diff options
Diffstat (limited to 'tools/perf/builtin-stat.c')
| -rw-r--r-- | tools/perf/builtin-stat.c | 3562 |
1 files changed, 2521 insertions, 1041 deletions
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 352fbd7ff4a1..ab40d85fb125 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -1,3 +1,4 @@ +// SPDX-License-Identifier: GPL-2.0-only /* * builtin-stat.c * @@ -37,15 +38,15 @@ * Mike Galbraith <efault@gmx.de> * Paul Mackerras <paulus@samba.org> * Jaswinder Singh Rajput <jaswinder@kernel.org> - * - * Released under the GPL v2. (and only v2, not any later version) */ -#include "perf.h" #include "builtin.h" -#include "util/util.h" -#include "util/parse-options.h" +#include "util/cgroup.h" +#include <subcmd/parse-options.h> #include "util/parse-events.h" +#include "util/pmus.h" +#include "util/pmu.h" +#include "util/tool_pmu.h" #include "util/event.h" #include "util/evlist.h" #include "util/evsel.h" @@ -54,477 +55,1030 @@ #include "util/stat.h" #include "util/header.h" #include "util/cpumap.h" -#include "util/thread.h" #include "util/thread_map.h" - +#include "util/counts.h" +#include "util/topdown.h" +#include "util/session.h" +#include "util/tool.h" +#include "util/string2.h" +#include "util/metricgroup.h" +#include "util/synthetic-events.h" +#include "util/target.h" +#include "util/time-utils.h" +#include "util/top.h" +#include "util/affinity.h" +#include "util/pfm.h" +#include "util/bpf_counter.h" +#include "util/iostat.h" +#include "util/util.h" +#include "util/intel-tpebs.h" +#include "asm/bug.h" + +#include <linux/list_sort.h> +#include <linux/time64.h> +#include <linux/zalloc.h> +#include <api/fs/fs.h> +#include <errno.h> +#include <signal.h> #include <stdlib.h> #include <sys/prctl.h> +#include <inttypes.h> #include <locale.h> +#include <math.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <sys/wait.h> +#include <unistd.h> +#include <sys/time.h> +#include <sys/resource.h> +#include <linux/err.h> + +#include <linux/ctype.h> +#include <perf/evlist.h> +#include <internal/threadmap.h> + +#ifdef HAVE_BPF_SKEL +#include "util/bpf_skel/bperf_cgroup.h" +#endif #define DEFAULT_SEPARATOR " " -#define CNTR_NOT_SUPPORTED "<not supported>" -#define CNTR_NOT_COUNTED "<not counted>" +#define FREEZE_ON_SMI_PATH "bus/event_source/devices/cpu/freeze_on_smi" -static void print_stat(int argc, const char **argv); -static void print_counter_aggr(struct perf_evsel *counter, char *prefix); -static void print_counter(struct perf_evsel *counter, char *prefix); -static void print_aggr(char *prefix); +struct rusage_stats { + struct stats ru_utime_usec_stat; + struct stats ru_stime_usec_stat; +}; -static struct perf_evlist *evsel_list; +static void print_counters(struct timespec *ts, int argc, const char **argv); -static struct perf_target target = { - .uid = UINT_MAX, +static struct evlist *evsel_list; +static struct parse_events_option_args parse_events_option_args = { + .evlistp = &evsel_list, }; -enum aggr_mode { - AGGR_NONE, - AGGR_GLOBAL, - AGGR_SOCKET, - AGGR_CORE, -}; +static bool all_counters_use_bpf = true; + +static struct target target; -static int run_count = 1; -static bool no_inherit = false; -static bool scale = true; -static enum aggr_mode aggr_mode = AGGR_GLOBAL; -static volatile pid_t child_pid = -1; -static bool null_run = false; +static volatile sig_atomic_t child_pid = -1; static int detailed_run = 0; -static bool big_num = true; +static bool transaction_run; +static bool topdown_run = false; +static bool smi_cost = false; +static bool smi_reset = false; static int big_num_opt = -1; -static const char *csv_sep = NULL; -static bool csv_output = false; -static bool group = false; -static FILE *output = NULL; static const char *pre_cmd = NULL; static const char *post_cmd = NULL; static bool sync_run = false; -static unsigned int interval = 0; static bool forever = false; +static bool force_metric_only = false; static struct timespec ref_time; -static struct cpu_map *aggr_map; -static int (*aggr_get_id)(struct cpu_map *m, int cpu); - -static volatile int done = 0; +static bool append_file; +static bool interval_count; +static const char *output_name; +static int output_fd; +static char *metrics; +static struct rusage_stats ru_stats; struct perf_stat { - struct stats res_stats[3]; + bool record; + struct perf_data data; + struct perf_session *session; + u64 bytes_written; + struct perf_tool tool; + bool maps_allocated; + struct perf_cpu_map *cpus; + struct perf_thread_map *threads; + enum aggr_mode aggr_mode; + u32 aggr_level; }; +static struct perf_stat perf_stat; +#define STAT_RECORD perf_stat.record + +static volatile sig_atomic_t done = 0; + +/* Options set from the command line. */ +struct opt_aggr_mode { + bool node, socket, die, cluster, cache, core, thread, no_aggr; +}; + +/* Turn command line option into most generic aggregation mode setting. */ +static enum aggr_mode opt_aggr_mode_to_aggr_mode(struct opt_aggr_mode *opt_mode) +{ + enum aggr_mode mode = AGGR_GLOBAL; + + if (opt_mode->node) + mode = AGGR_NODE; + if (opt_mode->socket) + mode = AGGR_SOCKET; + if (opt_mode->die) + mode = AGGR_DIE; + if (opt_mode->cluster) + mode = AGGR_CLUSTER; + if (opt_mode->cache) + mode = AGGR_CACHE; + if (opt_mode->core) + mode = AGGR_CORE; + if (opt_mode->thread) + mode = AGGR_THREAD; + if (opt_mode->no_aggr) + mode = AGGR_NONE; + return mode; +} + +static void evlist__check_cpu_maps(struct evlist *evlist) +{ + struct evsel *evsel, *warned_leader = NULL; + + evlist__for_each_entry(evlist, evsel) { + struct evsel *leader = evsel__leader(evsel); + + /* Check that leader matches cpus with each member. */ + if (leader == evsel) + continue; + if (perf_cpu_map__equal(leader->core.cpus, evsel->core.cpus)) + continue; + + /* If there's mismatch disable the group and warn user. */ + if (warned_leader != leader) { + char buf[200]; + + pr_warning("WARNING: grouped events cpus do not match.\n" + "Events with CPUs not matching the leader will " + "be removed from the group.\n"); + evsel__group_desc(leader, buf, sizeof(buf)); + pr_warning(" %s\n", buf); + warned_leader = leader; + } + if (verbose > 0) { + char buf[200]; + + cpu_map__snprint(leader->core.cpus, buf, sizeof(buf)); + pr_warning(" %s: %s\n", leader->name, buf); + cpu_map__snprint(evsel->core.cpus, buf, sizeof(buf)); + pr_warning(" %s: %s\n", evsel->name, buf); + } + + evsel__remove_from_group(evsel, leader); + } +} + static inline void diff_timespec(struct timespec *r, struct timespec *a, struct timespec *b) { r->tv_sec = a->tv_sec - b->tv_sec; if (a->tv_nsec < b->tv_nsec) { - r->tv_nsec = a->tv_nsec + 1000000000L - b->tv_nsec; + r->tv_nsec = a->tv_nsec + NSEC_PER_SEC - b->tv_nsec; r->tv_sec--; } else { r->tv_nsec = a->tv_nsec - b->tv_nsec ; } } -static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel) +static void perf_stat__reset_stats(void) { - return (evsel->cpus && !target.cpu_list) ? evsel->cpus : evsel_list->cpus; + evlist__reset_stats(evsel_list); + memset(stat_config.walltime_nsecs_stats, 0, sizeof(*stat_config.walltime_nsecs_stats)); } -static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel) +static int process_synthesized_event(const struct perf_tool *tool __maybe_unused, + union perf_event *event, + struct perf_sample *sample __maybe_unused, + struct machine *machine __maybe_unused) { - return perf_evsel__cpus(evsel)->nr; + if (perf_data__write(&perf_stat.data, event, event->header.size) < 0) { + pr_err("failed to write perf data, error: %m\n"); + return -1; + } + + perf_stat.bytes_written += event->header.size; + return 0; } -static void perf_evsel__reset_stat_priv(struct perf_evsel *evsel) +static int write_stat_round_event(u64 tm, u64 type) { - memset(evsel->priv, 0, sizeof(struct perf_stat)); + return perf_event__synthesize_stat_round(NULL, tm, type, + process_synthesized_event, + NULL); } -static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel) +#define WRITE_STAT_ROUND_EVENT(time, interval) \ + write_stat_round_event(time, PERF_STAT_ROUND_TYPE__ ## interval) + +#define SID(e, x, y) xyarray__entry(e->core.sample_id, x, y) + +static int evsel__write_stat_event(struct evsel *counter, int cpu_map_idx, u32 thread, + struct perf_counts_values *count) { - evsel->priv = zalloc(sizeof(struct perf_stat)); - return evsel->priv == NULL ? -ENOMEM : 0; + struct perf_sample_id *sid = SID(counter, cpu_map_idx, thread); + struct perf_cpu cpu = perf_cpu_map__cpu(evsel__cpus(counter), cpu_map_idx); + + return perf_event__synthesize_stat(NULL, cpu, thread, sid->id, count, + process_synthesized_event, NULL); } -static void perf_evsel__free_stat_priv(struct perf_evsel *evsel) +static int read_single_counter(struct evsel *counter, int cpu_map_idx, int thread) { - free(evsel->priv); - evsel->priv = NULL; + int err = evsel__read_counter(counter, cpu_map_idx, thread); + + /* + * Reading user and system time will fail when the process + * terminates. Use the wait4 values in that case. + */ + if (err && cpu_map_idx == 0 && + (evsel__tool_event(counter) == TOOL_PMU__EVENT_USER_TIME || + evsel__tool_event(counter) == TOOL_PMU__EVENT_SYSTEM_TIME)) { + struct perf_counts_values *count = + perf_counts(counter->counts, cpu_map_idx, thread); + struct perf_counts_values *old_count = NULL; + u64 val; + + if (counter->prev_raw_counts) + old_count = perf_counts(counter->prev_raw_counts, cpu_map_idx, thread); + + if (evsel__tool_event(counter) == TOOL_PMU__EVENT_USER_TIME) + val = ru_stats.ru_utime_usec_stat.mean; + else + val = ru_stats.ru_stime_usec_stat.mean; + + count->val = val; + if (old_count) { + count->run = old_count->run + 1; + count->ena = old_count->ena + 1; + } else { + count->run++; + count->ena++; + } + return 0; + } + return err; } -static int perf_evsel__alloc_prev_raw_counts(struct perf_evsel *evsel) +/* + * Read out the results of a single counter: + * do not aggregate counts across CPUs in system-wide mode + */ +static int read_counter_cpu(struct evsel *counter, int cpu_map_idx) { - void *addr; - size_t sz; + int nthreads = perf_thread_map__nr(evsel_list->core.threads); + int thread; + + if (!counter->supported) + return -ENOENT; + + for (thread = 0; thread < nthreads; thread++) { + struct perf_counts_values *count; + + count = perf_counts(counter->counts, cpu_map_idx, thread); + + /* + * The leader's group read loads data into its group members + * (via evsel__read_counter()) and sets their count->loaded. + */ + if (!perf_counts__is_loaded(counter->counts, cpu_map_idx, thread) && + read_single_counter(counter, cpu_map_idx, thread)) { + counter->counts->scaled = -1; + perf_counts(counter->counts, cpu_map_idx, thread)->ena = 0; + perf_counts(counter->counts, cpu_map_idx, thread)->run = 0; + return -1; + } - sz = sizeof(*evsel->counts) + - (perf_evsel__nr_cpus(evsel) * sizeof(struct perf_counts_values)); + perf_counts__set_loaded(counter->counts, cpu_map_idx, thread, false); - addr = zalloc(sz); - if (!addr) - return -ENOMEM; + if (STAT_RECORD) { + if (evsel__write_stat_event(counter, cpu_map_idx, thread, count)) { + pr_err("failed to write stat event\n"); + return -1; + } + } - evsel->prev_raw_counts = addr; + if (verbose > 1) { + fprintf(stat_config.output, + "%s: %d: %" PRIu64 " %" PRIu64 " %" PRIu64 "\n", + evsel__name(counter), + perf_cpu_map__cpu(evsel__cpus(counter), + cpu_map_idx).cpu, + count->val, count->ena, count->run); + } + } return 0; } -static void perf_evsel__free_prev_raw_counts(struct perf_evsel *evsel) +static int read_counters_with_affinity(void) { - free(evsel->prev_raw_counts); - evsel->prev_raw_counts = NULL; + struct evlist_cpu_iterator evlist_cpu_itr; + struct affinity saved_affinity, *affinity; + + if (all_counters_use_bpf) + return 0; + + if (!target__has_cpu(&target) || target__has_per_thread(&target)) + affinity = NULL; + else if (affinity__setup(&saved_affinity) < 0) + return -1; + else + affinity = &saved_affinity; + + evlist__for_each_cpu(evlist_cpu_itr, evsel_list, affinity) { + struct evsel *counter = evlist_cpu_itr.evsel; + + if (evsel__is_bpf(counter)) + continue; + + if (evsel__is_tool(counter)) + continue; + + if (!counter->err) + counter->err = read_counter_cpu(counter, evlist_cpu_itr.cpu_map_idx); + } + if (affinity) + affinity__cleanup(&saved_affinity); + + return 0; } -static void perf_evlist__free_stats(struct perf_evlist *evlist) +static int read_bpf_map_counters(void) { - struct perf_evsel *evsel; + struct evsel *counter; + int err; + + evlist__for_each_entry(evsel_list, counter) { + if (!evsel__is_bpf(counter)) + continue; - list_for_each_entry(evsel, &evlist->entries, node) { - perf_evsel__free_stat_priv(evsel); - perf_evsel__free_counts(evsel); - perf_evsel__free_prev_raw_counts(evsel); + err = bpf_counter__read(counter); + if (err) + return err; } + return 0; } -static int perf_evlist__alloc_stats(struct perf_evlist *evlist, bool alloc_raw) +static int read_tool_counters(void) { - struct perf_evsel *evsel; + struct evsel *counter; - list_for_each_entry(evsel, &evlist->entries, node) { - if (perf_evsel__alloc_stat_priv(evsel) < 0 || - perf_evsel__alloc_counts(evsel, perf_evsel__nr_cpus(evsel)) < 0 || - (alloc_raw && perf_evsel__alloc_prev_raw_counts(evsel) < 0)) - goto out_free; - } + evlist__for_each_entry(evsel_list, counter) { + int idx; - return 0; + if (!evsel__is_tool(counter)) + continue; -out_free: - perf_evlist__free_stats(evlist); - return -1; + perf_cpu_map__for_each_idx(idx, counter->core.cpus) { + if (!counter->err) + counter->err = read_counter_cpu(counter, idx); + } + } + return 0; } -static struct stats runtime_nsecs_stats[MAX_NR_CPUS]; -static struct stats runtime_cycles_stats[MAX_NR_CPUS]; -static struct stats runtime_stalled_cycles_front_stats[MAX_NR_CPUS]; -static struct stats runtime_stalled_cycles_back_stats[MAX_NR_CPUS]; -static struct stats runtime_branches_stats[MAX_NR_CPUS]; -static struct stats runtime_cacherefs_stats[MAX_NR_CPUS]; -static struct stats runtime_l1_dcache_stats[MAX_NR_CPUS]; -static struct stats runtime_l1_icache_stats[MAX_NR_CPUS]; -static struct stats runtime_ll_cache_stats[MAX_NR_CPUS]; -static struct stats runtime_itlb_cache_stats[MAX_NR_CPUS]; -static struct stats runtime_dtlb_cache_stats[MAX_NR_CPUS]; -static struct stats walltime_nsecs_stats; - -static void perf_stat__reset_stats(struct perf_evlist *evlist) +static int read_counters(void) { - struct perf_evsel *evsel; + int ret; + + if (stat_config.stop_read_counter) + return 0; - list_for_each_entry(evsel, &evlist->entries, node) { - perf_evsel__reset_stat_priv(evsel); - perf_evsel__reset_counts(evsel, perf_evsel__nr_cpus(evsel)); + // Read all BPF counters first. + ret = read_bpf_map_counters(); + if (ret) + return ret; + + // Read non-BPF and non-tool counters next. + ret = read_counters_with_affinity(); + if (ret) + return ret; + + // Read the tool counters last. This way the duration_time counter + // should always be greater than any other counter's enabled time. + return read_tool_counters(); +} + +static void process_counters(void) +{ + struct evsel *counter; + + evlist__for_each_entry(evsel_list, counter) { + if (counter->err) + pr_debug("failed to read counter %s\n", counter->name); + if (counter->err == 0 && perf_stat_process_counter(&stat_config, counter)) + pr_warning("failed to process counter %s\n", counter->name); + counter->err = 0; } - memset(runtime_nsecs_stats, 0, sizeof(runtime_nsecs_stats)); - memset(runtime_cycles_stats, 0, sizeof(runtime_cycles_stats)); - memset(runtime_stalled_cycles_front_stats, 0, sizeof(runtime_stalled_cycles_front_stats)); - memset(runtime_stalled_cycles_back_stats, 0, sizeof(runtime_stalled_cycles_back_stats)); - memset(runtime_branches_stats, 0, sizeof(runtime_branches_stats)); - memset(runtime_cacherefs_stats, 0, sizeof(runtime_cacherefs_stats)); - memset(runtime_l1_dcache_stats, 0, sizeof(runtime_l1_dcache_stats)); - memset(runtime_l1_icache_stats, 0, sizeof(runtime_l1_icache_stats)); - memset(runtime_ll_cache_stats, 0, sizeof(runtime_ll_cache_stats)); - memset(runtime_itlb_cache_stats, 0, sizeof(runtime_itlb_cache_stats)); - memset(runtime_dtlb_cache_stats, 0, sizeof(runtime_dtlb_cache_stats)); - memset(&walltime_nsecs_stats, 0, sizeof(walltime_nsecs_stats)); + perf_stat_merge_counters(&stat_config, evsel_list); + perf_stat_process_percore(&stat_config, evsel_list); } -static int create_perf_stat_counter(struct perf_evsel *evsel) +static void process_interval(void) { - struct perf_event_attr *attr = &evsel->attr; + struct timespec ts, rs; - if (scale) - attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | - PERF_FORMAT_TOTAL_TIME_RUNNING; + clock_gettime(CLOCK_MONOTONIC, &ts); + diff_timespec(&rs, &ts, &ref_time); - attr->inherit = !no_inherit; + evlist__reset_aggr_stats(evsel_list); - if (perf_target__has_cpu(&target)) - return perf_evsel__open_per_cpu(evsel, perf_evsel__cpus(evsel)); + if (read_counters() == 0) + process_counters(); - if (!perf_target__has_task(&target) && - perf_evsel__is_group_leader(evsel)) { - attr->disabled = 1; - attr->enable_on_exec = 1; + if (STAT_RECORD) { + if (WRITE_STAT_ROUND_EVENT(rs.tv_sec * NSEC_PER_SEC + rs.tv_nsec, INTERVAL)) + pr_err("failed to write stat round event\n"); } - return perf_evsel__open_per_thread(evsel, evsel_list->threads); + init_stats(stat_config.walltime_nsecs_stats); + update_stats(stat_config.walltime_nsecs_stats, stat_config.interval * 1000000ULL); + print_counters(&rs, 0, NULL); } -/* - * Does the counter have nsecs as a unit? - */ -static inline int nsec_counter(struct perf_evsel *evsel) +static bool handle_interval(unsigned int interval, int *times) { - if (perf_evsel__match(evsel, SOFTWARE, SW_CPU_CLOCK) || - perf_evsel__match(evsel, SOFTWARE, SW_TASK_CLOCK)) - return 1; + if (interval) { + process_interval(); + if (interval_count && !(--(*times))) + return true; + } + return false; +} + +static int enable_counters(void) +{ + struct evsel *evsel; + int err; + + evlist__for_each_entry(evsel_list, evsel) { + if (!evsel__is_bpf(evsel)) + continue; + + err = bpf_counter__enable(evsel); + if (err) + return err; + } + if (!target__enable_on_exec(&target)) { + if (!all_counters_use_bpf) + evlist__enable(evsel_list); + } return 0; } -/* - * Update various tracking values we maintain to print - * more semantic information such as miss/hit ratios, - * instruction rates, etc: - */ -static void update_shadow_stats(struct perf_evsel *counter, u64 *count) -{ - if (perf_evsel__match(counter, SOFTWARE, SW_TASK_CLOCK)) - update_stats(&runtime_nsecs_stats[0], count[0]); - else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES)) - update_stats(&runtime_cycles_stats[0], count[0]); - else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND)) - update_stats(&runtime_stalled_cycles_front_stats[0], count[0]); - else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_BACKEND)) - update_stats(&runtime_stalled_cycles_back_stats[0], count[0]); - else if (perf_evsel__match(counter, HARDWARE, HW_BRANCH_INSTRUCTIONS)) - update_stats(&runtime_branches_stats[0], count[0]); - else if (perf_evsel__match(counter, HARDWARE, HW_CACHE_REFERENCES)) - update_stats(&runtime_cacherefs_stats[0], count[0]); - else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1D)) - update_stats(&runtime_l1_dcache_stats[0], count[0]); - else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1I)) - update_stats(&runtime_l1_icache_stats[0], count[0]); - else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_LL)) - update_stats(&runtime_ll_cache_stats[0], count[0]); - else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_DTLB)) - update_stats(&runtime_dtlb_cache_stats[0], count[0]); - else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB)) - update_stats(&runtime_itlb_cache_stats[0], count[0]); +static void disable_counters(void) +{ + struct evsel *counter; + + /* + * If we don't have tracee (attaching to task or cpu), counters may + * still be running. To get accurate group ratios, we must stop groups + * from counting before reading their constituent counters. + */ + if (!target__none(&target)) { + evlist__for_each_entry(evsel_list, counter) + bpf_counter__disable(counter); + if (!all_counters_use_bpf) + evlist__disable(evsel_list); + } } +static volatile sig_atomic_t workload_exec_errno; + /* - * Read out the results of a single counter: - * aggregate counts across CPUs in system-wide mode + * evlist__prepare_workload will send a SIGUSR1 + * if the fork fails, since we asked by setting its + * want_signal to true. */ -static int read_counter_aggr(struct perf_evsel *counter) +static void workload_exec_failed_signal(int signo __maybe_unused, siginfo_t *info, + void *ucontext __maybe_unused) +{ + workload_exec_errno = info->si_value.sival_int; +} + +static bool evsel__should_store_id(struct evsel *counter) { - struct perf_stat *ps = counter->priv; - u64 *count = counter->counts->aggr.values; + return STAT_RECORD || counter->core.attr.read_format & PERF_FORMAT_ID; +} + +static bool is_target_alive(struct target *_target, + struct perf_thread_map *threads) +{ + struct stat st; int i; - if (__perf_evsel__read(counter, perf_evsel__nr_cpus(counter), - thread_map__nr(evsel_list->threads), scale) < 0) - return -1; + if (!target__has_task(_target)) + return true; - for (i = 0; i < 3; i++) - update_stats(&ps->res_stats[i], count[i]); + for (i = 0; i < threads->nr; i++) { + char path[PATH_MAX]; - if (verbose) { - fprintf(output, "%s: %" PRIu64 " %" PRIu64 " %" PRIu64 "\n", - perf_evsel__name(counter), count[0], count[1], count[2]); + scnprintf(path, PATH_MAX, "%s/%d", procfs__mountpoint(), + threads->map[i].pid); + + if (!stat(path, &st)) + return true; } - /* - * Save the full runtime - to allow normalization during printout: - */ - update_shadow_stats(counter, count); + return false; +} - return 0; +static void process_evlist(struct evlist *evlist, unsigned int interval) +{ + enum evlist_ctl_cmd cmd = EVLIST_CTL_CMD_UNSUPPORTED; + + if (evlist__ctlfd_process(evlist, &cmd) > 0) { + switch (cmd) { + case EVLIST_CTL_CMD_ENABLE: + fallthrough; + case EVLIST_CTL_CMD_DISABLE: + if (interval) + process_interval(); + break; + case EVLIST_CTL_CMD_SNAPSHOT: + case EVLIST_CTL_CMD_ACK: + case EVLIST_CTL_CMD_UNSUPPORTED: + case EVLIST_CTL_CMD_EVLIST: + case EVLIST_CTL_CMD_STOP: + case EVLIST_CTL_CMD_PING: + default: + break; + } + } } -/* - * Read out the results of a single counter: - * do not aggregate counts across CPUs in system-wide mode - */ -static int read_counter(struct perf_evsel *counter) +static void compute_tts(struct timespec *time_start, struct timespec *time_stop, + int *time_to_sleep) { - u64 *count; - int cpu; + int tts = *time_to_sleep; + struct timespec time_diff; - for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) { - if (__perf_evsel__read_on_cpu(counter, cpu, 0, scale) < 0) - return -1; + diff_timespec(&time_diff, time_stop, time_start); + + tts -= time_diff.tv_sec * MSEC_PER_SEC + + time_diff.tv_nsec / NSEC_PER_MSEC; + + if (tts < 0) + tts = 0; + + *time_to_sleep = tts; +} - count = counter->counts->cpu[cpu].values; +static int dispatch_events(bool forks, int timeout, int interval, int *times) +{ + int child_exited = 0, status = 0; + int time_to_sleep, sleep_time; + struct timespec time_start, time_stop; + + if (interval) + sleep_time = interval; + else if (timeout) + sleep_time = timeout; + else + sleep_time = 1000; + + time_to_sleep = sleep_time; + + while (!done) { + if (forks) + child_exited = waitpid(child_pid, &status, WNOHANG); + else + child_exited = !is_target_alive(&target, evsel_list->core.threads) ? 1 : 0; + + if (child_exited) + break; - update_shadow_stats(counter, count); + clock_gettime(CLOCK_MONOTONIC, &time_start); + if (!(evlist__poll(evsel_list, time_to_sleep) > 0)) { /* poll timeout or EINTR */ + if (timeout || handle_interval(interval, times)) + break; + time_to_sleep = sleep_time; + } else { /* fd revent */ + process_evlist(evsel_list, interval); + clock_gettime(CLOCK_MONOTONIC, &time_stop); + compute_tts(&time_start, &time_stop, &time_to_sleep); + } } - return 0; + return status; } -static void print_interval(void) +enum counter_recovery { + COUNTER_SKIP, + COUNTER_RETRY, +}; + +static enum counter_recovery stat_handle_error(struct evsel *counter, int err) { - static int num_print_interval; - struct perf_evsel *counter; - struct perf_stat *ps; - struct timespec ts, rs; - char prefix[64]; + char msg[BUFSIZ]; - if (aggr_mode == AGGR_GLOBAL) { - list_for_each_entry(counter, &evsel_list->entries, node) { - ps = counter->priv; - memset(ps->res_stats, 0, sizeof(ps->res_stats)); - read_counter_aggr(counter); + assert(!counter->supported); + + /* + * PPC returns ENXIO for HW counters until 2.6.37 + * (behavior changed with commit b0a873e). + */ + if (err == EINVAL || err == ENOSYS || err == ENOENT || err == ENXIO) { + if (verbose > 0) { + evsel__open_strerror(counter, &target, err, msg, sizeof(msg)); + ui__warning("%s event is not supported by the kernel.\n%s\n", + evsel__name(counter), msg); } - } else { - list_for_each_entry(counter, &evsel_list->entries, node) { - ps = counter->priv; - memset(ps->res_stats, 0, sizeof(ps->res_stats)); - read_counter(counter); + return COUNTER_SKIP; + } + if (evsel__fallback(counter, &target, err, msg, sizeof(msg))) { + if (verbose > 0) + ui__warning("%s\n", msg); + counter->supported = true; + return COUNTER_RETRY; + } + if (target__has_per_thread(&target) && err != EOPNOTSUPP && + evsel_list->core.threads && evsel_list->core.threads->err_thread != -1) { + /* + * For global --per-thread case, skip current + * error thread. + */ + if (!thread_map__remove(evsel_list->core.threads, + evsel_list->core.threads->err_thread)) { + evsel_list->core.threads->err_thread = -1; + counter->supported = true; + return COUNTER_RETRY; } } + if (verbose > 0) { + evsel__open_strerror(counter, &target, err, msg, sizeof(msg)); + ui__warning(err == EOPNOTSUPP + ? "%s event is not supported by the kernel.\n%s\n" + : "skipping event %s that kernel failed to open.\n%s\n", + evsel__name(counter), msg); + } + return COUNTER_SKIP; +} - clock_gettime(CLOCK_MONOTONIC, &ts); - diff_timespec(&rs, &ts, &ref_time); - sprintf(prefix, "%6lu.%09lu%s", rs.tv_sec, rs.tv_nsec, csv_sep); +static int create_perf_stat_counter(struct evsel *evsel, + struct perf_stat_config *config, + int cpu_map_idx) +{ + struct perf_event_attr *attr = &evsel->core.attr; + struct evsel *leader = evsel__leader(evsel); - if (num_print_interval == 0 && !csv_output) { - switch (aggr_mode) { - case AGGR_SOCKET: - fprintf(output, "# time socket cpus counts events\n"); - break; - case AGGR_CORE: - fprintf(output, "# time core cpus counts events\n"); - break; - case AGGR_NONE: - fprintf(output, "# time CPU counts events\n"); - break; - case AGGR_GLOBAL: - default: - fprintf(output, "# time counts events\n"); - } + /* Reset supported flag as creating a stat counter is retried. */ + attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | + PERF_FORMAT_TOTAL_TIME_RUNNING; + + /* + * The event is part of non trivial group, let's enable + * the group read (for leader) and ID retrieval for all + * members. + */ + if (leader->core.nr_members > 1) + attr->read_format |= PERF_FORMAT_ID|PERF_FORMAT_GROUP; + + attr->inherit = !config->no_inherit && list_empty(&evsel->bpf_counter_list); + + /* + * Some events get initialized with sample_(period/type) set, + * like tracepoints. Clear it up for counting. + */ + attr->sample_period = 0; + + if (config->identifier) + attr->sample_type = PERF_SAMPLE_IDENTIFIER; + + if (config->all_user) { + attr->exclude_kernel = 1; + attr->exclude_user = 0; + } + + if (config->all_kernel) { + attr->exclude_kernel = 0; + attr->exclude_user = 1; } - if (++num_print_interval == 25) - num_print_interval = 0; + /* + * Disabling all counters initially, they will be enabled + * either manually by us or by kernel via enable_on_exec + * set later. + */ + if (evsel__is_group_leader(evsel)) { + attr->disabled = 1; - switch (aggr_mode) { - case AGGR_CORE: - case AGGR_SOCKET: - print_aggr(prefix); - break; - case AGGR_NONE: - list_for_each_entry(counter, &evsel_list->entries, node) - print_counter(counter, prefix); - break; - case AGGR_GLOBAL: - default: - list_for_each_entry(counter, &evsel_list->entries, node) - print_counter_aggr(counter, prefix); + if (target__enable_on_exec(&target)) + attr->enable_on_exec = 1; } + + return evsel__open_per_cpu_and_thread(evsel, evsel__cpus(evsel), cpu_map_idx, + evsel->core.threads); +} + +static void update_rusage_stats(const struct rusage *rusage) +{ + const u64 us_to_ns = 1000; + const u64 s_to_ns = 1000000000; + + update_stats(&ru_stats.ru_utime_usec_stat, + (rusage->ru_utime.tv_usec * us_to_ns + rusage->ru_utime.tv_sec * s_to_ns)); + update_stats(&ru_stats.ru_stime_usec_stat, + (rusage->ru_stime.tv_usec * us_to_ns + rusage->ru_stime.tv_sec * s_to_ns)); } -static int __run_perf_stat(int argc, const char **argv) +static int __run_perf_stat(int argc, const char **argv, int run_idx) { - char msg[512]; + int interval = stat_config.interval; + int times = stat_config.times; + int timeout = stat_config.timeout; + char msg[BUFSIZ]; unsigned long long t0, t1; - struct perf_evsel *counter; - struct timespec ts; + struct evsel *counter; + size_t l; int status = 0; const bool forks = (argc > 0); - - if (interval) { - ts.tv_sec = interval / 1000; - ts.tv_nsec = (interval % 1000) * 1000000; - } else { - ts.tv_sec = 1; - ts.tv_nsec = 0; - } + bool is_pipe = STAT_RECORD ? perf_stat.data.is_pipe : false; + struct evlist_cpu_iterator evlist_cpu_itr; + struct affinity saved_affinity, *affinity = NULL; + int err, open_err = 0; + bool second_pass = false, has_supported_counters; if (forks) { - if (perf_evlist__prepare_workload(evsel_list, &target, argv, - false, false) < 0) { + if (evlist__prepare_workload(evsel_list, &target, argv, is_pipe, workload_exec_failed_signal) < 0) { perror("failed to prepare workload"); return -1; } + child_pid = evsel_list->workload.pid; } - if (group) - perf_evlist__set_leader(evsel_list); + if (!cpu_map__is_dummy(evsel_list->core.user_requested_cpus)) { + if (affinity__setup(&saved_affinity) < 0) { + err = -1; + goto err_out; + } + affinity = &saved_affinity; + } + + evlist__for_each_entry(evsel_list, counter) { + counter->reset_group = false; + if (bpf_counter__load(counter, &target)) { + err = -1; + goto err_out; + } + if (!(evsel__is_bperf(counter))) + all_counters_use_bpf = false; + } + + evlist__reset_aggr_stats(evsel_list); + + evlist__for_each_cpu(evlist_cpu_itr, evsel_list, affinity) { + counter = evlist_cpu_itr.evsel; - list_for_each_entry(counter, &evsel_list->entries, node) { - if (create_perf_stat_counter(counter) < 0) { + /* + * bperf calls evsel__open_per_cpu() in bperf__load(), so + * no need to call it again here. + */ + if (target.use_bpf) + break; + + if (counter->reset_group || !counter->supported) + continue; + if (evsel__is_bperf(counter)) + continue; + + while (true) { + if (create_perf_stat_counter(counter, &stat_config, + evlist_cpu_itr.cpu_map_idx) == 0) + break; + + open_err = errno; /* - * PPC returns ENXIO for HW counters until 2.6.37 - * (behavior changed with commit b0a873e). + * Weak group failed. We cannot just undo this here + * because earlier CPUs might be in group mode, and the kernel + * doesn't support mixing group and non group reads. Defer + * it to later. + * Don't close here because we're in the wrong affinity. */ - if (errno == EINVAL || errno == ENOSYS || - errno == ENOENT || errno == EOPNOTSUPP || - errno == ENXIO) { - if (verbose) - ui__warning("%s event is not supported by the kernel.\n", - perf_evsel__name(counter)); - counter->supported = false; + if ((open_err == EINVAL || open_err == EBADF) && + evsel__leader(counter) != counter && + counter->weak_group) { + evlist__reset_weak_group(evsel_list, counter, false); + assert(counter->reset_group); + counter->supported = true; + second_pass = true; + break; + } + + if (stat_handle_error(counter, open_err) != COUNTER_RETRY) + break; + } + } + + if (second_pass) { + /* + * Now redo all the weak group after closing them, + * and also close errored counters. + */ + + /* First close errored or weak retry */ + evlist__for_each_cpu(evlist_cpu_itr, evsel_list, affinity) { + counter = evlist_cpu_itr.evsel; + + if (!counter->reset_group && counter->supported) continue; + + perf_evsel__close_cpu(&counter->core, evlist_cpu_itr.cpu_map_idx); + } + /* Now reopen weak */ + evlist__for_each_cpu(evlist_cpu_itr, evsel_list, affinity) { + counter = evlist_cpu_itr.evsel; + + if (!counter->reset_group) + continue; + + while (true) { + pr_debug2("reopening weak %s\n", evsel__name(counter)); + if (create_perf_stat_counter(counter, &stat_config, + evlist_cpu_itr.cpu_map_idx) == 0) + break; + + open_err = errno; + if (stat_handle_error(counter, open_err) != COUNTER_RETRY) + break; } + } + } + affinity__cleanup(affinity); + affinity = NULL; - perf_evsel__open_strerror(counter, &target, - errno, msg, sizeof(msg)); - ui__error("%s\n", msg); + has_supported_counters = false; + evlist__for_each_entry(evsel_list, counter) { + if (!counter->supported) { + perf_evsel__free_fd(&counter->core); + continue; + } + has_supported_counters = true; - if (child_pid != -1) - kill(child_pid, SIGTERM); + l = strlen(counter->unit); + if (l > stat_config.unit_width) + stat_config.unit_width = l; - return -1; + if (evsel__should_store_id(counter) && + evsel__store_ids(counter, evsel_list)) { + err = -1; + goto err_out; } - counter->supported = true; } + if (!has_supported_counters && !stat_config.null_run) { + if (open_err) { + evsel__open_strerror(evlist__first(evsel_list), &target, open_err, + msg, sizeof(msg)); + } + ui__error("No supported events found.\n%s\n", msg); - if (perf_evlist__apply_filters(evsel_list)) { - error("failed to set filter with %d (%s)\n", errno, - strerror(errno)); + if (child_pid != -1) + kill(child_pid, SIGTERM); + err = -1; + goto err_out; + } + + if (evlist__apply_filters(evsel_list, &counter, &target)) { + pr_err("failed to set filter \"%s\" on event %s with %d (%s)\n", + counter->filter, evsel__name(counter), errno, + str_error_r(errno, msg, sizeof(msg))); return -1; } - /* - * Enable counters and exec the command: - */ + if (STAT_RECORD) { + int fd = perf_data__fd(&perf_stat.data); + + if (is_pipe) { + err = perf_header__write_pipe(perf_data__fd(&perf_stat.data)); + } else { + err = perf_session__write_header(perf_stat.session, evsel_list, + fd, false); + } + + if (err < 0) + goto err_out; + + err = perf_event__synthesize_stat_events(&stat_config, NULL, evsel_list, + process_synthesized_event, is_pipe); + if (err < 0) + goto err_out; + + } + + if (target.initial_delay) { + pr_info(EVLIST_DISABLED_MSG); + } else { + err = enable_counters(); + if (err) { + err = -1; + goto err_out; + } + } + + /* Exec the command, if any */ + if (forks) + evlist__start_workload(evsel_list); + + if (target.initial_delay > 0) { + usleep(target.initial_delay * USEC_PER_MSEC); + err = enable_counters(); + if (err) { + err = -1; + goto err_out; + } + + pr_info(EVLIST_ENABLED_MSG); + } + t0 = rdclock(); clock_gettime(CLOCK_MONOTONIC, &ref_time); if (forks) { - perf_evlist__start_workload(evsel_list); + if (interval || timeout || evlist__ctlfd_initialized(evsel_list)) + status = dispatch_events(forks, timeout, interval, ×); + if (child_pid != -1) { + if (timeout) + kill(child_pid, SIGTERM); + wait4(child_pid, &status, 0, &stat_config.ru_data); + } - if (interval) { - while (!waitpid(child_pid, &status, WNOHANG)) { - nanosleep(&ts, NULL); - print_interval(); - } + if (workload_exec_errno) { + const char *emsg = str_error_r(workload_exec_errno, msg, sizeof(msg)); + pr_err("Workload failed: %s\n", emsg); + err = -1; + goto err_out; } - wait(&status); - if (WIFSIGNALED(status)) + + if (WIFSIGNALED(status)) { + /* + * We want to indicate failure to stop a repeat run, + * hence negative. We want the value to be the exit code + * of perf, which for termination by a signal is 128 + * plus the signal number. + */ + err = 0 - (128 + WTERMSIG(status)); psignal(WTERMSIG(status), argv[0]); - } else { - while (!done) { - nanosleep(&ts, NULL); - if (interval) - print_interval(); + } else { + err = WEXITSTATUS(status); } + } else { + err = dispatch_events(forks, timeout, interval, ×); } + disable_counters(); + t1 = rdclock(); - update_stats(&walltime_nsecs_stats, t1 - t0); + if (stat_config.walltime_run_table) + stat_config.walltime_run[run_idx] = t1 - t0; - if (aggr_mode == AGGR_GLOBAL) { - list_for_each_entry(counter, &evsel_list->entries, node) { - read_counter_aggr(counter); - perf_evsel__close_fd(counter, perf_evsel__nr_cpus(counter), - thread_map__nr(evsel_list->threads)); - } + if (interval && stat_config.summary) { + stat_config.interval = 0; + stat_config.stop_read_counter = true; + init_stats(stat_config.walltime_nsecs_stats); + update_stats(stat_config.walltime_nsecs_stats, t1 - t0); + + evlist__copy_prev_raw_counts(evsel_list); + evlist__reset_prev_raw_counts(evsel_list); + evlist__reset_aggr_stats(evsel_list); } else { - list_for_each_entry(counter, &evsel_list->entries, node) { - read_counter(counter); - perf_evsel__close_fd(counter, perf_evsel__nr_cpus(counter), 1); - } + update_stats(stat_config.walltime_nsecs_stats, t1 - t0); + update_rusage_stats(&stat_config.ru_data); } - return WEXITSTATUS(status); + /* + * Closing a group leader splits the group, and as we only disable + * group leaders, results in remaining events becoming enabled. To + * avoid arbitrary skew, we must read all counters before closing any + * group leaders. + */ + if (read_counters() == 0) + process_counters(); + + /* + * We need to keep evsel_list alive, because it's processed + * later the evsel_list will be closed after. + */ + if (!STAT_RECORD) + evlist__close(evsel_list); + + return err; + +err_out: + if (forks) + evlist__cancel_workload(evsel_list); + + affinity__cleanup(affinity); + return err; } -static int run_perf_stat(int argc __maybe_unused, const char **argv) +/* + * Returns -1 for fatal errors which signifies to not continue + * when in repeat mode. + * + * Returns < -1 error codes when stat record is used. These + * result in the stat information being displayed, but writing + * to the file fails and is non fatal. + */ +static int run_perf_stat(int argc, const char **argv, int run_idx) { int ret; @@ -537,7 +1091,7 @@ static int run_perf_stat(int argc __maybe_unused, const char **argv) if (sync_run) sync(); - ret = __run_perf_stat(argc, argv); + ret = __run_perf_stat(argc, argv, run_idx); if (ret) return ret; @@ -550,908 +1104,1589 @@ static int run_perf_stat(int argc __maybe_unused, const char **argv) return ret; } -static void print_noise_pct(double total, double avg) +static void print_counters(struct timespec *ts, int argc, const char **argv) +{ + /* Do not print anything if we record to the pipe. */ + if (STAT_RECORD && perf_stat.data.is_pipe) + return; + if (quiet) + return; + + evlist__print_counters(evsel_list, &stat_config, &target, ts, argc, argv); +} + +static volatile sig_atomic_t signr = -1; + +static void skip_signal(int signo) { - double pct = rel_stddev_stats(total, avg); + if ((child_pid == -1) || stat_config.interval) + done = 1; - if (csv_output) - fprintf(output, "%s%.2f%%", csv_sep, pct); - else if (pct) - fprintf(output, " ( +-%6.2f%% )", pct); + signr = signo; + /* + * render child_pid harmless + * won't send SIGTERM to a random + * process in case of race condition + * and fast PID recycling + */ + child_pid = -1; } -static void print_noise(struct perf_evsel *evsel, double avg) +static void sig_atexit(void) { - struct perf_stat *ps; + sigset_t set, oset; + + /* + * avoid race condition with SIGCHLD handler + * in skip_signal() which is modifying child_pid + * goal is to avoid send SIGTERM to a random + * process + */ + sigemptyset(&set); + sigaddset(&set, SIGCHLD); + sigprocmask(SIG_BLOCK, &set, &oset); + + if (child_pid != -1) + kill(child_pid, SIGTERM); + + sigprocmask(SIG_SETMASK, &oset, NULL); - if (run_count == 1) + if (signr == -1) return; - ps = evsel->priv; - print_noise_pct(stddev_stats(&ps->res_stats[0]), avg); + signal(signr, SIG_DFL); + kill(getpid(), signr); } -static void aggr_printout(struct perf_evsel *evsel, int id, int nr) +static int stat__set_big_num(const struct option *opt __maybe_unused, + const char *s __maybe_unused, int unset) { - switch (aggr_mode) { - case AGGR_CORE: - fprintf(output, "S%d-C%*d%s%*d%s", - cpu_map__id_to_socket(id), - csv_output ? 0 : -8, - cpu_map__id_to_cpu(id), - csv_sep, - csv_output ? 0 : 4, - nr, - csv_sep); - break; - case AGGR_SOCKET: - fprintf(output, "S%*d%s%*d%s", - csv_output ? 0 : -5, - id, - csv_sep, - csv_output ? 0 : 4, - nr, - csv_sep); - break; - case AGGR_NONE: - fprintf(output, "CPU%*d%s", - csv_output ? 0 : -4, - perf_evsel__cpus(evsel)->map[id], csv_sep); - break; - case AGGR_GLOBAL: - default: - break; - } + big_num_opt = unset ? 0 : 1; + perf_stat__set_big_num(!unset); + return 0; } -static void nsec_printout(int cpu, int nr, struct perf_evsel *evsel, double avg) +static int enable_metric_only(const struct option *opt __maybe_unused, + const char *s __maybe_unused, int unset) { - double msecs = avg / 1e6; - const char *fmt = csv_output ? "%.6f%s%s" : "%18.6f%s%-25s"; - - aggr_printout(evsel, cpu, nr); + force_metric_only = true; + stat_config.metric_only = !unset; + return 0; +} - fprintf(output, fmt, msecs, csv_sep, perf_evsel__name(evsel)); +static int append_metric_groups(const struct option *opt __maybe_unused, + const char *str, + int unset __maybe_unused) +{ + if (metrics) { + char *tmp; - if (evsel->cgrp) - fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); + if (asprintf(&tmp, "%s,%s", metrics, str) < 0) + return -ENOMEM; + free(metrics); + metrics = tmp; + } else { + metrics = strdup(str); + if (!metrics) + return -ENOMEM; + } + return 0; +} - if (csv_output || interval) - return; +static int parse_control_option(const struct option *opt, + const char *str, + int unset __maybe_unused) +{ + struct perf_stat_config *config = opt->value; - if (perf_evsel__match(evsel, SOFTWARE, SW_TASK_CLOCK)) - fprintf(output, " # %8.3f CPUs utilized ", - avg / avg_stats(&walltime_nsecs_stats)); - else - fprintf(output, " "); + return evlist__parse_control(str, &config->ctl_fd, &config->ctl_fd_ack, &config->ctl_fd_close); } -/* used for get_ratio_color() */ -enum grc_type { - GRC_STALLED_CYCLES_FE, - GRC_STALLED_CYCLES_BE, - GRC_CACHE_MISSES, - GRC_MAX_NR -}; +static int parse_stat_cgroups(const struct option *opt, + const char *str, int unset) +{ + if (stat_config.cgroup_list) { + pr_err("--cgroup and --for-each-cgroup cannot be used together\n"); + return -1; + } -static const char *get_ratio_color(enum grc_type type, double ratio) + return parse_cgroups(opt, str, unset); +} + +static int parse_cputype(const struct option *opt, + const char *str, + int unset __maybe_unused) { - static const double grc_table[GRC_MAX_NR][3] = { - [GRC_STALLED_CYCLES_FE] = { 50.0, 30.0, 10.0 }, - [GRC_STALLED_CYCLES_BE] = { 75.0, 50.0, 20.0 }, - [GRC_CACHE_MISSES] = { 20.0, 10.0, 5.0 }, - }; - const char *color = PERF_COLOR_NORMAL; + const struct perf_pmu *pmu; + struct evlist *evlist = *(struct evlist **)opt->value; - if (ratio > grc_table[type][0]) - color = PERF_COLOR_RED; - else if (ratio > grc_table[type][1]) - color = PERF_COLOR_MAGENTA; - else if (ratio > grc_table[type][2]) - color = PERF_COLOR_YELLOW; + if (!list_empty(&evlist->core.entries)) { + fprintf(stderr, "Must define cputype before events/metrics\n"); + return -1; + } + + pmu = perf_pmus__pmu_for_pmu_filter(str); + if (!pmu) { + fprintf(stderr, "--cputype %s is not supported!\n", str); + return -1; + } + parse_events_option_args.pmu_filter = pmu->name; - return color; + return 0; } -static void print_stalled_cycles_frontend(int cpu, - struct perf_evsel *evsel - __maybe_unused, double avg) +static int parse_cache_level(const struct option *opt, + const char *str, + int unset __maybe_unused) { - double total, ratio = 0.0; - const char *color; + int level; + struct opt_aggr_mode *opt_aggr_mode = (struct opt_aggr_mode *)opt->value; + u32 *aggr_level = (u32 *)opt->data; + + /* + * If no string is specified, aggregate based on the topology of + * Last Level Cache (LLC). Since the LLC level can change from + * architecture to architecture, set level greater than + * MAX_CACHE_LVL which will be interpreted as LLC. + */ + if (str == NULL) { + level = MAX_CACHE_LVL + 1; + goto out; + } + + /* + * The format to specify cache level is LX or lX where X is the + * cache level. + */ + if (strlen(str) != 2 || (str[0] != 'l' && str[0] != 'L')) { + pr_err("Cache level must be of form L[1-%d], or l[1-%d]\n", + MAX_CACHE_LVL, + MAX_CACHE_LVL); + return -EINVAL; + } + + level = atoi(&str[1]); + if (level < 1) { + pr_err("Cache level must be of form L[1-%d], or l[1-%d]\n", + MAX_CACHE_LVL, + MAX_CACHE_LVL); + return -EINVAL; + } + + if (level > MAX_CACHE_LVL) { + pr_err("perf only supports max cache level of %d.\n" + "Consider increasing MAX_CACHE_LVL\n", MAX_CACHE_LVL); + return -EINVAL; + } +out: + opt_aggr_mode->cache = true; + *aggr_level = level; + return 0; +} - total = avg_stats(&runtime_cycles_stats[cpu]); +/** + * Calculate the cache instance ID from the map in + * /sys/devices/system/cpu/cpuX/cache/indexY/shared_cpu_list + * Cache instance ID is the first CPU reported in the shared_cpu_list file. + */ +static int cpu__get_cache_id_from_map(struct perf_cpu cpu, char *map) +{ + int id; + struct perf_cpu_map *cpu_map = perf_cpu_map__new(map); - if (total) - ratio = avg / total * 100.0; + /* + * If the map contains no CPU, consider the current CPU to + * be the first online CPU in the cache domain else use the + * first online CPU of the cache domain as the ID. + */ + id = perf_cpu_map__min(cpu_map).cpu; + if (id == -1) + id = cpu.cpu; - color = get_ratio_color(GRC_STALLED_CYCLES_FE, ratio); + /* Free the perf_cpu_map used to find the cache ID */ + perf_cpu_map__put(cpu_map); - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " frontend cycles idle "); + return id; } -static void print_stalled_cycles_backend(int cpu, - struct perf_evsel *evsel - __maybe_unused, double avg) +/** + * cpu__get_cache_id - Returns 0 if successful in populating the + * cache level and cache id. Cache level is read from + * /sys/devices/system/cpu/cpuX/cache/indexY/level where as cache instance ID + * is the first CPU reported by + * /sys/devices/system/cpu/cpuX/cache/indexY/shared_cpu_list + */ +static int cpu__get_cache_details(struct perf_cpu cpu, struct perf_cache *cache) { - double total, ratio = 0.0; - const char *color; + int ret = 0; + u32 cache_level = stat_config.aggr_level; + struct cpu_cache_level caches[MAX_CACHE_LVL]; + u32 i = 0, caches_cnt = 0; + + cache->cache_lvl = (cache_level > MAX_CACHE_LVL) ? 0 : cache_level; + cache->cache = -1; + + ret = build_caches_for_cpu(cpu.cpu, caches, &caches_cnt); + if (ret) { + /* + * If caches_cnt is not 0, cpu_cache_level data + * was allocated when building the topology. + * Free the allocated data before returning. + */ + if (caches_cnt) + goto free_caches; + + return ret; + } + + if (!caches_cnt) + return -1; - total = avg_stats(&runtime_cycles_stats[cpu]); + /* + * Save the data for the highest level if no + * level was specified by the user. + */ + if (cache_level > MAX_CACHE_LVL) { + int max_level_index = 0; + + for (i = 1; i < caches_cnt; ++i) { + if (caches[i].level > caches[max_level_index].level) + max_level_index = i; + } - if (total) - ratio = avg / total * 100.0; + cache->cache_lvl = caches[max_level_index].level; + cache->cache = cpu__get_cache_id_from_map(cpu, caches[max_level_index].map); - color = get_ratio_color(GRC_STALLED_CYCLES_BE, ratio); + /* Reset i to 0 to free entire caches[] */ + i = 0; + goto free_caches; + } + + for (i = 0; i < caches_cnt; ++i) { + if (caches[i].level == cache_level) { + cache->cache_lvl = cache_level; + cache->cache = cpu__get_cache_id_from_map(cpu, caches[i].map); + } - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " backend cycles idle "); + cpu_cache_level__free(&caches[i]); + } + +free_caches: + /* + * Free all the allocated cpu_cache_level data. + */ + while (i < caches_cnt) + cpu_cache_level__free(&caches[i++]); + + return ret; } -static void print_branch_misses(int cpu, - struct perf_evsel *evsel __maybe_unused, - double avg) +/** + * aggr_cpu_id__cache - Create an aggr_cpu_id with cache instache ID, cache + * level, die and socket populated with the cache instache ID, cache level, + * die and socket for cpu. The function signature is compatible with + * aggr_cpu_id_get_t. + */ +static struct aggr_cpu_id aggr_cpu_id__cache(struct perf_cpu cpu, void *data) { - double total, ratio = 0.0; - const char *color; + int ret; + struct aggr_cpu_id id; + struct perf_cache cache; - total = avg_stats(&runtime_branches_stats[cpu]); + id = aggr_cpu_id__die(cpu, data); + if (aggr_cpu_id__is_empty(&id)) + return id; - if (total) - ratio = avg / total * 100.0; + ret = cpu__get_cache_details(cpu, &cache); + if (ret) + return id; - color = get_ratio_color(GRC_CACHE_MISSES, ratio); + id.cache_lvl = cache.cache_lvl; + id.cache = cache.cache; + return id; +} - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " of all branches "); +static const char *const aggr_mode__string[] = { + [AGGR_CORE] = "core", + [AGGR_CACHE] = "cache", + [AGGR_CLUSTER] = "cluster", + [AGGR_DIE] = "die", + [AGGR_GLOBAL] = "global", + [AGGR_NODE] = "node", + [AGGR_NONE] = "none", + [AGGR_SOCKET] = "socket", + [AGGR_THREAD] = "thread", + [AGGR_UNSET] = "unset", +}; + +static struct aggr_cpu_id perf_stat__get_socket(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return aggr_cpu_id__socket(cpu, /*data=*/NULL); } -static void print_l1_dcache_misses(int cpu, - struct perf_evsel *evsel __maybe_unused, - double avg) +static struct aggr_cpu_id perf_stat__get_die(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) { - double total, ratio = 0.0; - const char *color; + return aggr_cpu_id__die(cpu, /*data=*/NULL); +} + +static struct aggr_cpu_id perf_stat__get_cache_id(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return aggr_cpu_id__cache(cpu, /*data=*/NULL); +} - total = avg_stats(&runtime_l1_dcache_stats[cpu]); +static struct aggr_cpu_id perf_stat__get_cluster(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return aggr_cpu_id__cluster(cpu, /*data=*/NULL); +} - if (total) - ratio = avg / total * 100.0; +static struct aggr_cpu_id perf_stat__get_core(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return aggr_cpu_id__core(cpu, /*data=*/NULL); +} - color = get_ratio_color(GRC_CACHE_MISSES, ratio); +static struct aggr_cpu_id perf_stat__get_node(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return aggr_cpu_id__node(cpu, /*data=*/NULL); +} - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " of all L1-dcache hits "); +static struct aggr_cpu_id perf_stat__get_global(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return aggr_cpu_id__global(cpu, /*data=*/NULL); } -static void print_l1_icache_misses(int cpu, - struct perf_evsel *evsel __maybe_unused, - double avg) +static struct aggr_cpu_id perf_stat__get_cpu(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) { - double total, ratio = 0.0; - const char *color; + return aggr_cpu_id__cpu(cpu, /*data=*/NULL); +} - total = avg_stats(&runtime_l1_icache_stats[cpu]); +static struct aggr_cpu_id perf_stat__get_aggr(struct perf_stat_config *config, + aggr_get_id_t get_id, struct perf_cpu cpu) +{ + struct aggr_cpu_id id; - if (total) - ratio = avg / total * 100.0; + /* per-process mode - should use global aggr mode */ + if (cpu.cpu == -1 || cpu.cpu >= config->cpus_aggr_map->nr) + return get_id(config, cpu); - color = get_ratio_color(GRC_CACHE_MISSES, ratio); + if (aggr_cpu_id__is_empty(&config->cpus_aggr_map->map[cpu.cpu])) + config->cpus_aggr_map->map[cpu.cpu] = get_id(config, cpu); - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " of all L1-icache hits "); + id = config->cpus_aggr_map->map[cpu.cpu]; + return id; } -static void print_dtlb_cache_misses(int cpu, - struct perf_evsel *evsel __maybe_unused, - double avg) +static struct aggr_cpu_id perf_stat__get_socket_cached(struct perf_stat_config *config, + struct perf_cpu cpu) { - double total, ratio = 0.0; - const char *color; + return perf_stat__get_aggr(config, perf_stat__get_socket, cpu); +} - total = avg_stats(&runtime_dtlb_cache_stats[cpu]); +static struct aggr_cpu_id perf_stat__get_die_cached(struct perf_stat_config *config, + struct perf_cpu cpu) +{ + return perf_stat__get_aggr(config, perf_stat__get_die, cpu); +} - if (total) - ratio = avg / total * 100.0; +static struct aggr_cpu_id perf_stat__get_cluster_cached(struct perf_stat_config *config, + struct perf_cpu cpu) +{ + return perf_stat__get_aggr(config, perf_stat__get_cluster, cpu); +} - color = get_ratio_color(GRC_CACHE_MISSES, ratio); +static struct aggr_cpu_id perf_stat__get_cache_id_cached(struct perf_stat_config *config, + struct perf_cpu cpu) +{ + return perf_stat__get_aggr(config, perf_stat__get_cache_id, cpu); +} - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " of all dTLB cache hits "); +static struct aggr_cpu_id perf_stat__get_core_cached(struct perf_stat_config *config, + struct perf_cpu cpu) +{ + return perf_stat__get_aggr(config, perf_stat__get_core, cpu); } -static void print_itlb_cache_misses(int cpu, - struct perf_evsel *evsel __maybe_unused, - double avg) +static struct aggr_cpu_id perf_stat__get_node_cached(struct perf_stat_config *config, + struct perf_cpu cpu) { - double total, ratio = 0.0; - const char *color; + return perf_stat__get_aggr(config, perf_stat__get_node, cpu); +} - total = avg_stats(&runtime_itlb_cache_stats[cpu]); +static struct aggr_cpu_id perf_stat__get_global_cached(struct perf_stat_config *config, + struct perf_cpu cpu) +{ + return perf_stat__get_aggr(config, perf_stat__get_global, cpu); +} - if (total) - ratio = avg / total * 100.0; +static struct aggr_cpu_id perf_stat__get_cpu_cached(struct perf_stat_config *config, + struct perf_cpu cpu) +{ + return perf_stat__get_aggr(config, perf_stat__get_cpu, cpu); +} - color = get_ratio_color(GRC_CACHE_MISSES, ratio); +static aggr_cpu_id_get_t aggr_mode__get_aggr(enum aggr_mode aggr_mode) +{ + switch (aggr_mode) { + case AGGR_SOCKET: + return aggr_cpu_id__socket; + case AGGR_DIE: + return aggr_cpu_id__die; + case AGGR_CLUSTER: + return aggr_cpu_id__cluster; + case AGGR_CACHE: + return aggr_cpu_id__cache; + case AGGR_CORE: + return aggr_cpu_id__core; + case AGGR_NODE: + return aggr_cpu_id__node; + case AGGR_NONE: + return aggr_cpu_id__cpu; + case AGGR_GLOBAL: + return aggr_cpu_id__global; + case AGGR_THREAD: + case AGGR_UNSET: + case AGGR_MAX: + default: + return NULL; + } +} - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " of all iTLB cache hits "); +static aggr_get_id_t aggr_mode__get_id(enum aggr_mode aggr_mode) +{ + switch (aggr_mode) { + case AGGR_SOCKET: + return perf_stat__get_socket_cached; + case AGGR_DIE: + return perf_stat__get_die_cached; + case AGGR_CLUSTER: + return perf_stat__get_cluster_cached; + case AGGR_CACHE: + return perf_stat__get_cache_id_cached; + case AGGR_CORE: + return perf_stat__get_core_cached; + case AGGR_NODE: + return perf_stat__get_node_cached; + case AGGR_NONE: + return perf_stat__get_cpu_cached; + case AGGR_GLOBAL: + return perf_stat__get_global_cached; + case AGGR_THREAD: + case AGGR_UNSET: + case AGGR_MAX: + default: + return NULL; + } } -static void print_ll_cache_misses(int cpu, - struct perf_evsel *evsel __maybe_unused, - double avg) +static int perf_stat_init_aggr_mode(void) { - double total, ratio = 0.0; - const char *color; + int nr; + aggr_cpu_id_get_t get_id = aggr_mode__get_aggr(stat_config.aggr_mode); + + if (get_id) { + bool needs_sort = stat_config.aggr_mode != AGGR_NONE; + stat_config.aggr_map = cpu_aggr_map__new(evsel_list->core.user_requested_cpus, + get_id, /*data=*/NULL, needs_sort); + if (!stat_config.aggr_map) { + pr_err("cannot build %s map\n", aggr_mode__string[stat_config.aggr_mode]); + return -1; + } + stat_config.aggr_get_id = aggr_mode__get_id(stat_config.aggr_mode); + } - total = avg_stats(&runtime_ll_cache_stats[cpu]); + if (stat_config.aggr_mode == AGGR_THREAD) { + nr = perf_thread_map__nr(evsel_list->core.threads); + stat_config.aggr_map = cpu_aggr_map__empty_new(nr); + if (stat_config.aggr_map == NULL) + return -ENOMEM; - if (total) - ratio = avg / total * 100.0; + for (int s = 0; s < nr; s++) { + struct aggr_cpu_id id = aggr_cpu_id__empty(); - color = get_ratio_color(GRC_CACHE_MISSES, ratio); + id.thread_idx = s; + stat_config.aggr_map->map[s] = id; + } + return 0; + } - fprintf(output, " # "); - color_fprintf(output, color, "%6.2f%%", ratio); - fprintf(output, " of all LL-cache hits "); + /* + * The evsel_list->cpus is the base we operate on, + * taking the highest cpu number to be the size of + * the aggregation translate cpumap. + */ + nr = perf_cpu_map__max(evsel_list->core.all_cpus).cpu + 1; + stat_config.cpus_aggr_map = cpu_aggr_map__empty_new(nr); + return stat_config.cpus_aggr_map ? 0 : -ENOMEM; } -static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg) +static void cpu_aggr_map__delete(struct cpu_aggr_map *map) { - double total, ratio = 0.0; - const char *fmt; + free(map); +} - if (csv_output) - fmt = "%.0f%s%s"; - else if (big_num) - fmt = "%'18.0f%s%-25s"; - else - fmt = "%18.0f%s%-25s"; +static void perf_stat__exit_aggr_mode(void) +{ + cpu_aggr_map__delete(stat_config.aggr_map); + cpu_aggr_map__delete(stat_config.cpus_aggr_map); + stat_config.aggr_map = NULL; + stat_config.cpus_aggr_map = NULL; +} + +static struct aggr_cpu_id perf_env__get_socket_aggr_by_cpu(struct perf_cpu cpu, void *data) +{ + struct perf_env *env = data; + struct aggr_cpu_id id = aggr_cpu_id__empty(); + + if (cpu.cpu != -1) + id.socket = env->cpu[cpu.cpu].socket_id; - aggr_printout(evsel, cpu, nr); + return id; +} - if (aggr_mode == AGGR_GLOBAL) - cpu = 0; +static struct aggr_cpu_id perf_env__get_die_aggr_by_cpu(struct perf_cpu cpu, void *data) +{ + struct perf_env *env = data; + struct aggr_cpu_id id = aggr_cpu_id__empty(); + + if (cpu.cpu != -1) { + /* + * die_id is relative to socket, so start + * with the socket ID and then add die to + * make a unique ID. + */ + id.socket = env->cpu[cpu.cpu].socket_id; + id.die = env->cpu[cpu.cpu].die_id; + } - fprintf(output, fmt, avg, csv_sep, perf_evsel__name(evsel)); + return id; +} + +static void perf_env__get_cache_id_for_cpu(struct perf_cpu cpu, struct perf_env *env, + u32 cache_level, struct aggr_cpu_id *id) +{ + int i; + int caches_cnt = env->caches_cnt; + struct cpu_cache_level *caches = env->caches; - if (evsel->cgrp) - fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); + id->cache_lvl = (cache_level > MAX_CACHE_LVL) ? 0 : cache_level; + id->cache = -1; - if (csv_output || interval) + if (!caches_cnt) return; - if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) { - total = avg_stats(&runtime_cycles_stats[cpu]); - if (total) - ratio = avg / total; + for (i = caches_cnt - 1; i > -1; --i) { + struct perf_cpu_map *cpu_map; + int map_contains_cpu; - fprintf(output, " # %5.2f insns per cycle ", ratio); + /* + * If user has not specified a level, find the fist level with + * the cpu in the map. Since building the map is expensive, do + * this only if levels match. + */ + if (cache_level <= MAX_CACHE_LVL && caches[i].level != cache_level) + continue; - total = avg_stats(&runtime_stalled_cycles_front_stats[cpu]); - total = max(total, avg_stats(&runtime_stalled_cycles_back_stats[cpu])); + cpu_map = perf_cpu_map__new(caches[i].map); + map_contains_cpu = perf_cpu_map__idx(cpu_map, cpu); + perf_cpu_map__put(cpu_map); - if (total && avg) { - ratio = total / avg; - fprintf(output, "\n # %5.2f stalled cycles per insn", ratio); + if (map_contains_cpu != -1) { + id->cache_lvl = caches[i].level; + id->cache = cpu__get_cache_id_from_map(cpu, caches[i].map); + return; } + } +} - } else if (perf_evsel__match(evsel, HARDWARE, HW_BRANCH_MISSES) && - runtime_branches_stats[cpu].n != 0) { - print_branch_misses(cpu, evsel, avg); - } else if ( - evsel->attr.type == PERF_TYPE_HW_CACHE && - evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1D | - ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | - ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && - runtime_l1_dcache_stats[cpu].n != 0) { - print_l1_dcache_misses(cpu, evsel, avg); - } else if ( - evsel->attr.type == PERF_TYPE_HW_CACHE && - evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1I | - ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | - ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && - runtime_l1_icache_stats[cpu].n != 0) { - print_l1_icache_misses(cpu, evsel, avg); - } else if ( - evsel->attr.type == PERF_TYPE_HW_CACHE && - evsel->attr.config == ( PERF_COUNT_HW_CACHE_DTLB | - ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | - ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && - runtime_dtlb_cache_stats[cpu].n != 0) { - print_dtlb_cache_misses(cpu, evsel, avg); - } else if ( - evsel->attr.type == PERF_TYPE_HW_CACHE && - evsel->attr.config == ( PERF_COUNT_HW_CACHE_ITLB | - ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | - ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && - runtime_itlb_cache_stats[cpu].n != 0) { - print_itlb_cache_misses(cpu, evsel, avg); - } else if ( - evsel->attr.type == PERF_TYPE_HW_CACHE && - evsel->attr.config == ( PERF_COUNT_HW_CACHE_LL | - ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | - ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && - runtime_ll_cache_stats[cpu].n != 0) { - print_ll_cache_misses(cpu, evsel, avg); - } else if (perf_evsel__match(evsel, HARDWARE, HW_CACHE_MISSES) && - runtime_cacherefs_stats[cpu].n != 0) { - total = avg_stats(&runtime_cacherefs_stats[cpu]); - - if (total) - ratio = avg * 100 / total; - - fprintf(output, " # %8.3f %% of all cache refs ", ratio); - - } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_FRONTEND)) { - print_stalled_cycles_frontend(cpu, evsel, avg); - } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_BACKEND)) { - print_stalled_cycles_backend(cpu, evsel, avg); - } else if (perf_evsel__match(evsel, HARDWARE, HW_CPU_CYCLES)) { - total = avg_stats(&runtime_nsecs_stats[cpu]); - - if (total) - ratio = 1.0 * avg / total; - - fprintf(output, " # %8.3f GHz ", ratio); - } else if (runtime_nsecs_stats[cpu].n != 0) { - char unit = 'M'; - - total = avg_stats(&runtime_nsecs_stats[cpu]); - - if (total) - ratio = 1000.0 * avg / total; - if (ratio < 0.001) { - ratio *= 1000; - unit = 'K'; - } +static struct aggr_cpu_id perf_env__get_cache_aggr_by_cpu(struct perf_cpu cpu, + void *data) +{ + struct perf_env *env = data; + struct aggr_cpu_id id = aggr_cpu_id__empty(); - fprintf(output, " # %8.3f %c/sec ", ratio, unit); - } else { - fprintf(output, " "); + if (cpu.cpu != -1) { + u32 cache_level = (perf_stat.aggr_level) ?: stat_config.aggr_level; + + id.socket = env->cpu[cpu.cpu].socket_id; + id.die = env->cpu[cpu.cpu].die_id; + perf_env__get_cache_id_for_cpu(cpu, env, cache_level, &id); } + + return id; } -static void print_aggr(char *prefix) +static struct aggr_cpu_id perf_env__get_cluster_aggr_by_cpu(struct perf_cpu cpu, + void *data) { - struct perf_evsel *counter; - int cpu, cpu2, s, s2, id, nr; - u64 ena, run, val; + struct perf_env *env = data; + struct aggr_cpu_id id = aggr_cpu_id__empty(); - if (!(aggr_map || aggr_get_id)) - return; + if (cpu.cpu != -1) { + id.socket = env->cpu[cpu.cpu].socket_id; + id.die = env->cpu[cpu.cpu].die_id; + id.cluster = env->cpu[cpu.cpu].cluster_id; + } - for (s = 0; s < aggr_map->nr; s++) { - id = aggr_map->map[s]; - list_for_each_entry(counter, &evsel_list->entries, node) { - val = ena = run = 0; - nr = 0; - for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) { - cpu2 = perf_evsel__cpus(counter)->map[cpu]; - s2 = aggr_get_id(evsel_list->cpus, cpu2); - if (s2 != id) - continue; - val += counter->counts->cpu[cpu].val; - ena += counter->counts->cpu[cpu].ena; - run += counter->counts->cpu[cpu].run; - nr++; - } - if (prefix) - fprintf(output, "%s", prefix); + return id; +} - if (run == 0 || ena == 0) { - aggr_printout(counter, id, nr); +static struct aggr_cpu_id perf_env__get_core_aggr_by_cpu(struct perf_cpu cpu, void *data) +{ + struct perf_env *env = data; + struct aggr_cpu_id id = aggr_cpu_id__empty(); + + if (cpu.cpu != -1) { + /* + * core_id is relative to socket, die and cluster, we need a + * global id. So we set socket, die id, cluster id and core id. + */ + id.socket = env->cpu[cpu.cpu].socket_id; + id.die = env->cpu[cpu.cpu].die_id; + id.cluster = env->cpu[cpu.cpu].cluster_id; + id.core = env->cpu[cpu.cpu].core_id; + } - fprintf(output, "%*s%s%*s", - csv_output ? 0 : 18, - counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED, - csv_sep, - csv_output ? 0 : -24, - perf_evsel__name(counter)); + return id; +} - if (counter->cgrp) - fprintf(output, "%s%s", - csv_sep, counter->cgrp->name); +static struct aggr_cpu_id perf_env__get_cpu_aggr_by_cpu(struct perf_cpu cpu, void *data) +{ + struct perf_env *env = data; + struct aggr_cpu_id id = aggr_cpu_id__empty(); + + if (cpu.cpu != -1) { + /* + * core_id is relative to socket and die, + * we need a global id. So we set + * socket, die id and core id + */ + id.socket = env->cpu[cpu.cpu].socket_id; + id.die = env->cpu[cpu.cpu].die_id; + id.core = env->cpu[cpu.cpu].core_id; + id.cpu = cpu; + } - fputc('\n', output); - continue; - } + return id; +} - if (nsec_counter(counter)) - nsec_printout(id, nr, counter, val); - else - abs_printout(id, nr, counter, val); +static struct aggr_cpu_id perf_env__get_node_aggr_by_cpu(struct perf_cpu cpu, void *data) +{ + struct aggr_cpu_id id = aggr_cpu_id__empty(); - if (!csv_output) { - print_noise(counter, 1.0); + id.node = perf_env__numa_node(data, cpu); + return id; +} - if (run != ena) - fprintf(output, " (%.2f%%)", - 100.0 * run / ena); - } - fputc('\n', output); - } - } +static struct aggr_cpu_id perf_env__get_global_aggr_by_cpu(struct perf_cpu cpu __maybe_unused, + void *data __maybe_unused) +{ + struct aggr_cpu_id id = aggr_cpu_id__empty(); + + /* it always aggregates to the cpu 0 */ + id.cpu = (struct perf_cpu){ .cpu = 0 }; + return id; } -/* - * Print out the results of a single counter: - * aggregated counts in system-wide mode - */ -static void print_counter_aggr(struct perf_evsel *counter, char *prefix) +static struct aggr_cpu_id perf_stat__get_socket_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return perf_env__get_socket_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} +static struct aggr_cpu_id perf_stat__get_die_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return perf_env__get_die_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} + +static struct aggr_cpu_id perf_stat__get_cluster_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) { - struct perf_stat *ps = counter->priv; - double avg = avg_stats(&ps->res_stats[0]); - int scaled = counter->counts->scaled; + return perf_env__get_cluster_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} - if (prefix) - fprintf(output, "%s", prefix); +static struct aggr_cpu_id perf_stat__get_cache_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return perf_env__get_cache_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} - if (scaled == -1) { - fprintf(output, "%*s%s%*s", - csv_output ? 0 : 18, - counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED, - csv_sep, - csv_output ? 0 : -24, - perf_evsel__name(counter)); +static struct aggr_cpu_id perf_stat__get_core_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return perf_env__get_core_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} - if (counter->cgrp) - fprintf(output, "%s%s", csv_sep, counter->cgrp->name); +static struct aggr_cpu_id perf_stat__get_cpu_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return perf_env__get_cpu_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} - fputc('\n', output); - return; +static struct aggr_cpu_id perf_stat__get_node_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return perf_env__get_node_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} + +static struct aggr_cpu_id perf_stat__get_global_file(struct perf_stat_config *config __maybe_unused, + struct perf_cpu cpu) +{ + return perf_env__get_global_aggr_by_cpu(cpu, perf_session__env(perf_stat.session)); +} + +static aggr_cpu_id_get_t aggr_mode__get_aggr_file(enum aggr_mode aggr_mode) +{ + switch (aggr_mode) { + case AGGR_SOCKET: + return perf_env__get_socket_aggr_by_cpu; + case AGGR_DIE: + return perf_env__get_die_aggr_by_cpu; + case AGGR_CLUSTER: + return perf_env__get_cluster_aggr_by_cpu; + case AGGR_CACHE: + return perf_env__get_cache_aggr_by_cpu; + case AGGR_CORE: + return perf_env__get_core_aggr_by_cpu; + case AGGR_NODE: + return perf_env__get_node_aggr_by_cpu; + case AGGR_GLOBAL: + return perf_env__get_global_aggr_by_cpu; + case AGGR_NONE: + return perf_env__get_cpu_aggr_by_cpu; + case AGGR_THREAD: + case AGGR_UNSET: + case AGGR_MAX: + default: + return NULL; } +} - if (nsec_counter(counter)) - nsec_printout(-1, 0, counter, avg); - else - abs_printout(-1, 0, counter, avg); +static aggr_get_id_t aggr_mode__get_id_file(enum aggr_mode aggr_mode) +{ + switch (aggr_mode) { + case AGGR_SOCKET: + return perf_stat__get_socket_file; + case AGGR_DIE: + return perf_stat__get_die_file; + case AGGR_CLUSTER: + return perf_stat__get_cluster_file; + case AGGR_CACHE: + return perf_stat__get_cache_file; + case AGGR_CORE: + return perf_stat__get_core_file; + case AGGR_NODE: + return perf_stat__get_node_file; + case AGGR_GLOBAL: + return perf_stat__get_global_file; + case AGGR_NONE: + return perf_stat__get_cpu_file; + case AGGR_THREAD: + case AGGR_UNSET: + case AGGR_MAX: + default: + return NULL; + } +} + +static int perf_stat_init_aggr_mode_file(struct perf_stat *st) +{ + struct perf_env *env = perf_session__env(st->session); + aggr_cpu_id_get_t get_id = aggr_mode__get_aggr_file(stat_config.aggr_mode); + bool needs_sort = stat_config.aggr_mode != AGGR_NONE; - print_noise(counter, avg); + if (stat_config.aggr_mode == AGGR_THREAD) { + int nr = perf_thread_map__nr(evsel_list->core.threads); - if (csv_output) { - fputc('\n', output); - return; + stat_config.aggr_map = cpu_aggr_map__empty_new(nr); + if (stat_config.aggr_map == NULL) + return -ENOMEM; + + for (int s = 0; s < nr; s++) { + struct aggr_cpu_id id = aggr_cpu_id__empty(); + + id.thread_idx = s; + stat_config.aggr_map->map[s] = id; + } + return 0; } - if (scaled) { - double avg_enabled, avg_running; + if (!get_id) + return 0; - avg_enabled = avg_stats(&ps->res_stats[1]); - avg_running = avg_stats(&ps->res_stats[2]); + stat_config.aggr_map = cpu_aggr_map__new(evsel_list->core.user_requested_cpus, + get_id, env, needs_sort); + if (!stat_config.aggr_map) { + pr_err("cannot build %s map\n", aggr_mode__string[stat_config.aggr_mode]); + return -1; + } + stat_config.aggr_get_id = aggr_mode__get_id_file(stat_config.aggr_mode); + return 0; +} - fprintf(output, " [%5.2f%%]", 100 * avg_running / avg_enabled); +static int default_evlist_evsel_cmp(void *priv __maybe_unused, + const struct list_head *l, + const struct list_head *r) +{ + const struct perf_evsel *lhs_core = container_of(l, struct perf_evsel, node); + const struct evsel *lhs = container_of(lhs_core, struct evsel, core); + const struct perf_evsel *rhs_core = container_of(r, struct perf_evsel, node); + const struct evsel *rhs = container_of(rhs_core, struct evsel, core); + + if (evsel__leader(lhs) == evsel__leader(rhs)) { + /* Within the same group, respect the original order. */ + return lhs_core->idx - rhs_core->idx; } - fprintf(output, "\n"); + + /* Sort default metrics evsels first, and default show events before those. */ + if (lhs->default_metricgroup != rhs->default_metricgroup) + return lhs->default_metricgroup ? -1 : 1; + + if (lhs->default_show_events != rhs->default_show_events) + return lhs->default_show_events ? -1 : 1; + + /* Sort by PMU type (prefers legacy types first). */ + if (lhs->pmu != rhs->pmu) + return lhs->pmu->type - rhs->pmu->type; + + /* Sort by name. */ + return strcmp(evsel__name((struct evsel *)lhs), evsel__name((struct evsel *)rhs)); } /* - * Print out the results of a single counter: - * does not use aggregated count in system-wide + * Add default events, if there were no attributes specified or + * if -d/--detailed, -d -d or -d -d -d is used: */ -static void print_counter(struct perf_evsel *counter, char *prefix) -{ - u64 ena, run, val; - int cpu; - - for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) { - val = counter->counts->cpu[cpu].val; - ena = counter->counts->cpu[cpu].ena; - run = counter->counts->cpu[cpu].run; - - if (prefix) - fprintf(output, "%s", prefix); - - if (run == 0 || ena == 0) { - fprintf(output, "CPU%*d%s%*s%s%*s", - csv_output ? 0 : -4, - perf_evsel__cpus(counter)->map[cpu], csv_sep, - csv_output ? 0 : 18, - counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED, - csv_sep, - csv_output ? 0 : -24, - perf_evsel__name(counter)); - - if (counter->cgrp) - fprintf(output, "%s%s", - csv_sep, counter->cgrp->name); - - fputc('\n', output); - continue; +static int add_default_events(void) +{ + const char *pmu = parse_events_option_args.pmu_filter ?: "all"; + struct parse_events_error err; + struct evlist *evlist = evlist__new(); + struct evsel *evsel; + int ret = 0; + + if (!evlist) + return -ENOMEM; + + parse_events_error__init(&err); + + /* Set attrs if no event is selected and !null_run: */ + if (stat_config.null_run) + goto out; + + if (transaction_run) { + /* Handle -T as -M transaction. Once platform specific metrics + * support has been added to the json files, all architectures + * will use this approach. To determine transaction support + * on an architecture test for such a metric name. + */ + if (!metricgroup__has_metric_or_groups(pmu, "transaction")) { + pr_err("Missing transaction metrics\n"); + ret = -1; + goto out; } + ret = metricgroup__parse_groups(evlist, pmu, "transaction", + stat_config.metric_no_group, + stat_config.metric_no_merge, + stat_config.metric_no_threshold, + stat_config.user_requested_cpu_list, + stat_config.system_wide, + stat_config.hardware_aware_grouping); + goto out; + } - if (nsec_counter(counter)) - nsec_printout(cpu, 0, counter, val); - else - abs_printout(cpu, 0, counter, val); + if (smi_cost) { + int smi; + + if (sysfs__read_int(FREEZE_ON_SMI_PATH, &smi) < 0) { + pr_err("freeze_on_smi is not supported.\n"); + ret = -1; + goto out; + } - if (!csv_output) { - print_noise(counter, 1.0); + if (!smi) { + if (sysfs__write_int(FREEZE_ON_SMI_PATH, 1) < 0) { + pr_err("Failed to set freeze_on_smi.\n"); + ret = -1; + goto out; + } + smi_reset = true; + } - if (run != ena) - fprintf(output, " (%.2f%%)", - 100.0 * run / ena); + if (!metricgroup__has_metric_or_groups(pmu, "smi")) { + pr_err("Missing smi metrics\n"); + ret = -1; + goto out; } - fputc('\n', output); + + if (!force_metric_only) + stat_config.metric_only = true; + + ret = metricgroup__parse_groups(evlist, pmu, "smi", + stat_config.metric_no_group, + stat_config.metric_no_merge, + stat_config.metric_no_threshold, + stat_config.user_requested_cpu_list, + stat_config.system_wide, + stat_config.hardware_aware_grouping); + goto out; } -} -static void print_stat(int argc, const char **argv) -{ - struct perf_evsel *counter; - int i; + if (topdown_run) { + unsigned int max_level = metricgroups__topdown_max_level(); + char str[] = "TopdownL1"; - fflush(stdout); - - if (!csv_output) { - fprintf(output, "\n"); - fprintf(output, " Performance counter stats for "); - if (!perf_target__has_task(&target)) { - fprintf(output, "\'%s", argv[0]); - for (i = 1; i < argc; i++) - fprintf(output, " %s", argv[i]); - } else if (target.pid) - fprintf(output, "process id \'%s", target.pid); - else - fprintf(output, "thread id \'%s", target.tid); + if (!force_metric_only) + stat_config.metric_only = true; - fprintf(output, "\'"); - if (run_count > 1) - fprintf(output, " (%d runs)", run_count); - fprintf(output, ":\n\n"); + if (!max_level) { + pr_err("Topdown requested but the topdown metric groups aren't present.\n" + "(See perf list the metric groups have names like TopdownL1)\n"); + ret = -1; + goto out; + } + if (stat_config.topdown_level > max_level) { + pr_err("Invalid top-down metrics level. The max level is %u.\n", max_level); + ret = -1; + goto out; + } else if (!stat_config.topdown_level) { + stat_config.topdown_level = 1; + } + if (!stat_config.interval && !stat_config.metric_only) { + fprintf(stat_config.output, + "Topdown accuracy may decrease when measuring long periods.\n" + "Please print the result regularly, e.g. -I1000\n"); + } + str[8] = stat_config.topdown_level + '0'; + if (metricgroup__parse_groups(evlist, + pmu, str, + /*metric_no_group=*/false, + /*metric_no_merge=*/false, + /*metric_no_threshold=*/true, + stat_config.user_requested_cpu_list, + stat_config.system_wide, + stat_config.hardware_aware_grouping) < 0) { + ret = -1; + goto out; + } } - switch (aggr_mode) { - case AGGR_CORE: - case AGGR_SOCKET: - print_aggr(NULL); - break; - case AGGR_GLOBAL: - list_for_each_entry(counter, &evsel_list->entries, node) - print_counter_aggr(counter, NULL); - break; - case AGGR_NONE: - list_for_each_entry(counter, &evsel_list->entries, node) - print_counter(counter, NULL); - break; - default: - break; - } - - if (!csv_output) { - if (!null_run) - fprintf(output, "\n"); - fprintf(output, " %17.9f seconds time elapsed", - avg_stats(&walltime_nsecs_stats)/1e9); - if (run_count > 1) { - fprintf(output, " "); - print_noise_pct(stddev_stats(&walltime_nsecs_stats), - avg_stats(&walltime_nsecs_stats)); + if (!stat_config.topdown_level) + stat_config.topdown_level = 1; + + if (!evlist->core.nr_entries && !evsel_list->core.nr_entries) { + /* + * Add Default metrics. To minimize multiplexing, don't request + * threshold computation, but it will be computed if the events + * are present. + */ + const char *default_metricgroup_names[] = { + "Default", "Default2", "Default3", "Default4", + }; + + for (size_t i = 0; i < ARRAY_SIZE(default_metricgroup_names); i++) { + struct evlist *metric_evlist; + + if (!metricgroup__has_metric_or_groups(pmu, default_metricgroup_names[i])) + continue; + + if ((int)i > detailed_run) + break; + + metric_evlist = evlist__new(); + if (!metric_evlist) { + ret = -ENOMEM; + break; + } + if (metricgroup__parse_groups(metric_evlist, pmu, default_metricgroup_names[i], + /*metric_no_group=*/false, + /*metric_no_merge=*/false, + /*metric_no_threshold=*/true, + stat_config.user_requested_cpu_list, + stat_config.system_wide, + stat_config.hardware_aware_grouping) < 0) { + evlist__delete(metric_evlist); + ret = -1; + break; + } + + evlist__for_each_entry(metric_evlist, evsel) + evsel->default_metricgroup = true; + + evlist__splice_list_tail(evlist, &metric_evlist->core.entries); + metricgroup__copy_metric_events(evlist, /*cgrp=*/NULL, + &evlist->metric_events, + &metric_evlist->metric_events); + evlist__delete(metric_evlist); } - fprintf(output, "\n\n"); + list_sort(/*priv=*/NULL, &evlist->core.entries, default_evlist_evsel_cmp); + } +out: + if (!ret) { + evlist__for_each_entry(evlist, evsel) { + /* + * Make at least one event non-skippable so fatal errors are visible. + * 'cycles' always used to be default and non-skippable, so use that. + */ + if (!evsel__match(evsel, HARDWARE, HW_CPU_CYCLES)) + evsel->skippable = true; + } + } + parse_events_error__exit(&err); + evlist__splice_list_tail(evsel_list, &evlist->core.entries); + metricgroup__copy_metric_events(evsel_list, /*cgrp=*/NULL, + &evsel_list->metric_events, + &evlist->metric_events); + evlist__delete(evlist); + return ret; } -static volatile int signr = -1; +static const char * const stat_record_usage[] = { + "perf stat record [<options>]", + NULL, +}; -static void skip_signal(int signo) +static void init_features(struct perf_session *session) { - if ((child_pid == -1) || interval) - done = 1; + int feat; - signr = signo; - /* - * render child_pid harmless - * won't send SIGTERM to a random - * process in case of race condition - * and fast PID recycling - */ - child_pid = -1; + for (feat = HEADER_FIRST_FEATURE; feat < HEADER_LAST_FEATURE; feat++) + perf_header__set_feat(&session->header, feat); + + perf_header__clear_feat(&session->header, HEADER_DIR_FORMAT); + perf_header__clear_feat(&session->header, HEADER_BUILD_ID); + perf_header__clear_feat(&session->header, HEADER_TRACING_DATA); + perf_header__clear_feat(&session->header, HEADER_BRANCH_STACK); + perf_header__clear_feat(&session->header, HEADER_AUXTRACE); } -static void sig_atexit(void) +static int __cmd_record(const struct option stat_options[], struct opt_aggr_mode *opt_mode, + int argc, const char **argv) { - sigset_t set, oset; + struct perf_session *session; + struct perf_data *data = &perf_stat.data; - /* - * avoid race condition with SIGCHLD handler - * in skip_signal() which is modifying child_pid - * goal is to avoid send SIGTERM to a random - * process - */ - sigemptyset(&set); - sigaddset(&set, SIGCHLD); - sigprocmask(SIG_BLOCK, &set, &oset); + argc = parse_options(argc, argv, stat_options, stat_record_usage, + PARSE_OPT_STOP_AT_NON_OPTION); + stat_config.aggr_mode = opt_aggr_mode_to_aggr_mode(opt_mode); - if (child_pid != -1) - kill(child_pid, SIGTERM); + if (output_name) + data->path = output_name; - sigprocmask(SIG_SETMASK, &oset, NULL); + if (stat_config.run_count != 1 || forever) { + pr_err("Cannot use -r option with perf stat record.\n"); + return -1; + } - if (signr == -1) - return; + session = perf_session__new(data, NULL); + if (IS_ERR(session)) { + pr_err("Perf session creation failed\n"); + return PTR_ERR(session); + } - signal(signr, SIG_DFL); - kill(getpid(), signr); + init_features(session); + + session->evlist = evsel_list; + perf_stat.session = session; + perf_stat.record = true; + return argc; } -static int stat__set_big_num(const struct option *opt __maybe_unused, - const char *s __maybe_unused, int unset) +static int process_stat_round_event(const struct perf_tool *tool __maybe_unused, + struct perf_session *session, + union perf_event *event) { - big_num_opt = unset ? 0 : 1; + struct perf_record_stat_round *stat_round = &event->stat_round; + struct timespec tsh, *ts = NULL; + struct perf_env *env = perf_session__env(session); + const char **argv = env->cmdline_argv; + int argc = env->nr_cmdline; + + process_counters(); + + if (stat_round->type == PERF_STAT_ROUND_TYPE__FINAL) + update_stats(stat_config.walltime_nsecs_stats, stat_round->time); + + if (stat_config.interval && stat_round->time) { + tsh.tv_sec = stat_round->time / NSEC_PER_SEC; + tsh.tv_nsec = stat_round->time % NSEC_PER_SEC; + ts = &tsh; + } + + print_counters(ts, argc, argv); return 0; } -static int perf_stat_init_aggr_mode(void) +static +int process_stat_config_event(const struct perf_tool *tool, + struct perf_session *session, + union perf_event *event) { - switch (aggr_mode) { - case AGGR_SOCKET: - if (cpu_map__build_socket_map(evsel_list->cpus, &aggr_map)) { - perror("cannot build socket map"); - return -1; - } - aggr_get_id = cpu_map__get_socket; - break; - case AGGR_CORE: - if (cpu_map__build_core_map(evsel_list->cpus, &aggr_map)) { - perror("cannot build core map"); + struct perf_stat *st = container_of(tool, struct perf_stat, tool); + + perf_event__read_stat_config(&stat_config, &event->stat_config); + + if (perf_cpu_map__is_empty(st->cpus)) { + if (st->aggr_mode != AGGR_UNSET) + pr_warning("warning: processing task data, aggregation mode not set\n"); + } else if (st->aggr_mode != AGGR_UNSET) { + stat_config.aggr_mode = st->aggr_mode; + } + + if (perf_stat.data.is_pipe) + perf_stat_init_aggr_mode(); + else + perf_stat_init_aggr_mode_file(st); + + if (stat_config.aggr_map) { + int nr_aggr = stat_config.aggr_map->nr; + + if (evlist__alloc_aggr_stats(session->evlist, nr_aggr) < 0) { + pr_err("cannot allocate aggr counts\n"); return -1; } - aggr_get_id = cpu_map__get_core; - break; - case AGGR_NONE: - case AGGR_GLOBAL: - default: - break; } return 0; } +static int set_maps(struct perf_stat *st) +{ + if (!st->cpus || !st->threads) + return 0; -/* - * Add default attributes, if there were no attributes specified or - * if -d/--detailed, -d -d or -d -d -d is used: - */ -static int add_default_attributes(void) + if (WARN_ONCE(st->maps_allocated, "stats double allocation\n")) + return -EINVAL; + + perf_evlist__set_maps(&evsel_list->core, st->cpus, st->threads); + + if (evlist__alloc_stats(&stat_config, evsel_list, /*alloc_raw=*/true)) + return -ENOMEM; + + st->maps_allocated = true; + return 0; +} + +static +int process_thread_map_event(const struct perf_tool *tool, + struct perf_session *session __maybe_unused, + union perf_event *event) { - struct perf_event_attr default_attrs[] = { + struct perf_stat *st = container_of(tool, struct perf_stat, tool); - { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_TASK_CLOCK }, - { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CONTEXT_SWITCHES }, - { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CPU_MIGRATIONS }, - { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_PAGE_FAULTS }, + if (st->threads) { + pr_warning("Extra thread map event, ignoring.\n"); + return 0; + } - { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CPU_CYCLES }, - { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_STALLED_CYCLES_FRONTEND }, - { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_STALLED_CYCLES_BACKEND }, - { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_INSTRUCTIONS }, - { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS }, - { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_MISSES }, + st->threads = thread_map__new_event(&event->thread_map); + if (!st->threads) + return -ENOMEM; -}; + return set_maps(st); +} -/* - * Detailed stats (-d), covering the L1 and last level data caches: - */ - struct perf_event_attr detailed_attrs[] = { - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_L1D << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_L1D << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_LL << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_LL << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, -}; +static +int process_cpu_map_event(const struct perf_tool *tool, + struct perf_session *session __maybe_unused, + union perf_event *event) +{ + struct perf_stat *st = container_of(tool, struct perf_stat, tool); + struct perf_cpu_map *cpus; -/* - * Very detailed stats (-d -d), covering the instruction cache and the TLB caches: - */ - struct perf_event_attr very_detailed_attrs[] = { - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_L1I << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_L1I << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_DTLB << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_DTLB << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_ITLB << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_ITLB << 0 | - (PERF_COUNT_HW_CACHE_OP_READ << 8) | - (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, + if (st->cpus) { + pr_warning("Extra cpu map event, ignoring.\n"); + return 0; + } + + cpus = cpu_map__new_data(&event->cpu_map.data); + if (!cpus) + return -ENOMEM; + st->cpus = cpus; + return set_maps(st); +} + +static const char * const stat_report_usage[] = { + "perf stat report [<options>]", + NULL, }; -/* - * Very, very detailed stats (-d -d -d), adding prefetch events: - */ - struct perf_event_attr very_very_detailed_attrs[] = { - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_L1D << 0 | - (PERF_COUNT_HW_CACHE_OP_PREFETCH << 8) | - (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, - - { .type = PERF_TYPE_HW_CACHE, - .config = - PERF_COUNT_HW_CACHE_L1D << 0 | - (PERF_COUNT_HW_CACHE_OP_PREFETCH << 8) | - (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, +static struct perf_stat perf_stat = { + .aggr_mode = AGGR_UNSET, + .aggr_level = 0, }; - /* Set attrs if no event is selected and !null_run: */ - if (null_run) - return 0; +static int __cmd_report(int argc, const char **argv) +{ + struct perf_session *session; + const struct option options[] = { + OPT_STRING('i', "input", &input_name, "file", "input file name"), + OPT_SET_UINT(0, "per-socket", &perf_stat.aggr_mode, + "aggregate counts per processor socket", AGGR_SOCKET), + OPT_SET_UINT(0, "per-die", &perf_stat.aggr_mode, + "aggregate counts per processor die", AGGR_DIE), + OPT_SET_UINT(0, "per-cluster", &perf_stat.aggr_mode, + "aggregate counts perf processor cluster", AGGR_CLUSTER), + OPT_CALLBACK_OPTARG(0, "per-cache", &perf_stat.aggr_mode, &perf_stat.aggr_level, + "cache level", + "aggregate count at this cache level (Default: LLC)", + parse_cache_level), + OPT_SET_UINT(0, "per-core", &perf_stat.aggr_mode, + "aggregate counts per physical processor core", AGGR_CORE), + OPT_SET_UINT(0, "per-node", &perf_stat.aggr_mode, + "aggregate counts per numa node", AGGR_NODE), + OPT_SET_UINT('A', "no-aggr", &perf_stat.aggr_mode, + "disable CPU count aggregation", AGGR_NONE), + OPT_END() + }; + struct stat st; + int ret; - if (!evsel_list->nr_entries) { - if (perf_evlist__add_default_attrs(evsel_list, default_attrs) < 0) - return -1; + argc = parse_options(argc, argv, options, stat_report_usage, 0); + + if (!input_name || !strlen(input_name)) { + if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) + input_name = "-"; + else + input_name = "perf.data"; } - /* Detailed events get appended to the event list: */ + perf_stat.data.path = input_name; + perf_stat.data.mode = PERF_DATA_MODE_READ; - if (detailed_run < 1) - return 0; + perf_tool__init(&perf_stat.tool, /*ordered_events=*/false); + perf_stat.tool.attr = perf_event__process_attr; + perf_stat.tool.event_update = perf_event__process_event_update; + perf_stat.tool.thread_map = process_thread_map_event; + perf_stat.tool.cpu_map = process_cpu_map_event; + perf_stat.tool.stat_config = process_stat_config_event; + perf_stat.tool.stat = perf_event__process_stat_event; + perf_stat.tool.stat_round = process_stat_round_event; - /* Append detailed run extra attributes: */ - if (perf_evlist__add_default_attrs(evsel_list, detailed_attrs) < 0) - return -1; + session = perf_session__new(&perf_stat.data, &perf_stat.tool); + if (IS_ERR(session)) + return PTR_ERR(session); - if (detailed_run < 2) - return 0; + perf_stat.session = session; + stat_config.output = stderr; + evlist__delete(evsel_list); + evsel_list = session->evlist; - /* Append very detailed run extra attributes: */ - if (perf_evlist__add_default_attrs(evsel_list, very_detailed_attrs) < 0) - return -1; + ret = perf_session__process_events(session); + if (ret) + return ret; - if (detailed_run < 3) - return 0; + perf_session__delete(session); + return 0; +} + +static void setup_system_wide(int forks) +{ + /* + * Make system wide (-a) the default target if + * no target was specified and one of following + * conditions is met: + * + * - there's no workload specified + * - there is workload specified but all requested + * events are system wide events + */ + if (!target__none(&target)) + return; - /* Append very, very detailed run extra attributes: */ - return perf_evlist__add_default_attrs(evsel_list, very_very_detailed_attrs); + if (!forks) + target.system_wide = true; + else { + struct evsel *counter; + + evlist__for_each_entry(evsel_list, counter) { + if (!counter->core.requires_cpu && + !evsel__name_is(counter, "duration_time")) { + return; + } + } + + if (evsel_list->core.nr_entries) + target.system_wide = true; + } } -int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) +#ifdef HAVE_ARCH_X86_64_SUPPORT +static int parse_tpebs_mode(const struct option *opt, const char *str, + int unset __maybe_unused) { - bool append_file = false; - int output_fd = 0; - const char *output_name = NULL; - const struct option options[] = { - OPT_CALLBACK('e', "event", &evsel_list, "event", - "event selector. use 'perf list' to list available events", - parse_events_option), - OPT_CALLBACK(0, "filter", &evsel_list, "filter", - "event filter", parse_filter), - OPT_BOOLEAN('i', "no-inherit", &no_inherit, - "child tasks do not inherit counters"), - OPT_STRING('p', "pid", &target.pid, "pid", - "stat events on existing process id"), - OPT_STRING('t', "tid", &target.tid, "tid", - "stat events on existing thread id"), - OPT_BOOLEAN('a', "all-cpus", &target.system_wide, - "system-wide collection from all CPUs"), - OPT_BOOLEAN('g', "group", &group, - "put the counters into a counter group"), - OPT_BOOLEAN('c', "scale", &scale, "scale/normalize counters"), - OPT_INCR('v', "verbose", &verbose, - "be more verbose (show counter open errors, etc)"), - OPT_INTEGER('r', "repeat", &run_count, - "repeat command and print average + stddev (max: 100, forever: 0)"), - OPT_BOOLEAN('n', "null", &null_run, - "null run - dont start any counters"), - OPT_INCR('d', "detailed", &detailed_run, - "detailed run - start a lot of events"), - OPT_BOOLEAN('S', "sync", &sync_run, - "call sync() before starting a run"), - OPT_CALLBACK_NOOPT('B', "big-num", NULL, NULL, - "print large numbers with thousands\' separators", - stat__set_big_num), - OPT_STRING('C', "cpu", &target.cpu_list, "cpu", - "list of cpus to monitor in system-wide"), - OPT_SET_UINT('A', "no-aggr", &aggr_mode, - "disable CPU count aggregation", AGGR_NONE), - OPT_STRING('x', "field-separator", &csv_sep, "separator", - "print counts with custom separator"), - OPT_CALLBACK('G', "cgroup", &evsel_list, "name", - "monitor event in cgroup name only", parse_cgroups), - OPT_STRING('o', "output", &output_name, "file", "output file name"), - OPT_BOOLEAN(0, "append", &append_file, "append to the output file"), - OPT_INTEGER(0, "log-fd", &output_fd, - "log output to fd, instead of stderr"), - OPT_STRING(0, "pre", &pre_cmd, "command", + enum tpebs_mode *mode = opt->value; + + if (!strcasecmp("mean", str)) { + *mode = TPEBS_MODE__MEAN; + return 0; + } + if (!strcasecmp("min", str)) { + *mode = TPEBS_MODE__MIN; + return 0; + } + if (!strcasecmp("max", str)) { + *mode = TPEBS_MODE__MAX; + return 0; + } + if (!strcasecmp("last", str)) { + *mode = TPEBS_MODE__LAST; + return 0; + } + return -1; +} +#endif // HAVE_ARCH_X86_64_SUPPORT + +int cmd_stat(int argc, const char **argv) +{ + struct opt_aggr_mode opt_mode = {}; + struct option stat_options[] = { + OPT_BOOLEAN('T', "transaction", &transaction_run, + "hardware transaction statistics"), + OPT_CALLBACK('e', "event", &parse_events_option_args, "event", + "event selector. use 'perf list' to list available events", + parse_events_option), + OPT_CALLBACK(0, "filter", &evsel_list, "filter", + "event filter", parse_filter), + OPT_BOOLEAN('i', "no-inherit", &stat_config.no_inherit, + "child tasks do not inherit counters"), + OPT_STRING('p', "pid", &target.pid, "pid", + "stat events on existing process id"), + OPT_STRING('t', "tid", &target.tid, "tid", + "stat events on existing thread id"), +#ifdef HAVE_BPF_SKEL + OPT_STRING('b', "bpf-prog", &target.bpf_str, "bpf-prog-id", + "stat events on existing bpf program id"), + OPT_BOOLEAN(0, "bpf-counters", &target.use_bpf, + "use bpf program to count events"), + OPT_STRING(0, "bpf-attr-map", &target.attr_map, "attr-map-path", + "path to perf_event_attr map"), +#endif + OPT_BOOLEAN('a', "all-cpus", &target.system_wide, + "system-wide collection from all CPUs"), + OPT_BOOLEAN(0, "scale", &stat_config.scale, + "Use --no-scale to disable counter scaling for multiplexing"), + OPT_INCR('v', "verbose", &verbose, + "be more verbose (show counter open errors, etc)"), + OPT_INTEGER('r', "repeat", &stat_config.run_count, + "repeat command and print average + stddev (max: 100, forever: 0)"), + OPT_BOOLEAN(0, "table", &stat_config.walltime_run_table, + "display details about each run (only with -r option)"), + OPT_BOOLEAN('n', "null", &stat_config.null_run, + "null run - dont start any counters"), + OPT_INCR('d', "detailed", &detailed_run, + "detailed run - start a lot of events"), + OPT_BOOLEAN('S', "sync", &sync_run, + "call sync() before starting a run"), + OPT_CALLBACK_NOOPT('B', "big-num", NULL, NULL, + "print large numbers with thousands\' separators", + stat__set_big_num), + OPT_STRING('C', "cpu", &target.cpu_list, "cpu", + "list of cpus to monitor in system-wide"), + OPT_BOOLEAN('A', "no-aggr", &opt_mode.no_aggr, + "disable aggregation across CPUs or PMUs"), + OPT_BOOLEAN(0, "no-merge", &opt_mode.no_aggr, + "disable aggregation the same as -A or -no-aggr"), + OPT_BOOLEAN(0, "hybrid-merge", &stat_config.hybrid_merge, + "Merge identical named hybrid events"), + OPT_STRING('x', "field-separator", &stat_config.csv_sep, "separator", + "print counts with custom separator"), + OPT_BOOLEAN('j', "json-output", &stat_config.json_output, + "print counts in JSON format"), + OPT_CALLBACK('G', "cgroup", &evsel_list, "name", + "monitor event in cgroup name only", parse_stat_cgroups), + OPT_STRING(0, "for-each-cgroup", &stat_config.cgroup_list, "name", + "expand events for each cgroup"), + OPT_STRING('o', "output", &output_name, "file", "output file name"), + OPT_BOOLEAN(0, "append", &append_file, "append to the output file"), + OPT_INTEGER(0, "log-fd", &output_fd, + "log output to fd, instead of stderr"), + OPT_STRING(0, "pre", &pre_cmd, "command", "command to run prior to the measured command"), - OPT_STRING(0, "post", &post_cmd, "command", + OPT_STRING(0, "post", &post_cmd, "command", "command to run after to the measured command"), - OPT_UINTEGER('I', "interval-print", &interval, - "print counts at regular interval in ms (>= 100)"), - OPT_SET_UINT(0, "per-socket", &aggr_mode, - "aggregate counts per processor socket", AGGR_SOCKET), - OPT_SET_UINT(0, "per-core", &aggr_mode, - "aggregate counts per physical processor core", AGGR_CORE), - OPT_END() + OPT_UINTEGER('I', "interval-print", &stat_config.interval, + "print counts at regular interval in ms " + "(overhead is possible for values <= 100ms)"), + OPT_INTEGER(0, "interval-count", &stat_config.times, + "print counts for fixed number of times"), + OPT_BOOLEAN(0, "interval-clear", &stat_config.interval_clear, + "clear screen in between new interval"), + OPT_UINTEGER(0, "timeout", &stat_config.timeout, + "stop workload and print counts after a timeout period in ms (>= 10ms)"), + OPT_BOOLEAN(0, "per-socket", &opt_mode.socket, + "aggregate counts per processor socket"), + OPT_BOOLEAN(0, "per-die", &opt_mode.die, "aggregate counts per processor die"), + OPT_BOOLEAN(0, "per-cluster", &opt_mode.cluster, + "aggregate counts per processor cluster"), + OPT_CALLBACK_OPTARG(0, "per-cache", &opt_mode, &stat_config.aggr_level, + "cache level", "aggregate count at this cache level (Default: LLC)", + parse_cache_level), + OPT_BOOLEAN(0, "per-core", &opt_mode.core, + "aggregate counts per physical processor core"), + OPT_BOOLEAN(0, "per-thread", &opt_mode.thread, "aggregate counts per thread"), + OPT_BOOLEAN(0, "per-node", &opt_mode.node, "aggregate counts per numa node"), + OPT_INTEGER('D', "delay", &target.initial_delay, + "ms to wait before starting measurement after program start (-1: start with events disabled)"), + OPT_CALLBACK_NOOPT(0, "metric-only", &stat_config.metric_only, NULL, + "Only print computed metrics. No raw values", enable_metric_only), + OPT_BOOLEAN(0, "metric-no-group", &stat_config.metric_no_group, + "don't group metric events, impacts multiplexing"), + OPT_BOOLEAN(0, "metric-no-merge", &stat_config.metric_no_merge, + "don't try to share events between metrics in a group"), + OPT_BOOLEAN(0, "metric-no-threshold", &stat_config.metric_no_threshold, + "disable adding events for the metric threshold calculation"), + OPT_BOOLEAN(0, "topdown", &topdown_run, + "measure top-down statistics"), +#ifdef HAVE_ARCH_X86_64_SUPPORT + OPT_BOOLEAN(0, "record-tpebs", &tpebs_recording, + "enable recording for tpebs when retire_latency required"), + OPT_CALLBACK(0, "tpebs-mode", &tpebs_mode, "tpebs-mode", + "Mode of TPEBS recording: mean, min or max", + parse_tpebs_mode), +#endif + OPT_UINTEGER(0, "td-level", &stat_config.topdown_level, + "Set the metrics level for the top-down statistics (0: max level)"), + OPT_BOOLEAN(0, "smi-cost", &smi_cost, + "measure SMI cost"), + OPT_CALLBACK('M', "metrics", &evsel_list, "metric/metric group list", + "monitor specified metrics or metric groups (separated by ,)", + append_metric_groups), + OPT_BOOLEAN_FLAG(0, "all-kernel", &stat_config.all_kernel, + "Configure all used events to run in kernel space.", + PARSE_OPT_EXCLUSIVE), + OPT_BOOLEAN_FLAG(0, "all-user", &stat_config.all_user, + "Configure all used events to run in user space.", + PARSE_OPT_EXCLUSIVE), + OPT_BOOLEAN(0, "percore-show-thread", &stat_config.percore_show_thread, + "Use with 'percore' event qualifier to show the event " + "counts of one hardware thread by sum up total hardware " + "threads of same physical core"), + OPT_BOOLEAN(0, "summary", &stat_config.summary, + "print summary for interval mode"), + OPT_BOOLEAN(0, "no-csv-summary", &stat_config.no_csv_summary, + "don't print 'summary' for CSV summary output"), + OPT_BOOLEAN(0, "quiet", &quiet, + "don't print any output, messages or warnings (useful with record)"), + OPT_CALLBACK(0, "cputype", &evsel_list, "hybrid cpu type", + "Only enable events on applying cpu with this type " + "for hybrid platform (e.g. core or atom)", + parse_cputype), +#ifdef HAVE_LIBPFM + OPT_CALLBACK(0, "pfm-events", &evsel_list, "event", + "libpfm4 event selector. use 'perf list' to list available events", + parse_libpfm_events_option), +#endif + OPT_CALLBACK(0, "control", &stat_config, "fd:ctl-fd[,ack-fd] or fifo:ctl-fifo[,ack-fifo]", + "Listen on ctl-fd descriptor for command to control measurement ('enable': enable events, 'disable': disable events).\n" + "\t\t\t Optionally send control command completion ('ack\\n') to ack-fd descriptor.\n" + "\t\t\t Alternatively, ctl-fifo / ack-fifo will be opened and used as ctl-fd / ack-fd.", + parse_control_option), + OPT_CALLBACK_OPTARG(0, "iostat", &evsel_list, &stat_config, "default", + "measure I/O performance metrics provided by arch/platform", + iostat_parse), + OPT_END() }; const char * const stat_usage[] = { "perf stat [<options>] [<command>]", NULL }; - int status = -ENOMEM, run_idx; + int status = -EINVAL, run_idx, err; const char *mode; + FILE *output = stderr; + unsigned int interval, timeout; + const char * const stat_subcommands[] = { "record", "report" }; + char errbuf[BUFSIZ]; + struct evsel *counter; setlocale(LC_ALL, ""); - evsel_list = perf_evlist__new(); + evsel_list = evlist__new(); if (evsel_list == NULL) return -ENOMEM; - argc = parse_options(argc, argv, options, stat_usage, - PARSE_OPT_STOP_AT_NON_OPTION); + parse_events__shrink_config_terms(); + + /* String-parsing callback-based options would segfault when negated */ + set_option_flag(stat_options, 'e', "event", PARSE_OPT_NONEG); + set_option_flag(stat_options, 'M', "metrics", PARSE_OPT_NONEG); + set_option_flag(stat_options, 'G', "cgroup", PARSE_OPT_NONEG); - output = stderr; - if (output_name && strcmp(output_name, "-")) + argc = parse_options_subcommand(argc, argv, stat_options, stat_subcommands, + (const char **) stat_usage, + PARSE_OPT_STOP_AT_NON_OPTION); + + stat_config.aggr_mode = opt_aggr_mode_to_aggr_mode(&opt_mode); + + if (stat_config.csv_sep) { + stat_config.csv_output = true; + if (!strcmp(stat_config.csv_sep, "\\t")) + stat_config.csv_sep = "\t"; + } else + stat_config.csv_sep = DEFAULT_SEPARATOR; + + if (argc && strlen(argv[0]) > 2 && strstarts("record", argv[0])) { + argc = __cmd_record(stat_options, &opt_mode, argc, argv); + if (argc < 0) + return -1; + } else if (argc && strlen(argv[0]) > 2 && strstarts("report", argv[0])) + return __cmd_report(argc, argv); + + interval = stat_config.interval; + timeout = stat_config.timeout; + + /* + * For record command the -o is already taken care of. + */ + if (!STAT_RECORD && output_name && strcmp(output_name, "-")) output = NULL; if (output_name && output_fd) { fprintf(stderr, "cannot use both --output and --log-fd\n"); - usage_with_options(stat_usage, options); + parse_options_usage(stat_usage, stat_options, "o", 1); + parse_options_usage(NULL, stat_options, "log-fd", 0); + goto out; + } + + if (stat_config.metric_only && stat_config.aggr_mode == AGGR_THREAD) { + fprintf(stderr, "--metric-only is not supported with --per-thread\n"); + goto out; + } + + if (stat_config.metric_only && stat_config.run_count > 1) { + fprintf(stderr, "--metric-only is not supported with -r\n"); + goto out; + } + + if (stat_config.csv_output || (stat_config.metric_only && stat_config.json_output)) { + /* + * Current CSV and metric-only JSON output doesn't display the + * metric threshold so don't compute it. + */ + stat_config.metric_no_threshold = true; + } + + if (stat_config.walltime_run_table && stat_config.run_count <= 1) { + fprintf(stderr, "--table is only supported with -r\n"); + parse_options_usage(stat_usage, stat_options, "r", 1); + parse_options_usage(NULL, stat_options, "table", 0); + goto out; } if (output_fd < 0) { fprintf(stderr, "argument to --log-fd must be a > 0\n"); - usage_with_options(stat_usage, options); + parse_options_usage(stat_usage, stat_options, "log-fd", 0); + goto out; } - if (!output) { + if (!output && !quiet) { struct timespec tm; mode = append_file ? "a" : "w"; @@ -1460,8 +2695,10 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) perror("failed to create output file"); return -1; } - clock_gettime(CLOCK_REALTIME, &tm); - fprintf(output, "# started on %s\n", ctime(&tm.tv_sec)); + if (!stat_config.json_output) { + clock_gettime(CLOCK_REALTIME, &tm); + fprintf(output, "# started on %s\n", ctime(&tm.tv_sec)); + } } else if (output_fd > 0) { mode = append_file ? "a" : "w"; output = fdopen(output_fd, mode); @@ -1471,71 +2708,249 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) } } - if (csv_sep) { - csv_output = true; - if (!strcmp(csv_sep, "\\t")) - csv_sep = "\t"; - } else - csv_sep = DEFAULT_SEPARATOR; + if (stat_config.interval_clear && !isatty(fileno(output))) { + fprintf(stderr, "--interval-clear does not work with output\n"); + parse_options_usage(stat_usage, stat_options, "o", 1); + parse_options_usage(NULL, stat_options, "log-fd", 0); + parse_options_usage(NULL, stat_options, "interval-clear", 0); + return -1; + } + + stat_config.output = output; /* * let the spreadsheet do the pretty-printing */ - if (csv_output) { + if (stat_config.csv_output) { /* User explicitly passed -B? */ if (big_num_opt == 1) { fprintf(stderr, "-B option not supported with -x\n"); - usage_with_options(stat_usage, options); + parse_options_usage(stat_usage, stat_options, "B", 1); + parse_options_usage(NULL, stat_options, "x", 1); + goto out; } else /* Nope, so disable big number formatting */ - big_num = false; + stat_config.big_num = false; } else if (big_num_opt == 0) /* User passed --no-big-num */ - big_num = false; + stat_config.big_num = false; + + target.inherit = !stat_config.no_inherit; + err = target__validate(&target); + if (err) { + target__strerror(&target, err, errbuf, BUFSIZ); + pr_warning("%s\n", errbuf); + } - if (!argc && !perf_target__has_task(&target)) - usage_with_options(stat_usage, options); - if (run_count < 0) { - usage_with_options(stat_usage, options); - } else if (run_count == 0) { + setup_system_wide(argc); + + /* + * Display user/system times only for single + * run and when there's specified tracee. + */ + if ((stat_config.run_count == 1) && target__none(&target)) + stat_config.ru_display = true; + + if (stat_config.run_count < 0) { + pr_err("Run count must be a positive number\n"); + parse_options_usage(stat_usage, stat_options, "r", 1); + goto out; + } else if (stat_config.run_count == 0) { forever = true; - run_count = 1; + stat_config.run_count = 1; + } + + if (stat_config.walltime_run_table) { + stat_config.walltime_run = zalloc(stat_config.run_count * sizeof(stat_config.walltime_run[0])); + if (!stat_config.walltime_run) { + pr_err("failed to setup -r option"); + goto out; + } } - /* no_aggr, cgroup are for system-wide only */ - if ((aggr_mode != AGGR_GLOBAL || nr_cgroups) - && !perf_target__has_cpu(&target)) { + if ((stat_config.aggr_mode == AGGR_THREAD) && + !target__has_task(&target)) { + if (!target.system_wide || target.cpu_list) { + fprintf(stderr, "The --per-thread option is only " + "available when monitoring via -p -t -a " + "options or only --per-thread.\n"); + parse_options_usage(NULL, stat_options, "p", 1); + parse_options_usage(NULL, stat_options, "t", 1); + goto out; + } + } + + /* + * no_aggr, cgroup are for system-wide only + * --per-thread is aggregated per thread, we dont mix it with cpu mode + */ + if (((stat_config.aggr_mode != AGGR_GLOBAL && + stat_config.aggr_mode != AGGR_THREAD) || + (nr_cgroups || stat_config.cgroup_list)) && + !target__has_cpu(&target)) { fprintf(stderr, "both cgroup and no-aggregation " "modes only available in system-wide mode\n"); - usage_with_options(stat_usage, options); - return -1; + parse_options_usage(stat_usage, stat_options, "G", 1); + parse_options_usage(NULL, stat_options, "A", 1); + parse_options_usage(NULL, stat_options, "a", 1); + parse_options_usage(NULL, stat_options, "for-each-cgroup", 0); + goto out; + } + + if (stat_config.iostat_run) { + status = iostat_prepare(evsel_list, &stat_config); + if (status) + goto out; + if (iostat_mode == IOSTAT_LIST) { + iostat_list(evsel_list, &stat_config); + goto out; + } else if (verbose > 0) + iostat_list(evsel_list, &stat_config); + if (iostat_mode == IOSTAT_RUN && !target__has_cpu(&target)) + target.system_wide = true; + } + + if ((stat_config.aggr_mode == AGGR_THREAD) && (target.system_wide)) + target.per_thread = true; + + stat_config.system_wide = target.system_wide; + if (target.cpu_list) { + stat_config.user_requested_cpu_list = strdup(target.cpu_list); + if (!stat_config.user_requested_cpu_list) { + status = -ENOMEM; + goto out; + } + } + + /* + * Metric parsing needs to be delayed as metrics may optimize events + * knowing the target is system-wide. + */ + if (metrics) { + const char *pmu = parse_events_option_args.pmu_filter ?: "all"; + int ret = metricgroup__parse_groups(evsel_list, pmu, metrics, + stat_config.metric_no_group, + stat_config.metric_no_merge, + stat_config.metric_no_threshold, + stat_config.user_requested_cpu_list, + stat_config.system_wide, + stat_config.hardware_aware_grouping); + + zfree(&metrics); + if (ret) { + status = ret; + goto out; + } } - if (add_default_attributes()) + if (add_default_events()) goto out; - perf_target__validate(&target); + if (stat_config.cgroup_list) { + if (nr_cgroups > 0) { + pr_err("--cgroup and --for-each-cgroup cannot be used together\n"); + parse_options_usage(stat_usage, stat_options, "G", 1); + parse_options_usage(NULL, stat_options, "for-each-cgroup", 0); + goto out; + } + + if (evlist__expand_cgroup(evsel_list, stat_config.cgroup_list, true) < 0) { + parse_options_usage(stat_usage, stat_options, + "for-each-cgroup", 0); + goto out; + } + } +#ifdef HAVE_BPF_SKEL + if (target.use_bpf && nr_cgroups && + (evsel_list->core.nr_entries / nr_cgroups) > BPERF_CGROUP__MAX_EVENTS) { + pr_warning("Disabling BPF counters due to more events (%d) than the max (%d)\n", + evsel_list->core.nr_entries / nr_cgroups, BPERF_CGROUP__MAX_EVENTS); + target.use_bpf = false; + } +#endif // HAVE_BPF_SKEL + evlist__warn_user_requested_cpus(evsel_list, target.cpu_list); + + evlist__for_each_entry(evsel_list, counter) { + /* + * Setup BPF counters to require CPUs as any(-1) isn't + * supported. evlist__create_maps below will propagate this + * information to the evsels. Note, evsel__is_bperf isn't yet + * set up, and this change must happen early, so directly use + * the bpf_counter variable and target information. + */ + if ((counter->bpf_counter || target.use_bpf) && !target__has_cpu(&target)) + counter->core.requires_cpu = true; + } - if (perf_evlist__create_maps(evsel_list, &target) < 0) { - if (perf_target__has_task(&target)) + if (evlist__create_maps(evsel_list, &target) < 0) { + if (target__has_task(&target)) { pr_err("Problems finding threads of monitor\n"); - if (perf_target__has_cpu(&target)) + parse_options_usage(stat_usage, stat_options, "p", 1); + parse_options_usage(NULL, stat_options, "t", 1); + } else if (target__has_cpu(&target)) { perror("failed to parse CPUs map"); + parse_options_usage(stat_usage, stat_options, "C", 1); + parse_options_usage(NULL, stat_options, "a", 1); + } + goto out; + } - usage_with_options(stat_usage, options); - return -1; + evlist__check_cpu_maps(evsel_list); + + /* + * Initialize thread_map with comm names, + * so we could print it out on output. + */ + if (stat_config.aggr_mode == AGGR_THREAD) { + thread_map__read_comms(evsel_list->core.threads); } - if (interval && interval < 100) { - pr_err("print interval must be >= 100ms\n"); - usage_with_options(stat_usage, options); - return -1; + + if (stat_config.aggr_mode == AGGR_NODE) + cpu__setup_cpunode_map(); + + if (stat_config.times && interval) + interval_count = true; + else if (stat_config.times && !interval) { + pr_err("interval-count option should be used together with " + "interval-print.\n"); + parse_options_usage(stat_usage, stat_options, "interval-count", 0); + parse_options_usage(stat_usage, stat_options, "I", 1); + goto out; } - if (perf_evlist__alloc_stats(evsel_list, interval)) - goto out_free_maps; + if (timeout && timeout < 100) { + if (timeout < 10) { + pr_err("timeout must be >= 10ms.\n"); + parse_options_usage(stat_usage, stat_options, "timeout", 0); + goto out; + } else + pr_warning("timeout < 100ms. " + "The overhead percentage could be high in some cases. " + "Please proceed with caution.\n"); + } + if (timeout && interval) { + pr_err("timeout option is not supported with interval-print.\n"); + parse_options_usage(stat_usage, stat_options, "timeout", 0); + parse_options_usage(stat_usage, stat_options, "I", 1); + goto out; + } if (perf_stat_init_aggr_mode()) goto out; + if (evlist__alloc_stats(&stat_config, evsel_list, interval)) + goto out; + + /* + * Set sample_type to PERF_SAMPLE_IDENTIFIER, which should be harmless + * while avoiding that older tools show confusing messages. + * + * However for pipe sessions we need to keep it zero, + * because script's perf_evsel__check_attr is triggered + * by attr->sample_type != 0, and we can't run it on + * stat sessions. + */ + stat_config.identifier = !(STAT_RECORD && perf_stat.data.is_pipe); + /* * We dont want to block the signals - that would cause * child tasks to inherit that and Ctrl-C would not work. @@ -1549,26 +2964,91 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) signal(SIGALRM, skip_signal); signal(SIGABRT, skip_signal); + if (evlist__initialize_ctlfd(evsel_list, stat_config.ctl_fd, stat_config.ctl_fd_ack)) + goto out; + + /* Enable ignoring missing threads when -p option is defined. */ + evlist__first(evsel_list)->ignore_missing_thread = target.pid; status = 0; - for (run_idx = 0; forever || run_idx < run_count; run_idx++) { - if (run_count != 1 && verbose) + for (run_idx = 0; forever || run_idx < stat_config.run_count; run_idx++) { + if (stat_config.run_count != 1 && verbose > 0) fprintf(output, "[ perf stat: executing run #%d ... ]\n", run_idx + 1); - status = run_perf_stat(argc, argv); - if (forever && status != -1) { - print_stat(argc, argv); - perf_stat__reset_stats(evsel_list); + if (run_idx != 0) + evlist__reset_prev_raw_counts(evsel_list); + + status = run_perf_stat(argc, argv, run_idx); + if (status < 0) + break; + + if (forever && !interval) { + print_counters(NULL, argc, argv); + perf_stat__reset_stats(); } } - if (!forever && status != -1 && !interval) - print_stat(argc, argv); + if (!forever && status != -1 && (!interval || stat_config.summary)) { + if (stat_config.run_count > 1) + evlist__copy_res_stats(&stat_config, evsel_list); + print_counters(NULL, argc, argv); + } + + evlist__finalize_ctlfd(evsel_list); + + if (STAT_RECORD) { + /* + * We synthesize the kernel mmap record just so that older tools + * don't emit warnings about not being able to resolve symbols + * due to /proc/sys/kernel/kptr_restrict settings and instead provide + * a saner message about no samples being in the perf.data file. + * + * This also serves to suppress a warning about f_header.data.size == 0 + * in header.c at the moment 'perf stat record' gets introduced, which + * is not really needed once we start adding the stat specific PERF_RECORD_ + * records, but the need to suppress the kptr_restrict messages in older + * tools remain -acme + */ + int fd = perf_data__fd(&perf_stat.data); + + err = perf_event__synthesize_kernel_mmap((void *)&perf_stat, + process_synthesized_event, + &perf_stat.session->machines.host); + if (err) { + pr_warning("Couldn't synthesize the kernel mmap record, harmless, " + "older tools may produce warnings about this file\n."); + } + + if (!interval) { + if (WRITE_STAT_ROUND_EVENT(stat_config.walltime_nsecs_stats->max, FINAL)) + pr_err("failed to write stat round event\n"); + } + + if (!perf_stat.data.is_pipe) { + perf_stat.session->header.data_size += perf_stat.bytes_written; + perf_session__write_header(perf_stat.session, evsel_list, fd, true); + } + + evlist__close(evsel_list); + perf_session__delete(perf_stat.session); + } - perf_evlist__free_stats(evsel_list); -out_free_maps: - perf_evlist__delete_maps(evsel_list); + perf_stat__exit_aggr_mode(); + evlist__free_stats(evsel_list); out: - perf_evlist__delete(evsel_list); - return status; + if (stat_config.iostat_run) + iostat_release(evsel_list); + + zfree(&stat_config.walltime_run); + zfree(&stat_config.user_requested_cpu_list); + + if (smi_cost && smi_reset) + sysfs__write_int(FREEZE_ON_SMI_PATH, 0); + + evlist__delete(evsel_list); + + evlist__close_control(stat_config.ctl_fd, stat_config.ctl_fd_ack, &stat_config.ctl_fd_close); + + /* Only the low byte of status becomes the exit code. */ + return abs(status); } |