1/* Written 2003 by Andi Kleen, based on a kernel by Evandro Menezes */ 2 3#include <linux/linkage.h> 4#include <asm/dwarf2.h> 5#include <asm/alternative-asm.h> 6 7 ALIGN 8copy_page_c: 9 CFI_STARTPROC 10 movl $4096/8,%ecx 11 rep movsq 12 ret 13 CFI_ENDPROC 14ENDPROC(copy_page_c) 15 16/* Don't use streaming store because it's better when the target 17 ends up in cache. */ 18 19/* Could vary the prefetch distance based on SMP/UP */ 20 21ENTRY(copy_page) 22 CFI_STARTPROC 23 subq $2*8,%rsp 24 CFI_ADJUST_CFA_OFFSET 2*8 25 movq %rbx,(%rsp) 26 CFI_REL_OFFSET rbx, 0 27 movq %r12,1*8(%rsp) 28 CFI_REL_OFFSET r12, 1*8 29 30 movl $(4096/64)-5,%ecx 31 .p2align 4 32.Loop64: 33 dec %rcx 34 35 movq (%rsi), %rax 36 movq 8 (%rsi), %rbx 37 movq 16 (%rsi), %rdx 38 movq 24 (%rsi), %r8 39 movq 32 (%rsi), %r9 40 movq 40 (%rsi), %r10 41 movq 48 (%rsi), %r11 42 movq 56 (%rsi), %r12 43 44 prefetcht0 5*64(%rsi) 45 46 movq %rax, (%rdi) 47 movq %rbx, 8 (%rdi) 48 movq %rdx, 16 (%rdi) 49 movq %r8, 24 (%rdi) 50 movq %r9, 32 (%rdi) 51 movq %r10, 40 (%rdi) 52 movq %r11, 48 (%rdi) 53 movq %r12, 56 (%rdi) 54 55 leaq 64 (%rsi), %rsi 56 leaq 64 (%rdi), %rdi 57 58 jnz .Loop64 59 60 movl $5,%ecx 61 .p2align 4 62.Loop2: 63 decl %ecx 64 65 movq (%rsi), %rax 66 movq 8 (%rsi), %rbx 67 movq 16 (%rsi), %rdx 68 movq 24 (%rsi), %r8 69 movq 32 (%rsi), %r9 70 movq 40 (%rsi), %r10 71 movq 48 (%rsi), %r11 72 movq 56 (%rsi), %r12 73 74 movq %rax, (%rdi) 75 movq %rbx, 8 (%rdi) 76 movq %rdx, 16 (%rdi) 77 movq %r8, 24 (%rdi) 78 movq %r9, 32 (%rdi) 79 movq %r10, 40 (%rdi) 80 movq %r11, 48 (%rdi) 81 movq %r12, 56 (%rdi) 82 83 leaq 64(%rdi),%rdi 84 leaq 64(%rsi),%rsi 85 86 jnz .Loop2 87 88 movq (%rsp),%rbx 89 CFI_RESTORE rbx 90 movq 1*8(%rsp),%r12 91 CFI_RESTORE r12 92 addq $2*8,%rsp 93 CFI_ADJUST_CFA_OFFSET -2*8 94 ret 95.Lcopy_page_end: 96 CFI_ENDPROC 97ENDPROC(copy_page) 98 99 /* Some CPUs run faster using the string copy instructions. 100 It is also a lot simpler. Use this when possible */ 101 102#include <asm/cpufeature.h> 103 104 .section .altinstr_replacement,"ax" 1051: .byte 0xeb /* jmp <disp8> */ 106 .byte (copy_page_c - copy_page) - (2f - 1b) /* offset */ 1072: 108 .previous 109 .section .altinstructions,"a" 110 altinstruction_entry copy_page, 1b, X86_FEATURE_REP_GOOD, \ 111 .Lcopy_page_end-copy_page, 2b-1b 112 .previous 113