x86/util/intel-bts.c

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * intel-bts.c: Intel Processor Trace support
0004  * Copyright (c) 2013-2015, Intel Corporation.
0005  */
0006
0007 #include <errno.h>
0008 #include <linux/kernel.h>
0009 #include <linux/types.h>
0010 #include <linux/bitops.h>
0011 #include <linux/log2.h>
0012 #include <linux/zalloc.h>
0013
0014 #include "../../../util/cpumap.h"
0015 #include "../../../util/event.h"
0016 #include "../../../util/evsel.h"
0017 #include "../../../util/evlist.h"
0018 #include "../../../util/mmap.h"
0019 #include "../../../util/session.h"
0020 #include "../../../util/pmu.h"
0021 #include "../../../util/debug.h"
0022 #include "../../../util/record.h"
0023 #include "../../../util/tsc.h"
0024 #include "../../../util/auxtrace.h"
0025 #include "../../../util/intel-bts.h"
0026 #include <internal/lib.h> // page_size
0027
0028 #define KiB(x) ((x) * 1024)
0029 #define MiB(x) ((x) * 1024 * 1024)
0030 #define KiB_MASK(x) (KiB(x) - 1)
0031 #define MiB_MASK(x) (MiB(x) - 1)
0032
0033 struct intel_bts_snapshot_ref {
0034     void    *ref_buf;
0035     size_t  ref_offset;
0036     bool    wrapped;
0037 };
0038
0039 struct intel_bts_recording {
0040     struct auxtrace_record      itr;
0041     struct perf_pmu         *intel_bts_pmu;
0042     struct evlist       *evlist;
0043     bool                snapshot_mode;
0044     size_t              snapshot_size;
0045     int             snapshot_ref_cnt;
0046     struct intel_bts_snapshot_ref   *snapshot_refs;
0047 };
0048
0049 struct branch {
0050     u64 from;
0051     u64 to;
0052     u64 misc;
0053 };
0054
0055 static size_t
0056 intel_bts_info_priv_size(struct auxtrace_record *itr __maybe_unused,
0057              struct evlist *evlist __maybe_unused)
0058 {
0059     return INTEL_BTS_AUXTRACE_PRIV_SIZE;
0060 }
0061
0062 static int intel_bts_info_fill(struct auxtrace_record *itr,
0063                    struct perf_session *session,
0064                    struct perf_record_auxtrace_info *auxtrace_info,
0065                    size_t priv_size)
0066 {
0067     struct intel_bts_recording *btsr =
0068             container_of(itr, struct intel_bts_recording, itr);
0069     struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
0070     struct perf_event_mmap_page *pc;
0071     struct perf_tsc_conversion tc = { .time_mult = 0, };
0072     bool cap_user_time_zero = false;
0073     int err;
0074
0075     if (priv_size != INTEL_BTS_AUXTRACE_PRIV_SIZE)
0076         return -EINVAL;
0077
0078     if (!session->evlist->core.nr_mmaps)
0079         return -EINVAL;
0080
0081     pc = session->evlist->mmap[0].core.base;
0082     if (pc) {
0083         err = perf_read_tsc_conversion(pc, &tc);
0084         if (err) {
0085             if (err != -EOPNOTSUPP)
0086                 return err;
0087         } else {
0088             cap_user_time_zero = tc.time_mult != 0;
0089         }
0090         if (!cap_user_time_zero)
0091             ui__warning("Intel BTS: TSC not available\n");
0092     }
0093
0094     auxtrace_info->type = PERF_AUXTRACE_INTEL_BTS;
0095     auxtrace_info->priv[INTEL_BTS_PMU_TYPE] = intel_bts_pmu->type;
0096     auxtrace_info->priv[INTEL_BTS_TIME_SHIFT] = tc.time_shift;
0097     auxtrace_info->priv[INTEL_BTS_TIME_MULT] = tc.time_mult;
0098     auxtrace_info->priv[INTEL_BTS_TIME_ZERO] = tc.time_zero;
0099     auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO] = cap_user_time_zero;
0100     auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE] = btsr->snapshot_mode;
0101
0102     return 0;
0103 }
0104
0105 static int intel_bts_recording_options(struct auxtrace_record *itr,
0106                        struct evlist *evlist,
0107                        struct record_opts *opts)
0108 {
0109     struct intel_bts_recording *btsr =
0110             container_of(itr, struct intel_bts_recording, itr);
0111     struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
0112     struct evsel *evsel, *intel_bts_evsel = NULL;
0113     const struct perf_cpu_map *cpus = evlist->core.user_requested_cpus;
0114     bool privileged = perf_event_paranoid_check(-1);
0115
0116     if (opts->auxtrace_sample_mode) {
0117         pr_err("Intel BTS does not support AUX area sampling\n");
0118         return -EINVAL;
0119     }
0120
0121     btsr->evlist = evlist;
0122     btsr->snapshot_mode = opts->auxtrace_snapshot_mode;
0123
0124     evlist__for_each_entry(evlist, evsel) {
0125         if (evsel->core.attr.type == intel_bts_pmu->type) {
0126             if (intel_bts_evsel) {
0127                 pr_err("There may be only one " INTEL_BTS_PMU_NAME " event\n");
0128                 return -EINVAL;
0129             }
0130             evsel->core.attr.freq = 0;
0131             evsel->core.attr.sample_period = 1;
0132             evsel->needs_auxtrace_mmap = true;
0133             intel_bts_evsel = evsel;
0134             opts->full_auxtrace = true;
0135         }
0136     }
0137
0138     if (opts->auxtrace_snapshot_mode && !opts->full_auxtrace) {
0139         pr_err("Snapshot mode (-S option) requires " INTEL_BTS_PMU_NAME " PMU event (-e " INTEL_BTS_PMU_NAME ")\n");
0140         return -EINVAL;
0141     }
0142
0143     if (!opts->full_auxtrace)
0144         return 0;
0145
0146     if (opts->full_auxtrace && !perf_cpu_map__empty(cpus)) {
0147         pr_err(INTEL_BTS_PMU_NAME " does not support per-cpu recording\n");
0148         return -EINVAL;
0149     }
0150
0151     /* Set default sizes for snapshot mode */
0152     if (opts->auxtrace_snapshot_mode) {
0153         if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) {
0154             if (privileged) {
0155                 opts->auxtrace_mmap_pages = MiB(4) / page_size;
0156             } else {
0157                 opts->auxtrace_mmap_pages = KiB(128) / page_size;
0158                 if (opts->mmap_pages == UINT_MAX)
0159                     opts->mmap_pages = KiB(256) / page_size;
0160             }
0161         } else if (!opts->auxtrace_mmap_pages && !privileged &&
0162                opts->mmap_pages == UINT_MAX) {
0163             opts->mmap_pages = KiB(256) / page_size;
0164         }
0165         if (!opts->auxtrace_snapshot_size)
0166             opts->auxtrace_snapshot_size =
0167                 opts->auxtrace_mmap_pages * (size_t)page_size;
0168         if (!opts->auxtrace_mmap_pages) {
0169             size_t sz = opts->auxtrace_snapshot_size;
0170
0171             sz = round_up(sz, page_size) / page_size;
0172             opts->auxtrace_mmap_pages = roundup_pow_of_two(sz);
0173         }
0174         if (opts->auxtrace_snapshot_size >
0175                 opts->auxtrace_mmap_pages * (size_t)page_size) {
0176             pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n",
0177                    opts->auxtrace_snapshot_size,
0178                    opts->auxtrace_mmap_pages * (size_t)page_size);
0179             return -EINVAL;
0180         }
0181         if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) {
0182             pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n");
0183             return -EINVAL;
0184         }
0185         pr_debug2("Intel BTS snapshot size: %zu\n",
0186               opts->auxtrace_snapshot_size);
0187     }
0188
0189     /* Set default sizes for full trace mode */
0190     if (opts->full_auxtrace && !opts->auxtrace_mmap_pages) {
0191         if (privileged) {
0192             opts->auxtrace_mmap_pages = MiB(4) / page_size;
0193         } else {
0194             opts->auxtrace_mmap_pages = KiB(128) / page_size;
0195             if (opts->mmap_pages == UINT_MAX)
0196                 opts->mmap_pages = KiB(256) / page_size;
0197         }
0198     }
0199
0200     /* Validate auxtrace_mmap_pages */
0201     if (opts->auxtrace_mmap_pages) {
0202         size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size;
0203         size_t min_sz;
0204
0205         if (opts->auxtrace_snapshot_mode)
0206             min_sz = KiB(4);
0207         else
0208             min_sz = KiB(8);
0209
0210         if (sz < min_sz || !is_power_of_2(sz)) {
0211             pr_err("Invalid mmap size for Intel BTS: must be at least %zuKiB and a power of 2\n",
0212                    min_sz / 1024);
0213             return -EINVAL;
0214         }
0215     }
0216
0217     if (intel_bts_evsel) {
0218         /*
0219          * To obtain the auxtrace buffer file descriptor, the auxtrace event
0220          * must come first.
0221          */
0222         evlist__to_front(evlist, intel_bts_evsel);
0223         /*
0224          * In the case of per-cpu mmaps, we need the CPU on the
0225          * AUX event.
0226          */
0227         if (!perf_cpu_map__empty(cpus))
0228             evsel__set_sample_bit(intel_bts_evsel, CPU);
0229     }
0230
0231     /* Add dummy event to keep tracking */
0232     if (opts->full_auxtrace) {
0233         struct evsel *tracking_evsel;
0234         int err;
0235
0236         err = parse_event(evlist, "dummy:u");
0237         if (err)
0238             return err;
0239
0240         tracking_evsel = evlist__last(evlist);
0241
0242         evlist__set_tracking_event(evlist, tracking_evsel);
0243
0244         tracking_evsel->core.attr.freq = 0;
0245         tracking_evsel->core.attr.sample_period = 1;
0246     }
0247
0248     return 0;
0249 }
0250
0251 static int intel_bts_parse_snapshot_options(struct auxtrace_record *itr,
0252                         struct record_opts *opts,
0253                         const char *str)
0254 {
0255     struct intel_bts_recording *btsr =
0256             container_of(itr, struct intel_bts_recording, itr);
0257     unsigned long long snapshot_size = 0;
0258     char *endptr;
0259
0260     if (str) {
0261         snapshot_size = strtoull(str, &endptr, 0);
0262         if (*endptr || snapshot_size > SIZE_MAX)
0263             return -1;
0264     }
0265
0266     opts->auxtrace_snapshot_mode = true;
0267     opts->auxtrace_snapshot_size = snapshot_size;
0268
0269     btsr->snapshot_size = snapshot_size;
0270
0271     return 0;
0272 }
0273
0274 static u64 intel_bts_reference(struct auxtrace_record *itr __maybe_unused)
0275 {
0276     return rdtsc();
0277 }
0278
0279 static int intel_bts_alloc_snapshot_refs(struct intel_bts_recording *btsr,
0280                      int idx)
0281 {
0282     const size_t sz = sizeof(struct intel_bts_snapshot_ref);
0283     int cnt = btsr->snapshot_ref_cnt, new_cnt = cnt * 2;
0284     struct intel_bts_snapshot_ref *refs;
0285
0286     if (!new_cnt)
0287         new_cnt = 16;
0288
0289     while (new_cnt <= idx)
0290         new_cnt *= 2;
0291
0292     refs = calloc(new_cnt, sz);
0293     if (!refs)
0294         return -ENOMEM;
0295
0296     memcpy(refs, btsr->snapshot_refs, cnt * sz);
0297
0298     btsr->snapshot_refs = refs;
0299     btsr->snapshot_ref_cnt = new_cnt;
0300
0301     return 0;
0302 }
0303
0304 static void intel_bts_free_snapshot_refs(struct intel_bts_recording *btsr)
0305 {
0306     int i;
0307
0308     for (i = 0; i < btsr->snapshot_ref_cnt; i++)
0309         zfree(&btsr->snapshot_refs[i].ref_buf);
0310     zfree(&btsr->snapshot_refs);
0311 }
0312
0313 static void intel_bts_recording_free(struct auxtrace_record *itr)
0314 {
0315     struct intel_bts_recording *btsr =
0316             container_of(itr, struct intel_bts_recording, itr);
0317
0318     intel_bts_free_snapshot_refs(btsr);
0319     free(btsr);
0320 }
0321
0322 static int intel_bts_snapshot_start(struct auxtrace_record *itr)
0323 {
0324     struct intel_bts_recording *btsr =
0325             container_of(itr, struct intel_bts_recording, itr);
0326     struct evsel *evsel;
0327
0328     evlist__for_each_entry(btsr->evlist, evsel) {
0329         if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
0330             return evsel__disable(evsel);
0331     }
0332     return -EINVAL;
0333 }
0334
0335 static int intel_bts_snapshot_finish(struct auxtrace_record *itr)
0336 {
0337     struct intel_bts_recording *btsr =
0338             container_of(itr, struct intel_bts_recording, itr);
0339     struct evsel *evsel;
0340
0341     evlist__for_each_entry(btsr->evlist, evsel) {
0342         if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
0343             return evsel__enable(evsel);
0344     }
0345     return -EINVAL;
0346 }
0347
0348 static bool intel_bts_first_wrap(u64 *data, size_t buf_size)
0349 {
0350     int i, a, b;
0351
0352     b = buf_size >> 3;
0353     a = b - 512;
0354     if (a < 0)
0355         a = 0;
0356
0357     for (i = a; i < b; i++) {
0358         if (data[i])
0359             return true;
0360     }
0361
0362     return false;
0363 }
0364
0365 static int intel_bts_find_snapshot(struct auxtrace_record *itr, int idx,
0366                    struct auxtrace_mmap *mm, unsigned char *data,
0367                    u64 *head, u64 *old)
0368 {
0369     struct intel_bts_recording *btsr =
0370             container_of(itr, struct intel_bts_recording, itr);
0371     bool wrapped;
0372     int err;
0373
0374     pr_debug3("%s: mmap index %d old head %zu new head %zu\n",
0375           __func__, idx, (size_t)*old, (size_t)*head);
0376
0377     if (idx >= btsr->snapshot_ref_cnt) {
0378         err = intel_bts_alloc_snapshot_refs(btsr, idx);
0379         if (err)
0380             goto out_err;
0381     }
0382
0383     wrapped = btsr->snapshot_refs[idx].wrapped;
0384     if (!wrapped && intel_bts_first_wrap((u64 *)data, mm->len)) {
0385         btsr->snapshot_refs[idx].wrapped = true;
0386         wrapped = true;
0387     }
0388
0389     /*
0390      * In full trace mode 'head' continually increases.  However in snapshot
0391      * mode 'head' is an offset within the buffer.  Here 'old' and 'head'
0392      * are adjusted to match the full trace case which expects that 'old' is
0393      * always less than 'head'.
0394      */
0395     if (wrapped) {
0396         *old = *head;
0397         *head += mm->len;
0398     } else {
0399         if (mm->mask)
0400             *old &= mm->mask;
0401         else
0402             *old %= mm->len;
0403         if (*old > *head)
0404             *head += mm->len;
0405     }
0406
0407     pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n",
0408           __func__, wrapped ? "" : "not ", (size_t)*old, (size_t)*head);
0409
0410     return 0;
0411
0412 out_err:
0413     pr_err("%s: failed, error %d\n", __func__, err);
0414     return err;
0415 }
0416
0417 struct auxtrace_record *intel_bts_recording_init(int *err)
0418 {
0419     struct perf_pmu *intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
0420     struct intel_bts_recording *btsr;
0421
0422     if (!intel_bts_pmu)
0423         return NULL;
0424
0425     if (setenv("JITDUMP_USE_ARCH_TIMESTAMP", "1", 1)) {
0426         *err = -errno;
0427         return NULL;
0428     }
0429
0430     btsr = zalloc(sizeof(struct intel_bts_recording));
0431     if (!btsr) {
0432         *err = -ENOMEM;
0433         return NULL;
0434     }
0435
0436     btsr->intel_bts_pmu = intel_bts_pmu;
0437     btsr->itr.pmu = intel_bts_pmu;
0438     btsr->itr.recording_options = intel_bts_recording_options;
0439     btsr->itr.info_priv_size = intel_bts_info_priv_size;
0440     btsr->itr.info_fill = intel_bts_info_fill;
0441     btsr->itr.free = intel_bts_recording_free;
0442     btsr->itr.snapshot_start = intel_bts_snapshot_start;
0443     btsr->itr.snapshot_finish = intel_bts_snapshot_finish;
0444     btsr->itr.find_snapshot = intel_bts_find_snapshot;
0445     btsr->itr.parse_snapshot_options = intel_bts_parse_snapshot_options;
0446     btsr->itr.reference = intel_bts_reference;
0447     btsr->itr.read_finish = auxtrace_record__read_finish;
0448     btsr->itr.alignment = sizeof(struct branch);
0449     return &btsr->itr;
0450 }