +136,6 @@
 #else
 #define SOLARIS_SYSCALL_TRAP TRAP(solaris_syscall)
 #endif
-/* FIXME: Write these actually */	
-#define NETBSD_SYSCALL_TRAP TRAP(netbsd_syscall)
 #define BREAKPOINT_TRAP TRAP(breakpoint_trap)
 
 #define TRAP_IRQ(routine, level)			\
@@ -182,6 +180,26 @@
 #define KPROBES_TRAP(lvl) TRAP_ARG(bad_trap, lvl)
 #endif
 
+#define SUN4V_ITSB_MISS					\
+	ldxa	[%g0] ASI_SCRATCHPAD, %g2;		\
+	ldx	[%g2 + HV_FAULT_I_ADDR_OFFSET], %g4;	\
+	ldx	[%g2 + HV_FAULT_I_CTX_OFFSET], %g5;	\
+	srlx	%g4, 22, %g6;				\
+	ba,pt	%xcc, sun4v_itsb_miss;			\
+	 nop;						\
+	nop;						\
+	nop;
+
+#define SUN4V_DTSB_MISS					\
+	ldxa	[%g0] ASI_SCRATCHPAD, %g2;		\
+	ldx	[%g2 + HV_FAULT_D_ADDR_OFFSET], %g4;	\
+	ldx	[%g2 + HV_FAULT_D_CTX_OFFSET], %g5;	\
+	srlx	%g4, 22, %g6;				\
+	ba,pt	%xcc, sun4v_dtsb_miss;			\
+	 nop;						\
+	nop;						\
+	nop;
+
 /* Before touching these macros, you owe it to yourself to go and
  * see how arch/sparc64/kernel/winfixup.S works... -DaveM
  *
@@ -221,6 +239,31 @@
 	saved; retry; nop; nop; nop; nop; nop; nop;	\
 	nop; nop; nop; nop; nop; nop; nop; nop;
 
+#define SPILL_0_NORMAL_ETRAP				\
+etrap_kernel_spill:					\
+	stx	%l0, [%sp + STACK_BIAS + 0x00];		\
+	stx	%l1, [%sp + STACK_BIAS + 0x08];		\
+	stx	%l2, [%sp + STACK_BIAS + 0x10];		\
+	stx	%l3, [%sp + STACK_BIAS + 0x18];		\
+	stx	%l4, [%sp + STACK_BIAS + 0x20];		\
+	stx	%l5, [%sp + STACK_BIAS + 0x28];		\
+	stx	%l6, [%sp + STACK_BIAS + 0x30];		\
+	stx	%l7, [%sp + STACK_BIAS + 0x38];		\
+	stx	%i0, [%sp + STACK_BIAS + 0x40];		\
+	stx	%i1, [%sp + STACK_BIAS + 0x48];		\
+	stx	%i2, [%sp + STACK_BIAS + 0x50];		\
+	stx	%i3, [%sp + STACK_BIAS + 0x58];		\
+	stx	%i4, [%sp + STACK_BIAS + 0x60];		\
+	stx	%i5, [%sp + STACK_BIAS + 0x68];		\
+	stx	%i6, [%sp + STACK_BIAS + 0x70];		\
+	stx	%i7, [%sp + STACK_BIAS + 0x78];		\
+	saved;						\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	wrpr	%g1, %cwp;				\
+	nop; nop; nop; nop; nop; nop; nop; nop;		\
+	nop; nop; nop; nop;
+
 /* Normal 64bit spill */
 #define SPILL_1_GENERIC(ASI)				\
 	add	%sp, STACK_BIAS + 0x00, %g1;		\
@@ -254,6 +297,67 @@
 	b,a,pt	%xcc, spill_fixup_mna;			\
 	b,a,pt	%xcc, spill_fixup;
 
+#define SPILL_1_GENERIC_ETRAP				\
+etrap_user_spill_64bit:					\
+	stxa	%l0, [%sp + STACK_BIAS + 0x00] %asi;	\
+	stxa	%l1, [%sp + STACK_BIAS + 0x08] %asi;	\
+	stxa	%l2, [%sp + STACK_BIAS + 0x10] %asi;	\
+	stxa	%l3, [%sp + STACK_BIAS + 0x18] %asi;	\
+	stxa	%l4, [%sp + STACK_BIAS + 0x20] %asi;	\
+	stxa	%l5, [%sp + STACK_BIAS + 0x28] %asi;	\
+	stxa	%l6, [%sp + STACK_BIAS + 0x30] %asi;	\
+	stxa	%l7, [%sp + STACK_BIAS + 0x38] %asi;	\
+	stxa	%i0, [%sp + STACK_BIAS + 0x40] %asi;	\
+	stxa	%i1, [%sp + STACK_BIAS + 0x48] %asi;	\
+	stxa	%i2, [%sp + STACK_BIAS + 0x50] %asi;	\
+	stxa	%i3, [%sp + STACK_BIAS + 0x58] %asi;	\
+	stxa	%i4, [%sp + STACK_BIAS + 0x60] %asi;	\
+	stxa	%i5, [%sp + STACK_BIAS + 0x68] %asi;	\
+	stxa	%i6, [%sp + STACK_BIAS + 0x70] %asi;	\
+	stxa	%i7, [%sp + STACK_BIAS + 0x78] %asi;	\
+	saved;						\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop;				\
+	ba,a,pt	%xcc, etrap_spill_fixup_64bit;		\
+	ba,a,pt	%xcc, etrap_spill_fixup_64bit;		\
+	ba,a,pt	%xcc, etrap_spill_fixup_64bit;
+
+#define SPILL_1_GENERIC_ETRAP_FIXUP			\
+etrap_spill_fixup_64bit:				\
+	ldub	[%g6 + TI_WSAVED], %g1;			\
+	sll	%g1, 3, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stx	%sp, [%g3 + TI_RWIN_SPTRS];		\
+	sll	%g1, 7, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stx	%l0, [%g3 + TI_REG_WINDOW + 0x00];	\
+	stx	%l1, [%g3 + TI_REG_WINDOW + 0x08];	\
+	stx	%l2, [%g3 + TI_REG_WINDOW + 0x10];	\
+	stx	%l3, [%g3 + TI_REG_WINDOW + 0x18];	\
+	stx	%l4, [%g3 + TI_REG_WINDOW + 0x20];	\
+	stx	%l5, [%g3 + TI_REG_WINDOW + 0x28];	\
+	stx	%l6, [%g3 + TI_REG_WINDOW + 0x30];	\
+	stx	%l7, [%g3 + TI_REG_WINDOW + 0x38];	\
+	stx	%i0, [%g3 + TI_REG_WINDOW + 0x40];	\
+	stx	%i1, [%g3 + TI_REG_WINDOW + 0x48];	\
+	stx	%i2, [%g3 + TI_REG_WINDOW + 0x50];	\
+	stx	%i3, [%g3 + TI_REG_WINDOW + 0x58];	\
+	stx	%i4, [%g3 + TI_REG_WINDOW + 0x60];	\
+	stx	%i5, [%g3 + TI_REG_WINDOW + 0x68];	\
+	stx	%i6, [%g3 + TI_REG_WINDOW + 0x70];	\
+	stx	%i7, [%g3 + TI_REG_WINDOW + 0x78];	\
+	add	%g1, 1, %g1;				\
+	stb	%g1, [%g6 + TI_WSAVED];			\
+	saved;						\
+	rdpr	%cwp, %g1;				\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop
+
 /* Normal 32bit spill */
 #define SPILL_2_GENERIC(ASI)				\
 	srl	%sp, 0, %sp;				\
@@ -287,6 +391,68 @@
 	b,a,pt	%xcc, spill_fixup_mna;			\
 	b,a,pt	%xcc, spill_fixup;
 
+#define SPILL_2_GENERIC_ETRAP		\
+etrap_user_spill_32bit:			\
+	srl	%sp, 0, %sp;		\
+	stwa	%l0, [%sp + 0x00] %asi;	\
+	stwa	%l1, [%sp + 0x04] %asi;	\
+	stwa	%l2, [%sp + 0x08] %asi;	\
+	stwa	%l3, [%sp + 0x0c] %asi;	\
+	stwa	%l4, [%sp + 0x10] %asi;	\
+	stwa	%l5, [%sp + 0x14] %asi;	\
+	stwa	%l6, [%sp + 0x18] %asi;	\
+	stwa	%l7, [%sp + 0x1c] %asi;	\
+	stwa	%i0, [%sp + 0x20] %asi;	\
+	stwa	%i1, [%sp + 0x24] %asi;	\
+	stwa	%i2, [%sp + 0x28] %asi;	\
+	stwa	%i3, [%sp + 0x2c] %asi;	\
+	stwa	%i4, [%sp + 0x30] %asi;	\
+	stwa	%i5, [%sp + 0x34] %asi;	\
+	stwa	%i6, [%sp + 0x38] %asi;	\
+	stwa	%i7, [%sp + 0x3c] %asi;	\
+	saved;				\
+	sub	%g1, 2, %g1;		\
+	ba,pt	%xcc, etrap_save;	\
+	 wrpr	%g1, %cwp;		\
+	nop; nop; nop; nop;		\
+	nop; nop; nop; nop;		\
+	ba,a,pt	%xcc, etrap_spill_fixup_32bit; \
+	ba,a,pt	%xcc, etrap_spill_fixup_32bit; \
+	ba,a,pt	%xcc, etrap_spill_fixup_32bit;
+
+#define SPILL_2_GENERIC_ETRAP_FIXUP			\
+etrap_spill_fixup_32bit:				\
+	ldub	[%g6 + TI_WSAVED], %g1;			\
+	sll	%g1, 3, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stx	%sp, [%g3 + TI_RWIN_SPTRS];		\
+	sll	%g1, 7, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stw	%l0, [%g3 + TI_REG_WINDOW + 0x00];	\
+	stw	%l1, [%g3 + TI_REG_WINDOW + 0x04];	\
+	stw	%l2, [%g3 + TI_REG_WINDOW + 0x08];	\
+	stw	%l3, [%g3 + TI_REG_WINDOW + 0x0c];	\
+	stw	%l4, [%g3 + TI_REG_WINDOW + 0x10];	\
+	stw	%l5, [%g3 + TI_REG_WINDOW + 0x14];	\
+	stw	%l6, [%g3 + TI_REG_WINDOW + 0x18];	\
+	stw	%l7, [%g3 + TI_REG_WINDOW + 0x1c];	\
+	stw	%i0, [%g3 + TI_REG_WINDOW + 0x20];	\
+	stw	%i1, [%g3 + TI_REG_WINDOW + 0x24];	\
+	stw	%i2, [%g3 + TI_REG_WINDOW + 0x28];	\
+	stw	%i3, [%g3 + TI_REG_WINDOW + 0x2c];	\
+	stw	%i4, [%g3 + TI_REG_WINDOW + 0x30];	\
+	stw	%i5, [%g3 + TI_REG_WINDOW + 0x34];	\
+	stw	%i6, [%g3 + TI_REG_WINDOW + 0x38];	\
+	stw	%i7, [%g3 + TI_REG_WINDOW + 0x3c];	\
+	add	%g1, 1, %g1;				\
+	stb	%g1, [%g6 + TI_WSAVED];			\
+	saved;						\
+	rdpr	%cwp, %g1;				\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop
+
 #define SPILL_1_NORMAL SPILL_1_GENERIC(ASI_AIUP)
 #define SPILL_2_NORMAL SPILL_2_GENERIC(ASI_AIUP)
 #define SPILL_3_NORMAL SPILL_0_NORMAL
@@ -325,6 +491,35 @@
 	restored; retry; nop; nop; nop; nop; nop; nop;	\
 	nop; nop; nop; nop; nop; nop; nop; nop;
 
+#define FILL_0_NORMAL_RTRAP				\
+kern_rtt_fill:						\
+	rdpr	%cwp, %g1;				\
+	sub	%g1, 1, %g1;				\
+	wrpr	%g1, %cwp;				\
+	ldx	[%sp + STACK_BIAS + 0x00], %l0;		\
+	ldx	[%sp + STACK_BIAS + 0x08], %l1;		\
+	ldx	[%sp + STACK_BIAS + 0x10], %l2;		\
+	ldx	[%sp + STACK_BIAS + 0x18], %l3;		\
+	ldx	[%sp + STACK_BIAS + 0x20], %l4;		\
+	ldx	[%sp + STACK_BIAS + 0x28], %l5;		\
+	ldx	[%sp + STACK_BIAS + 0x30], %l6;		\
+	ldx	[%sp + STACK_BIAS + 0x38], %l7;		\
+	ldx	[%sp + STACK_BIAS + 0x40], %i0;		\
+	ldx	[%sp + STACK_BIAS + 0x48], %i1;		\
+	ldx	[%sp + STACK_BIAS + 0x50], %i2;		\
+	ldx	[%sp + STACK_BIAS + 0x58], %i3;		\
+	ldx	[%sp + STACK_BIAS + 0x60], %i4;		\
+	ldx	[%sp + STACK_BIAS + 0x68], %i5;		\
+	ldx	[%sp + STACK_BIAS + 0x70], %i6;		\
+	ldx	[%sp + STACK_BIAS + 0x78], %i7;		\
+	restored;					\
+	add	%g1, 1, %g1;				\
+	ba,pt	%xcc, kern_rtt_restore;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop;
+
+
 /* Normal 64bit fill */
 #define FILL_1_GENERIC(ASI)				\
 	add	%sp, STACK_BIAS + 0x00, %g1;		\
@@ -356,6 +551,33 @@
 	b,a,pt	%xcc, fill_fixup_mna;			\
 	b,a,pt	%xcc, fill_fixup;
 
+#define FILL_1_GENERIC_RTRAP				\
+user_rtt_fill_64bit:					\
+	ldxa	[%sp + STACK_BIAS + 0x00] %asi, %l0;	\
+	ldxa	[%sp + STACK_BIAS + 0x08] %asi, %l1;	\
+	ldxa	[%sp + STACK_BIAS + 0x10] %asi, %l2;	\
+	ldxa	[%sp + STACK_BIAS + 0x18] %asi, %l3;	\
+	ldxa	[%sp + STACK_BIAS + 0x20] %asi, %l4;	\
+	ldxa	[%sp + STACK_BIAS + 0x28] %asi, %l5;	\
+	ldxa	[%sp + STACK_BIAS + 0x30] %asi, %l6;	\
+	ldxa	[%sp + STACK_BIAS + 0x38] %asi, %l7;	\
+	ldxa	[%sp + STACK_BIAS + 0x40] %asi, %i0;	\
+	ldxa	[%sp + STACK_BIAS + 0x48] %asi, %i1;	\
+	ldxa	[%sp + STACK_BIAS + 0x50] %asi, %i2;	\
+	ldxa	[%sp + STACK_BIAS + 0x58] %asi, %i3;	\
+	ldxa	[%sp + STACK_BIAS + 0x60] %asi, %i4;	\
+	ldxa	[%sp + STACK_BIAS + 0x68] %asi, %i5;	\
+	ldxa	[%sp + STACK_BIAS + 0x70] %asi, %i6;	\
+	ldxa	[%sp + STACK_BIAS + 0x78] %asi, %i7;	\
+	ba,pt	%xcc, user_rtt_pre_restore;		\
+	 restored;					\
+	nop; nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop; nop;			\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;
+
+
 /* Normal 32bit fill */
 #define FILL_2_GENERIC(ASI)				\
 	srl	%sp, 0, %sp;				\
@@ -387,6 +609,34 @@
 	b,a,pt	%xcc, fill_fixup_mna;			\
 	b,a,pt	%xcc, fill_fixup;
 
+#define FILL_2_GENERIC_RTRAP				\
+user_rtt_fill_32bit:					\
+	srl	%sp, 0, %sp;				\
+	lduwa	[%sp + 0x00] %asi, %l0;			\
+	lduwa	[%sp + 0x04] %asi, %l1;			\
+	lduwa	[%sp + 0x08] %asi, %l2;			\
+	lduwa	[%sp + 0x0c] %asi, %l3;			\
+	lduwa	[%sp + 0x10] %asi, %l4;			\
+	lduwa	[%sp + 0x14] %asi, %l5;			\
+	lduwa	[%sp + 0x18] %asi, %l6;			\
+	lduwa	[%sp + 0x1c] %asi, %l7;			\
+	lduwa	[%sp + 0x20] %asi, %i0;			\
+	lduwa	[%sp + 0x24] %asi, %i1;			\
+	lduwa	[%sp + 0x28] %asi, %i2;			\
+	lduwa	[%sp + 0x2c] %asi, %i3;			\
+	lduwa	[%sp + 0x30] %asi, %i4;			\
+	lduwa	[%sp + 0x34] %asi, %i5;			\
+	lduwa	[%sp + 0x38] %asi, %i6;			\
+	lduwa	[%sp + 0x3c] %asi, %i7;			\
+	ba,pt	%xcc, user_rtt_pre_restore;		\
+	 restored;					\
+	nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop; nop;			\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;
+		
+
 #define FILL_1_NORMAL FILL_1_GENERIC(ASI_AIUP)
 #define FILL_2_NORMAL FILL_2_GENERIC(ASI_AIUP)
 #define FILL_3_NORMAL FILL_0_NORMAL
diff --git a/include/asm-sparc64/uaccess.h b/include/asm-sparc64/uaccess.h
index c91d1e3..afe236b 100644
--- a/include/asm-sparc64/uaccess.h
+++ b/include/asm-sparc64/uaccess.h
@@ -114,16 +114,6 @@ case 8: __put_user_asm(data,x,addr,__pu_ret); break; \
 default: __pu_ret = __put_user_bad(); break; \
 } __pu_ret; })
 
-#define __put_user_nocheck_ret(data,addr,size,retval) ({ \
-register int __foo __asm__ ("l1"); \
-switch (size) { \
-case 1: __put_user_asm_ret(data,b,addr,retval,__foo); break; \
-case 2: __put_user_asm_ret(data,h,addr,retval,__foo); break; \
-case 4: __put_user_asm_ret(data,w,addr,retval,__foo); break; \
-case 8: __put_user_asm_ret(data,x,addr,retval,__foo); break; \
-default: if (__put_user_bad()) return retval; break; \
-} })
-
 #define __put_user_asm(x,size,addr,ret)					\
 __asm__ __volatile__(							\
 	"/* Put user asm, inline. */\n"					\
@@ -143,33 +133,6 @@ __asm__ __volatile__(							\
        : "=r" (ret) : "r" (x), "r" (__m(addr)),				\
 	 "i" (-EFAULT))
 
-#define __put_user_asm_ret(x,size,addr,ret,foo)				\
-if (__builtin_constant_p(ret) && ret == -EFAULT)			\
-__asm__ __volatile__(							\
-	"/* Put user asm ret, inline. */\n"				\
-"1:\t"	"st"#size "a %1, [%2] %%asi\n\n\t"				\
-	".section __ex_table,\"a\"\n\t"					\
-	".align	4\n\t"							\
-	".word	1b, __ret_efault\n\n\t"					\
-	".previous\n\n\t"						\
-       : "=r" (foo) : "r" (x), "r" (__m(addr)));			\
-else									\
-__asm__ __volatile__(							\
-	"/* Put user asm ret, inline. */\n"				\
-"1:\t"	"st"#size "a %1, [%2] %%asi\n\n\t"				\
-	".section .fixup,#alloc,#execinstr\n\t"				\
-	".align	4\n"							\
-"3:\n\t"								\
-	"ret\n\t"							\
-	" restore %%g0, %3, %%o0\n\n\t"					\
-	".previous\n\t"							\
-	".section __ex_table,\"a\"\n\t"					\
-	".align	4\n\t"							\
-	".word	1b, 3b\n\n\t"						\
-	".previous\n\n\t"						\
-       : "=r" (foo) : "r" (x), "r" (__m(addr)),				\
-         "i" (ret))
-
 extern int __put_user_bad(void);
 
 #define __get_user_nocheck(data,addr,size,type) ({ \
@@ -289,14 +252,7 @@ copy_in_user(void __user *to, void __user *from, unsigned long size)
 }
 #define __copy_in_user copy_in_user
 
-extern unsigned long __must_check __bzero_noasi(void __user *, unsigned long);
-
-static inline unsigned long __must_check
-__clear_user(void __user *addr, unsigned long size)
-{
-	
-	return __bzero_noasi(addr, size);
-}
+extern unsigned long __must_check __clear_user(void __user *, unsigned long);
 
 #define clear_user __clear_user
 
diff --git a/include/asm-sparc64/vdev.h b/include/asm-sparc64/vdev.h
new file mode 100644
index 0000000..996e6be
--- /dev/null
+++ b/include/asm-sparc64/vdev.h
@@ -0,0 +1,16 @@
+/* vdev.h: SUN4V virtual device interfaces and defines.
+ *
+ * Copyright (C) 2006 David S. Miller <davem@davemloft.net>
+ */
+
+#ifndef _SPARC64_VDEV_H
+#define _SPARC64_VDEV_H
+
+#include <linux/types.h>
+
+extern u32 sun4v_vdev_devhandle;
+extern int sun4v_vdev_root;
+
+extern unsigned int sun4v_vdev_device_interrupt(unsigned int);
+
+#endif /* !(_SPARC64_VDEV_H) */
diff --git a/include/asm-sparc64/xor.h b/include/asm-sparc64/xor.h
index 8b3a7e4..8ce3f18 100644
--- a/include/asm-sparc64/xor.h
+++ b/include/asm-sparc64/xor.h
@@ -2,9 +2,11 @@
  * include/asm-sparc64/xor.h
  *
  * High speed xor_block operation for RAID4/5 utilizing the
- * UltraSparc Visual Instruction Set.
+ * UltraSparc Visual Instruction Set and Niagara block-init
+ * twin-load instructions.
  *
  * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
+ * Copyright (C) 2006 David S. Miller <davem@davemloft.net>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -16,8 +18,7 @@
  * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  */
 
-#include <asm/pstate.h>
-#include <asm/asi.h>
+#include <asm/spitfire.h>
 
 extern void xor_vis_2(unsigned long, unsigned long *, unsigned long *);
 extern void xor_vis_3(unsigned long, unsigned long *, unsigned long *,
@@ -37,4 +38,29 @@ static struct xor_block_template xor_block_VIS = {
         .do_5	= xor_vis_5,
 };
 
-#define XOR_TRY_TEMPLATES       xor_speed(&xor_block_VIS)
+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 *);
+
+static struct xor_block_template xor_block_niagara = {
+        .name	= "Niagara",
+        .do_2	= xor_niagara_2,
+        .do_3	= xor_niagara_3,
+        .do_4	= xor_niagara_4,
+        .do_5	= xor_niagara_5,
+};
+
+#undef XOR_TRY_TEMPLATES
+#define XOR_TRY_TEMPLATES				\
+	do {						\
+		xor_speed(&xor_block_VIS);		\
+		xor_speed(&xor_block_niagara);		\
+	} while (0)
+
+/* For VIS for everything except Niagara.  */
+#define XOR_SELECT_TEMPLATE(FASTEST) \
+	(tlb_type == hypervisor ? &xor_block_niagara : &xor_block_VIS)
diff --git a/include/linux/serial_core.h b/include/linux/serial_core.h
index 4041122..57abcea 100644
--- a/include/linux/serial_core.h
+++ b/include/linux/serial_core.h
@@ -127,6 +127,9 @@
 /* Hilscher netx */
 #define PORT_NETX	71
 
+/* SUN4V Hypervisor Console */
+#define PORT_SUNHV	72
+
 #ifdef __KERNEL__
 
 #include <linux/config.h>