diff options
Diffstat (limited to 'tools/perf/builtin-top.c')
-rw-r--r-- | tools/perf/builtin-top.c | 1030 |
1 files changed, 313 insertions, 717 deletions
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 5a29d9c..417f757 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -20,11 +20,16 @@ #include "perf.h" +#include "util/annotate.h" +#include "util/cache.h" #include "util/color.h" +#include "util/evlist.h" #include "util/evsel.h" #include "util/session.h" #include "util/symbol.h" #include "util/thread.h" +#include "util/thread_map.h" +#include "util/top.h" #include "util/util.h" #include <linux/rbtree.h> #include "util/parse-options.h" @@ -45,7 +50,6 @@ #include <errno.h> #include <time.h> #include <sched.h> -#include <pthread.h> #include <sys/syscall.h> #include <sys/ioctl.h> @@ -60,85 +64,42 @@ #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) +static struct perf_top top = { + .count_filter = 5, + .delay_secs = 2, + .display_weighted = -1, + .target_pid = -1, + .target_tid = -1, + .active_symbols = LIST_HEAD_INIT(top.active_symbols), + .active_symbols_lock = PTHREAD_MUTEX_INITIALIZER, + .active_symbols_cond = PTHREAD_COND_INITIALIZER, + .freq = 1000, /* 1 KHz */ +}; + static bool system_wide = false; -static int default_interval = 0; +static bool use_tui, use_stdio; -static int count_filter = 5; -static int print_entries; +static int default_interval = 0; -static int target_pid = -1; -static int target_tid = -1; -static struct thread_map *threads; static bool inherit = false; -static struct cpu_map *cpus; static int realtime_prio = 0; static bool group = false; static unsigned int page_size; -static unsigned int mmap_pages = 16; -static int freq = 1000; /* 1 KHz */ +static unsigned int mmap_pages = 128; -static int delay_secs = 2; -static bool zero = false; static bool dump_symtab = false; -static bool hide_kernel_symbols = false; -static bool hide_user_symbols = false; static struct winsize winsize; -/* - * Source - */ - -struct source_line { - u64 eip; - unsigned long count[MAX_COUNTERS]; - char *line; - struct source_line *next; -}; - static const char *sym_filter = NULL; -struct sym_entry *sym_filter_entry = NULL; struct sym_entry *sym_filter_entry_sched = NULL; static int sym_pcnt_filter = 5; -static int sym_counter = 0; -static struct perf_evsel *sym_evsel = NULL; -static int display_weighted = -1; -static const char *cpu_list; - -/* - * Symbols - */ - -struct sym_entry_source { - struct source_line *source; - struct source_line *lines; - struct source_line **lines_tail; - pthread_mutex_t lock; -}; - -struct sym_entry { - struct rb_node rb_node; - struct list_head node; - unsigned long snap_count; - double weight; - int skip; - u16 name_len; - u8 origin; - struct map *map; - struct sym_entry_source *src; - unsigned long count[0]; -}; /* * Source functions */ -static inline struct symbol *sym_entry__symbol(struct sym_entry *self) -{ - return ((void *)self) + symbol_conf.priv_size; -} - void get_term_dimensions(struct winsize *ws) { char *s = getenv("LINES"); @@ -163,10 +124,10 @@ void get_term_dimensions(struct winsize *ws) static void update_print_entries(struct winsize *ws) { - print_entries = ws->ws_row; + top.print_entries = ws->ws_row; - if (print_entries > 9) - print_entries -= 9; + if (top.print_entries > 9) + top.print_entries -= 9; } static void sig_winch_handler(int sig __used) @@ -178,12 +139,9 @@ static void sig_winch_handler(int sig __used) static int parse_source(struct sym_entry *syme) { struct symbol *sym; - struct sym_entry_source *source; + struct annotation *notes; struct map *map; - FILE *file; - char command[PATH_MAX*2]; - const char *path; - u64 len; + int err = -1; if (!syme) return -1; @@ -194,411 +152,137 @@ static int parse_source(struct sym_entry *syme) /* * We can't annotate with just /proc/kallsyms */ - if (map->dso->origin == DSO__ORIG_KERNEL) + if (map->dso->origin == DSO__ORIG_KERNEL) { + pr_err("Can't annotate %s: No vmlinux file was found in the " + "path\n", sym->name); + sleep(1); return -1; - - if (syme->src == NULL) { - syme->src = zalloc(sizeof(*source)); - if (syme->src == NULL) - return -1; - pthread_mutex_init(&syme->src->lock, NULL); } - source = syme->src; - - if (source->lines) { - pthread_mutex_lock(&source->lock); + notes = symbol__annotation(sym); + if (notes->src != NULL) { + pthread_mutex_lock(¬es->lock); goto out_assign; } - path = map->dso->long_name; - - len = sym->end - sym->start; - - sprintf(command, - "objdump --start-address=%#0*" PRIx64 " --stop-address=%#0*" PRIx64 " -dS %s", - BITS_PER_LONG / 4, map__rip_2objdump(map, sym->start), - BITS_PER_LONG / 4, map__rip_2objdump(map, sym->end), path); - - file = popen(command, "r"); - if (!file) - return -1; - - pthread_mutex_lock(&source->lock); - source->lines_tail = &source->lines; - while (!feof(file)) { - struct source_line *src; - size_t dummy = 0; - char *c, *sep; - src = malloc(sizeof(struct source_line)); - assert(src != NULL); - memset(src, 0, sizeof(struct source_line)); + pthread_mutex_lock(¬es->lock); - if (getline(&src->line, &dummy, file) < 0) - break; - if (!src->line) - break; - - c = strchr(src->line, '\n'); - if (c) - *c = 0; - - src->next = NULL; - *source->lines_tail = src; - source->lines_tail = &src->next; - - src->eip = strtoull(src->line, &sep, 16); - if (*sep == ':') - src->eip = map__objdump_2ip(map, src->eip); - else /* this line has no ip info (e.g. source line) */ - src->eip = 0; + if (symbol__alloc_hist(sym, top.evlist->nr_entries) < 0) { + pthread_mutex_unlock(¬es->lock); + pr_err("Not enough memory for annotating '%s' symbol!\n", + sym->name); + sleep(1); + return err; } - pclose(file); + + err = symbol__annotate(sym, syme->map, 0); + if (err == 0) { out_assign: - sym_filter_entry = syme; - pthread_mutex_unlock(&source->lock); - return 0; + top.sym_filter_entry = syme; + } + + pthread_mutex_unlock(¬es->lock); + return err; } static void __zero_source_counters(struct sym_entry *syme) { - int i; - struct source_line *line; - - line = syme->src->lines; - while (line) { - for (i = 0; i < nr_counters; i++) - line->count[i] = 0; - line = line->next; - } + struct symbol *sym = sym_entry__symbol(syme); + symbol__annotate_zero_histograms(sym); } static void record_precise_ip(struct sym_entry *syme, int counter, u64 ip) { - struct source_line *line; - - if (syme != sym_filter_entry) - return; + struct annotation *notes; + struct symbol *sym; - if (pthread_mutex_trylock(&syme->src->lock)) + if (syme != top.sym_filter_entry) return; - if (syme->src == NULL || syme->src->source == NULL) - goto out_unlock; - - for (line = syme->src->lines; line; line = line->next) { - /* skip lines without IP info */ - if (line->eip == 0) - continue; - if (line->eip == ip) { - line->count[counter]++; - break; - } - if (line->eip > ip) - break; - } -out_unlock: - pthread_mutex_unlock(&syme->src->lock); -} - -#define PATTERN_LEN (BITS_PER_LONG / 4 + 2) - -static void lookup_sym_source(struct sym_entry *syme) -{ - struct symbol *symbol = sym_entry__symbol(syme); - struct source_line *line; - char pattern[PATTERN_LEN + 1]; - - sprintf(pattern, "%0*" PRIx64 " <", BITS_PER_LONG / 4, - map__rip_2objdump(syme->map, symbol->start)); - - pthread_mutex_lock(&syme->src->lock); - for (line = syme->src->lines; line; line = line->next) { - if (memcmp(line->line, pattern, PATTERN_LEN) == 0) { - syme->src->source = line; - break; - } - } - pthread_mutex_unlock(&syme->src->lock); -} + sym = sym_entry__symbol(syme); + notes = symbol__annotation(sym); -static void show_lines(struct source_line *queue, int count, int total) -{ - int i; - struct source_line *line; + if (pthread_mutex_trylock(¬es->lock)) + return; - line = queue; - for (i = 0; i < count; i++) { - float pcnt = 100.0*(float)line->count[sym_counter]/(float)total; + ip = syme->map->map_ip(syme->map, ip); + symbol__inc_addr_samples(sym, syme->map, counter, ip); - printf("%8li %4.1f%%\t%s\n", line->count[sym_counter], pcnt, line->line); - line = line->next; - } + pthread_mutex_unlock(¬es->lock); } -#define TRACE_COUNT 3 - static void show_details(struct sym_entry *syme) { + struct annotation *notes; struct symbol *symbol; - struct source_line *line; - struct source_line *line_queue = NULL; - int displayed = 0; - int line_queue_count = 0, total = 0, more = 0; + int more; if (!syme) return; - if (!syme->src->source) - lookup_sym_source(syme); - - if (!syme->src->source) - return; - symbol = sym_entry__symbol(syme); - printf("Showing %s for %s\n", event_name(sym_evsel), symbol->name); - printf(" Events Pcnt (>=%d%%)\n", sym_pcnt_filter); - - pthread_mutex_lock(&syme->src->lock); - line = syme->src->source; - while (line) { - total += line->count[sym_counter]; - line = line->next; - } - - line = syme->src->source; - while (line) { - float pcnt = 0.0; - - if (!line_queue_count) - line_queue = line; - line_queue_count++; - - if (line->count[sym_counter]) - pcnt = 100.0 * line->count[sym_counter] / (float)total; - if (pcnt >= (float)sym_pcnt_filter) { - if (displayed <= print_entries) - show_lines(line_queue, line_queue_count, total); - else more++; - displayed += line_queue_count; - line_queue_count = 0; - line_queue = NULL; - } else if (line_queue_count > TRACE_COUNT) { - line_queue = line_queue->next; - line_queue_count--; - } - - line->count[sym_counter] = zero ? 0 : line->count[sym_counter] * 7 / 8; - line = line->next; - } - pthread_mutex_unlock(&syme->src->lock); - if (more) - printf("%d lines not displayed, maybe increase display entries [e]\n", more); -} + notes = symbol__annotation(symbol); -/* - * Symbols will be added here in event__process_sample and will get out - * after decayed. - */ -static LIST_HEAD(active_symbols); -static pthread_mutex_t active_symbols_lock = PTHREAD_MUTEX_INITIALIZER; - -/* - * Ordering weight: count-1 * count-2 * ... / count-n - */ -static double sym_weight(const struct sym_entry *sym) -{ - double weight = sym->snap_count; - int counter; - - if (!display_weighted) - return weight; + pthread_mutex_lock(¬es->lock); - for (counter = 1; counter < nr_counters-1; counter++) - weight *= sym->count[counter]; + if (notes->src == NULL) + goto out_unlock; - weight /= (sym->count[counter] + 1); + printf("Showing %s for %s\n", event_name(top.sym_evsel), symbol->name); + printf(" Events Pcnt (>=%d%%)\n", sym_pcnt_filter); - return weight; + more = symbol__annotate_printf(symbol, syme->map, top.sym_evsel->idx, + 0, sym_pcnt_filter, top.print_entries, 4); + if (top.zero) + symbol__annotate_zero_histogram(symbol, top.sym_evsel->idx); + else + symbol__annotate_decay_histogram(symbol, top.sym_evsel->idx); + if (more != 0) + printf("%d lines not displayed, maybe increase display entries [e]\n", more); +out_unlock: + pthread_mutex_unlock(¬es->lock); } -static long samples; -static long kernel_samples, us_samples; -static long exact_samples; -static long guest_us_samples, guest_kernel_samples; static const char CONSOLE_CLEAR[] = "[H[2J"; static void __list_insert_active_sym(struct sym_entry *syme) { - list_add(&syme->node, &active_symbols); -} - -static void list_remove_active_sym(struct sym_entry *syme) -{ - pthread_mutex_lock(&active_symbols_lock); - list_del_init(&syme->node); - pthread_mutex_unlock(&active_symbols_lock); + list_add(&syme->node, &top.active_symbols); } -static void rb_insert_active_sym(struct rb_root *tree, struct sym_entry *se) +static void print_sym_table(struct perf_session *session) { - struct rb_node **p = &tree->rb_node; - struct rb_node *parent = NULL; - struct sym_entry *iter; - - while (*p != NULL) { - parent = *p; - iter = rb_entry(parent, struct sym_entry, rb_node); - - if (se->weight > iter->weight) - p = &(*p)->rb_left; - else - p = &(*p)->rb_right; - } - - rb_link_node(&se->rb_node, parent, p); - rb_insert_color(&se->rb_node, tree); -} - -static void print_sym_table(void) -{ - int printed = 0, j; - struct perf_evsel *counter; - int snap = !display_weighted ? sym_counter : 0; - float samples_per_sec = samples/delay_secs; - float ksamples_per_sec = kernel_samples/delay_secs; - float us_samples_per_sec = (us_samples)/delay_secs; - float guest_kernel_samples_per_sec = (guest_kernel_samples)/delay_secs; - float guest_us_samples_per_sec = (guest_us_samples)/delay_secs; - float esamples_percent = (100.0*exact_samples)/samples; - float sum_ksamples = 0.0; - struct sym_entry *syme, *n; - struct rb_root tmp = RB_ROOT; + char bf[160]; + int printed = 0; struct rb_node *nd; - int sym_width = 0, dso_width = 0, dso_short_width = 0; + struct sym_entry *syme; + struct rb_root tmp = RB_ROOT; const int win_width = winsize.ws_col - 1; - - samples = us_samples = kernel_samples = exact_samples = 0; - guest_kernel_samples = guest_us_samples = 0; - - /* Sort the active symbols */ - pthread_mutex_lock(&active_symbols_lock); - syme = list_entry(active_symbols.next, struct sym_entry, node); - pthread_mutex_unlock(&active_symbols_lock); - - list_for_each_entry_safe_from(syme, n, &active_symbols, node) { - syme->snap_count = syme->count[snap]; - if (syme->snap_count != 0) { - - if ((hide_user_symbols && - syme->origin == PERF_RECORD_MISC_USER) || - (hide_kernel_symbols && - syme->origin == PERF_RECORD_MISC_KERNEL)) { - list_remove_active_sym(syme); - continue; - } - syme->weight = sym_weight(syme); - rb_insert_active_sym(&tmp, syme); - sum_ksamples += syme->snap_count; - - for (j = 0; j < nr_counters; j++) - syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8; - } else - list_remove_active_sym(syme); - } + int sym_width, dso_width, dso_short_width; + float sum_ksamples = perf_top__decay_samples(&top, &tmp); puts(CONSOLE_CLEAR); - printf("%-*.*s\n", win_width, win_width, graph_dotted_line); - if (!perf_guest) { - printf(" PerfTop:%8.0f irqs/sec kernel:%4.1f%%" - " exact: %4.1f%% [", - samples_per_sec, - 100.0 - (100.0 * ((samples_per_sec - ksamples_per_sec) / - samples_per_sec)), - esamples_percent); - } else { - printf(" PerfTop:%8.0f irqs/sec kernel:%4.1f%% us:%4.1f%%" - " guest kernel:%4.1f%% guest us:%4.1f%%" - " exact: %4.1f%% [", - samples_per_sec, - 100.0 - (100.0 * ((samples_per_sec-ksamples_per_sec) / - samples_per_sec)), - 100.0 - (100.0 * ((samples_per_sec-us_samples_per_sec) / - samples_per_sec)), - 100.0 - (100.0 * ((samples_per_sec - - guest_kernel_samples_per_sec) / - samples_per_sec)), - 100.0 - (100.0 * ((samples_per_sec - - guest_us_samples_per_sec) / - samples_per_sec)), - esamples_percent); - } - - if (nr_counters == 1 || !display_weighted) { - struct perf_evsel *first; - first = list_entry(evsel_list.next, struct perf_evsel, node); - printf("%" PRIu64, (uint64_t)first->attr.sample_period); - if (freq) - printf("Hz "); - else - printf(" "); - } - - if (!display_weighted) - printf("%s", event_name(sym_evsel)); - else list_for_each_entry(counter, &evsel_list, node) { - if (counter->idx) - printf("/"); - - printf("%s", event_name(counter)); - } + perf_top__header_snprintf(&top, bf, sizeof(bf)); + printf("%s\n", bf); - printf( "], "); - - if (target_pid != -1) - printf(" (target_pid: %d", target_pid); - else if (target_tid != -1) - printf(" (target_tid: %d", target_tid); - else - printf(" (all"); - - if (cpu_list) - printf(", CPU%s: %s)\n", cpus->nr > 1 ? "s" : "", cpu_list); - else { - if (target_tid != -1) - printf(")\n"); - else - printf(", %d CPU%s)\n", cpus->nr, cpus->nr > 1 ? "s" : ""); - } + perf_top__reset_sample_counters(&top); printf("%-*.*s\n", win_width, win_width, graph_dotted_line); - if (sym_filter_entry) { - show_details(sym_filter_entry); - return; + if (session->hists.stats.total_lost != 0) { + color_fprintf(stdout, PERF_COLOR_RED, "WARNING:"); + printf(" LOST %" PRIu64 " events, Check IO/CPU overload\n", + session->hists.stats.total_lost); } - /* - * Find the longest symbol name that will be displayed - */ - for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) { - syme = rb_entry(nd, struct sym_entry, rb_node); - if (++printed > print_entries || - (int)syme->snap_count < count_filter) - continue; - - if (syme->map->dso->long_name_len > dso_width) - dso_width = syme->map->dso->long_name_len; - - if (syme->map->dso->short_name_len > dso_short_width) - dso_short_width = syme->map->dso->short_name_len; - - if (syme->name_len > sym_width) - sym_width = syme->name_len; + if (top.sym_filter_entry) { + show_details(top.sym_filter_entry); + return; } - printed = 0; + perf_top__find_widths(&top, &tmp, &dso_width, &dso_short_width, + &sym_width); if (sym_width + dso_width > winsize.ws_col - 29) { dso_width = dso_short_width; @@ -606,7 +290,7 @@ static void print_sym_table(void) sym_width = winsize.ws_col - dso_width - 29; } putchar('\n'); - if (nr_counters == 1) + if (top.evlist->nr_entries == 1) printf(" samples pcnt"); else printf(" weight samples pcnt"); @@ -615,7 +299,7 @@ static void print_sym_table(void) printf(" RIP "); printf(" %-*.*s DSO\n", sym_width, sym_width, "function"); printf(" %s _______ _____", - nr_counters == 1 ? " " : "______"); + top.evlist->nr_entries == 1 ? " " : "______"); if (verbose) printf(" ________________"); printf(" %-*.*s", sym_width, sym_width, graph_line); @@ -628,13 +312,14 @@ static void print_sym_table(void) syme = rb_entry(nd, struct sym_entry, rb_node); sym = sym_entry__symbol(syme); - if (++printed > print_entries || (int)syme->snap_count < count_filter) + if (++printed > top.print_entries || + (int)syme->snap_count < top.count_filter) continue; pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) / sum_ksamples)); - if (nr_counters == 1 || !display_weighted) + if (top.evlist->nr_entries == 1 || !top.display_weighted) printf("%20.2f ", syme->weight); else printf("%9.1f %10ld ", syme->weight, syme->snap_count); @@ -693,10 +378,8 @@ static void prompt_symbol(struct sym_entry **target, const char *msg) /* zero counters of active symbol */ if (syme) { - pthread_mutex_lock(&syme->src->lock); __zero_source_counters(syme); *target = NULL; - pthread_mutex_unlock(&syme->src->lock); } fprintf(stdout, "\n%s: ", msg); @@ -707,11 +390,11 @@ static void prompt_symbol(struct sym_entry **target, const char *msg) if (p) *p = 0; - pthread_mutex_lock(&active_symbols_lock); - syme = list_entry(active_symbols.next, struct sym_entry, node); - pthread_mutex_unlock(&active_symbols_lock); + pthread_mutex_lock(&top.active_symbols_lock); + syme = list_entry(top.active_symbols.next, struct sym_entry, node); + pthread_mutex_unlock(&top.active_symbols_lock); - list_for_each_entry_safe_from(syme, n, &active_symbols, node) { + list_for_each_entry_safe_from(syme, n, &top.active_symbols, node) { struct symbol *sym = sym_entry__symbol(syme); if (!strcmp(buf, sym->name)) { @@ -735,34 +418,34 @@ static void print_mapped_keys(void) { char *name = NULL; - if (sym_filter_entry) { - struct symbol *sym = sym_entry__symbol(sym_filter_entry); + if (top.sym_filter_entry) { + struct symbol *sym = sym_entry__symbol(top.sym_filter_entry); name = sym->name; } fprintf(stdout, "\nMapped keys:\n"); - fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", delay_secs); - fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", print_entries); + fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", top.delay_secs); + fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", top.print_entries); - if (nr_counters > 1) - fprintf(stdout, "\t[E] active event counter. \t(%s)\n", event_name(sym_evsel)); + if (top.evlist->nr_entries > 1) + fprintf(stdout, "\t[E] active event counter. \t(%s)\n", event_name(top.sym_evsel)); - fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", count_filter); + fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", top.count_filter); fprintf(stdout, "\t[F] annotate display filter (percent). \t(%d%%)\n", sym_pcnt_filter); fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL"); fprintf(stdout, "\t[S] stop annotation.\n"); - if (nr_counters > 1) - fprintf(stdout, "\t[w] toggle display weighted/count[E]r. \t(%d)\n", display_weighted ? 1 : 0); + if (top.evlist->nr_entries > 1) + fprintf(stdout, "\t[w] toggle display weighted/count[E]r. \t(%d)\n", top.display_weighted ? 1 : 0); fprintf(stdout, "\t[K] hide kernel_symbols symbols. \t(%s)\n", - hide_kernel_symbols ? "yes" : "no"); + top.hide_kernel_symbols ? "yes" : "no"); fprintf(stdout, "\t[U] hide user symbols. \t(%s)\n", - hide_user_symbols ? "yes" : "no"); - fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", zero ? 1 : 0); + top.hide_user_symbols ? "yes" : "no"); + fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", top.zero ? 1 : 0); fprintf(stdout, "\t[qQ] quit.\n"); } @@ -783,7 +466,7 @@ static int key_mapped(int c) return 1; case 'E': case 'w': - return nr_counters > 1 ? 1 : 0; + return top.evlist->nr_entries > 1 ? 1 : 0; default: break; } @@ -818,47 +501,47 @@ static void handle_keypress(struct perf_session *session, int c) switch (c) { case 'd': - prompt_integer(&delay_secs, "Enter display delay"); - if (delay_secs < 1) - delay_secs = 1; + prompt_integer(&top.delay_secs, "Enter display delay"); + if (top.delay_secs < 1) + top.delay_secs = 1; break; case 'e': - prompt_integer(&print_entries, "Enter display entries (lines)"); - if (print_entries == 0) { + prompt_integer(&top.print_entries, "Enter display entries (lines)"); + if (top.print_entries == 0) { sig_winch_handler(SIGWINCH); signal(SIGWINCH, sig_winch_handler); } else signal(SIGWINCH, SIG_DFL); break; case 'E': - if (nr_counters > 1) { + if (top.evlist->nr_entries > 1) { fprintf(stderr, "\nAvailable events:"); - list_for_each_entry(sym_evsel, &evsel_list, node) - fprintf(stderr, "\n\t%d %s", sym_evsel->idx, event_name(sym_evsel)); + list_for_each_entry(top.sym_evsel, &top.evlist->entries, node) + fprintf(stderr, "\n\t%d %s", top.sym_evsel->idx, event_name(top.sym_evsel)); - prompt_integer(&sym_counter, "Enter details event counter"); + prompt_integer(&top.sym_counter, "Enter details event counter"); - if (sym_counter >= nr_counters) { - sym_evsel = list_entry(evsel_list.next, struct perf_evsel, node); - sym_counter = 0; - fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(sym_evsel)); + if (top.sym_counter >= top.evlist->nr_entries) { + top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node); + top.sym_counter = 0; + fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(top.sym_evsel)); sleep(1); break; } - list_for_each_entry(sym_evsel, &evsel_list, node) - if (sym_evsel->idx == sym_counter) + list_for_each_entry(top.sym_evsel, &top.evlist->entries, node) + if (top.sym_evsel->idx == top.sym_counter) break; - } else sym_counter = 0; + } else top.sym_counter = 0; break; case 'f': - prompt_integer(&count_filter, "Enter display event count filter"); + prompt_integer(&top.count_filter, "Enter display event count filter"); break; case 'F': prompt_percent(&sym_pcnt_filter, "Enter details display event filter (percent)"); break; case 'K': - hide_kernel_symbols = !hide_kernel_symbols; + top.hide_kernel_symbols = !top.hide_kernel_symbols; break; case 'q': case 'Q': @@ -867,34 +550,50 @@ static void handle_keypress(struct perf_session *session, int c) perf_session__fprintf_dsos(session, stderr); exit(0); case 's': - prompt_symbol(&sym_filter_entry, "Enter details symbol"); + prompt_symbol(&top.sym_filter_entry, "Enter details symbol"); break; case 'S': - if (!sym_filter_entry) + if (!top.sym_filter_entry) break; else { - struct sym_entry *syme = sym_filter_entry; + struct sym_entry *syme = top.sym_filter_entry; - pthread_mutex_lock(&syme->src->lock); - sym_filter_entry = NULL; + top.sym_filter_entry = NULL; __zero_source_counters(syme); - pthread_mutex_unlock(&syme->src->lock); } break; case 'U': - hide_user_symbols = !hide_user_symbols; + top.hide_user_symbols = !top.hide_user_symbols; break; case 'w': - display_weighted = ~display_weighted; + top.display_weighted = ~top.display_weighted; break; case 'z': - zero = !zero; + top.zero = !top.zero; break; default: break; } } +static void *display_thread_tui(void *arg __used) +{ + int err = 0; + pthread_mutex_lock(&top.active_symbols_lock); + while (list_empty(&top.active_symbols)) { + err = pthread_cond_wait(&top.active_symbols_cond, + &top.active_symbols_lock); + if (err) + break; + } + pthread_mutex_unlock(&top.active_symbols_lock); + if (!err) + perf_top__tui_browser(&top); + exit_browser(0); + exit(0); + return NULL; +} + static void *display_thread(void *arg __used) { struct pollfd stdin_poll = { .fd = 0, .events = POLLIN }; @@ -909,13 +608,13 @@ static void *display_thread(void *arg __used) tc.c_cc[VTIME] = 0; repeat: - delay_msecs = delay_secs * 1000; + delay_msecs = top.delay_secs * 1000; tcsetattr(0, TCSANOW, &tc); /* trash return*/ getc(stdin); do { - print_sym_table(); + print_sym_table(session); } while (!poll(&stdin_poll, 1, delay_msecs) == 1); c = getc(stdin); @@ -930,6 +629,7 @@ repeat: /* Tag samples to be skipped. */ static const char *skip_symbols[] = { "default_idle", + "native_safe_halt", "cpu_idle", "enter_idle", "exit_idle", @@ -965,9 +665,9 @@ static int symbol_filter(struct map *map, struct symbol *sym) syme = symbol__priv(sym); syme->map = map; - syme->src = NULL; + symbol__annotate_init(map, sym); - if (!sym_filter_entry && sym_filter && !strcmp(name, sym_filter)) { + if (!top.sym_filter_entry && sym_filter && !strcmp(name, sym_filter)) { /* schedule initial sym_filter_entry setup */ sym_filter_entry_sched = syme; sym_filter = NULL; @@ -980,44 +680,40 @@ static int symbol_filter(struct map *map, struct symbol *sym) } } - if (!syme->skip) - syme->name_len = strlen(sym->name); - return 0; } -static void event__process_sample(const event_t *self, - struct sample_data *sample, - struct perf_session *session, - struct perf_evsel *evsel) +static void perf_event__process_sample(const union perf_event *event, + struct perf_sample *sample, + struct perf_session *session) { - u64 ip = self->ip.ip; + u64 ip = event->ip.ip; struct sym_entry *syme; struct addr_location al; struct machine *machine; - u8 origin = self->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; + u8 origin = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; - ++samples; + ++top.samples; switch (origin) { case PERF_RECORD_MISC_USER: - ++us_samples; - if (hide_user_symbols) + ++top.us_samples; + if (top.hide_user_symbols) return; machine = perf_session__find_host_machine(session); break; case PERF_RECORD_MISC_KERNEL: - ++kernel_samples; - if (hide_kernel_symbols) + ++top.kernel_samples; + if (top.hide_kernel_symbols) return; machine = perf_session__find_host_machine(session); break; case PERF_RECORD_MISC_GUEST_KERNEL: - ++guest_kernel_samples; - machine = perf_session__find_machine(session, self->ip.pid); + ++top.guest_kernel_samples; + machine = perf_session__find_machine(session, event->ip.pid); break; case PERF_RECORD_MISC_GUEST_USER: - ++guest_us_samples; + ++top.guest_us_samples; /* * TODO: we don't process guest user from host side * except simple counting. @@ -1029,15 +725,15 @@ static void event__process_sample(const event_t *self, if (!machine && perf_guest) { pr_err("Can't find guest [%d]'s kernel information\n", - self->ip.pid); + event->ip.pid); return; } - if (self->header.misc & PERF_RECORD_MISC_EXACT_IP) - exact_samples++; + if (event->header.misc & PERF_RECORD_MISC_EXACT_IP) + top.exact_samples++; - if (event__preprocess_sample(self, session, &al, sample, - symbol_filter) < 0 || + if (perf_event__preprocess_sample(event, session, &al, sample, + symbol_filter) < 0 || al.filtered) return; @@ -1055,8 +751,9 @@ static void event__process_sample(const event_t *self, */ if (al.map == machine->vmlinux_maps[MAP__FUNCTION] && RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) { - pr_err("The %s file can't be used\n", - symbol_conf.vmlinux_name); + ui__warning("The %s file can't be used\n", + symbol_conf.vmlinux_name); + exit_browser(0); exit(1); } @@ -1065,13 +762,13 @@ static void event__process_sample(const event_t *self, /* let's see, whether we need to install initial sym_filter_entry */ if (sym_filter_entry_sched) { - sym_filter_entry = sym_filter_entry_sched; + top.sym_filter_entry = sym_filter_entry_sched; sym_filter_entry_sched = NULL; - if (parse_source(sym_filter_entry) < 0) { - struct symbol *sym = sym_entry__symbol(sym_filter_entry); + if (parse_source(top.sym_filter_entry) < 0) { + struct symbol *sym = sym_entry__symbol(top.sym_filter_entry); pr_err("Can't annotate %s", sym->name); - if (sym_filter_entry->map->dso->origin == DSO__ORIG_KERNEL) { + if (top.sym_filter_entry->map->dso->origin == DSO__ORIG_KERNEL) { pr_err(": No vmlinux file was found in the path:\n"); machine__fprintf_vmlinux_path(machine, stderr); } else @@ -1082,166 +779,73 @@ static void event__process_sample(const event_t *self, syme = symbol__priv(al.sym); if (!syme->skip) { - syme->count[evsel->idx]++; + struct perf_evsel *evsel; + syme->origin = origin; + evsel = perf_evlist__id2evsel(top.evlist, sample->id); + assert(evsel != NULL); + syme->count[evsel->idx]++; record_precise_ip(syme, evsel->idx, ip); - pthread_mutex_lock(&active_symbols_lock); - if (list_empty(&syme->node) || !syme->node.next) + pthread_mutex_lock(&top.active_symbols_lock); + if (list_empty(&syme->node) || !syme->node.next) { + static bool first = true; __list_insert_active_sym(syme); - pthread_mutex_unlock(&active_symbols_lock); + if (first) { + pthread_cond_broadcast(&top.active_symbols_cond); + first = false; + } + } + pthread_mutex_unlock(&top.active_symbols_lock); } } -struct mmap_data { - void *base; - int mask; - unsigned int prev; -}; - -static int perf_evsel__alloc_mmap_per_thread(struct perf_evsel *evsel, - int ncpus, int nthreads) -{ - evsel->priv = xyarray__new(ncpus, nthreads, sizeof(struct mmap_data)); - return evsel->priv != NULL ? 0 : -ENOMEM; -} - -static void perf_evsel__free_mmap(struct perf_evsel *evsel) -{ - xyarray__delete(evsel->priv); - evsel->priv = NULL; -} - -static unsigned int mmap_read_head(struct mmap_data *md) -{ - struct perf_event_mmap_page *pc = md->base; - int head; - - head = pc->data_head; - rmb(); - - return head; -} - -static void perf_session__mmap_read_counter(struct perf_session *self, - struct perf_evsel *evsel, - int cpu, int thread_idx) +static void perf_session__mmap_read_cpu(struct perf_session *self, int cpu) { - struct xyarray *mmap_array = evsel->priv; - struct mmap_data *md = xyarray__entry(mmap_array, cpu, thread_idx); - unsigned int head = mmap_read_head(md); - unsigned int old = md->prev; - unsigned char *data = md->base + page_size; - struct sample_data sample; - int diff; - - /* - * If we're further behind than half the buffer, there's a chance - * the writer will bite our tail and mess up the samples under us. - * - * If we somehow ended up ahead of the head, we got messed up. - * - * In either case, truncate and restart at head. - */ - diff = head - old; - if (diff > md->mask / 2 || diff < 0) { - fprintf(stderr, "WARNING: failed to keep up with mmap data.\n"); - - /* - * head points to a known good entry, start there. - */ - old = head; - } - - for (; old != head;) { - event_t *event = (event_t *)&data[old & md->mask]; - - event_t event_copy; + struct perf_sample sample; + union perf_event *event; - size_t size = event->header.size; + while ((event = perf_evlist__read_on_cpu(top.evlist, cpu)) != NULL) { + perf_session__parse_sample(self, event, &sample); - /* - * Event straddles the mmap boundary -- header should always - * be inside due to u64 alignment of output. - */ - if ((old & md->mask) + size != ((old + size) & md->mask)) { - unsigned int offset = old; - unsigned int len = min(sizeof(*event), size), cpy; - void *dst = &event_copy; - - do { - cpy = min(md->mask + 1 - (offset & md->mask), len); - memcpy(dst, &data[offset & md->mask], cpy); - offset += cpy; - dst += cpy; - len -= cpy; - } while (len); - - event = &event_copy; - } - - event__parse_sample(event, self, &sample); if (event->header.type == PERF_RECORD_SAMPLE) - event__process_sample(event, &sample, self, evsel); + perf_event__process_sample(event, &sample, self); else - event__process(event, &sample, self); - old += size; + perf_event__process(event, &sample, self); } - - md->prev = old; } -static struct pollfd *event_array; - static void perf_session__mmap_read(struct perf_session *self) { - struct perf_evsel *counter; - int i, thread_index; - - for (i = 0; i < cpus->nr; i++) { - list_for_each_entry(counter, &evsel_list, node) { - for (thread_index = 0; - thread_index < threads->nr; - thread_index++) { - perf_session__mmap_read_counter(self, - counter, i, thread_index); - } - } - } -} + int i; -int nr_poll; -int group_fd; + for (i = 0; i < top.evlist->cpus->nr; i++) + perf_session__mmap_read_cpu(self, i); +} -static void start_counter(int i, struct perf_evsel *evsel) +static void start_counters(struct perf_evlist *evlist) { - struct xyarray *mmap_array = evsel->priv; - struct mmap_data *mm; - struct perf_event_attr *attr; - int cpu = -1; - int thread_index; - - if (target_tid == -1) - cpu = cpus->map[i]; + struct perf_evsel *counter; - attr = &evsel->attr; + list_for_each_entry(counter, &evlist->entries, node) { + struct perf_event_attr *attr = &counter->attr; - attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID; + attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID; - if (freq) { - attr->sample_type |= PERF_SAMPLE_PERIOD; - attr->freq = 1; - attr->sample_freq = freq; - } + if (top.freq) { + attr->sample_type |= PERF_SAMPLE_PERIOD; + attr->freq = 1; + attr->sample_freq = top.freq; + } - attr->inherit = (cpu < 0) && inherit; - attr->mmap = 1; + if (evlist->nr_entries > 1) { + attr->sample_type |= PERF_SAMPLE_ID; + attr->read_format |= PERF_FORMAT_ID; + } - for (thread_index = 0; thread_index < threads->nr; thread_index++) { + attr->mmap = 1; try_again: - FD(evsel, i, thread_index) = sys_perf_event_open(attr, - threads->map[thread_index], cpu, group_fd, 0); - - if (FD(evsel, i, thread_index) < 0) { + if (perf_evsel__open(counter, top.evlist->cpus, + top.evlist->threads, group, inherit) < 0) { int err = errno; if (err == EPERM || err == EACCES) @@ -1253,8 +857,8 @@ try_again: * based cpu-clock-tick sw counter, which * is always available even if no PMU support: */ - if (attr->type == PERF_TYPE_HARDWARE - && attr->config == PERF_COUNT_HW_CPU_CYCLES) { + if (attr->type == PERF_TYPE_HARDWARE && + attr->config == PERF_COUNT_HW_CPU_CYCLES) { if (verbose) warning(" ... trying to fall back to cpu-clock-ticks\n"); @@ -1264,39 +868,23 @@ try_again: goto try_again; } printf("\n"); - error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n", - FD(evsel, i, thread_index), strerror(err)); + error("sys_perf_event_open() syscall returned with %d " + "(%s). /bin/dmesg may provide additional information.\n", + err, strerror(err)); die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); exit(-1); } - assert(FD(evsel, i, thread_index) >= 0); - fcntl(FD(evsel, i, thread_index), F_SETFL, O_NONBLOCK); - - /* - * First counter acts as the group leader: - */ - if (group && group_fd == -1) - group_fd = FD(evsel, i, thread_index); - - event_array[nr_poll].fd = FD(evsel, i, thread_index); - event_array[nr_poll].events = POLLIN; - nr_poll++; - - mm = xyarray__entry(mmap_array, i, thread_index); - mm->prev = 0; - mm->mask = mmap_pages*page_size - 1; - mm->base = mmap(NULL, (mmap_pages+1)*page_size, - PROT_READ, MAP_SHARED, FD(evsel, i, thread_index), 0); - if (mm->base == MAP_FAILED) - die("failed to mmap with %d (%s)\n", errno, strerror(errno)); } + + if (perf_evlist__mmap(evlist, mmap_pages, false) < 0) + die("failed to mmap with %d (%s)\n", errno, strerror(errno)); } static int __cmd_top(void) { pthread_t thread; - struct perf_evsel *counter; - int i, ret; + struct perf_evsel *first; + int ret __used; /* * FIXME: perf_session__new should allow passing a O_MMAP, so that all this * mmap reading, etc is encapsulated in it. Use O_WRONLY for now. @@ -1305,23 +893,23 @@ static int __cmd_top(void) if (session == NULL) return -ENOMEM; - if (target_tid != -1) - event__synthesize_thread_map(threads, event__process, session); + if (top.target_tid != -1) + perf_event__synthesize_thread_map(top.evlist->threads, + perf_event__process, session); else - event__synthesize_threads(event__process, session); + perf_event__synthesize_threads(perf_event__process, session); - for (i = 0; i < cpus->nr; i++) { - group_fd = -1; - list_for_each_entry(counter, &evsel_list, node) - start_counter(i, counter); - } + start_counters(top.evlist); + first = list_entry(top.evlist->entries.next, struct perf_evsel, node); + perf_session__set_sample_type(session, first->attr.sample_type); /* Wait for a minimal set of events before starting the snapshot */ - poll(&event_array[0], nr_poll, 100); + poll(top.evlist->pollfd, top.evlist->nr_fds, 100); perf_session__mmap_read(session); - if (pthread_create(&thread, NULL, display_thread, session)) { + if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui : + display_thread), session)) { printf("Could not create display thread.\n"); exit(-1); } @@ -1337,12 +925,12 @@ static int __cmd_top(void) } while (1) { - int hits = samples; + u64 hits = top.samples; perf_session__mmap_read(session); - if (hits == samples) - ret = poll(event_array, nr_poll, 100); + if (hits == top.samples) + ret = poll(top.evlist->pollfd, top.evlist->nr_fds, 100); } return 0; @@ -1354,31 +942,31 @@ static const char * const top_usage[] = { }; static const struct option options[] = { - OPT_CALLBACK('e', "event", NULL, "event", + OPT_CALLBACK('e', "event", &top.evlist, "event", "event selector. use 'perf list' to list available events", parse_events), OPT_INTEGER('c', "count", &default_interval, "event period to sample"), - OPT_INTEGER('p', "pid", &target_pid, + OPT_INTEGER('p', "pid", &top.target_pid, "profile events on existing process id"), - OPT_INTEGER('t', "tid", &target_tid, + OPT_INTEGER('t', "tid", &top.target_tid, "profile events on existing thread id"), OPT_BOOLEAN('a', "all-cpus", &system_wide, "system-wide collection from all CPUs"), - OPT_STRING('C', "cpu", &cpu_list, "cpu", + OPT_STRING('C', "cpu", &top.cpu_list, "cpu", "list of cpus to monitor"), OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, "file", "vmlinux pathname"), - OPT_BOOLEAN('K', "hide_kernel_symbols", &hide_kernel_symbols, + OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols, "hide kernel symbols"), OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"), OPT_INTEGER('r', "realtime", &realtime_prio, "collect data with this RT SCHED_FIFO priority"), - OPT_INTEGER('d', "delay", &delay_secs, + OPT_INTEGER('d', "delay", &top.delay_secs, "number of seconds to delay between refreshes"), OPT_BOOLEAN('D', "dump-symtab", &dump_symtab, "dump the symbol table used for profiling"), - OPT_INTEGER('f', "count-filter", &count_filter, + OPT_INTEGER('f', "count-filter", &top.count_filter, "only display functions with more events than this"), OPT_BOOLEAN('g', "group", &group, "put the counters into a counter group"), @@ -1386,14 +974,16 @@ static const struct option options[] = { "child tasks inherit counters"), OPT_STRING('s', "sym-annotate", &sym_filter, "symbol name", "symbol to annotate"), - OPT_BOOLEAN('z', "zero", &zero, + OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"), - OPT_INTEGER('F', "freq", &freq, + OPT_INTEGER('F', "freq", &top.freq, "profile at this frequency"), - OPT_INTEGER('E', "entries", &print_entries, + OPT_INTEGER('E', "entries", &top.print_entries, "display this many functions"), - OPT_BOOLEAN('U', "hide_user_symbols", &hide_user_symbols, + OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols, "hide user symbols"), + OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"), + OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"), OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"), OPT_END() @@ -1404,64 +994,68 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) struct perf_evsel *pos; int status = -ENOMEM; + top.evlist = perf_evlist__new(NULL, NULL); + if (top.evlist == NULL) + return -ENOMEM; + page_size = sysconf(_SC_PAGE_SIZE); argc = parse_options(argc, argv, options, top_usage, 0); if (argc) usage_with_options(top_usage, options); - if (target_pid != -1) - target_tid = target_pid; + /* + * XXX For now start disabled, only using TUI if explicitely asked for. + * Change that when handle_keys equivalent gets written, live annotation + * done, etc. + */ + use_browser = 0; - threads = thread_map__new(target_pid, target_tid); - if (threads == NULL) { - pr_err("Problems finding threads of monitor\n"); - usage_with_options(top_usage, options); - } + if (use_stdio) + use_browser = 0; + else if (use_tui) + use_browser = 1; - event_array = malloc((sizeof(struct pollfd) * - MAX_NR_CPUS * MAX_COUNTERS * threads->nr)); - if (!event_array) - return -ENOMEM; + setup_browser(false); /* CPU and PID are mutually exclusive */ - if (target_tid > 0 && cpu_list) { + if (top.target_tid > 0 && top.cpu_list) { printf("WARNING: PID switch overriding CPU\n"); sleep(1); - cpu_list = NULL; + top.cpu_list = NULL; } - if (!nr_counters && perf_evsel_list__create_default() < 0) { + if (top.target_pid != -1) + top.target_tid = top.target_pid; + + if (perf_evlist__create_maps(top.evlist, top.target_pid, + top.target_tid, top.cpu_list) < 0) + usage_with_options(top_usage, options); + + if (!top.evlist->nr_entries && + perf_evlist__add_default(top.evlist) < 0) { pr_err("Not enough memory for event selector list\n"); return -ENOMEM; } - if (delay_secs < 1) - delay_secs = 1; + if (top.delay_secs < 1) + top.delay_secs = 1; /* * User specified count overrides default frequency. */ if (default_interval) - freq = 0; - else if (freq) { - default_interval = freq; + top.freq = 0; + else if (top.freq) { + default_interval = top.freq; } else { fprintf(stderr, "frequency and count are zero, aborting\n"); exit(EXIT_FAILURE); } - if (target_tid != -1) - cpus = cpu_map__dummy_new(); - else - cpus = cpu_map__new(cpu_list); - - if (cpus == NULL) - usage_with_options(top_usage, options); - - list_for_each_entry(pos, &evsel_list, node) { - if (perf_evsel__alloc_mmap_per_thread(pos, cpus->nr, threads->nr) < 0 || - perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) + list_for_each_entry(pos, &top.evlist->entries, node) { + if (perf_evsel__alloc_fd(pos, top.evlist->cpus->nr, + top.evlist->threads->nr) < 0) goto out_free_fd; /* * Fill in the ones not specifically initialized via -c: @@ -1472,26 +1066,28 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) pos->attr.sample_period = default_interval; } - sym_evsel = list_entry(evsel_list.next, struct perf_evsel, node); + if (perf_evlist__alloc_pollfd(top.evlist) < 0 || + perf_evlist__alloc_mmap(top.evlist) < 0) + goto out_free_fd; + + top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node); - symbol_conf.priv_size = (sizeof(struct sym_entry) + - (nr_counters + 1) * sizeof(unsigned long)); + symbol_conf.priv_size = (sizeof(struct sym_entry) + sizeof(struct annotation) + + (top.evlist->nr_entries + 1) * sizeof(unsigned long)); symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL); if (symbol__init() < 0) return -1; get_term_dimensions(&winsize); - if (print_entries == 0) { + if (top.print_entries == 0) { update_print_entries(&winsize); signal(SIGWINCH, sig_winch_handler); } status = __cmd_top(); out_free_fd: - list_for_each_entry(pos, &evsel_list, node) - perf_evsel__free_mmap(pos); - perf_evsel_list__delete(); + perf_evlist__delete(top.evlist); return status; } |