1/* PLT trampolines. x86-64 version.
2 Copyright (C) 2009-2016 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, see
17 <http://www.gnu.org/licenses/>. */
18
19#undef REGISTER_SAVE_AREA_RAW
20#ifdef __ILP32__
21/* X32 saves RCX, RDX, RSI, RDI, R8 and R9 plus RAX as well as VEC0 to
22 VEC7. */
23# define REGISTER_SAVE_AREA_RAW (8 * 7 + VEC_SIZE * 8)
24#else
25/* X86-64 saves RCX, RDX, RSI, RDI, R8 and R9 plus RAX as well as
26 BND0, BND1, BND2, BND3 and VEC0 to VEC7. */
27# define REGISTER_SAVE_AREA_RAW (8 * 7 + 16 * 4 + VEC_SIZE * 8)
28#endif
29
30#undef REGISTER_SAVE_AREA
31#undef LOCAL_STORAGE_AREA
32#undef BASE
33#if DL_RUNIME_RESOLVE_REALIGN_STACK
34# define REGISTER_SAVE_AREA (REGISTER_SAVE_AREA_RAW + 8)
35/* Local stack area before jumping to function address: RBX. */
36# define LOCAL_STORAGE_AREA 8
37# define BASE rbx
38# if (REGISTER_SAVE_AREA % VEC_SIZE) != 0
39# error REGISTER_SAVE_AREA must be multples of VEC_SIZE
40# endif
41#else
42# define REGISTER_SAVE_AREA REGISTER_SAVE_AREA_RAW
43/* Local stack area before jumping to function address: All saved
44 registers. */
45# define LOCAL_STORAGE_AREA REGISTER_SAVE_AREA
46# define BASE rsp
47# if (REGISTER_SAVE_AREA % 16) != 8
48# error REGISTER_SAVE_AREA must be odd multples of 8
49# endif
50#endif
51
52 .text
53 .globl _dl_runtime_resolve
54 .hidden _dl_runtime_resolve
55 .type _dl_runtime_resolve, @function
56 .align 16
57 cfi_startproc
58_dl_runtime_resolve:
59 cfi_adjust_cfa_offset(16) # Incorporate PLT
60#if DL_RUNIME_RESOLVE_REALIGN_STACK
61# if LOCAL_STORAGE_AREA != 8
62# error LOCAL_STORAGE_AREA must be 8
63# endif
64 pushq %rbx # push subtracts stack by 8.
65 cfi_adjust_cfa_offset(8)
66 cfi_rel_offset(%rbx, 0)
67 mov %RSP_LP, %RBX_LP
68 cfi_def_cfa_register(%rbx)
69 and $-VEC_SIZE, %RSP_LP
70#endif
71 sub $REGISTER_SAVE_AREA, %RSP_LP
72 cfi_adjust_cfa_offset(REGISTER_SAVE_AREA)
73 # Preserve registers otherwise clobbered.
74 movq %rax, REGISTER_SAVE_RAX(%rsp)
75 movq %rcx, REGISTER_SAVE_RCX(%rsp)
76 movq %rdx, REGISTER_SAVE_RDX(%rsp)
77 movq %rsi, REGISTER_SAVE_RSI(%rsp)
78 movq %rdi, REGISTER_SAVE_RDI(%rsp)
79 movq %r8, REGISTER_SAVE_R8(%rsp)
80 movq %r9, REGISTER_SAVE_R9(%rsp)
81 VMOV %VEC(0), (REGISTER_SAVE_VEC_OFF)(%rsp)
82 VMOV %VEC(1), (REGISTER_SAVE_VEC_OFF + VEC_SIZE)(%rsp)
83 VMOV %VEC(2), (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 2)(%rsp)
84 VMOV %VEC(3), (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 3)(%rsp)
85 VMOV %VEC(4), (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 4)(%rsp)
86 VMOV %VEC(5), (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 5)(%rsp)
87 VMOV %VEC(6), (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 6)(%rsp)
88 VMOV %VEC(7), (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 7)(%rsp)
89#ifndef __ILP32__
90 # We also have to preserve bound registers. These are nops if
91 # Intel MPX isn't available or disabled.
92# ifdef HAVE_MPX_SUPPORT
93 bndmov %bnd0, REGISTER_SAVE_BND0(%rsp)
94 bndmov %bnd1, REGISTER_SAVE_BND1(%rsp)
95 bndmov %bnd2, REGISTER_SAVE_BND2(%rsp)
96 bndmov %bnd3, REGISTER_SAVE_BND3(%rsp)
97# else
98# if REGISTER_SAVE_BND0 == 0
99 .byte 0x66,0x0f,0x1b,0x04,0x24
100# else
101 .byte 0x66,0x0f,0x1b,0x44,0x24,REGISTER_SAVE_BND0
102# endif
103 .byte 0x66,0x0f,0x1b,0x4c,0x24,REGISTER_SAVE_BND1
104 .byte 0x66,0x0f,0x1b,0x54,0x24,REGISTER_SAVE_BND2
105 .byte 0x66,0x0f,0x1b,0x5c,0x24,REGISTER_SAVE_BND3
106# endif
107#endif
108 # Copy args pushed by PLT in register.
109 # %rdi: link_map, %rsi: reloc_index
110 mov (LOCAL_STORAGE_AREA + 8)(%BASE), %RSI_LP
111 mov LOCAL_STORAGE_AREA(%BASE), %RDI_LP
112 call _dl_fixup # Call resolver.
113 mov %RAX_LP, %R11_LP # Save return value
114#ifndef __ILP32__
115 # Restore bound registers. These are nops if Intel MPX isn't
116 # avaiable or disabled.
117# ifdef HAVE_MPX_SUPPORT
118 bndmov REGISTER_SAVE_BND3(%rsp), %bnd3
119 bndmov REGISTER_SAVE_BND2(%rsp), %bnd2
120 bndmov REGISTER_SAVE_BND1(%rsp), %bnd1
121 bndmov REGISTER_SAVE_BND0(%rsp), %bnd0
122# else
123 .byte 0x66,0x0f,0x1a,0x5c,0x24,REGISTER_SAVE_BND3
124 .byte 0x66,0x0f,0x1a,0x54,0x24,REGISTER_SAVE_BND2
125 .byte 0x66,0x0f,0x1a,0x4c,0x24,REGISTER_SAVE_BND1
126# if REGISTER_SAVE_BND0 == 0
127 .byte 0x66,0x0f,0x1a,0x04,0x24
128# else
129 .byte 0x66,0x0f,0x1a,0x44,0x24,REGISTER_SAVE_BND0
130# endif
131# endif
132#endif
133 # Get register content back.
134 movq REGISTER_SAVE_R9(%rsp), %r9
135 movq REGISTER_SAVE_R8(%rsp), %r8
136 movq REGISTER_SAVE_RDI(%rsp), %rdi
137 movq REGISTER_SAVE_RSI(%rsp), %rsi
138 movq REGISTER_SAVE_RDX(%rsp), %rdx
139 movq REGISTER_SAVE_RCX(%rsp), %rcx
140 movq REGISTER_SAVE_RAX(%rsp), %rax
141 VMOV (REGISTER_SAVE_VEC_OFF)(%rsp), %VEC(0)
142 VMOV (REGISTER_SAVE_VEC_OFF + VEC_SIZE)(%rsp), %VEC(1)
143 VMOV (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 2)(%rsp), %VEC(2)
144 VMOV (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 3)(%rsp), %VEC(3)
145 VMOV (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 4)(%rsp), %VEC(4)
146 VMOV (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 5)(%rsp), %VEC(5)
147 VMOV (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 6)(%rsp), %VEC(6)
148 VMOV (REGISTER_SAVE_VEC_OFF + VEC_SIZE * 7)(%rsp), %VEC(7)
149#if DL_RUNIME_RESOLVE_REALIGN_STACK
150 mov %RBX_LP, %RSP_LP
151 cfi_def_cfa_register(%rsp)
152 movq (%rsp), %rbx
153 cfi_restore(%rbx)
154#endif
155 # Adjust stack(PLT did 2 pushes)
156 add $(LOCAL_STORAGE_AREA + 16), %RSP_LP
157 cfi_adjust_cfa_offset(-(LOCAL_STORAGE_AREA + 16))
158 # Preserve bound registers.
159 PRESERVE_BND_REGS_PREFIX
160 jmp *%r11 # Jump to function address.
161 cfi_endproc
162 .size _dl_runtime_resolve, .-_dl_runtime_resolve
163
164
165#ifndef PROF
166# if (LR_VECTOR_OFFSET % VEC_SIZE) != 0
167# error LR_VECTOR_OFFSET must be multples of VEC_SIZE
168# endif
169
170 .globl _dl_runtime_profile
171 .hidden _dl_runtime_profile
172 .type _dl_runtime_profile, @function
173 .align 16
174_dl_runtime_profile:
175 cfi_startproc
176 cfi_adjust_cfa_offset(16) # Incorporate PLT
177 /* The La_x86_64_regs data structure pointed to by the
178 fourth paramater must be VEC_SIZE-byte aligned. This must
179 be explicitly enforced. We have the set up a dynamically
180 sized stack frame. %rbx points to the top half which
181 has a fixed size and preserves the original stack pointer. */
182
183 sub $32, %RSP_LP # Allocate the local storage.
184 cfi_adjust_cfa_offset(32)
185 movq %rbx, (%rsp)
186 cfi_rel_offset(%rbx, 0)
187
188 /* On the stack:
189 56(%rbx) parameter #1
190 48(%rbx) return address
191
192 40(%rbx) reloc index
193 32(%rbx) link_map
194
195 24(%rbx) La_x86_64_regs pointer
196 16(%rbx) framesize
197 8(%rbx) rax
198 (%rbx) rbx
199 */
200
201 movq %rax, 8(%rsp)
202 mov %RSP_LP, %RBX_LP
203 cfi_def_cfa_register(%rbx)
204
205 /* Actively align the La_x86_64_regs structure. */
206 and $-VEC_SIZE, %RSP_LP
207 /* sizeof(La_x86_64_regs). Need extra space for 8 SSE registers
208 to detect if any xmm0-xmm7 registers are changed by audit
209 module. */
210 sub $(LR_SIZE + XMM_SIZE*8), %RSP_LP
211 movq %rsp, 24(%rbx)
212
213 /* Fill the La_x86_64_regs structure. */
214 movq %rdx, LR_RDX_OFFSET(%rsp)
215 movq %r8, LR_R8_OFFSET(%rsp)
216 movq %r9, LR_R9_OFFSET(%rsp)
217 movq %rcx, LR_RCX_OFFSET(%rsp)
218 movq %rsi, LR_RSI_OFFSET(%rsp)
219 movq %rdi, LR_RDI_OFFSET(%rsp)
220 movq %rbp, LR_RBP_OFFSET(%rsp)
221
222 lea 48(%rbx), %RAX_LP
223 movq %rax, LR_RSP_OFFSET(%rsp)
224
225 /* We always store the XMM registers even if AVX is available.
226 This is to provide backward binary compatibility for existing
227 audit modules. */
228 movaps %xmm0, (LR_XMM_OFFSET)(%rsp)
229 movaps %xmm1, (LR_XMM_OFFSET + XMM_SIZE)(%rsp)
230 movaps %xmm2, (LR_XMM_OFFSET + XMM_SIZE*2)(%rsp)
231 movaps %xmm3, (LR_XMM_OFFSET + XMM_SIZE*3)(%rsp)
232 movaps %xmm4, (LR_XMM_OFFSET + XMM_SIZE*4)(%rsp)
233 movaps %xmm5, (LR_XMM_OFFSET + XMM_SIZE*5)(%rsp)
234 movaps %xmm6, (LR_XMM_OFFSET + XMM_SIZE*6)(%rsp)
235 movaps %xmm7, (LR_XMM_OFFSET + XMM_SIZE*7)(%rsp)
236
237# ifndef __ILP32__
238# ifdef HAVE_MPX_SUPPORT
239 bndmov %bnd0, (LR_BND_OFFSET)(%rsp) # Preserve bound
240 bndmov %bnd1, (LR_BND_OFFSET + BND_SIZE)(%rsp) # registers. Nops if
241 bndmov %bnd2, (LR_BND_OFFSET + BND_SIZE*2)(%rsp) # MPX not available
242 bndmov %bnd3, (LR_BND_OFFSET + BND_SIZE*3)(%rsp) # or disabled.
243# else
244 .byte 0x66,0x0f,0x1b,0x84,0x24;.long (LR_BND_OFFSET)
245 .byte 0x66,0x0f,0x1b,0x8c,0x24;.long (LR_BND_OFFSET + BND_SIZE)
246 .byte 0x66,0x0f,0x1b,0x94,0x24;.long (LR_BND_OFFSET + BND_SIZE*2)
247 .byte 0x66,0x0f,0x1b,0x9c,0x24;.long (LR_BND_OFFSET + BND_SIZE*3)
248# endif
249# endif
250
251# ifdef RESTORE_AVX
252 /* This is to support AVX audit modules. */
253 VMOVA %VEC(0), (LR_VECTOR_OFFSET)(%rsp)
254 VMOVA %VEC(1), (LR_VECTOR_OFFSET + VECTOR_SIZE)(%rsp)
255 VMOVA %VEC(2), (LR_VECTOR_OFFSET + VECTOR_SIZE*2)(%rsp)
256 VMOVA %VEC(3), (LR_VECTOR_OFFSET + VECTOR_SIZE*3)(%rsp)
257 VMOVA %VEC(4), (LR_VECTOR_OFFSET + VECTOR_SIZE*4)(%rsp)
258 VMOVA %VEC(5), (LR_VECTOR_OFFSET + VECTOR_SIZE*5)(%rsp)
259 VMOVA %VEC(6), (LR_VECTOR_OFFSET + VECTOR_SIZE*6)(%rsp)
260 VMOVA %VEC(7), (LR_VECTOR_OFFSET + VECTOR_SIZE*7)(%rsp)
261
262 /* Save xmm0-xmm7 registers to detect if any of them are
263 changed by audit module. */
264 vmovdqa %xmm0, (LR_SIZE)(%rsp)
265 vmovdqa %xmm1, (LR_SIZE + XMM_SIZE)(%rsp)
266 vmovdqa %xmm2, (LR_SIZE + XMM_SIZE*2)(%rsp)
267 vmovdqa %xmm3, (LR_SIZE + XMM_SIZE*3)(%rsp)
268 vmovdqa %xmm4, (LR_SIZE + XMM_SIZE*4)(%rsp)
269 vmovdqa %xmm5, (LR_SIZE + XMM_SIZE*5)(%rsp)
270 vmovdqa %xmm6, (LR_SIZE + XMM_SIZE*6)(%rsp)
271 vmovdqa %xmm7, (LR_SIZE + XMM_SIZE*7)(%rsp)
272# endif
273
274 mov %RSP_LP, %RCX_LP # La_x86_64_regs pointer to %rcx.
275 mov 48(%rbx), %RDX_LP # Load return address if needed.
276 mov 40(%rbx), %RSI_LP # Copy args pushed by PLT in register.
277 mov 32(%rbx), %RDI_LP # %rdi: link_map, %rsi: reloc_index
278 lea 16(%rbx), %R8_LP # Address of framesize
279 call _dl_profile_fixup # Call resolver.
280
281 mov %RAX_LP, %R11_LP # Save return value.
282
283 movq 8(%rbx), %rax # Get back register content.
284 movq LR_RDX_OFFSET(%rsp), %rdx
285 movq LR_R8_OFFSET(%rsp), %r8
286 movq LR_R9_OFFSET(%rsp), %r9
287
288 movaps (LR_XMM_OFFSET)(%rsp), %xmm0
289 movaps (LR_XMM_OFFSET + XMM_SIZE)(%rsp), %xmm1
290 movaps (LR_XMM_OFFSET + XMM_SIZE*2)(%rsp), %xmm2
291 movaps (LR_XMM_OFFSET + XMM_SIZE*3)(%rsp), %xmm3
292 movaps (LR_XMM_OFFSET + XMM_SIZE*4)(%rsp), %xmm4
293 movaps (LR_XMM_OFFSET + XMM_SIZE*5)(%rsp), %xmm5
294 movaps (LR_XMM_OFFSET + XMM_SIZE*6)(%rsp), %xmm6
295 movaps (LR_XMM_OFFSET + XMM_SIZE*7)(%rsp), %xmm7
296
297# ifdef RESTORE_AVX
298 /* Check if any xmm0-xmm7 registers are changed by audit
299 module. */
300 vpcmpeqq (LR_SIZE)(%rsp), %xmm0, %xmm8
301 vpmovmskb %xmm8, %esi
302 cmpl $0xffff, %esi
303 je 2f
304 vmovdqa %xmm0, (LR_VECTOR_OFFSET)(%rsp)
305 jmp 1f
3062: VMOVA (LR_VECTOR_OFFSET)(%rsp), %VEC(0)
307 vmovdqa %xmm0, (LR_XMM_OFFSET)(%rsp)
308
3091: vpcmpeqq (LR_SIZE + XMM_SIZE)(%rsp), %xmm1, %xmm8
310 vpmovmskb %xmm8, %esi
311 cmpl $0xffff, %esi
312 je 2f
313 vmovdqa %xmm1, (LR_VECTOR_OFFSET + VECTOR_SIZE)(%rsp)
314 jmp 1f
3152: VMOVA (LR_VECTOR_OFFSET + VECTOR_SIZE)(%rsp), %VEC(1)
316 vmovdqa %xmm1, (LR_XMM_OFFSET + XMM_SIZE)(%rsp)
317
3181: vpcmpeqq (LR_SIZE + XMM_SIZE*2)(%rsp), %xmm2, %xmm8
319 vpmovmskb %xmm8, %esi
320 cmpl $0xffff, %esi
321 je 2f
322 vmovdqa %xmm2, (LR_VECTOR_OFFSET + VECTOR_SIZE*2)(%rsp)
323 jmp 1f
3242: VMOVA (LR_VECTOR_OFFSET + VECTOR_SIZE*2)(%rsp), %VEC(2)
325 vmovdqa %xmm2, (LR_XMM_OFFSET + XMM_SIZE*2)(%rsp)
326
3271: vpcmpeqq (LR_SIZE + XMM_SIZE*3)(%rsp), %xmm3, %xmm8
328 vpmovmskb %xmm8, %esi
329 cmpl $0xffff, %esi
330 je 2f
331 vmovdqa %xmm3, (LR_VECTOR_OFFSET + VECTOR_SIZE*3)(%rsp)
332 jmp 1f
3332: VMOVA (LR_VECTOR_OFFSET + VECTOR_SIZE*3)(%rsp), %VEC(3)
334 vmovdqa %xmm3, (LR_XMM_OFFSET + XMM_SIZE*3)(%rsp)
335
3361: vpcmpeqq (LR_SIZE + XMM_SIZE*4)(%rsp), %xmm4, %xmm8
337 vpmovmskb %xmm8, %esi
338 cmpl $0xffff, %esi
339 je 2f
340 vmovdqa %xmm4, (LR_VECTOR_OFFSET + VECTOR_SIZE*4)(%rsp)
341 jmp 1f
3422: VMOVA (LR_VECTOR_OFFSET + VECTOR_SIZE*4)(%rsp), %VEC(4)
343 vmovdqa %xmm4, (LR_XMM_OFFSET + XMM_SIZE*4)(%rsp)
344
3451: vpcmpeqq (LR_SIZE + XMM_SIZE*5)(%rsp), %xmm5, %xmm8
346 vpmovmskb %xmm8, %esi
347 cmpl $0xffff, %esi
348 je 2f
349 vmovdqa %xmm5, (LR_VECTOR_OFFSET + VECTOR_SIZE*5)(%rsp)
350 jmp 1f
3512: VMOVA (LR_VECTOR_OFFSET + VECTOR_SIZE*5)(%rsp), %VEC(5)
352 vmovdqa %xmm5, (LR_XMM_OFFSET + XMM_SIZE*5)(%rsp)
353
3541: vpcmpeqq (LR_SIZE + XMM_SIZE*6)(%rsp), %xmm6, %xmm8
355 vpmovmskb %xmm8, %esi
356 cmpl $0xffff, %esi
357 je 2f
358 vmovdqa %xmm6, (LR_VECTOR_OFFSET + VECTOR_SIZE*6)(%rsp)
359 jmp 1f
3602: VMOVA (LR_VECTOR_OFFSET + VECTOR_SIZE*6)(%rsp), %VEC(6)
361 vmovdqa %xmm6, (LR_XMM_OFFSET + XMM_SIZE*6)(%rsp)
362
3631: vpcmpeqq (LR_SIZE + XMM_SIZE*7)(%rsp), %xmm7, %xmm8
364 vpmovmskb %xmm8, %esi
365 cmpl $0xffff, %esi
366 je 2f
367 vmovdqa %xmm7, (LR_VECTOR_OFFSET + VECTOR_SIZE*7)(%rsp)
368 jmp 1f
3692: VMOVA (LR_VECTOR_OFFSET + VECTOR_SIZE*7)(%rsp), %VEC(7)
370 vmovdqa %xmm7, (LR_XMM_OFFSET + XMM_SIZE*7)(%rsp)
371
3721:
373# endif
374
375# ifndef __ILP32__
376# ifdef HAVE_MPX_SUPPORT
377 bndmov (LR_BND_OFFSET)(%rsp), %bnd0 # Restore bound
378 bndmov (LR_BND_OFFSET + BND_SIZE)(%rsp), %bnd1 # registers.
379 bndmov (LR_BND_OFFSET + BND_SIZE*2)(%rsp), %bnd2
380 bndmov (LR_BND_OFFSET + BND_SIZE*3)(%rsp), %bnd3
381# else
382 .byte 0x66,0x0f,0x1a,0x84,0x24;.long (LR_BND_OFFSET)
383 .byte 0x66,0x0f,0x1a,0x8c,0x24;.long (LR_BND_OFFSET + BND_SIZE)
384 .byte 0x66,0x0f,0x1a,0x94,0x24;.long (LR_BND_OFFSET + BND_SIZE*2)
385 .byte 0x66,0x0f,0x1a,0x9c,0x24;.long (LR_BND_OFFSET + BND_SIZE*3)
386# endif
387# endif
388
389 mov 16(%rbx), %R10_LP # Anything in framesize?
390 test %R10_LP, %R10_LP
391 PRESERVE_BND_REGS_PREFIX
392 jns 3f
393
394 /* There's nothing in the frame size, so there
395 will be no call to the _dl_call_pltexit. */
396
397 /* Get back registers content. */
398 movq LR_RCX_OFFSET(%rsp), %rcx
399 movq LR_RSI_OFFSET(%rsp), %rsi
400 movq LR_RDI_OFFSET(%rsp), %rdi
401
402 mov %RBX_LP, %RSP_LP
403 movq (%rsp), %rbx
404 cfi_restore(%rbx)
405 cfi_def_cfa_register(%rsp)
406
407 add $48, %RSP_LP # Adjust the stack to the return value
408 # (eats the reloc index and link_map)
409 cfi_adjust_cfa_offset(-48)
410 PRESERVE_BND_REGS_PREFIX
411 jmp *%r11 # Jump to function address.
412
4133:
414 cfi_adjust_cfa_offset(48)
415 cfi_rel_offset(%rbx, 0)
416 cfi_def_cfa_register(%rbx)
417
418 /* At this point we need to prepare new stack for the function
419 which has to be called. We copy the original stack to a
420 temporary buffer of the size specified by the 'framesize'
421 returned from _dl_profile_fixup */
422
423 lea LR_RSP_OFFSET(%rbx), %RSI_LP # stack
424 add $8, %R10_LP
425 and $-16, %R10_LP
426 mov %R10_LP, %RCX_LP
427 sub %R10_LP, %RSP_LP
428 mov %RSP_LP, %RDI_LP
429 shr $3, %RCX_LP
430 rep
431 movsq
432
433 movq 24(%rdi), %rcx # Get back register content.
434 movq 32(%rdi), %rsi
435 movq 40(%rdi), %rdi
436
437 PRESERVE_BND_REGS_PREFIX
438 call *%r11
439
440 mov 24(%rbx), %RSP_LP # Drop the copied stack content
441
442 /* Now we have to prepare the La_x86_64_retval structure for the
443 _dl_call_pltexit. The La_x86_64_regs is being pointed by rsp now,
444 so we just need to allocate the sizeof(La_x86_64_retval) space on
445 the stack, since the alignment has already been taken care of. */
446# ifdef RESTORE_AVX
447 /* sizeof(La_x86_64_retval). Need extra space for 2 SSE
448 registers to detect if xmm0/xmm1 registers are changed
449 by audit module. */
450 sub $(LRV_SIZE + XMM_SIZE*2), %RSP_LP
451# else
452 sub $LRV_SIZE, %RSP_LP # sizeof(La_x86_64_retval)
453# endif
454 mov %RSP_LP, %RCX_LP # La_x86_64_retval argument to %rcx.
455
456 /* Fill in the La_x86_64_retval structure. */
457 movq %rax, LRV_RAX_OFFSET(%rcx)
458 movq %rdx, LRV_RDX_OFFSET(%rcx)
459
460 movaps %xmm0, LRV_XMM0_OFFSET(%rcx)
461 movaps %xmm1, LRV_XMM1_OFFSET(%rcx)
462
463# ifdef RESTORE_AVX
464 /* This is to support AVX audit modules. */
465 VMOVA %VEC(0), LRV_VECTOR0_OFFSET(%rcx)
466 VMOVA %VEC(1), LRV_VECTOR1_OFFSET(%rcx)
467
468 /* Save xmm0/xmm1 registers to detect if they are changed
469 by audit module. */
470 vmovdqa %xmm0, (LRV_SIZE)(%rcx)
471 vmovdqa %xmm1, (LRV_SIZE + XMM_SIZE)(%rcx)
472# endif
473
474# ifndef __ILP32__
475# ifdef HAVE_MPX_SUPPORT
476 bndmov %bnd0, LRV_BND0_OFFSET(%rcx) # Preserve returned bounds.
477 bndmov %bnd1, LRV_BND1_OFFSET(%rcx)
478# else
479 .byte 0x66,0x0f,0x1b,0x81;.long (LRV_BND0_OFFSET)
480 .byte 0x66,0x0f,0x1b,0x89;.long (LRV_BND1_OFFSET)
481# endif
482# endif
483
484 fstpt LRV_ST0_OFFSET(%rcx)
485 fstpt LRV_ST1_OFFSET(%rcx)
486
487 movq 24(%rbx), %rdx # La_x86_64_regs argument to %rdx.
488 movq 40(%rbx), %rsi # Copy args pushed by PLT in register.
489 movq 32(%rbx), %rdi # %rdi: link_map, %rsi: reloc_index
490 call _dl_call_pltexit
491
492 /* Restore return registers. */
493 movq LRV_RAX_OFFSET(%rsp), %rax
494 movq LRV_RDX_OFFSET(%rsp), %rdx
495
496 movaps LRV_XMM0_OFFSET(%rsp), %xmm0
497 movaps LRV_XMM1_OFFSET(%rsp), %xmm1
498
499# ifdef RESTORE_AVX
500 /* Check if xmm0/xmm1 registers are changed by audit module. */
501 vpcmpeqq (LRV_SIZE)(%rsp), %xmm0, %xmm2
502 vpmovmskb %xmm2, %esi
503 cmpl $0xffff, %esi
504 jne 1f
505 VMOVA LRV_VECTOR0_OFFSET(%rsp), %VEC(0)
506
5071: vpcmpeqq (LRV_SIZE + XMM_SIZE)(%rsp), %xmm1, %xmm2
508 vpmovmskb %xmm2, %esi
509 cmpl $0xffff, %esi
510 jne 1f
511 VMOVA LRV_VECTOR1_OFFSET(%rsp), %VEC(1)
512
5131:
514# endif
515
516# ifndef __ILP32__
517# ifdef HAVE_MPX_SUPPORT
518 bndmov LRV_BND0_OFFSET(%rsp), %bnd0 # Restore bound registers.
519 bndmov LRV_BND1_OFFSET(%rsp), %bnd1
520# else
521 .byte 0x66,0x0f,0x1a,0x84,0x24;.long (LRV_BND0_OFFSET)
522 .byte 0x66,0x0f,0x1a,0x8c,0x24;.long (LRV_BND1_OFFSET)
523# endif
524# endif
525
526 fldt LRV_ST1_OFFSET(%rsp)
527 fldt LRV_ST0_OFFSET(%rsp)
528
529 mov %RBX_LP, %RSP_LP
530 movq (%rsp), %rbx
531 cfi_restore(%rbx)
532 cfi_def_cfa_register(%rsp)
533
534 add $48, %RSP_LP # Adjust the stack to the return value
535 # (eats the reloc index and link_map)
536 cfi_adjust_cfa_offset(-48)
537 PRESERVE_BND_REGS_PREFIX
538 retq
539
540 cfi_endproc
541 .size _dl_runtime_profile, .-_dl_runtime_profile
542#endif
543