aboutsummaryrefslogtreecommitdiffstats
path: root/arch/sparc
diff options
context:
space:
mode:
authorDavid S. Miller <davem@davemloft.net>2011-08-01 21:18:57 -0400
committerDavid S. Miller <davem@davemloft.net>2011-08-03 00:28:32 -0400
commite95ade083939dcb4b0c51c1a2c8504ea9ef3d6ef (patch)
treead81930a3160a865a78869ece5066484a854cff0 /arch/sparc
parentac85fe8b21248054851e05bfaa352562e5b06dd3 (diff)
sparc: Minor tweaks to Niagara page copy/clear.
Don't use floating point on Niagara2, use the traditional plain Niagara code instead. Unroll Niagara loops to 128 bytes for copy, and 256 bytes for clear. Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc')
-rw-r--r--arch/sparc/kernel/head_64.S2
-rw-r--r--arch/sparc/lib/Makefile2
-rw-r--r--arch/sparc/lib/NG2page.S61
-rw-r--r--arch/sparc/lib/NGpage.S114
4 files changed, 77 insertions, 102 deletions
diff --git a/arch/sparc/kernel/head_64.S b/arch/sparc/kernel/head_64.S
index c752603a7c0d..0eac1b2fc53d 100644
--- a/arch/sparc/kernel/head_64.S
+++ b/arch/sparc/kernel/head_64.S
@@ -559,7 +559,7 @@ niagara2_patch:
559 nop 559 nop
560 call niagara_patch_bzero 560 call niagara_patch_bzero
561 nop 561 nop
562 call niagara2_patch_pageops 562 call niagara_patch_pageops
563 nop 563 nop
564 564
565 ba,a,pt %xcc, 80f 565 ba,a,pt %xcc, 80f
diff --git a/arch/sparc/lib/Makefile b/arch/sparc/lib/Makefile
index 7f01b8fce8bc..c25f94d28df8 100644
--- a/arch/sparc/lib/Makefile
+++ b/arch/sparc/lib/Makefile
@@ -31,7 +31,7 @@ lib-$(CONFIG_SPARC64) += NGmemcpy.o NGcopy_from_user.o NGcopy_to_user.o
31lib-$(CONFIG_SPARC64) += NGpatch.o NGpage.o NGbzero.o 31lib-$(CONFIG_SPARC64) += NGpatch.o NGpage.o NGbzero.o
32 32
33lib-$(CONFIG_SPARC64) += NG2memcpy.o NG2copy_from_user.o NG2copy_to_user.o 33lib-$(CONFIG_SPARC64) += NG2memcpy.o NG2copy_from_user.o NG2copy_to_user.o
34lib-$(CONFIG_SPARC64) += NG2patch.o NG2page.o 34lib-$(CONFIG_SPARC64) += NG2patch.o
35 35
36lib-$(CONFIG_SPARC64) += GENmemcpy.o GENcopy_from_user.o GENcopy_to_user.o 36lib-$(CONFIG_SPARC64) += GENmemcpy.o GENcopy_from_user.o GENcopy_to_user.o
37lib-$(CONFIG_SPARC64) += GENpatch.o GENpage.o GENbzero.o 37lib-$(CONFIG_SPARC64) += GENpatch.o GENpage.o GENbzero.o
diff --git a/arch/sparc/lib/NG2page.S b/arch/sparc/lib/NG2page.S
deleted file mode 100644
index 73b6b7c72cbf..000000000000
--- a/arch/sparc/lib/NG2page.S
+++ /dev/null
@@ -1,61 +0,0 @@
1/* NG2page.S: Niagara-2 optimized clear and copy page.
2 *
3 * Copyright (C) 2007 (davem@davemloft.net)
4 */
5
6#include <asm/asi.h>
7#include <asm/page.h>
8#include <asm/visasm.h>
9
10 .text
11 .align 32
12
13 /* This is heavily simplified from the sun4u variants
14 * because Niagara-2 does not have any D-cache aliasing issues.
15 */
16NG2copy_user_page: /* %o0=dest, %o1=src, %o2=vaddr */
17 prefetch [%o1 + 0x00], #one_read
18 prefetch [%o1 + 0x40], #one_read
19 VISEntryHalf
20 set PAGE_SIZE, %g7
21 sub %o0, %o1, %g3
221: stxa %g0, [%o1 + %g3] ASI_BLK_INIT_QUAD_LDD_P
23 subcc %g7, 64, %g7
24 ldda [%o1] ASI_BLK_P, %f0
25 stda %f0, [%o1 + %g3] ASI_BLK_P
26 add %o1, 64, %o1
27 bne,pt %xcc, 1b
28 prefetch [%o1 + 0x40], #one_read
29 membar #Sync
30 VISExitHalf
31 retl
32 nop
33
34#define BRANCH_ALWAYS 0x10680000
35#define NOP 0x01000000
36#define NG_DO_PATCH(OLD, NEW) \
37 sethi %hi(NEW), %g1; \
38 or %g1, %lo(NEW), %g1; \
39 sethi %hi(OLD), %g2; \
40 or %g2, %lo(OLD), %g2; \
41 sub %g1, %g2, %g1; \
42 sethi %hi(BRANCH_ALWAYS), %g3; \
43 sll %g1, 11, %g1; \
44 srl %g1, 11 + 2, %g1; \
45 or %g3, %lo(BRANCH_ALWAYS), %g3; \
46 or %g3, %g1, %g3; \
47 stw %g3, [%g2]; \
48 sethi %hi(NOP), %g3; \
49 or %g3, %lo(NOP), %g3; \
50 stw %g3, [%g2 + 0x4]; \
51 flush %g2;
52
53 .globl niagara2_patch_pageops
54 .type niagara2_patch_pageops,#function
55niagara2_patch_pageops:
56 NG_DO_PATCH(copy_user_page, NG2copy_user_page)
57 NG_DO_PATCH(_clear_page, NGclear_page)
58 NG_DO_PATCH(clear_user_page, NGclear_user_page)
59 retl
60 nop
61 .size niagara2_patch_pageops,.-niagara2_patch_pageops
diff --git a/arch/sparc/lib/NGpage.S b/arch/sparc/lib/NGpage.S
index 428920de05ba..b9e790b9c6b8 100644
--- a/arch/sparc/lib/NGpage.S
+++ b/arch/sparc/lib/NGpage.S
@@ -16,55 +16,91 @@
16 */ 16 */
17 17
18NGcopy_user_page: /* %o0=dest, %o1=src, %o2=vaddr */ 18NGcopy_user_page: /* %o0=dest, %o1=src, %o2=vaddr */
19 prefetch [%o1 + 0x00], #one_read 19 save %sp, -192, %sp
20 mov 8, %g1 20 rd %asi, %g3
21 mov 16, %g2 21 wr %g0, ASI_BLK_INIT_QUAD_LDD_P, %asi
22 mov 24, %g3
23 set PAGE_SIZE, %g7 22 set PAGE_SIZE, %g7
23 prefetch [%i1 + 0x00], #one_read
24 prefetch [%i1 + 0x40], #one_read
24 25
251: ldda [%o1 + %g0] ASI_BLK_INIT_QUAD_LDD_P, %o2 261: prefetch [%i1 + 0x80], #one_read
26 ldda [%o1 + %g2] ASI_BLK_INIT_QUAD_LDD_P, %o4 27 prefetch [%i1 + 0xc0], #one_read
27 prefetch [%o1 + 0x40], #one_read 28 ldda [%i1 + 0x00] %asi, %o2
28 add %o1, 32, %o1 29 ldda [%i1 + 0x10] %asi, %o4
29 stxa %o2, [%o0 + %g0] ASI_BLK_INIT_QUAD_LDD_P 30 ldda [%i1 + 0x20] %asi, %l2
30 stxa %o3, [%o0 + %g1] ASI_BLK_INIT_QUAD_LDD_P 31 ldda [%i1 + 0x30] %asi, %l4
31 ldda [%o1 + %g0] ASI_BLK_INIT_QUAD_LDD_P, %o2 32 stxa %o2, [%i0 + 0x00] %asi
32 stxa %o4, [%o0 + %g2] ASI_BLK_INIT_QUAD_LDD_P 33 stxa %o3, [%i0 + 0x08] %asi
33 stxa %o5, [%o0 + %g3] ASI_BLK_INIT_QUAD_LDD_P 34 stxa %o4, [%i0 + 0x10] %asi
34 ldda [%o1 + %g2] ASI_BLK_INIT_QUAD_LDD_P, %o4 35 stxa %o5, [%i0 + 0x18] %asi
35 add %o1, 32, %o1 36 stxa %l2, [%i0 + 0x20] %asi
36 add %o0, 32, %o0 37 stxa %l3, [%i0 + 0x28] %asi
37 stxa %o2, [%o0 + %g0] ASI_BLK_INIT_QUAD_LDD_P 38 stxa %l4, [%i0 + 0x30] %asi
38 stxa %o3, [%o0 + %g1] ASI_BLK_INIT_QUAD_LDD_P 39 stxa %l5, [%i0 + 0x38] %asi
39 stxa %o4, [%o0 + %g2] ASI_BLK_INIT_QUAD_LDD_P 40 ldda [%i1 + 0x40] %asi, %o2
40 stxa %o5, [%o0 + %g3] ASI_BLK_INIT_QUAD_LDD_P 41 ldda [%i1 + 0x50] %asi, %o4
41 subcc %g7, 64, %g7 42 ldda [%i1 + 0x60] %asi, %l2
43 ldda [%i1 + 0x70] %asi, %l4
44 stxa %o2, [%i0 + 0x40] %asi
45 stxa %o3, [%i0 + 0x48] %asi
46 stxa %o4, [%i0 + 0x50] %asi
47 stxa %o5, [%i0 + 0x58] %asi
48 stxa %l2, [%i0 + 0x60] %asi
49 stxa %l3, [%i0 + 0x68] %asi
50 stxa %l4, [%i0 + 0x70] %asi
51 stxa %l5, [%i0 + 0x78] %asi
52 add %i1, 128, %i1
53 subcc %g7, 128, %g7
42 bne,pt %xcc, 1b 54 bne,pt %xcc, 1b
43 add %o0, 32, %o0 55 add %i0, 128, %i0
56 wr %g3, 0x0, %asi
44 membar #Sync 57 membar #Sync
45 retl 58 ret
46 nop 59 restore
47 60
48 .globl NGclear_page, NGclear_user_page 61 .align 32
49NGclear_page: /* %o0=dest */ 62NGclear_page: /* %o0=dest */
50NGclear_user_page: /* %o0=dest, %o1=vaddr */ 63NGclear_user_page: /* %o0=dest, %o1=vaddr */
51 mov 8, %g1 64 rd %asi, %g3
52 mov 16, %g2 65 wr %g0, ASI_BLK_INIT_QUAD_LDD_P, %asi
53 mov 24, %g3
54 set PAGE_SIZE, %g7 66 set PAGE_SIZE, %g7
55 67
561: stxa %g0, [%o0 + %g0] ASI_BLK_INIT_QUAD_LDD_P 681: stxa %g0, [%o0 + 0x00] %asi
57 stxa %g0, [%o0 + %g1] ASI_BLK_INIT_QUAD_LDD_P 69 stxa %g0, [%o0 + 0x08] %asi
58 stxa %g0, [%o0 + %g2] ASI_BLK_INIT_QUAD_LDD_P 70 stxa %g0, [%o0 + 0x10] %asi
59 stxa %g0, [%o0 + %g3] ASI_BLK_INIT_QUAD_LDD_P 71 stxa %g0, [%o0 + 0x18] %asi
60 add %o0, 32, %o0 72 stxa %g0, [%o0 + 0x20] %asi
61 stxa %g0, [%o0 + %g0] ASI_BLK_INIT_QUAD_LDD_P 73 stxa %g0, [%o0 + 0x28] %asi
62 stxa %g0, [%o0 + %g1] ASI_BLK_INIT_QUAD_LDD_P 74 stxa %g0, [%o0 + 0x30] %asi
63 stxa %g0, [%o0 + %g2] ASI_BLK_INIT_QUAD_LDD_P 75 stxa %g0, [%o0 + 0x38] %asi
64 stxa %g0, [%o0 + %g3] ASI_BLK_INIT_QUAD_LDD_P 76 stxa %g0, [%o0 + 0x40] %asi
65 subcc %g7, 64, %g7 77 stxa %g0, [%o0 + 0x48] %asi
78 stxa %g0, [%o0 + 0x50] %asi
79 stxa %g0, [%o0 + 0x58] %asi
80 stxa %g0, [%o0 + 0x60] %asi
81 stxa %g0, [%o0 + 0x68] %asi
82 stxa %g0, [%o0 + 0x70] %asi
83 stxa %g0, [%o0 + 0x78] %asi
84 stxa %g0, [%o0 + 0x80] %asi
85 stxa %g0, [%o0 + 0x88] %asi
86 stxa %g0, [%o0 + 0x90] %asi
87 stxa %g0, [%o0 + 0x98] %asi
88 stxa %g0, [%o0 + 0xa0] %asi
89 stxa %g0, [%o0 + 0xa8] %asi
90 stxa %g0, [%o0 + 0xb0] %asi
91 stxa %g0, [%o0 + 0xb8] %asi
92 stxa %g0, [%o0 + 0xc0] %asi
93 stxa %g0, [%o0 + 0xc8] %asi
94 stxa %g0, [%o0 + 0xd0] %asi
95 stxa %g0, [%o0 + 0xd8] %asi
96 stxa %g0, [%o0 + 0xe0] %asi
97 stxa %g0, [%o0 + 0xe8] %asi
98 stxa %g0, [%o0 + 0xf0] %asi
99 stxa %g0, [%o0 + 0xf8] %asi
100 subcc %g7, 256, %g7
66 bne,pt %xcc, 1b 101 bne,pt %xcc, 1b
67 add %o0, 32, %o0 102 add %o0, 256, %o0
103 wr %g3, 0x0, %asi
68 membar #Sync 104 membar #Sync
69 retl 105 retl
70 nop 106 nop
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-15 14:26:45 -0500