Merge tag 'ntb-5.11' of git://github.com/jonmason/ntb
[linux/fpc-iii.git] / tools / perf / arch / x86 / util / intel-bts.c
blob4a76d49d25d606055648e9ed70a59d5239e70f1c
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 */
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>
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
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)
33 struct intel_bts_snapshot_ref {
34 void *ref_buf;
35 size_t ref_offset;
36 bool wrapped;
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;
49 struct branch {
50 u64 from;
51 u64 to;
52 u64 misc;
55 static size_t
56 intel_bts_info_priv_size(struct auxtrace_record *itr __maybe_unused,
57 struct evlist *evlist __maybe_unused)
59 return INTEL_BTS_AUXTRACE_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)
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;
75 if (priv_size != INTEL_BTS_AUXTRACE_PRIV_SIZE)
76 return -EINVAL;
78 if (!session->evlist->core.nr_mmaps)
79 return -EINVAL;
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;
90 if (!cap_user_time_zero)
91 ui__warning("Intel BTS: TSC not available\n");
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;
102 return 0;
105 static int intel_bts_recording_options(struct auxtrace_record *itr,
106 struct evlist *evlist,
107 struct record_opts *opts)
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);
116 if (opts->auxtrace_sample_mode) {
117 pr_err("Intel BTS does not support AUX area sampling\n");
118 return -EINVAL;
121 btsr->evlist = evlist;
122 btsr->snapshot_mode = opts->auxtrace_snapshot_mode;
124 evlist__for_each_entry(evlist, evsel) {
125 if (evsel->core.attr.type == intel_bts_pmu->type) {
126 if (intel_bts_evsel) {
127 pr_err("There may be only one " INTEL_BTS_PMU_NAME " event\n");
128 return -EINVAL;
130 evsel->core.attr.freq = 0;
131 evsel->core.attr.sample_period = 1;
132 intel_bts_evsel = evsel;
133 opts->full_auxtrace = true;
137 if (opts->auxtrace_snapshot_mode && !opts->full_auxtrace) {
138 pr_err("Snapshot mode (-S option) requires " INTEL_BTS_PMU_NAME " PMU event (-e " INTEL_BTS_PMU_NAME ")\n");
139 return -EINVAL;
142 if (!opts->full_auxtrace)
143 return 0;
145 if (opts->full_auxtrace && !perf_cpu_map__empty(cpus)) {
146 pr_err(INTEL_BTS_PMU_NAME " does not support per-cpu recording\n");
147 return -EINVAL;
150 /* Set default sizes for snapshot mode */
151 if (opts->auxtrace_snapshot_mode) {
152 if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) {
153 if (privileged) {
154 opts->auxtrace_mmap_pages = MiB(4) / page_size;
155 } else {
156 opts->auxtrace_mmap_pages = KiB(128) / page_size;
157 if (opts->mmap_pages == UINT_MAX)
158 opts->mmap_pages = KiB(256) / page_size;
160 } else if (!opts->auxtrace_mmap_pages && !privileged &&
161 opts->mmap_pages == UINT_MAX) {
162 opts->mmap_pages = KiB(256) / page_size;
164 if (!opts->auxtrace_snapshot_size)
165 opts->auxtrace_snapshot_size =
166 opts->auxtrace_mmap_pages * (size_t)page_size;
167 if (!opts->auxtrace_mmap_pages) {
168 size_t sz = opts->auxtrace_snapshot_size;
170 sz = round_up(sz, page_size) / page_size;
171 opts->auxtrace_mmap_pages = roundup_pow_of_two(sz);
173 if (opts->auxtrace_snapshot_size >
174 opts->auxtrace_mmap_pages * (size_t)page_size) {
175 pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n",
176 opts->auxtrace_snapshot_size,
177 opts->auxtrace_mmap_pages * (size_t)page_size);
178 return -EINVAL;
180 if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) {
181 pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n");
182 return -EINVAL;
184 pr_debug2("Intel BTS snapshot size: %zu\n",
185 opts->auxtrace_snapshot_size);
188 /* Set default sizes for full trace mode */
189 if (opts->full_auxtrace && !opts->auxtrace_mmap_pages) {
190 if (privileged) {
191 opts->auxtrace_mmap_pages = MiB(4) / page_size;
192 } else {
193 opts->auxtrace_mmap_pages = KiB(128) / page_size;
194 if (opts->mmap_pages == UINT_MAX)
195 opts->mmap_pages = KiB(256) / page_size;
199 /* Validate auxtrace_mmap_pages */
200 if (opts->auxtrace_mmap_pages) {
201 size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size;
202 size_t min_sz;
204 if (opts->auxtrace_snapshot_mode)
205 min_sz = KiB(4);
206 else
207 min_sz = KiB(8);
209 if (sz < min_sz || !is_power_of_2(sz)) {
210 pr_err("Invalid mmap size for Intel BTS: must be at least %zuKiB and a power of 2\n",
211 min_sz / 1024);
212 return -EINVAL;
216 if (intel_bts_evsel) {
218 * To obtain the auxtrace buffer file descriptor, the auxtrace event
219 * must come first.
221 evlist__to_front(evlist, intel_bts_evsel);
223 * In the case of per-cpu mmaps, we need the CPU on the
224 * AUX event.
226 if (!perf_cpu_map__empty(cpus))
227 evsel__set_sample_bit(intel_bts_evsel, CPU);
230 /* Add dummy event to keep tracking */
231 if (opts->full_auxtrace) {
232 struct evsel *tracking_evsel;
233 int err;
235 err = parse_events(evlist, "dummy:u", NULL);
236 if (err)
237 return err;
239 tracking_evsel = evlist__last(evlist);
241 evlist__set_tracking_event(evlist, tracking_evsel);
243 tracking_evsel->core.attr.freq = 0;
244 tracking_evsel->core.attr.sample_period = 1;
247 return 0;
250 static int intel_bts_parse_snapshot_options(struct auxtrace_record *itr,
251 struct record_opts *opts,
252 const char *str)
254 struct intel_bts_recording *btsr =
255 container_of(itr, struct intel_bts_recording, itr);
256 unsigned long long snapshot_size = 0;
257 char *endptr;
259 if (str) {
260 snapshot_size = strtoull(str, &endptr, 0);
261 if (*endptr || snapshot_size > SIZE_MAX)
262 return -1;
265 opts->auxtrace_snapshot_mode = true;
266 opts->auxtrace_snapshot_size = snapshot_size;
268 btsr->snapshot_size = snapshot_size;
270 return 0;
273 static u64 intel_bts_reference(struct auxtrace_record *itr __maybe_unused)
275 return rdtsc();
278 static int intel_bts_alloc_snapshot_refs(struct intel_bts_recording *btsr,
279 int idx)
281 const size_t sz = sizeof(struct intel_bts_snapshot_ref);
282 int cnt = btsr->snapshot_ref_cnt, new_cnt = cnt * 2;
283 struct intel_bts_snapshot_ref *refs;
285 if (!new_cnt)
286 new_cnt = 16;
288 while (new_cnt <= idx)
289 new_cnt *= 2;
291 refs = calloc(new_cnt, sz);
292 if (!refs)
293 return -ENOMEM;
295 memcpy(refs, btsr->snapshot_refs, cnt * sz);
297 btsr->snapshot_refs = refs;
298 btsr->snapshot_ref_cnt = new_cnt;
300 return 0;
303 static void intel_bts_free_snapshot_refs(struct intel_bts_recording *btsr)
305 int i;
307 for (i = 0; i < btsr->snapshot_ref_cnt; i++)
308 zfree(&btsr->snapshot_refs[i].ref_buf);
309 zfree(&btsr->snapshot_refs);
312 static void intel_bts_recording_free(struct auxtrace_record *itr)
314 struct intel_bts_recording *btsr =
315 container_of(itr, struct intel_bts_recording, itr);
317 intel_bts_free_snapshot_refs(btsr);
318 free(btsr);
321 static int intel_bts_snapshot_start(struct auxtrace_record *itr)
323 struct intel_bts_recording *btsr =
324 container_of(itr, struct intel_bts_recording, itr);
325 struct evsel *evsel;
327 evlist__for_each_entry(btsr->evlist, evsel) {
328 if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
329 return evsel__disable(evsel);
331 return -EINVAL;
334 static int intel_bts_snapshot_finish(struct auxtrace_record *itr)
336 struct intel_bts_recording *btsr =
337 container_of(itr, struct intel_bts_recording, itr);
338 struct evsel *evsel;
340 evlist__for_each_entry(btsr->evlist, evsel) {
341 if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
342 return evsel__enable(evsel);
344 return -EINVAL;
347 static bool intel_bts_first_wrap(u64 *data, size_t buf_size)
349 int i, a, b;
351 b = buf_size >> 3;
352 a = b - 512;
353 if (a < 0)
354 a = 0;
356 for (i = a; i < b; i++) {
357 if (data[i])
358 return true;
361 return false;
364 static int intel_bts_find_snapshot(struct auxtrace_record *itr, int idx,
365 struct auxtrace_mmap *mm, unsigned char *data,
366 u64 *head, u64 *old)
368 struct intel_bts_recording *btsr =
369 container_of(itr, struct intel_bts_recording, itr);
370 bool wrapped;
371 int err;
373 pr_debug3("%s: mmap index %d old head %zu new head %zu\n",
374 __func__, idx, (size_t)*old, (size_t)*head);
376 if (idx >= btsr->snapshot_ref_cnt) {
377 err = intel_bts_alloc_snapshot_refs(btsr, idx);
378 if (err)
379 goto out_err;
382 wrapped = btsr->snapshot_refs[idx].wrapped;
383 if (!wrapped && intel_bts_first_wrap((u64 *)data, mm->len)) {
384 btsr->snapshot_refs[idx].wrapped = true;
385 wrapped = true;
389 * In full trace mode 'head' continually increases. However in snapshot
390 * mode 'head' is an offset within the buffer. Here 'old' and 'head'
391 * are adjusted to match the full trace case which expects that 'old' is
392 * always less than 'head'.
394 if (wrapped) {
395 *old = *head;
396 *head += mm->len;
397 } else {
398 if (mm->mask)
399 *old &= mm->mask;
400 else
401 *old %= mm->len;
402 if (*old > *head)
403 *head += mm->len;
406 pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n",
407 __func__, wrapped ? "" : "not ", (size_t)*old, (size_t)*head);
409 return 0;
411 out_err:
412 pr_err("%s: failed, error %d\n", __func__, err);
413 return err;
416 struct auxtrace_record *intel_bts_recording_init(int *err)
418 struct perf_pmu *intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
419 struct intel_bts_recording *btsr;
421 if (!intel_bts_pmu)
422 return NULL;
424 if (setenv("JITDUMP_USE_ARCH_TIMESTAMP", "1", 1)) {
425 *err = -errno;
426 return NULL;
429 btsr = zalloc(sizeof(struct intel_bts_recording));
430 if (!btsr) {
431 *err = -ENOMEM;
432 return NULL;
435 btsr->intel_bts_pmu = intel_bts_pmu;
436 btsr->itr.pmu = intel_bts_pmu;
437 btsr->itr.recording_options = intel_bts_recording_options;
438 btsr->itr.info_priv_size = intel_bts_info_priv_size;
439 btsr->itr.info_fill = intel_bts_info_fill;
440 btsr->itr.free = intel_bts_recording_free;
441 btsr->itr.snapshot_start = intel_bts_snapshot_start;
442 btsr->itr.snapshot_finish = intel_bts_snapshot_finish;
443 btsr->itr.find_snapshot = intel_bts_find_snapshot;
444 btsr->itr.parse_snapshot_options = intel_bts_parse_snapshot_options;
445 btsr->itr.reference = intel_bts_reference;
446 btsr->itr.read_finish = auxtrace_record__read_finish;
447 btsr->itr.alignment = sizeof(struct branch);
448 return &btsr->itr;