[SPARC64]: Niagara optimized memset/bzero/clear_user.

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

2 files changed, 163 insertions, 1 deletions

diff --git a/arch/sparc64/lib/Makefile b/arch/sparc64/lib/Makefile
index 3d0e9a24d7a3..8812ded19f01 100644
--- a/arch/sparc64/lib/Makefile
+++ b/arch/sparc64/lib/Makefile
@@ -12,7 +12,7 @@ lib-y := PeeCeeI.o copy_page.o clear_page.o strlen.o strncmp.o \
 	 U1memcpy.o U1copy_from_user.o U1copy_to_user.o \
 	 U3memcpy.o U3copy_from_user.o U3copy_to_user.o U3patch.o \
 	 NGmemcpy.o NGcopy_from_user.o NGcopy_to_user.o NGpatch.o \
-	 NGpage.o \
+	 NGpage.o NGbzero.o \
 	 copy_in_user.o user_fixup.o memmove.o \
 	 mcount.o ipcsum.o rwsem.o xor.o find_bit.o delay.o
 
diff --git a/arch/sparc64/lib/NGbzero.S b/arch/sparc64/lib/NGbzero.S
new file mode 100644
index 000000000000..fef584f745dc
--- /dev/null
+++ b/arch/sparc64/lib/NGbzero.S
@@ -0,0 +1,162 @@
+/* NGbzero.S: Niagara optimized memset/clear_user.
+ *
+ * Copyright (C) 2006 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;
+
+	.text
+
+	.globl		NGmemset
+	.type		NGmemset, #function
+NGmemset:		/* %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		NGbzero
+	.type		NGbzero, #function
+NGbzero:
+	clr		%o2
+1:	brz,pn		%o1, NGbzero_return
+	 mov		%o0, %o3
+
+	/* %o5: saved %asi, restored at NGbzero_done
+	 * %g7: store-init %asi to use
+	 * %o4:	non-store-init %asi to use
+	 */
+	rd		%asi, %o5
+	mov		ASI_BLK_INIT_QUAD_LDD_P, %g7
+	mov		ASI_P, %o4
+	wr		%o4, 0x0, %asi
+
+NGbzero_from_clear_user:
+	cmp		%o1, 15
+	bl,pn		%icc, NGbzero_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, NGbzero_medium
+	 andcc		%o0, (64 - 1), %g1
+	be,pt		%xcc, NGbzero_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
+
+NGbzero_pre_loop:
+	wr		%g7, 0x0, %asi
+	andn		%o1, (64 - 1), %g1
+	sub		%o1, %g1, %o1
+NGbzero_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, NGbzero_loop
+	 add		%o0, 64, %o0
+
+	wr		%o4, 0x0, %asi
+	brz,pn		%o1, NGbzero_done
+NGbzero_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, NGbzero_done
+	 nop
+
+NGbzero_tiny:
+1:	EX_ST(stba %o2, [%o0 + 0x00] %asi)
+	subcc		%o1, 1, %o1
+	bne,pt		%icc, 1b
+	 add		%o0, 1, %o0
+
+	/* fallthrough */
+
+NGbzero_done:
+	wr		%o5, 0x0, %asi
+
+NGbzero_return:
+	retl
+	 mov		%o3, %o0
+	.size		NGbzero, .-NGbzero
+	.size		NGmemset, .-NGmemset
+
+	.globl		NGclear_user
+	.type		NGclear_user, #function
+NGclear_user:		/* %o0=buf, %o1=len */
+	rd		%asi, %o5
+	brz,pn		%o1, NGbzero_done
+	 clr		%o3
+	cmp		%o5, ASI_AIUS
+	bne,pn		%icc, NGbzero
+	 clr		%o2
+	mov		ASI_BLK_INIT_QUAD_LDD_AIUS, %g7
+	ba,pt		%xcc, NGbzero_from_clear_user
+	 mov		ASI_AIUS, %o4
+	.size		NGclear_user, .-NGclear_user
+
+#define BRANCH_ALWAYS	0x10680000
+#define NOP		0x01000000
+#define NG_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	niagara_patch_bzero
+	.type	niagara_patch_bzero,#function
+niagara_patch_bzero:
+	NG_DO_PATCH(memset, NGmemset)
+	NG_DO_PATCH(__bzero, NGbzero)
+	NG_DO_PATCH(__clear_user, NGclear_user)
+	retl
+	 nop
+	.size	niagara_patch_bzero,.-niagara_patch_bzero