1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * intel-bts.c: Intel Processor Trace support
4 * Copyright (c) 2013-2015, Intel Corporation.
5 */
6
7 #include <errno.h>
8 #include <linux/kernel.h>
9 #include <linux/types.h>
10 #include <linux/bitops.h>
11 #include <linux/log2.h>
12 #include <linux/zalloc.h>
13
14 #include "../../util/cpumap.h"
15 #include "../../util/event.h"
16 #include "../../util/evsel.h"
17 #include "../../util/evlist.h"
18 #include "../../util/mmap.h"
19 #include "../../util/session.h"
20 #include "../../util/pmu.h"
21 #include "../../util/debug.h"
22 #include "../../util/record.h"
23 #include "../../util/tsc.h"
24 #include "../../util/auxtrace.h"
25 #include "../../util/intel-bts.h"
26 #include <internal/lib.h> // page_size
27
28 #define KiB(x) ((x) * 1024)
29 #define MiB(x) ((x) * 1024 * 1024)
30 #define KiB_MASK(x) (KiB(x) - 1)
31 #define MiB_MASK(x) (MiB(x) - 1)
32
33 struct intel_bts_snapshot_ref {
34 void *ref_buf;
35 size_t ref_offset;
36 bool wrapped;
37 };
38
39 struct intel_bts_recording {
40 struct auxtrace_record itr;
41 struct perf_pmu *intel_bts_pmu;
42 struct evlist *evlist;
43 bool snapshot_mode;
44 size_t snapshot_size;
45 int snapshot_ref_cnt;
46 struct intel_bts_snapshot_ref *snapshot_refs;
47 };
48
49 struct branch {
50 u64 from;
51 u64 to;
52 u64 misc;
53 };
54
55 static size_t
intel_bts_info_priv_size(struct auxtrace_record * itr __maybe_unused,struct evlist * evlist __maybe_unused)56 intel_bts_info_priv_size(struct auxtrace_record *itr __maybe_unused,
57 struct evlist *evlist __maybe_unused)
58 {
59 return INTEL_BTS_AUXTRACE_PRIV_SIZE;
60 }
61
intel_bts_info_fill(struct auxtrace_record * itr,struct perf_session * session,struct perf_record_auxtrace_info * auxtrace_info,size_t priv_size)62 static int intel_bts_info_fill(struct auxtrace_record *itr,
63 struct perf_session *session,
64 struct perf_record_auxtrace_info *auxtrace_info,
65 size_t priv_size)
66 {
67 struct intel_bts_recording *btsr =
68 container_of(itr, struct intel_bts_recording, itr);
69 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
70 struct perf_event_mmap_page *pc;
71 struct perf_tsc_conversion tc = { .time_mult = 0, };
72 bool cap_user_time_zero = false;
73 int err;
74
75 if (priv_size != INTEL_BTS_AUXTRACE_PRIV_SIZE)
76 return -EINVAL;
77
78 if (!session->evlist->core.nr_mmaps)
79 return -EINVAL;
80
81 pc = session->evlist->mmap[0].core.base;
82 if (pc) {
83 err = perf_read_tsc_conversion(pc, &tc);
84 if (err) {
85 if (err != -EOPNOTSUPP)
86 return err;
87 } else {
88 cap_user_time_zero = tc.time_mult != 0;
89 }
90 if (!cap_user_time_zero)
91 ui__warning("Intel BTS: TSC not available\n");
92 }
93
94 auxtrace_info->type = PERF_AUXTRACE_INTEL_BTS;
95 auxtrace_info->priv[INTEL_BTS_PMU_TYPE] = intel_bts_pmu->type;
96 auxtrace_info->priv[INTEL_BTS_TIME_SHIFT] = tc.time_shift;
97 auxtrace_info->priv[INTEL_BTS_TIME_MULT] = tc.time_mult;
98 auxtrace_info->priv[INTEL_BTS_TIME_ZERO] = tc.time_zero;
99 auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO] = cap_user_time_zero;
100 auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE] = btsr->snapshot_mode;
101
102 return 0;
103 }
104
intel_bts_recording_options(struct auxtrace_record * itr,struct evlist * evlist,struct record_opts * opts)105 static int intel_bts_recording_options(struct auxtrace_record *itr,
106 struct evlist *evlist,
107 struct record_opts *opts)
108 {
109 struct intel_bts_recording *btsr =
110 container_of(itr, struct intel_bts_recording, itr);
111 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
112 struct evsel *evsel, *intel_bts_evsel = NULL;
113 const struct perf_cpu_map *cpus = evlist->core.cpus;
114 bool privileged = perf_event_paranoid_check(-1);
115
116 btsr->evlist = evlist;
117 btsr->snapshot_mode = opts->auxtrace_snapshot_mode;
118
119 evlist__for_each_entry(evlist, evsel) {
120 if (evsel->core.attr.type == intel_bts_pmu->type) {
121 if (intel_bts_evsel) {
122 pr_err("There may be only one " INTEL_BTS_PMU_NAME " event\n");
123 return -EINVAL;
124 }
125 evsel->core.attr.freq = 0;
126 evsel->core.attr.sample_period = 1;
127 intel_bts_evsel = evsel;
128 opts->full_auxtrace = true;
129 }
130 }
131
132 if (opts->auxtrace_snapshot_mode && !opts->full_auxtrace) {
133 pr_err("Snapshot mode (-S option) requires " INTEL_BTS_PMU_NAME " PMU event (-e " INTEL_BTS_PMU_NAME ")\n");
134 return -EINVAL;
135 }
136
137 if (!opts->full_auxtrace)
138 return 0;
139
140 if (opts->full_auxtrace && !perf_cpu_map__empty(cpus)) {
141 pr_err(INTEL_BTS_PMU_NAME " does not support per-cpu recording\n");
142 return -EINVAL;
143 }
144
145 /* Set default sizes for snapshot mode */
146 if (opts->auxtrace_snapshot_mode) {
147 if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) {
148 if (privileged) {
149 opts->auxtrace_mmap_pages = MiB(4) / page_size;
150 } else {
151 opts->auxtrace_mmap_pages = KiB(128) / page_size;
152 if (opts->mmap_pages == UINT_MAX)
153 opts->mmap_pages = KiB(256) / page_size;
154 }
155 } else if (!opts->auxtrace_mmap_pages && !privileged &&
156 opts->mmap_pages == UINT_MAX) {
157 opts->mmap_pages = KiB(256) / page_size;
158 }
159 if (!opts->auxtrace_snapshot_size)
160 opts->auxtrace_snapshot_size =
161 opts->auxtrace_mmap_pages * (size_t)page_size;
162 if (!opts->auxtrace_mmap_pages) {
163 size_t sz = opts->auxtrace_snapshot_size;
164
165 sz = round_up(sz, page_size) / page_size;
166 opts->auxtrace_mmap_pages = roundup_pow_of_two(sz);
167 }
168 if (opts->auxtrace_snapshot_size >
169 opts->auxtrace_mmap_pages * (size_t)page_size) {
170 pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n",
171 opts->auxtrace_snapshot_size,
172 opts->auxtrace_mmap_pages * (size_t)page_size);
173 return -EINVAL;
174 }
175 if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) {
176 pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n");
177 return -EINVAL;
178 }
179 pr_debug2("Intel BTS snapshot size: %zu\n",
180 opts->auxtrace_snapshot_size);
181 }
182
183 /* Set default sizes for full trace mode */
184 if (opts->full_auxtrace && !opts->auxtrace_mmap_pages) {
185 if (privileged) {
186 opts->auxtrace_mmap_pages = MiB(4) / page_size;
187 } else {
188 opts->auxtrace_mmap_pages = KiB(128) / page_size;
189 if (opts->mmap_pages == UINT_MAX)
190 opts->mmap_pages = KiB(256) / page_size;
191 }
192 }
193
194 /* Validate auxtrace_mmap_pages */
195 if (opts->auxtrace_mmap_pages) {
196 size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size;
197 size_t min_sz;
198
199 if (opts->auxtrace_snapshot_mode)
200 min_sz = KiB(4);
201 else
202 min_sz = KiB(8);
203
204 if (sz < min_sz || !is_power_of_2(sz)) {
205 pr_err("Invalid mmap size for Intel BTS: must be at least %zuKiB and a power of 2\n",
206 min_sz / 1024);
207 return -EINVAL;
208 }
209 }
210
211 if (intel_bts_evsel) {
212 /*
213 * To obtain the auxtrace buffer file descriptor, the auxtrace event
214 * must come first.
215 */
216 perf_evlist__to_front(evlist, intel_bts_evsel);
217 /*
218 * In the case of per-cpu mmaps, we need the CPU on the
219 * AUX event.
220 */
221 if (!perf_cpu_map__empty(cpus))
222 perf_evsel__set_sample_bit(intel_bts_evsel, CPU);
223 }
224
225 /* Add dummy event to keep tracking */
226 if (opts->full_auxtrace) {
227 struct evsel *tracking_evsel;
228 int err;
229
230 err = parse_events(evlist, "dummy:u", NULL);
231 if (err)
232 return err;
233
234 tracking_evsel = evlist__last(evlist);
235
236 perf_evlist__set_tracking_event(evlist, tracking_evsel);
237
238 tracking_evsel->core.attr.freq = 0;
239 tracking_evsel->core.attr.sample_period = 1;
240 }
241
242 return 0;
243 }
244
intel_bts_parse_snapshot_options(struct auxtrace_record * itr,struct record_opts * opts,const char * str)245 static int intel_bts_parse_snapshot_options(struct auxtrace_record *itr,
246 struct record_opts *opts,
247 const char *str)
248 {
249 struct intel_bts_recording *btsr =
250 container_of(itr, struct intel_bts_recording, itr);
251 unsigned long long snapshot_size = 0;
252 char *endptr;
253
254 if (str) {
255 snapshot_size = strtoull(str, &endptr, 0);
256 if (*endptr || snapshot_size > SIZE_MAX)
257 return -1;
258 }
259
260 opts->auxtrace_snapshot_mode = true;
261 opts->auxtrace_snapshot_size = snapshot_size;
262
263 btsr->snapshot_size = snapshot_size;
264
265 return 0;
266 }
267
intel_bts_reference(struct auxtrace_record * itr __maybe_unused)268 static u64 intel_bts_reference(struct auxtrace_record *itr __maybe_unused)
269 {
270 return rdtsc();
271 }
272
intel_bts_alloc_snapshot_refs(struct intel_bts_recording * btsr,int idx)273 static int intel_bts_alloc_snapshot_refs(struct intel_bts_recording *btsr,
274 int idx)
275 {
276 const size_t sz = sizeof(struct intel_bts_snapshot_ref);
277 int cnt = btsr->snapshot_ref_cnt, new_cnt = cnt * 2;
278 struct intel_bts_snapshot_ref *refs;
279
280 if (!new_cnt)
281 new_cnt = 16;
282
283 while (new_cnt <= idx)
284 new_cnt *= 2;
285
286 refs = calloc(new_cnt, sz);
287 if (!refs)
288 return -ENOMEM;
289
290 memcpy(refs, btsr->snapshot_refs, cnt * sz);
291
292 btsr->snapshot_refs = refs;
293 btsr->snapshot_ref_cnt = new_cnt;
294
295 return 0;
296 }
297
intel_bts_free_snapshot_refs(struct intel_bts_recording * btsr)298 static void intel_bts_free_snapshot_refs(struct intel_bts_recording *btsr)
299 {
300 int i;
301
302 for (i = 0; i < btsr->snapshot_ref_cnt; i++)
303 zfree(&btsr->snapshot_refs[i].ref_buf);
304 zfree(&btsr->snapshot_refs);
305 }
306
intel_bts_recording_free(struct auxtrace_record * itr)307 static void intel_bts_recording_free(struct auxtrace_record *itr)
308 {
309 struct intel_bts_recording *btsr =
310 container_of(itr, struct intel_bts_recording, itr);
311
312 intel_bts_free_snapshot_refs(btsr);
313 free(btsr);
314 }
315
intel_bts_snapshot_start(struct auxtrace_record * itr)316 static int intel_bts_snapshot_start(struct auxtrace_record *itr)
317 {
318 struct intel_bts_recording *btsr =
319 container_of(itr, struct intel_bts_recording, itr);
320 struct evsel *evsel;
321
322 evlist__for_each_entry(btsr->evlist, evsel) {
323 if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
324 return evsel__disable(evsel);
325 }
326 return -EINVAL;
327 }
328
intel_bts_snapshot_finish(struct auxtrace_record * itr)329 static int intel_bts_snapshot_finish(struct auxtrace_record *itr)
330 {
331 struct intel_bts_recording *btsr =
332 container_of(itr, struct intel_bts_recording, itr);
333 struct evsel *evsel;
334
335 evlist__for_each_entry(btsr->evlist, evsel) {
336 if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
337 return evsel__enable(evsel);
338 }
339 return -EINVAL;
340 }
341
intel_bts_first_wrap(u64 * data,size_t buf_size)342 static bool intel_bts_first_wrap(u64 *data, size_t buf_size)
343 {
344 int i, a, b;
345
346 b = buf_size >> 3;
347 a = b - 512;
348 if (a < 0)
349 a = 0;
350
351 for (i = a; i < b; i++) {
352 if (data[i])
353 return true;
354 }
355
356 return false;
357 }
358
intel_bts_find_snapshot(struct auxtrace_record * itr,int idx,struct auxtrace_mmap * mm,unsigned char * data,u64 * head,u64 * old)359 static int intel_bts_find_snapshot(struct auxtrace_record *itr, int idx,
360 struct auxtrace_mmap *mm, unsigned char *data,
361 u64 *head, u64 *old)
362 {
363 struct intel_bts_recording *btsr =
364 container_of(itr, struct intel_bts_recording, itr);
365 bool wrapped;
366 int err;
367
368 pr_debug3("%s: mmap index %d old head %zu new head %zu\n",
369 __func__, idx, (size_t)*old, (size_t)*head);
370
371 if (idx >= btsr->snapshot_ref_cnt) {
372 err = intel_bts_alloc_snapshot_refs(btsr, idx);
373 if (err)
374 goto out_err;
375 }
376
377 wrapped = btsr->snapshot_refs[idx].wrapped;
378 if (!wrapped && intel_bts_first_wrap((u64 *)data, mm->len)) {
379 btsr->snapshot_refs[idx].wrapped = true;
380 wrapped = true;
381 }
382
383 /*
384 * In full trace mode 'head' continually increases. However in snapshot
385 * mode 'head' is an offset within the buffer. Here 'old' and 'head'
386 * are adjusted to match the full trace case which expects that 'old' is
387 * always less than 'head'.
388 */
389 if (wrapped) {
390 *old = *head;
391 *head += mm->len;
392 } else {
393 if (mm->mask)
394 *old &= mm->mask;
395 else
396 *old %= mm->len;
397 if (*old > *head)
398 *head += mm->len;
399 }
400
401 pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n",
402 __func__, wrapped ? "" : "not ", (size_t)*old, (size_t)*head);
403
404 return 0;
405
406 out_err:
407 pr_err("%s: failed, error %d\n", __func__, err);
408 return err;
409 }
410
intel_bts_read_finish(struct auxtrace_record * itr,int idx)411 static int intel_bts_read_finish(struct auxtrace_record *itr, int idx)
412 {
413 struct intel_bts_recording *btsr =
414 container_of(itr, struct intel_bts_recording, itr);
415 struct evsel *evsel;
416
417 evlist__for_each_entry(btsr->evlist, evsel) {
418 if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
419 return perf_evlist__enable_event_idx(btsr->evlist,
420 evsel, idx);
421 }
422 return -EINVAL;
423 }
424
intel_bts_recording_init(int * err)425 struct auxtrace_record *intel_bts_recording_init(int *err)
426 {
427 struct perf_pmu *intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
428 struct intel_bts_recording *btsr;
429
430 if (!intel_bts_pmu)
431 return NULL;
432
433 if (setenv("JITDUMP_USE_ARCH_TIMESTAMP", "1", 1)) {
434 *err = -errno;
435 return NULL;
436 }
437
438 btsr = zalloc(sizeof(struct intel_bts_recording));
439 if (!btsr) {
440 *err = -ENOMEM;
441 return NULL;
442 }
443
444 btsr->intel_bts_pmu = intel_bts_pmu;
445 btsr->itr.recording_options = intel_bts_recording_options;
446 btsr->itr.info_priv_size = intel_bts_info_priv_size;
447 btsr->itr.info_fill = intel_bts_info_fill;
448 btsr->itr.free = intel_bts_recording_free;
449 btsr->itr.snapshot_start = intel_bts_snapshot_start;
450 btsr->itr.snapshot_finish = intel_bts_snapshot_finish;
451 btsr->itr.find_snapshot = intel_bts_find_snapshot;
452 btsr->itr.parse_snapshot_options = intel_bts_parse_snapshot_options;
453 btsr->itr.reference = intel_bts_reference;
454 btsr->itr.read_finish = intel_bts_read_finish;
455 btsr->itr.alignment = sizeof(struct branch);
456 return &btsr->itr;
457 }
458
