arch/x86/lib/copy_page_64.S

   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
   8 copy_page_rep:
   9         CFI_STARTPROC
  10         movl    $4096/8, %ecx
  11         rep     movsq
  12         ret
  13         CFI_ENDPROC
  14 ENDPROC(copy_page_rep)
  15
  16 /*
  17  *  Don't use streaming copy unless the CPU indicates X86_FEATURE_REP_GOOD.
  18  *  Could vary the prefetch distance based on SMP/UP.
  19 */
  20
  21 ENTRY(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         movq    0x8*0(%rsi), %rax
  35         movq    0x8*1(%rsi), %rbx
  36         movq    0x8*2(%rsi), %rdx
  37         movq    0x8*3(%rsi), %r8
  38         movq    0x8*4(%rsi), %r9
  39         movq    0x8*5(%rsi), %r10
  40         movq    0x8*6(%rsi), %r11
  41         movq    0x8*7(%rsi), %r12
  42
  43         prefetcht0 5*64(%rsi)
  44
  45         movq    %rax, 0x8*0(%rdi)
  46         movq    %rbx, 0x8*1(%rdi)
  47         movq    %rdx, 0x8*2(%rdi)
  48         movq    %r8,  0x8*3(%rdi)
  49         movq    %r9,  0x8*4(%rdi)
  50         movq    %r10, 0x8*5(%rdi)
  51         movq    %r11, 0x8*6(%rdi)
  52         movq    %r12, 0x8*7(%rdi)
  53
  54         leaq    64 (%rsi), %rsi
  55         leaq    64 (%rdi), %rdi
  56
  57         jnz     .Loop64
  58
  59         movl    $5, %ecx
  60         .p2align 4
  61 .Loop2:
  62         decl    %ecx
  63
  64         movq    0x8*0(%rsi), %rax
  65         movq    0x8*1(%rsi), %rbx
  66         movq    0x8*2(%rsi), %rdx
  67         movq    0x8*3(%rsi), %r8
  68         movq    0x8*4(%rsi), %r9
  69         movq    0x8*5(%rsi), %r10
  70         movq    0x8*6(%rsi), %r11
  71         movq    0x8*7(%rsi), %r12
  72
  73         movq    %rax, 0x8*0(%rdi)
  74         movq    %rbx, 0x8*1(%rdi)
  75         movq    %rdx, 0x8*2(%rdi)
  76         movq    %r8,  0x8*3(%rdi)
  77         movq    %r9,  0x8*4(%rdi)
  78         movq    %r10, 0x8*5(%rdi)
  79         movq    %r11, 0x8*6(%rdi)
  80         movq    %r12, 0x8*7(%rdi)
  81
  82         leaq    64(%rdi), %rdi
  83         leaq    64(%rsi), %rsi
  84         jnz     .Loop2
  85
  86         movq    (%rsp), %rbx
  87         CFI_RESTORE rbx
  88         movq    1*8(%rsp), %r12
  89         CFI_RESTORE r12
  90         addq    $2*8, %rsp
  91         CFI_ADJUST_CFA_OFFSET -2*8
  92         ret
  93 .Lcopy_page_end:
  94         CFI_ENDPROC
  95 ENDPROC(copy_page)
  96
  97         /* Some CPUs run faster using the string copy instructions.
  98            It is also a lot simpler. Use this when possible */
  99
 100 #include <asm/cpufeature.h>
 101
 102         .section .altinstr_replacement,"ax"
 103 1:      .byte 0xeb                                      /* jmp <disp8> */
 104         .byte (copy_page_rep - copy_page) - (2f - 1b)   /* offset */
 105 2:
 106         .previous
 107         .section .altinstructions,"a"
 108         altinstruction_entry copy_page, 1b, X86_FEATURE_REP_GOOD,       \
 109                 .Lcopy_page_end-copy_page, 2b-1b
 110         .previous