| OLD | NEW |
|---|
| 1 /* | 1 /* |
| 2 * ==================================================== | 2 * ==================================================== |
| 3 * Copyright (C) 2007 by Ellips BV. All rights reserved. | 3 * Copyright (C) 2007 by Ellips BV. All rights reserved. |
| 4 * | 4 * |
| 5 * Permission to use, copy, modify, and distribute this | 5 * Permission to use, copy, modify, and distribute this |
| 6 * software is freely granted, provided that this notice | 6 * software is freely granted, provided that this notice |
| 7 * is preserved. | 7 * is preserved. |
| 8 * ==================================================== | 8 * ==================================================== |
| 9 */ | 9 */ |
| 10 | 10 |
| 11 #include "x86_64mach.h" | 11 #include "x86_64mach.h" |
| 12 | 12 |
| 13 .global SYM (memset) | 13 .global SYM (memset) |
| 14 SOTYPE_FUNCTION(memset) | 14 SOTYPE_FUNCTION(memset) |
| 15 | 15 |
| 16 /* |
| 17 * Avoid revealing the sandbox base address. In particular, we never read from |
| 18 * r11, and avoid using the string instructions (stos) because they leave the |
| 19 * full 64 bits in rdi. |
| 20 */ |
| 21 |
| 16 SYM (memset): | 22 SYM (memset): |
| 17 movl edi, r9d /* Save return value */ | 23 movl edi, r9d /* Save return value */ |
| 18 movl esi, eax | 24 movl esi, eax |
| 19 movl edx, ecx | 25 movl edx, ecx |
| 20 cmpl $16, edx | 26 cmpl $16, edx |
| 21 jb byte_set | 27 jb .Lbyte_set |
| 22 | 28 |
| 23 movl edi, r8d /* Align on quad word boundary */ | 29 movl edi, r8d /* Align on quad word boundary */ |
| 24 andl $7, r8d | 30 andl $7, r8d |
| 25 jz quadword_aligned | 31 jz .Lquadword_aligned |
| 26 movl $8, ecx | 32 movl $8, ecx |
| 27 subl r8d, ecx | 33 subl r8d, ecx |
| 28 subl ecx, edx | 34 subl ecx, edx |
| 29 rep stosb al, %nacl:(rdi), r15 | 35 .Lheader_loop: |
| 36 movb al, %nacl:(r15, rdi) |
| 37 inc edi |
| 38 dec ecx |
| 39 jnz .Lheader_loop |
| 40 |
| 30 movl edx, ecx | 41 movl edx, ecx |
| 31 | 42 |
| 32 quadword_aligned: | 43 .Lquadword_aligned: |
| 33 movabs $0x0101010101010101, r8 | 44 movabs $0x0101010101010101, r8 |
| 34 movzbl sil, eax | 45 movzbl sil, eax |
| 35 imul r8, rax | 46 imul r8, rax |
| 36 cmpl $256, edx | 47 cmpl $256, edx |
| 37 jb quadword_set | 48 jb .Lquadword_set |
| 38 | 49 |
| 39 shrl $7, ecx /* Store 128 bytes at a time with minimum cach
e polution */ | 50 shrl $7, ecx /* Store 128 bytes at a time */ |
| 51 |
| 52 /* Save the frame pointer and use it as a base to avoid repeated masking */ |
| 53 pushq $0 |
| 54 movl ebp, (rsp) |
| 55 naclrestbp edi, r15 |
| 40 | 56 |
| 41 .p2align 4 | 57 .p2align 4 |
| 42 loop: | 58 .Lquadword_aligned_loop: |
| 43 movntiq rax, %nacl: (r15,rdi) | |
| 44 movntiq rax, %nacl: 8 (r15,rdi) | |
| 45 movntiq rax, %nacl: 16 (r15,rdi) | |
| 46 movntiq rax, %nacl: 24 (r15,rdi) | |
| 47 movntiq rax, %nacl: 32 (r15,rdi) | |
| 48 movntiq rax, %nacl: 40 (r15,rdi) | |
| 49 movntiq rax, %nacl: 48 (r15,rdi) | |
| 50 movntiq rax, %nacl: 56 (r15,rdi) | |
| 51 movntiq rax, %nacl: 64 (r15,rdi) | |
| 52 movntiq rax, %nacl: 72 (r15,rdi) | |
| 53 movntiq rax, %nacl: 80 (r15,rdi) | |
| 54 movntiq rax, %nacl: 88 (r15,rdi) | |
| 55 movntiq rax, %nacl: 96 (r15,rdi) | |
| 56 movntiq rax, %nacl: 104 (r15,rdi) | |
| 57 movntiq rax, %nacl: 112 (r15,rdi) | |
| 58 movntiq rax, %nacl: 120 (r15,rdi) | |
| 59 | 59 |
| 60 leal 128 (rdi), edi | 60 movq rax, (rbp) |
| 61 movq rax, 8 (rbp) |
| 62 movq rax, 16 (rbp) |
| 63 movq rax, 24 (rbp) |
| 64 movq rax, 32 (rbp) |
| 65 movq rax, 40 (rbp) |
| 66 movq rax, 48 (rbp) |
| 67 movq rax, 56 (rbp) |
| 68 movq rax, 64 (rbp) |
| 69 movq rax, 72 (rbp) |
| 70 movq rax, 80 (rbp) |
| 71 movq rax, 88 (rbp) |
| 72 movq rax, 96 (rbp) |
| 73 movq rax, 104 (rbp) |
| 74 movq rax, 112 (rbp) |
| 75 movq rax, 120 (rbp) |
| 76 |
| 77 .bundle_lock |
| 78 leal 128 (rbp), ebp |
| 79 add r15, rbp |
| 80 .bundle_unlock |
| 61 | 81 |
| 62 dec ecx | 82 dec ecx |
| 63 jnz loop | 83 jnz .Lquadword_aligned_loop |
| 64 | 84 |
| 65 sfence | 85 movl ebp, edi |
| 86 popq r8 |
| 87 naclrestbp r8d, r15 |
| 88 |
| 66 movl edx, ecx | 89 movl edx, ecx |
| 67 andl $127, ecx | 90 andl $127, ecx |
| 68 rep stosb al, %nacl:(rdi), r15 | 91 jz .Lfooter_loop_end |
| 92 .Lfooter_loop: |
| 93 movb al, %nacl:(r15, rdi) |
| 94 inc edi |
| 95 dec ecx |
| 96 jnz .Lfooter_loop |
| 97 .Lfooter_loop_end: |
| 69 movl r9d, eax | 98 movl r9d, eax |
| 70 pop r11 | 99 pop r11 |
| 71 nacljmp r11d, r15 | 100 nacljmp r11d, r15 |
| 72 | 101 |
| 73 | 102 |
| 74 byte_set: | 103 .Lbyte_set: |
| 75 rep stosb al, %nacl:(rdi), r15 | 104 testl ecx, ecx |
| 105 jz .Lbyte_set_end |
| 106 .Lbyte_set_loop: |
| 107 movb al, %nacl:(r15, rdi) |
| 108 inc edi |
| 109 dec ecx |
| 110 jnz .Lbyte_set_loop |
| 111 .Lbyte_set_end: |
| 76 movl r9d, eax | 112 movl r9d, eax |
| 77 pop r11 | 113 pop r11 |
| 78 nacljmp r11d, r15 | 114 nacljmp r11d, r15 |
| 79 | 115 |
| 80 | 116 |
| 81 quadword_set: | 117 .Lquadword_set: |
| 82 shrl $3, ecx | 118 shrl $3, ecx |
| 119 jz .Lquadword_loop_end |
| 83 .p2align 4 | 120 .p2align 4 |
| 84 rep stosq rax, %nacl:(rdi), r15 | 121 .Lquadword_loop: |
| 122 movq rax, %nacl:(r15, rdi) |
| 123 add $8, edi |
| 124 dec ecx |
| 125 jnz .Lquadword_loop |
| 126 .Lquadword_loop_end: |
| 85 movl edx, ecx | 127 movl edx, ecx |
| 86 andl $7, ecx | 128 andl $7, ecx |
| 87 rep stosb al, %nacl:(rdi), r15 /* Store the remaining bytes */ | 129 jz .Lquadword_footer_end |
| 130 .Lquadword_footer: |
| 131 movb al, %nacl:(r15, rdi) |
| 132 inc edi |
| 133 dec ecx |
| 134 jnz .Lquadword_footer |
| 135 .Lquadword_footer_end: |
| 88 movl r9d, eax | 136 movl r9d, eax |
| 89 pop r11 | 137 pop r11 |
| 90 nacljmp r11d, r15 | 138 nacljmp r11d, r15 |
| 91 | 139 |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 957063002:
Avoid string instructions in x86_64 memset (Closed)
Base URL: https://chromium.googlesource.com/native_client/nacl-newlib.git@master