2 * intel-bts.c: Intel Processor Trace support
3 * Copyright (c) 2013-2015, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/bitops.h>
19 #include <linux/log2.h>
21 #include "../../util/cpumap.h"
22 #include "../../util/evsel.h"
23 #include "../../util/evlist.h"
24 #include "../../util/session.h"
25 #include "../../util/util.h"
26 #include "../../util/pmu.h"
27 #include "../../util/debug.h"
28 #include "../../util/tsc.h"
29 #include "../../util/auxtrace.h"
30 #include "../../util/intel-bts.h"
32 #define KiB(x) ((x) * 1024)
33 #define MiB(x) ((x) * 1024 * 1024)
34 #define KiB_MASK(x) (KiB(x) - 1)
35 #define MiB_MASK(x) (MiB(x) - 1)
37 #define INTEL_BTS_DFLT_SAMPLE_SIZE KiB(4)
39 #define INTEL_BTS_MAX_SAMPLE_SIZE KiB(60)
41 struct intel_bts_snapshot_ref {
47 struct intel_bts_recording {
48 struct auxtrace_record itr;
49 struct perf_pmu *intel_bts_pmu;
50 struct perf_evlist *evlist;
54 struct intel_bts_snapshot_ref *snapshot_refs;
64 intel_bts_info_priv_size(struct auxtrace_record *itr __maybe_unused,
65 struct perf_evlist *evlist __maybe_unused)
67 return INTEL_BTS_AUXTRACE_PRIV_SIZE;
70 static int intel_bts_info_fill(struct auxtrace_record *itr,
71 struct perf_session *session,
72 struct auxtrace_info_event *auxtrace_info,
75 struct intel_bts_recording *btsr =
76 container_of(itr, struct intel_bts_recording, itr);
77 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
78 struct perf_event_mmap_page *pc;
79 struct perf_tsc_conversion tc = { .time_mult = 0, };
80 bool cap_user_time_zero = false;
83 if (priv_size != INTEL_BTS_AUXTRACE_PRIV_SIZE)
86 if (!session->evlist->nr_mmaps)
89 pc = session->evlist->mmap[0].base;
91 err = perf_read_tsc_conversion(pc, &tc);
93 if (err != -EOPNOTSUPP)
96 cap_user_time_zero = tc.time_mult != 0;
98 if (!cap_user_time_zero)
99 ui__warning("Intel BTS: TSC not available\n");
102 auxtrace_info->type = PERF_AUXTRACE_INTEL_BTS;
103 auxtrace_info->priv[INTEL_BTS_PMU_TYPE] = intel_bts_pmu->type;
104 auxtrace_info->priv[INTEL_BTS_TIME_SHIFT] = tc.time_shift;
105 auxtrace_info->priv[INTEL_BTS_TIME_MULT] = tc.time_mult;
106 auxtrace_info->priv[INTEL_BTS_TIME_ZERO] = tc.time_zero;
107 auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO] = cap_user_time_zero;
108 auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE] = btsr->snapshot_mode;
113 static int intel_bts_recording_options(struct auxtrace_record *itr,
114 struct perf_evlist *evlist,
115 struct record_opts *opts)
117 struct intel_bts_recording *btsr =
118 container_of(itr, struct intel_bts_recording, itr);
119 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
120 struct perf_evsel *evsel, *intel_bts_evsel = NULL;
121 const struct cpu_map *cpus = evlist->cpus;
122 bool privileged = geteuid() == 0 || perf_event_paranoid() < 0;
124 btsr->evlist = evlist;
125 btsr->snapshot_mode = opts->auxtrace_snapshot_mode;
127 evlist__for_each(evlist, evsel) {
128 if (evsel->attr.type == intel_bts_pmu->type) {
129 if (intel_bts_evsel) {
130 pr_err("There may be only one " INTEL_BTS_PMU_NAME " event\n");
133 evsel->attr.freq = 0;
134 evsel->attr.sample_period = 1;
135 intel_bts_evsel = evsel;
136 opts->full_auxtrace = true;
140 if (opts->auxtrace_snapshot_mode && !opts->full_auxtrace) {
141 pr_err("Snapshot mode (-S option) requires " INTEL_BTS_PMU_NAME " PMU event (-e " INTEL_BTS_PMU_NAME ")\n");
145 if (!opts->full_auxtrace)
148 if (opts->full_auxtrace && !cpu_map__empty(cpus)) {
149 pr_err(INTEL_BTS_PMU_NAME " does not support per-cpu recording\n");
153 /* Set default sizes for snapshot mode */
154 if (opts->auxtrace_snapshot_mode) {
155 if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) {
157 opts->auxtrace_mmap_pages = MiB(4) / page_size;
159 opts->auxtrace_mmap_pages = KiB(128) / page_size;
160 if (opts->mmap_pages == UINT_MAX)
161 opts->mmap_pages = KiB(256) / page_size;
163 } else if (!opts->auxtrace_mmap_pages && !privileged &&
164 opts->mmap_pages == UINT_MAX) {
165 opts->mmap_pages = KiB(256) / page_size;
167 if (!opts->auxtrace_snapshot_size)
168 opts->auxtrace_snapshot_size =
169 opts->auxtrace_mmap_pages * (size_t)page_size;
170 if (!opts->auxtrace_mmap_pages) {
171 size_t sz = opts->auxtrace_snapshot_size;
173 sz = round_up(sz, page_size) / page_size;
174 opts->auxtrace_mmap_pages = roundup_pow_of_two(sz);
176 if (opts->auxtrace_snapshot_size >
177 opts->auxtrace_mmap_pages * (size_t)page_size) {
178 pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n",
179 opts->auxtrace_snapshot_size,
180 opts->auxtrace_mmap_pages * (size_t)page_size);
183 if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) {
184 pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n");
187 pr_debug2("Intel BTS snapshot size: %zu\n",
188 opts->auxtrace_snapshot_size);
191 /* Set default sizes for full trace mode */
192 if (opts->full_auxtrace && !opts->auxtrace_mmap_pages) {
194 opts->auxtrace_mmap_pages = MiB(4) / page_size;
196 opts->auxtrace_mmap_pages = KiB(128) / page_size;
197 if (opts->mmap_pages == UINT_MAX)
198 opts->mmap_pages = KiB(256) / page_size;
202 /* Validate auxtrace_mmap_pages */
203 if (opts->auxtrace_mmap_pages) {
204 size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size;
207 if (opts->auxtrace_snapshot_mode)
212 if (sz < min_sz || !is_power_of_2(sz)) {
213 pr_err("Invalid mmap size for Intel BTS: must be at least %zuKiB and a power of 2\n",
219 if (intel_bts_evsel) {
221 * To obtain the auxtrace buffer file descriptor, the auxtrace event
224 perf_evlist__to_front(evlist, intel_bts_evsel);
226 * In the case of per-cpu mmaps, we need the CPU on the
229 if (!cpu_map__empty(cpus))
230 perf_evsel__set_sample_bit(intel_bts_evsel, CPU);
233 /* Add dummy event to keep tracking */
234 if (opts->full_auxtrace) {
235 struct perf_evsel *tracking_evsel;
238 err = parse_events(evlist, "dummy:u", NULL);
242 tracking_evsel = perf_evlist__last(evlist);
244 perf_evlist__set_tracking_event(evlist, tracking_evsel);
246 tracking_evsel->attr.freq = 0;
247 tracking_evsel->attr.sample_period = 1;
253 static int intel_bts_parse_snapshot_options(struct auxtrace_record *itr,
254 struct record_opts *opts,
257 struct intel_bts_recording *btsr =
258 container_of(itr, struct intel_bts_recording, itr);
259 unsigned long long snapshot_size = 0;
263 snapshot_size = strtoull(str, &endptr, 0);
264 if (*endptr || snapshot_size > SIZE_MAX)
268 opts->auxtrace_snapshot_mode = true;
269 opts->auxtrace_snapshot_size = snapshot_size;
271 btsr->snapshot_size = snapshot_size;
276 static u64 intel_bts_reference(struct auxtrace_record *itr __maybe_unused)
281 static int intel_bts_alloc_snapshot_refs(struct intel_bts_recording *btsr,
284 const size_t sz = sizeof(struct intel_bts_snapshot_ref);
285 int cnt = btsr->snapshot_ref_cnt, new_cnt = cnt * 2;
286 struct intel_bts_snapshot_ref *refs;
291 while (new_cnt <= idx)
294 refs = calloc(new_cnt, sz);
298 memcpy(refs, btsr->snapshot_refs, cnt * sz);
300 btsr->snapshot_refs = refs;
301 btsr->snapshot_ref_cnt = new_cnt;
306 static void intel_bts_free_snapshot_refs(struct intel_bts_recording *btsr)
310 for (i = 0; i < btsr->snapshot_ref_cnt; i++)
311 zfree(&btsr->snapshot_refs[i].ref_buf);
312 zfree(&btsr->snapshot_refs);
315 static void intel_bts_recording_free(struct auxtrace_record *itr)
317 struct intel_bts_recording *btsr =
318 container_of(itr, struct intel_bts_recording, itr);
320 intel_bts_free_snapshot_refs(btsr);
324 static int intel_bts_snapshot_start(struct auxtrace_record *itr)
326 struct intel_bts_recording *btsr =
327 container_of(itr, struct intel_bts_recording, itr);
328 struct perf_evsel *evsel;
330 evlist__for_each(btsr->evlist, evsel) {
331 if (evsel->attr.type == btsr->intel_bts_pmu->type)
332 return perf_evsel__disable(evsel);
337 static int intel_bts_snapshot_finish(struct auxtrace_record *itr)
339 struct intel_bts_recording *btsr =
340 container_of(itr, struct intel_bts_recording, itr);
341 struct perf_evsel *evsel;
343 evlist__for_each(btsr->evlist, evsel) {
344 if (evsel->attr.type == btsr->intel_bts_pmu->type)
345 return perf_evsel__enable(evsel);
350 static bool intel_bts_first_wrap(u64 *data, size_t buf_size)
359 for (i = a; i < b; i++) {
367 static int intel_bts_find_snapshot(struct auxtrace_record *itr, int idx,
368 struct auxtrace_mmap *mm, unsigned char *data,
371 struct intel_bts_recording *btsr =
372 container_of(itr, struct intel_bts_recording, itr);
376 pr_debug3("%s: mmap index %d old head %zu new head %zu\n",
377 __func__, idx, (size_t)*old, (size_t)*head);
379 if (idx >= btsr->snapshot_ref_cnt) {
380 err = intel_bts_alloc_snapshot_refs(btsr, idx);
385 wrapped = btsr->snapshot_refs[idx].wrapped;
386 if (!wrapped && intel_bts_first_wrap((u64 *)data, mm->len)) {
387 btsr->snapshot_refs[idx].wrapped = true;
392 * In full trace mode 'head' continually increases. However in snapshot
393 * mode 'head' is an offset within the buffer. Here 'old' and 'head'
394 * are adjusted to match the full trace case which expects that 'old' is
395 * always less than 'head'.
409 pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n",
410 __func__, wrapped ? "" : "not ", (size_t)*old, (size_t)*head);
415 pr_err("%s: failed, error %d\n", __func__, err);
419 static int intel_bts_read_finish(struct auxtrace_record *itr, int idx)
421 struct intel_bts_recording *btsr =
422 container_of(itr, struct intel_bts_recording, itr);
423 struct perf_evsel *evsel;
425 evlist__for_each(btsr->evlist, evsel) {
426 if (evsel->attr.type == btsr->intel_bts_pmu->type)
427 return perf_evlist__enable_event_idx(btsr->evlist,
433 struct auxtrace_record *intel_bts_recording_init(int *err)
435 struct perf_pmu *intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
436 struct intel_bts_recording *btsr;
441 if (setenv("JITDUMP_USE_ARCH_TIMESTAMP", "1", 1)) {
446 btsr = zalloc(sizeof(struct intel_bts_recording));
452 btsr->intel_bts_pmu = intel_bts_pmu;
453 btsr->itr.recording_options = intel_bts_recording_options;
454 btsr->itr.info_priv_size = intel_bts_info_priv_size;
455 btsr->itr.info_fill = intel_bts_info_fill;
456 btsr->itr.free = intel_bts_recording_free;
457 btsr->itr.snapshot_start = intel_bts_snapshot_start;
458 btsr->itr.snapshot_finish = intel_bts_snapshot_finish;
459 btsr->itr.find_snapshot = intel_bts_find_snapshot;
460 btsr->itr.parse_snapshot_options = intel_bts_parse_snapshot_options;
461 btsr->itr.reference = intel_bts_reference;
462 btsr->itr.read_finish = intel_bts_read_finish;
463 btsr->itr.alignment = sizeof(struct branch);