diff options
| -rw-r--r-- | include/linux/perf_event.h | 12 | ||||
| -rw-r--r-- | kernel/perf_event.c | 255 | ||||
| -rw-r--r-- | tools/perf/Documentation/perf-record.txt | 4 | ||||
| -rw-r--r-- | tools/perf/builtin-record.c | 76 | ||||
| -rw-r--r-- | tools/perf/util/parse-events.c | 16 | ||||
| -rw-r--r-- | tools/perf/util/trace-event-parse.c | 2 |
6 files changed, 252 insertions, 113 deletions
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 6e96cc8225d..bf896d0b2e9 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -589,6 +589,14 @@ enum perf_group_flag { PERF_GROUP_SOFTWARE = 0x1, }; +#define SWEVENT_HLIST_BITS 8 +#define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) + +struct swevent_hlist { + struct hlist_head heads[SWEVENT_HLIST_SIZE]; + struct rcu_head rcu_head; +}; + /** * struct perf_event - performance event kernel representation: */ @@ -597,6 +605,7 @@ struct perf_event { struct list_head group_entry; struct list_head event_entry; struct list_head sibling_list; + struct hlist_node hlist_entry; int nr_siblings; int group_flags; struct perf_event *group_leader; @@ -744,6 +753,9 @@ struct perf_cpu_context { int active_oncpu; int max_pertask; int exclusive; + struct swevent_hlist *swevent_hlist; + struct mutex hlist_mutex; + int hlist_refcount; /* * Recursion avoidance: diff --git a/kernel/perf_event.c b/kernel/perf_event.c index fcf42dcd608..07b7a435bf0 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -16,6 +16,7 @@ #include <linux/file.h> #include <linux/poll.h> #include <linux/slab.h> +#include <linux/hash.h> #include <linux/sysfs.h> #include <linux/dcache.h> #include <linux/percpu.h> @@ -3966,36 +3967,6 @@ static void perf_swevent_add(struct perf_event *event, u64 nr, perf_swevent_overflow(event, 0, nmi, data, regs); } -static int perf_swevent_is_counting(struct perf_event *event) -{ - /* - * The event is active, we're good! - */ - if (event->state == PERF_EVENT_STATE_ACTIVE) - return 1; - - /* - * The event is off/error, not counting. - */ - if (event->state != PERF_EVENT_STATE_INACTIVE) - return 0; - - /* - * The event is inactive, if the context is active - * we're part of a group that didn't make it on the 'pmu', - * not counting. - */ - if (event->ctx->is_active) - return 0; - - /* - * We're inactive and the context is too, this means the - * task is scheduled out, we're counting events that happen - * to us, like migration events. - */ - return 1; -} - static int perf_tp_event_match(struct perf_event *event, struct perf_sample_data *data); @@ -4019,12 +3990,6 @@ static int perf_swevent_match(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs) { - if (event->cpu != -1 && event->cpu != smp_processor_id()) - return 0; - - if (!perf_swevent_is_counting(event)) - return 0; - if (event->attr.type != type) return 0; @@ -4041,18 +4006,53 @@ static int perf_swevent_match(struct perf_event *event, return 1; } -static void perf_swevent_ctx_event(struct perf_event_context *ctx, - enum perf_type_id type, - u32 event_id, u64 nr, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) +static inline u64 swevent_hash(u64 type, u32 event_id) { + u64 val = event_id | (type << 32); + + return hash_64(val, SWEVENT_HLIST_BITS); +} + +static struct hlist_head * +find_swevent_head(struct perf_cpu_context *ctx, u64 type, u32 event_id) +{ + u64 hash; + struct swevent_hlist *hlist; + + hash = swevent_hash(type, event_id); + + hlist = rcu_dereference(ctx->swevent_hlist); + if (!hlist) + return NULL; + + return &hlist->heads[hash]; +} + +static void do_perf_sw_event(enum perf_type_id type, u32 event_id, + u64 nr, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + struct perf_cpu_context *cpuctx; struct perf_event *event; + struct hlist_node *node; + struct hlist_head *head; - list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { + cpuctx = &__get_cpu_var(perf_cpu_context); + + rcu_read_lock(); + + head = find_swevent_head(cpuctx, type, event_id); + + if (!head) + goto end; + + hlist_for_each_entry_rcu(event, node, head, hlist_entry) { if (perf_swevent_match(event, type, event_id, data, regs)) perf_swevent_add(event, nr, nmi, data, regs); } +end: + rcu_read_unlock(); } int perf_swevent_get_recursion_context(void) @@ -4090,27 +4090,6 @@ void perf_swevent_put_recursion_context(int rctx) } EXPORT_SYMBOL_GPL(perf_swevent_put_recursion_context); -static void do_perf_sw_event(enum perf_type_id type, u32 event_id, - u64 nr, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) -{ - struct perf_cpu_context *cpuctx; - struct perf_event_context *ctx; - - cpuctx = &__get_cpu_var(perf_cpu_context); - rcu_read_lock(); - perf_swevent_ctx_event(&cpuctx->ctx, type, event_id, - nr, nmi, data, regs); - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ - ctx = rcu_dereference(current->perf_event_ctxp); - if (ctx) - perf_swevent_ctx_event(ctx, type, event_id, nr, nmi, data, regs); - rcu_read_unlock(); -} void __perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) @@ -4136,16 +4115,28 @@ static void perf_swevent_read(struct perf_event *event) static int perf_swevent_enable(struct perf_event *event) { struct hw_perf_event *hwc = &event->hw; + struct perf_cpu_context *cpuctx; + struct hlist_head *head; + + cpuctx = &__get_cpu_var(perf_cpu_context); if (hwc->sample_period) { hwc->last_period = hwc->sample_period; perf_swevent_set_period(event); } + + head = find_swevent_head(cpuctx, event->attr.type, event->attr.config); + if (WARN_ON_ONCE(!head)) + return -EINVAL; + + hlist_add_head_rcu(&event->hlist_entry, head); + return 0; } static void perf_swevent_disable(struct perf_event *event) { + hlist_del_rcu(&event->hlist_entry); } static const struct pmu perf_ops_generic = { @@ -4173,15 +4164,8 @@ static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer) perf_sample_data_init(&data, 0); data.period = event->hw.last_period; regs = get_irq_regs(); - /* - * In case we exclude kernel IPs or are somehow not in interrupt - * context, provide the next best thing, the user IP. - */ - if ((event->attr.exclude_kernel || !regs) && - !event->attr.exclude_user) - regs = task_pt_regs(current); - if (regs) { + if (regs && !perf_exclude_event(event, regs)) { if (!(event->attr.exclude_idle && current->pid == 0)) if (perf_event_overflow(event, 0, &data, regs)) ret = HRTIMER_NORESTART; @@ -4329,6 +4313,105 @@ static const struct pmu perf_ops_task_clock = { .read = task_clock_perf_event_read, }; +static void swevent_hlist_release_rcu(struct rcu_head *rcu_head) +{ + struct swevent_hlist *hlist; + + hlist = container_of(rcu_head, struct swevent_hlist, rcu_head); + kfree(hlist); +} + +static void swevent_hlist_release(struct perf_cpu_context *cpuctx) +{ + struct swevent_hlist *hlist; + + if (!cpuctx->swevent_hlist) + return; + + hlist = cpuctx->swevent_hlist; + rcu_assign_pointer(cpuctx->swevent_hlist, NULL); + call_rcu(&hlist->rcu_head, swevent_hlist_release_rcu); +} + +static void swevent_hlist_put_cpu(struct perf_event *event, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + + mutex_lock(&cpuctx->hlist_mutex); + + if (!--cpuctx->hlist_refcount) + swevent_hlist_release(cpuctx); + + mutex_unlock(&cpuctx->hlist_mutex); +} + +static void swevent_hlist_put(struct perf_event *event) +{ + int cpu; + + if (event->cpu != -1) { + swevent_hlist_put_cpu(event, event->cpu); + return; + } + + for_each_possible_cpu(cpu) + swevent_hlist_put_cpu(event, cpu); +} + +static int swevent_hlist_get_cpu(struct perf_event *event, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + int err = 0; + + mutex_lock(&cpuctx->hlist_mutex); + + if (!cpuctx->swevent_hlist && cpu_online(cpu)) { + struct swevent_hlist *hlist; + + hlist = kzalloc(sizeof(*hlist), GFP_KERNEL); + if (!hlist) { + err = -ENOMEM; + goto exit; + } + rcu_assign_pointer(cpuctx->swevent_hlist, hlist); + } + cpuctx->hlist_refcount++; + exit: + mutex_unlock(&cpuctx->hlist_mutex); + + return err; +} + +static int swevent_hlist_get(struct perf_event *event) +{ + int err; + int cpu, failed_cpu; + + if (event->cpu != -1) + return swevent_hlist_get_cpu(event, event->cpu); + + get_online_cpus(); + for_each_possible_cpu(cpu) { + err = swevent_hlist_get_cpu(event, cpu); + if (err) { + failed_cpu = cpu; + goto fail; + } + } + put_online_cpus(); + + return 0; + fail: + for_each_possible_cpu(cpu) { + if (cpu == failed_cpu) + break; + swevent_hlist_put_cpu(event, cpu); + } + + put_online_cpus(); + return err; +} + #ifdef CONFIG_EVENT_TRACING void perf_tp_event(int event_id, u64 addr, u64 count, void *record, @@ -4362,10 +4445,13 @@ static int perf_tp_event_match(struct perf_event *event, static void tp_perf_event_destroy(struct perf_event *event) { perf_trace_disable(event->attr.config); + swevent_hlist_put(event); } static const struct pmu *tp_perf_event_init(struct perf_event *event) { + int err; + /* * Raw tracepoint data is a severe data leak, only allow root to * have these. @@ -4379,6 +4465,11 @@ static const struct pmu *tp_perf_event_init(struct perf_event *event) return NULL; event->destroy = tp_perf_event_destroy; + err = swevent_hlist_get(event); + if (err) { + perf_trace_disable(event->attr.config); + return ERR_PTR(err); + } return &perf_ops_generic; } @@ -4479,6 +4570,7 @@ static void sw_perf_event_destroy(struct perf_event *event) WARN_ON(event->parent); atomic_dec(&perf_swevent_enabled[event_id]); + swevent_hlist_put(event); } static const struct pmu *sw_perf_event_init(struct perf_event *event) @@ -4517,6 +4609,12 @@ static const struct pmu *sw_perf_event_init(struct perf_event *event) case PERF_COUNT_SW_ALIGNMENT_FAULTS: case PERF_COUNT_SW_EMULATION_FAULTS: if (!event->parent) { + int err; + + err = swevent_hlist_get(event); + if (err) + return ERR_PTR(err); + atomic_inc(&perf_swevent_enabled[event_id]); event->destroy = sw_perf_event_destroy; } @@ -5389,6 +5487,7 @@ static void __init perf_event_init_all_cpus(void) for_each_possible_cpu(cpu) { cpuctx = &per_cpu(perf_cpu_context, cpu); + mutex_init(&cpuctx->hlist_mutex); __perf_event_init_context(&cpuctx->ctx, NULL); } } @@ -5402,6 +5501,16 @@ static void __cpuinit perf_event_init_cpu(int cpu) spin_lock(&perf_resource_lock); cpuctx->max_pertask = perf_max_events - perf_reserved_percpu; spin_unlock(&perf_resource_lock); + + mutex_lock(&cpuctx->hlist_mutex); + if (cpuctx->hlist_refcount > 0) { + struct swevent_hlist *hlist; + + hlist = kzalloc(sizeof(*hlist), GFP_KERNEL); + WARN_ON_ONCE(!hlist); + rcu_assign_pointer(cpuctx->swevent_hlist, hlist); + } + mutex_unlock(&cpuctx->hlist_mutex); } #ifdef CONFIG_HOTPLUG_CPU @@ -5421,6 +5530,10 @@ static void perf_event_exit_cpu(int cpu) struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); struct perf_event_context *ctx = &cpuctx->ctx; + mutex_lock(&cpuctx->hlist_mutex); + swevent_hlist_release(cpuctx); + mutex_unlock(&cpuctx->hlist_mutex); + mutex_lock(&ctx->mutex); smp_call_function_single(cpu, __perf_event_exit_cpu, NULL, 1); mutex_unlock(&ctx->mutex); diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt index fc46c0b40f6..020d871c793 100644 --- a/tools/perf/Documentation/perf-record.txt +++ b/tools/perf/Documentation/perf-record.txt @@ -58,7 +58,7 @@ OPTIONS -f:: --force:: - Overwrite existing data file. + Overwrite existing data file. (deprecated) -c:: --count=:: @@ -101,7 +101,7 @@ OPTIONS -R:: --raw-samples:: -Collect raw sample records from all opened counters (typically for tracepoint counters). +Collect raw sample records from all opened counters (default for tracepoint counters). SEE ALSO -------- diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 0bde31bc8e2..a1b99eeac3c 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -26,13 +26,20 @@ #include <unistd.h> #include <sched.h> +enum write_mode_t { + WRITE_FORCE, + WRITE_APPEND +}; + static int *fd[MAX_NR_CPUS][MAX_COUNTERS]; +static unsigned int user_interval = UINT_MAX; static long default_interval = 0; static int nr_cpus = 0; static unsigned int page_size; static unsigned int mmap_pages = 128; +static unsigned int user_freq = UINT_MAX; static int freq = 1000; static int output; static int pipe_output = 0; @@ -48,8 +55,7 @@ static pid_t *all_tids = NULL; static int thread_num = 0; static pid_t child_pid = -1; static bool inherit = true; -static bool force = false; -static bool append_file = false; +static enum write_mode_t write_mode = WRITE_FORCE; static bool call_graph = false; static bool inherit_stat = false; static bool no_samples = false; @@ -257,10 +263,19 @@ static void create_counter(int counter, int cpu) if (nr_counters > 1) attr->sample_type |= PERF_SAMPLE_ID; - if (freq) { - attr->sample_type |= PERF_SAMPLE_PERIOD; - attr->freq = 1; - attr->sample_freq = freq; + /* + * We default some events to a 1 default interval. But keep + * it a weak assumption overridable by the user. + */ + if (!attr->sample_period || (user_freq != UINT_MAX && + user_interval != UINT_MAX)) { + if (freq) { + attr->sample_type |= PERF_SAMPLE_PERIOD; + attr->freq = 1; + attr->sample_freq = freq; + } else { + attr->sample_period = default_interval; + } } if (no_samples) @@ -467,26 +482,19 @@ static int __cmd_record(int argc, const char **argv) if (!strcmp(output_name, "-")) pipe_output = 1; else if (!stat(output_name, &st) && st.st_size) { - if (!force) { - if (!append_file) { - pr_err("Error, output file %s exists, use -A " - "to append or -f to overwrite.\n", - output_name); - exit(-1); - } - } else { + if (write_mode == WRITE_FORCE) { char oldname[PATH_MAX]; snprintf(oldname, sizeof(oldname), "%s.old", output_name); unlink(oldname); rename(output_name, oldname); } - } else { - append_file = false; + } else if (write_mode == WRITE_APPEND) { + write_mode = WRITE_FORCE; } flags = O_CREAT|O_RDWR; - if (append_file) + if (write_mode == WRITE_APPEND) file_new = 0; else flags |= O_TRUNC; @@ -500,7 +508,8 @@ static int __cmd_record(int argc, const char **argv) exit(-1); } - session = perf_session__new(output_name, O_WRONLY, force); + session = perf_session__new(output_name, O_WRONLY, + write_mode == WRITE_FORCE); if (session == NULL) { pr_err("Not enough memory for reading perf file header\n"); return -1; @@ -721,6 +730,8 @@ static const char * const record_usage[] = { NULL }; +static bool force, append_file; + static const struct option options[] = { OPT_CALLBACK('e', "event", NULL, "event", "event selector. use 'perf list' to list available events", @@ -742,14 +753,14 @@ static const struct option options[] = { OPT_INTEGER('C', "profile_cpu", &profile_cpu, "CPU to profile on"), OPT_BOOLEAN('f', "force", &force, - "overwrite existing data file"), - OPT_LONG('c', "count", &default_interval, + "overwrite existing data file (deprecated)"), + OPT_LONG('c', "count", &user_interval, "event period to sample"), OPT_STRING('o', "output", &output_name, "file", "output file name"), OPT_BOOLEAN('i', "inherit", &inherit, "child tasks inherit counters"), - OPT_INTEGER('F', "freq", &freq, + OPT_INTEGER('F', "freq", &user_freq, "profile at this frequency"), OPT_INTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"), @@ -770,7 +781,6 @@ static const struct option options[] = { int cmd_record(int argc, const char **argv, const char *prefix __used) { - int counter; int i,j; argc = parse_options(argc, argv, options, record_usage, @@ -779,6 +789,16 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) !system_wide && profile_cpu == -1) usage_with_options(record_usage, options); + if (force && append_file) { + fprintf(stderr, "Can't overwrite and append at the same time." + " You need to choose between -f and -A"); + usage_with_options(record_usage, options); + } else if (append_file) { + write_mode = WRITE_APPEND; + } else { + write_mode = WRITE_FORCE; + } + symbol__init(); if (!nr_counters) { @@ -818,6 +838,11 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) if (!event_array) return -ENOMEM; + if (user_interval != UINT_MAX) + default_interval = user_interval; + if (user_freq != UINT_MAX) + freq = user_freq; + /* * User specified count overrides default frequency. */ @@ -830,12 +855,5 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) exit(EXIT_FAILURE); } - for (counter = 0; counter < nr_counters; counter++) { - if (attrs[counter].sample_period) - continue; - - attrs[counter].sample_period = default_interval; - } - return __cmd_record(argc, argv); } diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 435781e0c20..3b4ec679756 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -410,7 +410,6 @@ static enum event_result parse_single_tracepoint_event(char *sys_name, const char *evt_name, unsigned int evt_length, - char *flags, struct perf_event_attr *attr, const char **strp) { @@ -419,13 +418,11 @@ parse_single_tracepoint_event(char *sys_name, u64 id; int fd; - if (flags) { - if (!strncmp(flags, "record", strlen(flags))) { - attr->sample_type |= PERF_SAMPLE_RAW; - attr->sample_type |= PERF_SAMPLE_TIME; - attr->sample_type |= PERF_SAMPLE_CPU; - } - } + attr->sample_type |= PERF_SAMPLE_RAW; + attr->sample_type |= PERF_SAMPLE_TIME; + attr->sample_type |= PERF_SAMPLE_CPU; + + attr->sample_period = 1; snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id", debugfs_path, sys_name, evt_name); @@ -533,8 +530,7 @@ static enum event_result parse_tracepoint_event(const char **strp, flags); } else return parse_single_tracepoint_event(sys_name, evt_name, - evt_length, flags, - attr, strp); + evt_length, attr, strp); } static enum event_result diff --git a/tools/perf/util/trace-event-parse.c b/tools/perf/util/trace-event-parse.c index 17d6d66ed76..d6ef414075a 100644 --- a/tools/perf/util/trace-event-parse.c +++ b/tools/perf/util/trace-event-parse.c @@ -761,7 +761,7 @@ static int field_is_string(struct format_field *field) static int field_is_dynamic(struct format_field *field) { - if (!strcmp(field->type, "__data_loc")) + if (!strncmp(field->type, "__data_loc", 10)) return 1; return 0; |