summaryrefslogtreecommitdiffstats
path: root/arch/sparc64/lib
diff options
context:
space:
mode:
authorDavid S. Miller <davem@sunset.davemloft.net>2007-08-08 17:11:39 -0700
committerDavid S. Miller <davem@sunset.davemloft.net>2007-08-08 17:33:45 -0700
commit6c70b6fc7b6fc321636a014082d9e32333da1f80 (patch)
treeff0e52bb7ba43b058b7cbb88b952fd268fbe3ad2 /arch/sparc64/lib
parent68c9f9fd336dc7e793cecad25f8ac40ccaa7a256 (diff)
downloadop-kernel-dev-6c70b6fc7b6fc321636a014082d9e32333da1f80.zip
op-kernel-dev-6c70b6fc7b6fc321636a014082d9e32333da1f80.tar.gz
[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>
Diffstat (limited to 'arch/sparc64/lib')
-rw-r--r--arch/sparc64/lib/GENbzero.S160
-rw-r--r--arch/sparc64/lib/GENcopy_from_user.S34
-rw-r--r--arch/sparc64/lib/GENcopy_to_user.S38
-rw-r--r--arch/sparc64/lib/GENmemcpy.S121
-rw-r--r--arch/sparc64/lib/GENpage.S77
-rw-r--r--arch/sparc64/lib/GENpatch.S33
-rw-r--r--arch/sparc64/lib/Makefile4
7 files changed, 466 insertions, 1 deletions
diff --git a/arch/sparc64/lib/GENbzero.S b/arch/sparc64/lib/GENbzero.S
new file mode 100644
index 0000000..f9c71d6
--- /dev/null
+++ b/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
diff --git a/arch/sparc64/lib/GENcopy_from_user.S b/arch/sparc64/lib/GENcopy_from_user.S
new file mode 100644
index 0000000..2b9df99
--- /dev/null
+++ b/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"
diff --git a/arch/sparc64/lib/GENcopy_to_user.S b/arch/sparc64/lib/GENcopy_to_user.S
new file mode 100644
index 0000000..bb3f708
--- /dev/null
+++ b/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"
diff --git a/arch/sparc64/lib/GENmemcpy.S b/arch/sparc64/lib/GENmemcpy.S
new file mode 100644
index 0000000..89358ee
--- /dev/null
+++ b/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
diff --git a/arch/sparc64/lib/GENpage.S b/arch/sparc64/lib/GENpage.S
new file mode 100644
index 0000000..2ef9d05
--- /dev/null
+++ b/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
diff --git a/arch/sparc64/lib/GENpatch.S b/arch/sparc64/lib/GENpatch.S
new file mode 100644
index 0000000..fab9e89
--- /dev/null
+++ b/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
diff --git a/arch/sparc64/lib/Makefile b/arch/sparc64/lib/Makefile
index c4a6d6e..f95fbfa 100644
--- a/arch/sparc64/lib/Makefile
+++ b/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
OpenPOWER on IntegriCloud