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
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
|
/* Linux-dependent part of branch trace support for GDB, and GDBserver.
Copyright (C) 2013-2015 Free Software Foundation, Inc.
Contributed by Intel Corp. <markus.t.metzger@intel.com>
This file is part of GDB.
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 3 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, see <http://www.gnu.org/licenses/>. */
#include "common-defs.h"
#include "linux-btrace.h"
#include "common-regcache.h"
#include "gdb_wait.h"
#include "x86-cpuid.h"
#ifdef HAVE_SYS_SYSCALL_H
#include <sys/syscall.h>
#endif
#if HAVE_LINUX_PERF_EVENT_H && defined(SYS_perf_event_open)
#include <stdint.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/user.h>
#include <sys/ptrace.h>
#include <sys/types.h>
#include <signal.h>
/* A branch trace record in perf_event. */
struct perf_event_bts
{
/* The linear address of the branch source. */
uint64_t from;
/* The linear address of the branch destination. */
uint64_t to;
};
/* A perf_event branch trace sample. */
struct perf_event_sample
{
/* The perf_event sample header. */
struct perf_event_header header;
/* The perf_event branch tracing payload. */
struct perf_event_bts bts;
};
/* Return non-zero if there is new data in PEVENT; zero otherwise. */
static int
perf_event_new_data (const struct perf_event_buffer *pev)
{
return *pev->data_head != pev->last_head;
}
/* Check whether an address is in the kernel. */
static inline int
perf_event_is_kernel_addr (const struct btrace_target_info *tinfo,
uint64_t addr)
{
uint64_t mask;
/* If we don't know the size of a pointer, we can't check. Let's assume it's
not a kernel address in this case. */
if (tinfo->ptr_bits == 0)
return 0;
/* A bit mask for the most significant bit in an address. */
mask = (uint64_t) 1 << (tinfo->ptr_bits - 1);
/* Check whether the most significant bit in the address is set. */
return (addr & mask) != 0;
}
/* Check whether a perf event record should be skipped. */
static inline int
perf_event_skip_record (const struct btrace_target_info *tinfo,
const struct perf_event_bts *bts)
{
/* The hardware may report branches from kernel into user space. Branches
from user into kernel space will be suppressed. We filter the former to
provide a consistent branch trace excluding kernel. */
return perf_event_is_kernel_addr (tinfo, bts->from);
}
/* Perform a few consistency checks on a perf event sample record. This is
meant to catch cases when we get out of sync with the perf event stream. */
static inline int
perf_event_sample_ok (const struct perf_event_sample *sample)
{
if (sample->header.type != PERF_RECORD_SAMPLE)
return 0;
if (sample->header.size != sizeof (*sample))
return 0;
return 1;
}
/* Branch trace is collected in a circular buffer [begin; end) as pairs of from
and to addresses (plus a header).
Start points into that buffer at the next sample position.
We read the collected samples backwards from start.
While reading the samples, we convert the information into a list of blocks.
For two adjacent samples s1 and s2, we form a block b such that b.begin =
s1.to and b.end = s2.from.
In case the buffer overflows during sampling, one sample may have its lower
part at the end and its upper part at the beginning of the buffer. */
static VEC (btrace_block_s) *
perf_event_read_bts (struct btrace_target_info* tinfo, const uint8_t *begin,
const uint8_t *end, const uint8_t *start,
unsigned long long size)
{
VEC (btrace_block_s) *btrace = NULL;
struct perf_event_sample sample;
unsigned long long read = 0;
struct btrace_block block = { 0, 0 };
struct regcache *regcache;
gdb_assert (begin <= start);
gdb_assert (start <= end);
/* The first block ends at the current pc. */
regcache = get_thread_regcache_for_ptid (tinfo->ptid);
block.end = regcache_read_pc (regcache);
/* The buffer may contain a partial record as its last entry (i.e. when the
buffer size is not a multiple of the sample size). */
read = sizeof (sample) - 1;
for (; read < size; read += sizeof (sample))
{
const struct perf_event_sample *psample;
/* Find the next perf_event sample in a backwards traversal. */
start -= sizeof (sample);
/* If we're still inside the buffer, we're done. */
if (begin <= start)
psample = (const struct perf_event_sample *) start;
else
{
int missing;
/* We're to the left of the ring buffer, we will wrap around and
reappear at the very right of the ring buffer. */
missing = (begin - start);
start = (end - missing);
/* If the entire sample is missing, we're done. */
if (missing == sizeof (sample))
psample = (const struct perf_event_sample *) start;
else
{
uint8_t *stack;
/* The sample wrapped around. The lower part is at the end and
the upper part is at the beginning of the buffer. */
stack = (uint8_t *) &sample;
/* Copy the two parts so we have a contiguous sample. */
memcpy (stack, start, missing);
memcpy (stack + missing, begin, sizeof (sample) - missing);
psample = &sample;
}
}
if (!perf_event_sample_ok (psample))
{
warning (_("Branch trace may be incomplete."));
break;
}
if (perf_event_skip_record (tinfo, &psample->bts))
continue;
/* We found a valid sample, so we can complete the current block. */
block.begin = psample->bts.to;
VEC_safe_push (btrace_block_s, btrace, &block);
/* Start the next block. */
block.end = psample->bts.from;
}
/* Push the last block (i.e. the first one of inferior execution), as well.
We don't know where it ends, but we know where it starts. If we're
reading delta trace, we can fill in the start address later on.
Otherwise we will prune it. */
block.begin = 0;
VEC_safe_push (btrace_block_s, btrace, &block);
return btrace;
}
/* Check whether the kernel supports BTS. */
static int
kernel_supports_bts (void)
{
struct perf_event_attr attr;
pid_t child, pid;
int status, file;
errno = 0;
child = fork ();
switch (child)
{
case -1:
warning (_("test bts: cannot fork: %s."), strerror (errno));
return 0;
case 0:
status = ptrace (PTRACE_TRACEME, 0, NULL, NULL);
if (status != 0)
{
warning (_("test bts: cannot PTRACE_TRACEME: %s."),
strerror (errno));
_exit (1);
}
status = raise (SIGTRAP);
if (status != 0)
{
warning (_("test bts: cannot raise SIGTRAP: %s."),
strerror (errno));
_exit (1);
}
_exit (1);
default:
pid = waitpid (child, &status, 0);
if (pid != child)
{
warning (_("test bts: bad pid %ld, error: %s."),
(long) pid, strerror (errno));
return 0;
}
if (!WIFSTOPPED (status))
{
warning (_("test bts: expected stop. status: %d."),
status);
return 0;
}
memset (&attr, 0, sizeof (attr));
attr.type = PERF_TYPE_HARDWARE;
attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
attr.sample_period = 1;
attr.sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_ADDR;
attr.exclude_kernel = 1;
attr.exclude_hv = 1;
attr.exclude_idle = 1;
file = syscall (SYS_perf_event_open, &attr, child, -1, -1, 0);
if (file >= 0)
close (file);
kill (child, SIGKILL);
ptrace (PTRACE_KILL, child, NULL, NULL);
pid = waitpid (child, &status, 0);
if (pid != child)
{
warning (_("test bts: bad pid %ld, error: %s."),
(long) pid, strerror (errno));
if (!WIFSIGNALED (status))
warning (_("test bts: expected killed. status: %d."),
status);
}
return (file >= 0);
}
}
/* Check whether an Intel cpu supports BTS. */
static int
intel_supports_bts (void)
{
unsigned int cpuid, model, family;
if (!x86_cpuid (1, &cpuid, NULL, NULL, NULL))
return 0;
family = (cpuid >> 8) & 0xf;
model = (cpuid >> 4) & 0xf;
switch (family)
{
case 0x6:
model += (cpuid >> 12) & 0xf0;
switch (model)
{
case 0x1a: /* Nehalem */
case 0x1f:
case 0x1e:
case 0x2e:
case 0x25: /* Westmere */
case 0x2c:
case 0x2f:
case 0x2a: /* Sandy Bridge */
case 0x2d:
case 0x3a: /* Ivy Bridge */
/* AAJ122: LBR, BTM, or BTS records may have incorrect branch
"from" information afer an EIST transition, T-states, C1E, or
Adaptive Thermal Throttling. */
return 0;
}
}
return 1;
}
/* Check whether the cpu supports BTS. */
static int
cpu_supports_bts (void)
{
unsigned int ebx, ecx, edx;
if (!x86_cpuid (0, NULL, &ebx, &ecx, &edx))
return 0;
if (ebx == signature_INTEL_ebx && ecx == signature_INTEL_ecx
&& edx == signature_INTEL_edx)
return intel_supports_bts ();
/* Don't know about others. Let's assume they do. */
return 1;
}
/* Check whether the linux target supports BTS. */
static int
linux_supports_bts (void)
{
static int cached;
if (cached == 0)
{
if (!kernel_supports_bts ())
cached = -1;
else if (!cpu_supports_bts ())
cached = -1;
else
cached = 1;
}
return cached > 0;
}
/* See linux-btrace.h. */
int
linux_supports_btrace (struct target_ops *ops, enum btrace_format format)
{
switch (format)
{
case BTRACE_FORMAT_NONE:
return 0;
case BTRACE_FORMAT_BTS:
return linux_supports_bts ();
}
internal_error (__FILE__, __LINE__, _("Unknown branch trace format"));
}
/* See linux-btrace.h. */
struct btrace_target_info *
linux_enable_btrace (ptid_t ptid)
{
struct perf_event_mmap_page *header;
struct btrace_target_info *tinfo;
int pid, pg;
tinfo = xzalloc (sizeof (*tinfo));
tinfo->ptid = ptid;
tinfo->attr.size = sizeof (tinfo->attr);
tinfo->attr.type = PERF_TYPE_HARDWARE;
tinfo->attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
tinfo->attr.sample_period = 1;
/* We sample from and to address. */
tinfo->attr.sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_ADDR;
tinfo->attr.exclude_kernel = 1;
tinfo->attr.exclude_hv = 1;
tinfo->attr.exclude_idle = 1;
tinfo->ptr_bits = 0;
pid = ptid_get_lwp (ptid);
if (pid == 0)
pid = ptid_get_pid (ptid);
errno = 0;
tinfo->file = syscall (SYS_perf_event_open, &tinfo->attr, pid, -1, -1, 0);
if (tinfo->file < 0)
goto err;
/* We try to allocate as much buffer as we can get.
We could allow the user to specify the size of the buffer, but then
we'd leave this search for the maximum buffer size to him. */
for (pg = 4; pg >= 0; --pg)
{
/* The number of pages we request needs to be a power of two. */
header = mmap (NULL, ((1 << pg) + 1) * PAGE_SIZE, PROT_READ, MAP_SHARED,
tinfo->file, 0);
if (header != MAP_FAILED)
break;
}
if (header == MAP_FAILED)
goto err_file;
tinfo->header = header;
tinfo->bts.mem = ((const uint8_t *) header) + PAGE_SIZE;
tinfo->bts.size = (1 << pg) * PAGE_SIZE;
tinfo->bts.data_head = &header->data_head;
tinfo->bts.last_head = 0;
return tinfo;
err_file:
/* We were not able to allocate any buffer. */
close (tinfo->file);
err:
xfree (tinfo);
return NULL;
}
/* See linux-btrace.h. */
enum btrace_error
linux_disable_btrace (struct btrace_target_info *tinfo)
{
munmap((void *) tinfo->header, tinfo->bts.size + PAGE_SIZE);
close (tinfo->file);
xfree (tinfo);
return BTRACE_ERR_NONE;
}
/* Read branch trace data in BTS format for the thread given by TINFO into
BTRACE using the TYPE reading method. */
static enum btrace_error
linux_read_bts (struct btrace_data_bts *btrace,
struct btrace_target_info *tinfo,
enum btrace_read_type type)
{
struct perf_event_buffer *pevent;
const uint8_t *begin, *end, *start;
unsigned long long data_head, data_tail, buffer_size, size;
unsigned int retries = 5;
pevent = &tinfo->bts;
/* For delta reads, we return at least the partial last block containing
the current PC. */
if (type == BTRACE_READ_NEW && !perf_event_new_data (pevent))
return BTRACE_ERR_NONE;
buffer_size = pevent->size;
data_tail = pevent->last_head;
/* We may need to retry reading the trace. See below. */
while (retries--)
{
data_head = *pevent->data_head;
/* Delete any leftover trace from the previous iteration. */
VEC_free (btrace_block_s, btrace->blocks);
if (type == BTRACE_READ_DELTA)
{
/* Determine the number of bytes to read and check for buffer
overflows. */
/* Check for data head overflows. We might be able to recover from
those but they are very unlikely and it's not really worth the
effort, I think. */
if (data_head < data_tail)
return BTRACE_ERR_OVERFLOW;
/* If the buffer is smaller than the trace delta, we overflowed. */
size = data_head - data_tail;
if (buffer_size < size)
return BTRACE_ERR_OVERFLOW;
}
else
{
/* Read the entire buffer. */
size = buffer_size;
/* Adjust the size if the buffer has not overflowed, yet. */
if (data_head < size)
size = data_head;
}
/* Data_head keeps growing; the buffer itself is circular. */
begin = pevent->mem;
start = begin + data_head % buffer_size;
if (data_head <= buffer_size)
end = start;
else
end = begin + pevent->size;
btrace->blocks = perf_event_read_bts (tinfo, begin, end, start, size);
/* The stopping thread notifies its ptracer before it is scheduled out.
On multi-core systems, the debugger might therefore run while the
kernel might be writing the last branch trace records.
Let's check whether the data head moved while we read the trace. */
if (data_head == *pevent->data_head)
break;
}
pevent->last_head = data_head;
/* Prune the incomplete last block (i.e. the first one of inferior execution)
if we're not doing a delta read. There is no way of filling in its zeroed
BEGIN element. */
if (!VEC_empty (btrace_block_s, btrace->blocks)
&& type != BTRACE_READ_DELTA)
VEC_pop (btrace_block_s, btrace->blocks);
return BTRACE_ERR_NONE;
}
/* See linux-btrace.h. */
enum btrace_error
linux_read_btrace (struct btrace_data *btrace,
struct btrace_target_info *tinfo,
enum btrace_read_type type)
{
/* We read btrace in BTS format. */
btrace->format = BTRACE_FORMAT_BTS;
btrace->variant.bts.blocks = NULL;
return linux_read_bts (&btrace->variant.bts, tinfo, type);
}
#else /* !HAVE_LINUX_PERF_EVENT_H */
/* See linux-btrace.h. */
int
linux_supports_btrace (struct target_ops *ops, enum btrace_format format)
{
return 0;
}
/* See linux-btrace.h. */
struct btrace_target_info *
linux_enable_btrace (ptid_t ptid)
{
return NULL;
}
/* See linux-btrace.h. */
enum btrace_error
linux_disable_btrace (struct btrace_target_info *tinfo)
{
return BTRACE_ERR_NOT_SUPPORTED;
}
/* See linux-btrace.h. */
enum btrace_error
linux_read_btrace (struct btrace_data *btrace,
struct btrace_target_info *tinfo,
enum btrace_read_type type)
{
return BTRACE_ERR_NOT_SUPPORTED;
}
#endif /* !HAVE_LINUX_PERF_EVENT_H */
|