summaryrefslogtreecommitdiffstats
path: root/arch/sparc/lib
diff options
context:
space:
mode:
Diffstat (limited to 'arch/sparc/lib')
-rw-r--r--arch/sparc/lib/Makefile2
-rw-r--r--arch/sparc/lib/NG2memcpy.S46
-rw-r--r--arch/sparc/lib/NG4clear_page.S29
-rw-r--r--arch/sparc/lib/NG4copy_page.S16
-rw-r--r--arch/sparc/lib/NG4memset.S105
-rw-r--r--arch/sparc/lib/NG4patch.S15
6 files changed, 179 insertions, 34 deletions
diff --git a/arch/sparc/lib/Makefile b/arch/sparc/lib/Makefile
index 30f6ab51c55..8410065f286 100644
--- a/arch/sparc/lib/Makefile
+++ b/arch/sparc/lib/Makefile
@@ -33,7 +33,7 @@ lib-$(CONFIG_SPARC64) += NG2memcpy.o NG2copy_from_user.o NG2copy_to_user.o
lib-$(CONFIG_SPARC64) += NG2patch.o
lib-$(CONFIG_SPARC64) += NG4memcpy.o NG4copy_from_user.o NG4copy_to_user.o
-lib-$(CONFIG_SPARC64) += NG4patch.o NG4copy_page.o
+lib-$(CONFIG_SPARC64) += NG4patch.o NG4copy_page.o NG4clear_page.o NG4memset.o
lib-$(CONFIG_SPARC64) += GENmemcpy.o GENcopy_from_user.o GENcopy_to_user.o
lib-$(CONFIG_SPARC64) += GENpatch.o GENpage.o GENbzero.o
diff --git a/arch/sparc/lib/NG2memcpy.S b/arch/sparc/lib/NG2memcpy.S
index 03eadf66b0d..2c20ad63ddb 100644
--- a/arch/sparc/lib/NG2memcpy.S
+++ b/arch/sparc/lib/NG2memcpy.S
@@ -14,7 +14,7 @@
#define FPRS_FEF 0x04
#ifdef MEMCPY_DEBUG
#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs; \
- clr %g1; clr %g2; clr %g3; subcc %g0, %g0, %g0;
+ clr %g1; clr %g2; clr %g3; clr %g5; subcc %g0, %g0, %g0;
#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
#else
#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs
@@ -182,13 +182,13 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
cmp %g2, 0
tne %xcc, 5
PREAMBLE
- mov %o0, GLOBAL_SPARE
+ mov %o0, %o3
cmp %o2, 0
be,pn %XCC, 85f
- or %o0, %o1, %o3
+ or %o0, %o1, GLOBAL_SPARE
cmp %o2, 16
blu,a,pn %XCC, 80f
- or %o3, %o2, %o3
+ or GLOBAL_SPARE, %o2, GLOBAL_SPARE
/* 2 blocks (128 bytes) is the minimum we can do the block
* copy with. We need to ensure that we'll iterate at least
@@ -202,7 +202,7 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
*/
cmp %o2, (4 * 64)
blu,pt %XCC, 75f
- andcc %o3, 0x7, %g0
+ andcc GLOBAL_SPARE, 0x7, %g0
/* %o0: dst
* %o1: src
@@ -404,13 +404,13 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
* over. If anything is left, we copy it one byte at a time.
*/
brz,pt %o2, 85f
- sub %o0, %o1, %o3
+ sub %o0, %o1, GLOBAL_SPARE
ba,a,pt %XCC, 90f
.align 64
75: /* 16 < len <= 64 */
bne,pn %XCC, 75f
- sub %o0, %o1, %o3
+ sub %o0, %o1, GLOBAL_SPARE
72:
andn %o2, 0xf, %o4
@@ -420,9 +420,9 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
add %o1, 0x08, %o1
EX_LD(LOAD(ldx, %o1, %g1))
sub %o1, 0x08, %o1
- EX_ST(STORE(stx, %o5, %o1 + %o3))
+ EX_ST(STORE(stx, %o5, %o1 + GLOBAL_SPARE))
add %o1, 0x8, %o1
- EX_ST(STORE(stx, %g1, %o1 + %o3))
+ EX_ST(STORE(stx, %g1, %o1 + GLOBAL_SPARE))
bgu,pt %XCC, 1b
add %o1, 0x8, %o1
73: andcc %o2, 0x8, %g0
@@ -430,14 +430,14 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
nop
sub %o2, 0x8, %o2
EX_LD(LOAD(ldx, %o1, %o5))
- EX_ST(STORE(stx, %o5, %o1 + %o3))
+ EX_ST(STORE(stx, %o5, %o1 + GLOBAL_SPARE))
add %o1, 0x8, %o1
1: andcc %o2, 0x4, %g0
be,pt %XCC, 1f
nop
sub %o2, 0x4, %o2
EX_LD(LOAD(lduw, %o1, %o5))
- EX_ST(STORE(stw, %o5, %o1 + %o3))
+ EX_ST(STORE(stw, %o5, %o1 + GLOBAL_SPARE))
add %o1, 0x4, %o1
1: cmp %o2, 0
be,pt %XCC, 85f
@@ -454,11 +454,11 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
1: subcc %g1, 1, %g1
EX_LD(LOAD(ldub, %o1, %o5))
- EX_ST(STORE(stb, %o5, %o1 + %o3))
+ EX_ST(STORE(stb, %o5, %o1 + GLOBAL_SPARE))
bgu,pt %icc, 1b
add %o1, 1, %o1
-2: add %o1, %o3, %o0
+2: add %o1, GLOBAL_SPARE, %o0
andcc %o1, 0x7, %g1
bne,pt %icc, 8f
sll %g1, 3, %g1
@@ -468,16 +468,16 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
nop
ba,a,pt %xcc, 73b
-8: mov 64, %o3
+8: mov 64, GLOBAL_SPARE
andn %o1, 0x7, %o1
EX_LD(LOAD(ldx, %o1, %g2))
- sub %o3, %g1, %o3
+ sub GLOBAL_SPARE, %g1, GLOBAL_SPARE
andn %o2, 0x7, %o4
sllx %g2, %g1, %g2
1: add %o1, 0x8, %o1
EX_LD(LOAD(ldx, %o1, %g3))
subcc %o4, 0x8, %o4
- srlx %g3, %o3, %o5
+ srlx %g3, GLOBAL_SPARE, %o5
or %o5, %g2, %o5
EX_ST(STORE(stx, %o5, %o0))
add %o0, 0x8, %o0
@@ -489,32 +489,32 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
be,pn %icc, 85f
add %o1, %g1, %o1
ba,pt %xcc, 90f
- sub %o0, %o1, %o3
+ sub %o0, %o1, GLOBAL_SPARE
.align 64
80: /* 0 < len <= 16 */
- andcc %o3, 0x3, %g0
+ andcc GLOBAL_SPARE, 0x3, %g0
bne,pn %XCC, 90f
- sub %o0, %o1, %o3
+ sub %o0, %o1, GLOBAL_SPARE
1:
subcc %o2, 4, %o2
EX_LD(LOAD(lduw, %o1, %g1))
- EX_ST(STORE(stw, %g1, %o1 + %o3))
+ EX_ST(STORE(stw, %g1, %o1 + GLOBAL_SPARE))
bgu,pt %XCC, 1b
add %o1, 4, %o1
85: retl
- mov EX_RETVAL(GLOBAL_SPARE), %o0
+ mov EX_RETVAL(%o3), %o0
.align 32
90:
subcc %o2, 1, %o2
EX_LD(LOAD(ldub, %o1, %g1))
- EX_ST(STORE(stb, %g1, %o1 + %o3))
+ EX_ST(STORE(stb, %g1, %o1 + GLOBAL_SPARE))
bgu,pt %XCC, 90b
add %o1, 1, %o1
retl
- mov EX_RETVAL(GLOBAL_SPARE), %o0
+ mov EX_RETVAL(%o3), %o0
.size FUNC_NAME, .-FUNC_NAME
diff --git a/arch/sparc/lib/NG4clear_page.S b/arch/sparc/lib/NG4clear_page.S
new file mode 100644
index 00000000000..e16c88204a4
--- /dev/null
+++ b/arch/sparc/lib/NG4clear_page.S
@@ -0,0 +1,29 @@
+/* NG4copy_page.S: Niagara-4 optimized clear page.
+ *
+ * Copyright (C) 2012 (davem@davemloft.net)
+ */
+
+#include <asm/asi.h>
+#include <asm/page.h>
+
+ .text
+
+ .register %g3, #scratch
+
+ .align 32
+ .globl NG4clear_page
+ .globl NG4clear_user_page
+NG4clear_page: /* %o0=dest */
+NG4clear_user_page: /* %o0=dest, %o1=vaddr */
+ set PAGE_SIZE, %g7
+ mov 0x20, %g3
+1: stxa %g0, [%o0 + %g0] ASI_ST_BLKINIT_MRU_P
+ subcc %g7, 0x40, %g7
+ stxa %g0, [%o0 + %g3] ASI_ST_BLKINIT_MRU_P
+ bne,pt %xcc, 1b
+ add %o0, 0x40, %o0
+ membar #StoreLoad|#StoreStore
+ retl
+ nop
+ .size NG4clear_page,.-NG4clear_page
+ .size NG4clear_user_page,.-NG4clear_user_page \ No newline at end of file
diff --git a/arch/sparc/lib/NG4copy_page.S b/arch/sparc/lib/NG4copy_page.S
index f30ec10bbca..28504e88c53 100644
--- a/arch/sparc/lib/NG4copy_page.S
+++ b/arch/sparc/lib/NG4copy_page.S
@@ -30,25 +30,25 @@ NG4copy_user_page: /* %o0=dest, %o1=src, %o2=vaddr */
ldx [%o1 + 0x10], %o4
ldx [%o1 + 0x18], %o5
ldx [%o1 + 0x20], %g1
- stxa %o2, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o2, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
ldx [%o1 + 0x28], %g2
- stxa %o3, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o3, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
ldx [%o1 + 0x30], %g3
- stxa %o4, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o4, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
ldx [%o1 + 0x38], %o2
add %o1, 0x40, %o1
- stxa %o5, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o5, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
- stxa %g1, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %g1, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
- stxa %g2, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %g2, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
- stxa %g3, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %g3, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
- stxa %o2, [%o0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o2, [%o0] ASI_ST_BLKINIT_MRU_P
add %o0, 0x08, %o0
bne,pt %icc, 1b
prefetch [%o1 + 0x200], #n_reads_strong
diff --git a/arch/sparc/lib/NG4memset.S b/arch/sparc/lib/NG4memset.S
new file mode 100644
index 00000000000..41da4bdd95c
--- /dev/null
+++ b/arch/sparc/lib/NG4memset.S
@@ -0,0 +1,105 @@
+/* NG4memset.S: Niagara-4 optimized memset/bzero.
+ *
+ * Copyright (C) 2012 David S. Miller (davem@davemloft.net)
+ */
+
+#include <asm/asi.h>
+
+ .register %g2, #scratch
+ .register %g3, #scratch
+
+ .text
+ .align 32
+ .globl NG4memset
+NG4memset:
+ andcc %o1, 0xff, %o4
+ be,pt %icc, 1f
+ mov %o2, %o1
+ sllx %o4, 8, %g1
+ or %g1, %o4, %o2
+ sllx %o2, 16, %g1
+ or %g1, %o2, %o2
+ sllx %o2, 32, %g1
+ ba,pt %icc, 1f
+ or %g1, %o2, %o4
+ .size NG4memset,.-NG4memset
+
+ .align 32
+ .globl NG4bzero
+NG4bzero:
+ clr %o4
+1: cmp %o1, 16
+ ble %icc, .Ltiny
+ mov %o0, %o3
+ sub %g0, %o0, %g1
+ and %g1, 0x7, %g1
+ brz,pt %g1, .Laligned8
+ sub %o1, %g1, %o1
+1: stb %o4, [%o0 + 0x00]
+ subcc %g1, 1, %g1
+ bne,pt %icc, 1b
+ add %o0, 1, %o0
+.Laligned8:
+ cmp %o1, 64 + (64 - 8)
+ ble .Lmedium
+ sub %g0, %o0, %g1
+ andcc %g1, (64 - 1), %g1
+ brz,pn %g1, .Laligned64
+ sub %o1, %g1, %o1
+1: stx %o4, [%o0 + 0x00]
+ subcc %g1, 8, %g1
+ bne,pt %icc, 1b
+ add %o0, 0x8, %o0
+.Laligned64:
+ andn %o1, 64 - 1, %g1
+ sub %o1, %g1, %o1
+ brnz,pn %o4, .Lnon_bzero_loop
+ mov 0x20, %g2
+1: stxa %o4, [%o0 + %g0] ASI_BLK_INIT_QUAD_LDD_P
+ subcc %g1, 0x40, %g1
+ stxa %o4, [%o0 + %g2] ASI_BLK_INIT_QUAD_LDD_P
+ bne,pt %icc, 1b
+ add %o0, 0x40, %o0
+.Lpostloop:
+ cmp %o1, 8
+ bl,pn %icc, .Ltiny
+ membar #StoreStore|#StoreLoad
+.Lmedium:
+ andn %o1, 0x7, %g1
+ sub %o1, %g1, %o1
+1: stx %o4, [%o0 + 0x00]
+ subcc %g1, 0x8, %g1
+ bne,pt %icc, 1b
+ add %o0, 0x08, %o0
+ andcc %o1, 0x4, %g1
+ be,pt %icc, .Ltiny
+ sub %o1, %g1, %o1
+ stw %o4, [%o0 + 0x00]
+ add %o0, 0x4, %o0
+.Ltiny:
+ cmp %o1, 0
+ be,pn %icc, .Lexit
+1: subcc %o1, 1, %o1
+ stb %o4, [%o0 + 0x00]
+ bne,pt %icc, 1b
+ add %o0, 1, %o0
+.Lexit:
+ retl
+ mov %o3, %o0
+.Lnon_bzero_loop:
+ mov 0x08, %g3
+ mov 0x28, %o5
+1: stxa %o4, [%o0 + %g0] ASI_BLK_INIT_QUAD_LDD_P
+ subcc %g1, 0x40, %g1
+ stxa %o4, [%o0 + %g2] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o4, [%o0 + %g3] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o4, [%o0 + %o5] ASI_BLK_INIT_QUAD_LDD_P
+ add %o0, 0x10, %o0
+ stxa %o4, [%o0 + %g0] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o4, [%o0 + %g2] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o4, [%o0 + %g3] ASI_BLK_INIT_QUAD_LDD_P
+ stxa %o4, [%o0 + %o5] ASI_BLK_INIT_QUAD_LDD_P
+ bne,pt %icc, 1b
+ add %o0, 0x30, %o0
+ ba,a,pt %icc, .Lpostloop
+ .size NG4bzero,.-NG4bzero
diff --git a/arch/sparc/lib/NG4patch.S b/arch/sparc/lib/NG4patch.S
index c21c34c61dd..a114cbcf2a4 100644
--- a/arch/sparc/lib/NG4patch.S
+++ b/arch/sparc/lib/NG4patch.S
@@ -32,12 +32,23 @@ niagara4_patch_copyops:
nop
.size niagara4_patch_copyops,.-niagara4_patch_copyops
+ .globl niagara4_patch_bzero
+ .type niagara4_patch_bzero,#function
+niagara4_patch_bzero:
+ NG_DO_PATCH(memset, NG4memset)
+ NG_DO_PATCH(__bzero, NG4bzero)
+ NG_DO_PATCH(__clear_user, NGclear_user)
+ NG_DO_PATCH(tsb_init, NGtsb_init)
+ retl
+ nop
+ .size niagara4_patch_bzero,.-niagara4_patch_bzero
+
.globl niagara4_patch_pageops
.type niagara4_patch_pageops,#function
niagara4_patch_pageops:
NG_DO_PATCH(copy_user_page, NG4copy_user_page)
- NG_DO_PATCH(_clear_page, NGclear_page)
- NG_DO_PATCH(clear_user_page, NGclear_user_page)
+ NG_DO_PATCH(_clear_page, NG4clear_page)
+ NG_DO_PATCH(clear_user_page, NG4clear_user_page)
retl
nop
.size niagara4_patch_pageops,.-niagara4_patch_pageops
generated by cgit v1.2.3-70-g09d2 (git 2.46.0) at 2025-03-23 13:06:43 +0000