arch/x86/lib/copy_user_nocache_64.S

   1 /*
   2  * Copyright 2008 Vitaly Mayatskikh <vmayatsk@redhat.com>
   3  * Copyright 2002 Andi Kleen, SuSE Labs.
   4  * Subject to the GNU Public License v2.
   5  *
   6  * Functions to copy from and to user space.
   7  */
   8
   9 #include <linux/linkage.h>
  10 #include <asm/dwarf2.h>
  11
  12 #define FIX_ALIGNMENT 1
  13
  14 #include <asm/current.h>
  15 #include <asm/asm-offsets.h>
  16 #include <asm/thread_info.h>
  17
  18         .macro ALIGN_DESTINATION
  19 #ifdef FIX_ALIGNMENT
  20         /* check for bad alignment of destination */
  21         movl %edi,%ecx
  22         andl $7,%ecx
  23         jz 102f                         /* already aligned */
  24         subl $8,%ecx
  25         negl %ecx
  26         subl %ecx,%edx
  27 100:    movb (%rsi),%al
  28 101:    movb %al,(%rdi)
  29         incq %rsi
  30         incq %rdi
  31         decl %ecx
  32         jnz 100b
  33 102:
  34         .section .fixup,"ax"
  35 103:    addl %r8d,%edx                  /* ecx is zerorest also */
  36         jmp copy_user_handle_tail
  37         .previous
  38
  39         .section __ex_table,"a"
  40         .align 8
  41         .quad 100b,103b
  42         .quad 101b,103b
  43         .previous
  44 #endif
  45         .endm
  46
  47 /*
  48  * copy_user_nocache - Uncached memory copy with exception handling
  49  * This will force destination/source out of cache for more performance.
  50  */
  51 ENTRY(__copy_user_nocache)
  52         CFI_STARTPROC
  53         cmpl $8,%edx
  54         jb 20f          /* less then 8 bytes, go to byte copy loop */
  55         ALIGN_DESTINATION
  56         movl %edx,%ecx
  57         andl $63,%edx
  58         shrl $6,%ecx
  59         jz 17f
  60 1:      movq (%rsi),%r8
  61 2:      movq 1*8(%rsi),%r9
  62 3:      movq 2*8(%rsi),%r10
  63 4:      movq 3*8(%rsi),%r11
  64 5:      movnti %r8,(%rdi)
  65 6:      movnti %r9,1*8(%rdi)
  66 7:      movnti %r10,2*8(%rdi)
  67 8:      movnti %r11,3*8(%rdi)
  68 9:      movq 4*8(%rsi),%r8
  69 10:     movq 5*8(%rsi),%r9
  70 11:     movq 6*8(%rsi),%r10
  71 12:     movq 7*8(%rsi),%r11
  72 13:     movnti %r8,4*8(%rdi)
  73 14:     movnti %r9,5*8(%rdi)
  74 15:     movnti %r10,6*8(%rdi)
  75 16:     movnti %r11,7*8(%rdi)
  76         leaq 64(%rsi),%rsi
  77         leaq 64(%rdi),%rdi
  78         decl %ecx
  79         jnz 1b
  80 17:     movl %edx,%ecx
  81         andl $7,%edx
  82         shrl $3,%ecx
  83         jz 20f
  84 18:     movq (%rsi),%r8
  85 19:     movnti %r8,(%rdi)
  86         leaq 8(%rsi),%rsi
  87         leaq 8(%rdi),%rdi
  88         decl %ecx
  89         jnz 18b
  90 20:     andl %edx,%edx
  91         jz 23f
  92         movl %edx,%ecx
  93 21:     movb (%rsi),%al
  94 22:     movb %al,(%rdi)
  95         incq %rsi
  96         incq %rdi
  97         decl %ecx
  98         jnz 21b
  99 23:     xorl %eax,%eax
 100         sfence
 101         ret
 102
 103         .section .fixup,"ax"
 104 30:     shll $6,%ecx
 105         addl %ecx,%edx
 106         jmp 60f
 107 40:     lea (%rdx,%rcx,8),%rdx
 108         jmp 60f
 109 50:     movl %ecx,%edx
 110 60:     sfence
 111         movl %r8d,%ecx
 112         jmp copy_user_handle_tail
 113         .previous
 114
 115         .section __ex_table,"a"
 116         .quad 1b,30b
 117         .quad 2b,30b
 118         .quad 3b,30b
 119         .quad 4b,30b
 120         .quad 5b,30b
 121         .quad 6b,30b
 122         .quad 7b,30b
 123         .quad 8b,30b
 124         .quad 9b,30b
 125         .quad 10b,30b
 126         .quad 11b,30b
 127         .quad 12b,30b
 128         .quad 13b,30b
 129         .quad 14b,30b
 130         .quad 15b,30b
 131         .quad 16b,30b
 132         .quad 18b,40b
 133         .quad 19b,40b
 134         .quad 21b,50b
 135         .quad 22b,50b
 136         .previous
 137         CFI_ENDPROC
 138 ENDPROC(__copy_user_nocache)