Merge branch 'master'

28 files changed, 1060 insertions, 1450 deletions

diff --git a/arch/sparc64/kernel/cpu.c b/arch/sparc64/kernel/cpu.c
index 48756958116..77ef5df4e5a 100644
--- a/arch/sparc64/kernel/cpu.c
+++ b/arch/sparc64/kernel/cpu.c
@@ -39,6 +39,8 @@ struct cpu_fp_info linux_sparc_fpu[] = {
   { 0x3e, 0x15, 0, "UltraSparc III+ integrated FPU"},
   { 0x3e, 0x16, 0, "UltraSparc IIIi integrated FPU"},
   { 0x3e, 0x18, 0, "UltraSparc IV integrated FPU"},
+  { 0x3e, 0x19, 0, "UltraSparc IV+ integrated FPU"},
+  { 0x3e, 0x22, 0, "UltraSparc IIIi+ integrated FPU"},
 };
 
 #define NSPARCFPU  (sizeof(linux_sparc_fpu)/sizeof(struct cpu_fp_info))
@@ -53,6 +55,8 @@ struct cpu_iu_info linux_sparc_chips[] = {
   { 0x3e, 0x15, "TI UltraSparc III+ (Cheetah+)"},
   { 0x3e, 0x16, "TI UltraSparc IIIi (Jalapeno)"},
   { 0x3e, 0x18, "TI UltraSparc IV (Jaguar)"},
+  { 0x3e, 0x19, "TI UltraSparc IV+ (Panther)"},
+  { 0x3e, 0x22, "TI UltraSparc IIIi+ (Serrano)"},
 };
 
 #define NSPARCCHIPS  (sizeof(linux_sparc_chips)/sizeof(struct cpu_iu_info))
diff --git a/arch/sparc64/kernel/devices.c b/arch/sparc64/kernel/devices.c
index d710274e516..df9a1ca8fd7 100644
--- a/arch/sparc64/kernel/devices.c
+++ b/arch/sparc64/kernel/devices.c
@@ -135,6 +135,28 @@ void __init device_scan(void)
 		cpu_data(0).clock_tick = prom_getintdefault(cpu_node,
 							    "clock-frequency",
 							    0);
+		cpu_data(0).dcache_size = prom_getintdefault(cpu_node,
+							     "dcache-size",
+							     16 * 1024);
+		cpu_data(0).dcache_line_size =
+			prom_getintdefault(cpu_node, "dcache-line-size", 32);
+		cpu_data(0).icache_size = prom_getintdefault(cpu_node,
+							     "icache-size",
+							     16 * 1024);
+		cpu_data(0).icache_line_size =
+			prom_getintdefault(cpu_node, "icache-line-size", 32);
+		cpu_data(0).ecache_size = prom_getintdefault(cpu_node,
+							     "ecache-size",
+							     4 * 1024 * 1024);
+		cpu_data(0).ecache_line_size =
+			prom_getintdefault(cpu_node, "ecache-line-size", 64);
+		printk("CPU[0]: Caches "
+		       "D[sz(%d):line_sz(%d)] "
+		       "I[sz(%d):line_sz(%d)] "
+		       "E[sz(%d):line_sz(%d)]\n",
+		       cpu_data(0).dcache_size, cpu_data(0).dcache_line_size,
+		       cpu_data(0).icache_size, cpu_data(0).icache_line_size,
+		       cpu_data(0).ecache_size, cpu_data(0).ecache_line_size);
 	}
 #endif
 
diff --git a/arch/sparc64/kernel/dtlb_backend.S b/arch/sparc64/kernel/dtlb_backend.S
index 538522848ad..acc889a7f9c 100644
--- a/arch/sparc64/kernel/dtlb_backend.S
+++ b/arch/sparc64/kernel/dtlb_backend.S
@@ -9,17 +9,7 @@
 #include <asm/pgtable.h>
 #include <asm/mmu.h>
 
-#if PAGE_SHIFT == 13
-#define SZ_BITS		_PAGE_SZ8K
-#elif PAGE_SHIFT == 16
-#define SZ_BITS		_PAGE_SZ64K
-#elif PAGE_SHIFT == 19
-#define SZ_BITS		_PAGE_SZ512K
-#elif PAGE_SHIFT == 22
-#define SZ_BITS		_PAGE_SZ4MB
-#endif
-
-#define VALID_SZ_BITS	(_PAGE_VALID | SZ_BITS)
+#define VALID_SZ_BITS	(_PAGE_VALID | _PAGE_SZBITS)
 
 #define VPTE_BITS		(_PAGE_CP | _PAGE_CV | _PAGE_P )
 #define VPTE_SHIFT		(PAGE_SHIFT - 3)
@@ -163,7 +153,6 @@ sparc64_vpte_continue:
 	stxa		%g4, [%g1 + %g1] ASI_DMMU	! Restore previous TAG_ACCESS
 	retry						! Load PTE once again
 
-#undef SZ_BITS
 #undef VALID_SZ_BITS
 #undef VPTE_SHIFT
 #undef VPTE_BITS
diff --git a/arch/sparc64/kernel/dtlb_base.S b/arch/sparc64/kernel/dtlb_base.S
index ded2fed23fc..6528786840c 100644
--- a/arch/sparc64/kernel/dtlb_base.S
+++ b/arch/sparc64/kernel/dtlb_base.S
@@ -53,39 +53,36 @@
  * be guaranteed to be 0 ... mmu_context.h does guarantee this
  * by only using 10 bits in the hwcontext value.
  */
-#define CREATE_VPTE_OFFSET1(r1, r2)
+#define CREATE_VPTE_OFFSET1(r1, r2) nop
 #define CREATE_VPTE_OFFSET2(r1, r2) \
 				srax	r1, 10, r2
-#define CREATE_VPTE_NOP		nop
 #else
 #define CREATE_VPTE_OFFSET1(r1, r2) \
 				srax	r1, PAGE_SHIFT, r2
 #define CREATE_VPTE_OFFSET2(r1, r2) \
 				sllx	r2, 3, r2
-#define CREATE_VPTE_NOP
 #endif
 
 /* DTLB ** ICACHE line 1: Quick user TLB misses		*/
+	mov		TLB_SFSR, %g1
 	ldxa		[%g1 + %g1] ASI_DMMU, %g4	! Get TAG_ACCESS
 	andcc		%g4, TAG_CONTEXT_BITS, %g0	! From Nucleus?
 from_tl1_trap:
 	rdpr		%tl, %g5			! For TL==3 test
 	CREATE_VPTE_OFFSET1(%g4, %g6)			! Create VPTE offset
-	be,pn		%xcc, 3f			! Yep, special processing
+	be,pn		%xcc, kvmap			! Yep, special processing
 	 CREATE_VPTE_OFFSET2(%g4, %g6)			! Create VPTE offset
 	cmp		%g5, 4				! Last trap level?
-	be,pn		%xcc, longpath			! Yep, cannot risk VPTE miss
-	 nop						! delay slot
 
 /* DTLB ** ICACHE line 2: User finish + quick kernel TLB misses	*/
+	be,pn		%xcc, longpath			! Yep, cannot risk VPTE miss
+	 nop						! delay slot
 	ldxa		[%g3 + %g6] ASI_S, %g5		! Load VPTE
 1:	brgez,pn	%g5, longpath			! Invalid, branch out
 	 nop						! Delay-slot
 9:	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Reload TLB
 	retry						! Trap return
-3:	brlz,pt		%g4, 9b				! Kernel virtual map?
-	 xor		%g2, %g4, %g5			! Finish bit twiddles
-	ba,a,pt		%xcc, kvmap			! Yep, go check for obp/vmalloc
+	nop
 
 /* DTLB ** ICACHE line 3: winfixups+real_faults		*/
 longpath:
@@ -106,8 +103,7 @@ longpath:
 	nop
 	nop
 	nop
-	CREATE_VPTE_NOP
+	nop
 
 #undef CREATE_VPTE_OFFSET1
 #undef CREATE_VPTE_OFFSET2
-#undef CREATE_VPTE_NOP
diff --git a/arch/sparc64/kernel/dtlb_prot.S b/arch/sparc64/kernel/dtlb_prot.S
index d848bb7374b..e0a92016260 100644
--- a/arch/sparc64/kernel/dtlb_prot.S
+++ b/arch/sparc64/kernel/dtlb_prot.S
@@ -14,14 +14,14 @@
  */
 
 /* PROT ** ICACHE line 1: User DTLB protection trap	*/
-	stxa		%g0, [%g1] ASI_DMMU		! Clear SFSR FaultValid bit
-	membar		#Sync				! Synchronize ASI stores
-	rdpr		%pstate, %g5			! Move into alternate globals
+	mov		TLB_SFSR, %g1
+	stxa		%g0, [%g1] ASI_DMMU		! Clear FaultValid bit
+	membar		#Sync				! Synchronize stores
+	rdpr		%pstate, %g5			! Move into alt-globals
 	wrpr		%g5, PSTATE_AG|PSTATE_MG, %pstate
-	rdpr		%tl, %g1			! Need to do a winfixup?
+	rdpr		%tl, %g1			! Need a winfixup?
 	cmp		%g1, 1				! Trap level >1?
-	mov		TLB_TAG_ACCESS, %g4		! Prepare reload of vaddr
-	nop
+	mov		TLB_TAG_ACCESS, %g4		! For reload of vaddr
 
 /* PROT ** ICACHE line 2: More real fault processing */
 	bgu,pn		%xcc, winfix_trampoline		! Yes, perform winfixup
diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S
index b4834952785..11a848402fb 100644
--- a/arch/sparc64/kernel/entry.S
+++ b/arch/sparc64/kernel/entry.S
@@ -30,163 +30,10 @@
 	.text
 	.align		32
 
-	.globl		sparc64_vpte_patchme1
-	.globl		sparc64_vpte_patchme2
-/*
- * On a second level vpte miss, check whether the original fault is to the OBP 
- * range (note that this is only possible for instruction miss, data misses to
- * obp range do not use vpte). If so, go back directly to the faulting address.
- * This is because we want to read the tpc, otherwise we have no way of knowing
- * the 8k aligned faulting address if we are using >8k kernel pagesize. This
- * also ensures no vpte range addresses are dropped into tlb while obp is
- * executing (see inherit_locked_prom_mappings() rant).
- */
-sparc64_vpte_nucleus:
-	/* Note that kvmap below has verified that the address is
-	 * in the range MODULES_VADDR --> VMALLOC_END already.  So
-	 * here we need only check if it is an OBP address or not.
-	 */
-	sethi		%hi(LOW_OBP_ADDRESS), %g5
-	cmp		%g4, %g5
-	blu,pn		%xcc, sparc64_vpte_patchme1
-	 mov		0x1, %g5
-	sllx		%g5, 32, %g5
-	cmp		%g4, %g5
-	blu,pn		%xcc, obp_iaddr_patch
-	 nop
-
-	/* These two instructions are patched by paginig_init().  */
-sparc64_vpte_patchme1:
-	sethi		%hi(0), %g5
-sparc64_vpte_patchme2:
-	or		%g5, %lo(0), %g5
-
-	/* With kernel PGD in %g5, branch back into dtlb_backend.  */
-	ba,pt		%xcc, sparc64_kpte_continue
-	 andn		%g1, 0x3, %g1	/* Finish PMD offset adjustment.  */
-
-vpte_noent:
-	/* Restore previous TAG_ACCESS, %g5 is zero, and we will
-	 * skip over the trap instruction so that the top level
-	 * TLB miss handler will thing this %g5 value is just an
-	 * invalid PTE, thus branching to full fault processing.
-	 */
-	mov		TLB_SFSR, %g1
-	stxa		%g4, [%g1 + %g1] ASI_DMMU
-	done
-
-	.globl		obp_iaddr_patch
-obp_iaddr_patch:
-	/* These two instructions patched by inherit_prom_mappings().  */
-	sethi		%hi(0), %g5
-	or		%g5, %lo(0), %g5
-
-	/* Behave as if we are at TL0.  */
-	wrpr		%g0, 1, %tl
-	rdpr		%tpc, %g4	/* Find original faulting iaddr */
-	srlx		%g4, 13, %g4	/* Throw out context bits */
-	sllx		%g4, 13, %g4	/* g4 has vpn + ctx0 now */
-
-	/* Restore previous TAG_ACCESS.  */
-	mov		TLB_SFSR, %g1
-	stxa		%g4, [%g1 + %g1] ASI_IMMU
-
-	/* Get PMD offset.  */
-	srlx		%g4, 23, %g6
-	and		%g6, 0x7ff, %g6
-	sllx		%g6, 2, %g6
-
-	/* Load PMD, is it valid?  */
-	lduwa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brz,pn		%g5, longpath
-	 sllx		%g5, 11, %g5
-
-	/* Get PTE offset.  */
-	srlx		%g4, 13, %g6
-	and		%g6, 0x3ff, %g6
-	sllx		%g6, 3, %g6
-
-	/* Load PTE.  */
-	ldxa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brgez,pn	%g5, longpath
-	 nop
-
-	/* TLB load and return from trap.  */
-	stxa		%g5, [%g0] ASI_ITLB_DATA_IN
-	retry
-
-	.globl		obp_daddr_patch
-obp_daddr_patch:
-	/* These two instructions patched by inherit_prom_mappings().  */
-	sethi		%hi(0), %g5
-	or		%g5, %lo(0), %g5
-
-	/* Get PMD offset.  */
-	srlx		%g4, 23, %g6
-	and		%g6, 0x7ff, %g6
-	sllx		%g6, 2, %g6
-
-	/* Load PMD, is it valid?  */
-	lduwa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brz,pn		%g5, longpath
-	 sllx		%g5, 11, %g5
-
-	/* Get PTE offset.  */
-	srlx		%g4, 13, %g6
-	and		%g6, 0x3ff, %g6
-	sllx		%g6, 3, %g6
-
-	/* Load PTE.  */
-	ldxa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brgez,pn	%g5, longpath
-	 nop
-
-	/* TLB load and return from trap.  */
-	stxa		%g5, [%g0] ASI_DTLB_DATA_IN
-	retry
-
-/*
- * On a first level data miss, check whether this is to the OBP range (note
- * that such accesses can be made by prom, as well as by kernel using
- * prom_getproperty on "address"), and if so, do not use vpte access ...
- * rather, use information saved during inherit_prom_mappings() using 8k
- * pagesize.
- */
-	.align		32
-kvmap:
-	sethi		%hi(MODULES_VADDR), %g5
-	cmp		%g4, %g5
-	blu,pn		%xcc, longpath
-	 mov		(VMALLOC_END >> 24), %g5
-	sllx		%g5, 24, %g5
-	cmp		%g4, %g5
-	bgeu,pn		%xcc, longpath
-	 nop
-
-kvmap_check_obp:
-	sethi		%hi(LOW_OBP_ADDRESS), %g5
-	cmp		%g4, %g5
-	blu,pn		%xcc, kvmap_vmalloc_addr
-	 mov		0x1, %g5
-	sllx		%g5, 32, %g5
-	cmp		%g4, %g5
-	blu,pn		%xcc, obp_daddr_patch
-	 nop
-
-kvmap_vmalloc_addr:
-	/* If we get here, a vmalloc addr was accessed, load kernel VPTE.  */
-	ldxa		[%g3 + %g6] ASI_N, %g5
-	brgez,pn	%g5, longpath
-	 nop
-
-	/* PTE is valid, load into TLB and return from trap.  */
-	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Reload TLB
-	retry
-
 	/* This is trivial with the new code... */
 	.globl		do_fpdis
 do_fpdis:
-	sethi		%hi(TSTATE_PEF), %g4					! IEU0
+	sethi		%hi(TSTATE_PEF), %g4
 	rdpr		%tstate, %g5
 	andcc		%g5, %g4, %g0
 	be,pt		%xcc, 1f
@@ -203,18 +50,18 @@ do_fpdis:
 	add		%g0, %g0, %g0
 	ba,a,pt		%xcc, rtrap_clr_l6
 
-1:	ldub		[%g6 + TI_FPSAVED], %g5					! Load	Group
-	wr		%g0, FPRS_FEF, %fprs					! LSU	Group+4bubbles
-	andcc		%g5, FPRS_FEF, %g0					! IEU1	Group
-	be,a,pt		%icc, 1f						! CTI
-	 clr		%g7							! IEU0
-	ldx		[%g6 + TI_GSR], %g7					! Load	Group
-1:	andcc		%g5, FPRS_DL, %g0					! IEU1
-	bne,pn		%icc, 2f						! CTI
-	 fzero		%f0							! FPA
-	andcc		%g5, FPRS_DU, %g0					! IEU1  Group
-	bne,pn		%icc, 1f						! CTI
-	 fzero		%f2							! FPA
+1:	ldub		[%g6 + TI_FPSAVED], %g5
+	wr		%g0, FPRS_FEF, %fprs
+	andcc		%g5, FPRS_FEF, %g0
+	be,a,pt		%icc, 1f
+	 clr		%g7
+	ldx		[%g6 + TI_GSR], %g7
+1:	andcc		%g5, FPRS_DL, %g0
+	bne,pn		%icc, 2f
+	 fzero		%f0
+	andcc		%g5, FPRS_DU, %g0
+	bne,pn		%icc, 1f
+	 fzero		%f2
 	faddd		%f0, %f2, %f4
 	fmuld		%f0, %f2, %f6
 	faddd		%f0, %f2, %f8
@@ -250,15 +97,17 @@ do_fpdis:
 	faddd		%f0, %f2, %f4
 	fmuld		%f0, %f2, %f6
 	ldxa		[%g3] ASI_DMMU, %g5
-cplus_fptrap_insn_1:
-	sethi		%hi(0), %g2
+	sethi		%hi(sparc64_kern_sec_context), %g2
+	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
 	stxa		%g2, [%g3] ASI_DMMU
 	membar		#Sync
 	add		%g6, TI_FPREGS + 0xc0, %g2
 	faddd		%f0, %f2, %f8
 	fmuld		%f0, %f2, %f10
-	ldda		[%g1] ASI_BLK_S, %f32	! grrr, where is ASI_BLK_NUCLEUS 8-(
+	membar		#Sync
+	ldda		[%g1] ASI_BLK_S, %f32
 	ldda		[%g2] ASI_BLK_S, %f48
+	membar		#Sync
 	faddd		%f0, %f2, %f12
 	fmuld		%f0, %f2, %f14
 	faddd		%f0, %f2, %f16
@@ -269,7 +118,6 @@ cplus_fptrap_insn_1:
 	fmuld		%f0, %f2, %f26
 	faddd		%f0, %f2, %f28
 	fmuld		%f0, %f2, %f30
-	membar		#Sync
 	b,pt		%xcc, fpdis_exit
 	 nop
 2:	andcc		%g5, FPRS_DU, %g0
@@ -279,15 +127,17 @@ cplus_fptrap_insn_1:
 	fzero		%f34
 	ldxa		[%g3] ASI_DMMU, %g5
 	add		%g6, TI_FPREGS, %g1
-cplus_fptrap_insn_2:
-	sethi		%hi(0), %g2
+	sethi		%hi(sparc64_kern_sec_context), %g2
+	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
 	stxa		%g2, [%g3] ASI_DMMU
 	membar		#Sync
 	add		%g6, TI_FPREGS + 0x40, %g2
 	faddd		%f32, %f34, %f36
 	fmuld		%f32, %f34, %f38
-	ldda		[%g1] ASI_BLK_S, %f0	! grrr, where is ASI_BLK_NUCLEUS 8-(
+	membar		#Sync
+	ldda		[%g1] ASI_BLK_S, %f0
 	ldda		[%g2] ASI_BLK_S, %f16
+	membar		#Sync
 	faddd		%f32, %f34, %f40
 	fmuld		%f32, %f34, %f42
 	faddd		%f32, %f34, %f44
@@ -300,18 +150,18 @@ cplus_fptrap_insn_2:
 	fmuld		%f32, %f34, %f58
 	faddd		%f32, %f34, %f60
 	fmuld		%f32, %f34, %f62
-	membar		#Sync
 	ba,pt		%xcc, fpdis_exit
 	 nop
 3:	mov		SECONDARY_CONTEXT, %g3
 	add		%g6, TI_FPREGS, %g1
 	ldxa		[%g3] ASI_DMMU, %g5
-cplus_fptrap_insn_3:
-	sethi		%hi(0), %g2
+	sethi		%hi(sparc64_kern_sec_context), %g2
+	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
 	stxa		%g2, [%g3] ASI_DMMU
 	membar		#Sync
 	mov		0x40, %g2
-	ldda		[%g1] ASI_BLK_S, %f0		! grrr, where is ASI_BLK_NUCLEUS 8-(
+	membar		#Sync
+	ldda		[%g1] ASI_BLK_S, %f0
 	ldda		[%g1 + %g2] ASI_BLK_S, %f16
 	add		%g1, 0x80, %g1
 	ldda		[%g1] ASI_BLK_S, %f32
@@ -472,8 +322,8 @@ do_fptrap_after_fsr:
 	stx		%g3, [%g6 + TI_GSR]
 	mov		SECONDARY_CONTEXT, %g3
 	ldxa		[%g3] ASI_DMMU, %g5
-cplus_fptrap_insn_4:
-	sethi		%hi(0), %g2
+	sethi		%hi(sparc64_kern_sec_context), %g2
+	ldx		[%g2 + %lo(sparc64_kern_sec_context)], %g2
 	stxa		%g2, [%g3] ASI_DMMU
 	membar		#Sync
 	add		%g6, TI_FPREGS, %g2
@@ -494,45 +344,17 @@ cplus_fptrap_insn_4:
 	ba,pt		%xcc, etrap
 	 wr		%g0, 0, %fprs
 
-cplus_fptrap_1:
-	sethi		%hi(CTX_CHEETAH_PLUS_CTX0), %g2
-
-	.globl		cheetah_plus_patch_fpdis
-cheetah_plus_patch_fpdis:
-	/* We configure the dTLB512_0 for 4MB pages and the
-	 * dTLB512_1 for 8K pages when in context zero.
-	 */
-	sethi			%hi(cplus_fptrap_1), %o0
-	lduw			[%o0 + %lo(cplus_fptrap_1)], %o1
-
-	set			cplus_fptrap_insn_1, %o2
-	stw			%o1, [%o2]
-	flush			%o2
-	set			cplus_fptrap_insn_2, %o2
-	stw			%o1, [%o2]
-	flush			%o2
-	set			cplus_fptrap_insn_3, %o2
-	stw			%o1, [%o2]
-	flush			%o2
-	set			cplus_fptrap_insn_4, %o2
-	stw			%o1, [%o2]
-	flush			%o2
-
-	retl
-	 nop
-
 	/* The registers for cross calls will be:
 	 *
 	 * DATA 0: [low 32-bits]  Address of function to call, jmp to this
 	 *         [high 32-bits] MMU Context Argument 0, place in %g5
-	 * DATA 1: Address Argument 1, place in %g6
+	 * DATA 1: Address Argument 1, place in %g1
 	 * DATA 2: Address Argument 2, place in %g7
 	 *
 	 * With this method we can do most of the cross-call tlb/cache
 	 * flushing very quickly.
 	 *
-	 * Current CPU's IRQ worklist table is locked into %g1,
-	 * don't touch.
+	 * Current CPU's IRQ worklist table is locked into %g6, don't touch.
 	 */
 	.text
 	.align		32
@@ -1006,13 +828,14 @@ cheetah_plus_dcpe_trap_vector:
 	nop
 
 do_cheetah_plus_data_parity:
-	ba,pt		%xcc, etrap
+	rdpr		%pil, %g2
+	wrpr		%g0, 15, %pil
+	ba,pt		%xcc, etrap_irq
 	 rd		%pc, %g7
 	mov		0x0, %o0
 	call		cheetah_plus_parity_error
 	 add		%sp, PTREGS_OFF, %o1
-	ba,pt		%xcc, rtrap
-	 clr		%l6
+	ba,a,pt		%xcc, rtrap_irq
 
 cheetah_plus_dcpe_trap_vector_tl1:
 	membar		#Sync
@@ -1036,13 +859,14 @@ cheetah_plus_icpe_trap_vector:
 	nop
 
 do_cheetah_plus_insn_parity:
-	ba,pt		%xcc, etrap
+	rdpr		%pil, %g2
+	wrpr		%g0, 15, %pil
+	ba,pt		%xcc, etrap_irq
 	 rd		%pc, %g7
 	mov		0x1, %o0
 	call		cheetah_plus_parity_error
 	 add		%sp, PTREGS_OFF, %o1
-	ba,pt		%xcc, rtrap
-	 clr		%l6
+	ba,a,pt		%xcc, rtrap_irq
 
 cheetah_plus_icpe_trap_vector_tl1:
 	membar		#Sync
@@ -1075,6 +899,10 @@ do_dcpe_tl1:
 	 nop
 	wrpr		%g1, %tl		! Restore original trap level
 do_dcpe_tl1_nonfatal:	/* Ok we may use interrupt globals safely. */
+	sethi		%hi(dcache_parity_tl1_occurred), %g2
+	lduw		[%g2 + %lo(dcache_parity_tl1_occurred)], %g1
+	add		%g1, 1, %g1
+	stw		%g1, [%g2 + %lo(dcache_parity_tl1_occurred)]
 	/* Reset D-cache parity */
 	sethi		%hi(1 << 16), %g1	! D-cache size
 	mov		(1 << 5), %g2		! D-cache line size
@@ -1121,6 +949,10 @@ do_icpe_tl1:
 	 nop
 	wrpr		%g1, %tl		! Restore original trap level
 do_icpe_tl1_nonfatal:	/* Ok we may use interrupt globals safely. */
+	sethi		%hi(icache_parity_tl1_occurred), %g2
+	lduw		[%g2 + %lo(icache_parity_tl1_occurred)], %g1
+	add		%g1, 1, %g1
+	stw		%g1, [%g2 + %lo(icache_parity_tl1_occurred)]
 	/* Flush I-cache */
 	sethi		%hi(1 << 15), %g1	! I-cache size
 	mov		(1 << 5), %g2		! I-cache line size
diff --git a/arch/sparc64/kernel/etrap.S b/arch/sparc64/kernel/etrap.S
index 50d2af1d98a..0d8eba21111 100644
--- a/arch/sparc64/kernel/etrap.S
+++ b/arch/sparc64/kernel/etrap.S
@@ -68,12 +68,8 @@ etrap_irq:
 
 		wrpr	%g3, 0, %otherwin
 		wrpr	%g2, 0, %wstate
-cplus_etrap_insn_1:
-		sethi	%hi(0), %g3
-		sllx	%g3, 32, %g3
-cplus_etrap_insn_2:
-		sethi	%hi(0), %g2
-		or	%g3, %g2, %g3
+		sethi	%hi(sparc64_kern_pri_context), %g2
+		ldx	[%g2 + %lo(sparc64_kern_pri_context)], %g3
 		stxa	%g3, [%l4] ASI_DMMU
 		flush	%l6
 		wr	%g0, ASI_AIUS, %asi
@@ -215,12 +211,8 @@ scetrap:	rdpr	%pil, %g2
 		mov	PRIMARY_CONTEXT, %l4
 		wrpr	%g3, 0, %otherwin
 		wrpr	%g2, 0, %wstate
-cplus_etrap_insn_3:
-		sethi	%hi(0), %g3
-		sllx	%g3, 32, %g3
-cplus_etrap_insn_4:
-		sethi	%hi(0), %g2
-		or	%g3, %g2, %g3
+		sethi	%hi(sparc64_kern_pri_context), %g2
+		ldx	[%g2 + %lo(sparc64_kern_pri_context)], %g3
 		stxa	%g3, [%l4] ASI_DMMU
 		flush	%l6
 
@@ -264,38 +256,3 @@ cplus_etrap_insn_4:
 
 #undef TASK_REGOFF
 #undef ETRAP_PSTATE1
-
-cplus_einsn_1:
-		sethi			%uhi(CTX_CHEETAH_PLUS_NUC), %g3
-cplus_einsn_2:
-		sethi			%hi(CTX_CHEETAH_PLUS_CTX0), %g2
-
-		.globl			cheetah_plus_patch_etrap
-cheetah_plus_patch_etrap:
-		/* We configure the dTLB512_0 for 4MB pages and the
-		 * dTLB512_1 for 8K pages when in context zero.
-		 */
-		sethi			%hi(cplus_einsn_1), %o0
-		sethi			%hi(cplus_etrap_insn_1), %o2
-		lduw			[%o0 + %lo(cplus_einsn_1)], %o1
-		or			%o2, %lo(cplus_etrap_insn_1), %o2
-		stw			%o1, [%o2]
-		flush			%o2
-		sethi			%hi(cplus_etrap_insn_3), %o2
-		or			%o2, %lo(cplus_etrap_insn_3), %o2
-		stw			%o1, [%o2]
-		flush			%o2
-
-		sethi			%hi(cplus_einsn_2), %o0
-		sethi			%hi(cplus_etrap_insn_2), %o2
-		lduw			[%o0 + %lo(cplus_einsn_2)], %o1
-		or			%o2, %lo(cplus_etrap_insn_2), %o2
-		stw			%o1, [%o2]
-		flush			%o2
-		sethi			%hi(cplus_etrap_insn_4), %o2
-		or			%o2, %lo(cplus_etrap_insn_4), %o2
-		stw			%o1, [%o2]
-		flush			%o2
-
-		retl
-		 nop
diff --git a/arch/sparc64/kernel/head.S b/arch/sparc64/kernel/head.S
index 1fa06c4e3bd..b49dcd4504b 100644
--- a/arch/sparc64/kernel/head.S
+++ b/arch/sparc64/kernel/head.S
@@ -28,19 +28,14 @@
 #include <asm/mmu.h>
 	
 /* This section from from _start to sparc64_boot_end should fit into
- * 0x0000.0000.0040.4000 to 0x0000.0000.0040.8000 and will be sharing space
- * with bootup_user_stack, which is from 0x0000.0000.0040.4000 to
- * 0x0000.0000.0040.6000 and empty_bad_page, which is from
- * 0x0000.0000.0040.6000 to 0x0000.0000.0040.8000. 
+ * 0x0000000000404000 to 0x0000000000408000.
  */
-
 	.text
 	.globl	start, _start, stext, _stext
 _start:
 start:
 _stext:
 stext:
-bootup_user_stack:
 ! 0x0000000000404000
 	b	sparc64_boot
 	 flushw					/* Flush register file.      */
@@ -80,15 +75,169 @@ sparc_ramdisk_image64:
 	.xword	0
 	.word	_end
 
-	/* We must be careful, 32-bit OpenBOOT will get confused if it
-	 * tries to save away a register window to a 64-bit kernel
-	 * stack address.  Flush all windows, disable interrupts,
-	 * remap if necessary, jump onto kernel trap table, then kernel
-	 * stack, or else we die.
+	/* PROM cif handler code address is in %o4.  */
+sparc64_boot:
+1:	rd	%pc, %g7
+	set	1b, %g1
+	cmp	%g1, %g7
+	be,pn	%xcc, sparc64_boot_after_remap
+	 mov	%o4, %l7
+
+	/* We need to remap the kernel.  Use position independant
+	 * code to remap us to KERNBASE.
 	 *
-	 * PROM entry point is on %o4
+	 * SILO can invoke us with 32-bit address masking enabled,
+	 * so make sure that's clear.
 	 */
-sparc64_boot:
+	rdpr	%pstate, %g1
+	andn	%g1, PSTATE_AM, %g1
+	wrpr	%g1, 0x0, %pstate
+	ba,a,pt	%xcc, 1f
+
+	.globl	prom_finddev_name, prom_chosen_path
+	.globl	prom_getprop_name, prom_mmu_name
+	.globl	prom_callmethod_name, prom_translate_name
+	.globl	prom_map_name, prom_unmap_name, prom_mmu_ihandle_cache
+	.globl	prom_boot_mapped_pc, prom_boot_mapping_mode
+	.globl	prom_boot_mapping_phys_high, prom_boot_mapping_phys_low
+prom_finddev_name:
+	.asciz	"finddevice"
+prom_chosen_path:
+	.asciz	"/chosen"
+prom_getprop_name:
+	.asciz	"getprop"
+prom_mmu_name:
+	.asciz	"mmu"
+prom_callmethod_name:
+	.asciz	"call-method"
+prom_translate_name:
+	.asciz	"translate"
+prom_map_name:
+	.asciz	"map"
+prom_unmap_name:
+	.asciz	"unmap"
+	.align	4
+prom_mmu_ihandle_cache:
+	.word	0
+prom_boot_mapped_pc:
+	.word	0
+prom_boot_mapping_mode:
+	.word	0
+	.align	8
+prom_boot_mapping_phys_high:
+	.xword	0
+prom_boot_mapping_phys_low:
+	.xword	0
+1:
+	rd	%pc, %l0
+	mov	(1b - prom_finddev_name), %l1
+	mov	(1b - prom_chosen_path), %l2
+	mov	(1b - prom_boot_mapped_pc), %l3
+	sub	%l0, %l1, %l1
+	sub	%l0, %l2, %l2
+	sub	%l0, %l3, %l3
+	stw	%l0, [%l3]
+	sub	%sp, (192 + 128), %sp
+
+	/* chosen_node = prom_finddevice("/chosen") */
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "finddevice"
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 1
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	stx	%l2, [%sp + 2047 + 128 + 0x18]	! arg1, "/chosen"
+	stx	%g0, [%sp + 2047 + 128 + 0x20]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	ldx	[%sp + 2047 + 128 + 0x20], %l4	! chosen device node
+
+	mov	(1b - prom_getprop_name), %l1
+	mov	(1b - prom_mmu_name), %l2
+	mov	(1b - prom_mmu_ihandle_cache), %l5
+	sub	%l0, %l1, %l1
+	sub	%l0, %l2, %l2
+	sub	%l0, %l5, %l5
+
+	/* prom_mmu_ihandle_cache = prom_getint(chosen_node, "mmu") */
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "getprop"
+	mov	4, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 4
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	stx	%l4, [%sp + 2047 + 128 + 0x18]	! arg1, chosen_node
+	stx	%l2, [%sp + 2047 + 128 + 0x20]	! arg2, "mmu"
+	stx	%l5, [%sp + 2047 + 128 + 0x28]	! arg3, &prom_mmu_ihandle_cache
+	mov	4, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x30]	! arg4, sizeof(arg3)
+	stx	%g0, [%sp + 2047 + 128 + 0x38]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	mov	(1b - prom_callmethod_name), %l1
+	mov	(1b - prom_translate_name), %l2
+	sub	%l0, %l1, %l1
+	sub	%l0, %l2, %l2
+	lduw	[%l5], %l5			! prom_mmu_ihandle_cache
+
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "call-method"
+	mov	3, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 3
+	mov	5, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 5
+	stx	%l2, [%sp + 2047 + 128 + 0x18]	! arg1: "translate"
+	stx	%l5, [%sp + 2047 + 128 + 0x20]	! arg2: prom_mmu_ihandle_cache
+	/* PAGE align */
+	srlx	%l0, 13, %l3
+	sllx	%l3, 13, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x28]	! arg3: vaddr, our PC
+	stx	%g0, [%sp + 2047 + 128 + 0x30]	! res1
+	stx	%g0, [%sp + 2047 + 128 + 0x38]	! res2
+	stx	%g0, [%sp + 2047 + 128 + 0x40]	! res3
+	stx	%g0, [%sp + 2047 + 128 + 0x48]	! res4
+	stx	%g0, [%sp + 2047 + 128 + 0x50]	! res5
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	ldx	[%sp + 2047 + 128 + 0x40], %l1	! translation mode
+	mov	(1b - prom_boot_mapping_mode), %l4
+	sub	%l0, %l4, %l4
+	stw	%l1, [%l4]
+	mov	(1b - prom_boot_mapping_phys_high), %l4
+	sub	%l0, %l4, %l4
+	ldx	[%sp + 2047 + 128 + 0x48], %l2	! physaddr high
+	stx	%l2, [%l4 + 0x0]
+	ldx	[%sp + 2047 + 128 + 0x50], %l3	! physaddr low
+	/* 4MB align */
+	srlx	%l3, 22, %l3
+	sllx	%l3, 22, %l3
+	stx	%l3, [%l4 + 0x8]
+
+	/* Leave service as-is, "call-method" */
+	mov	7, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 7
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	mov	(1b - prom_map_name), %l3
+	sub	%l0, %l3, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x18]	! arg1: "map"
+	/* Leave arg2 as-is, prom_mmu_ihandle_cache */
+	mov	-1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x28]	! arg3: mode (-1 default)
+	sethi	%hi(8 * 1024 * 1024), %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x30]	! arg4: size (8MB)
+	sethi	%hi(KERNBASE), %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x38]	! arg5: vaddr (KERNBASE)
+	stx	%g0, [%sp + 2047 + 128 + 0x40]	! arg6: empty
+	mov	(1b - prom_boot_mapping_phys_low), %l3
+	sub	%l0, %l3, %l3
+	ldx	[%l3], %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x48]	! arg7: phys addr
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	add	%sp, (192 + 128), %sp
+
+sparc64_boot_after_remap:
 	BRANCH_IF_CHEETAH_BASE(g1,g7,cheetah_boot)
 	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,cheetah_plus_boot)
 	ba,pt	%xcc, spitfire_boot
@@ -125,185 +274,7 @@ cheetah_generic_boot:
 	stxa	%g0, [%g3] ASI_IMMU
 	membar	#Sync
 
-	wrpr    %g0, (PSTATE_PRIV|PSTATE_PEF|PSTATE_IE), %pstate
-	wr	%g0, 0, %fprs
-
-	/* Just like for Spitfire, we probe itlb-2 for a mapping which
-	 * matches our current %pc.  We take the physical address in
-	 * that mapping and use it to make our own.
-	 */
-
-	/* %g5 holds the tlb data */
-        sethi   %uhi(_PAGE_VALID | _PAGE_SZ4MB), %g5
-        sllx    %g5, 32, %g5
-        or      %g5, (_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_L | _PAGE_W | _PAGE_G), %g5
-
-	/* Put PADDR tlb data mask into %g3. */
-	sethi	%uhi(_PAGE_PADDR), %g3
-	or	%g3, %ulo(_PAGE_PADDR), %g3
-	sllx	%g3, 32, %g3
-	sethi	%hi(_PAGE_PADDR), %g7
-	or	%g7, %lo(_PAGE_PADDR), %g7
-	or	%g3, %g7, %g3
-
-	set	2 << 16, %l0		/* TLB entry walker. */
-	set	0x1fff, %l2		/* Page mask. */
-	rd	%pc, %l3
-	andn	%l3, %l2, %g2		/* vaddr comparator */
-
-1:	ldxa	[%l0] ASI_ITLB_TAG_READ, %g1
-	membar	#Sync
-	andn	%g1, %l2, %g1
-	cmp	%g1, %g2
-	be,pn	%xcc, cheetah_got_tlbentry
-	 nop
-	and	%l0, (127 << 3), %g1
-	cmp	%g1, (127 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-	/* Search the small TLB.  OBP never maps us like that but
-	 * newer SILO can.
-	 */
-	clr	%l0
-
-1:	ldxa	[%l0] ASI_ITLB_TAG_READ, %g1
-	membar	#Sync
-	andn	%g1, %l2, %g1
-	cmp	%g1, %g2
-	be,pn	%xcc, cheetah_got_tlbentry
-	 nop
-	cmp	%l0, (15 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-	/* BUG() if we get here... */
-	ta	0x5
-
-cheetah_got_tlbentry:
-	ldxa	[%l0] ASI_ITLB_DATA_ACCESS, %g0
-	ldxa	[%l0] ASI_ITLB_DATA_ACCESS, %g1
-	membar	#Sync
-	and	%g1, %g3, %g1
-	set	0x5fff, %l0
-	andn	%g1, %l0, %g1
-	or	%g5, %g1, %g5
-
-	/* Clear out any KERNBASE area entries. */
-	set	2 << 16, %l0
-	sethi	%hi(KERNBASE), %g3
-	sethi	%hi(KERNBASE<<1), %g7
-	mov	TLB_TAG_ACCESS, %l7
-
-	/* First, check ITLB */
-1:	ldxa	[%l0] ASI_ITLB_TAG_READ, %g1
-	membar	#Sync
-	andn	%g1, %l2, %g1
-	cmp	%g1, %g3
-	blu,pn	%xcc, 2f
-	 cmp	%g1, %g7
-	bgeu,pn	%xcc, 2f
-	 nop
-	stxa	%g0, [%l7] ASI_IMMU
-	membar	#Sync
-	stxa	%g0, [%l0] ASI_ITLB_DATA_ACCESS
-	membar	#Sync
-
-2:	and	%l0, (127 << 3), %g1
-	cmp	%g1, (127 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-	/* Next, check DTLB */
-	set	2 << 16, %l0
-1:	ldxa	[%l0] ASI_DTLB_TAG_READ, %g1
-	membar	#Sync
-	andn	%g1, %l2, %g1
-	cmp	%g1, %g3
-	blu,pn	%xcc, 2f
-	 cmp	%g1, %g7
-	bgeu,pn	%xcc, 2f
-	 nop
-	stxa	%g0, [%l7] ASI_DMMU
-	membar	#Sync
-	stxa	%g0, [%l0] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-	
-2:	and	%l0, (511 << 3), %g1
-	cmp	%g1, (511 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-	/* On Cheetah+, have to check second DTLB.  */
-	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,l0,2f)
-	ba,pt	%xcc, 9f
-	 nop
-
-2:	set	3 << 16, %l0
-1:	ldxa	[%l0] ASI_DTLB_TAG_READ, %g1
-	membar	#Sync
-	andn	%g1, %l2, %g1
-	cmp	%g1, %g3
-	blu,pn	%xcc, 2f
-	 cmp	%g1, %g7
-	bgeu,pn	%xcc, 2f
-	 nop
-	stxa	%g0, [%l7] ASI_DMMU
-	membar	#Sync
-	stxa	%g0, [%l0] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-	
-2:	and	%l0, (511 << 3), %g1
-	cmp	%g1, (511 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-9:
-
-	/* Now lock the TTE we created into ITLB-0 and DTLB-0,
-	 * entry 15 (and maybe 14 too).
-	 */
-	sethi	%hi(KERNBASE), %g3
-	set	(0 << 16) | (15 << 3), %g7
-	stxa	%g3, [%l7] ASI_DMMU
-	membar	#Sync
-	stxa	%g5, [%g7] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-	stxa	%g3, [%l7] ASI_IMMU
-	membar	#Sync
-	stxa	%g5, [%g7] ASI_ITLB_DATA_ACCESS
-	membar	#Sync
-	flush	%g3
-	membar	#Sync
-	sethi	%hi(_end), %g3			/* Check for bigkernel case */
-	or	%g3, %lo(_end), %g3
-	srl	%g3, 23, %g3			/* Check if _end > 8M */
-	brz,pt	%g3, 1f
-	 sethi	%hi(KERNBASE), %g3		/* Restore for fixup code below */
-	sethi	%hi(0x400000), %g3
-	or	%g3, %lo(0x400000), %g3
-	add	%g5, %g3, %g5			/* New tte data */
-	andn	%g5, (_PAGE_G), %g5
-	sethi	%hi(KERNBASE+0x400000), %g3
-	or	%g3, %lo(KERNBASE+0x400000), %g3
-	set	(0 << 16) | (14 << 3), %g7
-	stxa	%g3, [%l7] ASI_DMMU
-	membar	#Sync
-	stxa	%g5, [%g7] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-	stxa	%g3, [%l7] ASI_IMMU
-	membar	#Sync
-	stxa	%g5, [%g7] ASI_ITLB_DATA_ACCESS
-	membar	#Sync
-	flush	%g3
-	membar	#Sync
-	sethi	%hi(KERNBASE), %g3		/* Restore for fixup code below */
-	ba,pt	%xcc, 1f
-	 nop
-
-1:	set	sun4u_init, %g2
-	jmpl    %g2 + %g0, %g0
-	 nop
+	ba,a,pt	%xcc, jump_to_sun4u_init
 
 spitfire_boot:
 	/* Typically PROM has already enabled both MMU's and both on-chip
@@ -313,6 +284,7 @@ spitfire_boot:
 	stxa	%g1, [%g0] ASI_LSU_CONTROL
 	membar	#Sync
 
+jump_to_sun4u_init:
 	/*
 	 * Make sure we are in privileged mode, have address masking,
          * using the ordinary globals and have enabled floating
@@ -324,151 +296,6 @@ spitfire_boot:
 	wrpr    %g0, (PSTATE_PRIV|PSTATE_PEF|PSTATE_IE), %pstate
 	wr	%g0, 0, %fprs
 
-spitfire_create_mappings:
-	/* %g5 holds the tlb data */
-        sethi   %uhi(_PAGE_VALID | _PAGE_SZ4MB), %g5
-        sllx    %g5, 32, %g5
-        or      %g5, (_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_L | _PAGE_W | _PAGE_G), %g5
-
-	/* Base of physical memory cannot reliably be assumed to be
-	 * at 0x0!  Figure out where it happens to be. -DaveM
-	 */
-
-	/* Put PADDR tlb data mask into %g3. */
-	sethi	%uhi(_PAGE_PADDR_SF), %g3
-	or	%g3, %ulo(_PAGE_PADDR_SF), %g3
-	sllx	%g3, 32, %g3
-	sethi	%hi(_PAGE_PADDR_SF), %g7
-	or	%g7, %lo(_PAGE_PADDR_SF), %g7
-	or	%g3, %g7, %g3
-
-	/* Walk through entire ITLB, looking for entry which maps
-	 * our %pc currently, stick PADDR from there into %g5 tlb data.
-	 */
-	clr	%l0			/* TLB entry walker. */
-	set	0x1fff, %l2		/* Page mask. */
-	rd	%pc, %l3
-	andn	%l3, %l2, %g2		/* vaddr comparator */
-1:
-	/* Yes, the nops seem to be necessary for now, don't ask me why. -DaveM */
-	ldxa	[%l0] ASI_ITLB_TAG_READ, %g1
-	nop
-	nop
-	nop
-	andn	%g1, %l2, %g1		/* Get vaddr */
-	cmp	%g1, %g2
-	be,a,pn	%xcc, spitfire_got_tlbentry
-	 ldxa	[%l0] ASI_ITLB_DATA_ACCESS, %g1
-	cmp	%l0, (63 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-	/* BUG() if we get here... */
-	ta	0x5
-
-spitfire_got_tlbentry:
-	/* Nops here again, perhaps Cheetah/Blackbird are better behaved... */
-	nop
-	nop
-	nop
-	and	%g1, %g3, %g1		/* Mask to just get paddr bits.       */
-	set	0x5fff, %l3		/* Mask offset to get phys base.      */
-	andn	%g1, %l3, %g1
-
-	/* NOTE: We hold on to %g1 paddr base as we need it below to lock
-	 * NOTE: the PROM cif code into the TLB.
-	 */
-
-	or	%g5, %g1, %g5		/* Or it into TAG being built.        */
-
-	clr	%l0			/* TLB entry walker. */
-	sethi	%hi(KERNBASE), %g3	/* 4M lower limit */
-	sethi	%hi(KERNBASE<<1), %g7	/* 8M upper limit */
-	mov	TLB_TAG_ACCESS, %l7
-1:
-	/* Yes, the nops seem to be necessary for now, don't ask me why. -DaveM */
-	ldxa	[%l0] ASI_ITLB_TAG_READ, %g1
-	nop
-	nop
-	nop
-	andn	%g1, %l2, %g1		/* Get vaddr */
-	cmp	%g1, %g3
-	blu,pn	%xcc, 2f
-	 cmp	%g1, %g7
-	bgeu,pn	%xcc, 2f
-	 nop
-	stxa	%g0, [%l7] ASI_IMMU
-	stxa	%g0, [%l0] ASI_ITLB_DATA_ACCESS
-	membar	#Sync
-2:
-	cmp	%l0, (63 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-	nop; nop; nop
-
-	clr	%l0			/* TLB entry walker. */
-1:
-	/* Yes, the nops seem to be necessary for now, don't ask me why. -DaveM */
-	ldxa	[%l0] ASI_DTLB_TAG_READ, %g1
-	nop
-	nop
-	nop
-	andn	%g1, %l2, %g1		/* Get vaddr */
-	cmp	%g1, %g3
-	blu,pn	%xcc, 2f
-	 cmp	%g1, %g7
-	bgeu,pn	%xcc, 2f
-	 nop
-	stxa	%g0, [%l7] ASI_DMMU
-	stxa	%g0, [%l0] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-2:
-	cmp	%l0, (63 << 3)
-	blu,pt	%xcc, 1b
-	 add	%l0, (1 << 3), %l0
-
-	nop; nop; nop
-
-
-	/* PROM never puts any TLB entries into the MMU with the lock bit
-	 * set.  So we gladly use tlb entry 63 for KERNBASE. And maybe 62 too.
-	 */
-
-	sethi	%hi(KERNBASE), %g3
-	mov	(63 << 3), %g7
-	stxa	%g3, [%l7] ASI_DMMU		/* KERNBASE into TLB TAG	*/
-	stxa	%g5, [%g7] ASI_DTLB_DATA_ACCESS	/* TTE into TLB DATA		*/
-	membar	#Sync
-	stxa	%g3, [%l7] ASI_IMMU		/* KERNBASE into TLB TAG	*/
-	stxa	%g5, [%g7] ASI_ITLB_DATA_ACCESS	/* TTE into TLB DATA		*/
-	membar	#Sync
-	flush	%g3
-	membar	#Sync
-	sethi	%hi(_end), %g3			/* Check for bigkernel case */
-	or	%g3, %lo(_end), %g3
-	srl	%g3, 23, %g3			/* Check if _end > 8M */
-	brz,pt	%g3, 2f
-	 sethi	%hi(KERNBASE), %g3		/* Restore for fixup code below */
-	sethi	%hi(0x400000), %g3
-	or	%g3, %lo(0x400000), %g3
-	add	%g5, %g3, %g5			/* New tte data */
-	andn	%g5, (_PAGE_G), %g5
-	sethi	%hi(KERNBASE+0x400000), %g3
-	or	%g3, %lo(KERNBASE+0x400000), %g3
-	mov	(62 << 3), %g7
-	stxa	%g3, [%l7] ASI_DMMU
-	stxa	%g5, [%g7] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-	stxa	%g3, [%l7] ASI_IMMU
-	stxa	%g5, [%g7] ASI_ITLB_DATA_ACCESS
-	membar	#Sync
-	flush	%g3
-	membar	#Sync
-	sethi	%hi(KERNBASE), %g3		/* Restore for fixup code below */
-2:	ba,pt	%xcc, 1f
-	 nop
-1:
 	set	sun4u_init, %g2
 	jmpl    %g2 + %g0, %g0
 	 nop
@@ -483,38 +310,12 @@ sun4u_init:
 	stxa	%g0, [%g7] ASI_DMMU
 	membar	#Sync
 
-	/* We are now safely (we hope) in Nucleus context (0), rewrite
-	 * the KERNBASE TTE's so they no longer have the global bit set.
-	 * Don't forget to setup TAG_ACCESS first 8-)
-	 */
-	mov	TLB_TAG_ACCESS, %g2
-	stxa	%g3, [%g2] ASI_IMMU
-	stxa	%g3, [%g2] ASI_DMMU
-	membar	#Sync
-
 	BRANCH_IF_ANY_CHEETAH(g1,g7,cheetah_tlb_fixup)
 
 	ba,pt	%xcc, spitfire_tlb_fixup
 	 nop
 
 cheetah_tlb_fixup:
-	set	(0 << 16) | (15 << 3), %g7
-	ldxa	[%g7] ASI_ITLB_DATA_ACCESS, %g0
-	ldxa	[%g7] ASI_ITLB_DATA_ACCESS, %g1
-	andn	%g1, (_PAGE_G), %g1
-	stxa	%g1, [%g7] ASI_ITLB_DATA_ACCESS
-	membar	#Sync
-
-	ldxa	[%g7] ASI_DTLB_DATA_ACCESS, %g0
-	ldxa	[%g7] ASI_DTLB_DATA_ACCESS, %g1
-	andn	%g1, (_PAGE_G), %g1
-	stxa	%g1, [%g7] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-
-	/* Kill instruction prefetch queues. */
-	flush	%g3
-	membar	#Sync
-
 	mov	2, %g2		/* Set TLB type to cheetah+. */
 	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f)
 
@@ -523,23 +324,7 @@ cheetah_tlb_fixup:
 1:	sethi	%hi(tlb_type), %g1
 	stw	%g2, [%g1 + %lo(tlb_type)]
 
-	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f)
-	ba,pt	%xcc, 2f
-	 nop
-
-1:	/* Patch context register writes to support nucleus page
-	 * size correctly.
-	 */
-	call	cheetah_plus_patch_etrap
-	 nop
-	call	cheetah_plus_patch_rtrap
-	 nop
-	call	cheetah_plus_patch_fpdis
-	 nop
-	call	cheetah_plus_patch_winfixup
-	 nop
-
-2:	/* Patch copy/page operations to cheetah optimized versions. */
+	/* Patch copy/page operations to cheetah optimized versions. */
 	call	cheetah_patch_copyops
 	 nop
 	call	cheetah_patch_copy_page
@@ -551,21 +336,6 @@ cheetah_tlb_fixup:
 	 nop
 
 spitfire_tlb_fixup:
-	mov	(63 << 3), %g7
-	ldxa	[%g7] ASI_ITLB_DATA_ACCESS, %g1
-	andn	%g1, (_PAGE_G), %g1
-	stxa	%g1, [%g7] ASI_ITLB_DATA_ACCESS
-	membar	#Sync
-
-	ldxa	[%g7] ASI_DTLB_DATA_ACCESS, %g1
-	andn	%g1, (_PAGE_G), %g1
-	stxa	%g1, [%g7] ASI_DTLB_DATA_ACCESS
-	membar	#Sync
-
-	/* Kill instruction prefetch queues. */
-	flush	%g3
-	membar	#Sync
-
 	/* Set TLB type to spitfire. */
 	mov	0, %g2
 	sethi	%hi(tlb_type), %g1
@@ -578,24 +348,6 @@ tlb_fixup_done:
 	mov	%sp, %l6
 	mov	%o4, %l7
 
-#if 0	/* We don't do it like this anymore, but for historical hack value
-	 * I leave this snippet here to show how crazy we can be sometimes. 8-)
-	 */
-
-	/* Setup "Linux Current Register", thanks Sun 8-) */
-	wr	%g0, 0x1, %pcr
-
-	/* Blackbird errata workaround.  See commentary in
-	 * smp.c:smp_percpu_timer_interrupt() for more
-	 * information.
-	 */
-	ba,pt	%xcc, 99f
-	 nop
-	.align	64
-99:	wr	%g6, %g0, %pic
-	rd	%pic, %g0
-#endif
-
 	wr	%g0, ASI_P, %asi
 	mov	1, %g1
 	sllx	%g1, THREAD_SHIFT, %g1
@@ -629,32 +381,78 @@ tlb_fixup_done:
 	 nop
 	/* Not reached... */
 
-/* IMPORTANT NOTE: Whenever making changes here, check
- * trampoline.S as well. -jj */
-	.globl	setup_tba
-setup_tba:	/* i0 = is_starfire */
-	save	%sp, -160, %sp
+	/* This is meant to allow the sharing of this code between
+	 * boot processor invocation (via setup_tba() below) and
+	 * secondary processor startup (via trampoline.S).  The
+	 * former does use this code, the latter does not yet due
+	 * to some complexities.  That should be fixed up at some
+	 * point.
+	 *
+	 * There used to be enormous complexity wrt. transferring
+	 * over from the firwmare's trap table to the Linux kernel's.
+	 * For example, there was a chicken & egg problem wrt. building
+	 * the OBP page tables, yet needing to be on the Linux kernel
+	 * trap table (to translate PAGE_OFFSET addresses) in order to
+	 * do that.
+	 *
+	 * We now handle OBP tlb misses differently, via linear lookups
+	 * into the prom_trans[] array.  So that specific problem no
+	 * longer exists.  Yet, unfortunately there are still some issues
+	 * preventing trampoline.S from using this code... ho hum.
+	 */
+	.globl	setup_trap_table
+setup_trap_table:
+	save	%sp, -192, %sp
 
-	rdpr	%tba, %g7
-	sethi	%hi(prom_tba), %o1
-	or	%o1, %lo(prom_tba), %o1
-	stx	%g7, [%o1]
+	/* Force interrupts to be disabled. */
+	rdpr	%pstate, %o1
+	andn	%o1, PSTATE_IE, %o1
+	wrpr	%o1, 0x0, %pstate
+	wrpr	%g0, 15, %pil
+
+	/* Make the firmware call to jump over to the Linux trap table.  */
+	call	prom_set_trap_table
+	 sethi	%hi(sparc64_ttable_tl0), %o0
+
+	/* Start using proper page size encodings in ctx register.  */
+	sethi	%hi(sparc64_kern_pri_context), %g3
+	ldx	[%g3 + %lo(sparc64_kern_pri_context)], %g2
+	mov	PRIMARY_CONTEXT, %g1
+	stxa	%g2, [%g1] ASI_DMMU
+	membar	#Sync
+
+	/* The Linux trap handlers expect various trap global registers
+	 * to be setup with some fixed values.  So here we set these
+	 * up very carefully.  These globals are:
+	 *
+	 * Alternate Globals (PSTATE_AG):
+	 *
+	 * %g6			--> current_thread_info()
+	 *
+	 * MMU Globals (PSTATE_MG):
+	 *
+	 * %g1			--> TLB_SFSR
+	 * %g2			--> ((_PAGE_VALID | _PAGE_SZ4MB |
+	 *			      _PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W)
+	 *			     ^ 0xfffff80000000000)
+	 * (this %g2 value is used for computing the PAGE_OFFSET kernel
+	 *  TLB entries quickly, the virtual address of the fault XOR'd
+	 *  with this %g2 value is the PTE to load into the TLB)
+	 * %g3			--> VPTE_BASE_CHEETAH or VPTE_BASE_SPITFIRE
+	 *
+	 * Interrupt Globals (PSTATE_IG, setup by init_irqwork_curcpu()):
+	 *
+	 * %g6			--> __irq_work[smp_processor_id()]
+	 */
 
-	/* Setup "Linux" globals 8-) */
 	rdpr	%pstate, %o1
 	mov	%g6, %o2
-	wrpr	%o1, (PSTATE_AG|PSTATE_IE), %pstate
-	sethi	%hi(sparc64_ttable_tl0), %g1
-	wrpr	%g1, %tba
+	wrpr	%o1, PSTATE_AG, %pstate
 	mov	%o2, %g6
 
-	/* Set up MMU globals */
-	wrpr	%o1, (PSTATE_MG|PSTATE_IE), %pstate
-
-	/* Set fixed globals used by dTLB miss handler. */
 #define KERN_HIGHBITS		((_PAGE_VALID|_PAGE_SZ4MB)^0xfffff80000000000)
 #define KERN_LOWBITS		(_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W)
-
+	wrpr	%o1, PSTATE_MG, %pstate
 	mov	TSB_REG, %g1
 	stxa	%g0, [%g1] ASI_DMMU
 	membar	#Sync
@@ -666,17 +464,17 @@ setup_tba:	/* i0 = is_starfire */
 	sllx	%g2, 32, %g2
 	or	%g2, KERN_LOWBITS, %g2
 
-	BRANCH_IF_ANY_CHEETAH(g3,g7,cheetah_vpte_base)
-	ba,pt	%xcc, spitfire_vpte_base
+	BRANCH_IF_ANY_CHEETAH(g3,g7,8f)
+	ba,pt	%xcc, 9f
 	 nop
 
-cheetah_vpte_base:
+8:
 	sethi		%uhi(VPTE_BASE_CHEETAH), %g3
 	or		%g3, %ulo(VPTE_BASE_CHEETAH), %g3
 	ba,pt		%xcc, 2f
 	 sllx		%g3, 32, %g3
 
-spitfire_vpte_base:
+9:
 	sethi		%uhi(VPTE_BASE_SPITFIRE), %g3
 	or		%g3, %ulo(VPTE_BASE_SPITFIRE), %g3
 	sllx		%g3, 32, %g3
@@ -702,48 +500,55 @@ spitfire_vpte_base:
 	sllx	%o2, 32, %o2
 	wr	%o2, %asr25
 
-	/* Ok, we're done setting up all the state our trap mechanims needs,
-	 * now get back into normal globals and let the PROM know what is up.
-	 */
 2:
 	wrpr	%g0, %g0, %wstate
-	wrpr	%o1, PSTATE_IE, %pstate
+	wrpr	%o1, 0x0, %pstate
 
 	call	init_irqwork_curcpu
 	 nop
 
-	call	prom_set_trap_table
-	 sethi	%hi(sparc64_ttable_tl0), %o0
-
-	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g2,g3,1f)
-	ba,pt	%xcc, 2f
-	 nop
-
-1:	/* Start using proper page size encodings in ctx register.  */
-	sethi	%uhi(CTX_CHEETAH_PLUS_NUC), %g3
-	mov	PRIMARY_CONTEXT, %g1
-	sllx	%g3, 32, %g3
-	sethi	%hi(CTX_CHEETAH_PLUS_CTX0), %g2
-	or	%g3, %g2, %g3
-	stxa	%g3, [%g1] ASI_DMMU
-	membar	#Sync
-
-2:
+	/* Now we can turn interrupts back on. */
 	rdpr	%pstate, %o1
 	or	%o1, PSTATE_IE, %o1
 	wrpr	%o1, 0, %pstate
+	wrpr	%g0, 0x0, %pil
+
+	ret
+	 restore
+
+	.globl	setup_tba
+setup_tba:	/* i0 = is_starfire */
+	save	%sp, -192, %sp
+
+	/* The boot processor is the only cpu which invokes this
+	 * routine, the other cpus set things up via trampoline.S.
+	 * So save the OBP trap table address here.
+	 */
+	rdpr	%tba, %g7
+	sethi	%hi(prom_tba), %o1
+	or	%o1, %lo(prom_tba), %o1
+	stx	%g7, [%o1]
+
+	call	setup_trap_table
+	 nop
 
 	ret
 	 restore
+sparc64_boot_end:
+
+#include "systbls.S"
+#include "ktlb.S"
+#include "etrap.S"
+#include "rtrap.S"
+#include "winfixup.S"
+#include "entry.S"
 
 /*
- * The following skips make sure the trap table in ttable.S is aligned
+ * The following skip makes sure the trap table in ttable.S is aligned
  * on a 32K boundary as required by the v9 specs for TBA register.
  */
-sparc64_boot_end:
-	.skip	0x2000 + _start - sparc64_boot_end
-bootup_user_stack_end:
-	.skip	0x2000
+1:
+	.skip	0x4000 + _start - 1b
 
 #ifdef CONFIG_SBUS
 /* This is just a hack to fool make depend config.h discovering
@@ -755,20 +560,6 @@ bootup_user_stack_end:
 ! 0x0000000000408000
 
 #include "ttable.S"
-#include "systbls.S"
-
-	.align	1024
-	.globl	swapper_pg_dir
-swapper_pg_dir:
-	.word	0
-
-#include "etrap.S"
-#include "rtrap.S"
-#include "winfixup.S"
-#include "entry.S"
-
-	/* This is just anal retentiveness on my part... */
-	.align	16384
 
 	.data
 	.align	8
@@ -776,8 +567,11 @@ swapper_pg_dir:
 prom_tba:	.xword	0
 tlb_type:	.word	0	/* Must NOT end up in BSS */
 	.section	".fixup",#alloc,#execinstr
-	.globl	__ret_efault
+
+	.globl	__ret_efault, __retl_efault
 __ret_efault:
 	ret
 	 restore %g0, -EFAULT, %o0
-
+__retl_efault:
+	retl
+	 mov	-EFAULT, %o0
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index c9b69167632..233526ba3ab 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -27,6 +27,7 @@
 #include <asm/atomic.h>
 #include <asm/system.h>
 #include <asm/irq.h>
+#include <asm/io.h>
 #include <asm/sbus.h>
 #include <asm/iommu.h>
 #include <asm/upa.h>
diff --git a/arch/sparc64/kernel/itlb_base.S b/arch/sparc64/kernel/itlb_base.S
index b5e32dfa4fb..4951ff8f687 100644
--- a/arch/sparc64/kernel/itlb_base.S
+++ b/arch/sparc64/kernel/itlb_base.S
@@ -15,14 +15,12 @@
  */
 #define CREATE_VPTE_OFFSET1(r1, r2) \
 				srax	r1, 10, r2
-#define CREATE_VPTE_OFFSET2(r1, r2)
-#define CREATE_VPTE_NOP		nop
+#define CREATE_VPTE_OFFSET2(r1, r2) nop
 #else /* PAGE_SHIFT */
 #define CREATE_VPTE_OFFSET1(r1, r2) \
 				srax	r1, PAGE_SHIFT, r2
 #define CREATE_VPTE_OFFSET2(r1, r2) \
 				sllx	r2, 3, r2
-#define CREATE_VPTE_NOP
 #endif /* PAGE_SHIFT */
 
 
@@ -36,6 +34,7 @@
  */
 
 /* ITLB ** ICACHE line 1: Quick user TLB misses		*/
+	mov		TLB_SFSR, %g1
 	ldxa		[%g1 + %g1] ASI_IMMU, %g4	! Get TAG_ACCESS
 	CREATE_VPTE_OFFSET1(%g4, %g6)			! Create VPTE offset
 	CREATE_VPTE_OFFSET2(%g4, %g6)			! Create VPTE offset
@@ -43,41 +42,38 @@
 1:	brgez,pn	%g5, 3f				! Not valid, branch out
 	 sethi		%hi(_PAGE_EXEC), %g4		! Delay-slot
 	andcc		%g5, %g4, %g0			! Executable?
+
+/* ITLB ** ICACHE line 2: Real faults			*/
 	be,pn		%xcc, 3f			! Nope, branch.
 	 nop						! Delay-slot
 2:	stxa		%g5, [%g0] ASI_ITLB_DATA_IN	! Load PTE into TLB
 	retry						! Trap return
-3:	rdpr		%pstate, %g4			! Move into alternate globals
-
-/* ITLB ** ICACHE line 2: Real faults			*/
+3:	rdpr		%pstate, %g4			! Move into alt-globals
 	wrpr		%g4, PSTATE_AG|PSTATE_MG, %pstate
 	rdpr		%tpc, %g5			! And load faulting VA
 	mov		FAULT_CODE_ITLB, %g4		! It was read from ITLB
-sparc64_realfault_common:				! Called by TL0 dtlb_miss too
+
+/* ITLB ** ICACHE line 3: Finish faults	*/
+sparc64_realfault_common:				! Called by dtlb_miss
 	stb		%g4, [%g6 + TI_FAULT_CODE]
 	stx		%g5, [%g6 + TI_FAULT_ADDR]
 	ba,pt		%xcc, etrap			! Save state
 1:	 rd		%pc, %g7			! ...
-	nop
-
-/* ITLB ** ICACHE line 3: Finish faults + window fixups	*/
 	call		do_sparc64_fault		! Call fault handler
 	 add		%sp, PTREGS_OFF, %o0! Compute pt_regs arg
 	ba,pt		%xcc, rtrap_clr_l6		! Restore cpu state
 	 nop
+
+/* ITLB ** ICACHE line 4: Window fixups */
 winfix_trampoline:
 	rdpr		%tpc, %g3			! Prepare winfixup TNPC
-	or		%g3, 0x7c, %g3			! Compute offset to branch
+	or		%g3, 0x7c, %g3			! Compute branch offset
 	wrpr		%g3, %tnpc			! Write it into TNPC
 	done						! Do it to it
-
-/* ITLB ** ICACHE line 4: Unused...	*/
 	nop
 	nop
 	nop
 	nop
-	CREATE_VPTE_NOP
 
 #undef CREATE_VPTE_OFFSET1
 #undef CREATE_VPTE_OFFSET2
-#undef CREATE_VPTE_NOP
diff --git a/arch/sparc64/kernel/ktlb.S b/arch/sparc64/kernel/ktlb.S
new file mode 100644
index 00000000000..d9244d3c9f7
--- /dev/null
+++ b/arch/sparc64/kernel/ktlb.S
@@ -0,0 +1,194 @@
+/* arch/sparc64/kernel/ktlb.S: Kernel mapping TLB miss handling.
+ *
+ * Copyright (C) 1995, 1997, 2005 David S. Miller <davem@davemloft.net>
+ * Copyright (C) 1996 Eddie C. Dost        (ecd@brainaid.de)
+ * Copyright (C) 1996 Miguel de Icaza      (miguel@nuclecu.unam.mx)
+ * Copyright (C) 1996,98,99 Jakub Jelinek  (jj@sunsite.mff.cuni.cz)
+*/
+
+#include <linux/config.h>
+#include <asm/head.h>
+#include <asm/asi.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+
+	.text
+	.align		32
+
+/*
+ * On a second level vpte miss, check whether the original fault is to the OBP 
+ * range (note that this is only possible for instruction miss, data misses to
+ * obp range do not use vpte). If so, go back directly to the faulting address.
+ * This is because we want to read the tpc, otherwise we have no way of knowing
+ * the 8k aligned faulting address if we are using >8k kernel pagesize. This
+ * also ensures no vpte range addresses are dropped into tlb while obp is
+ * executing (see inherit_locked_prom_mappings() rant).
+ */
+sparc64_vpte_nucleus:
+	/* Note that kvmap below has verified that the address is
+	 * in the range MODULES_VADDR --> VMALLOC_END already.  So
+	 * here we need only check if it is an OBP address or not.
+	 */
+	sethi		%hi(LOW_OBP_ADDRESS), %g5
+	cmp		%g4, %g5
+	blu,pn		%xcc, kern_vpte
+	 mov		0x1, %g5
+	sllx		%g5, 32, %g5
+	cmp		%g4, %g5
+	blu,pn		%xcc, vpte_insn_obp
+	 nop
+
+	/* These two instructions are patched by paginig_init().  */
+kern_vpte:
+	sethi		%hi(swapper_pgd_zero), %g5
+	lduw		[%g5 + %lo(swapper_pgd_zero)], %g5
+
+	/* With kernel PGD in %g5, branch back into dtlb_backend.  */
+	ba,pt		%xcc, sparc64_kpte_continue
+	 andn		%g1, 0x3, %g1	/* Finish PMD offset adjustment.  */
+
+vpte_noent:
+	/* Restore previous TAG_ACCESS, %g5 is zero, and we will
+	 * skip over the trap instruction so that the top level
+	 * TLB miss handler will thing this %g5 value is just an
+	 * invalid PTE, thus branching to full fault processing.
+	 */
+	mov		TLB_SFSR, %g1
+	stxa		%g4, [%g1 + %g1] ASI_DMMU
+	done
+
+vpte_insn_obp:
+	/* Behave as if we are at TL0.  */
+	wrpr		%g0, 1, %tl
+	rdpr		%tpc, %g4	/* Find original faulting iaddr */
+	srlx		%g4, 13, %g4	/* Throw out context bits */
+	sllx		%g4, 13, %g4	/* g4 has vpn + ctx0 now */
+
+	/* Restore previous TAG_ACCESS.  */
+	mov		TLB_SFSR, %g1
+	stxa		%g4, [%g1 + %g1] ASI_IMMU
+
+	sethi		%hi(prom_trans), %g5
+	or		%g5, %lo(prom_trans), %g5
+
+1:	ldx		[%g5 + 0x00], %g6	! base
+	brz,a,pn	%g6, longpath		! no more entries, fail
+	 mov		TLB_SFSR, %g1		! and restore %g1
+	ldx		[%g5 + 0x08], %g1	! len
+	add		%g6, %g1, %g1		! end
+	cmp		%g6, %g4
+	bgu,pt		%xcc, 2f
+	 cmp		%g4, %g1
+	bgeu,pt		%xcc, 2f
+	 ldx		[%g5 + 0x10], %g1	! PTE
+
+	/* TLB load, restore %g1, and return from trap.  */
+	sub		%g4, %g6, %g6
+	add		%g1, %g6, %g5
+	mov		TLB_SFSR, %g1
+	stxa		%g5, [%g0] ASI_ITLB_DATA_IN
+	retry
+
+2:	ba,pt		%xcc, 1b
+	 add		%g5, (3 * 8), %g5	! next entry
+
+kvmap_do_obp:
+	sethi		%hi(prom_trans), %g5
+	or		%g5, %lo(prom_trans), %g5
+	srlx		%g4, 13, %g4
+	sllx		%g4, 13, %g4
+
+1:	ldx		[%g5 + 0x00], %g6	! base
+	brz,a,pn	%g6, longpath		! no more entries, fail
+	 mov		TLB_SFSR, %g1		! and restore %g1
+	ldx		[%g5 + 0x08], %g1	! len
+	add		%g6, %g1, %g1		! end
+	cmp		%g6, %g4
+	bgu,pt		%xcc, 2f
+	 cmp		%g4, %g1
+	bgeu,pt		%xcc, 2f
+	 ldx		[%g5 + 0x10], %g1	! PTE
+
+	/* TLB load, restore %g1, and return from trap.  */
+	sub		%g4, %g6, %g6
+	add		%g1, %g6, %g5
+	mov		TLB_SFSR, %g1
+	stxa		%g5, [%g0] ASI_DTLB_DATA_IN
+	retry
+
+2:	ba,pt		%xcc, 1b
+	 add		%g5, (3 * 8), %g5	! next entry
+
+/*
+ * On a first level data miss, check whether this is to the OBP range (note
+ * that such accesses can be made by prom, as well as by kernel using
+ * prom_getproperty on "address"), and if so, do not use vpte access ...
+ * rather, use information saved during inherit_prom_mappings() using 8k
+ * pagesize.
+ */
+	.align		32
+kvmap:
+	brgez,pn	%g4, kvmap_nonlinear
+	 nop
+
+#ifdef CONFIG_DEBUG_PAGEALLOC
+	.globl		kvmap_linear_patch
+kvmap_linear_patch:
+#endif
+	ba,pt		%xcc, kvmap_load
+	 xor		%g2, %g4, %g5
+
+#ifdef CONFIG_DEBUG_PAGEALLOC
+	sethi		%hi(swapper_pg_dir), %g5
+	or		%g5, %lo(swapper_pg_dir), %g5
+	sllx		%g4, 64 - (PGDIR_SHIFT + PGDIR_BITS), %g6
+	srlx		%g6, 64 - PAGE_SHIFT, %g6
+	andn		%g6, 0x3, %g6
+	lduw		[%g5 + %g6], %g5
+	brz,pn		%g5, longpath
+	 sllx		%g4, 64 - (PMD_SHIFT + PMD_BITS), %g6
+	srlx		%g6, 64 - PAGE_SHIFT, %g6
+	sllx		%g5, 11, %g5
+	andn		%g6, 0x3, %g6
+	lduwa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
+	brz,pn		%g5, longpath
+	 sllx		%g4, 64 - PMD_SHIFT, %g6
+	srlx		%g6, 64 - PAGE_SHIFT, %g6
+	sllx		%g5, 11, %g5
+	andn		%g6, 0x7, %g6
+	ldxa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
+	brz,pn		%g5, longpath
+	 nop
+	ba,a,pt		%xcc, kvmap_load
+#endif
+
+kvmap_nonlinear:
+	sethi		%hi(MODULES_VADDR), %g5
+	cmp		%g4, %g5
+	blu,pn		%xcc, longpath
+	 mov		(VMALLOC_END >> 24), %g5
+	sllx		%g5, 24, %g5
+	cmp		%g4, %g5
+	bgeu,pn		%xcc, longpath
+	 nop
+
+kvmap_check_obp:
+	sethi		%hi(LOW_OBP_ADDRESS), %g5
+	cmp		%g4, %g5
+	blu,pn		%xcc, kvmap_vmalloc_addr
+	 mov		0x1, %g5
+	sllx		%g5, 32, %g5
+	cmp		%g4, %g5
+	blu,pn		%xcc, kvmap_do_obp
+	 nop
+
+kvmap_vmalloc_addr:
+	/* If we get here, a vmalloc addr was accessed, load kernel VPTE.  */
+	ldxa		[%g3 + %g6] ASI_N, %g5
+	brgez,pn	%g5, longpath
+	 nop
+
+kvmap_load:
+	/* PTE is valid, load into TLB and return from trap.  */
+	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Reload TLB
+	retry
diff --git a/arch/sparc64/kernel/pci_iommu.c b/arch/sparc64/kernel/pci_iommu.c
index 425c60cfea1..a11910be101 100644
--- a/arch/sparc64/kernel/pci_iommu.c
+++ b/arch/sparc64/kernel/pci_iommu.c
@@ -49,12 +49,6 @@ static void __iommu_flushall(struct pci_iommu *iommu)
 
 	/* Ensure completion of previous PIO writes. */
 	(void) pci_iommu_read(iommu->write_complete_reg);
-
-	/* Now update everyone's flush point. */
-	for (entry = 0; entry < PBM_NCLUSTERS; entry++) {
-		iommu->alloc_info[entry].flush =
-			iommu->alloc_info[entry].next;
-	}
 }
 
 #define IOPTE_CONSISTENT(CTX) \
@@ -80,120 +74,117 @@ static void inline iopte_make_dummy(struct pci_iommu *iommu, iopte_t *iopte)
 	iopte_val(*iopte) = val;
 }
 
-void pci_iommu_table_init(struct pci_iommu *iommu, int tsbsize)
+/* Based largely upon the ppc64 iommu allocator.  */
+static long pci_arena_alloc(struct pci_iommu *iommu, unsigned long npages)
 {
-	int i;
-
-	tsbsize /= sizeof(iopte_t);
-
-	for (i = 0; i < tsbsize; i++)
-		iopte_make_dummy(iommu, &iommu->page_table[i]);
-}
-
-static iopte_t *alloc_streaming_cluster(struct pci_iommu *iommu, unsigned long npages)
-{
-	iopte_t *iopte, *limit, *first;
-	unsigned long cnum, ent, flush_point;
-
-	cnum = 0;
-	while ((1UL << cnum) < npages)
-		cnum++;
-	iopte  = (iommu->page_table +
-		  (cnum << (iommu->page_table_sz_bits - PBM_LOGCLUSTERS)));
-
-	if (cnum == 0)
-		limit = (iommu->page_table +
-			 iommu->lowest_consistent_map);
-	else
-		limit = (iopte +
-			 (1 << (iommu->page_table_sz_bits - PBM_LOGCLUSTERS)));
-
-	iopte += ((ent = iommu->alloc_info[cnum].next) << cnum);
-	flush_point = iommu->alloc_info[cnum].flush;
-	
-	first = iopte;
-	for (;;) {
-		if (IOPTE_IS_DUMMY(iommu, iopte)) {
-			if ((iopte + (1 << cnum)) >= limit)
-				ent = 0;
-			else
-				ent = ent + 1;
-			iommu->alloc_info[cnum].next = ent;
-			if (ent == flush_point)
-				__iommu_flushall(iommu);
-			break;
+	struct pci_iommu_arena *arena = &iommu->arena;
+	unsigned long n, i, start, end, limit;
+	int pass;
+
+	limit = arena->limit;
+	start = arena->hint;
+	pass = 0;
+
+again:
+	n = find_next_zero_bit(arena->map, limit, start);
+	end = n + npages;
+	if (unlikely(end >= limit)) {
+		if (likely(pass < 1)) {
+			limit = start;
+			start = 0;
+			__iommu_flushall(iommu);
+			pass++;
+			goto again;
+		} else {
+			/* Scanned the whole thing, give up. */
+			return -1;
 		}
-		iopte += (1 << cnum);
-		ent++;
-		if (iopte >= limit) {
-			iopte = (iommu->page_table +
-				 (cnum <<
-				  (iommu->page_table_sz_bits - PBM_LOGCLUSTERS)));
-			ent = 0;
+	}
+
+	for (i = n; i < end; i++) {
+		if (test_bit(i, arena->map)) {
+			start = i + 1;
+			goto again;
 		}
-		if (ent == flush_point)
-			__iommu_flushall(iommu);
-		if (iopte == first)
-			goto bad;
 	}
 
-	/* I've got your streaming cluster right here buddy boy... */
-	return iopte;
+	for (i = n; i < end; i++)
+		__set_bit(i, arena->map);
 
-bad:
-	printk(KERN_EMERG "pci_iommu: alloc_streaming_cluster of npages(%ld) failed!\n",
-	       npages);
-	return NULL;
+	arena->hint = end;
+
+	return n;
 }
 
-static void free_streaming_cluster(struct pci_iommu *iommu, dma_addr_t base,
-				   unsigned long npages, unsigned long ctx)
+static void pci_arena_free(struct pci_iommu_arena *arena, unsigned long base, unsigned long npages)
 {
-	unsigned long cnum, ent;
+	unsigned long i;
 
-	cnum = 0;
-	while ((1UL << cnum) < npages)
-		cnum++;
+	for (i = base; i < (base + npages); i++)
+		__clear_bit(i, arena->map);
+}
 
-	ent = (base << (32 - IO_PAGE_SHIFT + PBM_LOGCLUSTERS - iommu->page_table_sz_bits))
-		>> (32 + PBM_LOGCLUSTERS + cnum - iommu->page_table_sz_bits);
+void pci_iommu_table_init(struct pci_iommu *iommu, int tsbsize, u32 dma_offset, u32 dma_addr_mask)
+{
+	unsigned long i, tsbbase, order, sz, num_tsb_entries;
+
+	num_tsb_entries = tsbsize / sizeof(iopte_t);
+
+	/* Setup initial software IOMMU state. */
+	spin_lock_init(&iommu->lock);
+	iommu->ctx_lowest_free = 1;
+	iommu->page_table_map_base = dma_offset;
+	iommu->dma_addr_mask = dma_addr_mask;
+
+	/* Allocate and initialize the free area map.  */
+	sz = num_tsb_entries / 8;
+	sz = (sz + 7UL) & ~7UL;
+	iommu->arena.map = kmalloc(sz, GFP_KERNEL);
+	if (!iommu->arena.map) {
+		prom_printf("PCI_IOMMU: Error, kmalloc(arena.map) failed.\n");
+		prom_halt();
+	}
+	memset(iommu->arena.map, 0, sz);
+	iommu->arena.limit = num_tsb_entries;
 
-	/* If the global flush might not have caught this entry,
-	 * adjust the flush point such that we will flush before
-	 * ever trying to reuse it.
+	/* Allocate and initialize the dummy page which we
+	 * set inactive IO PTEs to point to.
 	 */
-#define between(X,Y,Z)	(((Z) - (Y)) >= ((X) - (Y)))
-	if (between(ent, iommu->alloc_info[cnum].next, iommu->alloc_info[cnum].flush))
-		iommu->alloc_info[cnum].flush = ent;
-#undef between
+	iommu->dummy_page = __get_free_pages(GFP_KERNEL, 0);
+	if (!iommu->dummy_page) {
+		prom_printf("PCI_IOMMU: Error, gfp(dummy_page) failed.\n");
+		prom_halt();
+	}
+	memset((void *)iommu->dummy_page, 0, PAGE_SIZE);
+	iommu->dummy_page_pa = (unsigned long) __pa(iommu->dummy_page);
+
+	/* Now allocate and setup the IOMMU page table itself.  */
+	order = get_order(tsbsize);
+	tsbbase = __get_free_pages(GFP_KERNEL, order);
+	if (!tsbbase) {
+		prom_printf("PCI_IOMMU: Error, gfp(tsb) failed.\n");
+		prom_halt();
+	}
+	iommu->page_table = (iopte_t *)tsbbase;
+
+	for (i = 0; i < num_tsb_entries; i++)
+		iopte_make_dummy(iommu, &iommu->page_table[i]);
 }
 
-/* We allocate consistent mappings from the end of cluster zero. */
-static iopte_t *alloc_consistent_cluster(struct pci_iommu *iommu, unsigned long npages)
+static inline iopte_t *alloc_npages(struct pci_iommu *iommu, unsigned long npages)
 {
-	iopte_t *iopte;
+	long entry;
 
-	iopte = iommu->page_table + (1 << (iommu->page_table_sz_bits - PBM_LOGCLUSTERS));
-	while (iopte > iommu->page_table) {
-		iopte--;
-		if (IOPTE_IS_DUMMY(iommu, iopte)) {
-			unsigned long tmp = npages;
+	entry = pci_arena_alloc(iommu, npages);
+	if (unlikely(entry < 0))
+		return NULL;
 
-			while (--tmp) {
-				iopte--;
-				if (!IOPTE_IS_DUMMY(iommu, iopte))
-					break;
-			}
-			if (tmp == 0) {
-				u32 entry = (iopte - iommu->page_table);
+	return iommu->page_table + entry;
+}
 
-				if (entry < iommu->lowest_consistent_map)
-					iommu->lowest_consistent_map = entry;
-				return iopte;
-			}
-		}
-	}
-	return NULL;
+static inline void free_npages(struct pci_iommu *iommu, dma_addr_t base, unsigned long npages)
+{
+	pci_arena_free(&iommu->arena, base >> IO_PAGE_SHIFT, npages);
 }
 
 static int iommu_alloc_ctx(struct pci_iommu *iommu)
@@ -233,7 +224,7 @@ void *pci_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_ad
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
 	iopte_t *iopte;
-	unsigned long flags, order, first_page, ctx;
+	unsigned long flags, order, first_page;
 	void *ret;
 	int npages;
 
@@ -251,9 +242,10 @@ void *pci_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_ad
 	iommu = pcp->pbm->iommu;
 
 	spin_lock_irqsave(&iommu->lock, flags);
-	iopte = alloc_consistent_cluster(iommu, size >> IO_PAGE_SHIFT);
-	if (iopte == NULL) {
-		spin_unlock_irqrestore(&iommu->lock, flags);
+	iopte = alloc_npages(iommu, size >> IO_PAGE_SHIFT);
+	spin_unlock_irqrestore(&iommu->lock, flags);
+
+	if (unlikely(iopte == NULL)) {
 		free_pages(first_page, order);
 		return NULL;
 	}
@@ -262,31 +254,15 @@ void *pci_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_ad
 		      ((iopte - iommu->page_table) << IO_PAGE_SHIFT));
 	ret = (void *) first_page;
 	npages = size >> IO_PAGE_SHIFT;
-	ctx = 0;
-	if (iommu->iommu_ctxflush)
-		ctx = iommu_alloc_ctx(iommu);
 	first_page = __pa(first_page);
 	while (npages--) {
-		iopte_val(*iopte) = (IOPTE_CONSISTENT(ctx) |
+		iopte_val(*iopte) = (IOPTE_CONSISTENT(0UL) |
 				     IOPTE_WRITE |
 				     (first_page & IOPTE_PAGE));
 		iopte++;
 		first_page += IO_PAGE_SIZE;
 	}
 
-	{
-		int i;
-		u32 daddr = *dma_addrp;
-
-		npages = size >> IO_PAGE_SHIFT;
-		for (i = 0; i < npages; i++) {
-			pci_iommu_write(iommu->iommu_flush, daddr);
-			daddr += IO_PAGE_SIZE;
-		}
-	}
-
-	spin_unlock_irqrestore(&iommu->lock, flags);
-
 	return ret;
 }
 
@@ -296,7 +272,7 @@ void pci_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_
 	struct pcidev_cookie *pcp;
 	struct pci_iommu *iommu;
 	iopte_t *iopte;
-	unsigned long flags, order, npages, i, ctx;
+	unsigned long flags, order, npages;
 
 	npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
 	pcp = pdev->sysdata;
@@ -306,46 +282,7 @@ void pci_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	if ((iopte - iommu->page_table) ==
-	    iommu->lowest_consistent_map) {
-		iopte_t *walk = iopte + npages;
-		iopte_t *limit;
-
-		limit = (iommu->page_table +
-			 (1 << (iommu->page_table_sz_bits - PBM_LOGCLUSTERS)));
-		while (walk < limit) {
-			if (!IOPTE_IS_DUMMY(iommu, walk))
-				break;
-			walk++;
-		}
-		iommu->lowest_consistent_map =
-			(walk - iommu->page_table);
-	}
-
-	/* Data for consistent mappings cannot enter the streaming
-	 * buffers, so we only need to update the TSB.  We flush
-	 * the IOMMU here as well to prevent conflicts with the
-	 * streaming mapping deferred tlb flush scheme.
-	 */
-
-	ctx = 0;
-	if (iommu->iommu_ctxflush)
-		ctx = (iopte_val(*iopte) & IOPTE_CONTEXT) >> 47UL;
-
-	for (i = 0; i < npages; i++, iopte++)
-		iopte_make_dummy(iommu, iopte);
-
-	if (iommu->iommu_ctxflush) {
-		pci_iommu_write(iommu->iommu_ctxflush, ctx);
-	} else {
-		for (i = 0; i < npages; i++) {
-			u32 daddr = dvma + (i << IO_PAGE_SHIFT);
-
-			pci_iommu_write(iommu->iommu_flush, daddr);
-		}
-	}
-
-	iommu_free_ctx(iommu, ctx);
+	free_npages(iommu, dvma, npages);
 
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
@@ -372,25 +309,27 @@ dma_addr_t pci_map_single(struct pci_dev *pdev, void *ptr, size_t sz, int direct
 	iommu = pcp->pbm->iommu;
 	strbuf = &pcp->pbm->stc;
 
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	if (unlikely(direction == PCI_DMA_NONE))
+		goto bad_no_ctx;
 
 	oaddr = (unsigned long)ptr;
 	npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
 	npages >>= IO_PAGE_SHIFT;
 
 	spin_lock_irqsave(&iommu->lock, flags);
+	base = alloc_npages(iommu, npages);
+	ctx = 0;
+	if (iommu->iommu_ctxflush)
+		ctx = iommu_alloc_ctx(iommu);
+	spin_unlock_irqrestore(&iommu->lock, flags);
 
-	base = alloc_streaming_cluster(iommu, npages);
-	if (base == NULL)
+	if (unlikely(!base))
 		goto bad;
+
 	bus_addr = (iommu->page_table_map_base +
 		    ((base - iommu->page_table) << IO_PAGE_SHIFT));
 	ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
 	base_paddr = __pa(oaddr & IO_PAGE_MASK);
-	ctx = 0;
-	if (iommu->iommu_ctxflush)
-		ctx = iommu_alloc_ctx(iommu);
 	if (strbuf->strbuf_enabled)
 		iopte_protection = IOPTE_STREAMING(ctx);
 	else
@@ -401,12 +340,13 @@ dma_addr_t pci_map_single(struct pci_dev *pdev, void *ptr, size_t sz, int direct
 	for (i = 0; i < npages; i++, base++, base_paddr += IO_PAGE_SIZE)
 		iopte_val(*base) = iopte_protection | base_paddr;
 
-	spin_unlock_irqrestore(&iommu->lock, flags);
-
 	return ret;
 
 bad:
-	spin_unlock_irqrestore(&iommu->lock, flags);
+	iommu_free_ctx(iommu, ctx);
+bad_no_ctx:
+	if (printk_ratelimit())
+		WARN_ON(1);
 	return PCI_DMA_ERROR_CODE;
 }
 
@@ -481,10 +421,13 @@ void pci_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int
 	struct pci_iommu *iommu;
 	struct pci_strbuf *strbuf;
 	iopte_t *base;
-	unsigned long flags, npages, ctx;
+	unsigned long flags, npages, ctx, i;
 
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	if (unlikely(direction == PCI_DMA_NONE)) {
+		if (printk_ratelimit())
+			WARN_ON(1);
+		return;
+	}
 
 	pcp = pdev->sysdata;
 	iommu = pcp->pbm->iommu;
@@ -510,13 +453,14 @@ void pci_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int
 
 	/* Step 1: Kick data out of streaming buffers if necessary. */
 	if (strbuf->strbuf_enabled)
-		pci_strbuf_flush(strbuf, iommu, bus_addr, ctx, npages, direction);
+		pci_strbuf_flush(strbuf, iommu, bus_addr, ctx,
+				 npages, direction);
 
-	/* Step 2: Clear out first TSB entry. */
-	iopte_make_dummy(iommu, base);
+	/* Step 2: Clear out TSB entries. */
+	for (i = 0; i < npages; i++)
+		iopte_make_dummy(iommu, base + i);
 
-	free_streaming_cluster(iommu, bus_addr - iommu->page_table_map_base,
-			       npages, ctx);
+	free_npages(iommu, bus_addr - iommu->page_table_map_base, npages);
 
 	iommu_free_ctx(iommu, ctx);
 
@@ -621,6 +565,8 @@ int pci_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int
 			pci_map_single(pdev,
 				       (page_address(sglist->page) + sglist->offset),
 				       sglist->length, direction);
+		if (unlikely(sglist->dma_address == PCI_DMA_ERROR_CODE))
+			return 0;
 		sglist->dma_length = sglist->length;
 		return 1;
 	}
@@ -629,21 +575,29 @@ int pci_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int
 	iommu = pcp->pbm->iommu;
 	strbuf = &pcp->pbm->stc;
 	
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	if (unlikely(direction == PCI_DMA_NONE))
+		goto bad_no_ctx;
 
 	/* Step 1: Prepare scatter list. */
 
 	npages = prepare_sg(sglist, nelems);
 
-	/* Step 2: Allocate a cluster. */
+	/* Step 2: Allocate a cluster and context, if necessary. */
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	base = alloc_streaming_cluster(iommu, npages);
+	base = alloc_npages(iommu, npages);
+	ctx = 0;
+	if (iommu->iommu_ctxflush)
+		ctx = iommu_alloc_ctx(iommu);
+
+	spin_unlock_irqrestore(&iommu->lock, flags);
+
 	if (base == NULL)
 		goto bad;
-	dma_base = iommu->page_table_map_base + ((base - iommu->page_table) << IO_PAGE_SHIFT);
+
+	dma_base = iommu->page_table_map_base +
+		((base - iommu->page_table) << IO_PAGE_SHIFT);
 
 	/* Step 3: Normalize DMA addresses. */
 	used = nelems;
@@ -656,30 +610,28 @@ int pci_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int
 	}
 	used = nelems - used;
 
-	/* Step 4: Choose a context if necessary. */
-	ctx = 0;
-	if (iommu->iommu_ctxflush)
-		ctx = iommu_alloc_ctx(iommu);
-
-	/* Step 5: Create the mappings. */
+	/* Step 4: Create the mappings. */
 	if (strbuf->strbuf_enabled)
 		iopte_protection = IOPTE_STREAMING(ctx);
 	else
 		iopte_protection = IOPTE_CONSISTENT(ctx);
 	if (direction != PCI_DMA_TODEVICE)
 		iopte_protection |= IOPTE_WRITE;
-	fill_sg (base, sglist, used, nelems, iopte_protection);
+
+	fill_sg(base, sglist, used, nelems, iopte_protection);
+
 #ifdef VERIFY_SG
 	verify_sglist(sglist, nelems, base, npages);
 #endif
 
-	spin_unlock_irqrestore(&iommu->lock, flags);
-
 	return used;
 
 bad:
-	spin_unlock_irqrestore(&iommu->lock, flags);
-	return PCI_DMA_ERROR_CODE;
+	iommu_free_ctx(iommu, ctx);
+bad_no_ctx:
+	if (printk_ratelimit())
+		WARN_ON(1);
+	return 0;
 }
 
 /* Unmap a set of streaming mode DMA translations. */
@@ -692,8 +644,10 @@ void pci_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems,
 	unsigned long flags, ctx, i, npages;
 	u32 bus_addr;
 
-	if (direction == PCI_DMA_NONE)
-		BUG();
+	if (unlikely(direction == PCI_DMA_NONE)) {
+		if (printk_ratelimit())
+			WARN_ON(1);
+	}
 
 	pcp = pdev->sysdata;
 	iommu = pcp->pbm->iommu;
@@ -705,7 +659,8 @@ void pci_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems,
 		if (sglist[i].dma_length == 0)
 			break;
 	i--;
-	npages = (IO_PAGE_ALIGN(sglist[i].dma_address + sglist[i].dma_length) - bus_addr) >> IO_PAGE_SHIFT;
+	npages = (IO_PAGE_ALIGN(sglist[i].dma_address + sglist[i].dma_length) -
+		  bus_addr) >> IO_PAGE_SHIFT;
 
 	base = iommu->page_table +
 		((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
@@ -726,11 +681,11 @@ void pci_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems,
 	if (strbuf->strbuf_enabled)
 		pci_strbuf_flush(strbuf, iommu, bus_addr, ctx, npages, direction);
 
-	/* Step 2: Clear out first TSB entry. */
-	iopte_make_dummy(iommu, base);
+	/* Step 2: Clear out the TSB entries. */
+	for (i = 0; i < npages; i++)
+		iopte_make_dummy(iommu, base + i);
 
-	free_streaming_cluster(iommu, bus_addr - iommu->page_table_map_base,
-			       npages, ctx);
+	free_npages(iommu, bus_addr - iommu->page_table_map_base, npages);
 
 	iommu_free_ctx(iommu, ctx);
 
diff --git a/arch/sparc64/kernel/pci_psycho.c b/arch/sparc64/kernel/pci_psycho.c
index 6ed1ef25e0a..c03ed5f49d3 100644
--- a/arch/sparc64/kernel/pci_psycho.c
+++ b/arch/sparc64/kernel/pci_psycho.c
@@ -1207,13 +1207,9 @@ static void psycho_scan_bus(struct pci_controller_info *p)
 static void psycho_iommu_init(struct pci_controller_info *p)
 {
 	struct pci_iommu *iommu = p->pbm_A.iommu;
-	unsigned long tsbbase, i;
+	unsigned long i;
 	u64 control;
 
-	/* Setup initial software IOMMU state. */
-	spin_lock_init(&iommu->lock);
-	iommu->ctx_lowest_free = 1;
-
 	/* Register addresses. */
 	iommu->iommu_control  = p->pbm_A.controller_regs + PSYCHO_IOMMU_CONTROL;
 	iommu->iommu_tsbbase  = p->pbm_A.controller_regs + PSYCHO_IOMMU_TSBBASE;
@@ -1240,40 +1236,10 @@ static void psycho_iommu_init(struct pci_controller_info *p)
 	/* Leave diag mode enabled for full-flushing done
 	 * in pci_iommu.c
 	 */
+	pci_iommu_table_init(iommu, IO_TSB_SIZE, 0xc0000000, 0xffffffff);
 
-	iommu->dummy_page = __get_free_pages(GFP_KERNEL, 0);
-	if (!iommu->dummy_page) {
-		prom_printf("PSYCHO_IOMMU: Error, gfp(dummy_page) failed.\n");
-		prom_halt();
-	}
-	memset((void *)iommu->dummy_page, 0, PAGE_SIZE);
-	iommu->dummy_page_pa = (unsigned long) __pa(iommu->dummy_page);
-
-	/* Using assumed page size 8K with 128K entries we need 1MB iommu page
-	 * table (128K ioptes * 8 bytes per iopte).  This is
-	 * page order 7 on UltraSparc.
-	 */
-	tsbbase = __get_free_pages(GFP_KERNEL, get_order(IO_TSB_SIZE));
-	if (!tsbbase) {
-		prom_printf("PSYCHO_IOMMU: Error, gfp(tsb) failed.\n");
-		prom_halt();
-	}
-	iommu->page_table = (iopte_t *)tsbbase;
-	iommu->page_table_sz_bits = 17;
-	iommu->page_table_map_base = 0xc0000000;
-	iommu->dma_addr_mask = 0xffffffff;
-	pci_iommu_table_init(iommu, IO_TSB_SIZE);
-
-	/* We start with no consistent mappings. */
-	iommu->lowest_consistent_map =
-		1 << (iommu->page_table_sz_bits - PBM_LOGCLUSTERS);
-
-	for (i = 0; i < PBM_NCLUSTERS; i++) {
-		iommu->alloc_info[i].flush = 0;
-		iommu->alloc_info[i].next = 0;
-	}
-
-	psycho_write(p->pbm_A.controller_regs + PSYCHO_IOMMU_TSBBASE, __pa(tsbbase));
+	psycho_write(p->pbm_A.controller_regs + PSYCHO_IOMMU_TSBBASE,
+		     __pa(iommu->page_table));
 
 	control = psycho_read(p->pbm_A.controller_regs + PSYCHO_IOMMU_CONTROL);
 	control &= ~(PSYCHO_IOMMU_CTRL_TSBSZ | PSYCHO_IOMMU_CTRL_TBWSZ);
@@ -1281,7 +1247,7 @@ static void psycho_iommu_init(struct pci_controller_info *p)
 	psycho_write(p->pbm_A.controller_regs + PSYCHO_IOMMU_CONTROL, control);
 
 	/* If necessary, hook us up for starfire IRQ translations. */
-	if(this_is_starfire)
+	if (this_is_starfire)
 		p->starfire_cookie = starfire_hookup(p->pbm_A.portid);
 	else
 		p->starfire_cookie = NULL;
diff --git a/arch/sparc64/kernel/pci_sabre.c b/arch/sparc64/kernel/pci_sabre.c
index 0ee6bd5b9ac..da8e1364194 100644
--- a/arch/sparc64/kernel/pci_sabre.c
+++ b/arch/sparc64/kernel/pci_sabre.c
@@ -1267,13 +1267,9 @@ static void sabre_iommu_init(struct pci_controller_info *p,
 			     u32 dma_mask)
 {
 	struct pci_iommu *iommu = p->pbm_A.iommu;
-	unsigned long tsbbase, i, order;
+	unsigned long i;
 	u64 control;
 
-	/* Setup initial software IOMMU state. */
-	spin_lock_init(&iommu->lock);
-	iommu->ctx_lowest_free = 1;
-
 	/* Register addresses. */
 	iommu->iommu_control  = p->pbm_A.controller_regs + SABRE_IOMMU_CONTROL;
 	iommu->iommu_tsbbase  = p->pbm_A.controller_regs + SABRE_IOMMU_TSBBASE;
@@ -1295,26 +1291,10 @@ static void sabre_iommu_init(struct pci_controller_info *p,
 	/* Leave diag mode enabled for full-flushing done
 	 * in pci_iommu.c
 	 */
+	pci_iommu_table_init(iommu, tsbsize * 1024 * 8, dvma_offset, dma_mask);
 
-	iommu->dummy_page = __get_free_pages(GFP_KERNEL, 0);
-	if (!iommu->dummy_page) {
-		prom_printf("PSYCHO_IOMMU: Error, gfp(dummy_page) failed.\n");
-		prom_halt();
-	}
-	memset((void *)iommu->dummy_page, 0, PAGE_SIZE);
-	iommu->dummy_page_pa = (unsigned long) __pa(iommu->dummy_page);
-
-	tsbbase = __get_free_pages(GFP_KERNEL, order = get_order(tsbsize * 1024 * 8));
-	if (!tsbbase) {
-		prom_printf("SABRE_IOMMU: Error, gfp(tsb) failed.\n");
-		prom_halt();
-	}
-	iommu->page_table = (iopte_t *)tsbbase;
-	iommu->page_table_map_base = dvma_offset;
-	iommu->dma_addr_mask = dma_mask;
-	pci_iommu_table_init(iommu, PAGE_SIZE << order);
-
-	sabre_write(p->pbm_A.controller_regs + SABRE_IOMMU_TSBBASE, __pa(tsbbase));
+	sabre_write(p->pbm_A.controller_regs + SABRE_IOMMU_TSBBASE,
+		    __pa(iommu->page_table));
 
 	control = sabre_read(p->pbm_A.controller_regs + SABRE_IOMMU_CONTROL);
 	control &= ~(SABRE_IOMMUCTRL_TSBSZ | SABRE_IOMMUCTRL_TBWSZ);
@@ -1322,11 +1302,9 @@ static void sabre_iommu_init(struct pci_controller_info *p,
 	switch(tsbsize) {
 	case 64:
 		control |= SABRE_IOMMU_TSBSZ_64K;
-		iommu->page_table_sz_bits = 16;
 		break;
 	case 128:
 		control |= SABRE_IOMMU_TSBSZ_128K;
-		iommu->page_table_sz_bits = 17;
 		break;
 	default:
 		prom_printf("iommu_init: Illegal TSB size %d\n", tsbsize);
@@ -1334,15 +1312,6 @@ static void sabre_iommu_init(struct pci_controller_info *p,
 		break;
 	}
 	sabre_write(p->pbm_A.controller_regs + SABRE_IOMMU_CONTROL, control);
-
-	/* We start with no consistent mappings. */
-	iommu->lowest_consistent_map =
-		1 << (iommu->page_table_sz_bits - PBM_LOGCLUSTERS);
-
-	for (i = 0; i < PBM_NCLUSTERS; i++) {
-		iommu->alloc_info[i].flush = 0;
-		iommu->alloc_info[i].next = 0;
-	}
 }
 
 static void pbm_register_toplevel_resources(struct pci_controller_info *p,
diff --git a/arch/sparc64/kernel/pci_schizo.c b/arch/sparc64/kernel/pci_schizo.c
index 331382e1a75..d8c4e0919b4 100644
--- a/arch/sparc64/kernel/pci_schizo.c
+++ b/arch/sparc64/kernel/pci_schizo.c
@@ -330,7 +330,7 @@ static int schizo_ino_to_pil(struct pci_dev *pdev, unsigned int ino)
 static void tomatillo_wsync_handler(struct ino_bucket *bucket, void *_arg1, void *_arg2)
 {
 	unsigned long sync_reg = (unsigned long) _arg2;
-	u64 mask = 1 << (__irq_ino(__irq(bucket)) & IMAP_INO);
+	u64 mask = 1UL << (__irq_ino(__irq(bucket)) & IMAP_INO);
 	u64 val;
 	int limit;
 
@@ -1765,7 +1765,7 @@ static void schizo_pbm_strbuf_init(struct pci_pbm_info *pbm)
 static void schizo_pbm_iommu_init(struct pci_pbm_info *pbm)
 {
 	struct pci_iommu *iommu = pbm->iommu;
-	unsigned long tsbbase, i, tagbase, database, order;
+	unsigned long i, tagbase, database;
 	u32 vdma[2], dma_mask;
 	u64 control;
 	int err, tsbsize;
@@ -1800,10 +1800,6 @@ static void schizo_pbm_iommu_init(struct pci_pbm_info *pbm)
 			prom_halt();
 	};
 
-	/* Setup initial software IOMMU state. */
-	spin_lock_init(&iommu->lock);
-	iommu->ctx_lowest_free = 1;
-
 	/* Register addresses, SCHIZO has iommu ctx flushing. */
 	iommu->iommu_control  = pbm->pbm_regs + SCHIZO_IOMMU_CONTROL;
 	iommu->iommu_tsbbase  = pbm->pbm_regs + SCHIZO_IOMMU_TSBBASE;
@@ -1832,56 +1828,9 @@ static void schizo_pbm_iommu_init(struct pci_pbm_info *pbm)
 	/* Leave diag mode enabled for full-flushing done
 	 * in pci_iommu.c
 	 */
+	pci_iommu_table_init(iommu, tsbsize * 8 * 1024, vdma[0], dma_mask);
 
-	iommu->dummy_page = __get_free_pages(GFP_KERNEL, 0);
-	if (!iommu->dummy_page) {
-		prom_printf("PSYCHO_IOMMU: Error, gfp(dummy_page) failed.\n");
-		prom_halt();
-	}
-	memset((void *)iommu->dummy_page, 0, PAGE_SIZE);
-	iommu->dummy_page_pa = (unsigned long) __pa(iommu->dummy_page);
-
-	/* Using assumed page size 8K with 128K entries we need 1MB iommu page
-	 * table (128K ioptes * 8 bytes per iopte).  This is
-	 * page order 7 on UltraSparc.
-	 */
-	order = get_order(tsbsize * 8 * 1024);
-	tsbbase = __get_free_pages(GFP_KERNEL, order);
-	if (!tsbbase) {
-		prom_printf("%s: Error, gfp(tsb) failed.\n", pbm->name);
-		prom_halt();
-	}
-
-	iommu->page_table = (iopte_t *)tsbbase;
-	iommu->page_table_map_base = vdma[0];
-	iommu->dma_addr_mask = dma_mask;
-	pci_iommu_table_init(iommu, PAGE_SIZE << order);
-
-	switch (tsbsize) {
-	case 64:
-		iommu->page_table_sz_bits = 16;
-		break;
-
-	case 128:
-		iommu->page_table_sz_bits = 17;
-		break;
-
-	default:
-		prom_printf("iommu_init: Illegal TSB size %d\n", tsbsize);
-		prom_halt();
-		break;
-	};
-
-	/* We start with no consistent mappings. */
-	iommu->lowest_consistent_map =
-		1 << (iommu->page_table_sz_bits - PBM_LOGCLUSTERS);
-
-	for (i = 0; i < PBM_NCLUSTERS; i++) {
-		iommu->alloc_info[i].flush = 0;
-		iommu->alloc_info[i].next = 0;
-	}
-
-	schizo_write(iommu->iommu_tsbbase, __pa(tsbbase));
+	schizo_write(iommu->iommu_tsbbase, __pa(iommu->page_table));
 
 	control = schizo_read(iommu->iommu_control);
 	control &= ~(SCHIZO_IOMMU_CTRL_TSBSZ | SCHIZO_IOMMU_CTRL_TBWSZ);
diff --git a/arch/sparc64/kernel/power.c b/arch/sparc64/kernel/power.c
index 946cee0257e..9e8362ea310 100644
--- a/arch/sparc64/kernel/power.c
+++ b/arch/sparc64/kernel/power.c
@@ -17,6 +17,7 @@
 
 #include <asm/system.h>
 #include <asm/ebus.h>
+#include <asm/isa.h>
 #include <asm/auxio.h>
 
 #include <linux/unistd.h>
@@ -100,46 +101,83 @@ again:
 	return 0;
 }
 
-static int __init has_button_interrupt(struct linux_ebus_device *edev)
+static int __init has_button_interrupt(unsigned int irq, int prom_node)
 {
-	if (edev->irqs[0] == PCI_IRQ_NONE)
+	if (irq == PCI_IRQ_NONE)
 		return 0;
-	if (!prom_node_has_property(edev->prom_node, "button"))
+	if (!prom_node_has_property(prom_node, "button"))
 		return 0;
 
 	return 1;
 }
 
-void __init power_init(void)
+static int __init power_probe_ebus(struct resource **resp, unsigned int *irq_p, int *prom_node_p)
 {
 	struct linux_ebus *ebus;
 	struct linux_ebus_device *edev;
+
+	for_each_ebus(ebus) {
+		for_each_ebusdev(edev, ebus) {
+			if (!strcmp(edev->prom_name, "power")) {
+				*resp = &edev->resource[0];
+				*irq_p = edev->irqs[0];
+				*prom_node_p = edev->prom_node;
+				return 0;
+			}
+		}
+	}
+	return -ENODEV;
+}
+
+static int __init power_probe_isa(struct resource **resp, unsigned int *irq_p, int *prom_node_p)
+{
+	struct sparc_isa_bridge *isa_bus;
+	struct sparc_isa_device *isa_dev;
+
+	for_each_isa(isa_bus) {
+		for_each_isadev(isa_dev, isa_bus) {
+			if (!strcmp(isa_dev->prom_name, "power")) {
+				*resp = &isa_dev->resource;
+				*irq_p = isa_dev->irq;
+				*prom_node_p = isa_dev->prom_node;
+				return 0;
+			}
+		}
+	}
+	return -ENODEV;
+}
+
+void __init power_init(void)
+{
+	struct resource *res = NULL;
+	unsigned int irq;
+	int prom_node;
 	static int invoked;
 
 	if (invoked)
 		return;
 	invoked = 1;
 
-	for_each_ebus(ebus) {
-		for_each_ebusdev(edev, ebus) {
-			if (!strcmp(edev->prom_name, "power"))
-				goto found;
-		}
-	}
+	if (!power_probe_ebus(&res, &irq, &prom_node))
+		goto found;
+
+	if (!power_probe_isa(&res, &irq, &prom_node))
+		goto found;
+
 	return;
 
 found:
-	power_reg = ioremap(edev->resource[0].start, 0x4);
+	power_reg = ioremap(res->start, 0x4);
 	printk("power: Control reg at %p ... ", power_reg);
 	poweroff_method = machine_halt;  /* able to use the standard halt */
-	if (has_button_interrupt(edev)) {
+	if (has_button_interrupt(irq, prom_node)) {
 		if (kernel_thread(powerd, NULL, CLONE_FS) < 0) {
 			printk("Failed to start power daemon.\n");
 			return;
 		}
 		printk("powerd running.\n");
 
-		if (request_irq(edev->irqs[0],
+		if (request_irq(irq,
 				power_handler, SA_SHIRQ, "power", NULL) < 0)
 			printk("power: Error, cannot register IRQ handler.\n");
 	} else {
diff --git a/arch/sparc64/kernel/ptrace.c b/arch/sparc64/kernel/ptrace.c
index 5efbff90d66..774ecbb8a03 100644
--- a/arch/sparc64/kernel/ptrace.c
+++ b/arch/sparc64/kernel/ptrace.c
@@ -31,6 +31,7 @@
 #include <asm/visasm.h>
 #include <asm/spitfire.h>
 #include <asm/page.h>
+#include <asm/cpudata.h>
 
 /* Returning from ptrace is a bit tricky because the syscall return
  * low level code assumes any value returned which is negative and
@@ -132,12 +133,16 @@ void flush_ptrace_access(struct vm_area_struct *vma, struct page *page,
 	if ((uaddr ^ (unsigned long) kaddr) & (1UL << 13)) {
 		unsigned long start = __pa(kaddr);
 		unsigned long end = start + len;
+		unsigned long dcache_line_size;
+
+		dcache_line_size = local_cpu_data().dcache_line_size;
 
 		if (tlb_type == spitfire) {
-			for (; start < end; start += 32)
+			for (; start < end; start += dcache_line_size)
 				spitfire_put_dcache_tag(start & 0x3fe0, 0x0);
 		} else {
-			for (; start < end; start += 32)
+			start &= ~(dcache_line_size - 1);
+			for (; start < end; start += dcache_line_size)
 				__asm__ __volatile__(
 					"stxa %%g0, [%0] %1\n\t"
 					"membar #Sync"
@@ -150,8 +155,11 @@ void flush_ptrace_access(struct vm_area_struct *vma, struct page *page,
 	if (write && tlb_type == spitfire) {
 		unsigned long start = (unsigned long) kaddr;
 		unsigned long end = start + len;
+		unsigned long icache_line_size;
+
+		icache_line_size = local_cpu_data().icache_line_size;
 
-		for (; start < end; start += 32)
+		for (; start < end; start += icache_line_size)
 			flushi(start);
 	}
 }
diff --git a/arch/sparc64/kernel/rtrap.S b/arch/sparc64/kernel/rtrap.S
index fafd227735f..090dcca00d2 100644
--- a/arch/sparc64/kernel/rtrap.S
+++ b/arch/sparc64/kernel/rtrap.S
@@ -256,9 +256,8 @@ rt_continue:	ldx			[%sp + PTREGS_OFF + PT_V9_G1], %g1
 		brnz,pn			%l3, kern_rtt
 		 mov			PRIMARY_CONTEXT, %l7
 		ldxa			[%l7 + %l7] ASI_DMMU, %l0
-cplus_rtrap_insn_1:
-		sethi			%hi(0), %l1
-		sllx			%l1, 32, %l1
+		sethi			%hi(sparc64_kern_pri_nuc_bits), %l1
+		ldx			[%l1 + %lo(sparc64_kern_pri_nuc_bits)], %l1
 		or			%l0, %l1, %l0
 		stxa			%l0, [%l7] ASI_DMMU
 		flush			%g6
@@ -313,53 +312,36 @@ kern_fpucheck:	ldub			[%g6 + TI_FPDEPTH], %l5
 		wr			%g1, FPRS_FEF, %fprs
 		ldx			[%o1 + %o5], %g1
 		add			%g6, TI_XFSR, %o1
-		membar			#StoreLoad | #LoadLoad
 		sll			%o0, 8, %o2
 		add			%g6, TI_FPREGS, %o3
 		brz,pn			%l6, 1f
 		 add			%g6, TI_FPREGS+0x40, %o4
 
+		membar			#Sync
 		ldda			[%o3 + %o2] ASI_BLK_P, %f0
 		ldda			[%o4 + %o2] ASI_BLK_P, %f16
+		membar			#Sync
 1:		andcc			%l2, FPRS_DU, %g0
 		be,pn			%icc, 1f
 		 wr			%g1, 0, %gsr
 		add			%o2, 0x80, %o2
+		membar			#Sync
 		ldda			[%o3 + %o2] ASI_BLK_P, %f32
 		ldda			[%o4 + %o2] ASI_BLK_P, %f48
-
 1:		membar			#Sync
 		ldx			[%o1 + %o5], %fsr
 2:		stb			%l5, [%g6 + TI_FPDEPTH]
 		ba,pt			%xcc, rt_continue
 		 nop
 5:		wr			%g0, FPRS_FEF, %fprs
-		membar			#StoreLoad | #LoadLoad
 		sll			%o0, 8, %o2
 
 		add			%g6, TI_FPREGS+0x80, %o3
 		add			%g6, TI_FPREGS+0xc0, %o4
+		membar			#Sync
 		ldda			[%o3 + %o2] ASI_BLK_P, %f32
 		ldda			[%o4 + %o2] ASI_BLK_P, %f48
 		membar			#Sync
 		wr			%g0, FPRS_DU, %fprs
 		ba,pt			%xcc, rt_continue
 		 stb			%l5, [%g6 + TI_FPDEPTH]
-
-cplus_rinsn_1:
-		sethi			%uhi(CTX_CHEETAH_PLUS_NUC), %l1
-
-		.globl			cheetah_plus_patch_rtrap
-cheetah_plus_patch_rtrap:
-		/* We configure the dTLB512_0 for 4MB pages and the
-		 * dTLB512_1 for 8K pages when in context zero.
-		 */
-		sethi			%hi(cplus_rinsn_1), %o0
-		sethi			%hi(cplus_rtrap_insn_1), %o2
-		lduw			[%o0 + %lo(cplus_rinsn_1)], %o1
-		or			%o2, %lo(cplus_rtrap_insn_1), %o2
-		stw			%o1, [%o2]
-		flush			%o2
-
-		retl
-		 nop
diff --git a/arch/sparc64/kernel/setup.c b/arch/sparc64/kernel/setup.c
index ddbed3341a2..c1f34237cdf 100644
--- a/arch/sparc64/kernel/setup.c
+++ b/arch/sparc64/kernel/setup.c
@@ -187,17 +187,13 @@ int prom_callback(long *args)
 		}
 
 		if ((va >= KERNBASE) && (va < (KERNBASE + (4 * 1024 * 1024)))) {
-			unsigned long kernel_pctx = 0;
-
-			if (tlb_type == cheetah_plus)
-				kernel_pctx |= (CTX_CHEETAH_PLUS_NUC |
-						CTX_CHEETAH_PLUS_CTX0);
+			extern unsigned long sparc64_kern_pri_context;
 
 			/* Spitfire Errata #32 workaround */
 			__asm__ __volatile__("stxa	%0, [%1] %2\n\t"
 					     "flush	%%g6"
 					     : /* No outputs */
-					     : "r" (kernel_pctx),
+					     : "r" (sparc64_kern_pri_context),
 					       "r" (PRIMARY_CONTEXT),
 					       "i" (ASI_DMMU));
 
@@ -464,8 +460,6 @@ static void __init boot_flags_init(char *commands)
 	}
 }
 
-extern int prom_probe_memory(void);
-extern unsigned long start, end;
 extern void panic_setup(char *, int *);
 
 extern unsigned short root_flags;
@@ -492,13 +486,8 @@ void register_prom_callbacks(void)
 		   "' linux-.soft2 to .soft2");
 }
 
-extern void paging_init(void);
-
 void __init setup_arch(char **cmdline_p)
 {
-	unsigned long highest_paddr;
-	int i;
-
 	/* Initialize PROM console and command line. */
 	*cmdline_p = prom_getbootargs();
 	strcpy(saved_command_line, *cmdline_p);
@@ -517,40 +506,6 @@ void __init setup_arch(char **cmdline_p)
 	boot_flags_init(*cmdline_p);
 
 	idprom_init();
-	(void) prom_probe_memory();
-
-	/* In paging_init() we tip off this value to see if we need
-	 * to change init_mm.pgd to point to the real alias mapping.
-	 */
-	phys_base = 0xffffffffffffffffUL;
-	highest_paddr = 0UL;
-	for (i = 0; sp_banks[i].num_bytes != 0; i++) {
-		unsigned long top;
-
-		if (sp_banks[i].base_addr < phys_base)
-			phys_base = sp_banks[i].base_addr;
-		top = sp_banks[i].base_addr +
-			sp_banks[i].num_bytes;
-		if (highest_paddr < top)
-			highest_paddr = top;
-	}
-	pfn_base = phys_base >> PAGE_SHIFT;
-
-	switch (tlb_type) {
-	default:
-	case spitfire:
-		kern_base = spitfire_get_itlb_data(sparc64_highest_locked_tlbent());
-		kern_base &= _PAGE_PADDR_SF;
-		break;
-
-	case cheetah:
-	case cheetah_plus:
-		kern_base = cheetah_get_litlb_data(sparc64_highest_locked_tlbent());
-		kern_base &= _PAGE_PADDR;
-		break;
-	};
-
-	kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
 
 	if (!root_flags)
 		root_mountflags &= ~MS_RDONLY;
@@ -625,6 +580,9 @@ extern void smp_info(struct seq_file *);
 extern void smp_bogo(struct seq_file *);
 extern void mmu_info(struct seq_file *);
 
+unsigned int dcache_parity_tl1_occurred;
+unsigned int icache_parity_tl1_occurred;
+
 static int show_cpuinfo(struct seq_file *m, void *__unused)
 {
 	seq_printf(m, 
@@ -635,6 +593,8 @@ static int show_cpuinfo(struct seq_file *m, void *__unused)
 		   "type\t\t: sun4u\n"
 		   "ncpus probed\t: %ld\n"
 		   "ncpus active\t: %ld\n"
+		   "D$ parity tl1\t: %u\n"
+		   "I$ parity tl1\t: %u\n"
 #ifndef CONFIG_SMP
 		   "Cpu0Bogo\t: %lu.%02lu\n"
 		   "Cpu0ClkTck\t: %016lx\n"
@@ -647,7 +607,9 @@ static int show_cpuinfo(struct seq_file *m, void *__unused)
 		   (prom_prev >> 8) & 0xff,
 		   prom_prev & 0xff,
 		   (long)num_possible_cpus(),
-		   (long)num_online_cpus()
+		   (long)num_online_cpus(),
+		   dcache_parity_tl1_occurred,
+		   icache_parity_tl1_occurred
 #ifndef CONFIG_SMP
 		   , cpu_data(0).udelay_val/(500000/HZ),
 		   (cpu_data(0).udelay_val/(5000/HZ)) % 100,
diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c
index b4fc6a5462b..b137fd63f5e 100644
--- a/arch/sparc64/kernel/smp.c
+++ b/arch/sparc64/kernel/smp.c
@@ -93,6 +93,27 @@ void __init smp_store_cpu_info(int id)
 	cpu_data(id).pte_cache[1]		= NULL;
 	cpu_data(id).pgd_cache			= NULL;
 	cpu_data(id).idle_volume		= 1;
+
+	cpu_data(id).dcache_size = prom_getintdefault(cpu_node, "dcache-size",
+						      16 * 1024);
+	cpu_data(id).dcache_line_size =
+		prom_getintdefault(cpu_node, "dcache-line-size", 32);
+	cpu_data(id).icache_size = prom_getintdefault(cpu_node, "icache-size",
+						      16 * 1024);
+	cpu_data(id).icache_line_size =
+		prom_getintdefault(cpu_node, "icache-line-size", 32);
+	cpu_data(id).ecache_size = prom_getintdefault(cpu_node, "ecache-size",
+						      4 * 1024 * 1024);
+	cpu_data(id).ecache_line_size =
+		prom_getintdefault(cpu_node, "ecache-line-size", 64);
+	printk("CPU[%d]: Caches "
+	       "D[sz(%d):line_sz(%d)] "
+	       "I[sz(%d):line_sz(%d)] "
+	       "E[sz(%d):line_sz(%d)]\n",
+	       id,
+	       cpu_data(id).dcache_size, cpu_data(id).dcache_line_size,
+	       cpu_data(id).icache_size, cpu_data(id).icache_line_size,
+	       cpu_data(id).ecache_size, cpu_data(id).ecache_line_size);
 }
 
 static void smp_setup_percpu_timer(void);
@@ -980,13 +1001,6 @@ void smp_penguin_jailcell(int irq, struct pt_regs *regs)
 	preempt_enable();
 }
 
-extern unsigned long xcall_promstop;
-
-void smp_promstop_others(void)
-{
-	smp_cross_call(&xcall_promstop, 0, 0, 0);
-}
-
 #define prof_multiplier(__cpu)		cpu_data(__cpu).multiplier
 #define prof_counter(__cpu)		cpu_data(__cpu).counter
 
diff --git a/arch/sparc64/kernel/sys32.S b/arch/sparc64/kernel/sys32.S
index 5f9e4fae612..9cd272ac3ac 100644
--- a/arch/sparc64/kernel/sys32.S
+++ b/arch/sparc64/kernel/sys32.S
@@ -157,173 +157,199 @@ sys32_socketcall:	/* %o0=call, %o1=args */
 	or		%g2, %lo(__socketcall_table_begin), %g2
 	jmpl		%g2 + %o0, %g0
 	 nop
+do_einval:
+	retl
+	 mov		-EINVAL, %o0
 
-	/* Each entry is exactly 32 bytes. */
 	.align		32
 __socketcall_table_begin:
+
+	/* Each entry is exactly 32 bytes. */
 do_sys_socket: /* sys_socket(int, int, int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+1:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_socket), %g1
-	ldswa		[%o1 + 0x8] %asi, %o2
+2:	ldswa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(sys_socket), %g0
-	 ldswa		[%o1 + 0x4] %asi, %o1
+3:	 ldswa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 do_sys_bind: /* sys_bind(int fd, struct sockaddr *, int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+4:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_bind), %g1
-	ldswa		[%o1 + 0x8] %asi, %o2
+5:	ldswa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(sys_bind), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+6:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 do_sys_connect: /* sys_connect(int, struct sockaddr *, int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+7:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_connect), %g1
-	ldswa		[%o1 + 0x8] %asi, %o2
+8:	ldswa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(sys_connect), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+9:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 do_sys_listen: /* sys_listen(int, int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+10:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_listen), %g1
 	jmpl		%g1 + %lo(sys_listen), %g0
-	 ldswa		[%o1 + 0x4] %asi, %o1
+11:	 ldswa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 	nop
 do_sys_accept: /* sys_accept(int, struct sockaddr *, int *) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+12:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_accept), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
+13:	lduwa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(sys_accept), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+14:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 do_sys_getsockname: /* sys_getsockname(int, struct sockaddr *, int *) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+15:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_getsockname), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
+16:	lduwa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(sys_getsockname), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+17:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 do_sys_getpeername: /* sys_getpeername(int, struct sockaddr *, int *) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+18:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_getpeername), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
+19:	lduwa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(sys_getpeername), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+20:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 do_sys_socketpair: /* sys_socketpair(int, int, int, int *) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+21:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_socketpair), %g1
-	ldswa		[%o1 + 0x8] %asi, %o2
-	lduwa		[%o1 + 0xc] %asi, %o3
+22:	ldswa		[%o1 + 0x8] %asi, %o2
+23:	lduwa		[%o1 + 0xc] %asi, %o3
 	jmpl		%g1 + %lo(sys_socketpair), %g0
-	 ldswa		[%o1 + 0x4] %asi, %o1
+24:	 ldswa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 do_sys_send: /* sys_send(int, void *, size_t, unsigned int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+25:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_send), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
-	lduwa		[%o1 + 0xc] %asi, %o3
+26:	lduwa		[%o1 + 0x8] %asi, %o2
+27:	lduwa		[%o1 + 0xc] %asi, %o3
 	jmpl		%g1 + %lo(sys_send), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+28:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 do_sys_recv: /* sys_recv(int, void *, size_t, unsigned int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+29:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_recv), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
-	lduwa		[%o1 + 0xc] %asi, %o3
+30:	lduwa		[%o1 + 0x8] %asi, %o2
+31:	lduwa		[%o1 + 0xc] %asi, %o3
 	jmpl		%g1 + %lo(sys_recv), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+32:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 do_sys_sendto: /* sys_sendto(int, u32, compat_size_t, unsigned int, u32, int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+33:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_sendto), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
-	lduwa		[%o1 + 0xc] %asi, %o3
-	lduwa		[%o1 + 0x10] %asi, %o4
-	ldswa		[%o1 + 0x14] %asi, %o5
+34:	lduwa		[%o1 + 0x8] %asi, %o2
+35:	lduwa		[%o1 + 0xc] %asi, %o3
+36:	lduwa		[%o1 + 0x10] %asi, %o4
+37:	ldswa		[%o1 + 0x14] %asi, %o5
 	jmpl		%g1 + %lo(sys_sendto), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+38:	 lduwa		[%o1 + 0x4] %asi, %o1
 do_sys_recvfrom: /* sys_recvfrom(int, u32, compat_size_t, unsigned int, u32, u32) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+39:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_recvfrom), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
-	lduwa		[%o1 + 0xc] %asi, %o3
-	lduwa		[%o1 + 0x10] %asi, %o4
-	lduwa		[%o1 + 0x14] %asi, %o5
+40:	lduwa		[%o1 + 0x8] %asi, %o2
+41:	lduwa		[%o1 + 0xc] %asi, %o3
+42:	lduwa		[%o1 + 0x10] %asi, %o4
+43:	lduwa		[%o1 + 0x14] %asi, %o5
 	jmpl		%g1 + %lo(sys_recvfrom), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+44:	 lduwa		[%o1 + 0x4] %asi, %o1
 do_sys_shutdown: /* sys_shutdown(int, int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+45:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(sys_shutdown), %g1
 	jmpl		%g1 + %lo(sys_shutdown), %g0
-	 ldswa		[%o1 + 0x4] %asi, %o1
+46:	 ldswa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 	nop
 do_sys_setsockopt: /* compat_sys_setsockopt(int, int, int, char *, int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+47:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(compat_sys_setsockopt), %g1
-	ldswa		[%o1 + 0x8] %asi, %o2
-	lduwa		[%o1 + 0xc] %asi, %o3
-	ldswa		[%o1 + 0x10] %asi, %o4
+48:	ldswa		[%o1 + 0x8] %asi, %o2
+49:	lduwa		[%o1 + 0xc] %asi, %o3
+50:	ldswa		[%o1 + 0x10] %asi, %o4
 	jmpl		%g1 + %lo(compat_sys_setsockopt), %g0
-	 ldswa		[%o1 + 0x4] %asi, %o1
+51:	 ldswa		[%o1 + 0x4] %asi, %o1
 	nop
 do_sys_getsockopt: /* compat_sys_getsockopt(int, int, int, u32, u32) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+52:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(compat_sys_getsockopt), %g1
-	ldswa		[%o1 + 0x8] %asi, %o2
-	lduwa		[%o1 + 0xc] %asi, %o3
-	lduwa		[%o1 + 0x10] %asi, %o4
+53:	ldswa		[%o1 + 0x8] %asi, %o2
+54:	lduwa		[%o1 + 0xc] %asi, %o3
+55:	lduwa		[%o1 + 0x10] %asi, %o4
 	jmpl		%g1 + %lo(compat_sys_getsockopt), %g0
-	 ldswa		[%o1 + 0x4] %asi, %o1
+56:	 ldswa		[%o1 + 0x4] %asi, %o1
 	nop
 do_sys_sendmsg: /* compat_sys_sendmsg(int, struct compat_msghdr *, unsigned int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+57:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(compat_sys_sendmsg), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
+58:	lduwa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(compat_sys_sendmsg), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+59:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
 do_sys_recvmsg: /* compat_sys_recvmsg(int, struct compat_msghdr *, unsigned int) */
-	ldswa		[%o1 + 0x0] %asi, %o0
+60:	ldswa		[%o1 + 0x0] %asi, %o0
 	sethi		%hi(compat_sys_recvmsg), %g1
-	lduwa		[%o1 + 0x8] %asi, %o2
+61:	lduwa		[%o1 + 0x8] %asi, %o2
 	jmpl		%g1 + %lo(compat_sys_recvmsg), %g0
-	 lduwa		[%o1 + 0x4] %asi, %o1
+62:	 lduwa		[%o1 + 0x4] %asi, %o1
 	nop
 	nop
 	nop
-__socketcall_table_end:
-
-do_einval:
-	retl
-	 mov		-EINVAL, %o0
-do_efault:
-	retl
-	 mov		-EFAULT, %o0
 
 	.section	__ex_table
 	.align		4
-	.word		__socketcall_table_begin, 0, __socketcall_table_end, do_efault
+	.word		1b, __retl_efault, 2b, __retl_efault
+	.word		3b, __retl_efault, 4b, __retl_efault
+	.word		5b, __retl_efault, 6b, __retl_efault
+	.word		7b, __retl_efault, 8b, __retl_efault
+	.word		9b, __retl_efault, 10b, __retl_efault
+	.word		11b, __retl_efault, 12b, __retl_efault
+	.word		13b, __retl_efault, 14b, __retl_efault
+	.word		15b, __retl_efault, 16b, __retl_efault
+	.word		17b, __retl_efault, 18b, __retl_efault
+	.word		19b, __retl_efault, 20b, __retl_efault
+	.word		21b, __retl_efault, 22b, __retl_efault
+	.word		23b, __retl_efault, 24b, __retl_efault
+	.word		25b, __retl_efault, 26b, __retl_efault
+	.word		27b, __retl_efault, 28b, __retl_efault
+	.word		29b, __retl_efault, 30b, __retl_efault
+	.word		31b, __retl_efault, 32b, __retl_efault
+	.word		33b, __retl_efault, 34b, __retl_efault
+	.word		35b, __retl_efault, 36b, __retl_efault
+	.word		37b, __retl_efault, 38b, __retl_efault
+	.word		39b, __retl_efault, 40b, __retl_efault
+	.word		41b, __retl_efault, 42b, __retl_efault
+	.word		43b, __retl_efault, 44b, __retl_efault
+	.word		45b, __retl_efault, 46b, __retl_efault
+	.word		47b, __retl_efault, 48b, __retl_efault
+	.word		49b, __retl_efault, 50b, __retl_efault
+	.word		51b, __retl_efault, 52b, __retl_efault
+	.word		53b, __retl_efault, 54b, __retl_efault
+	.word		55b, __retl_efault, 56b, __retl_efault
+	.word		57b, __retl_efault, 58b, __retl_efault
+	.word		59b, __retl_efault, 60b, __retl_efault
+	.word		61b, __retl_efault, 62b, __retl_efault
 	.previous
diff --git a/arch/sparc64/kernel/trampoline.S b/arch/sparc64/kernel/trampoline.S
index 3a145fc39cf..9478551cb02 100644
--- a/arch/sparc64/kernel/trampoline.S
+++ b/arch/sparc64/kernel/trampoline.S
@@ -119,8 +119,8 @@ startup_continue:
 	sethi		%hi(itlb_load), %g2
 	or		%g2, %lo(itlb_load), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x18]
-	sethi		%hi(mmu_ihandle_cache), %g2
-	lduw		[%g2 + %lo(mmu_ihandle_cache)], %g2
+	sethi		%hi(prom_mmu_ihandle_cache), %g2
+	lduw		[%g2 + %lo(prom_mmu_ihandle_cache)], %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x20]
 	sethi		%hi(KERNBASE), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x28]
@@ -156,8 +156,8 @@ startup_continue:
 	sethi		%hi(itlb_load), %g2
 	or		%g2, %lo(itlb_load), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x18]
-	sethi		%hi(mmu_ihandle_cache), %g2
-	lduw		[%g2 + %lo(mmu_ihandle_cache)], %g2
+	sethi		%hi(prom_mmu_ihandle_cache), %g2
+	lduw		[%g2 + %lo(prom_mmu_ihandle_cache)], %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x20]
 	sethi		%hi(KERNBASE + 0x400000), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x28]
@@ -190,8 +190,8 @@ do_dtlb:
 	sethi		%hi(dtlb_load), %g2
 	or		%g2, %lo(dtlb_load), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x18]
-	sethi		%hi(mmu_ihandle_cache), %g2
-	lduw		[%g2 + %lo(mmu_ihandle_cache)], %g2
+	sethi		%hi(prom_mmu_ihandle_cache), %g2
+	lduw		[%g2 + %lo(prom_mmu_ihandle_cache)], %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x20]
 	sethi		%hi(KERNBASE), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x28]
@@ -228,8 +228,8 @@ do_dtlb:
 	sethi		%hi(dtlb_load), %g2
 	or		%g2, %lo(dtlb_load), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x18]
-	sethi		%hi(mmu_ihandle_cache), %g2
-	lduw		[%g2 + %lo(mmu_ihandle_cache)], %g2
+	sethi		%hi(prom_mmu_ihandle_cache), %g2
+	lduw		[%g2 + %lo(prom_mmu_ihandle_cache)], %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x20]
 	sethi		%hi(KERNBASE + 0x400000), %g2
 	stx		%g2, [%sp + 2047 + 128 + 0x28]
@@ -336,20 +336,13 @@ do_unlock:
 	call		init_irqwork_curcpu
 	 nop
 
-	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g2,g3,1f)
-	ba,pt	%xcc, 2f
-	 nop
-
-1:	/* Start using proper page size encodings in ctx register.  */
-	sethi	%uhi(CTX_CHEETAH_PLUS_NUC), %g3
+	/* Start using proper page size encodings in ctx register.  */
+	sethi	%hi(sparc64_kern_pri_context), %g3
+	ldx	[%g3 + %lo(sparc64_kern_pri_context)], %g2
 	mov	PRIMARY_CONTEXT, %g1
-	sllx	%g3, 32, %g3
-	sethi	%hi(CTX_CHEETAH_PLUS_CTX0), %g2
-	or	%g3, %g2, %g3
-	stxa	%g3, [%g1] ASI_DMMU
+	stxa	%g2, [%g1] ASI_DMMU
 	membar	#Sync
 
-2:
 	rdpr		%pstate, %o1
 	or		%o1, PSTATE_IE, %o1
 	wrpr		%o1, 0, %pstate
diff --git a/arch/sparc64/kernel/traps.c b/arch/sparc64/kernel/traps.c
index b280b2ef674..5570e7bb22b 100644
--- a/arch/sparc64/kernel/traps.c
+++ b/arch/sparc64/kernel/traps.c
@@ -189,19 +189,18 @@ void spitfire_data_access_exception(struct pt_regs *regs, unsigned long sfsr, un
 
 	if (regs->tstate & TSTATE_PRIV) {
 		/* Test if this comes from uaccess places. */
-		unsigned long fixup;
-		unsigned long g2 = regs->u_regs[UREG_G2];
+		const struct exception_table_entry *entry;
 
-		if ((fixup = search_extables_range(regs->tpc, &g2))) {
-			/* Ouch, somebody is trying ugly VM hole tricks on us... */
+		entry = search_exception_tables(regs->tpc);
+		if (entry) {
+			/* Ouch, somebody is trying VM hole tricks on us... */
 #ifdef DEBUG_EXCEPTIONS
 			printk("Exception: PC<%016lx> faddr<UNKNOWN>\n", regs->tpc);
-			printk("EX_TABLE: insn<%016lx> fixup<%016lx> "
-			       "g2<%016lx>\n", regs->tpc, fixup, g2);
+			printk("EX_TABLE: insn<%016lx> fixup<%016lx>\n",
+			       regs->tpc, entry->fixup);
 #endif
-			regs->tpc = fixup;
+			regs->tpc = entry->fixup;
 			regs->tnpc = regs->tpc + 4;
-			regs->u_regs[UREG_G2] = g2;
 			return;
 		}
 		/* Shit... */
@@ -758,26 +757,12 @@ void __init cheetah_ecache_flush_init(void)
 	ecache_flush_size = (2 * largest_size);
 	ecache_flush_linesize = smallest_linesize;
 
-	/* Discover a physically contiguous chunk of physical
-	 * memory in 'sp_banks' of size ecache_flush_size calculated
-	 * above.  Store the physical base of this area at
-	 * ecache_flush_physbase.
-	 */
-	for (node = 0; ; node++) {
-		if (sp_banks[node].num_bytes == 0)
-			break;
-		if (sp_banks[node].num_bytes >= ecache_flush_size) {
-			ecache_flush_physbase = sp_banks[node].base_addr;
-			break;
-		}
-	}
+	ecache_flush_physbase = find_ecache_flush_span(ecache_flush_size);
 
-	/* Note: Zero would be a valid value of ecache_flush_physbase so
-	 * don't use that as the success test. :-)
-	 */
-	if (sp_banks[node].num_bytes == 0) {
+	if (ecache_flush_physbase == ~0UL) {
 		prom_printf("cheetah_ecache_flush_init: Cannot find %d byte "
-			    "contiguous physical memory.\n", ecache_flush_size);
+			    "contiguous physical memory.\n",
+			    ecache_flush_size);
 		prom_halt();
 	}
 
@@ -869,14 +854,19 @@ static void cheetah_flush_ecache_line(unsigned long physaddr)
  */
 static void __cheetah_flush_icache(void)
 {
-	unsigned long i;
+	unsigned int icache_size, icache_line_size;
+	unsigned long addr;
+
+	icache_size = local_cpu_data().icache_size;
+	icache_line_size = local_cpu_data().icache_line_size;
 
 	/* Clear the valid bits in all the tags. */
-	for (i = 0; i < (1 << 15); i += (1 << 5)) {
+	for (addr = 0; addr < icache_size; addr += icache_line_size) {
 		__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
 				     "membar #Sync"
 				     : /* no outputs */
-				     : "r" (i | (2 << 3)), "i" (ASI_IC_TAG));
+				     : "r" (addr | (2 << 3)),
+				       "i" (ASI_IC_TAG));
 	}
 }
 
@@ -904,13 +894,17 @@ static void cheetah_flush_icache(void)
 
 static void cheetah_flush_dcache(void)
 {
-	unsigned long i;
+	unsigned int dcache_size, dcache_line_size;
+	unsigned long addr;
 
-	for (i = 0; i < (1 << 16); i += (1 << 5)) {
+	dcache_size = local_cpu_data().dcache_size;
+	dcache_line_size = local_cpu_data().dcache_line_size;
+
+	for (addr = 0; addr < dcache_size; addr += dcache_line_size) {
 		__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
 				     "membar #Sync"
 				     : /* no outputs */
-				     : "r" (i), "i" (ASI_DCACHE_TAG));
+				     : "r" (addr), "i" (ASI_DCACHE_TAG));
 	}
 }
 
@@ -921,24 +915,29 @@ static void cheetah_flush_dcache(void)
  */
 static void cheetah_plus_zap_dcache_parity(void)
 {
-	unsigned long i;
+	unsigned int dcache_size, dcache_line_size;
+	unsigned long addr;
+
+	dcache_size = local_cpu_data().dcache_size;
+	dcache_line_size = local_cpu_data().dcache_line_size;
 
-	for (i = 0; i < (1 << 16); i += (1 << 5)) {
-		unsigned long tag = (i >> 14);
-		unsigned long j;
+	for (addr = 0; addr < dcache_size; addr += dcache_line_size) {
+		unsigned long tag = (addr >> 14);
+		unsigned long line;
 
 		__asm__ __volatile__("membar	#Sync\n\t"
 				     "stxa	%0, [%1] %2\n\t"
 				     "membar	#Sync"
 				     : /* no outputs */
-				     : "r" (tag), "r" (i),
+				     : "r" (tag), "r" (addr),
 				       "i" (ASI_DCACHE_UTAG));
-		for (j = i; j < i + (1 << 5); j += (1 << 3))
+		for (line = addr; line < addr + dcache_line_size; line += 8)
 			__asm__ __volatile__("membar	#Sync\n\t"
 					     "stxa	%%g0, [%0] %1\n\t"
 					     "membar	#Sync"
 					     : /* no outputs */
-					     : "r" (j), "i" (ASI_DCACHE_DATA));
+					     : "r" (line),
+					       "i" (ASI_DCACHE_DATA));
 	}
 }
 
@@ -1332,16 +1331,12 @@ static int cheetah_fix_ce(unsigned long physaddr)
 /* Return non-zero if PADDR is a valid physical memory address. */
 static int cheetah_check_main_memory(unsigned long paddr)
 {
-	int i;
+	unsigned long vaddr = PAGE_OFFSET + paddr;
 
-	for (i = 0; ; i++) {
-		if (sp_banks[i].num_bytes == 0)
-			break;
-		if (paddr >= sp_banks[i].base_addr &&
-		    paddr < (sp_banks[i].base_addr + sp_banks[i].num_bytes))
-			return 1;
-	}
-	return 0;
+	if (vaddr > (unsigned long) high_memory)
+		return 0;
+
+	return kern_addr_valid(vaddr);
 }
 
 void cheetah_cee_handler(struct pt_regs *regs, unsigned long afsr, unsigned long afar)
@@ -1596,10 +1591,10 @@ void cheetah_deferred_handler(struct pt_regs *regs, unsigned long afsr, unsigned
 			/* OK, usermode access. */
 			recoverable = 1;
 		} else {
-			unsigned long g2 = regs->u_regs[UREG_G2];
-			unsigned long fixup = search_extables_range(regs->tpc, &g2);
+			const struct exception_table_entry *entry;
 
-			if (fixup != 0UL) {
+			entry = search_exception_tables(regs->tpc);
+			if (entry) {
 				/* OK, kernel access to userspace. */
 				recoverable = 1;
 
@@ -1618,9 +1613,8 @@ void cheetah_deferred_handler(struct pt_regs *regs, unsigned long afsr, unsigned
 				 * recoverable condition.
 				 */
 				if (recoverable) {
-					regs->tpc = fixup;
+					regs->tpc = entry->fixup;
 					regs->tnpc = regs->tpc + 4;
-					regs->u_regs[UREG_G2] = g2;
 				}
 			}
 		}
diff --git a/arch/sparc64/kernel/una_asm.S b/arch/sparc64/kernel/una_asm.S
index da48400bcc9..1f5b5b708ce 100644
--- a/arch/sparc64/kernel/una_asm.S
+++ b/arch/sparc64/kernel/una_asm.S
@@ -6,13 +6,6 @@
 
 	.text
 
-kernel_unaligned_trap_fault:
-	call	kernel_mna_trap_fault
-	 nop
-	retl
-	 nop
-	.size	kern_unaligned_trap_fault, .-kern_unaligned_trap_fault
-
 	.globl	__do_int_store
 __do_int_store:
 	rd	%asi, %o4
@@ -51,24 +44,24 @@ __do_int_store:
 0:
 	wr	%o4, 0x0, %asi
 	retl
-	 nop
+	 mov	0, %o0
 	.size	__do_int_store, .-__do_int_store
 
 	.section	__ex_table
-	.word		4b, kernel_unaligned_trap_fault
-	.word		5b, kernel_unaligned_trap_fault
-	.word		6b, kernel_unaligned_trap_fault
-	.word		7b, kernel_unaligned_trap_fault
-	.word		8b, kernel_unaligned_trap_fault
-	.word		9b, kernel_unaligned_trap_fault
-	.word		10b, kernel_unaligned_trap_fault
-	.word		11b, kernel_unaligned_trap_fault
-	.word		12b, kernel_unaligned_trap_fault
-	.word		13b, kernel_unaligned_trap_fault
-	.word		14b, kernel_unaligned_trap_fault
-	.word		15b, kernel_unaligned_trap_fault
-	.word		16b, kernel_unaligned_trap_fault
-	.word		17b, kernel_unaligned_trap_fault
+	.word		4b, __retl_efault
+	.word		5b, __retl_efault
+	.word		6b, __retl_efault
+	.word		7b, __retl_efault
+	.word		8b, __retl_efault
+	.word		9b, __retl_efault
+	.word		10b, __retl_efault
+	.word		11b, __retl_efault
+	.word		12b, __retl_efault
+	.word		13b, __retl_efault
+	.word		14b, __retl_efault
+	.word		15b, __retl_efault
+	.word		16b, __retl_efault
+	.word		17b, __retl_efault
 	.previous
 
 	.globl	do_int_load
@@ -133,21 +126,21 @@ do_int_load:
 0:
 	wr	%o5, 0x0, %asi
 	retl
-	 nop
+	 mov	0, %o0
 	.size	__do_int_load, .-__do_int_load
 
 	.section	__ex_table
-	.word		4b, kernel_unaligned_trap_fault
-	.word		5b, kernel_unaligned_trap_fault
-	.word		6b, kernel_unaligned_trap_fault
-	.word		7b, kernel_unaligned_trap_fault
-	.word		8b, kernel_unaligned_trap_fault
-	.word		9b, kernel_unaligned_trap_fault
-	.word		10b, kernel_unaligned_trap_fault
-	.word		11b, kernel_unaligned_trap_fault
-	.word		12b, kernel_unaligned_trap_fault
-	.word		13b, kernel_unaligned_trap_fault
-	.word		14b, kernel_unaligned_trap_fault
-	.word		15b, kernel_unaligned_trap_fault
-	.word		16b, kernel_unaligned_trap_fault
+	.word		4b, __retl_efault
+	.word		5b, __retl_efault
+	.word		6b, __retl_efault
+	.word		7b, __retl_efault
+	.word		8b, __retl_efault
+	.word		9b, __retl_efault
+	.word		10b, __retl_efault
+	.word		11b, __retl_efault
+	.word		12b, __retl_efault
+	.word		13b, __retl_efault
+	.word		14b, __retl_efault
+	.word		15b, __retl_efault
+	.word		16b, __retl_efault
 	.previous
diff --git a/arch/sparc64/kernel/unaligned.c b/arch/sparc64/kernel/unaligned.c
index 42718f6a7d3..70faf630603 100644
--- a/arch/sparc64/kernel/unaligned.c
+++ b/arch/sparc64/kernel/unaligned.c
@@ -180,14 +180,14 @@ static void __attribute_used__ unaligned_panic(char *str, struct pt_regs *regs)
 	die_if_kernel(str, regs);
 }
 
-extern void do_int_load(unsigned long *dest_reg, int size,
-			unsigned long *saddr, int is_signed, int asi);
+extern int do_int_load(unsigned long *dest_reg, int size,
+		       unsigned long *saddr, int is_signed, int asi);
 	
-extern void __do_int_store(unsigned long *dst_addr, int size,
-			   unsigned long src_val, int asi);
+extern int __do_int_store(unsigned long *dst_addr, int size,
+			  unsigned long src_val, int asi);
 
-static inline void do_int_store(int reg_num, int size, unsigned long *dst_addr,
-				struct pt_regs *regs, int asi, int orig_asi)
+static inline int do_int_store(int reg_num, int size, unsigned long *dst_addr,
+			       struct pt_regs *regs, int asi, int orig_asi)
 {
 	unsigned long zero = 0;
 	unsigned long *src_val_p = &zero;
@@ -219,7 +219,7 @@ static inline void do_int_store(int reg_num, int size, unsigned long *dst_addr,
 			break;
 		};
 	}
-	__do_int_store(dst_addr, size, src_val, asi);
+	return __do_int_store(dst_addr, size, src_val, asi);
 }
 
 static inline void advance(struct pt_regs *regs)
@@ -242,14 +242,14 @@ static inline int ok_for_kernel(unsigned int insn)
 	return !floating_point_load_or_store_p(insn);
 }
 
-void kernel_mna_trap_fault(void)
+static void kernel_mna_trap_fault(void)
 {
 	struct pt_regs *regs = current_thread_info()->kern_una_regs;
 	unsigned int insn = current_thread_info()->kern_una_insn;
-	unsigned long g2 = regs->u_regs[UREG_G2];
-	unsigned long fixup = search_extables_range(regs->tpc, &g2);
+	const struct exception_table_entry *entry;
 
-	if (!fixup) {
+	entry = search_exception_tables(regs->tpc);
+	if (!entry) {
 		unsigned long address;
 
 		address = compute_effective_address(regs, insn,
@@ -270,9 +270,8 @@ void kernel_mna_trap_fault(void)
 	        die_if_kernel("Oops", regs);
 		/* Not reached */
 	}
-	regs->tpc = fixup;
+	regs->tpc = entry->fixup;
 	regs->tnpc = regs->tpc + 4;
-	regs->u_regs [UREG_G2] = g2;
 
 	regs->tstate &= ~TSTATE_ASI;
 	regs->tstate |= (ASI_AIUS << 24UL);
@@ -294,8 +293,8 @@ asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn, u
 
 		kernel_mna_trap_fault();
 	} else {
-		unsigned long addr;
-		int orig_asi, asi;
+		unsigned long addr, *reg_addr;
+		int orig_asi, asi, err;
 
 		addr = compute_effective_address(regs, insn,
 						 ((insn >> 25) & 0x1f));
@@ -319,11 +318,12 @@ asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn, u
 		};
 		switch (dir) {
 		case load:
-			do_int_load(fetch_reg_addr(((insn>>25)&0x1f), regs),
-				    size, (unsigned long *) addr,
-				    decode_signedness(insn), asi);
-			if (unlikely(asi != orig_asi)) {
-				unsigned long val_in = *(unsigned long *) addr;
+			reg_addr = fetch_reg_addr(((insn>>25)&0x1f), regs);
+			err = do_int_load(reg_addr, size,
+					  (unsigned long *) addr,
+					  decode_signedness(insn), asi);
+			if (likely(!err) && unlikely(asi != orig_asi)) {
+				unsigned long val_in = *reg_addr;
 				switch (size) {
 				case 2:
 					val_in = swab16(val_in);
@@ -339,21 +339,24 @@ asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn, u
 					BUG();
 					break;
 				};
-				*(unsigned long *) addr = val_in;
+				*reg_addr = val_in;
 			}
 			break;
 
 		case store:
-			do_int_store(((insn>>25)&0x1f), size,
-				     (unsigned long *) addr, regs,
-				     asi, orig_asi);
+			err = do_int_store(((insn>>25)&0x1f), size,
+					   (unsigned long *) addr, regs,
+					   asi, orig_asi);
 			break;
 
 		default:
 			panic("Impossible kernel unaligned trap.");
 			/* Not reached... */
 		}
-		advance(regs);
+		if (unlikely(err))
+			kernel_mna_trap_fault();
+		else
+			advance(regs);
 	}
 }
 
diff --git a/arch/sparc64/kernel/us3_cpufreq.c b/arch/sparc64/kernel/us3_cpufreq.c
index 9080e7cd4bb..0340041f614 100644
--- a/arch/sparc64/kernel/us3_cpufreq.c
+++ b/arch/sparc64/kernel/us3_cpufreq.c
@@ -208,7 +208,10 @@ static int __init us3_freq_init(void)
 	impl  = ((ver >> 32) & 0xffff);
 
 	if (manuf == CHEETAH_MANUF &&
-	    (impl == CHEETAH_IMPL || impl == CHEETAH_PLUS_IMPL)) {
+	    (impl == CHEETAH_IMPL ||
+	     impl == CHEETAH_PLUS_IMPL ||
+	     impl == JAGUAR_IMPL ||
+	     impl == PANTHER_IMPL)) {
 		struct cpufreq_driver *driver;
 
 		ret = -ENOMEM;
diff --git a/arch/sparc64/kernel/vmlinux.lds.S b/arch/sparc64/kernel/vmlinux.lds.S
index f47d0be3937..2af0cf0a864 100644
--- a/arch/sparc64/kernel/vmlinux.lds.S
+++ b/arch/sparc64/kernel/vmlinux.lds.S
@@ -9,8 +9,7 @@ ENTRY(_start)
 jiffies = jiffies_64;
 SECTIONS
 {
-  swapper_pmd_dir = 0x0000000000402000;
-  empty_pg_dir = 0x0000000000403000;
+  swapper_low_pmd_dir = 0x0000000000402000;
   . = 0x4000;
   .text 0x0000000000404000 :
   {
diff --git a/arch/sparc64/kernel/winfixup.S b/arch/sparc64/kernel/winfixup.S
index 99c809a1e5a..39160926267 100644
--- a/arch/sparc64/kernel/winfixup.S
+++ b/arch/sparc64/kernel/winfixup.S
@@ -16,23 +16,14 @@
 	.text
 
 set_pcontext:
-cplus_winfixup_insn_1:
-	sethi	%hi(0), %l1
+	sethi	%hi(sparc64_kern_pri_context), %l1
+	ldx	[%l1 + %lo(sparc64_kern_pri_context)], %l1
 	mov	PRIMARY_CONTEXT, %g1
-	sllx	%l1, 32, %l1
-cplus_winfixup_insn_2:
-	sethi	%hi(0), %g2
-	or	%l1, %g2, %l1
 	stxa	%l1, [%g1] ASI_DMMU
 	flush	%g6
 	retl
 	 nop
 
-cplus_wfinsn_1:
-	sethi	%uhi(CTX_CHEETAH_PLUS_NUC), %l1
-cplus_wfinsn_2:
-	sethi	%hi(CTX_CHEETAH_PLUS_CTX0), %g2
-
 	.align	32
 
 	/* Here are the rules, pay attention.
@@ -395,23 +386,3 @@ window_dax_from_user_common:
 	 add		%sp, PTREGS_OFF, %o0
 	ba,pt		%xcc, rtrap
 	 clr		%l6
-	
-
-	.globl		cheetah_plus_patch_winfixup
-cheetah_plus_patch_winfixup:
-	sethi			%hi(cplus_wfinsn_1), %o0
-	sethi			%hi(cplus_winfixup_insn_1), %o2
-	lduw			[%o0 + %lo(cplus_wfinsn_1)], %o1
-	or			%o2, %lo(cplus_winfixup_insn_1), %o2
-	stw			%o1, [%o2]
-	flush			%o2
-
-	sethi			%hi(cplus_wfinsn_2), %o0
-	sethi			%hi(cplus_winfixup_insn_2), %o2
-	lduw			[%o0 + %lo(cplus_wfinsn_2)], %o1
-	or			%o2, %lo(cplus_winfixup_insn_2), %o2
-	stw			%o1, [%o2]
-	flush			%o2
-
-	retl
-	 nop