sysdeps/x86_64/dl-trampoline.S

   1 /* PLT trampolines.  x86-64 version.
   2    Copyright (C) 2004, 2005, 2007, 2009 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, write to the Free
  17    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  18    02111-1307 USA.  */
  19
  20 #include <config.h>
  21 #include <sysdep.h>
  22 #include <link-defines.h>
  23
  24         .text
  25         .globl _dl_runtime_resolve
  26         .type _dl_runtime_resolve, @function
  27         .align 16
  28         cfi_startproc
  29 _dl_runtime_resolve:
  30         subq $56,%rsp
  31         cfi_adjust_cfa_offset(72) # Incorporate PLT
  32         movq %rax,(%rsp)        # Preserve registers otherwise clobbered.
  33         movq %rcx, 8(%rsp)
  34         movq %rdx, 16(%rsp)
  35         movq %rsi, 24(%rsp)
  36         movq %rdi, 32(%rsp)
  37         movq %r8, 40(%rsp)
  38         movq %r9, 48(%rsp)
  39         movq 64(%rsp), %rsi     # Copy args pushed by PLT in register.
  40         movq 56(%rsp), %rdi     # %rdi: link_map, %rsi: reloc_index
  41         call _dl_fixup          # Call resolver.
  42         movq %rax, %r11         # Save return value
  43         movq 48(%rsp), %r9      # Get register content back.
  44         movq 40(%rsp), %r8
  45         movq 32(%rsp), %rdi
  46         movq 24(%rsp), %rsi
  47         movq 16(%rsp), %rdx
  48         movq 8(%rsp), %rcx
  49         movq (%rsp), %rax
  50         addq $72, %rsp          # Adjust stack(PLT did 2 pushes)
  51         cfi_adjust_cfa_offset(-72)
  52         jmp *%r11               # Jump to function address.
  53         cfi_endproc
  54         .size _dl_runtime_resolve, .-_dl_runtime_resolve
  55
  56
  57 #ifndef PROF
  58         .globl _dl_runtime_profile
  59         .type _dl_runtime_profile, @function
  60         .align 16
  61         cfi_startproc
  62
  63 _dl_runtime_profile:
  64         /* The La_x86_64_regs data structure pointed to by the
  65            fourth paramater must be 16-byte aligned.  This must
  66            be explicitly enforced.  We have the set up a dynamically
  67            sized stack frame.  %rbx points to the top half which
  68            has a fixed size and preserves the original stack pointer.  */
  69
  70         subq $32, %rsp          # Allocate the local storage.
  71         cfi_adjust_cfa_offset(48) # Incorporate PLT
  72         movq %rbx, (%rsp)
  73         cfi_rel_offset(%rbx, 0)
  74
  75         /* On the stack:
  76                 56(%rbx)        parameter #1
  77                 48(%rbx)        return address
  78
  79                 40(%rbx)        reloc index
  80                 32(%rbx)        link_map
  81
  82                 24(%rbx)        La_x86_64_regs pointer
  83                 16(%rbx)        framesize
  84                  8(%rbx)        rax
  85                   (%rbx)        rbx
  86         */
  87
  88         movq %rax, 8(%rsp)
  89         movq %rsp, %rbx
  90         cfi_def_cfa_register(%rbx)
  91
  92         /* Actively align the La_x86_64_regs structure.  */
  93         andq $0xfffffffffffffff0, %rsp
  94 # ifdef HAVE_AVX_SUPPORT
  95         /* sizeof(La_x86_64_regs).  Need extra space for 8 SSE registers
  96            to detect if any xmm0-xmm7 registers are changed by audit
  97            module.  */
  98         subq $(LR_SIZE + XMM_SIZE*8), %rsp
  99 # else
 100         subq $LR_SIZE, %rsp             # sizeof(La_x86_64_regs)
 101 # endif
 102         movq %rsp, 24(%rbx)
 103
 104         /* Fill the La_x86_64_regs structure.  */
 105         movq %rdx, LR_RDX_OFFSET(%rsp)
 106         movq %r8,  LR_R8_OFFSET(%rsp)
 107         movq %r9,  LR_R9_OFFSET(%rsp)
 108         movq %rcx, LR_RCX_OFFSET(%rsp)
 109         movq %rsi, LR_RSI_OFFSET(%rsp)
 110         movq %rdi, LR_RDI_OFFSET(%rsp)
 111         movq %rbp, LR_RBP_OFFSET(%rsp)
 112
 113         leaq 48(%rbx), %rax
 114         movq %rax, LR_RSP_OFFSET(%rsp)
 115
 116         /* We always store the XMM registers even if AVX is available.
 117            This is to provide backward binary compatility for existing
 118            audit modules.  */
 119         movaps %xmm0,              (LR_XMM_OFFSET)(%rsp)
 120         movaps %xmm1, (LR_XMM_OFFSET +   XMM_SIZE)(%rsp)
 121         movaps %xmm2, (LR_XMM_OFFSET + XMM_SIZE*2)(%rsp)
 122         movaps %xmm3, (LR_XMM_OFFSET + XMM_SIZE*3)(%rsp)
 123         movaps %xmm4, (LR_XMM_OFFSET + XMM_SIZE*4)(%rsp)
 124         movaps %xmm5, (LR_XMM_OFFSET + XMM_SIZE*5)(%rsp)
 125         movaps %xmm6, (LR_XMM_OFFSET + XMM_SIZE*6)(%rsp)
 126         movaps %xmm7, (LR_XMM_OFFSET + XMM_SIZE*7)(%rsp)
 127
 128 # ifdef HAVE_AVX_SUPPORT
 129         .data
 130 L(have_avx):
 131         .zero 4
 132         .size L(have_avx), 4
 133         .previous
 134
 135         cmpl    $0, L(have_avx)(%rip)
 136         jne     1f
 137         movq    %rbx, %r11              # Save rbx
 138         movl    $1, %eax
 139         cpuid
 140         movq    %r11,%rbx               # Restore rbx
 141         movl    $1, %eax
 142         testl   $(1 << 28), %ecx
 143         jne     2f
 144         negl    %eax
 145 2:      movl    %eax, L(have_avx)(%rip)
 146         cmpl    $0, %eax
 147
 148 1:      js      L(no_avx1)
 149
 150         /* This is to support AVX audit modules.  */
 151         vmovdqu %ymm0,                (LR_VECTOR_OFFSET)(%rsp)
 152         vmovdqu %ymm1, (LR_VECTOR_OFFSET +   VECTOR_SIZE)(%rsp)
 153         vmovdqu %ymm2, (LR_VECTOR_OFFSET + VECTOR_SIZE*2)(%rsp)
 154         vmovdqu %ymm3, (LR_VECTOR_OFFSET + VECTOR_SIZE*3)(%rsp)
 155         vmovdqu %ymm4, (LR_VECTOR_OFFSET + VECTOR_SIZE*4)(%rsp)
 156         vmovdqu %ymm5, (LR_VECTOR_OFFSET + VECTOR_SIZE*5)(%rsp)
 157         vmovdqu %ymm6, (LR_VECTOR_OFFSET + VECTOR_SIZE*6)(%rsp)
 158         vmovdqu %ymm7, (LR_VECTOR_OFFSET + VECTOR_SIZE*7)(%rsp)
 159
 160         /* Save xmm0-xmm7 registers to detect if any of them are
 161            changed by audit module.  */
 162         vmovdqa %xmm0,              (LR_SIZE)(%rsp)
 163         vmovdqa %xmm1, (LR_SIZE +   XMM_SIZE)(%rsp)
 164         vmovdqa %xmm2, (LR_SIZE + XMM_SIZE*2)(%rsp)
 165         vmovdqa %xmm3, (LR_SIZE + XMM_SIZE*3)(%rsp)
 166         vmovdqa %xmm4, (LR_SIZE + XMM_SIZE*4)(%rsp)
 167         vmovdqa %xmm5, (LR_SIZE + XMM_SIZE*5)(%rsp)
 168         vmovdqa %xmm6, (LR_SIZE + XMM_SIZE*6)(%rsp)
 169         vmovdqa %xmm7, (LR_SIZE + XMM_SIZE*7)(%rsp)
 170
 171 L(no_avx1):
 172 # endif
 173
 174         movq %rsp, %rcx         # La_x86_64_regs pointer to %rcx.
 175         movq 48(%rbx), %rdx     # Load return address if needed.
 176         movq 40(%rbx), %rsi     # Copy args pushed by PLT in register.
 177         movq 32(%rbx), %rdi     # %rdi: link_map, %rsi: reloc_index
 178         leaq 16(%rbx), %r8
 179         call _dl_profile_fixup  # Call resolver.
 180
 181         movq %rax, %r11         # Save return value.
 182
 183         movq 8(%rbx), %rax      # Get back register content.
 184         movq LR_RDX_OFFSET(%rsp), %rdx
 185         movq  LR_R8_OFFSET(%rsp), %r8
 186         movq  LR_R9_OFFSET(%rsp), %r9
 187
 188         movaps              (LR_XMM_OFFSET)(%rsp), %xmm0
 189         movaps   (LR_XMM_OFFSET + XMM_SIZE)(%rsp), %xmm1
 190         movaps (LR_XMM_OFFSET + XMM_SIZE*2)(%rsp), %xmm2
 191         movaps (LR_XMM_OFFSET + XMM_SIZE*3)(%rsp), %xmm3
 192         movaps (LR_XMM_OFFSET + XMM_SIZE*4)(%rsp), %xmm4
 193         movaps (LR_XMM_OFFSET + XMM_SIZE*5)(%rsp), %xmm5
 194         movaps (LR_XMM_OFFSET + XMM_SIZE*6)(%rsp), %xmm6
 195         movaps (LR_XMM_OFFSET + XMM_SIZE*7)(%rsp), %xmm7
 196
 197 # ifdef HAVE_AVX_SUPPORT
 198         cmpl    $0, L(have_avx)(%rip)
 199         js      L(no_avx2)
 200
 201         /* Check if any xmm0-xmm7 registers are changed by audit
 202            module.  */
 203         vpcmpeqq (LR_SIZE)(%rsp), %xmm0, %xmm8
 204         vpmovmskb %xmm8, %esi
 205         cmpl $0xffff, %esi
 206         je 1f
 207         vmovdqu                 (LR_VECTOR_OFFSET)(%rsp), %ymm0
 208
 209 1:      vpcmpeqq (LR_SIZE + XMM_SIZE)(%rsp), %xmm1, %xmm8
 210         vpmovmskb %xmm8, %esi
 211         cmpl $0xffff, %esi
 212         je 1f
 213         vmovdqu   (LR_VECTOR_OFFSET + VECTOR_SIZE)(%rsp), %ymm1
 214
 215 1:      vpcmpeqq (LR_SIZE + XMM_SIZE*2)(%rsp), %xmm2, %xmm8
 216         vpmovmskb %xmm8, %esi
 217         cmpl $0xffff, %esi
 218         je 1f
 219         vmovdqu (LR_VECTOR_OFFSET + VECTOR_SIZE*2)(%rsp), %ymm2
 220
 221 1:      vpcmpeqq (LR_SIZE + XMM_SIZE*3)(%rsp), %xmm3, %xmm8
 222         vpmovmskb %xmm8, %esi
 223         cmpl $0xffff, %esi
 224         je 1f
 225         vmovdqu (LR_VECTOR_OFFSET + VECTOR_SIZE*3)(%rsp), %ymm3
 226
 227 1:      vpcmpeqq (LR_SIZE + XMM_SIZE*4)(%rsp), %xmm4, %xmm8
 228         vpmovmskb %xmm8, %esi
 229         cmpl $0xffff, %esi
 230         je 1f
 231         vmovdqu (LR_VECTOR_OFFSET + VECTOR_SIZE*4)(%rsp), %ymm4
 232
 233 1:      vpcmpeqq (LR_SIZE + XMM_SIZE*5)(%rsp), %xmm5, %xmm8
 234         vpmovmskb %xmm8, %esi
 235         cmpl $0xffff, %esi
 236         je 1f
 237         vmovdqu (LR_VECTOR_OFFSET + VECTOR_SIZE*5)(%rsp), %ymm5
 238
 239 1:      vpcmpeqq (LR_SIZE + XMM_SIZE*6)(%rsp), %xmm6, %xmm8
 240         vpmovmskb %xmm8, %esi
 241         cmpl $0xffff, %esi
 242         je 1f
 243         vmovdqu (LR_VECTOR_OFFSET + VECTOR_SIZE*6)(%rsp), %ymm6
 244
 245 1:      vpcmpeqq (LR_SIZE + XMM_SIZE*7)(%rsp), %xmm7, %xmm8
 246         vpmovmskb %xmm8, %esi
 247         cmpl $0xffff, %esi
 248         je 1f
 249         vmovdqu (LR_VECTOR_OFFSET + VECTOR_SIZE*7)(%rsp), %ymm7
 250
 251 L(no_avx2):
 252 1:
 253 # endif
 254         movq 16(%rbx), %r10     # Anything in framesize?
 255         testq %r10, %r10
 256         jns 3f
 257
 258         /* There's nothing in the frame size, so there
 259            will be no call to the _dl_call_pltexit. */
 260
 261         /* Get back registers content.  */
 262         movq LR_RCX_OFFSET(%rsp), %rcx
 263         movq LR_RSI_OFFSET(%rsp), %rsi
 264         movq LR_RDI_OFFSET(%rsp), %rdi
 265
 266         movq %rbx, %rsp
 267         movq (%rsp), %rbx
 268         cfi_restore(rbx)
 269         cfi_def_cfa_register(%rsp)
 270
 271         addq $48, %rsp          # Adjust the stack to the return value
 272                                 # (eats the reloc index and link_map)
 273         cfi_adjust_cfa_offset(-48)
 274         jmp *%r11               # Jump to function address.
 275
 276 3:
 277         cfi_adjust_cfa_offset(48)
 278         cfi_rel_offset(%rbx, 0)
 279         cfi_def_cfa_register(%rbx)
 280
 281         /* At this point we need to prepare new stack for the function
 282            which has to be called.  We copy the original stack to a
 283            temporary buffer of the size specified by the 'framesize'
 284            returned from _dl_profile_fixup */
 285
 286         leaq LR_RSP_OFFSET(%rbx), %rsi  # stack
 287         addq $8, %r10
 288         andq $0xfffffffffffffff0, %r10
 289         movq %r10, %rcx
 290         subq %r10, %rsp
 291         movq %rsp, %rdi
 292         shrq $3, %rcx
 293         rep
 294         movsq
 295
 296         movq 24(%rdi), %rcx     # Get back register content.
 297         movq 32(%rdi), %rsi
 298         movq 40(%rdi), %rdi
 299
 300         call *%r11
 301
 302         mov 24(%rbx), %rsp      # Drop the copied stack content
 303
 304         /* Now we have to prepare the La_x86_64_retval structure for the
 305            _dl_call_pltexit.  The La_x86_64_regs is being pointed by rsp now,
 306            so we just need to allocate the sizeof(La_x86_64_retval) space on
 307            the stack, since the alignment has already been taken care of. */
 308 # ifdef HAVE_AVX_SUPPORT
 309         /* sizeof(La_x86_64_retval).  Need extra space for 2 SSE
 310            registers to detect if xmm0/xmm1 registers are changed
 311            by audit module.  */
 312         subq $(LRV_SIZE + XMM_SIZE*2), %rsp
 313 # else
 314         subq $LRV_SIZE, %rsp    # sizeof(La_x86_64_retval)
 315 # endif
 316         movq %rsp, %rcx         # La_x86_64_retval argument to %rcx.
 317
 318         /* Fill in the La_x86_64_retval structure.  */
 319         movq %rax, LRV_RAX_OFFSET(%rcx)
 320         movq %rdx, LRV_RDX_OFFSET(%rcx)
 321
 322         movaps %xmm0, LRV_XMM0_OFFSET(%rcx)
 323         movaps %xmm1, LRV_XMM1_OFFSET(%rcx)
 324
 325 # ifdef HAVE_AVX_SUPPORT
 326         cmpl    $0, L(have_avx)(%rip)
 327         js      L(no_avx3)
 328
 329         /* This is to support AVX audit modules.  */
 330         vmovdqu %ymm0, LRV_VECTOR0_OFFSET(%rcx)
 331         vmovdqu %ymm1, LRV_VECTOR1_OFFSET(%rcx)
 332
 333         /* Save xmm0/xmm1 registers to detect if they are changed
 334            by audit module.  */
 335         vmovdqa %xmm0,            (LRV_SIZE)(%rcx)
 336         vmovdqa %xmm1, (LRV_SIZE + XMM_SIZE)(%rcx)
 337
 338 L(no_avx3):
 339 # endif
 340
 341         fstpt LRV_ST0_OFFSET(%rcx)
 342         fstpt LRV_ST1_OFFSET(%rcx)
 343
 344         movq 24(%rbx), %rdx     # La_x86_64_regs argument to %rdx.
 345         movq 40(%rbx), %rsi     # Copy args pushed by PLT in register.
 346         movq 32(%rbx), %rdi     # %rdi: link_map, %rsi: reloc_index
 347         call _dl_call_pltexit
 348
 349         /* Restore return registers.  */
 350         movq LRV_RAX_OFFSET(%rsp), %rax
 351         movq LRV_RDX_OFFSET(%rsp), %rdx
 352
 353         movaps LRV_XMM0_OFFSET(%rsp), %xmm0
 354         movaps LRV_XMM1_OFFSET(%rsp), %xmm1
 355
 356 # ifdef HAVE_AVX_SUPPORT
 357         cmpl    $0, L(have_avx)(%rip)
 358         js      L(no_avx4)
 359
 360         /* Check if xmm0/xmm1 registers are changed by audit module.  */
 361         vpcmpeqq (LRV_SIZE)(%rsp), %xmm0, %xmm2
 362         vpmovmskb %xmm2, %esi
 363         cmpl $0xffff, %esi
 364         je 1f
 365         vmovdqu LRV_VECTOR0_OFFSET(%rsp), %ymm0
 366
 367 1:      vpcmpeqq (LRV_SIZE + XMM_SIZE)(%rsp), %xmm1, %xmm2
 368         vpmovmskb %xmm2, %esi
 369         cmpl $0xffff, %esi
 370         je 1f
 371         vmovdqu LRV_VECTOR1_OFFSET(%rsp), %ymm1
 372
 373 L(no_avx4):
 374 1:
 375 # endif
 376
 377         fldt LRV_ST1_OFFSET(%rsp)
 378         fldt LRV_ST0_OFFSET(%rsp)
 379
 380         movq %rbx, %rsp
 381         movq (%rsp), %rbx
 382         cfi_restore(rbx)
 383         cfi_def_cfa_register(%rsp)
 384
 385         addq $48, %rsp          # Adjust the stack to the return value
 386                                 # (eats the reloc index and link_map)
 387         cfi_adjust_cfa_offset(-48)
 388         retq
 389
 390         cfi_endproc
 391         .size _dl_runtime_profile, .-_dl_runtime_profile
 392 #endif
 393
 394
 395 #ifdef SHARED
 396         .globl _dl_x86_64_save_sse
 397         .type _dl_x86_64_save_sse, @function
 398         .align 16
 399         cfi_startproc
 400 _dl_x86_64_save_sse:
 401 # ifdef HAVE_AVX_SUPPORT
 402         cmpl    $0, L(have_avx)(%rip)
 403         jne     1f
 404         movq    %rbx, %r11              # Save rbx
 405         movl    $1, %eax
 406         cpuid
 407         movq    %r11,%rbx               # Restore rbx
 408         movl    $1, %eax
 409         testl   $(1 << 28), %ecx
 410         jne     2f
 411         negl    %eax
 412 2:      movl    %eax, L(have_avx)(%rip)
 413         cmpl    $0, %eax
 414
 415 1:      js      L(no_avx5)
 416
 417 #  define YMM_SIZE 32
 418         vmovdqa %ymm0, %fs:RTLD_SAVESPACE_SSE+0*YMM_SIZE
 419         vmovdqa %ymm1, %fs:RTLD_SAVESPACE_SSE+1*YMM_SIZE
 420         vmovdqa %ymm2, %fs:RTLD_SAVESPACE_SSE+2*YMM_SIZE
 421         vmovdqa %ymm3, %fs:RTLD_SAVESPACE_SSE+3*YMM_SIZE
 422         vmovdqa %ymm4, %fs:RTLD_SAVESPACE_SSE+4*YMM_SIZE
 423         vmovdqa %ymm5, %fs:RTLD_SAVESPACE_SSE+5*YMM_SIZE
 424         vmovdqa %ymm6, %fs:RTLD_SAVESPACE_SSE+6*YMM_SIZE
 425         vmovdqa %ymm7, %fs:RTLD_SAVESPACE_SSE+7*YMM_SIZE
 426         ret
 427 L(no_avx5):
 428 # endif
 429 # define YMM_SIZE 16
 430         movdqa  %xmm0, %fs:RTLD_SAVESPACE_SSE+0*XMM_SIZE
 431         movdqa  %xmm1, %fs:RTLD_SAVESPACE_SSE+1*XMM_SIZE
 432         movdqa  %xmm2, %fs:RTLD_SAVESPACE_SSE+2*XMM_SIZE
 433         movdqa  %xmm3, %fs:RTLD_SAVESPACE_SSE+3*XMM_SIZE
 434         movdqa  %xmm4, %fs:RTLD_SAVESPACE_SSE+4*XMM_SIZE
 435         movdqa  %xmm5, %fs:RTLD_SAVESPACE_SSE+5*XMM_SIZE
 436         movdqa  %xmm6, %fs:RTLD_SAVESPACE_SSE+6*XMM_SIZE
 437         movdqa  %xmm7, %fs:RTLD_SAVESPACE_SSE+7*XMM_SIZE
 438         ret
 439         cfi_endproc
 440         .size _dl_x86_64_save_sse, .-_dl_x86_64_save_sse
 441
 442
 443         .globl _dl_x86_64_restore_sse
 444         .type _dl_x86_64_restore_sse, @function
 445         .align 16
 446         cfi_startproc
 447 _dl_x86_64_restore_sse:
 448 # ifdef HAVE_AVX_SUPPORT
 449         cmpl    $0, L(have_avx)(%rip)
 450         js      L(no_avx6)
 451
 452         vmovdqa %fs:RTLD_SAVESPACE_SSE+0*YMM_SIZE, %ymm0
 453         vmovdqa %fs:RTLD_SAVESPACE_SSE+1*YMM_SIZE, %ymm1
 454         vmovdqa %fs:RTLD_SAVESPACE_SSE+2*YMM_SIZE, %ymm2
 455         vmovdqa %fs:RTLD_SAVESPACE_SSE+3*YMM_SIZE, %ymm3
 456         vmovdqa %fs:RTLD_SAVESPACE_SSE+4*YMM_SIZE, %ymm4
 457         vmovdqa %fs:RTLD_SAVESPACE_SSE+5*YMM_SIZE, %ymm5
 458         vmovdqa %fs:RTLD_SAVESPACE_SSE+6*YMM_SIZE, %ymm6
 459         vmovdqa %fs:RTLD_SAVESPACE_SSE+7*YMM_SIZE, %ymm7
 460         ret
 461 L(no_avx6):
 462 # endif
 463         movdqa  %fs:RTLD_SAVESPACE_SSE+0*XMM_SIZE, %xmm0
 464         movdqa  %fs:RTLD_SAVESPACE_SSE+1*XMM_SIZE, %xmm1
 465         movdqa  %fs:RTLD_SAVESPACE_SSE+2*XMM_SIZE, %xmm2
 466         movdqa  %fs:RTLD_SAVESPACE_SSE+3*XMM_SIZE, %xmm3
 467         movdqa  %fs:RTLD_SAVESPACE_SSE+4*XMM_SIZE, %xmm4
 468         movdqa  %fs:RTLD_SAVESPACE_SSE+5*XMM_SIZE, %xmm5
 469         movdqa  %fs:RTLD_SAVESPACE_SSE+6*XMM_SIZE, %xmm6
 470         movdqa  %fs:RTLD_SAVESPACE_SSE+7*XMM_SIZE, %xmm7
 471         ret
 472         cfi_endproc
 473         .size _dl_x86_64_restore_sse, .-_dl_x86_64_restore_sse
 474 #endif