aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/lib
diff options
context:
space:
mode:
authorFenghua Yu <fenghua.yu@intel.com>2011-05-17 18:29:15 -0400
committerH. Peter Anvin <hpa@linux.intel.com>2011-05-17 18:40:28 -0400
commit4307bec9344aed83f8107c3eb4285bd9d218fc10 (patch)
treed18d0b7e6c5d0a93fc84504d546add5f1ca09c5a /arch/x86/lib
parente365c9df2f2f001450decf9512412d2d5bd1cdef (diff)
x86, mem: copy_user_64.S: Support copy_to/from_user by enhanced REP MOVSB/STOSB
Support copy_to_user/copy_from_user() by enhanced REP MOVSB/STOSB. On processors supporting enhanced REP MOVSB/STOSB, the alternative copy_user_enhanced_fast_string function using enhanced rep movsb overrides the original function and the fast string function. Signed-off-by: Fenghua Yu <fenghua.yu@intel.com> Link: http://lkml.kernel.org/r/1305671358-14478-7-git-send-email-fenghua.yu@intel.com Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Diffstat (limited to 'arch/x86/lib')
-rw-r--r--arch/x86/lib/copy_user_64.S65
1 files changed, 55 insertions, 10 deletions
diff --git a/arch/x86/lib/copy_user_64.S b/arch/x86/lib/copy_user_64.S
index 99e482615195..d17a1170bbf5 100644
--- a/arch/x86/lib/copy_user_64.S
+++ b/arch/x86/lib/copy_user_64.S
@@ -15,23 +15,30 @@
15#include <asm/asm-offsets.h> 15#include <asm/asm-offsets.h>
16#include <asm/thread_info.h> 16#include <asm/thread_info.h>
17#include <asm/cpufeature.h> 17#include <asm/cpufeature.h>
18#include <asm/alternative-asm.h>
18 19
19 .macro ALTERNATIVE_JUMP feature,orig,alt 20/*
21 * By placing feature2 after feature1 in altinstructions section, we logically
22 * implement:
23 * If CPU has feature2, jmp to alt2 is used
24 * else if CPU has feature1, jmp to alt1 is used
25 * else jmp to orig is used.
26 */
27 .macro ALTERNATIVE_JUMP feature1,feature2,orig,alt1,alt2
200: 280:
21 .byte 0xe9 /* 32bit jump */ 29 .byte 0xe9 /* 32bit jump */
22 .long \orig-1f /* by default jump to orig */ 30 .long \orig-1f /* by default jump to orig */
231: 311:
24 .section .altinstr_replacement,"ax" 32 .section .altinstr_replacement,"ax"
252: .byte 0xe9 /* near jump with 32bit immediate */ 332: .byte 0xe9 /* near jump with 32bit immediate */
26 .long \alt-1b /* offset */ /* or alternatively to alt */ 34 .long \alt1-1b /* offset */ /* or alternatively to alt1 */
353: .byte 0xe9 /* near jump with 32bit immediate */
36 .long \alt2-1b /* offset */ /* or alternatively to alt2 */
27 .previous 37 .previous
38
28 .section .altinstructions,"a" 39 .section .altinstructions,"a"
29 .align 8 40 altinstruction_entry 0b,2b,\feature1,5,5
30 .quad 0b 41 altinstruction_entry 0b,3b,\feature2,5,5
31 .quad 2b
32 .word \feature /* when feature is set */
33 .byte 5
34 .byte 5
35 .previous 42 .previous
36 .endm 43 .endm
37 44
@@ -73,7 +80,9 @@ ENTRY(_copy_to_user)
73 jc bad_to_user 80 jc bad_to_user
74 cmpq TI_addr_limit(%rax),%rcx 81 cmpq TI_addr_limit(%rax),%rcx
75 jae bad_to_user 82 jae bad_to_user
76 ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,copy_user_generic_unrolled,copy_user_generic_string 83 ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,X86_FEATURE_ERMS, \
84 copy_user_generic_unrolled,copy_user_generic_string, \
85 copy_user_enhanced_fast_string
77 CFI_ENDPROC 86 CFI_ENDPROC
78ENDPROC(_copy_to_user) 87ENDPROC(_copy_to_user)
79 88
@@ -86,7 +95,9 @@ ENTRY(_copy_from_user)
86 jc bad_from_user 95 jc bad_from_user
87 cmpq TI_addr_limit(%rax),%rcx 96 cmpq TI_addr_limit(%rax),%rcx
88 jae bad_from_user 97 jae bad_from_user
89 ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,copy_user_generic_unrolled,copy_user_generic_string 98 ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,X86_FEATURE_ERMS, \
99 copy_user_generic_unrolled,copy_user_generic_string, \
100 copy_user_enhanced_fast_string
90 CFI_ENDPROC 101 CFI_ENDPROC
91ENDPROC(_copy_from_user) 102ENDPROC(_copy_from_user)
92 103
@@ -255,3 +266,37 @@ ENTRY(copy_user_generic_string)
255 .previous 266 .previous
256 CFI_ENDPROC 267 CFI_ENDPROC
257ENDPROC(copy_user_generic_string) 268ENDPROC(copy_user_generic_string)
269
270/*
271 * Some CPUs are adding enhanced REP MOVSB/STOSB instructions.
272 * It's recommended to use enhanced REP MOVSB/STOSB if it's enabled.
273 *
274 * Input:
275 * rdi destination
276 * rsi source
277 * rdx count
278 *
279 * Output:
280 * eax uncopied bytes or 0 if successful.
281 */
282ENTRY(copy_user_enhanced_fast_string)
283 CFI_STARTPROC
284 andl %edx,%edx
285 jz 2f
286 movl %edx,%ecx
2871: rep
288 movsb
2892: xorl %eax,%eax
290 ret
291
292 .section .fixup,"ax"
29312: movl %ecx,%edx /* ecx is zerorest also */
294 jmp copy_user_handle_tail
295 .previous
296
297 .section __ex_table,"a"
298 .align 8
299 .quad 1b,12b
300 .previous
301 CFI_ENDPROC
302ENDPROC(copy_user_enhanced_fast_string)