rdrand_asm.S

   1 /*
   2  * Copyright (c) 2011, Intel Corporation
   3  * Authors: Fenghua Yu <fenghua.yu@intel.com>,
   4  *          H. Peter Anvin <hpa@linux.intel.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms and conditions of the GNU General Public License,
   8  * version 2, as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along with
  16  * this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  */
  20
  21 #define ENTRY(x)          \
  22         .balign 64      ; \
  23         .globl  x       ; \
  24 x:
  25
  26 #define ENDPROC(x)                \
  27         .size   x, .-x          ; \
  28         .type   x, @function
  29
  30 #define RDRAND_RETRY_LIMIT      10
  31
  32 #if defined(__x86_64__)
  33
  34 ENTRY(x86_rdrand_nlong)
  35 1:
  36         mov     $RDRAND_RETRY_LIMIT, %eax
  37 2:
  38         .byte   0x48,0x0f,0xc7,0xf2     /* rdrand %rdx */
  39         jnc     3f
  40         mov     %rdx, (%rdi)
  41         add     $8, %rdi
  42         sub     $1, %esi
  43         jnz     1b
  44         ret
  45 3:
  46         sub     $1, %eax
  47         rep;nop
  48         jnz     2b
  49         ret
  50 ENDPROC(x86_rdrand_nlong)
  51
  52 #define SETPTR(var,ptr) leaq var(%rip),ptr
  53 #define PTR0    %rdi
  54 #define PTR1    %rsi
  55 #define PTR2    %rcx
  56 #define NPTR2   1       /* %rcx = %r1, only 0-7 valid here */
  57
  58 #elif defined(__i386__)
  59
  60 ENTRY(x86_rdrand_nlong)
  61         push    %ebp
  62         mov     %esp, %ebp
  63         push    %edi
  64         movl    8(%ebp), %ecx
  65         movl    12(%ebp), %edx
  66 1:
  67         mov     $RDRAND_RETRY_LIMIT, %eax
  68 2:
  69         .byte   0x0f,0xc7,0xf7          /* rdrand %edi */
  70         jnc     3f
  71         mov     %edi, (%ecx)
  72         add     $4, %ecx
  73         sub     $1, %edx
  74         jnz     2b
  75         pop     %edi
  76         pop     %ebp
  77         ret
  78 3:
  79         sub     $1, %eax
  80         rep;nop
  81         jnz     2b
  82         pop %edi
  83         pop     %ebp
  84         ret
  85 ENDPROC(x86_rdrand_nlong)
  86
  87 #define SETPTR(var,ptr) movl $(var),ptr
  88 #define PTR0    %eax
  89 #define PTR1    %edx
  90 #define PTR2    %ecx
  91 #define NPTR2   1       /* %rcx = %r1 */
  92
  93 #endif
  94
  95 #if defined(__i386__) || defined(__x86_64__)
  96
  97 ENTRY(x86_aes_mangle)
  98 #if defined(__i386__)
  99         push    %ebp
 100         mov     %esp, %ebp
 101         movl    8(%ebp), %eax
 102         movl    12(%ebp), %edx
 103 #endif
 104
 105         SETPTR(aes_round_keys, PTR2)
 106
 107         movdqa  (0*16)(PTR0), %xmm0
 108         movdqa  (1*16)(PTR0), %xmm1
 109         movdqa  (2*16)(PTR0), %xmm2
 110         movdqa  (3*16)(PTR0), %xmm3
 111         movdqa  (4*16)(PTR0), %xmm4
 112         movdqa  (5*16)(PTR0), %xmm5
 113         movdqa  (6*16)(PTR0), %xmm6
 114         movdqa  (7*16)(PTR0), %xmm7
 115
 116         pxor    (0*16)(PTR1), %xmm0
 117         pxor    (1*16)(PTR1), %xmm1
 118         pxor    (2*16)(PTR1), %xmm2
 119         pxor    (3*16)(PTR1), %xmm3
 120         pxor    (4*16)(PTR1), %xmm4
 121         pxor    (5*16)(PTR1), %xmm5
 122         pxor    (6*16)(PTR1), %xmm6
 123         pxor    (7*16)(PTR1), %xmm7
 124
 125         .rept 10
 126         .byte   0x66,0x0f,0x38,0xdc,0x00+NPTR2  /* aesenc (PTR2), %xmm0 */
 127         .byte   0x66,0x0f,0x38,0xdc,0x08+NPTR2  /* aesenc (PTR2), %xmm1 */
 128         .byte   0x66,0x0f,0x38,0xdc,0x10+NPTR2  /* aesenc (PTR2), %xmm2 */
 129         .byte   0x66,0x0f,0x38,0xdc,0x18+NPTR2  /* aesenc (PTR2), %xmm3 */
 130         .byte   0x66,0x0f,0x38,0xdc,0x20+NPTR2  /* aesenc (PTR2), %xmm4 */
 131         .byte   0x66,0x0f,0x38,0xdc,0x28+NPTR2  /* aesenc (PTR2), %xmm5 */
 132         .byte   0x66,0x0f,0x38,0xdc,0x30+NPTR2  /* aesenc (PTR2), %xmm6 */
 133         .byte   0x66,0x0f,0x38,0xdc,0x38+NPTR2  /* aesenc (PTR2), %xmm7 */
 134         add     $16, PTR2
 135         .endr
 136
 137         .byte   0x66,0x0f,0x38,0xdd,0x00+NPTR2  /* aesenclast (PTR2), %xmm0 */
 138         .byte   0x66,0x0f,0x38,0xdd,0x08+NPTR2  /* aesenclast (PTR2), %xmm1 */
 139         .byte   0x66,0x0f,0x38,0xdd,0x10+NPTR2  /* aesenclast (PTR2), %xmm2 */
 140         .byte   0x66,0x0f,0x38,0xdd,0x18+NPTR2  /* aesenclast (PTR2), %xmm3 */
 141         .byte   0x66,0x0f,0x38,0xdd,0x20+NPTR2  /* aesenclast (PTR2), %xmm4 */
 142         .byte   0x66,0x0f,0x38,0xdd,0x28+NPTR2  /* aesenclast (PTR2), %xmm5 */
 143         .byte   0x66,0x0f,0x38,0xdd,0x30+NPTR2  /* aesenclast (PTR2), %xmm6 */
 144         .byte   0x66,0x0f,0x38,0xdd,0x38+NPTR2  /* aesenclast (PTR2), %xmm7 */
 145
 146         movdqa  %xmm0, (0*16)(PTR0)
 147         movdqa  %xmm1, (1*16)(PTR0)
 148         movdqa  %xmm2, (2*16)(PTR0)
 149         movdqa  %xmm3, (3*16)(PTR0)
 150         movdqa  %xmm4, (4*16)(PTR0)
 151         movdqa  %xmm5, (5*16)(PTR0)
 152         movdqa  %xmm6, (6*16)(PTR0)
 153         movdqa  %xmm7, (7*16)(PTR0)
 154
 155         movdqa  %xmm0, (0*16)(PTR1)
 156         movdqa  %xmm1, (1*16)(PTR1)
 157         movdqa  %xmm2, (2*16)(PTR1)
 158         movdqa  %xmm3, (3*16)(PTR1)
 159         movdqa  %xmm4, (4*16)(PTR1)
 160         movdqa  %xmm5, (5*16)(PTR1)
 161         movdqa  %xmm6, (6*16)(PTR1)
 162         movdqa  %xmm7, (7*16)(PTR1)
 163
 164 #if defined(__i386__)
 165         pop     %ebp
 166 #endif
 167         ret
 168 ENDPROC(x86_aes_mangle)
 169         /*
 170          * AES round keys for an arbitrary key:
 171          * 00102030405060708090A0B0C0D0E0F0
 172          */
 173         .section ".rodata","a"
 174         .balign 16
 175 aes_round_keys:
 176         .long   0x30201000, 0x70605040, 0xB0A09080, 0xF0E0D0C0
 177         .long   0x8AACF171, 0xFACCA131, 0x4A6C31B1, 0xBA8CE171
 178         .long   0x2958958B, 0xD39434BA, 0x99F8050B, 0x2374E47A
 179         .long   0xF37E07E6, 0x20EA335C, 0xB9123657, 0x9A66D22D
 180         .long   0x2BC6345B, 0x0B2C0707, 0xB23E3150, 0x2858E37D
 181         .long   0xD4F25E5A, 0xDFDE595D, 0x6DE0680D, 0x45B88B70
 182         .long   0x859C3247, 0x5A426B1A, 0x37A20317, 0x721A8867
 183         .long   0x00DC90C3, 0x5A9EFBD9, 0x6D3CF8CE, 0x1F2670A9
 184         .long   0xD31C6712, 0x89829CCB, 0xE4BE6405, 0xFB9814AC
 185         .long   0x421321F3, 0xCB91BD38, 0x2F2FD93D, 0xD4B7CD91
 186         .long   0xC35B8878, 0x08CA3540, 0x27E5EC7D, 0xF35221EC
 187         .size   aes_round_keys, .-aes_round_keys
 188
 189         .bss
 190         .balign 16
 191 aes_fwd_state:
 192         .space  16
 193         .size   aes_fwd_state, .-aes_fwd_state
 194
 195 #endif /* i386 or x86_64 */
 196
 197 /*
 198  * This is necessary to keep the whole executable
 199  * from needing a writable stack.
 200  */
 201                 .section        .note.GNU-stack,"",%progbits