[SPARC64]: Do not assume sun4v chips have load-twin/store-init support.

Check the cpu type in the OBP device tree before committing to
using the optimized Niagara memcpy and memset implementation.

If we don't recognize the cpu type, use a completely generic
version.

Signed-off-by: David S. Miller <davem@davemloft.net>

arch/sparc64/kernel/cpu.c

@@ -13,7 +13,7 @@
 #include <asm/fpumacro.h>
 #include <asm/cpudata.h>
 #include <asm/spitfire.h>
-#include <asm/prom.h>
+#include <asm/oplib.h>
 
 DEFINE_PER_CPU(cpuinfo_sparc, __cpu_data) = { 0 };
 
@@ -69,36 +69,24 @@ unsigned int fsr_storage;
 
 static void __init sun4v_cpu_probe(void)
 {
-	struct device_node *dp;
+	switch (sun4v_chip_type) {
-	const char *compat;
+	case SUN4V_CHIP_NIAGARA1:
-	int len;
-
-	dp = of_find_node_by_name(NULL, "cpu");
-	if (!dp)
-		goto no_compat;
-
-	compat = of_get_property(dp, "compatible", &len);
-	if (!compat)
-		goto no_compat;
-
-	if (of_find_in_proplist(compat, "SUNW,UltraSPARC-T1", len)) {
 		sparc_cpu_type = "UltraSparc T1 (Niagara)";
 		sparc_fpu_type = "UltraSparc T1 integrated FPU";
-	} else if (of_find_in_proplist(compat, "SUNW,UltraSPARC-T2", len)) {
+		break;
+
+	case SUN4V_CHIP_NIAGARA2:
 		sparc_cpu_type = "UltraSparc T2 (Niagara2)";
 		sparc_fpu_type = "UltraSparc T2 integrated FPU";
-	} else
+		break;
-		goto unknown;
 
-	return;
+	default:
-
+		printk(KERN_WARNING "CPU: Unknown sun4v cpu type [%s]\n",
-no_compat:
+		       prom_cpu_compatible);
-	compat = "no property";
+		sparc_cpu_type = "Unknown SUN4V CPU";
-
+		sparc_fpu_type = "Unknown SUN4V FPU";
-unknown:
+		break;
-	printk(KERN_WARNING "CPU: Unknown sun4v cpu type [%s]\n", compat);
+	}
-	sparc_cpu_type = "Unknown SUN4V CPU";
-	sparc_fpu_type = "Unknown SUN4V FPU";
 }
 
 void __init cpu_probe(void)

arch/sparc64/kernel/head.S

@@ -97,7 +97,8 @@ sparc64_boot:
 	.globl	prom_map_name, prom_unmap_name, prom_mmu_ihandle_cache
 	.globl	prom_boot_mapped_pc, prom_boot_mapping_mode
 	.globl	prom_boot_mapping_phys_high, prom_boot_mapping_phys_low
-	.globl	is_sun4v
+	.globl	prom_compatible_name, prom_cpu_path, prom_cpu_compatible
+	.globl	is_sun4v, sun4v_chip_type
 prom_peer_name:
 	.asciz	"peer"
 prom_compatible_name:
@@ -106,6 +107,8 @@ prom_finddev_name:
 	.asciz	"finddevice"
 prom_chosen_path:
 	.asciz	"/chosen"
+prom_cpu_path:
+	.asciz	"/cpu"
 prom_getprop_name:
 	.asciz	"getprop"
 prom_mmu_name:
@@ -120,9 +123,13 @@ prom_unmap_name:
 	.asciz	"unmap"
 prom_sun4v_name:
 	.asciz	"sun4v"
+prom_niagara_prefix:
+	.asciz	"SUNW,UltraSPARC-T"
 	.align	4
 prom_root_compatible:
 	.skip	64
+prom_cpu_compatible:
+	.skip	64
 prom_root_node:
 	.word	0
 prom_mmu_ihandle_cache:
@@ -138,6 +145,8 @@ prom_boot_mapping_phys_low:
 	.xword	0
 is_sun4v:
 	.word	0
+sun4v_chip_type:
+	.word	SUN4V_CHIP_INVALID
 1:
 	rd	%pc, %l0
 
@@ -296,13 +305,13 @@ is_sun4v:
 	sethi	%hi(prom_sun4v_name), %g7
 	or	%g7, %lo(prom_sun4v_name), %g7
 	mov	5, %g3
-1:	ldub	[%g7], %g2
+90:	ldub	[%g7], %g2
 	ldub	[%g1], %g4
 	cmp	%g2, %g4
-	bne,pn	%icc, 2f
+	bne,pn	%icc, 80f
 	 add	%g7, 1, %g7
 	subcc	%g3, 1, %g3
-	bne,pt	%xcc, 1b
+	bne,pt	%xcc, 90b
 	 add	%g1, 1, %g1
 
 	sethi	%hi(is_sun4v), %g1
@@ -310,7 +319,80 @@ is_sun4v:
 	mov	1, %g7
 	stw	%g7, [%g1]
 
-2:
+	/* cpu_node = prom_finddevice("/cpu") */
+	mov	(1b - prom_finddev_name), %l1
+	mov	(1b - prom_cpu_path), %l2
+	sub	%l0, %l1, %l1
+	sub	%l0, %l2, %l2
+	sub	%sp, (192 + 128), %sp
+
+	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, "/cpu"
+	stx	%g0, [%sp + 2047 + 128 + 0x20]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	ldx	[%sp + 2047 + 128 + 0x20], %l4	! cpu device node
+
+	mov	(1b - prom_getprop_name), %l1
+	mov	(1b - prom_compatible_name), %l2
+	mov	(1b - prom_cpu_compatible), %l5
+	sub	%l0, %l1, %l1
+	sub	%l0, %l2, %l2
+	sub	%l0, %l5, %l5
+
+	/* prom_getproperty(cpu_node, "compatible",
+	 *                  &prom_cpu_compatible, 64)
+	 */
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "getprop"
+	mov	4, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 4
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	stx	%l4, [%sp + 2047 + 128 + 0x18]	! arg1, cpu_node
+	stx	%l2, [%sp + 2047 + 128 + 0x20]	! arg2, "compatible"
+	stx	%l5, [%sp + 2047 + 128 + 0x28]	! arg3, &prom_cpu_compatible
+	mov	64, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x30]	! arg4, size
+	stx	%g0, [%sp + 2047 + 128 + 0x38]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	add	%sp, (192 + 128), %sp
+
+	sethi	%hi(prom_cpu_compatible), %g1
+	or	%g1, %lo(prom_cpu_compatible), %g1
+	sethi	%hi(prom_niagara_prefix), %g7
+	or	%g7, %lo(prom_niagara_prefix), %g7
+	mov	17, %g3
+90:	ldub	[%g7], %g2
+	ldub	[%g1], %g4
+	cmp	%g2, %g4
+	bne,pn	%icc, 4f
+	 add	%g7, 1, %g7
+	subcc	%g3, 1, %g3
+	bne,pt	%xcc, 90b
+	 add	%g1, 1, %g1
+
+	sethi	%hi(prom_cpu_compatible), %g1
+	or	%g1, %lo(prom_cpu_compatible), %g1
+	ldub	[%g1 + 17], %g2
+	cmp	%g2, '1'
+	be,pt	%xcc, 5f
+	 mov	SUN4V_CHIP_NIAGARA1, %g4
+	cmp	%g2, '2'
+	be,pt	%xcc, 5f
+	 mov	SUN4V_CHIP_NIAGARA2, %g4
+4:
+	mov	SUN4V_CHIP_UNKNOWN, %g4
+5:	sethi	%hi(sun4v_chip_type), %g2
+	or	%g2, %lo(sun4v_chip_type), %g2
+	stw	%g4, [%g2]
+
+80:
 	BRANCH_IF_SUN4V(g1, jump_to_sun4u_init)
 	BRANCH_IF_CHEETAH_BASE(g1,g7,cheetah_boot)
 	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,cheetah_plus_boot)
@@ -414,6 +496,24 @@ niagara_tlb_fixup:
 	stw	%g2, [%g1 + %lo(tlb_type)]
 
 	/* Patch copy/clear ops.  */
+	sethi	%hi(sun4v_chip_type), %g1
+	lduw	[%g1 + %lo(sun4v_chip_type)], %g1
+	cmp	%g1, SUN4V_CHIP_NIAGARA1
+	be,pt	%xcc, niagara_patch
+	 cmp	%g1, SUN4V_CHIP_NIAGARA2
+	be,pt	%xcc, niagara_patch
+	 nop
+
+	call	generic_patch_copyops
+	 nop
+	call	generic_patch_bzero
+	 nop
+	call	generic_patch_pageops
+	 nop
+
+	ba,a,pt	%xcc, 80f
+
+niagara_patch:
 	call	niagara_patch_copyops
 	 nop
 	call	niagara_patch_bzero
@@ -421,6 +521,7 @@ niagara_tlb_fixup:
 	call	niagara_patch_pageops
 	 nop
 
+80:
 	/* Patch TLB/cache ops.  */
 	call	hypervisor_patch_cachetlbops
 	 nop

arch/sparc64/kernel/sparc64_ksyms.c

@@ -168,6 +168,7 @@ EXPORT_SYMBOL(change_bit);
 EXPORT_SYMBOL(__flushw_user);
 
 EXPORT_SYMBOL(tlb_type);
+EXPORT_SYMBOL(sun4v_chip_type);
 EXPORT_SYMBOL(get_fb_unmapped_area);
 EXPORT_SYMBOL(flush_icache_range);
 

arch/sparc64/lib/GENbzero.S

@@ -0,0 +1,160 @@
+/* GENbzero.S: Generic sparc64 memset/clear_user.
+ *
+ * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
+ */
+#include <asm/asi.h>
+
+#define EX_ST(x,y)		\
+98:	x,y;			\
+	.section .fixup;	\
+	.align 4;		\
+99:	retl;			\
+	 mov	%o1, %o0;	\
+	.section __ex_table;	\
+	.align 4;		\
+	.word 98b, 99b;		\
+	.text;			\
+	.align 4;
+
+	.align	32
+	.text
+
+	.globl		GENmemset
+	.type		GENmemset, #function
+GENmemset:		/* %o0=buf, %o1=pat, %o2=len */
+	and		%o1, 0xff, %o3
+	mov		%o2, %o1
+	sllx		%o3, 8, %g1
+	or		%g1, %o3, %o2
+	sllx		%o2, 16, %g1
+	or		%g1, %o2, %o2
+	sllx		%o2, 32, %g1
+	ba,pt		%xcc, 1f
+	 or		%g1, %o2, %o2
+
+	.globl		GENbzero
+	.type		GENbzero, #function
+GENbzero:
+	clr		%o2
+1:	brz,pn		%o1, GENbzero_return
+	 mov		%o0, %o3
+
+	/* %o5: saved %asi, restored at GENbzero_done
+	 * %o4:	store %asi to use
+	 */
+	rd		%asi, %o5
+	mov		ASI_P, %o4
+	wr		%o4, 0x0, %asi
+
+GENbzero_from_clear_user:
+	cmp		%o1, 15
+	bl,pn		%icc, GENbzero_tiny
+	 andcc		%o0, 0x7, %g1
+	be,pt		%xcc, 2f
+	 mov		8, %g2
+	sub		%g2, %g1, %g1
+	sub		%o1, %g1, %o1
+1:	EX_ST(stba %o2, [%o0 + 0x00] %asi)
+	subcc		%g1, 1, %g1
+	bne,pt		%xcc, 1b
+	 add		%o0, 1, %o0
+2:	cmp		%o1, 128
+	bl,pn		%icc, GENbzero_medium
+	 andcc		%o0, (64 - 1), %g1
+	be,pt		%xcc, GENbzero_pre_loop
+	 mov		64, %g2
+	sub		%g2, %g1, %g1
+	sub		%o1, %g1, %o1
+1:	EX_ST(stxa %o2, [%o0 + 0x00] %asi)
+	subcc		%g1, 8, %g1
+	bne,pt		%xcc, 1b
+	 add		%o0, 8, %o0
+
+GENbzero_pre_loop:
+	andn		%o1, (64 - 1), %g1
+	sub		%o1, %g1, %o1
+GENbzero_loop:
+	EX_ST(stxa %o2, [%o0 + 0x00] %asi)
+	EX_ST(stxa %o2, [%o0 + 0x08] %asi)
+	EX_ST(stxa %o2, [%o0 + 0x10] %asi)
+	EX_ST(stxa %o2, [%o0 + 0x18] %asi)
+	EX_ST(stxa %o2, [%o0 + 0x20] %asi)
+	EX_ST(stxa %o2, [%o0 + 0x28] %asi)
+	EX_ST(stxa %o2, [%o0 + 0x30] %asi)
+	EX_ST(stxa %o2, [%o0 + 0x38] %asi)
+	subcc		%g1, 64, %g1
+	bne,pt		%xcc, GENbzero_loop
+	 add		%o0, 64, %o0
+
+	membar		#Sync
+	wr		%o4, 0x0, %asi
+	brz,pn		%o1, GENbzero_done
+GENbzero_medium:
+	 andncc		%o1, 0x7, %g1
+	be,pn		%xcc, 2f
+	 sub		%o1, %g1, %o1
+1:	EX_ST(stxa %o2, [%o0 + 0x00] %asi)
+	subcc		%g1, 8, %g1
+	bne,pt		%xcc, 1b
+	 add		%o0, 8, %o0
+2:	brz,pt		%o1, GENbzero_done
+	 nop
+
+GENbzero_tiny:
+1:	EX_ST(stba %o2, [%o0 + 0x00] %asi)
+	subcc		%o1, 1, %o1
+	bne,pt		%icc, 1b
+	 add		%o0, 1, %o0
+
+	/* fallthrough */
+
+GENbzero_done:
+	wr		%o5, 0x0, %asi
+
+GENbzero_return:
+	retl
+	 mov		%o3, %o0
+	.size		GENbzero, .-GENbzero
+	.size		GENmemset, .-GENmemset
+
+	.globl		GENclear_user
+	.type		GENclear_user, #function
+GENclear_user:		/* %o0=buf, %o1=len */
+	rd		%asi, %o5
+	brz,pn		%o1, GENbzero_done
+	 clr		%o3
+	cmp		%o5, ASI_AIUS
+	bne,pn		%icc, GENbzero
+	 clr		%o2
+	ba,pt		%xcc, GENbzero_from_clear_user
+	 mov		ASI_AIUS, %o4
+	.size		GENclear_user, .-GENclear_user
+
+#define BRANCH_ALWAYS	0x10680000
+#define NOP		0x01000000
+#define GEN_DO_PATCH(OLD, NEW)	\
+	sethi	%hi(NEW), %g1; \
+	or	%g1, %lo(NEW), %g1; \
+	sethi	%hi(OLD), %g2; \
+	or	%g2, %lo(OLD), %g2; \
+	sub	%g1, %g2, %g1; \
+	sethi	%hi(BRANCH_ALWAYS), %g3; \
+	sll	%g1, 11, %g1; \
+	srl	%g1, 11 + 2, %g1; \
+	or	%g3, %lo(BRANCH_ALWAYS), %g3; \
+	or	%g3, %g1, %g3; \
+	stw	%g3, [%g2]; \
+	sethi	%hi(NOP), %g3; \
+	or	%g3, %lo(NOP), %g3; \
+	stw	%g3, [%g2 + 0x4]; \
+	flush	%g2;
+
+	.globl	generic_patch_bzero
+	.type	generic_patch_bzero,#function
+generic_patch_bzero:
+	GEN_DO_PATCH(memset, GENmemset)
+	GEN_DO_PATCH(__bzero, GENbzero)
+	GEN_DO_PATCH(__clear_user, GENclear_user)
+	retl
+	 nop
+	.size	generic_patch_bzero,.-generic_patch_bzero

arch/sparc64/lib/GENcopy_from_user.S

@@ -0,0 +1,34 @@
+/* GENcopy_from_user.S: Generic sparc64 copy from userspace.
+ *
+ * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
+ */
+
+#define EX_LD(x)		\
+98:	x;			\
+	.section .fixup;	\
+	.align 4;		\
+99:	retl;			\
+	 mov	1, %o0;		\
+	.section __ex_table,"a";\
+	.align 4;		\
+	.word 98b, 99b;		\
+	.text;			\
+	.align 4;
+
+#ifndef ASI_AIUS
+#define ASI_AIUS	0x11
+#endif
+
+#define FUNC_NAME		GENcopy_from_user
+#define LOAD(type,addr,dest)	type##a [addr] ASI_AIUS, dest
+#define EX_RETVAL(x)		0
+
+#ifdef __KERNEL__
+#define PREAMBLE					\
+	rd		%asi, %g1;			\
+	cmp		%g1, ASI_AIUS;			\
+	bne,pn		%icc, memcpy_user_stub;		\
+	 nop
+#endif
+
+#include "GENmemcpy.S"

arch/sparc64/lib/GENcopy_to_user.S

@@ -0,0 +1,38 @@
+/* GENcopy_to_user.S: Generic sparc64 copy to userspace.
+ *
+ * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
+ */
+
+#define EX_ST(x)		\
+98:	x;			\
+	.section .fixup;	\
+	.align 4;		\
+99:	retl;			\
+	 mov	1, %o0;		\
+	.section __ex_table,"a";\
+	.align 4;		\
+	.word 98b, 99b;		\
+	.text;			\
+	.align 4;
+
+#ifndef ASI_AIUS
+#define ASI_AIUS	0x11
+#endif
+
+#define FUNC_NAME		GENcopy_to_user
+#define STORE(type,src,addr)	type##a src, [addr] ASI_AIUS
+#define EX_RETVAL(x)		0
+
+#ifdef __KERNEL__
+	/* Writing to %asi is _expensive_ so we hardcode it.
+	 * Reading %asi to check for KERNEL_DS is comparatively
+	 * cheap.
+	 */
+#define PREAMBLE					\
+	rd		%asi, %g1;			\
+	cmp		%g1, ASI_AIUS;			\
+	bne,pn		%icc, memcpy_user_stub;		\
+	 nop
+#endif
+
+#include "GENmemcpy.S"

arch/sparc64/lib/GENmemcpy.S

@@ -0,0 +1,121 @@
+/* GENmemcpy.S: Generic sparc64 memcpy.
+ *
+ * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
+ */
+
+#ifdef __KERNEL__
+#define GLOBAL_SPARE	%g7
+#else
+#define GLOBAL_SPARE	%g5
+#endif
+
+#ifndef EX_LD
+#define EX_LD(x)	x
+#endif
+
+#ifndef EX_ST
+#define EX_ST(x)	x
+#endif
+
+#ifndef EX_RETVAL
+#define EX_RETVAL(x)	x
+#endif
+
+#ifndef LOAD
+#define LOAD(type,addr,dest)	type [addr], dest
+#endif
+
+#ifndef STORE
+#define STORE(type,src,addr)	type src, [addr]
+#endif
+
+#ifndef FUNC_NAME
+#define FUNC_NAME	GENmemcpy
+#endif
+
+#ifndef PREAMBLE
+#define PREAMBLE
+#endif
+
+#ifndef XCC
+#define XCC xcc
+#endif
+
+	.register	%g2,#scratch
+	.register	%g3,#scratch
+
+	.text
+	.align		64
+
+	.globl	FUNC_NAME
+	.type	FUNC_NAME,#function
+FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
+	srlx		%o2, 31, %g2
+	cmp		%g2, 0
+	tne		%XCC, 5
+	PREAMBLE
+	mov		%o0, GLOBAL_SPARE
+
+	cmp		%o2, 0
+	be,pn		%XCC, 85f
+	 or		%o0, %o1, %o3
+	cmp		%o2, 16
+	blu,a,pn	%XCC, 80f
+	 or		%o3, %o2, %o3
+
+	xor		%o0, %o1, %o4
+	andcc		%o4, 0x7, %g0
+	bne,a,pn	%XCC, 90f
+	 sub		%o0, %o1, %o3
+
+	and		%o0, 0x7, %o4
+	sub		%o4, 0x8, %o4
+	sub		%g0, %o4, %o4
+	sub		%o2, %o4, %o2
+1:	subcc		%o4, 1, %o4
+	EX_LD(LOAD(ldub, %o1, %g1))
+	EX_ST(STORE(stb, %g1, %o0))
+	add		%o1, 1, %o1
+	bne,pt		%XCC, 1b
+	add		%o0, 1, %o0
+
+	andn		%o2, 0x7, %g1
+	sub		%o2, %g1, %o2
+1:	subcc		%g1, 0x8, %g1
+	EX_LD(LOAD(ldx, %o1, %g2))
+	EX_ST(STORE(stx, %g2, %o0))
+	add		%o1, 0x8, %o1
+	bne,pt		%XCC, 1b
+	 add		%o0, 0x8, %o0
+
+	brz,pt		%o2, 85f
+	 sub		%o0, %o1, %o3
+	ba,a,pt		%XCC, 90f
+
+	.align		64
+80: /* 0 < len <= 16 */
+	andcc		%o3, 0x3, %g0
+	bne,pn		%XCC, 90f
+	 sub		%o0, %o1, %o3
+
+1:
+	subcc		%o2, 4, %o2
+	EX_LD(LOAD(lduw, %o1, %g1))
+	EX_ST(STORE(stw, %g1, %o1 + %o3))
+	bgu,pt		%XCC, 1b
+	 add		%o1, 4, %o1
+
+85:	retl
+	 mov		EX_RETVAL(GLOBAL_SPARE), %o0
+
+	.align		32
+90:
+	subcc		%o2, 1, %o2
+	EX_LD(LOAD(ldub, %o1, %g1))
+	EX_ST(STORE(stb, %g1, %o1 + %o3))
+	bgu,pt		%XCC, 90b
+	 add		%o1, 1, %o1
+	retl
+	 mov		EX_RETVAL(GLOBAL_SPARE), %o0
+
+	.size		FUNC_NAME, .-FUNC_NAME

arch/sparc64/lib/GENpage.S

@@ -0,0 +1,77 @@
+/* GENpage.S: Generic clear and copy page.
+ *
+ * Copyright (C) 2007 (davem@davemloft.net)
+ */
+#include <asm/page.h>
+
+	.text
+	.align	32
+
+GENcopy_user_page:
+	set	PAGE_SIZE, %g7
+1:	ldx	[%o1 + 0x00], %o2
+	ldx	[%o1 + 0x08], %o3
+	ldx	[%o1 + 0x10], %o4
+	ldx	[%o1 + 0x18], %o5
+	stx	%o2, [%o0 + 0x00]
+	stx	%o3, [%o0 + 0x08]
+	stx	%o4, [%o0 + 0x10]
+	stx	%o5, [%o0 + 0x18]
+	ldx	[%o1 + 0x20], %o2
+	ldx	[%o1 + 0x28], %o3
+	ldx	[%o1 + 0x30], %o4
+	ldx	[%o1 + 0x38], %o5
+	stx	%o2, [%o0 + 0x20]
+	stx	%o3, [%o0 + 0x28]
+	stx	%o4, [%o0 + 0x30]
+	stx	%o5, [%o0 + 0x38]
+	subcc	%g7, 64, %g7
+	add	%o1, 64, %o1
+	bne,pt	%xcc, 1b
+	 add	%o0, 64, %o0
+	retl
+	 nop
+
+GENclear_page:
+GENclear_user_page:
+	set	PAGE_SIZE, %g7
+1:	stx	%g0, [%o0 + 0x00]
+	stx	%g0, [%o0 + 0x08]
+	stx	%g0, [%o0 + 0x10]
+	stx	%g0, [%o0 + 0x18]
+	stx	%g0, [%o0 + 0x20]
+	stx	%g0, [%o0 + 0x28]
+	stx	%g0, [%o0 + 0x30]
+	stx	%g0, [%o0 + 0x38]
+	subcc	%g7, 64, %g7
+	bne,pt	%xcc, 1b
+	 add	%o0, 64, %o0
+
+#define BRANCH_ALWAYS	0x10680000
+#define NOP		0x01000000
+#define GEN_DO_PATCH(OLD, NEW)	\
+	sethi	%hi(NEW), %g1; \
+	or	%g1, %lo(NEW), %g1; \
+	sethi	%hi(OLD), %g2; \
+	or	%g2, %lo(OLD), %g2; \
+	sub	%g1, %g2, %g1; \
+	sethi	%hi(BRANCH_ALWAYS), %g3; \
+	sll	%g1, 11, %g1; \
+	srl	%g1, 11 + 2, %g1; \
+	or	%g3, %lo(BRANCH_ALWAYS), %g3; \
+	or	%g3, %g1, %g3; \
+	stw	%g3, [%g2]; \
+	sethi	%hi(NOP), %g3; \
+	or	%g3, %lo(NOP), %g3; \
+	stw	%g3, [%g2 + 0x4]; \
+	flush	%g2;
+
+	.globl	generic_patch_pageops
+	.type	generic_patch_pageops,#function
+generic_patch_pageops:
+	GEN_DO_PATCH(copy_user_page, GENcopy_user_page)
+	GEN_DO_PATCH(_clear_page, GENclear_page)
+	GEN_DO_PATCH(clear_user_page, GENclear_user_page)
+	retl
+	 nop
+	.size	generic_patch_pageops,.-generic_patch_pageops

arch/sparc64/lib/GENpatch.S

@@ -0,0 +1,33 @@
+/* GENpatch.S: Patch Ultra-I routines with generic variant.
+ *
+ * Copyright (C) 2007 David S. Miller <davem@davemloft.net>
+ */
+
+#define BRANCH_ALWAYS	0x10680000
+#define NOP		0x01000000
+#define GEN_DO_PATCH(OLD, NEW)	\
+	sethi	%hi(NEW), %g1; \
+	or	%g1, %lo(NEW), %g1; \
+	sethi	%hi(OLD), %g2; \
+	or	%g2, %lo(OLD), %g2; \
+	sub	%g1, %g2, %g1; \
+	sethi	%hi(BRANCH_ALWAYS), %g3; \
+	sll	%g1, 11, %g1; \
+	srl	%g1, 11 + 2, %g1; \
+	or	%g3, %lo(BRANCH_ALWAYS), %g3; \
+	or	%g3, %g1, %g3; \
+	stw	%g3, [%g2]; \
+	sethi	%hi(NOP), %g3; \
+	or	%g3, %lo(NOP), %g3; \
+	stw	%g3, [%g2 + 0x4]; \
+	flush	%g2;
+
+	.globl	generic_patch_copyops
+	.type	generic_patch_copyops,#function
+generic_patch_copyops:
+	GEN_DO_PATCH(memcpy, GENmemcpy)
+	GEN_DO_PATCH(___copy_from_user, GENcopy_from_user)
+	GEN_DO_PATCH(___copy_to_user, GENcopy_to_user)
+	retl
+	 nop
+	.size	generic_patch_copyops,.-generic_patch_copyops

arch/sparc64/lib/Makefile

@@ -1,4 +1,4 @@
-# $Id: Makefile,v 1.25 2000/12/14 22:57:25 davem Exp $
+#
 # Makefile for Sparc64 library files..
 #
 
@@ -13,6 +13,8 @@ lib-y := PeeCeeI.o copy_page.o clear_page.o strlen.o strncmp.o \
 	 U3memcpy.o U3copy_from_user.o U3copy_to_user.o U3patch.o \
 	 NGmemcpy.o NGcopy_from_user.o NGcopy_to_user.o NGpatch.o \
 	 NGpage.o NGbzero.o \
+	 GENmemcpy.o GENcopy_from_user.o GENcopy_to_user.o GENpatch.o \
+	 GENpage.o GENbzero.o \
 	 copy_in_user.o user_fixup.o memmove.o \
 	 mcount.o ipcsum.o rwsem.o xor.o
 

include/asm-sparc64/oplib.h

@@ -1,8 +1,7 @@
-/* $Id: oplib.h,v 1.14 2001/12/19 00:29:51 davem Exp $
+/* oplib.h:  Describes the interface and available routines in the
- * oplib.h:  Describes the interface and available routines in the
  *           Linux Prom library.
  *
- * Copyright (C) 1995 David S. Miller (davem@caip.rutgers.edu)
+ * Copyright (C) 1995, 2007 David S. Miller (davem@davemloft.net)
  * Copyright (C) 1996 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
  */
 
@@ -31,8 +30,10 @@ extern int prom_chosen_node;
 extern const char prom_peer_name[];
 extern const char prom_compatible_name[];
 extern const char prom_root_compatible[];
+extern const char prom_cpu_compatible[];
 extern const char prom_finddev_name[];
 extern const char prom_chosen_path[];
+extern const char prom_cpu_path[];
 extern const char prom_getprop_name[];
 extern const char prom_mmu_name[];
 extern const char prom_callmethod_name[];

include/asm-sparc64/spitfire.h

@@ -38,6 +38,11 @@
 
 #define L1DCACHE_SIZE		0x4000
 
+#define SUN4V_CHIP_INVALID	0x00
+#define SUN4V_CHIP_NIAGARA1	0x01
+#define SUN4V_CHIP_NIAGARA2	0x02
+#define SUN4V_CHIP_UNKNOWN	0xff
+
 #ifndef __ASSEMBLY__
 
 enum ultra_tlb_layout {
@@ -49,6 +54,8 @@ enum ultra_tlb_layout {
 
 extern enum ultra_tlb_layout tlb_type;
 
+extern int sun4v_chip_type;
+
 extern int cheetah_pcache_forced_on;
 extern void cheetah_enable_pcache(void);
 

include/asm-sparc64/xor.h

@@ -63,4 +63,8 @@ static struct xor_block_template xor_block_niagara = {
 
 /* For VIS for everything except Niagara.  */
 #define XOR_SELECT_TEMPLATE(FASTEST) \
-	(tlb_type == hypervisor ? &xor_block_niagara : &xor_block_VIS)
+	((tlb_type == hypervisor && \
+	  (sun4v_chip_type == SUN4V_CHIP_NIAGARA1 || \
+	   sun4v_chip_type == SUN4V_CHIP_NIAGARA2)) ? \
+	 &xor_block_niagara : \
+	 &xor_block_VIS)