arch/x86/lib/copy_user_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 #include <asm/cpufeature.h>
  18
  19         .macro ALTERNATIVE_JUMP feature,orig,alt
  20 0:
  21         .byte 0xe9      /* 32bit jump */
  22         .long \orig-1f  /* by default jump to orig */
  23 1:
  24         .section .altinstr_replacement,"ax"
  25 2:      .byte 0xe9                      /* near jump with 32bit immediate */
  26         .long \alt-1b /* offset */   /* or alternatively to alt */
  27         .previous
  28         .section .altinstructions,"a"
  29         .align 8
  30         .quad  0b
  31         .quad  2b
  32         .byte  \feature                 /* when feature is set */
  33         .byte  5
  34         .byte  5
  35         .previous
  36         .endm
  37
  38         .macro ALIGN_DESTINATION
  39 #ifdef FIX_ALIGNMENT
  40         /* check for bad alignment of destination */
  41         movl %edi,%ecx
  42         andl $7,%ecx
  43         jz 102f                         /* already aligned */
  44         subl $8,%ecx
  45         negl %ecx
  46         subl %ecx,%edx
  47 100:    movb (%rsi),%al
  48 101:    movb %al,(%rdi)
  49         incq %rsi
  50         incq %rdi
  51         decl %ecx
  52         jnz 100b
  53 102:
  54         .section .fixup,"ax"
  55 103:    addl %ecx,%edx                  /* ecx is zerorest also */
  56         jmp copy_user_handle_tail
  57         .previous
  58
  59         .section __ex_table,"a"
  60         .align 8
  61         .quad 100b,103b
  62         .quad 101b,103b
  63         .previous
  64 #endif
  65         .endm
  66
  67 /* Standard copy_to_user with segment limit checking */
  68 ENTRY(copy_to_user)
  69         CFI_STARTPROC
  70         GET_THREAD_INFO(%rax)
  71         movq %rdi,%rcx
  72         addq %rdx,%rcx
  73         jc bad_to_user
  74         cmpq TI_addr_limit(%rax),%rcx
  75         jae bad_to_user
  76         ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,copy_user_generic_unrolled,copy_user_generic_string
  77         CFI_ENDPROC
  78
  79 /* Standard copy_from_user with segment limit checking */
  80 ENTRY(copy_from_user)
  81         CFI_STARTPROC
  82         GET_THREAD_INFO(%rax)
  83         movq %rsi,%rcx
  84         addq %rdx,%rcx
  85         jc bad_from_user
  86         cmpq TI_addr_limit(%rax),%rcx
  87         jae bad_from_user
  88         ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,copy_user_generic_unrolled,copy_user_generic_string
  89         CFI_ENDPROC
  90 ENDPROC(copy_from_user)
  91
  92 ENTRY(copy_user_generic)
  93         CFI_STARTPROC
  94         ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,copy_user_generic_unrolled,copy_user_generic_string
  95         CFI_ENDPROC
  96 ENDPROC(copy_user_generic)
  97
  98 ENTRY(__copy_from_user_inatomic)
  99         CFI_STARTPROC
 100         ALTERNATIVE_JUMP X86_FEATURE_REP_GOOD,copy_user_generic_unrolled,copy_user_generic_string
 101         CFI_ENDPROC
 102 ENDPROC(__copy_from_user_inatomic)
 103
 104         .section .fixup,"ax"
 105         /* must zero dest */
 106 ENTRY(bad_from_user)
 107 bad_from_user:
 108         CFI_STARTPROC
 109         movl %edx,%ecx
 110         xorl %eax,%eax
 111         rep
 112         stosb
 113 bad_to_user:
 114         movl %edx,%eax
 115         ret
 116         CFI_ENDPROC
 117 ENDPROC(bad_from_user)
 118         .previous
 119
 120 /*
 121  * copy_user_generic_unrolled - memory copy with exception handling.
 122  * This version is for CPUs like P4 that don't have efficient micro
 123  * code for rep movsq
 124  *
 125  * Input:
 126  * rdi destination
 127  * rsi source
 128  * rdx count
 129  *
 130  * Output:
 131  * eax uncopied bytes or 0 if successfull.
 132  */
 133 ENTRY(copy_user_generic_unrolled)
 134         CFI_STARTPROC
 135         cmpl $8,%edx
 136         jb 20f          /* less then 8 bytes, go to byte copy loop */
 137         ALIGN_DESTINATION
 138         movl %edx,%ecx
 139         andl $63,%edx
 140         shrl $6,%ecx
 141         jz 17f
 142 1:      movq (%rsi),%r8
 143 2:      movq 1*8(%rsi),%r9
 144 3:      movq 2*8(%rsi),%r10
 145 4:      movq 3*8(%rsi),%r11
 146 5:      movq %r8,(%rdi)
 147 6:      movq %r9,1*8(%rdi)
 148 7:      movq %r10,2*8(%rdi)
 149 8:      movq %r11,3*8(%rdi)
 150 9:      movq 4*8(%rsi),%r8
 151 10:     movq 5*8(%rsi),%r9
 152 11:     movq 6*8(%rsi),%r10
 153 12:     movq 7*8(%rsi),%r11
 154 13:     movq %r8,4*8(%rdi)
 155 14:     movq %r9,5*8(%rdi)
 156 15:     movq %r10,6*8(%rdi)
 157 16:     movq %r11,7*8(%rdi)
 158         leaq 64(%rsi),%rsi
 159         leaq 64(%rdi),%rdi
 160         decl %ecx
 161         jnz 1b
 162 17:     movl %edx,%ecx
 163         andl $7,%edx
 164         shrl $3,%ecx
 165         jz 20f
 166 18:     movq (%rsi),%r8
 167 19:     movq %r8,(%rdi)
 168         leaq 8(%rsi),%rsi
 169         leaq 8(%rdi),%rdi
 170         decl %ecx
 171         jnz 18b
 172 20:     andl %edx,%edx
 173         jz 23f
 174         movl %edx,%ecx
 175 21:     movb (%rsi),%al
 176 22:     movb %al,(%rdi)
 177         incq %rsi
 178         incq %rdi
 179         decl %ecx
 180         jnz 21b
 181 23:     xor %eax,%eax
 182         ret
 183
 184         .section .fixup,"ax"
 185 30:     shll $6,%ecx
 186         addl %ecx,%edx
 187         jmp 60f
 188 40:     lea (%rdx,%rcx,8),%rdx
 189         jmp 60f
 190 50:     movl %ecx,%edx
 191 60:     jmp copy_user_handle_tail /* ecx is zerorest also */
 192         .previous
 193
 194         .section __ex_table,"a"
 195         .align 8
 196         .quad 1b,30b
 197         .quad 2b,30b
 198         .quad 3b,30b
 199         .quad 4b,30b
 200         .quad 5b,30b
 201         .quad 6b,30b
 202         .quad 7b,30b
 203         .quad 8b,30b
 204         .quad 9b,30b
 205         .quad 10b,30b
 206         .quad 11b,30b
 207         .quad 12b,30b
 208         .quad 13b,30b
 209         .quad 14b,30b
 210         .quad 15b,30b
 211         .quad 16b,30b
 212         .quad 18b,40b
 213         .quad 19b,40b
 214         .quad 21b,50b
 215         .quad 22b,50b
 216         .previous
 217         CFI_ENDPROC
 218 ENDPROC(copy_user_generic_unrolled)
 219
 220 /* Some CPUs run faster using the string copy instructions.
 221  * This is also a lot simpler. Use them when possible.
 222  *
 223  * Only 4GB of copy is supported. This shouldn't be a problem
 224  * because the kernel normally only writes from/to page sized chunks
 225  * even if user space passed a longer buffer.
 226  * And more would be dangerous because both Intel and AMD have
 227  * errata with rep movsq > 4GB. If someone feels the need to fix
 228  * this please consider this.
 229  *
 230  * Input:
 231  * rdi destination
 232  * rsi source
 233  * rdx count
 234  *
 235  * Output:
 236  * eax uncopied bytes or 0 if successful.
 237  */
 238 ENTRY(copy_user_generic_string)
 239         CFI_STARTPROC
 240         andl %edx,%edx
 241         jz 4f
 242         cmpl $8,%edx
 243         jb 2f           /* less than 8 bytes, go to byte copy loop */
 244         ALIGN_DESTINATION
 245         movl %edx,%ecx
 246         shrl $3,%ecx
 247         andl $7,%edx
 248 1:      rep
 249         movsq
 250 2:      movl %edx,%ecx
 251 3:      rep
 252         movsb
 253 4:      xorl %eax,%eax
 254         ret
 255
 256         .section .fixup,"ax"
 257 11:     lea (%rdx,%rcx,8),%rcx
 258 12:     movl %ecx,%edx          /* ecx is zerorest also */
 259         jmp copy_user_handle_tail
 260         .previous
 261
 262         .section __ex_table,"a"
 263         .align 8
 264         .quad 1b,11b
 265         .quad 3b,12b
 266         .previous
 267         CFI_ENDPROC
 268 ENDPROC(copy_user_generic_string)