1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
|
/*--------------------------------------------------------------------*/
/*--- Take snapshots of client stacks. m_stacktrace.c ---*/
/*--------------------------------------------------------------------*/
/*
This file is part of Valgrind, a dynamic binary instrumentation
framework.
Copyright (C) 2000-2009 Julian Seward
jseward@acm.org
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the
License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
02111-1307, USA.
The GNU General Public License is contained in the file COPYING.
*/
#include "pub_core_basics.h"
#include "pub_core_vki.h"
#include "pub_core_threadstate.h"
#include "pub_core_debuginfo.h" // XXX: circular dependency
#include "pub_core_aspacemgr.h" // For VG_(is_addressable)()
#include "pub_core_libcbase.h"
#include "pub_core_libcassert.h"
#include "pub_core_libcprint.h"
#include "pub_core_machine.h"
#include "pub_core_options.h"
#include "pub_core_stacks.h" // VG_(stack_limits)
#include "pub_core_stacktrace.h"
#include "pub_core_xarray.h"
#include "pub_core_clientstate.h" // VG_(client__dl_sysinfo_int80)
#include "pub_core_trampoline.h"
/*------------------------------------------------------------*/
/*--- Exported functions. ---*/
/*------------------------------------------------------------*/
/* Take a snapshot of the client's stack, putting the up to 'n_ips'
IPs into 'ips'. In order to be thread-safe, we pass in the
thread's IP SP, FP if that's meaningful, and LR if that's
meaningful. Returns number of IPs put in 'ips'.
If you know what the thread ID for this stack is, send that as the
first parameter, else send zero. This helps generate better stack
traces on ppc64-linux and has no effect on other platforms.
*/
UInt VG_(get_StackTrace_wrk) ( ThreadId tid_if_known,
/*OUT*/Addr* ips, UInt n_ips,
/*OUT*/Addr* sps, /*OUT*/Addr* fps,
Addr ip, Addr sp, Addr fp, Addr lr,
Addr fp_min, Addr fp_max_orig )
{
# if defined(VGP_ppc32_linux) || defined(VGP_ppc64_linux) \
|| defined(VGP_ppc32_aix5) \
|| defined(VGP_ppc64_aix5)
Bool lr_is_first_RA = False;
# endif
# if defined(VGP_ppc64_linux) || defined(VGP_ppc64_aix5) \
|| defined(VGP_ppc32_aix5)
Word redir_stack_size = 0;
Word redirs_used = 0;
# endif
Bool debug = False;
Int i;
Addr fp_max;
UInt n_found = 0;
vg_assert(sizeof(Addr) == sizeof(UWord));
vg_assert(sizeof(Addr) == sizeof(void*));
/* Snaffle IPs from the client's stack into ips[0 .. n_ips-1],
stopping when the trail goes cold, which we guess to be
when FP is not a reasonable stack location. */
// JRS 2002-sep-17: hack, to round up fp_max to the end of the
// current page, at least. Dunno if it helps.
// NJN 2002-sep-17: seems to -- stack traces look like 1.0.X again
fp_max = VG_PGROUNDUP(fp_max_orig);
if (fp_max >= sizeof(Addr))
fp_max -= sizeof(Addr);
if (debug)
VG_(printf)("n_ips=%d fp_min=0x%lx fp_max_orig=0x%lx, "
"fp_max=0x%lx ip=0x%lx fp=0x%lx\n",
n_ips, fp_min, fp_max_orig, fp_max, ip, fp);
/* Assertion broken before main() is reached in pthreaded programs; the
* offending stack traces only have one item. --njn, 2002-aug-16 */
/* vg_assert(fp_min <= fp_max);*/
if (fp_min + 512 >= fp_max) {
/* If the stack limits look bogus, don't poke around ... but
don't bomb out either. */
if (sps) sps[0] = sp;
if (fps) fps[0] = fp;
ips[0] = ip;
return 1;
}
/* Otherwise unwind the stack in a platform-specific way. Trying
to merge the x86, amd64, ppc32 and ppc64 logic into a single
piece of code is just too confusing and difficult to
performance-tune. */
# if defined(VGP_x86_linux)
/*--------------------- x86 ---------------------*/
/* fp is %ebp. sp is %esp. ip is %eip. */
if (sps) sps[0] = sp;
if (fps) fps[0] = fp;
ips[0] = ip;
i = 1;
/* Loop unwinding the stack. Note that the IP value we get on
* each pass (whether from CFI info or a stack frame) is a
* return address so is actually after the calling instruction
* in the calling function.
*
* Because of this we subtract one from the IP after each pass
* of the loop so that we find the right CFI block on the next
* pass - otherwise we can find the wrong CFI info if it happens
* to change after the calling instruction and that will mean
* that we will fail to unwind the next step.
*
* This most frequently happens at the end of a function when
* a tail call occurs and we wind up using the CFI info for the
* next function which is completely wrong.
*/
while (True) {
if (i >= n_ips)
break;
/* Try to derive a new (ip,sp,fp) triple from the current
set. */
/* On x86, first try the old-fashioned method of following the
%ebp-chain. Code which doesn't use this (that is, compiled
with -fomit-frame-pointer) is not ABI compliant and so
relatively rare. Besides, trying the CFI first almost always
fails, and is expensive. */
/* Deal with frames resulting from functions which begin "pushl%
ebp ; movl %esp, %ebp" which is the ABI-mandated preamble. */
if (fp_min <= fp && fp <= fp_max
- 1 * sizeof(UWord)/*see comment below*/) {
/* fp looks sane, so use it. */
ip = (((UWord*)fp)[1]);
sp = fp + sizeof(Addr) /*saved %ebp*/
+ sizeof(Addr) /*ra*/;
fp = (((UWord*)fp)[0]);
if (sps) sps[i] = sp;
if (fps) fps[i] = fp;
ips[i++] = ip - 1; /* -1: refer to calling insn, not the RA */
if (debug)
VG_(printf)(" ipsF[%d]=0x%08lx\n", i-1, ips[i-1]);
ip = ip - 1; /* as per comment at the head of this loop */
continue;
}
/* That didn't work out, so see if there is any CF info to hand
which can be used. */
if ( VG_(use_CF_info)( &ip, &sp, &fp, fp_min, fp_max ) ) {
if (sps) sps[i] = sp;
if (fps) fps[i] = fp;
ips[i++] = ip - 1; /* -1: refer to calling insn, not the RA */
if (debug)
VG_(printf)(" ipsC[%d]=0x%08lx\n", i-1, ips[i-1]);
ip = ip - 1; /* as per comment at the head of this loop */
continue;
}
/* And, similarly, try for MSVC FPO unwind info. */
if ( VG_(use_FPO_info)( &ip, &sp, &fp, fp_min, fp_max ) ) {
if (sps) sps[i] = sp;
if (fps) fps[i] = fp;
ips[i++] = ip;
if (debug)
VG_(printf)(" ipsC[%d]=0x%08lx\n", i-1, ips[i-1]);
ip = ip - 1;
continue;
}
/* No luck. We have to give up. */
break;
}
# elif defined(VGP_amd64_linux)
/*--------------------- amd64 ---------------------*/
/* fp is %rbp. sp is %rsp. ip is %rip. */
ips[0] = ip;
if (sps) sps[0] = sp;
if (fps) fps[0] = fp;
i = 1;
/* Loop unwinding the stack. Note that the IP value we get on
* each pass (whether from CFI info or a stack frame) is a
* return address so is actually after the calling instruction
* in the calling function.
*
* Because of this we subtract one from the IP after each pass
* of the loop so that we find the right CFI block on the next
* pass - otherwise we can find the wrong CFI info if it happens
* to change after the calling instruction and that will mean
* that we will fail to unwind the next step.
*
* This most frequently happens at the end of a function when
* a tail call occurs and we wind up using the CFI info for the
* next function which is completely wrong.
*/
while (True) {
if (i >= n_ips)
break;
/* Try to derive a new (ip,sp,fp) triple from the current
set. */
/* First off, see if there is any CFI info to hand which can
be used. */
if ( VG_(use_CF_info)( &ip, &sp, &fp, fp_min, fp_max ) ) {
if (sps) sps[i] = sp;
if (fps) fps[i] = fp;
ips[i++] = ip - 1; /* -1: refer to calling insn, not the RA */
if (debug)
VG_(printf)(" ipsC[%d]=%#08lx\n", i-1, ips[i-1]);
ip = ip - 1; /* as per comment at the head of this loop */
continue;
}
/* If VG_(use_CF_info) fails, it won't modify ip/sp/fp, so
we can safely try the old-fashioned method. */
/* This bit is supposed to deal with frames resulting from
functions which begin "pushq %rbp ; movq %rsp, %rbp".
Unfortunately, since we can't (easily) look at the insns at
the start of the fn, like GDB does, there's no reliable way
to tell. Hence the hack of first trying out CFI, and if that
fails, then use this as a fallback. */
/* Note: re "- 1 * sizeof(UWord)", need to take account of the
fact that we are prodding at & ((UWord*)fp)[1] and so need to
adjust the limit check accordingly. Omitting this has been
observed to cause segfaults on rare occasions. */
if (fp_min <= fp && fp <= fp_max - 1 * sizeof(UWord)) {
/* fp looks sane, so use it. */
ip = (((UWord*)fp)[1]);
sp = fp + sizeof(Addr) /*saved %rbp*/
+ sizeof(Addr) /*ra*/;
fp = (((UWord*)fp)[0]);
if (sps) sps[i] = sp;
if (fps) fps[i] = fp;
ips[i++] = ip - 1; /* -1: refer to calling insn, not the RA */
if (debug)
VG_(printf)(" ipsF[%d]=%#08lx\n", i-1, ips[i-1]);
ip = ip - 1; /* as per comment at the head of this loop */
continue;
}
/* Last-ditch hack (evidently GDB does something similar). We
are in the middle of nowhere and we have a nonsense value for
the frame pointer. If the stack pointer is still valid,
assume that what it points at is a return address. Yes,
desperate measures. Could do better here:
- check that the supposed return address is in
an executable page
- check that the supposed return address is just after a call insn
- given those two checks, don't just consider *sp as the return
address; instead scan a likely section of stack (eg sp .. sp+256)
and use suitable values found there.
*/
if (fp_min <= sp && sp < fp_max) {
ip = ((UWord*)sp)[0];
if (sps) sps[i] = sp;
if (fps) fps[i] = fp;
ips[i++] = ip == 0
? 0 /* sp[0] == 0 ==> stuck at the bottom of a
thread stack */
: ip - 1; /* -1: refer to calling insn, not the RA */
if (debug)
VG_(printf)(" ipsH[%d]=%#08lx\n", i-1, ips[i-1]);
ip = ip - 1; /* as per comment at the head of this loop */
sp += 8;
continue;
}
/* No luck at all. We have to give up. */
break;
}
# elif defined(VGP_ppc32_linux) || defined(VGP_ppc64_linux) \
|| defined(VGP_ppc32_aix5) || defined(VGP_ppc64_aix5)
/*--------------------- ppc32/64 ---------------------*/
/* fp is %r1. ip is %cia. Note, ppc uses r1 as both the stack and
frame pointers. */
# if defined(VGP_ppc64_linux) || defined(VGP_ppc64_aix5)
redir_stack_size = VEX_GUEST_PPC64_REDIR_STACK_SIZE;
redirs_used = 0;
# elif defined(VGP_ppc32_aix5)
redir_stack_size = VEX_GUEST_PPC32_REDIR_STACK_SIZE;
redirs_used = 0;
# endif
# if defined(VG_PLAT_USES_PPCTOC)
/* Deal with bogus LR values caused by function
interception/wrapping on ppc-TOC platforms; see comment on
similar code a few lines further down. */
if (ULong_to_Ptr(lr) == (void*)&VG_(ppctoc_magic_redirect_return_stub)
&& VG_(is_valid_tid)(tid_if_known)) {
Word hsp = VG_(threads)[tid_if_known].arch.vex.guest_REDIR_SP;
redirs_used++;
if (hsp >= 1 && hsp < redir_stack_size)
lr = VG_(threads)[tid_if_known]
.arch.vex.guest_REDIR_STACK[hsp-1];
}
# endif
/* We have to determine whether or not LR currently holds this fn
(call it F)'s return address. It might not if F has previously
called some other function, hence overwriting LR with a pointer
to some part of F. Hence if LR and IP point to the same
function then we conclude LR does not hold this function's
return address; instead the LR at entry must have been saved in
the stack by F's prologue and so we must get it from there
instead. Note all this guff only applies to the innermost
frame. */
lr_is_first_RA = False;
{
# define M_VG_ERRTXT 1000
UChar buf_lr[M_VG_ERRTXT], buf_ip[M_VG_ERRTXT];
/* The following conditional looks grossly inefficient and
surely could be majorly improved, with not much effort. */
if (VG_(get_fnname_raw) (lr, buf_lr, M_VG_ERRTXT))
if (VG_(get_fnname_raw) (ip, buf_ip, M_VG_ERRTXT))
if (VG_(strncmp)(buf_lr, buf_ip, M_VG_ERRTXT))
lr_is_first_RA = True;
# undef M_VG_ERRTXT
}
if (sps) sps[0] = fp; /* NB. not sp */
if (fps) fps[0] = fp;
ips[0] = ip;
i = 1;
if (fp_min <= fp && fp < fp_max-VG_WORDSIZE+1) {
/* initial FP is sane; keep going */
fp = (((UWord*)fp)[0]);
while (True) {
/* On ppc64-linux (ppc64-elf, really), and on AIX, the lr save
slot is 2 words back from sp, whereas on ppc32-elf(?) it's
only one word back. */
# if defined(VGP_ppc64_linux) \
|| defined(VGP_ppc32_aix5) || defined(VGP_ppc64_aix5)
const Int lr_offset = 2;
# else
const Int lr_offset = 1;
# endif
if (i >= n_ips)
break;
/* Try to derive a new (ip,fp) pair from the current set. */
if (fp_min <= fp && fp <= fp_max - lr_offset * sizeof(UWord)) {
/* fp looks sane, so use it. */
if (i == 1 && lr_is_first_RA)
ip = lr;
else
ip = (((UWord*)fp)[lr_offset]);
# if defined(VG_PLAT_USES_PPCTOC)
/* Nasty hack to do with function replacement/wrapping on
ppc64-linux/ppc64-aix/ppc32-aix. If LR points to our
magic return stub, then we are in a wrapped or
intercepted function, in which LR has been messed with.
The original LR will have been pushed onto the thread's
hidden REDIR stack one down from the top (top element
is the saved R2) and so we should restore the value
from there instead. Since nested redirections can and
do happen, we keep track of the number of nested LRs
used by the unwinding so far with 'redirs_used'. */
if (ip == (Addr)&VG_(ppctoc_magic_redirect_return_stub)
&& VG_(is_valid_tid)(tid_if_known)) {
Word hsp = VG_(threads)[tid_if_known]
.arch.vex.guest_REDIR_SP;
hsp -= 2 * redirs_used;
redirs_used ++;
if (hsp >= 1 && hsp < redir_stack_size)
ip = VG_(threads)[tid_if_known]
.arch.vex.guest_REDIR_STACK[hsp-1];
}
# endif
fp = (((UWord*)fp)[0]);
if (sps) sps[i] = fp; /* NB. not sp */
if (fps) fps[i] = fp;
ips[i++] = ip - 1; /* -1: refer to calling insn, not the RA */
if (debug)
VG_(printf)(" ipsF[%d]=%#08lx\n", i-1, ips[i-1]);
ip = ip - 1; /* ip is probably dead at this point, but
play safe, a la x86/amd64 above. See
extensive comments above. */
continue;
}
/* No luck there. We have to give up. */
break;
}
}
# else
# error "Unknown platform"
# endif
n_found = i;
return n_found;
}
UInt VG_(get_StackTrace) ( ThreadId tid,
/*OUT*/StackTrace ips, UInt n_ips,
/*OUT*/StackTrace sps,
/*OUT*/StackTrace fps,
Word first_ip_delta )
{
/* thread in thread table */
Addr ip = VG_(get_IP)(tid);
Addr fp = VG_(get_FP)(tid);
Addr sp = VG_(get_SP)(tid);
Addr lr = VG_(get_LR)(tid);
Addr stack_highest_word = VG_(threads)[tid].client_stack_highest_word;
Addr stack_lowest_word = 0;
# if defined(VGP_x86_linux)
/* Nasty little hack to deal with syscalls - if libc is using its
_dl_sysinfo_int80 function for syscalls (the TLS version does),
then ip will always appear to be in that function when doing a
syscall, not the actual libc function doing the syscall. This
check sees if IP is within that function, and pops the return
address off the stack so that ip is placed within the library
function calling the syscall. This makes stack backtraces much
more useful.
The function is assumed to look like this (from glibc-2.3.6 sources):
_dl_sysinfo_int80:
int $0x80
ret
That is 3 (2+1) bytes long. We could be more thorough and check
the 3 bytes of the function are as expected, but I can't be
bothered.
*/
if (VG_(client__dl_sysinfo_int80) != 0 /* we know its address */
&& ip >= VG_(client__dl_sysinfo_int80)
&& ip < VG_(client__dl_sysinfo_int80)+3
&& VG_(am_is_valid_for_client)(sp, sizeof(Addr), VKI_PROT_READ)) {
ip = *(Addr *)sp;
sp += sizeof(Addr);
}
# endif
/* See if we can get a better idea of the stack limits */
VG_(stack_limits)(sp, &stack_lowest_word, &stack_highest_word);
/* Take into account the first_ip_delta. */
vg_assert( sizeof(Addr) == sizeof(Word) );
ip += first_ip_delta;
if (0)
VG_(printf)("tid %d: stack_highest=0x%08lx ip=0x%08lx "
"sp=0x%08lx fp=0x%08lx\n",
tid, stack_highest_word, ip, sp, fp);
return VG_(get_StackTrace_wrk)(tid, ips, n_ips,
sps, fps,
ip, sp, fp, lr, sp,
stack_highest_word);
}
static void printIpDesc(UInt n, Addr ip)
{
#define BUF_LEN 4096
static UChar buf[BUF_LEN];
VG_(describe_IP)(ip, buf, BUF_LEN);
if (VG_(clo_xml)) {
VG_(message)(Vg_UserMsg, " %s", buf);
} else {
VG_(message)(Vg_UserMsg, " %s %s", ( n == 0 ? "at" : "by" ), buf);
}
}
/* Print a StackTrace. */
void VG_(pp_StackTrace) ( StackTrace ips, UInt n_ips )
{
vg_assert( n_ips > 0 );
if (VG_(clo_xml))
VG_(message)(Vg_UserMsg, " <stack>");
VG_(apply_StackTrace)( printIpDesc, ips, n_ips );
if (VG_(clo_xml))
VG_(message)(Vg_UserMsg, " </stack>");
}
/* Get and immediately print a StackTrace. */
void VG_(get_and_pp_StackTrace) ( ThreadId tid, UInt n_ips )
{
Addr ips[n_ips];
UInt n_ips_obtained
= VG_(get_StackTrace)(tid, ips, n_ips,
NULL/*array to dump SP values in*/,
NULL/*array to dump FP values in*/,
0/*first_ip_delta*/);
VG_(pp_StackTrace)(ips, n_ips_obtained);
}
void VG_(apply_StackTrace)( void(*action)(UInt n, Addr ip),
StackTrace ips, UInt n_ips )
{
Bool main_done = False;
Int i = 0;
vg_assert(n_ips > 0);
do {
Addr ip = ips[i];
// Stop after the first appearance of "main" or one of the other names
// (the appearance of which is a pretty good sign that we've gone past
// main without seeing it, for whatever reason)
if ( ! VG_(clo_show_below_main) ) {
Vg_FnNameKind kind = VG_(get_fnname_kind_from_IP)(ip);
if (Vg_FnNameMain == kind || Vg_FnNameBelowMain == kind) {
main_done = True;
}
}
// Act on the ip
action(i, ip);
i++;
// re 'while' condition: stop if we hit a zero value (the traditional
// end-of-stack marker) or a ~0 value. The latter because r8818
// (in this file) changes the meaning of entries [1] and above in a
// stack trace, by subtracting 1 from them. Hence stacks that used
// to end with a zero value now end in -1 and so we must detect
// that too.
} while (i < n_ips && ips[i] != 0 && ips[i] != ~(Addr)0 && !main_done);
#undef MYBUF_LEN
}
/*--------------------------------------------------------------------*/
/*--- end ---*/
/*--------------------------------------------------------------------*/
|