From 361c99a661a78ed22264649440e87fe4fe8da1f2 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 11 Jan 2011 20:56:53 -0200 Subject: [PATCH 01/29] perf evsel: Introduce perf_evlist Killing two more perf wide global variables: nr_counters and evsel_list as a list_head. There are more operations that will need more fields in perf_evlist, like the pollfd for polling all the fds in a list of evsel instances. Use option->value to pass the evsel_list to parse_{events,filters}. LKML-Reference: Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Makefile | 2 + tools/perf/builtin-record.c | 47 +++++++++++++--------- tools/perf/builtin-stat.c | 34 +++++++++------- tools/perf/builtin-top.c | 60 ++++++++++++++++------------ tools/perf/util/evlist.c | 53 ++++++++++++++++++++++++ tools/perf/util/evlist.h | 19 +++++++++ tools/perf/util/header.c | 17 ++++---- tools/perf/util/header.h | 7 +++- tools/perf/util/include/linux/list.h | 1 + tools/perf/util/parse-events.c | 51 +++++------------------ tools/perf/util/parse-events.h | 7 ---- 11 files changed, 180 insertions(+), 118 deletions(-) create mode 100644 tools/perf/util/evlist.c create mode 100644 tools/perf/util/evlist.h diff --git a/tools/perf/Makefile b/tools/perf/Makefile index 7141c42e1469..f20bc6f85611 100644 --- a/tools/perf/Makefile +++ b/tools/perf/Makefile @@ -402,6 +402,7 @@ LIB_H += util/debug.h LIB_H += util/debugfs.h LIB_H += util/event.h LIB_H += util/evsel.h +LIB_H += util/evlist.h LIB_H += util/exec_cmd.h LIB_H += util/types.h LIB_H += util/levenshtein.h @@ -440,6 +441,7 @@ LIB_OBJS += $(OUTPUT)util/ctype.o LIB_OBJS += $(OUTPUT)util/debugfs.o LIB_OBJS += $(OUTPUT)util/environment.o LIB_OBJS += $(OUTPUT)util/event.o +LIB_OBJS += $(OUTPUT)util/evlist.o LIB_OBJS += $(OUTPUT)util/evsel.o LIB_OBJS += $(OUTPUT)util/exec_cmd.o LIB_OBJS += $(OUTPUT)util/help.o diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index b2f729fdb317..252ace873d32 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -18,6 +18,7 @@ #include "util/header.h" #include "util/event.h" +#include "util/evlist.h" #include "util/evsel.h" #include "util/debug.h" #include "util/session.h" @@ -66,6 +67,7 @@ static bool sample_address = false; static bool sample_time = false; static bool no_buildid = false; static bool no_buildid_cache = false; +static struct perf_evlist *evsel_list; static long samples = 0; static u64 bytes_written = 0; @@ -229,7 +231,8 @@ static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int n return h_attr; } -static void create_counter(struct perf_evsel *evsel, int cpu) +static void create_counter(struct perf_evlist *evlist, + struct perf_evsel *evsel, int cpu) { char *filter = evsel->filter; struct perf_event_attr *attr = &evsel->attr; @@ -263,7 +266,7 @@ static void create_counter(struct perf_evsel *evsel, int cpu) attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID; - if (nr_counters > 1) + if (evlist->nr_entries > 1) attr->sample_type |= PERF_SAMPLE_ID; /* @@ -410,7 +413,7 @@ try_again: if (evsel->idx || thread_index) { struct perf_evsel *first; - first = list_entry(evsel_list.next, struct perf_evsel, node); + first = list_entry(evlist->entries.next, struct perf_evsel, node); ret = ioctl(FD(evsel, nr_cpu, thread_index), PERF_EVENT_IOC_SET_OUTPUT, FD(first, nr_cpu, 0)); @@ -449,14 +452,14 @@ try_again: sample_type = attr->sample_type; } -static void open_counters(int cpu) +static void open_counters(struct perf_evlist *evlist, int cpu) { struct perf_evsel *pos; group_fd = -1; - list_for_each_entry(pos, &evsel_list, node) - create_counter(pos, cpu); + list_for_each_entry(pos, &evlist->entries, node) + create_counter(evlist, pos, cpu); nr_cpu++; } @@ -481,9 +484,9 @@ static void atexit_header(void) if (!no_buildid) process_buildids(); - perf_header__write(&session->header, output, true); + perf_header__write(&session->header, evsel_list, output, true); perf_session__delete(session); - perf_evsel_list__delete(); + perf_evlist__delete(evsel_list); symbol__exit(); } } @@ -611,7 +614,7 @@ static int __cmd_record(int argc, const char **argv) goto out_delete_session; } - if (have_tracepoints(&evsel_list)) + if (have_tracepoints(&evsel_list->entries)) perf_header__set_feat(&session->header, HEADER_TRACE_INFO); /* @@ -674,10 +677,10 @@ static int __cmd_record(int argc, const char **argv) } if (!system_wide && no_inherit && !cpu_list) { - open_counters(-1); + open_counters(evsel_list, -1); } else { for (i = 0; i < cpus->nr; i++) - open_counters(cpus->map[i]); + open_counters(evsel_list, cpus->map[i]); } perf_session__set_sample_type(session, sample_type); @@ -687,7 +690,8 @@ static int __cmd_record(int argc, const char **argv) if (err < 0) return err; } else if (file_new) { - err = perf_header__write(&session->header, output, false); + err = perf_header__write(&session->header, evsel_list, + output, false); if (err < 0) return err; } @@ -712,7 +716,7 @@ static int __cmd_record(int argc, const char **argv) return err; } - if (have_tracepoints(&evsel_list)) { + if (have_tracepoints(&evsel_list->entries)) { /* * FIXME err <= 0 here actually means that * there were no tracepoints so its not really @@ -721,7 +725,7 @@ static int __cmd_record(int argc, const char **argv) * return this more properly and also * propagate errors that now are calling die() */ - err = event__synthesize_tracing_data(output, &evsel_list, + err = event__synthesize_tracing_data(output, evsel_list, process_synthesized_event, session); if (err <= 0) { @@ -797,7 +801,7 @@ static int __cmd_record(int argc, const char **argv) for (i = 0; i < nr_cpu; i++) { struct perf_evsel *pos; - list_for_each_entry(pos, &evsel_list, node) { + list_for_each_entry(pos, &evsel_list->entries, node) { for (thread = 0; thread < threads->nr; thread++) @@ -838,10 +842,10 @@ static const char * const record_usage[] = { static bool force, append_file; const struct option record_options[] = { - OPT_CALLBACK('e', "event", NULL, "event", + OPT_CALLBACK('e', "event", &evsel_list, "event", "event selector. use 'perf list' to list available events", parse_events), - OPT_CALLBACK(0, "filter", NULL, "filter", + OPT_CALLBACK(0, "filter", &evsel_list, "filter", "event filter", parse_filter), OPT_INTEGER('p', "pid", &target_pid, "record events on existing process id"), @@ -892,6 +896,10 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) int err = -ENOMEM; struct perf_evsel *pos; + evsel_list = perf_evlist__new(); + if (evsel_list == NULL) + return -ENOMEM; + argc = parse_options(argc, argv, record_options, record_usage, PARSE_OPT_STOP_AT_NON_OPTION); if (!argc && target_pid == -1 && target_tid == -1 && @@ -913,7 +921,8 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) if (no_buildid_cache || no_buildid) disable_buildid_cache(); - if (list_empty(&evsel_list) && perf_evsel_list__create_default() < 0) { + if (evsel_list->nr_entries == 0 && + perf_evlist__add_default(evsel_list) < 0) { pr_err("Not enough memory for event selector list\n"); goto out_symbol_exit; } @@ -933,7 +942,7 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) return -1; } - list_for_each_entry(pos, &evsel_list, node) { + list_for_each_entry(pos, &evsel_list->entries, node) { if (perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) goto out_free_fd; if (perf_header__push_event(pos->attr.config, event_name(pos))) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index a482a191a0ca..da9090245934 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -43,6 +43,7 @@ #include "util/parse-options.h" #include "util/parse-events.h" #include "util/event.h" +#include "util/evlist.h" #include "util/evsel.h" #include "util/debug.h" #include "util/header.h" @@ -71,6 +72,8 @@ static struct perf_event_attr default_attrs[] = { }; +struct perf_evlist *evsel_list; + static bool system_wide = false; static struct cpu_map *cpus; static int run_idx = 0; @@ -309,7 +312,7 @@ static int run_perf_stat(int argc __used, const char **argv) close(child_ready_pipe[0]); } - list_for_each_entry(counter, &evsel_list, node) { + list_for_each_entry(counter, &evsel_list->entries, node) { if (create_perf_stat_counter(counter) < 0) { if (errno == -EPERM || errno == -EACCES) { error("You may not have permission to collect %sstats.\n" @@ -347,12 +350,12 @@ static int run_perf_stat(int argc __used, const char **argv) update_stats(&walltime_nsecs_stats, t1 - t0); if (no_aggr) { - list_for_each_entry(counter, &evsel_list, node) { + list_for_each_entry(counter, &evsel_list->entries, node) { read_counter(counter); perf_evsel__close_fd(counter, cpus->nr, 1); } } else { - list_for_each_entry(counter, &evsel_list, node) { + list_for_each_entry(counter, &evsel_list->entries, node) { read_counter_aggr(counter); perf_evsel__close_fd(counter, cpus->nr, threads->nr); } @@ -555,10 +558,10 @@ static void print_stat(int argc, const char **argv) } if (no_aggr) { - list_for_each_entry(counter, &evsel_list, node) + list_for_each_entry(counter, &evsel_list->entries, node) print_counter(counter); } else { - list_for_each_entry(counter, &evsel_list, node) + list_for_each_entry(counter, &evsel_list->entries, node) print_counter_aggr(counter); } @@ -610,7 +613,7 @@ static int stat__set_big_num(const struct option *opt __used, } static const struct option options[] = { - OPT_CALLBACK('e', "event", NULL, "event", + OPT_CALLBACK('e', "event", &evsel_list, "event", "event selector. use 'perf list' to list available events", parse_events), OPT_BOOLEAN('i', "no-inherit", &no_inherit, @@ -648,6 +651,10 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used) setlocale(LC_ALL, ""); + evsel_list = perf_evlist__new(); + if (evsel_list == NULL) + return -ENOMEM; + argc = parse_options(argc, argv, options, stat_usage, PARSE_OPT_STOP_AT_NON_OPTION); @@ -679,17 +686,14 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used) usage_with_options(stat_usage, options); /* Set attrs and nr_counters if no event is selected and !null_run */ - if (!null_run && !nr_counters) { + if (!null_run && !evsel_list->nr_entries) { size_t c; - nr_counters = ARRAY_SIZE(default_attrs); - for (c = 0; c < ARRAY_SIZE(default_attrs); ++c) { - pos = perf_evsel__new(&default_attrs[c], - nr_counters); + pos = perf_evsel__new(&default_attrs[c], c); if (pos == NULL) goto out; - list_add(&pos->node, &evsel_list); + perf_evlist__add(evsel_list, pos); } } @@ -713,7 +717,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used) return -1; } - list_for_each_entry(pos, &evsel_list, node) { + list_for_each_entry(pos, &evsel_list->entries, node) { if (perf_evsel__alloc_stat_priv(pos) < 0 || perf_evsel__alloc_counts(pos, cpus->nr) < 0 || perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) @@ -741,9 +745,9 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used) if (status != -1) print_stat(argc, argv); out_free_fd: - list_for_each_entry(pos, &evsel_list, node) + list_for_each_entry(pos, &evsel_list->entries, node) perf_evsel__free_stat_priv(pos); - perf_evsel_list__delete(); + perf_evlist__delete(evsel_list); out: thread_map__delete(threads); threads = NULL; diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index b6998e055767..216b62ed4b89 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -21,6 +21,7 @@ #include "perf.h" #include "util/color.h" +#include "util/evlist.h" #include "util/evsel.h" #include "util/session.h" #include "util/symbol.h" @@ -60,6 +61,8 @@ #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) +struct perf_evlist *evsel_list; + static bool system_wide = false; static int default_interval = 0; @@ -267,7 +270,7 @@ static void __zero_source_counters(struct sym_entry *syme) line = syme->src->lines; while (line) { - for (i = 0; i < nr_counters; i++) + for (i = 0; i < evsel_list->nr_entries; i++) line->count[i] = 0; line = line->next; } @@ -414,7 +417,7 @@ static double sym_weight(const struct sym_entry *sym) if (!display_weighted) return weight; - for (counter = 1; counter < nr_counters-1; counter++) + for (counter = 1; counter < evsel_list->nr_entries - 1; counter++) weight *= sym->count[counter]; weight /= (sym->count[counter] + 1); @@ -501,7 +504,7 @@ static void print_sym_table(void) rb_insert_active_sym(&tmp, syme); sum_ksamples += syme->snap_count; - for (j = 0; j < nr_counters; j++) + for (j = 0; j < evsel_list->nr_entries; j++) syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8; } else list_remove_active_sym(syme); @@ -535,9 +538,9 @@ static void print_sym_table(void) esamples_percent); } - if (nr_counters == 1 || !display_weighted) { + if (evsel_list->nr_entries == 1 || !display_weighted) { struct perf_evsel *first; - first = list_entry(evsel_list.next, struct perf_evsel, node); + first = list_entry(evsel_list->entries.next, struct perf_evsel, node); printf("%" PRIu64, (uint64_t)first->attr.sample_period); if (freq) printf("Hz "); @@ -547,7 +550,7 @@ static void print_sym_table(void) if (!display_weighted) printf("%s", event_name(sym_evsel)); - else list_for_each_entry(counter, &evsel_list, node) { + else list_for_each_entry(counter, &evsel_list->entries, node) { if (counter->idx) printf("/"); @@ -606,7 +609,7 @@ static void print_sym_table(void) sym_width = winsize.ws_col - dso_width - 29; } putchar('\n'); - if (nr_counters == 1) + if (evsel_list->nr_entries == 1) printf(" samples pcnt"); else printf(" weight samples pcnt"); @@ -615,7 +618,7 @@ static void print_sym_table(void) printf(" RIP "); printf(" %-*.*s DSO\n", sym_width, sym_width, "function"); printf(" %s _______ _____", - nr_counters == 1 ? " " : "______"); + evsel_list->nr_entries == 1 ? " " : "______"); if (verbose) printf(" ________________"); printf(" %-*.*s", sym_width, sym_width, graph_line); @@ -634,7 +637,7 @@ static void print_sym_table(void) pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) / sum_ksamples)); - if (nr_counters == 1 || !display_weighted) + if (evsel_list->nr_entries == 1 || !display_weighted) printf("%20.2f ", syme->weight); else printf("%9.1f %10ld ", syme->weight, syme->snap_count); @@ -744,7 +747,7 @@ static void print_mapped_keys(void) fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", delay_secs); fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", print_entries); - if (nr_counters > 1) + if (evsel_list->nr_entries > 1) fprintf(stdout, "\t[E] active event counter. \t(%s)\n", event_name(sym_evsel)); fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", count_filter); @@ -753,7 +756,7 @@ static void print_mapped_keys(void) fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL"); fprintf(stdout, "\t[S] stop annotation.\n"); - if (nr_counters > 1) + if (evsel_list->nr_entries > 1) fprintf(stdout, "\t[w] toggle display weighted/count[E]r. \t(%d)\n", display_weighted ? 1 : 0); fprintf(stdout, @@ -783,7 +786,7 @@ static int key_mapped(int c) return 1; case 'E': case 'w': - return nr_counters > 1 ? 1 : 0; + return evsel_list->nr_entries > 1 ? 1 : 0; default: break; } @@ -831,22 +834,22 @@ static void handle_keypress(struct perf_session *session, int c) signal(SIGWINCH, SIG_DFL); break; case 'E': - if (nr_counters > 1) { + if (evsel_list->nr_entries > 1) { fprintf(stderr, "\nAvailable events:"); - list_for_each_entry(sym_evsel, &evsel_list, node) + list_for_each_entry(sym_evsel, &evsel_list->entries, node) fprintf(stderr, "\n\t%d %s", sym_evsel->idx, event_name(sym_evsel)); prompt_integer(&sym_counter, "Enter details event counter"); - if (sym_counter >= nr_counters) { - sym_evsel = list_entry(evsel_list.next, struct perf_evsel, node); + if (sym_counter >= evsel_list->nr_entries) { + sym_evsel = list_entry(evsel_list->entries.next, struct perf_evsel, node); sym_counter = 0; fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(sym_evsel)); sleep(1); break; } - list_for_each_entry(sym_evsel, &evsel_list, node) + list_for_each_entry(sym_evsel, &evsel_list->entries, node) if (sym_evsel->idx == sym_counter) break; } else sym_counter = 0; @@ -1198,7 +1201,7 @@ static void perf_session__mmap_read(struct perf_session *self) int i, thread_index; for (i = 0; i < cpus->nr; i++) { - list_for_each_entry(counter, &evsel_list, node) { + list_for_each_entry(counter, &evsel_list->entries, node) { for (thread_index = 0; thread_index < threads->nr; thread_index++) { @@ -1312,7 +1315,7 @@ static int __cmd_top(void) for (i = 0; i < cpus->nr; i++) { group_fd = -1; - list_for_each_entry(counter, &evsel_list, node) + list_for_each_entry(counter, &evsel_list->entries, node) start_counter(i, counter); } @@ -1354,7 +1357,7 @@ static const char * const top_usage[] = { }; static const struct option options[] = { - OPT_CALLBACK('e', "event", NULL, "event", + OPT_CALLBACK('e', "event", &evsel_list, "event", "event selector. use 'perf list' to list available events", parse_events), OPT_INTEGER('c', "count", &default_interval, @@ -1404,6 +1407,10 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) struct perf_evsel *pos; int status = -ENOMEM; + evsel_list = perf_evlist__new(); + if (evsel_list == NULL) + return -ENOMEM; + page_size = sysconf(_SC_PAGE_SIZE); argc = parse_options(argc, argv, options, top_usage, 0); @@ -1431,7 +1438,8 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) cpu_list = NULL; } - if (!nr_counters && perf_evsel_list__create_default() < 0) { + if (!evsel_list->nr_entries && + perf_evlist__add_default(evsel_list) < 0) { pr_err("Not enough memory for event selector list\n"); return -ENOMEM; } @@ -1459,7 +1467,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) if (cpus == NULL) usage_with_options(top_usage, options); - list_for_each_entry(pos, &evsel_list, node) { + list_for_each_entry(pos, &evsel_list->entries, node) { if (perf_evsel__alloc_mmap_per_thread(pos, cpus->nr, threads->nr) < 0 || perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) goto out_free_fd; @@ -1472,10 +1480,10 @@ 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); + sym_evsel = list_entry(evsel_list->entries.next, struct perf_evsel, node); symbol_conf.priv_size = (sizeof(struct sym_entry) + - (nr_counters + 1) * sizeof(unsigned long)); + (evsel_list->nr_entries + 1) * sizeof(unsigned long)); symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL); if (symbol__init() < 0) @@ -1489,9 +1497,9 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) status = __cmd_top(); out_free_fd: - list_for_each_entry(pos, &evsel_list, node) + list_for_each_entry(pos, &evsel_list->entries, node) perf_evsel__free_mmap(pos); - perf_evsel_list__delete(); + perf_evlist__delete(evsel_list); return status; } diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c new file mode 100644 index 000000000000..7b4faec23737 --- /dev/null +++ b/tools/perf/util/evlist.c @@ -0,0 +1,53 @@ +#include "evlist.h" +#include "evsel.h" +#include "util.h" + +struct perf_evlist *perf_evlist__new(void) +{ + struct perf_evlist *evlist = zalloc(sizeof(*evlist)); + + if (evlist != NULL) { + INIT_LIST_HEAD(&evlist->entries); + } + + return evlist; +} + +static void perf_evlist__purge(struct perf_evlist *evlist) +{ + struct perf_evsel *pos, *n; + + list_for_each_entry_safe(pos, n, &evlist->entries, node) { + list_del_init(&pos->node); + perf_evsel__delete(pos); + } + + evlist->nr_entries = 0; +} + +void perf_evlist__delete(struct perf_evlist *evlist) +{ + perf_evlist__purge(evlist); + free(evlist); +} + +void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry) +{ + list_add_tail(&entry->node, &evlist->entries); + ++evlist->nr_entries; +} + +int perf_evlist__add_default(struct perf_evlist *evlist) +{ + struct perf_event_attr attr = { + .type = PERF_TYPE_HARDWARE, + .config = PERF_COUNT_HW_CPU_CYCLES, + }; + struct perf_evsel *evsel = perf_evsel__new(&attr, 0); + + if (evsel == NULL) + return -ENOMEM; + + perf_evlist__add(evlist, evsel); + return 0; +} diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h new file mode 100644 index 000000000000..48db91a8abf3 --- /dev/null +++ b/tools/perf/util/evlist.h @@ -0,0 +1,19 @@ +#ifndef __PERF_EVLIST_H +#define __PERF_EVLIST_H 1 + +#include + +struct perf_evlist { + struct list_head entries; + int nr_entries; +}; + +struct perf_evsel; + +struct perf_evlist *perf_evlist__new(void); +void perf_evlist__delete(struct perf_evlist *evlist); + +void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry); +int perf_evlist__add_default(struct perf_evlist *evlist); + +#endif /* __PERF_EVLIST_H */ diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c index f6a929e74981..f0138d472339 100644 --- a/tools/perf/util/header.c +++ b/tools/perf/util/header.c @@ -8,6 +8,7 @@ #include #include +#include "evlist.h" #include "util.h" #include "header.h" #include "../perf.h" @@ -428,7 +429,8 @@ static bool perf_session__read_build_ids(struct perf_session *self, bool with_hi return ret; } -static int perf_header__adds_write(struct perf_header *self, int fd) +static int perf_header__adds_write(struct perf_header *self, + struct perf_evlist *evlist, int fd) { int nr_sections; struct perf_session *session; @@ -463,7 +465,7 @@ static int perf_header__adds_write(struct perf_header *self, int fd) /* Write trace info */ trace_sec->offset = lseek(fd, 0, SEEK_CUR); - read_tracing_data(fd, &evsel_list); + read_tracing_data(fd, &evlist->entries); trace_sec->size = lseek(fd, 0, SEEK_CUR) - trace_sec->offset; } @@ -513,7 +515,8 @@ int perf_header__write_pipe(int fd) return 0; } -int perf_header__write(struct perf_header *self, int fd, bool at_exit) +int perf_header__write(struct perf_header *self, struct perf_evlist *evlist, + int fd, bool at_exit) { struct perf_file_header f_header; struct perf_file_attr f_attr; @@ -566,7 +569,7 @@ int perf_header__write(struct perf_header *self, int fd, bool at_exit) self->data_offset = lseek(fd, 0, SEEK_CUR); if (at_exit) { - err = perf_header__adds_write(self, fd); + err = perf_header__adds_write(self, evlist, fd); if (err < 0) return err; } @@ -1133,7 +1136,7 @@ int event__process_event_type(event_t *self, return 0; } -int event__synthesize_tracing_data(int fd, struct list_head *pattrs, +int event__synthesize_tracing_data(int fd, struct perf_evlist *evlist, event__handler_t process, struct perf_session *session __unused) { @@ -1144,7 +1147,7 @@ int event__synthesize_tracing_data(int fd, struct list_head *pattrs, memset(&ev, 0, sizeof(ev)); ev.tracing_data.header.type = PERF_RECORD_HEADER_TRACING_DATA; - size = read_tracing_data_size(fd, pattrs); + size = read_tracing_data_size(fd, &evlist->entries); if (size <= 0) return size; aligned_size = ALIGN(size, sizeof(u64)); @@ -1154,7 +1157,7 @@ int event__synthesize_tracing_data(int fd, struct list_head *pattrs, process(&ev, NULL, session); - err = read_tracing_data(fd, pattrs); + err = read_tracing_data(fd, &evlist->entries); write_padded(fd, NULL, 0, padding); return aligned_size; diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h index 33f16be7b72f..65afd7f74e0d 100644 --- a/tools/perf/util/header.h +++ b/tools/perf/util/header.h @@ -65,8 +65,11 @@ struct perf_header { int perf_header__init(struct perf_header *self); void perf_header__exit(struct perf_header *self); +struct perf_evlist; + int perf_header__read(struct perf_session *session, int fd); -int perf_header__write(struct perf_header *self, int fd, bool at_exit); +int perf_header__write(struct perf_header *self, struct perf_evlist *evlist, + int fd, bool at_exit); int perf_header__write_pipe(int fd); int perf_header__add_attr(struct perf_header *self, @@ -113,7 +116,7 @@ int event__synthesize_event_types(event__handler_t process, int event__process_event_type(event_t *self, struct perf_session *session); -int event__synthesize_tracing_data(int fd, struct list_head *pattrs, +int event__synthesize_tracing_data(int fd, struct perf_evlist *evlist, event__handler_t process, struct perf_session *session); int event__process_tracing_data(event_t *self, diff --git a/tools/perf/util/include/linux/list.h b/tools/perf/util/include/linux/list.h index f5ca26e53fbb..356c7e467b83 100644 --- a/tools/perf/util/include/linux/list.h +++ b/tools/perf/util/include/linux/list.h @@ -1,3 +1,4 @@ +#include #include "../../../../include/linux/list.h" #ifndef PERF_LIST_H diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 135f69baf966..d3086cecd2dd 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -1,6 +1,7 @@ #include "../../../include/linux/hw_breakpoint.h" #include "util.h" #include "../perf.h" +#include "evlist.h" #include "evsel.h" #include "parse-options.h" #include "parse-events.h" @@ -11,10 +12,6 @@ #include "header.h" #include "debugfs.h" -int nr_counters; - -LIST_HEAD(evsel_list); - struct event_symbol { u8 type; u64 config; @@ -778,8 +775,9 @@ modifier: return ret; } -int parse_events(const struct option *opt __used, const char *str, int unset __used) +int parse_events(const struct option *opt, const char *str, int unset __used) { + struct perf_evlist *evlist = *(struct perf_evlist **)opt->value; struct perf_event_attr attr; enum event_result ret; @@ -794,12 +792,10 @@ int parse_events(const struct option *opt __used, const char *str, int unset __u if (ret != EVT_HANDLED_ALL) { struct perf_evsel *evsel; - evsel = perf_evsel__new(&attr, - nr_counters); + evsel = perf_evsel__new(&attr, evlist->nr_entries); if (evsel == NULL) return -1; - list_add_tail(&evsel->node, &evsel_list); - ++nr_counters; + perf_evlist__add(evlist, evsel); } if (*str == 0) @@ -813,13 +809,14 @@ int parse_events(const struct option *opt __used, const char *str, int unset __u return 0; } -int parse_filter(const struct option *opt __used, const char *str, +int parse_filter(const struct option *opt, const char *str, int unset __used) { + struct perf_evlist *evlist = *(struct perf_evlist **)opt->value; struct perf_evsel *last = NULL; - if (!list_empty(&evsel_list)) - last = list_entry(evsel_list.prev, struct perf_evsel, node); + if (evlist->nr_entries > 0) + last = list_entry(evlist->entries.prev, struct perf_evsel, node); if (last == NULL || last->attr.type != PERF_TYPE_TRACEPOINT) { fprintf(stderr, @@ -981,33 +978,3 @@ void print_events(void) exit(129); } - -int perf_evsel_list__create_default(void) -{ - struct perf_evsel *evsel; - struct perf_event_attr attr; - - memset(&attr, 0, sizeof(attr)); - attr.type = PERF_TYPE_HARDWARE; - attr.config = PERF_COUNT_HW_CPU_CYCLES; - - evsel = perf_evsel__new(&attr, 0); - - if (evsel == NULL) - return -ENOMEM; - - list_add(&evsel->node, &evsel_list); - ++nr_counters; - return 0; -} - -void perf_evsel_list__delete(void) -{ - struct perf_evsel *pos, *n; - - list_for_each_entry_safe(pos, n, &evsel_list, node) { - list_del_init(&pos->node); - perf_evsel__delete(pos); - } - nr_counters = 0; -} diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h index 458e3ecf17af..cf7e94abb676 100644 --- a/tools/perf/util/parse-events.h +++ b/tools/perf/util/parse-events.h @@ -9,11 +9,6 @@ struct list_head; struct perf_evsel; -extern struct list_head evsel_list; - -int perf_evsel_list__create_default(void); -void perf_evsel_list__delete(void); - struct option; struct tracepoint_path { @@ -25,8 +20,6 @@ struct tracepoint_path { extern struct tracepoint_path *tracepoint_id_to_path(u64 config); extern bool have_tracepoints(struct list_head *evlist); -extern int nr_counters; - const char *event_name(struct perf_evsel *event); extern const char *__event_name(int type, u64 config); From 5c581041cf97aa7980b442de81ddea8273d6dcde Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 11 Jan 2011 22:30:02 -0200 Subject: [PATCH 02/29] perf evlist: Adopt the pollfd array Allocating just the space needed for nr_cpus * nr_threads * nr_evsels, not the MAX_NR_CPUS and counters. LKML-Reference: Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 20 +++++++------------- tools/perf/builtin-top.c | 26 +++++++++++--------------- tools/perf/util/evlist.c | 9 +++++++++ tools/perf/util/evlist.h | 6 ++++++ 4 files changed, 33 insertions(+), 28 deletions(-) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 252ace873d32..1614d89b4765 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -72,9 +72,6 @@ static struct perf_evlist *evsel_list; static long samples = 0; static u64 bytes_written = 0; -static struct pollfd *event_array; - -static int nr_poll = 0; static int nr_cpu = 0; static int file_new = 1; @@ -432,9 +429,9 @@ try_again: exit(-1); } - event_array[nr_poll].fd = FD(evsel, nr_cpu, thread_index); - event_array[nr_poll].events = POLLIN; - nr_poll++; + evlist->pollfd[evlist->nr_fds].fd = FD(evsel, nr_cpu, thread_index); + evlist->pollfd[evlist->nr_fds].events = POLLIN; + evlist->nr_fds++; } if (filter != NULL) { @@ -793,7 +790,7 @@ static int __cmd_record(int argc, const char **argv) if (hits == samples) { if (done) break; - err = poll(event_array, nr_poll, -1); + err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1); waking++; } @@ -948,9 +945,8 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) if (perf_header__push_event(pos->attr.config, event_name(pos))) goto out_free_fd; } - event_array = malloc((sizeof(struct pollfd) * MAX_NR_CPUS * - MAX_COUNTERS * threads->nr)); - if (!event_array) + + if (perf_evlist__alloc_pollfd(evsel_list, cpus->nr, threads->nr) < 0) goto out_free_fd; if (user_interval != ULLONG_MAX) @@ -968,13 +964,11 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) } else { fprintf(stderr, "frequency and count are zero, aborting\n"); err = -EINVAL; - goto out_free_event_array; + goto out_free_fd; } err = __cmd_record(argc, argv); -out_free_event_array: - free(event_array); out_free_fd: thread_map__delete(threads); threads = NULL; diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 216b62ed4b89..1bc465215fc6 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -1193,8 +1193,6 @@ static void perf_session__mmap_read_counter(struct perf_session *self, md->prev = old; } -static struct pollfd *event_array; - static void perf_session__mmap_read(struct perf_session *self) { struct perf_evsel *counter; @@ -1212,10 +1210,10 @@ static void perf_session__mmap_read(struct perf_session *self) } } -int nr_poll; int group_fd; -static void start_counter(int i, struct perf_evsel *evsel) +static void start_counter(int i, struct perf_evlist *evlist, + struct perf_evsel *evsel) { struct xyarray *mmap_array = evsel->priv; struct mmap_data *mm; @@ -1281,9 +1279,9 @@ try_again: 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++; + evlist->pollfd[evlist->nr_fds].fd = FD(evsel, i, thread_index); + evlist->pollfd[evlist->nr_fds].events = POLLIN; + evlist->nr_fds++; mm = xyarray__entry(mmap_array, i, thread_index); mm->prev = 0; @@ -1316,11 +1314,11 @@ static int __cmd_top(void) for (i = 0; i < cpus->nr; i++) { group_fd = -1; list_for_each_entry(counter, &evsel_list->entries, node) - start_counter(i, counter); + start_counter(i, evsel_list, counter); } /* Wait for a minimal set of events before starting the snapshot */ - poll(&event_array[0], nr_poll, 100); + poll(evsel_list->pollfd, evsel_list->nr_fds, 100); perf_session__mmap_read(session); @@ -1345,7 +1343,7 @@ static int __cmd_top(void) perf_session__mmap_read(session); if (hits == samples) - ret = poll(event_array, nr_poll, 100); + ret = poll(evsel_list->pollfd, evsel_list->nr_fds, 100); } return 0; @@ -1426,11 +1424,6 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) usage_with_options(top_usage, options); } - event_array = malloc((sizeof(struct pollfd) * - MAX_NR_CPUS * MAX_COUNTERS * threads->nr)); - if (!event_array) - return -ENOMEM; - /* CPU and PID are mutually exclusive */ if (target_tid > 0 && cpu_list) { printf("WARNING: PID switch overriding CPU\n"); @@ -1480,6 +1473,9 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) pos->attr.sample_period = default_interval; } + if (perf_evlist__alloc_pollfd(evsel_list, cpus->nr, threads->nr) < 0) + goto out_free_fd; + sym_evsel = list_entry(evsel_list->entries.next, struct perf_evsel, node); symbol_conf.priv_size = (sizeof(struct sym_entry) + diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 7b4faec23737..2abf949259d0 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -1,3 +1,4 @@ +#include #include "evlist.h" #include "evsel.h" #include "util.h" @@ -28,6 +29,7 @@ static void perf_evlist__purge(struct perf_evlist *evlist) void perf_evlist__delete(struct perf_evlist *evlist) { perf_evlist__purge(evlist); + free(evlist->pollfd); free(evlist); } @@ -51,3 +53,10 @@ int perf_evlist__add_default(struct perf_evlist *evlist) perf_evlist__add(evlist, evsel); return 0; } + +int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads) +{ + int nfds = ncpus * nthreads * evlist->nr_entries; + evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); + return evlist->pollfd != NULL ? 0 : -ENOMEM; +} diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h index 48db91a8abf3..a7d7e122e3c6 100644 --- a/tools/perf/util/evlist.h +++ b/tools/perf/util/evlist.h @@ -3,9 +3,13 @@ #include +struct pollfd; + struct perf_evlist { struct list_head entries; int nr_entries; + int nr_fds; + struct pollfd *pollfd; }; struct perf_evsel; @@ -16,4 +20,6 @@ void perf_evlist__delete(struct perf_evlist *evlist); void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry); int perf_evlist__add_default(struct perf_evlist *evlist); +int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads); + #endif /* __PERF_EVLIST_H */ From f08199d314458610d4ca52f8e86e0a4ec7a7bc54 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 11 Jan 2011 23:42:19 -0200 Subject: [PATCH 03/29] perf evsel: Support event groups The perf_evsel__open now have an extra boolean argument specifying if event grouping is desired. The first file descriptor created on a CPU becomes the group leader. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-stat.c | 4 ++-- tools/perf/builtin-test.c | 4 ++-- tools/perf/util/evsel.c | 27 +++++++++++++++++---------- tools/perf/util/evsel.h | 10 ++++++---- 4 files changed, 27 insertions(+), 18 deletions(-) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index da9090245934..b5fe522f11dc 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -169,7 +169,7 @@ static int create_perf_stat_counter(struct perf_evsel *evsel) PERF_FORMAT_TOTAL_TIME_RUNNING; if (system_wide) - return perf_evsel__open_per_cpu(evsel, cpus); + return perf_evsel__open_per_cpu(evsel, cpus, false); attr->inherit = !no_inherit; if (target_pid == -1 && target_tid == -1) { @@ -177,7 +177,7 @@ static int create_perf_stat_counter(struct perf_evsel *evsel) attr->enable_on_exec = 1; } - return perf_evsel__open_per_thread(evsel, threads); + return perf_evsel__open_per_thread(evsel, threads, false); } /* diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index 5dcdba653d70..4282d671b161 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -289,7 +289,7 @@ static int test__open_syscall_event(void) goto out_thread_map_delete; } - if (perf_evsel__open_per_thread(evsel, threads) < 0) { + if (perf_evsel__open_per_thread(evsel, threads, false) < 0) { pr_debug("failed to open counter: %s, " "tweak /proc/sys/kernel/perf_event_paranoid?\n", strerror(errno)); @@ -364,7 +364,7 @@ static int test__open_syscall_event_on_all_cpus(void) goto out_thread_map_delete; } - if (perf_evsel__open(evsel, cpus, threads) < 0) { + if (perf_evsel__open(evsel, cpus, threads, false) < 0) { pr_debug("failed to open counter: %s, " "tweak /proc/sys/kernel/perf_event_paranoid?\n", strerror(errno)); diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index f5cfed60af98..da473ec93c75 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -128,7 +128,7 @@ int __perf_evsel__read(struct perf_evsel *evsel, } static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, - struct thread_map *threads) + struct thread_map *threads, bool group) { int cpu, thread; @@ -137,12 +137,18 @@ static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, return -1; for (cpu = 0; cpu < cpus->nr; cpu++) { + int group_fd = -1; + for (thread = 0; thread < threads->nr; thread++) { FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr, threads->map[thread], - cpus->map[cpu], -1, 0); + cpus->map[cpu], + group_fd, 0); if (FD(evsel, cpu, thread) < 0) goto out_close; + + if (group && group_fd == -1) + group_fd = FD(evsel, cpu, thread); } } @@ -175,10 +181,9 @@ static struct { .threads = { -1, }, }; -int perf_evsel__open(struct perf_evsel *evsel, - struct cpu_map *cpus, struct thread_map *threads) +int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, + struct thread_map *threads, bool group) { - if (cpus == NULL) { /* Work around old compiler warnings about strict aliasing */ cpus = &empty_cpu_map.map; @@ -187,15 +192,17 @@ int perf_evsel__open(struct perf_evsel *evsel, if (threads == NULL) threads = &empty_thread_map.map; - return __perf_evsel__open(evsel, cpus, threads); + return __perf_evsel__open(evsel, cpus, threads, group); } -int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus) +int perf_evsel__open_per_cpu(struct perf_evsel *evsel, + struct cpu_map *cpus, bool group) { - return __perf_evsel__open(evsel, cpus, &empty_thread_map.map); + return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group); } -int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads) +int perf_evsel__open_per_thread(struct perf_evsel *evsel, + struct thread_map *threads, bool group) { - return __perf_evsel__open(evsel, &empty_cpu_map.map, threads); + return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group); } diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index b2d755fe88a5..0962b500cb6d 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -45,10 +45,12 @@ int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus); void perf_evsel__free_fd(struct perf_evsel *evsel); void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads); -int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus); -int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads); -int perf_evsel__open(struct perf_evsel *evsel, - struct cpu_map *cpus, struct thread_map *threads); +int perf_evsel__open_per_cpu(struct perf_evsel *evsel, + struct cpu_map *cpus, bool group); +int perf_evsel__open_per_thread(struct perf_evsel *evsel, + struct thread_map *threads, bool group); +int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, + struct thread_map *threads, bool group); #define perf_evsel__match(evsel, t, c) \ (evsel->attr.type == PERF_TYPE_##t && \ From 9d04f1781772e11bd58806391555fc23ebb54377 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 00:08:18 -0200 Subject: [PATCH 04/29] perf evsel: Allow specifying if the inherit bit should be set As this is a per-cpu attribute, we can't set it up in advance and use it for all the calls. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-stat.c | 4 ++-- tools/perf/builtin-test.c | 4 ++-- tools/perf/util/evsel.c | 16 +++++++++------- tools/perf/util/evsel.h | 6 +++--- 4 files changed, 16 insertions(+), 14 deletions(-) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index b5fe522f11dc..e2a2d02c5c43 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -169,7 +169,7 @@ static int create_perf_stat_counter(struct perf_evsel *evsel) PERF_FORMAT_TOTAL_TIME_RUNNING; if (system_wide) - return perf_evsel__open_per_cpu(evsel, cpus, false); + return perf_evsel__open_per_cpu(evsel, cpus, false, false); attr->inherit = !no_inherit; if (target_pid == -1 && target_tid == -1) { @@ -177,7 +177,7 @@ static int create_perf_stat_counter(struct perf_evsel *evsel) attr->enable_on_exec = 1; } - return perf_evsel__open_per_thread(evsel, threads, false); + return perf_evsel__open_per_thread(evsel, threads, false, false); } /* diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index 4282d671b161..7287158c4830 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -289,7 +289,7 @@ static int test__open_syscall_event(void) goto out_thread_map_delete; } - if (perf_evsel__open_per_thread(evsel, threads, false) < 0) { + if (perf_evsel__open_per_thread(evsel, threads, false, false) < 0) { pr_debug("failed to open counter: %s, " "tweak /proc/sys/kernel/perf_event_paranoid?\n", strerror(errno)); @@ -364,7 +364,7 @@ static int test__open_syscall_event_on_all_cpus(void) goto out_thread_map_delete; } - if (perf_evsel__open(evsel, cpus, threads, false) < 0) { + if (perf_evsel__open(evsel, cpus, threads, false, false) < 0) { pr_debug("failed to open counter: %s, " "tweak /proc/sys/kernel/perf_event_paranoid?\n", strerror(errno)); diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index da473ec93c75..82a00536892a 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -128,7 +128,7 @@ int __perf_evsel__read(struct perf_evsel *evsel, } static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, - struct thread_map *threads, bool group) + struct thread_map *threads, bool group, bool inherit) { int cpu, thread; @@ -139,6 +139,8 @@ static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, for (cpu = 0; cpu < cpus->nr; cpu++) { int group_fd = -1; + evsel->attr.inherit = (cpus->map[cpu] < 0) && inherit; + for (thread = 0; thread < threads->nr; thread++) { FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr, threads->map[thread], @@ -182,7 +184,7 @@ static struct { }; int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, - struct thread_map *threads, bool group) + struct thread_map *threads, bool group, bool inherit) { if (cpus == NULL) { /* Work around old compiler warnings about strict aliasing */ @@ -192,17 +194,17 @@ int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, if (threads == NULL) threads = &empty_thread_map.map; - return __perf_evsel__open(evsel, cpus, threads, group); + return __perf_evsel__open(evsel, cpus, threads, group, inherit); } int perf_evsel__open_per_cpu(struct perf_evsel *evsel, - struct cpu_map *cpus, bool group) + struct cpu_map *cpus, bool group, bool inherit) { - return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group); + return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group, inherit); } int perf_evsel__open_per_thread(struct perf_evsel *evsel, - struct thread_map *threads, bool group) + struct thread_map *threads, bool group, bool inherit) { - return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group); + return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group, inherit); } diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 0962b500cb6d..1594696bd127 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -46,11 +46,11 @@ void perf_evsel__free_fd(struct perf_evsel *evsel); void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads); int perf_evsel__open_per_cpu(struct perf_evsel *evsel, - struct cpu_map *cpus, bool group); + struct cpu_map *cpus, bool group, bool inherit); int perf_evsel__open_per_thread(struct perf_evsel *evsel, - struct thread_map *threads, bool group); + struct thread_map *threads, bool group, bool inherit); int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, - struct thread_map *threads, bool group); + struct thread_map *threads, bool group, bool inherit); #define perf_evsel__match(evsel, t, c) \ (evsel->attr.type == PERF_TYPE_##t && \ From 72cb7013e08dec29631e0447f9496b7bacd3e14b Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 10:52:47 -0200 Subject: [PATCH 05/29] perf top: Use perf_evsel__open Now that it handles group_fd and inherit we can use it, sharing it with stat. Next step: 'perf record' should use, then move the mmap_array out of ->priv and into perf_evsel, with top and record sharing this, and at the same time, write a 'perf test' stress test. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-top.c | 122 ++++++++++++++++++--------------------- 1 file changed, 57 insertions(+), 65 deletions(-) diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 1bc465215fc6..15d89bede2fb 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -1210,75 +1210,17 @@ static void perf_session__mmap_read(struct perf_session *self) } } -int group_fd; - static void start_counter(int i, struct perf_evlist *evlist, struct perf_evsel *evsel) { 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]; - - attr = &evsel->attr; - - attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID; - - if (freq) { - attr->sample_type |= PERF_SAMPLE_PERIOD; - attr->freq = 1; - attr->sample_freq = freq; - } - - attr->inherit = (cpu < 0) && inherit; - attr->mmap = 1; - for (thread_index = 0; thread_index < threads->nr; thread_index++) { -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) { - int err = errno; - - if (err == EPERM || err == EACCES) - die("Permission error - are you root?\n" - "\t Consider tweaking" - " /proc/sys/kernel/perf_event_paranoid.\n"); - /* - * If it's cycles then fall back to hrtimer - * 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 (verbose) - warning(" ... trying to fall back to cpu-clock-ticks\n"); - - attr->type = PERF_TYPE_SOFTWARE; - attr->config = PERF_COUNT_SW_CPU_CLOCK; - 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)); - 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); - evlist->pollfd[evlist->nr_fds].fd = FD(evsel, i, thread_index); evlist->pollfd[evlist->nr_fds].events = POLLIN; evlist->nr_fds++; @@ -1293,11 +1235,65 @@ try_again: } } +static void start_counters(struct perf_evlist *evlist) +{ + struct perf_evsel *counter; + int i; + + list_for_each_entry(counter, &evlist->entries, node) { + struct perf_event_attr *attr = &counter->attr; + + attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID; + + if (freq) { + attr->sample_type |= PERF_SAMPLE_PERIOD; + attr->freq = 1; + attr->sample_freq = freq; + } + + attr->mmap = 1; +try_again: + if (perf_evsel__open(counter, cpus, threads, group, inherit) < 0) { + int err = errno; + + if (err == EPERM || err == EACCES) + die("Permission error - are you root?\n" + "\t Consider tweaking" + " /proc/sys/kernel/perf_event_paranoid.\n"); + /* + * If it's cycles then fall back to hrtimer + * 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 (verbose) + warning(" ... trying to fall back to cpu-clock-ticks\n"); + + attr->type = PERF_TYPE_SOFTWARE; + attr->config = PERF_COUNT_SW_CPU_CLOCK; + goto try_again; + } + printf("\n"); + 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); + } + } + + for (i = 0; i < cpus->nr; i++) { + list_for_each_entry(counter, &evlist->entries, node) + start_counter(i, evsel_list, counter); + } +} + static int __cmd_top(void) { pthread_t thread; - struct perf_evsel *counter; - int i, ret; + int ret; /* * 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. @@ -1311,11 +1307,7 @@ static int __cmd_top(void) else event__synthesize_threads(event__process, session); - for (i = 0; i < cpus->nr; i++) { - group_fd = -1; - list_for_each_entry(counter, &evsel_list->entries, node) - start_counter(i, evsel_list, counter); - } + start_counters(evsel_list); /* Wait for a minimal set of events before starting the snapshot */ poll(evsel_list->pollfd, evsel_list->nr_fds, 100); From dd7927f4f8ee75b032ff15aeef4bda49719a443a Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 14:28:51 -0200 Subject: [PATCH 06/29] perf record: Use perf_evsel__open Now its time to factor out the mmap handling bits into the perf_evsel class. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 233 +++++++++++++++++------------------- 1 file changed, 113 insertions(+), 120 deletions(-) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 1614d89b4765..ec43f2eb7b72 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -72,8 +72,6 @@ static struct perf_evlist *evsel_list; static long samples = 0; static u64 bytes_written = 0; -static int nr_cpu = 0; - static int file_new = 1; static off_t post_processing_offset; @@ -208,8 +206,6 @@ static void sig_atexit(void) kill(getpid(), signr); } -static int group_fd; - static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int nr) { struct perf_header_attr *h_attr; @@ -234,7 +230,6 @@ static void create_counter(struct perf_evlist *evlist, char *filter = evsel->filter; struct perf_event_attr *attr = &evsel->attr; struct perf_header_attr *h_attr; - int track = !evsel->idx; /* only the first counter needs these */ int thread_index; int ret; struct { @@ -243,19 +238,77 @@ static void create_counter(struct perf_evlist *evlist, u64 time_running; u64 id; } read_data; - /* - * Check if parse_single_tracepoint_event has already asked for - * PERF_SAMPLE_TIME. - * - * XXX this is kludgy but short term fix for problems introduced by - * eac23d1c that broke 'perf script' by having different sample_types - * when using multiple tracepoint events when we use a perf binary - * that tries to use sample_id_all on an older kernel. - * - * We need to move counter creation to perf_session, support - * different sample_types, etc. - */ - bool time_needed = attr->sample_type & PERF_SAMPLE_TIME; + + for (thread_index = 0; thread_index < threads->nr; thread_index++) { + h_attr = get_header_attr(attr, evsel->idx); + if (h_attr == NULL) + die("nomem\n"); + + if (!file_new) { + if (memcmp(&h_attr->attr, attr, sizeof(*attr))) { + fprintf(stderr, "incompatible append\n"); + exit(-1); + } + } + + if (read(FD(evsel, cpu, thread_index), &read_data, sizeof(read_data)) == -1) { + perror("Unable to read perf file descriptor"); + exit(-1); + } + + if (perf_header_attr__add_id(h_attr, read_data.id) < 0) { + pr_warning("Not enough memory to add id\n"); + exit(-1); + } + + assert(FD(evsel, cpu, thread_index) >= 0); + fcntl(FD(evsel, cpu, thread_index), F_SETFL, O_NONBLOCK); + + if (evsel->idx || thread_index) { + struct perf_evsel *first; + first = list_entry(evlist->entries.next, struct perf_evsel, node); + ret = ioctl(FD(evsel, cpu, thread_index), + PERF_EVENT_IOC_SET_OUTPUT, + FD(first, cpu, 0)); + if (ret) { + error("failed to set output: %d (%s)\n", errno, + strerror(errno)); + exit(-1); + } + } else { + mmap_array[cpu].prev = 0; + mmap_array[cpu].mask = mmap_pages*page_size - 1; + mmap_array[cpu].base = mmap(NULL, (mmap_pages+1)*page_size, + PROT_READ | PROT_WRITE, MAP_SHARED, FD(evsel, cpu, thread_index), 0); + if (mmap_array[cpu].base == MAP_FAILED) { + error("failed to mmap with %d (%s)\n", errno, strerror(errno)); + exit(-1); + } + + evlist->pollfd[evlist->nr_fds].fd = FD(evsel, cpu, thread_index); + evlist->pollfd[evlist->nr_fds].events = POLLIN; + evlist->nr_fds++; + } + + if (filter != NULL) { + ret = ioctl(FD(evsel, cpu, thread_index), + PERF_EVENT_IOC_SET_FILTER, filter); + if (ret) { + error("failed to set filter with %d (%s)\n", errno, + strerror(errno)); + exit(-1); + } + } + } + + if (!sample_type) + sample_type = attr->sample_type; +} + +static void config_attr(struct perf_evsel *evsel, struct perf_evlist *evlist) +{ + struct perf_event_attr *attr = &evsel->attr; + int track = !evsel->idx; /* only the first counter needs these */ attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING | @@ -315,19 +368,39 @@ static void create_counter(struct perf_evlist *evlist, attr->mmap = track; attr->comm = track; - attr->inherit = !no_inherit; + if (target_pid == -1 && target_tid == -1 && !system_wide) { attr->disabled = 1; attr->enable_on_exec = 1; } +} + +static void open_counters(struct perf_evlist *evlist) +{ + struct perf_evsel *pos; + int cpu; + + list_for_each_entry(pos, &evlist->entries, node) { + struct perf_event_attr *attr = &pos->attr; + /* + * Check if parse_single_tracepoint_event has already asked for + * PERF_SAMPLE_TIME. + * + * XXX this is kludgy but short term fix for problems introduced by + * eac23d1c that broke 'perf script' by having different sample_types + * when using multiple tracepoint events when we use a perf binary + * that tries to use sample_id_all on an older kernel. + * + * We need to move counter creation to perf_session, support + * different sample_types, etc. + */ + bool time_needed = attr->sample_type & PERF_SAMPLE_TIME; + + config_attr(pos, evlist); retry_sample_id: - attr->sample_id_all = sample_id_all_avail ? 1 : 0; - - for (thread_index = 0; thread_index < threads->nr; thread_index++) { + attr->sample_id_all = sample_id_all_avail ? 1 : 0; try_again: - FD(evsel, nr_cpu, thread_index) = sys_perf_event_open(attr, threads->map[thread_index], cpu, group_fd, 0); - - if (FD(evsel, nr_cpu, thread_index) < 0) { + if (perf_evsel__open(pos, cpus, threads, group, !no_inherit) < 0) { int err = errno; if (err == EPERM || err == EACCES) @@ -364,7 +437,7 @@ try_again: } printf("\n"); error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n", - FD(evsel, nr_cpu, thread_index), strerror(err)); + err, strerror(err)); #if defined(__i386__) || defined(__x86_64__) if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP) @@ -375,90 +448,13 @@ try_again: #endif die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); - exit(-1); - } - - h_attr = get_header_attr(attr, evsel->idx); - if (h_attr == NULL) - die("nomem\n"); - - if (!file_new) { - if (memcmp(&h_attr->attr, attr, sizeof(*attr))) { - fprintf(stderr, "incompatible append\n"); - exit(-1); - } - } - - if (read(FD(evsel, nr_cpu, thread_index), &read_data, sizeof(read_data)) == -1) { - perror("Unable to read perf file descriptor"); - exit(-1); - } - - if (perf_header_attr__add_id(h_attr, read_data.id) < 0) { - pr_warning("Not enough memory to add id\n"); - exit(-1); - } - - assert(FD(evsel, nr_cpu, thread_index) >= 0); - fcntl(FD(evsel, nr_cpu, thread_index), F_SETFL, O_NONBLOCK); - - /* - * First counter acts as the group leader: - */ - if (group && group_fd == -1) - group_fd = FD(evsel, nr_cpu, thread_index); - - if (evsel->idx || thread_index) { - struct perf_evsel *first; - first = list_entry(evlist->entries.next, struct perf_evsel, node); - ret = ioctl(FD(evsel, nr_cpu, thread_index), - PERF_EVENT_IOC_SET_OUTPUT, - FD(first, nr_cpu, 0)); - if (ret) { - error("failed to set output: %d (%s)\n", errno, - strerror(errno)); - exit(-1); - } - } else { - mmap_array[nr_cpu].prev = 0; - mmap_array[nr_cpu].mask = mmap_pages*page_size - 1; - mmap_array[nr_cpu].base = mmap(NULL, (mmap_pages+1)*page_size, - PROT_READ | PROT_WRITE, MAP_SHARED, FD(evsel, nr_cpu, thread_index), 0); - if (mmap_array[nr_cpu].base == MAP_FAILED) { - error("failed to mmap with %d (%s)\n", errno, strerror(errno)); - exit(-1); - } - - evlist->pollfd[evlist->nr_fds].fd = FD(evsel, nr_cpu, thread_index); - evlist->pollfd[evlist->nr_fds].events = POLLIN; - evlist->nr_fds++; - } - - if (filter != NULL) { - ret = ioctl(FD(evsel, nr_cpu, thread_index), - PERF_EVENT_IOC_SET_FILTER, filter); - if (ret) { - error("failed to set filter with %d (%s)\n", errno, - strerror(errno)); - exit(-1); - } } } - if (!sample_type) - sample_type = attr->sample_type; -} - -static void open_counters(struct perf_evlist *evlist, int cpu) -{ - struct perf_evsel *pos; - - group_fd = -1; - - list_for_each_entry(pos, &evlist->entries, node) - create_counter(evlist, pos, cpu); - - nr_cpu++; + for (cpu = 0; cpu < cpus->nr; ++cpu) { + list_for_each_entry(pos, &evlist->entries, node) + create_counter(evlist, pos, cpu); + } } static int process_buildids(void) @@ -533,7 +529,7 @@ static void mmap_read_all(void) { int i; - for (i = 0; i < nr_cpu; i++) { + for (i = 0; i < cpus->nr; i++) { if (mmap_array[i].base) mmap_read(&mmap_array[i]); } @@ -673,12 +669,7 @@ static int __cmd_record(int argc, const char **argv) close(child_ready_pipe[0]); } - if (!system_wide && no_inherit && !cpu_list) { - open_counters(evsel_list, -1); - } else { - for (i = 0; i < cpus->nr; i++) - open_counters(evsel_list, cpus->map[i]); - } + open_counters(evsel_list); perf_session__set_sample_type(session, sample_type); @@ -795,7 +786,7 @@ static int __cmd_record(int argc, const char **argv) } if (done) { - for (i = 0; i < nr_cpu; i++) { + for (i = 0; i < cpus->nr; i++) { struct perf_evsel *pos; list_for_each_entry(pos, &evsel_list->entries, node) { @@ -933,11 +924,13 @@ int cmd_record(int argc, const char **argv, const char *prefix __used) usage_with_options(record_usage, record_options); } - cpus = cpu_map__new(cpu_list); - if (cpus == NULL) { - perror("failed to parse CPUs map"); - return -1; - } + if (target_tid != -1) + cpus = cpu_map__dummy_new(); + else + cpus = cpu_map__new(cpu_list); + + if (cpus == NULL) + usage_with_options(record_usage, record_options); list_for_each_entry(pos, &evsel_list->entries, node) { if (perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) From 70082dd92c4b288bd723a77897e2b555f0e63113 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 17:03:24 -0200 Subject: [PATCH 07/29] perf evsel: Introduce mmap support Out of the code in 'perf top'. Record is next in line. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-top.c | 71 +++------------------------------------- tools/perf/perf.h | 14 ++++++++ tools/perf/util/evlist.c | 8 +++++ tools/perf/util/evlist.h | 1 + tools/perf/util/evsel.c | 71 ++++++++++++++++++++++++++++++++++++++++ tools/perf/util/evsel.h | 8 +++++ 6 files changed, 107 insertions(+), 66 deletions(-) diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 15d89bede2fb..7d723ad0bfa9 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -1095,43 +1095,12 @@ static void event__process_sample(const event_t *self, } } -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) { - struct xyarray *mmap_array = evsel->priv; - struct mmap_data *md = xyarray__entry(mmap_array, cpu, thread_idx); - unsigned int head = mmap_read_head(md); + struct perf_mmap *md = xyarray__entry(evsel->mmap, cpu, thread_idx); + unsigned int head = perf_mmap__read_head(md); unsigned int old = md->prev; unsigned char *data = md->base + page_size; struct sample_data sample; @@ -1210,35 +1179,9 @@ static void perf_session__mmap_read(struct perf_session *self) } } -static void start_counter(int i, struct perf_evlist *evlist, - struct perf_evsel *evsel) -{ - struct xyarray *mmap_array = evsel->priv; - struct mmap_data *mm; - int thread_index; - - for (thread_index = 0; thread_index < threads->nr; thread_index++) { - assert(FD(evsel, i, thread_index) >= 0); - fcntl(FD(evsel, i, thread_index), F_SETFL, O_NONBLOCK); - - evlist->pollfd[evlist->nr_fds].fd = FD(evsel, i, thread_index); - evlist->pollfd[evlist->nr_fds].events = POLLIN; - evlist->nr_fds++; - - 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)); - } -} - static void start_counters(struct perf_evlist *evlist) { struct perf_evsel *counter; - int i; list_for_each_entry(counter, &evlist->entries, node) { struct perf_event_attr *attr = &counter->attr; @@ -1282,11 +1225,9 @@ try_again: die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); exit(-1); } - } - for (i = 0; i < cpus->nr; i++) { - list_for_each_entry(counter, &evlist->entries, node) - start_counter(i, evsel_list, counter); + if (perf_evsel__mmap(counter, cpus, threads, mmap_pages, evlist) < 0) + die("failed to mmap with %d (%s)\n", errno, strerror(errno)); } } @@ -1453,7 +1394,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) usage_with_options(top_usage, options); list_for_each_entry(pos, &evsel_list->entries, node) { - if (perf_evsel__alloc_mmap_per_thread(pos, cpus->nr, threads->nr) < 0 || + if (perf_evsel__alloc_mmap(pos, cpus->nr, threads->nr) < 0 || perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) goto out_free_fd; /* @@ -1485,8 +1426,6 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) status = __cmd_top(); out_free_fd: - list_for_each_entry(pos, &evsel_list->entries, node) - perf_evsel__free_mmap(pos); perf_evlist__delete(evsel_list); return status; diff --git a/tools/perf/perf.h b/tools/perf/perf.h index 95aaf565c704..5fb5e1f11d1c 100644 --- a/tools/perf/perf.h +++ b/tools/perf/perf.h @@ -94,6 +94,20 @@ void get_term_dimensions(struct winsize *ws); #include "util/types.h" #include +struct perf_mmap { + void *base; + int mask; + unsigned int prev; +}; + +static inline unsigned int perf_mmap__read_head(struct perf_mmap *mm) +{ + struct perf_event_mmap_page *pc = mm->base; + int head = pc->data_head; + rmb(); + return head; +} + /* * prctl(PR_TASK_PERF_EVENTS_DISABLE) will (cheaply) disable all * counters in the current task. diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 2abf949259d0..6d4129214ee8 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -60,3 +60,11 @@ int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthread evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); return evlist->pollfd != NULL ? 0 : -ENOMEM; } + +void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) +{ + fcntl(fd, F_SETFL, O_NONBLOCK); + evlist->pollfd[evlist->nr_fds].fd = fd; + evlist->pollfd[evlist->nr_fds].events = POLLIN; + evlist->nr_fds++; +} diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h index a7d7e122e3c6..16bbfcba8ca8 100644 --- a/tools/perf/util/evlist.h +++ b/tools/perf/util/evlist.h @@ -21,5 +21,6 @@ void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry); int perf_evlist__add_default(struct perf_evlist *evlist); int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads); +void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd); #endif /* __PERF_EVLIST_H */ diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 82a00536892a..f5006958f8da 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -1,9 +1,13 @@ #include "evsel.h" +#include "evlist.h" #include "../perf.h" #include "util.h" #include "cpumap.h" #include "thread.h" +#include +#include + #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx) @@ -49,10 +53,32 @@ void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads) } } +void perf_evsel__munmap(struct perf_evsel *evsel, int ncpus, int nthreads) +{ + struct perf_mmap *mm; + int cpu, thread; + + for (cpu = 0; cpu < ncpus; cpu++) + for (thread = 0; thread < nthreads; ++thread) { + mm = xyarray__entry(evsel->mmap, cpu, thread); + if (mm->base != NULL) { + munmap(mm->base, evsel->mmap_len); + mm->base = NULL; + } + } +} + +int perf_evsel__alloc_mmap(struct perf_evsel *evsel, int ncpus, int nthreads) +{ + evsel->mmap = xyarray__new(ncpus, nthreads, sizeof(struct perf_mmap)); + return evsel->mmap != NULL ? 0 : -ENOMEM; +} + void perf_evsel__delete(struct perf_evsel *evsel) { assert(list_empty(&evsel->node)); xyarray__delete(evsel->fd); + xyarray__delete(evsel->mmap); free(evsel); } @@ -208,3 +234,48 @@ int perf_evsel__open_per_thread(struct perf_evsel *evsel, { return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group, inherit); } + +int perf_evsel__mmap(struct perf_evsel *evsel, struct cpu_map *cpus, + struct thread_map *threads, int pages, + struct perf_evlist *evlist) +{ + unsigned int page_size = sysconf(_SC_PAGE_SIZE); + int mask = pages * page_size - 1, cpu; + struct perf_mmap *mm; + int thread; + + if (evsel->mmap == NULL && + perf_evsel__alloc_mmap(evsel, cpus->nr, threads->nr) < 0) + return -ENOMEM; + + evsel->mmap_len = (pages + 1) * page_size; + + for (cpu = 0; cpu < cpus->nr; cpu++) { + for (thread = 0; thread < threads->nr; thread++) { + mm = xyarray__entry(evsel->mmap, cpu, thread); + mm->prev = 0; + mm->mask = mask; + mm->base = mmap(NULL, evsel->mmap_len, PROT_READ, + MAP_SHARED, FD(evsel, cpu, thread), 0); + if (mm->base == MAP_FAILED) + goto out_unmap; + + if (evlist != NULL) + perf_evlist__add_pollfd(evlist, FD(evsel, cpu, thread)); + } + } + + return 0; + +out_unmap: + do { + while (--thread >= 0) { + mm = xyarray__entry(evsel->mmap, cpu, thread); + munmap(mm->base, evsel->mmap_len); + mm->base = NULL; + } + thread = threads->nr; + } while (--cpu >= 0); + + return -1; +} diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 1594696bd127..c8fbef299436 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -29,19 +29,23 @@ struct perf_evsel { struct perf_event_attr attr; char *filter; struct xyarray *fd; + struct xyarray *mmap; struct perf_counts *counts; + size_t mmap_len; int idx; void *priv; }; struct cpu_map; struct thread_map; +struct perf_evlist; struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx); void perf_evsel__delete(struct perf_evsel *evsel); int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads); int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus); +int perf_evsel__alloc_mmap(struct perf_evsel *evsel, int ncpus, int nthreads); void perf_evsel__free_fd(struct perf_evsel *evsel); void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads); @@ -51,6 +55,10 @@ int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads, bool group, bool inherit); int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, struct thread_map *threads, bool group, bool inherit); +int perf_evsel__mmap(struct perf_evsel *evsel, struct cpu_map *cpus, + struct thread_map *threads, int pages, + struct perf_evlist *evlist); +void perf_evsel__munmap(struct perf_evsel *evsel, int ncpus, int nthreads); #define perf_evsel__match(evsel, t, c) \ (evsel->attr.type == PERF_TYPE_##t && \ From 744bd8aa3c8b43447f689a27872fa95e700b8a4f Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 17:07:28 -0200 Subject: [PATCH 08/29] perf record: Use struct perf_mmap and helpers Paving the way to using perf_evsel->mmap, do this to reduce the patch noise in the next ones. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 25 ++++--------------------- 1 file changed, 4 insertions(+), 21 deletions(-) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index ec43f2eb7b72..d89e2f106a62 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -78,26 +78,9 @@ static off_t post_processing_offset; static struct perf_session *session; static const char *cpu_list; -struct mmap_data { - void *base; - unsigned int mask; - unsigned int prev; -}; +static struct perf_mmap mmap_array[MAX_NR_CPUS]; -static struct mmap_data mmap_array[MAX_NR_CPUS]; - -static unsigned long mmap_read_head(struct mmap_data *md) -{ - struct perf_event_mmap_page *pc = md->base; - long head; - - head = pc->data_head; - rmb(); - - return head; -} - -static void mmap_write_tail(struct mmap_data *md, unsigned long tail) +static void mmap_write_tail(struct perf_mmap *md, unsigned long tail) { struct perf_event_mmap_page *pc = md->base; @@ -136,9 +119,9 @@ static int process_synthesized_event(event_t *event, return 0; } -static void mmap_read(struct mmap_data *md) +static void mmap_read(struct perf_mmap *md) { - unsigned int head = mmap_read_head(md); + unsigned int head = perf_mmap__read_head(md); unsigned int old = md->prev; unsigned char *data = md->base + page_size; unsigned long size; From 115d2d8963a426670ac3ce983fc4c4e001703943 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 17:11:53 -0200 Subject: [PATCH 09/29] perf record: Move perf_mmap__write_tail to perf.h Close to perf_mmap__read_head() and the perf_mmap struct definition. This is useful for any recorder, and we will need it in 'perf test'. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 13 +------------ tools/perf/perf.h | 12 ++++++++++++ 2 files changed, 13 insertions(+), 12 deletions(-) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index d89e2f106a62..109f3b269ac5 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -80,17 +80,6 @@ static const char *cpu_list; static struct perf_mmap mmap_array[MAX_NR_CPUS]; -static void mmap_write_tail(struct perf_mmap *md, unsigned long tail) -{ - struct perf_event_mmap_page *pc = md->base; - - /* - * ensure all reads are done before we write the tail out. - */ - /* mb(); */ - pc->data_tail = tail; -} - static void advance_output(size_t size) { bytes_written += size; @@ -165,7 +154,7 @@ static void mmap_read(struct perf_mmap *md) write_output(buf, size); md->prev = old; - mmap_write_tail(md, old); + perf_mmap__write_tail(md, old); } static volatile int done = 0; diff --git a/tools/perf/perf.h b/tools/perf/perf.h index 5fb5e1f11d1c..a5fc660c1f12 100644 --- a/tools/perf/perf.h +++ b/tools/perf/perf.h @@ -108,6 +108,18 @@ static inline unsigned int perf_mmap__read_head(struct perf_mmap *mm) return head; } +static inline void perf_mmap__write_tail(struct perf_mmap *md, + unsigned long tail) +{ + struct perf_event_mmap_page *pc = md->base; + + /* + * ensure all reads are done before we write the tail out. + */ + /* mb(); */ + pc->data_tail = tail; +} + /* * prctl(PR_TASK_PERF_EVENTS_DISABLE) will (cheaply) disable all * counters in the current task. From 70db7533caef02350ec8d6852e589491bca3a951 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 12 Jan 2011 22:39:13 -0200 Subject: [PATCH 10/29] perf evlist: Move the mmap array from perf_evsel Adopting the new model used in 'perf record', where we don't have a map per thread per cpu, instead we have an mmap per cpu, established on the first fd for that cpu and ask the kernel using the PERF_EVENT_IOC_SET_OUTPUT ioctl to send events for the other fds on that cpu for the one with the mmap. The methods moved from perf_evsel to perf_evlist, but for easing review they were modified in place, in evsel.c, the next patch will move the migrated methods to evlist.c. With this 'perf top' now uses the same mmap model used by 'perf record' and the next patches will make 'perf record' use these new routines, establishing a common codebase for both tools. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-top.c | 54 ++++++------- tools/perf/util/evlist.c | 27 +++++++ tools/perf/util/evlist.h | 9 +++ tools/perf/util/evsel.c | 158 +++++++++++++++++++++++++++++---------- tools/perf/util/evsel.h | 26 +++++-- 5 files changed, 199 insertions(+), 75 deletions(-) diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 7d723ad0bfa9..df85c1f9417b 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -78,7 +78,7 @@ 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 unsigned int mmap_pages = 128; static int freq = 1000; /* 1 KHz */ static int delay_secs = 2; @@ -991,8 +991,7 @@ static int symbol_filter(struct map *map, struct symbol *sym) static void event__process_sample(const event_t *self, struct sample_data *sample, - struct perf_session *session, - struct perf_evsel *evsel) + struct perf_session *session) { u64 ip = self->ip.ip; struct sym_entry *syme; @@ -1085,8 +1084,12 @@ 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(evsel_list, 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) @@ -1095,11 +1098,9 @@ static void event__process_sample(const event_t *self, } } -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 perf_mmap *md = xyarray__entry(evsel->mmap, cpu, thread_idx); + struct perf_mmap *md = &evsel_list->mmap[cpu]; unsigned int head = perf_mmap__read_head(md); unsigned int old = md->prev; unsigned char *data = md->base + page_size; @@ -1153,7 +1154,7 @@ static void perf_session__mmap_read_counter(struct perf_session *self, event__parse_sample(event, self, &sample); if (event->header.type == PERF_RECORD_SAMPLE) - event__process_sample(event, &sample, self, evsel); + event__process_sample(event, &sample, self); else event__process(event, &sample, self); old += size; @@ -1164,19 +1165,10 @@ static void perf_session__mmap_read_counter(struct perf_session *self, static void perf_session__mmap_read(struct perf_session *self) { - struct perf_evsel *counter; - int i, thread_index; + int i; - for (i = 0; i < cpus->nr; i++) { - list_for_each_entry(counter, &evsel_list->entries, node) { - for (thread_index = 0; - thread_index < threads->nr; - thread_index++) { - perf_session__mmap_read_counter(self, - counter, i, thread_index); - } - } - } + for (i = 0; i < cpus->nr; i++) + perf_session__mmap_read_cpu(self, i); } static void start_counters(struct perf_evlist *evlist) @@ -1194,6 +1186,11 @@ static void start_counters(struct perf_evlist *evlist) attr->sample_freq = freq; } + if (evlist->nr_entries > 1) { + attr->sample_type |= PERF_SAMPLE_ID; + attr->read_format |= PERF_FORMAT_ID; + } + attr->mmap = 1; try_again: if (perf_evsel__open(counter, cpus, threads, group, inherit) < 0) { @@ -1225,15 +1222,16 @@ try_again: die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); exit(-1); } - - if (perf_evsel__mmap(counter, cpus, threads, mmap_pages, evlist) < 0) - die("failed to mmap with %d (%s)\n", errno, strerror(errno)); } + + if (perf_evlist__mmap(evlist, cpus, threads, mmap_pages, true) < 0) + die("failed to mmap with %d (%s)\n", errno, strerror(errno)); } static int __cmd_top(void) { pthread_t thread; + struct perf_evsel *first; int ret; /* * FIXME: perf_session__new should allow passing a O_MMAP, so that all this @@ -1249,6 +1247,8 @@ static int __cmd_top(void) event__synthesize_threads(event__process, session); start_counters(evsel_list); + first = list_entry(evsel_list->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(evsel_list->pollfd, evsel_list->nr_fds, 100); @@ -1394,8 +1394,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) usage_with_options(top_usage, options); list_for_each_entry(pos, &evsel_list->entries, node) { - if (perf_evsel__alloc_mmap(pos, cpus->nr, threads->nr) < 0 || - perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) + if (perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) goto out_free_fd; /* * Fill in the ones not specifically initialized via -c: @@ -1406,7 +1405,8 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) pos->attr.sample_period = default_interval; } - if (perf_evlist__alloc_pollfd(evsel_list, cpus->nr, threads->nr) < 0) + if (perf_evlist__alloc_pollfd(evsel_list, cpus->nr, threads->nr) < 0 || + perf_evlist__alloc_mmap(evsel_list, cpus->nr) < 0) goto out_free_fd; sym_evsel = list_entry(evsel_list->entries.next, struct perf_evsel, node); diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 6d4129214ee8..deb82a4fc312 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -3,11 +3,18 @@ #include "evsel.h" #include "util.h" +#include +#include + struct perf_evlist *perf_evlist__new(void) { struct perf_evlist *evlist = zalloc(sizeof(*evlist)); if (evlist != NULL) { + int i; + + for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) + INIT_HLIST_HEAD(&evlist->heads[i]); INIT_LIST_HEAD(&evlist->entries); } @@ -29,6 +36,7 @@ static void perf_evlist__purge(struct perf_evlist *evlist) void perf_evlist__delete(struct perf_evlist *evlist) { perf_evlist__purge(evlist); + free(evlist->mmap); free(evlist->pollfd); free(evlist); } @@ -68,3 +76,22 @@ void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) evlist->pollfd[evlist->nr_fds].events = POLLIN; evlist->nr_fds++; } + +struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) +{ + struct hlist_head *head; + struct hlist_node *pos; + struct perf_sample_id *sid; + int hash; + + if (evlist->nr_entries == 1) + return list_entry(evlist->entries.next, struct perf_evsel, node); + + hash = hash_64(id, PERF_EVLIST__HLIST_BITS); + head = &evlist->heads[hash]; + + hlist_for_each_entry(sid, pos, head, node) + if (sid->id == id) + return sid->evsel; + return NULL; +} diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h index 16bbfcba8ca8..dbfcc79bb995 100644 --- a/tools/perf/util/evlist.h +++ b/tools/perf/util/evlist.h @@ -2,13 +2,20 @@ #define __PERF_EVLIST_H 1 #include +#include "../perf.h" struct pollfd; +#define PERF_EVLIST__HLIST_BITS 8 +#define PERF_EVLIST__HLIST_SIZE (1 << PERF_EVLIST__HLIST_BITS) + struct perf_evlist { struct list_head entries; + struct hlist_head heads[PERF_EVLIST__HLIST_SIZE]; int nr_entries; int nr_fds; + int mmap_len; + struct perf_mmap *mmap; struct pollfd *pollfd; }; @@ -23,4 +30,6 @@ int perf_evlist__add_default(struct perf_evlist *evlist); int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads); void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd); +struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id); + #endif /* __PERF_EVLIST_H */ diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index f5006958f8da..ee490356c817 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -8,7 +8,11 @@ #include #include +#include +#include + #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) +#define SID(e, x, y) xyarray__entry(e->id, x, y) struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx) { @@ -29,6 +33,12 @@ int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads) return evsel->fd != NULL ? 0 : -ENOMEM; } +int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads) +{ + evsel->id = xyarray__new(ncpus, nthreads, sizeof(struct perf_sample_id)); + return evsel->id != NULL ? 0 : -ENOMEM; +} + int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus) { evsel->counts = zalloc((sizeof(*evsel->counts) + @@ -42,6 +52,12 @@ void perf_evsel__free_fd(struct perf_evsel *evsel) evsel->fd = NULL; } +void perf_evsel__free_id(struct perf_evsel *evsel) +{ + xyarray__delete(evsel->id); + evsel->id = NULL; +} + void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads) { int cpu, thread; @@ -53,32 +69,29 @@ void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads) } } -void perf_evsel__munmap(struct perf_evsel *evsel, int ncpus, int nthreads) +void perf_evlist__munmap(struct perf_evlist *evlist, int ncpus) { - struct perf_mmap *mm; - int cpu, thread; + int cpu; - for (cpu = 0; cpu < ncpus; cpu++) - for (thread = 0; thread < nthreads; ++thread) { - mm = xyarray__entry(evsel->mmap, cpu, thread); - if (mm->base != NULL) { - munmap(mm->base, evsel->mmap_len); - mm->base = NULL; - } + for (cpu = 0; cpu < ncpus; cpu++) { + if (evlist->mmap[cpu].base != NULL) { + munmap(evlist->mmap[cpu].base, evlist->mmap_len); + evlist->mmap[cpu].base = NULL; } + } } -int perf_evsel__alloc_mmap(struct perf_evsel *evsel, int ncpus, int nthreads) +int perf_evlist__alloc_mmap(struct perf_evlist *evlist, int ncpus) { - evsel->mmap = xyarray__new(ncpus, nthreads, sizeof(struct perf_mmap)); - return evsel->mmap != NULL ? 0 : -ENOMEM; + evlist->mmap = zalloc(ncpus * sizeof(struct perf_mmap)); + return evlist->mmap != NULL ? 0 : -ENOMEM; } void perf_evsel__delete(struct perf_evsel *evsel) { assert(list_empty(&evsel->node)); xyarray__delete(evsel->fd); - xyarray__delete(evsel->mmap); + xyarray__delete(evsel->id); free(evsel); } @@ -235,47 +248,110 @@ int perf_evsel__open_per_thread(struct perf_evsel *evsel, return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group, inherit); } -int perf_evsel__mmap(struct perf_evsel *evsel, struct cpu_map *cpus, - struct thread_map *threads, int pages, - struct perf_evlist *evlist) +static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot, + int mask, int fd) +{ + evlist->mmap[cpu].prev = 0; + evlist->mmap[cpu].mask = mask; + evlist->mmap[cpu].base = mmap(NULL, evlist->mmap_len, prot, + MAP_SHARED, fd, 0); + if (evlist->mmap[cpu].base == MAP_FAILED) + return -1; + + perf_evlist__add_pollfd(evlist, fd); + return 0; +} + +static int perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *evsel, + int cpu, int thread, int fd) +{ + struct perf_sample_id *sid; + u64 read_data[4] = { 0, }; + int hash, id_idx = 1; /* The first entry is the counter value */ + + if (!(evsel->attr.read_format & PERF_FORMAT_ID) || + read(fd, &read_data, sizeof(read_data)) == -1) + return -1; + + if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + ++id_idx; + if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + ++id_idx; + + sid = SID(evsel, cpu, thread); + sid->id = read_data[id_idx]; + sid->evsel = evsel; + hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS); + hlist_add_head(&sid->node, &evlist->heads[hash]); + return 0; +} + +/** perf_evlist__mmap - Create per cpu maps to receive events + * + * @evlist - list of events + * @cpus - cpu map being monitored + * @threads - threads map being monitored + * @pages - map length in pages + * @overwrite - overwrite older events? + * + * If overwrite is false the user needs to signal event consuption using: + * + * struct perf_mmap *m = &evlist->mmap[cpu]; + * unsigned int head = perf_mmap__read_head(m); + * + * perf_mmap__write_tail(m, head) + */ +int perf_evlist__mmap(struct perf_evlist *evlist, struct cpu_map *cpus, + struct thread_map *threads, int pages, bool overwrite) { unsigned int page_size = sysconf(_SC_PAGE_SIZE); int mask = pages * page_size - 1, cpu; - struct perf_mmap *mm; - int thread; + struct perf_evsel *first_evsel, *evsel; + int thread, prot = PROT_READ | (overwrite ? 0 : PROT_WRITE); - if (evsel->mmap == NULL && - perf_evsel__alloc_mmap(evsel, cpus->nr, threads->nr) < 0) + if (evlist->mmap == NULL && + perf_evlist__alloc_mmap(evlist, cpus->nr) < 0) return -ENOMEM; - evsel->mmap_len = (pages + 1) * page_size; + if (evlist->pollfd == NULL && + perf_evlist__alloc_pollfd(evlist, cpus->nr, threads->nr) < 0) + return -ENOMEM; - for (cpu = 0; cpu < cpus->nr; cpu++) { - for (thread = 0; thread < threads->nr; thread++) { - mm = xyarray__entry(evsel->mmap, cpu, thread); - mm->prev = 0; - mm->mask = mask; - mm->base = mmap(NULL, evsel->mmap_len, PROT_READ, - MAP_SHARED, FD(evsel, cpu, thread), 0); - if (mm->base == MAP_FAILED) - goto out_unmap; + evlist->mmap_len = (pages + 1) * page_size; + first_evsel = list_entry(evlist->entries.next, struct perf_evsel, node); - if (evlist != NULL) - perf_evlist__add_pollfd(evlist, FD(evsel, cpu, thread)); + list_for_each_entry(evsel, &evlist->entries, node) { + if ((evsel->attr.read_format & PERF_FORMAT_ID) && + evsel->id == NULL && + perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0) + return -ENOMEM; + + for (cpu = 0; cpu < cpus->nr; cpu++) { + for (thread = 0; thread < threads->nr; thread++) { + int fd = FD(evsel, cpu, thread); + + if (evsel->idx || thread) { + if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, + FD(first_evsel, cpu, 0)) != 0) + goto out_unmap; + } else if (__perf_evlist__mmap(evlist, cpu, prot, mask, fd) < 0) + goto out_unmap; + + if ((evsel->attr.read_format & PERF_FORMAT_ID) && + perf_evlist__id_hash(evlist, evsel, cpu, thread, fd) < 0) + goto out_unmap; + } } } return 0; out_unmap: - do { - while (--thread >= 0) { - mm = xyarray__entry(evsel->mmap, cpu, thread); - munmap(mm->base, evsel->mmap_len); - mm->base = NULL; + for (cpu = 0; cpu < cpus->nr; cpu++) { + if (evlist->mmap[cpu].base != NULL) { + munmap(evlist->mmap[cpu].base, evlist->mmap_len); + evlist->mmap[cpu].base = NULL; } - thread = threads->nr; - } while (--cpu >= 0); - + } return -1; } diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index c8fbef299436..667ee4e2e35e 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -24,14 +24,25 @@ struct perf_counts { struct perf_counts_values cpu[]; }; +struct perf_evsel; + +/* + * Per fd, to map back from PERF_SAMPLE_ID to evsel, only used when there are + * more than one entry in the evlist. + */ +struct perf_sample_id { + struct hlist_node node; + u64 id; + struct perf_evsel *evsel; +}; + struct perf_evsel { struct list_head node; struct perf_event_attr attr; char *filter; struct xyarray *fd; - struct xyarray *mmap; + struct xyarray *id; struct perf_counts *counts; - size_t mmap_len; int idx; void *priv; }; @@ -44,9 +55,11 @@ struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx); void perf_evsel__delete(struct perf_evsel *evsel); int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads); +int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads); int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus); -int perf_evsel__alloc_mmap(struct perf_evsel *evsel, int ncpus, int nthreads); +int perf_evlist__alloc_mmap(struct perf_evlist *evlist, int ncpus); void perf_evsel__free_fd(struct perf_evsel *evsel); +void perf_evsel__free_id(struct perf_evsel *evsel); void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads); int perf_evsel__open_per_cpu(struct perf_evsel *evsel, @@ -55,10 +68,9 @@ int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads, bool group, bool inherit); int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, struct thread_map *threads, bool group, bool inherit); -int perf_evsel__mmap(struct perf_evsel *evsel, struct cpu_map *cpus, - struct thread_map *threads, int pages, - struct perf_evlist *evlist); -void perf_evsel__munmap(struct perf_evsel *evsel, int ncpus, int nthreads); +int perf_evlist__mmap(struct perf_evlist *evlist, struct cpu_map *cpus, + struct thread_map *threads, int pages, bool overwrite); +void perf_evlist__munmap(struct perf_evlist *evlist, int ncpus); #define perf_evsel__match(evsel, t, c) \ (evsel->attr.type == PERF_TYPE_##t && \ From 0a27d7f9f417c0305f7efa70631764a53c7af219 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Fri, 14 Jan 2011 15:50:51 -0200 Subject: [PATCH 11/29] perf record: Use perf_evlist__mmap There is more stuff that can go to the perf_ev{sel,list} layer, like detecting if sample_id_all is available, etc, but lets try using this in 'perf test' first. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 59 +++++++------------------------------ 1 file changed, 11 insertions(+), 48 deletions(-) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 109f3b269ac5..45a3689f9ed6 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -30,6 +30,7 @@ #include #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) +#define SID(e, x, y) xyarray__entry(e->id, x, y) enum write_mode_t { WRITE_FORCE, @@ -78,8 +79,6 @@ static off_t post_processing_offset; static struct perf_session *session; static const char *cpu_list; -static struct perf_mmap mmap_array[MAX_NR_CPUS]; - static void advance_output(size_t size) { bytes_written += size; @@ -196,20 +195,14 @@ static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int n return h_attr; } -static void create_counter(struct perf_evlist *evlist, - struct perf_evsel *evsel, int cpu) +static void create_counter(struct perf_evsel *evsel, int cpu) { char *filter = evsel->filter; struct perf_event_attr *attr = &evsel->attr; struct perf_header_attr *h_attr; + struct perf_sample_id *sid; int thread_index; int ret; - struct { - u64 count; - u64 time_enabled; - u64 time_running; - u64 id; - } read_data; for (thread_index = 0; thread_index < threads->nr; thread_index++) { h_attr = get_header_attr(attr, evsel->idx); @@ -223,45 +216,12 @@ static void create_counter(struct perf_evlist *evlist, } } - if (read(FD(evsel, cpu, thread_index), &read_data, sizeof(read_data)) == -1) { - perror("Unable to read perf file descriptor"); - exit(-1); - } - - if (perf_header_attr__add_id(h_attr, read_data.id) < 0) { + sid = SID(evsel, cpu, thread_index); + if (perf_header_attr__add_id(h_attr, sid->id) < 0) { pr_warning("Not enough memory to add id\n"); exit(-1); } - assert(FD(evsel, cpu, thread_index) >= 0); - fcntl(FD(evsel, cpu, thread_index), F_SETFL, O_NONBLOCK); - - if (evsel->idx || thread_index) { - struct perf_evsel *first; - first = list_entry(evlist->entries.next, struct perf_evsel, node); - ret = ioctl(FD(evsel, cpu, thread_index), - PERF_EVENT_IOC_SET_OUTPUT, - FD(first, cpu, 0)); - if (ret) { - error("failed to set output: %d (%s)\n", errno, - strerror(errno)); - exit(-1); - } - } else { - mmap_array[cpu].prev = 0; - mmap_array[cpu].mask = mmap_pages*page_size - 1; - mmap_array[cpu].base = mmap(NULL, (mmap_pages+1)*page_size, - PROT_READ | PROT_WRITE, MAP_SHARED, FD(evsel, cpu, thread_index), 0); - if (mmap_array[cpu].base == MAP_FAILED) { - error("failed to mmap with %d (%s)\n", errno, strerror(errno)); - exit(-1); - } - - evlist->pollfd[evlist->nr_fds].fd = FD(evsel, cpu, thread_index); - evlist->pollfd[evlist->nr_fds].events = POLLIN; - evlist->nr_fds++; - } - if (filter != NULL) { ret = ioctl(FD(evsel, cpu, thread_index), PERF_EVENT_IOC_SET_FILTER, filter); @@ -423,9 +383,12 @@ try_again: } } + if (perf_evlist__mmap(evlist, cpus, threads, mmap_pages, false) < 0) + die("failed to mmap with %d (%s)\n", errno, strerror(errno)); + for (cpu = 0; cpu < cpus->nr; ++cpu) { list_for_each_entry(pos, &evlist->entries, node) - create_counter(evlist, pos, cpu); + create_counter(pos, cpu); } } @@ -502,8 +465,8 @@ static void mmap_read_all(void) int i; for (i = 0; i < cpus->nr; i++) { - if (mmap_array[i].base) - mmap_read(&mmap_array[i]); + if (evsel_list->mmap[i].base) + mmap_read(&evsel_list->mmap[i]); } if (perf_header__has_feat(&session->header, HEADER_TRACE_INFO)) From 915fce20ecf8f7ff4189d0fff42b62aebf6a57cc Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Fri, 14 Jan 2011 16:19:12 -0200 Subject: [PATCH 12/29] perf tools: Add missing cpu_map__delete() Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/cpumap.c | 5 +++++ tools/perf/util/cpumap.h | 2 +- 2 files changed, 6 insertions(+), 1 deletion(-) diff --git a/tools/perf/util/cpumap.c b/tools/perf/util/cpumap.c index 3ccaa1043383..6893eec693ab 100644 --- a/tools/perf/util/cpumap.c +++ b/tools/perf/util/cpumap.c @@ -177,3 +177,8 @@ struct cpu_map *cpu_map__dummy_new(void) return cpus; } + +void cpu_map__delete(struct cpu_map *map) +{ + free(map); +} diff --git a/tools/perf/util/cpumap.h b/tools/perf/util/cpumap.h index f7a4f42f6307..072c0a374794 100644 --- a/tools/perf/util/cpumap.h +++ b/tools/perf/util/cpumap.h @@ -8,6 +8,6 @@ struct cpu_map { struct cpu_map *cpu_map__new(const char *cpu_list); struct cpu_map *cpu_map__dummy_new(void); -void *cpu_map__delete(struct cpu_map *map); +void cpu_map__delete(struct cpu_map *map); #endif /* __PERF_CPUMAP_H */ From d2af9687c96f3864178de1860e6d83873aeef224 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Fri, 14 Jan 2011 16:24:49 -0200 Subject: [PATCH 13/29] perf test: Check counts on all cpus in test__open_syscall_event_on_all_cpus We were bailing out after the first count mismatch, do it in all to see if only some CPUs are not getting the expected number of events. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-test.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index 7287158c4830..7cc6b2086947 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -408,6 +408,8 @@ static int test__open_syscall_event_on_all_cpus(void) goto out_close_fd; } + err = 0; + for (cpu = 0; cpu < cpus->nr; ++cpu) { unsigned int expected; @@ -416,18 +418,18 @@ static int test__open_syscall_event_on_all_cpus(void) if (perf_evsel__read_on_cpu(evsel, cpu, 0) < 0) { pr_debug("perf_evsel__open_read_on_cpu\n"); - goto out_close_fd; + err = -1; + break; } expected = nr_open_calls + cpu; if (evsel->counts->cpu[cpu].val != expected) { pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls on cpu %d, got %" PRIu64 "\n", expected, cpus->map[cpu], evsel->counts->cpu[cpu].val); - goto out_close_fd; + err = -1; } } - err = 0; out_close_fd: perf_evsel__close_fd(evsel, 1, threads->nr); out_evsel_delete: From 98d77b78504a423fca911a26a17bee00ef2fdda2 Mon Sep 17 00:00:00 2001 From: Han Pingtian Date: Sat, 15 Jan 2011 07:00:50 +0800 Subject: [PATCH 14/29] perf test: check if cpu_map__new() return NULL It looks like we should check if cpus is NULL after cpus = cpu_map__new(NULL); in test__open_syscall_event_on_all_cpus(). LKML-Reference: <20110114230050.GA7011@localhost> Signed-off-by: Han Pingtian Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-test.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index 7cc6b2086947..5a50e4755e6c 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -347,9 +347,9 @@ static int test__open_syscall_event_on_all_cpus(void) } cpus = cpu_map__new(NULL); - if (threads == NULL) { - pr_debug("thread_map__new\n"); - return -1; + if (cpus == NULL) { + pr_debug("cpu_map__new\n"); + goto out_thread_map_delete; } From 04391debc3e1195222a4dbb162ace6542dd89c1c Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sat, 15 Jan 2011 10:40:59 -0200 Subject: [PATCH 15/29] perf evlist: Steal mmap reading routine from 'perf top' Will be used in the upcoming 'perf test' entry for the evlist mmap routines. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-top.c | 56 ++---------------------------------- tools/perf/util/evlist.c | 62 ++++++++++++++++++++++++++++++++++++++++ tools/perf/util/evlist.h | 4 +++ 3 files changed, 69 insertions(+), 53 deletions(-) diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index df85c1f9417b..58352ad807c7 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -1100,67 +1100,17 @@ static void event__process_sample(const event_t *self, static void perf_session__mmap_read_cpu(struct perf_session *self, int cpu) { - struct perf_mmap *md = &evsel_list->mmap[cpu]; - unsigned int head = perf_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; - - size_t size = event->header.size; - - /* - * 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_t *event; + while ((event = perf_evlist__read_on_cpu(evsel_list, cpu)) != NULL) { event__parse_sample(event, self, &sample); + if (event->header.type == PERF_RECORD_SAMPLE) event__process_sample(event, &sample, self); else event__process(event, &sample, self); - old += size; } - - md->prev = old; } static void perf_session__mmap_read(struct perf_session *self) diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index deb82a4fc312..4b3b84cd71a1 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -95,3 +95,65 @@ struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) return sid->evsel; return NULL; } + +event_t *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu) +{ + /* XXX Move this to perf.c, making it generally available */ + unsigned int page_size = sysconf(_SC_PAGE_SIZE); + struct perf_mmap *md = &evlist->mmap[cpu]; + unsigned int head = perf_mmap__read_head(md); + unsigned int old = md->prev; + unsigned char *data = md->base + page_size; + event_t *event = NULL; + 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; + } + + if (old != head) { + size_t size; + + event = (event_t *)&data[old & md->mask]; + size = event->header.size; + + /* + * 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 = &evlist->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 = &evlist->event_copy; + } + + old += size; + } + + md->prev = old; + return event; +} diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h index dbfcc79bb995..28712063db97 100644 --- a/tools/perf/util/evlist.h +++ b/tools/perf/util/evlist.h @@ -3,6 +3,7 @@ #include #include "../perf.h" +#include "event.h" struct pollfd; @@ -15,6 +16,7 @@ struct perf_evlist { int nr_entries; int nr_fds; int mmap_len; + event_t event_copy; struct perf_mmap *mmap; struct pollfd *pollfd; }; @@ -32,4 +34,6 @@ void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd); struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id); +event_t *perf_evlist__read_on_cpu(struct perf_evlist *self, int cpu); + #endif /* __PERF_EVLIST_H */ From de5fa3a8a05cd60f59622e88cfeb90416760d78e Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sat, 15 Jan 2011 10:42:46 -0200 Subject: [PATCH 16/29] perf test: Add test for the evlist mmap routines This test will generate random numbers of calls to some getpid syscalls, then establish an mmap for a group of events that are created to monitor these syscalls. It will receive the events, using mmap, use its PERF_SAMPLE_ID generated sample.id field to map back to its respective perf_evsel instance. Then it checks if the number of syscalls reported as perf events by the kernel corresponds to the number of syscalls made. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-test.c | 171 +++++++++++++++++++++++++++++++++++++- 1 file changed, 169 insertions(+), 2 deletions(-) diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index 5a50e4755e6c..4fd34537c01d 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -7,7 +7,9 @@ #include "util/cache.h" #include "util/debug.h" +#include "util/evlist.h" #include "util/parse-options.h" +#include "util/parse-events.h" #include "util/session.h" #include "util/symbol.h" #include "util/thread.h" @@ -238,14 +240,14 @@ out: #include "util/evsel.h" #include -static int trace_event__id(const char *event_name) +static int trace_event__id(const char *evname) { char *filename; int err = -1, fd; if (asprintf(&filename, "/sys/kernel/debug/tracing/events/syscalls/%s/id", - event_name) < 0) + evname) < 0) return -1; fd = open(filename, O_RDONLY); @@ -439,6 +441,167 @@ out_thread_map_delete: return err; } +/* + * This test will generate random numbers of calls to some getpid syscalls, + * then establish an mmap for a group of events that are created to monitor + * the syscalls. + * + * It will receive the events, using mmap, use its PERF_SAMPLE_ID generated + * sample.id field to map back to its respective perf_evsel instance. + * + * Then it checks if the number of syscalls reported as perf events by + * the kernel corresponds to the number of syscalls made. + */ +static int test__basic_mmap(void) +{ + int err = -1; + event_t *event; + struct thread_map *threads; + struct perf_session session; + struct cpu_map *cpus; + struct perf_evlist *evlist; + struct perf_event_attr attr = { + .type = PERF_TYPE_TRACEPOINT, + .read_format = PERF_FORMAT_ID, + .sample_type = PERF_SAMPLE_ID, + .watermark = 0, + }; + cpu_set_t cpu_set; + const char *syscall_names[] = { "getsid", "getppid", "getpgrp", + "getpgid", }; + pid_t (*syscalls[])(void) = { (void *)getsid, getppid, getpgrp, + (void*)getpgid }; +#define nsyscalls ARRAY_SIZE(syscall_names) + int ids[nsyscalls]; + unsigned int nr_events[nsyscalls], + expected_nr_events[nsyscalls], i, j; + struct perf_evsel *evsels[nsyscalls], *evsel; + + for (i = 0; i < nsyscalls; ++i) { + char name[64]; + + snprintf(name, sizeof(name), "sys_enter_%s", syscall_names[i]); + ids[i] = trace_event__id(name); + if (ids[i] < 0) { + pr_debug("Is debugfs mounted on /sys/kernel/debug?\n"); + return -1; + } + nr_events[i] = 0; + expected_nr_events[i] = random() % 257; + } + + threads = thread_map__new(-1, getpid()); + if (threads == NULL) { + pr_debug("thread_map__new\n"); + return -1; + } + + cpus = cpu_map__new(NULL); + if (threads == NULL) { + pr_debug("thread_map__new\n"); + goto out_free_threads; + } + + CPU_ZERO(&cpu_set); + CPU_SET(cpus->map[0], &cpu_set); + sched_setaffinity(0, sizeof(cpu_set), &cpu_set); + if (sched_setaffinity(0, sizeof(cpu_set), &cpu_set) < 0) { + pr_debug("sched_setaffinity() failed on CPU %d: %s ", + cpus->map[0], strerror(errno)); + goto out_free_cpus; + } + + evlist = perf_evlist__new(); + if (threads == NULL) { + pr_debug("perf_evlist__new\n"); + goto out_free_cpus; + } + + /* anonymous union fields, can't be initialized above */ + attr.wakeup_events = 1; + attr.sample_period = 1; + + /* + * FIXME: use evsel->attr.sample_type in event__parse_sample. + * This will nicely remove the requirement that we have + * all the events with the same sample_type. + */ + session.sample_type = attr.sample_type; + + for (i = 0; i < nsyscalls; ++i) { + attr.config = ids[i]; + evsels[i] = perf_evsel__new(&attr, i); + if (evsels[i] == NULL) { + pr_debug("perf_evsel__new\n"); + goto out_free_evlist; + } + + perf_evlist__add(evlist, evsels[i]); + + if (perf_evsel__open(evsels[i], cpus, threads, false, false) < 0) { + pr_debug("failed to open counter: %s, " + "tweak /proc/sys/kernel/perf_event_paranoid?\n", + strerror(errno)); + goto out_close_fd; + } + } + + if (perf_evlist__mmap(evlist, cpus, threads, 128, true) < 0) { + pr_debug("failed to mmap events: %d (%s)\n", errno, + strerror(errno)); + goto out_close_fd; + } + + for (i = 0; i < nsyscalls; ++i) + for (j = 0; j < expected_nr_events[i]; ++j) { + int foo = syscalls[i](); + ++foo; + } + + while ((event = perf_evlist__read_on_cpu(evlist, 0)) != NULL) { + struct sample_data sample; + + if (event->header.type != PERF_RECORD_SAMPLE) { + pr_debug("unexpected %s event\n", + event__get_event_name(event->header.type)); + goto out_munmap; + } + + event__parse_sample(event, &session, &sample); + evsel = perf_evlist__id2evsel(evlist, sample.id); + if (evsel == NULL) { + pr_debug("event with id %" PRIu64 + " doesn't map to an evsel\n", sample.id); + goto out_munmap; + } + nr_events[evsel->idx]++; + } + + list_for_each_entry(evsel, &evlist->entries, node) { + if (nr_events[evsel->idx] != expected_nr_events[evsel->idx]) { + pr_debug("expected %d %s events, got %d\n", + expected_nr_events[evsel->idx], + event_name(evsel), nr_events[evsel->idx]); + goto out_munmap; + } + } + + err = 0; +out_munmap: + perf_evlist__munmap(evlist, 1); +out_close_fd: + for (i = 0; i < nsyscalls; ++i) + perf_evsel__close_fd(evsels[i], 1, threads->nr); +out_free_evlist: + perf_evlist__delete(evlist); +out_free_cpus: + cpu_map__delete(cpus); +out_free_threads: + thread_map__delete(threads); + return err; +#undef nsyscalls +} + static struct test { const char *desc; int (*func)(void); @@ -455,6 +618,10 @@ static struct test { .desc = "detect open syscall event on all cpus", .func = test__open_syscall_event_on_all_cpus, }, + { + .desc = "read samples using the mmap interface", + .func = test__basic_mmap, + }, { .func = NULL, }, From 1b3a0e9592ebf174af934b3908a2bf6a6fa86169 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Fri, 14 Jan 2011 04:51:58 +0100 Subject: [PATCH 17/29] perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-report.c | 25 ++--- tools/perf/util/callchain.c | 214 ++++++++++++++++++------------------ tools/perf/util/callchain.h | 66 ++++++++++- tools/perf/util/hist.c | 13 ++- tools/perf/util/hist.h | 2 + tools/perf/util/session.c | 22 ++-- tools/perf/util/session.h | 11 +- 7 files changed, 209 insertions(+), 144 deletions(-) diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c index c27e31f289e6..c95599a82f9e 100644 --- a/tools/perf/builtin-report.c +++ b/tools/perf/builtin-report.c @@ -81,18 +81,17 @@ static int perf_session__add_hist_entry(struct perf_session *self, struct addr_location *al, struct sample_data *data) { - struct map_symbol *syms = NULL; struct symbol *parent = NULL; - int err = -ENOMEM; + int err = 0; struct hist_entry *he; struct hists *hists; struct perf_event_attr *attr; if ((sort__has_parent || symbol_conf.use_callchain) && data->callchain) { - syms = perf_session__resolve_callchain(self, al->thread, - data->callchain, &parent); - if (syms == NULL) - return -ENOMEM; + err = perf_session__resolve_callchain(self, al->thread, + data->callchain, &parent); + if (err) + return err; } attr = perf_header__find_attr(data->id, &self->header); @@ -101,16 +100,17 @@ static int perf_session__add_hist_entry(struct perf_session *self, else hists = perf_session__hists_findnew(self, data->id, 0, 0); if (hists == NULL) - goto out_free_syms; + return -ENOMEM; + he = __hists__add_entry(hists, al, parent, data->period); if (he == NULL) - goto out_free_syms; - err = 0; + return -ENOMEM; + if (symbol_conf.use_callchain) { - err = callchain_append(he->callchain, data->callchain, syms, + err = callchain_append(he->callchain, &self->callchain_cursor, data->period); if (err) - goto out_free_syms; + return err; } /* * Only in the newt browser we are doing integrated annotation, @@ -119,8 +119,7 @@ static int perf_session__add_hist_entry(struct perf_session *self, */ if (use_browser > 0) err = hist_entry__inc_addr_samples(he, al->addr); -out_free_syms: - free(syms); + return err; } diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c index e12d539417b2..53a49e0cfc6c 100644 --- a/tools/perf/util/callchain.c +++ b/tools/perf/util/callchain.c @@ -1,5 +1,5 @@ /* - * Copyright (C) 2009-2010, Frederic Weisbecker + * Copyright (C) 2009-2011, Frederic Weisbecker * * Handle the callchains from the stream in an ad-hoc radix tree and then * sort them in an rbtree. @@ -195,26 +195,21 @@ create_child(struct callchain_node *parent, bool inherit_children) } -struct resolved_ip { - u64 ip; - struct map_symbol ms; -}; - -struct resolved_chain { - u64 nr; - struct resolved_ip ips[0]; -}; - - /* * Fill the node with callchain values */ static void -fill_node(struct callchain_node *node, struct resolved_chain *chain, int start) +fill_node(struct callchain_node *node, struct callchain_cursor *cursor) { - unsigned int i; + struct callchain_cursor_node *cursor_node; - for (i = start; i < chain->nr; i++) { + node->val_nr = cursor->nr - cursor->pos; + if (!node->val_nr) + pr_warning("Warning: empty node in callchain tree\n"); + + cursor_node = callchain_cursor_current(cursor); + + while (cursor_node) { struct callchain_list *call; call = zalloc(sizeof(*call)); @@ -222,23 +217,25 @@ fill_node(struct callchain_node *node, struct resolved_chain *chain, int start) perror("not enough memory for the code path tree"); return; } - call->ip = chain->ips[i].ip; - call->ms = chain->ips[i].ms; + call->ip = cursor_node->ip; + call->ms.sym = cursor_node->sym; + call->ms.map = cursor_node->map; list_add_tail(&call->list, &node->val); + + callchain_cursor_advance(cursor); + cursor_node = callchain_cursor_current(cursor); } - node->val_nr = chain->nr - start; - if (!node->val_nr) - pr_warning("Warning: empty node in callchain tree\n"); } static void -add_child(struct callchain_node *parent, struct resolved_chain *chain, - int start, u64 period) +add_child(struct callchain_node *parent, + struct callchain_cursor *cursor, + u64 period) { struct callchain_node *new; new = create_child(parent, false); - fill_node(new, chain, start); + fill_node(new, cursor); new->children_hit = 0; new->hit = period; @@ -250,9 +247,10 @@ add_child(struct callchain_node *parent, struct resolved_chain *chain, * Then create another child to host the given callchain of new branch */ static void -split_add_child(struct callchain_node *parent, struct resolved_chain *chain, - struct callchain_list *to_split, int idx_parents, int idx_local, - u64 period) +split_add_child(struct callchain_node *parent, + struct callchain_cursor *cursor, + struct callchain_list *to_split, + u64 idx_parents, u64 idx_local, u64 period) { struct callchain_node *new; struct list_head *old_tail; @@ -277,9 +275,9 @@ split_add_child(struct callchain_node *parent, struct resolved_chain *chain, parent->val_nr = idx_local; /* create a new child for the new branch if any */ - if (idx_total < chain->nr) { + if (idx_total < cursor->nr) { parent->hit = 0; - add_child(parent, chain, idx_total, period); + add_child(parent, cursor, period); parent->children_hit += period; } else { parent->hit = period; @@ -287,36 +285,41 @@ split_add_child(struct callchain_node *parent, struct resolved_chain *chain, } static int -append_chain(struct callchain_node *root, struct resolved_chain *chain, - unsigned int start, u64 period); +append_chain(struct callchain_node *root, + struct callchain_cursor *cursor, + u64 period); static void -append_chain_children(struct callchain_node *root, struct resolved_chain *chain, - unsigned int start, u64 period) +append_chain_children(struct callchain_node *root, + struct callchain_cursor *cursor, + u64 period) { struct callchain_node *rnode; /* lookup in childrens */ chain_for_each_child(rnode, root) { - unsigned int ret = append_chain(rnode, chain, start, period); + unsigned int ret = append_chain(rnode, cursor, period); if (!ret) goto inc_children_hit; } /* nothing in children, add to the current node */ - add_child(root, chain, start, period); + add_child(root, cursor, period); inc_children_hit: root->children_hit += period; } static int -append_chain(struct callchain_node *root, struct resolved_chain *chain, - unsigned int start, u64 period) +append_chain(struct callchain_node *root, + struct callchain_cursor *cursor, + u64 period) { + struct callchain_cursor_node *curr_snap = cursor->curr; struct callchain_list *cnode; - unsigned int i = start; + u64 start = cursor->pos; bool found = false; + u64 matches; /* * Lookup in the current node @@ -324,114 +327,95 @@ append_chain(struct callchain_node *root, struct resolved_chain *chain, * anywhere inside a function. */ list_for_each_entry(cnode, &root->val, list) { + struct callchain_cursor_node *node; struct symbol *sym; - if (i == chain->nr) + node = callchain_cursor_current(cursor); + if (!node) break; - sym = chain->ips[i].ms.sym; + sym = node->sym; if (cnode->ms.sym && sym) { if (cnode->ms.sym->start != sym->start) break; - } else if (cnode->ip != chain->ips[i].ip) + } else if (cnode->ip != node->ip) break; if (!found) found = true; - i++; + + callchain_cursor_advance(cursor); } /* matches not, relay on the parent */ - if (!found) + if (!found) { + cursor->curr = curr_snap; + cursor->pos = start; return -1; + } + + matches = cursor->pos - start; /* we match only a part of the node. Split it and add the new chain */ - if (i - start < root->val_nr) { - split_add_child(root, chain, cnode, start, i - start, period); + if (matches < root->val_nr) { + split_add_child(root, cursor, cnode, start, matches, period); return 0; } /* we match 100% of the path, increment the hit */ - if (i - start == root->val_nr && i == chain->nr) { + if (matches == root->val_nr && cursor->pos == cursor->nr) { root->hit += period; return 0; } /* We match the node and still have a part remaining */ - append_chain_children(root, chain, i, period); + append_chain_children(root, cursor, period); return 0; } -static void filter_context(struct ip_callchain *old, struct resolved_chain *new, - struct map_symbol *syms) +int callchain_append(struct callchain_root *root, + struct callchain_cursor *cursor, + u64 period) { - int i, j = 0; - - for (i = 0; i < (int)old->nr; i++) { - if (old->ips[i] >= PERF_CONTEXT_MAX) - continue; - - new->ips[j].ip = old->ips[i]; - new->ips[j].ms = syms[i]; - j++; - } - - new->nr = j; -} - - -int callchain_append(struct callchain_root *root, struct ip_callchain *chain, - struct map_symbol *syms, u64 period) -{ - struct resolved_chain *filtered; - - if (!chain->nr) + if (!cursor->nr) return 0; - filtered = zalloc(sizeof(*filtered) + - chain->nr * sizeof(struct resolved_ip)); - if (!filtered) - return -ENOMEM; + callchain_cursor_commit(cursor); - filter_context(chain, filtered, syms); + append_chain_children(&root->node, cursor, period); - if (!filtered->nr) - goto end; - - append_chain_children(&root->node, filtered, 0, period); - - if (filtered->nr > root->max_depth) - root->max_depth = filtered->nr; -end: - free(filtered); + if (cursor->nr > root->max_depth) + root->max_depth = cursor->nr; return 0; } static int -merge_chain_branch(struct callchain_node *dst, struct callchain_node *src, - struct resolved_chain *chain) +merge_chain_branch(struct callchain_cursor *cursor, + struct callchain_node *dst, struct callchain_node *src) { + struct callchain_cursor_node **old_last = cursor->last; struct callchain_node *child, *next_child; struct callchain_list *list, *next_list; - int old_pos = chain->nr; + int old_pos = cursor->nr; int err = 0; list_for_each_entry_safe(list, next_list, &src->val, list) { - chain->ips[chain->nr].ip = list->ip; - chain->ips[chain->nr].ms = list->ms; - chain->nr++; + callchain_cursor_append(cursor, list->ip, + list->ms.map, list->ms.sym); list_del(&list->list); free(list); } - if (src->hit) - append_chain_children(dst, chain, 0, src->hit); + if (src->hit) { + callchain_cursor_commit(cursor); + append_chain_children(dst, cursor, src->hit); + } chain_for_each_child_safe(child, next_child, src) { - err = merge_chain_branch(dst, child, chain); + err = merge_chain_branch(cursor, dst, child); if (err) break; @@ -439,26 +423,38 @@ merge_chain_branch(struct callchain_node *dst, struct callchain_node *src, free(child); } - chain->nr = old_pos; + cursor->nr = old_pos; + cursor->last = old_last; return err; } -int callchain_merge(struct callchain_root *dst, struct callchain_root *src) +int callchain_merge(struct callchain_cursor *cursor, + struct callchain_root *dst, struct callchain_root *src) { - struct resolved_chain *chain; - int err; - - chain = malloc(sizeof(*chain) + - src->max_depth * sizeof(struct resolved_ip)); - if (!chain) - return -ENOMEM; - - chain->nr = 0; - - err = merge_chain_branch(&dst->node, &src->node, chain); - - free(chain); - - return err; + return merge_chain_branch(cursor, &dst->node, &src->node); +} + +int callchain_cursor_append(struct callchain_cursor *cursor, + u64 ip, struct map *map, struct symbol *sym) +{ + struct callchain_cursor_node *node = *cursor->last; + + if (!node) { + node = calloc(sizeof(*node), 1); + if (!node) + return -ENOMEM; + + *cursor->last = node; + } + + node->ip = ip; + node->map = map; + node->sym = sym; + + cursor->nr++; + + cursor->last = &node->next; + + return 0; } diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h index c15fb8c24ad2..d74a19af4a44 100644 --- a/tools/perf/util/callchain.h +++ b/tools/perf/util/callchain.h @@ -49,6 +49,27 @@ struct callchain_list { struct list_head list; }; +/* + * A callchain cursor is a single linked list that + * let one feed a callchain progressively. + * It keeps persitent allocated entries to minimize + * allocations. + */ +struct callchain_cursor_node { + u64 ip; + struct map *map; + struct symbol *sym; + struct callchain_cursor_node *next; +}; + +struct callchain_cursor { + u64 nr; + struct callchain_cursor_node *first; + struct callchain_cursor_node **last; + u64 pos; + struct callchain_cursor_node *curr; +}; + static inline void callchain_init(struct callchain_root *root) { INIT_LIST_HEAD(&root->node.brothers); @@ -67,9 +88,48 @@ static inline u64 cumul_hits(struct callchain_node *node) } int register_callchain_param(struct callchain_param *param); -int callchain_append(struct callchain_root *root, struct ip_callchain *chain, - struct map_symbol *syms, u64 period); -int callchain_merge(struct callchain_root *dst, struct callchain_root *src); +int callchain_append(struct callchain_root *root, + struct callchain_cursor *cursor, + u64 period); + +int callchain_merge(struct callchain_cursor *cursor, + struct callchain_root *dst, struct callchain_root *src); bool ip_callchain__valid(struct ip_callchain *chain, const event_t *event); + +/* + * Initialize a cursor before adding entries inside, but keep + * the previously allocated entries as a cache. + */ +static inline void callchain_cursor_reset(struct callchain_cursor *cursor) +{ + cursor->nr = 0; + cursor->last = &cursor->first; +} + +int callchain_cursor_append(struct callchain_cursor *cursor, u64 ip, + struct map *map, struct symbol *sym); + +/* Close a cursor writing session. Initialize for the reader */ +static inline void callchain_cursor_commit(struct callchain_cursor *cursor) +{ + cursor->curr = cursor->first; + cursor->pos = 0; +} + +/* Cursor reading iteration helpers */ +static inline struct callchain_cursor_node * +callchain_cursor_current(struct callchain_cursor *cursor) +{ + if (cursor->pos == cursor->nr) + return NULL; + + return cursor->curr; +} + +static inline void callchain_cursor_advance(struct callchain_cursor *cursor) +{ + cursor->curr = cursor->curr->next; + cursor->pos++; +} #endif /* __PERF_CALLCHAIN_H */ diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c index 32f4f1f2f6e4..a438a0652d23 100644 --- a/tools/perf/util/hist.c +++ b/tools/perf/util/hist.c @@ -211,7 +211,9 @@ void hist_entry__free(struct hist_entry *he) * collapse the histogram */ -static bool collapse__insert_entry(struct rb_root *root, struct hist_entry *he) +static bool hists__collapse_insert_entry(struct hists *self, + struct rb_root *root, + struct hist_entry *he) { struct rb_node **p = &root->rb_node; struct rb_node *parent = NULL; @@ -226,8 +228,11 @@ static bool collapse__insert_entry(struct rb_root *root, struct hist_entry *he) if (!cmp) { iter->period += he->period; - if (symbol_conf.use_callchain) - callchain_merge(iter->callchain, he->callchain); + if (symbol_conf.use_callchain) { + callchain_cursor_reset(&self->callchain_cursor); + callchain_merge(&self->callchain_cursor, iter->callchain, + he->callchain); + } hist_entry__free(he); return false; } @@ -262,7 +267,7 @@ void hists__collapse_resort(struct hists *self) next = rb_next(&n->rb_node); rb_erase(&n->rb_node, &self->entries); - if (collapse__insert_entry(&tmp, n)) + if (hists__collapse_insert_entry(self, &tmp, n)) hists__inc_nr_entries(self, n); } diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h index ee789856a8c9..889559b86492 100644 --- a/tools/perf/util/hist.h +++ b/tools/perf/util/hist.h @@ -77,6 +77,8 @@ struct hists { u64 event_stream; u32 type; u16 col_len[HISTC_NR_COLS]; + /* Best would be to reuse the session callchain cursor */ + struct callchain_cursor callchain_cursor; }; struct hist_entry *__hists__add_entry(struct hists *self, diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c index 105f00bfd555..b58a48a5e5a9 100644 --- a/tools/perf/util/session.c +++ b/tools/perf/util/session.c @@ -242,17 +242,16 @@ static bool symbol__match_parent_regex(struct symbol *sym) return 0; } -struct map_symbol *perf_session__resolve_callchain(struct perf_session *self, - struct thread *thread, - struct ip_callchain *chain, - struct symbol **parent) +int perf_session__resolve_callchain(struct perf_session *self, + struct thread *thread, + struct ip_callchain *chain, + struct symbol **parent) { u8 cpumode = PERF_RECORD_MISC_USER; unsigned int i; - struct map_symbol *syms = calloc(chain->nr, sizeof(*syms)); + int err; - if (!syms) - return NULL; + callchain_cursor_reset(&self->callchain_cursor); for (i = 0; i < chain->nr; i++) { u64 ip = chain->ips[i]; @@ -281,12 +280,15 @@ struct map_symbol *perf_session__resolve_callchain(struct perf_session *self, *parent = al.sym; if (!symbol_conf.use_callchain) break; - syms[i].map = al.map; - syms[i].sym = al.sym; } + + err = callchain_cursor_append(&self->callchain_cursor, + ip, al.map, al.sym); + if (err) + return err; } - return syms; + return 0; } static int process_event_synth_stub(event_t *event __used, diff --git a/tools/perf/util/session.h b/tools/perf/util/session.h index decd83f274fd..e815468eb888 100644 --- a/tools/perf/util/session.h +++ b/tools/perf/util/session.h @@ -51,7 +51,8 @@ struct perf_session { int cwdlen; char *cwd; struct ordered_samples ordered_samples; - char filename[0]; + struct callchain_cursor callchain_cursor; + char filename[0]; }; struct perf_event_ops; @@ -94,10 +95,10 @@ int __perf_session__process_events(struct perf_session *self, int perf_session__process_events(struct perf_session *self, struct perf_event_ops *event_ops); -struct map_symbol *perf_session__resolve_callchain(struct perf_session *self, - struct thread *thread, - struct ip_callchain *chain, - struct symbol **parent); +int perf_session__resolve_callchain(struct perf_session *self, + struct thread *thread, + struct ip_callchain *chain, + struct symbol **parent); bool perf_session__has_traces(struct perf_session *self, const char *msg); From f08c3154ac439c4b5762a40107d84e839e08fbc5 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Fri, 14 Jan 2011 04:51:59 +0100 Subject: [PATCH 18/29] perf callchain: Rename cumul_hits into callchain_cumul_hits That makes the callchain API naming more consistent and reduce potential naming clashes. Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <1294977121-5700-3-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/callchain.c | 10 +++++----- tools/perf/util/callchain.h | 2 +- tools/perf/util/hist.c | 2 +- tools/perf/util/ui/browsers/hists.c | 2 +- 4 files changed, 8 insertions(+), 8 deletions(-) diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c index 53a49e0cfc6c..4c6360fc2d5b 100644 --- a/tools/perf/util/callchain.c +++ b/tools/perf/util/callchain.c @@ -38,14 +38,14 @@ rb_insert_callchain(struct rb_root *root, struct callchain_node *chain, struct rb_node **p = &root->rb_node; struct rb_node *parent = NULL; struct callchain_node *rnode; - u64 chain_cumul = cumul_hits(chain); + u64 chain_cumul = callchain_cumul_hits(chain); while (*p) { u64 rnode_cumul; parent = *p; rnode = rb_entry(parent, struct callchain_node, rb_node); - rnode_cumul = cumul_hits(rnode); + rnode_cumul = callchain_cumul_hits(rnode); switch (mode) { case CHAIN_FLAT: @@ -104,7 +104,7 @@ static void __sort_chain_graph_abs(struct callchain_node *node, chain_for_each_child(child, node) { __sort_chain_graph_abs(child, min_hit); - if (cumul_hits(child) >= min_hit) + if (callchain_cumul_hits(child) >= min_hit) rb_insert_callchain(&node->rb_root, child, CHAIN_GRAPH_ABS); } @@ -129,7 +129,7 @@ static void __sort_chain_graph_rel(struct callchain_node *node, chain_for_each_child(child, node) { __sort_chain_graph_rel(child, min_percent); - if (cumul_hits(child) >= min_hit) + if (callchain_cumul_hits(child) >= min_hit) rb_insert_callchain(&node->rb_root, child, CHAIN_GRAPH_REL); } @@ -270,7 +270,7 @@ split_add_child(struct callchain_node *parent, /* split the hits */ new->hit = parent->hit; new->children_hit = parent->children_hit; - parent->children_hit = cumul_hits(new); + parent->children_hit = callchain_cumul_hits(new); new->val_nr = parent->val_nr - idx_local; parent->val_nr = idx_local; diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h index d74a19af4a44..07f71e3e0a71 100644 --- a/tools/perf/util/callchain.h +++ b/tools/perf/util/callchain.h @@ -82,7 +82,7 @@ static inline void callchain_init(struct callchain_root *root) root->max_depth = 0; } -static inline u64 cumul_hits(struct callchain_node *node) +static inline u64 callchain_cumul_hits(struct callchain_node *node) { return node->hit + node->children_hit; } diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c index a438a0652d23..02ed318d7312 100644 --- a/tools/perf/util/hist.c +++ b/tools/perf/util/hist.c @@ -430,7 +430,7 @@ static size_t __callchain__fprintf_graph(FILE *fp, struct callchain_node *self, u64 cumul; child = rb_entry(node, struct callchain_node, rb_node); - cumul = cumul_hits(child); + cumul = callchain_cumul_hits(child); remaining -= cumul; /* diff --git a/tools/perf/util/ui/browsers/hists.c b/tools/perf/util/ui/browsers/hists.c index 60c463c16028..86428239fa65 100644 --- a/tools/perf/util/ui/browsers/hists.c +++ b/tools/perf/util/ui/browsers/hists.c @@ -377,7 +377,7 @@ static int hist_browser__show_callchain_node_rb_tree(struct hist_browser *self, while (node) { struct callchain_node *child = rb_entry(node, struct callchain_node, rb_node); struct rb_node *next = rb_next(node); - u64 cumul = cumul_hits(child); + u64 cumul = callchain_cumul_hits(child); struct callchain_list *chain; char folded_sign = ' '; int first = true; From 16537f1355017a285b904bfb6bf767464293e69c Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Fri, 14 Jan 2011 04:52:00 +0100 Subject: [PATCH 19/29] perf callchain: Rename register_callchain_param into callchain_register_param To make the callchain API naming more consistent. Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <1294977121-5700-4-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-report.c | 4 ++-- tools/perf/util/callchain.c | 2 +- tools/perf/util/callchain.h | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-) diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c index c95599a82f9e..f6a43493d1d0 100644 --- a/tools/perf/builtin-report.c +++ b/tools/perf/builtin-report.c @@ -221,7 +221,7 @@ static int perf_session__setup_sample_type(struct perf_session *self) } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE && !symbol_conf.use_callchain) { symbol_conf.use_callchain = true; - if (register_callchain_param(&callchain_param) < 0) { + if (callchain_register_param(&callchain_param) < 0) { fprintf(stderr, "Can't register callchain" " params\n"); return -EINVAL; @@ -423,7 +423,7 @@ parse_callchain_opt(const struct option *opt __used, const char *arg, if (tok2) callchain_param.print_limit = strtod(tok2, &endptr); setup: - if (register_callchain_param(&callchain_param) < 0) { + if (callchain_register_param(&callchain_param) < 0) { fprintf(stderr, "Can't register callchain params\n"); return -1; } diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c index 4c6360fc2d5b..5b3f09dd137d 100644 --- a/tools/perf/util/callchain.c +++ b/tools/perf/util/callchain.c @@ -143,7 +143,7 @@ sort_chain_graph_rel(struct rb_root *rb_root, struct callchain_root *chain_root, rb_root->rb_node = chain_root->node.rb_root.rb_node; } -int register_callchain_param(struct callchain_param *param) +int callchain_register_param(struct callchain_param *param) { switch (param->mode) { case CHAIN_GRAPH_ABS: diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h index 07f71e3e0a71..2bb5403010d2 100644 --- a/tools/perf/util/callchain.h +++ b/tools/perf/util/callchain.h @@ -87,7 +87,7 @@ static inline u64 callchain_cumul_hits(struct callchain_node *node) return node->hit + node->children_hit; } -int register_callchain_param(struct callchain_param *param); +int callchain_register_param(struct callchain_param *param); int callchain_append(struct callchain_root *root, struct callchain_cursor *cursor, u64 period); From 529363b76929beb85b81439c61063130af046a21 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Fri, 14 Jan 2011 04:52:01 +0100 Subject: [PATCH 20/29] perf callchain: Don't give arbitrary gender to callchain tree nodes Some little callchain tree nodes shyly asked me if they can have sisters. How cute! Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <1294977121-5700-5-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/callchain.c | 8 ++++---- tools/perf/util/callchain.h | 4 ++-- 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c index 5b3f09dd137d..f8c66d1435e0 100644 --- a/tools/perf/util/callchain.c +++ b/tools/perf/util/callchain.c @@ -26,10 +26,10 @@ bool ip_callchain__valid(struct ip_callchain *chain, const event_t *event) } #define chain_for_each_child(child, parent) \ - list_for_each_entry(child, &parent->children, brothers) + list_for_each_entry(child, &parent->children, siblings) #define chain_for_each_child_safe(child, next, parent) \ - list_for_each_entry_safe(child, next, &parent->children, brothers) + list_for_each_entry_safe(child, next, &parent->children, siblings) static void rb_insert_callchain(struct rb_root *root, struct callchain_node *chain, @@ -189,7 +189,7 @@ create_child(struct callchain_node *parent, bool inherit_children) chain_for_each_child(next, new) next->parent = new; } - list_add_tail(&new->brothers, &parent->children); + list_add_tail(&new->siblings, &parent->children); return new; } @@ -419,7 +419,7 @@ merge_chain_branch(struct callchain_cursor *cursor, if (err) break; - list_del(&child->brothers); + list_del(&child->siblings); free(child); } diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h index 2bb5403010d2..67137256a1cd 100644 --- a/tools/perf/util/callchain.h +++ b/tools/perf/util/callchain.h @@ -16,7 +16,7 @@ enum chain_mode { struct callchain_node { struct callchain_node *parent; - struct list_head brothers; + struct list_head siblings; struct list_head children; struct list_head val; struct rb_node rb_node; /* to sort nodes in an rbtree */ @@ -72,7 +72,7 @@ struct callchain_cursor { static inline void callchain_init(struct callchain_root *root) { - INIT_LIST_HEAD(&root->node.brothers); + INIT_LIST_HEAD(&root->node.siblings); INIT_LIST_HEAD(&root->node.children); INIT_LIST_HEAD(&root->node.val); From b0e8572f3b29c0760b66ba5627a6d5426c88c97d Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Sun, 16 Jan 2011 17:39:15 -0200 Subject: [PATCH 21/29] perf top: Add native_safe_halt to skip symbols Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-top.c | 1 + 1 file changed, 1 insertion(+) diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 58352ad807c7..31fbaf38d9c1 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -933,6 +933,7 @@ repeat: /* Tag samples to be skipped. */ static const char *skip_symbols[] = { "default_idle", + "native_safe_halt", "cpu_idle", "enter_idle", "exit_idle", From 4cc9cec636e7f78aba7f17606ac13cac07ea5787 Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Thu, 13 Jan 2011 21:45:58 +0900 Subject: [PATCH 22/29] perf probe: Introduce lines walker interface Introduce die_walk_lines() for walking on the line list of given die, and use it in line_range finder and probe point finder. Cc: 2nddept-manager@sdl.hitachi.co.jp Cc: Franck Bui-Huu Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Srikar Dronamraju Cc: Steven Rostedt LKML-Reference: <20110113124558.22426.48170.stgit@ltc236.sdl.hitachi.co.jp> Signed-off-by: Masami Hiramatsu [ committer note: s/%ld/%zd/ for a size_t nlines var that broke f14 x86 build] Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/probe-finder.c | 325 +++++++++++++++++---------------- 1 file changed, 169 insertions(+), 156 deletions(-) diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c index ab83b6ac5d65..508c017f566a 100644 --- a/tools/perf/util/probe-finder.c +++ b/tools/perf/util/probe-finder.c @@ -458,6 +458,124 @@ static Dwarf_Die *die_find_inlinefunc(Dwarf_Die *sp_die, Dwarf_Addr addr, return die_find_child(sp_die, __die_find_inline_cb, &addr, die_mem); } +/* Walker on lines (Note: line number will not be sorted) */ +typedef int (* line_walk_handler_t) (const char *fname, int lineno, + Dwarf_Addr addr, void *data); + +struct __line_walk_param { + line_walk_handler_t handler; + void *data; + int retval; +}; + +/* Walk on decl lines in given DIE */ +static int __die_walk_funclines(Dwarf_Die *sp_die, + line_walk_handler_t handler, void *data) +{ + const char *fname; + Dwarf_Addr addr; + int lineno, ret = 0; + + /* Handle function declaration line */ + fname = dwarf_decl_file(sp_die); + if (fname && dwarf_decl_line(sp_die, &lineno) == 0 && + dwarf_entrypc(sp_die, &addr) == 0) { + ret = handler(fname, lineno, addr, data); + } + + return ret; +} + +static int __die_walk_culines_cb(Dwarf_Die *sp_die, void *data) +{ + struct __line_walk_param *lw = data; + + lw->retval = __die_walk_funclines(sp_die, lw->handler, lw->data); + if (lw->retval != 0) + return DWARF_CB_ABORT; + + return DWARF_CB_OK; +} + +/* + * Walk on lines inside given PDIE. If the PDIE is subprogram, walk only on + * the lines inside the subprogram, otherwise PDIE must be a CU DIE. + */ +static int die_walk_lines(Dwarf_Die *pdie, line_walk_handler_t handler, + void *data) +{ + Dwarf_Lines *lines; + Dwarf_Line *line; + Dwarf_Addr addr; + const char *fname; + int lineno, ret = 0; + Dwarf_Die die_mem, *cu_die; + size_t nlines, i; + + /* Get the CU die */ + if (dwarf_tag(pdie) == DW_TAG_subprogram) + cu_die = dwarf_diecu(pdie, &die_mem, NULL, NULL); + else + cu_die = pdie; + if (!cu_die) { + pr_debug2("Failed to get CU from subprogram\n"); + return -EINVAL; + } + + /* Get lines list in the CU */ + if (dwarf_getsrclines(cu_die, &lines, &nlines) != 0) { + pr_debug2("Failed to get source lines on this CU.\n"); + return -ENOENT; + } + pr_debug2("Get %zd lines from this CU\n", nlines); + + /* Walk on the lines on lines list */ + for (i = 0; i < nlines; i++) { + line = dwarf_onesrcline(lines, i); + if (line == NULL || + dwarf_lineno(line, &lineno) != 0 || + dwarf_lineaddr(line, &addr) != 0) { + pr_debug2("Failed to get line info. " + "Possible error in debuginfo.\n"); + continue; + } + /* Filter lines based on address */ + if (pdie != cu_die) + /* + * Address filtering + * The line is included in given function, and + * no inline block includes it. + */ + if (!dwarf_haspc(pdie, addr) || + die_find_inlinefunc(pdie, addr, &die_mem)) + continue; + /* Get source line */ + fname = dwarf_linesrc(line, NULL, NULL); + + ret = handler(fname, lineno, addr, data); + if (ret != 0) + return ret; + } + + /* + * Dwarf lines doesn't include function declarations and inlined + * subroutines. We have to check functions list or given function. + */ + if (pdie != cu_die) + ret = __die_walk_funclines(pdie, handler, data); + else { + struct __line_walk_param param = { + .handler = handler, + .data = data, + .retval = 0, + }; + dwarf_getfuncs(cu_die, __die_walk_culines_cb, ¶m, 0); + ret = param.retval; + } + + return ret; +} + struct __find_variable_param { const char *name; Dwarf_Addr addr; @@ -1050,43 +1168,26 @@ static int call_probe_finder(Dwarf_Die *sp_die, struct probe_finder *pf) return ret; } +static int probe_point_line_walker(const char *fname, int lineno, + Dwarf_Addr addr, void *data) +{ + struct probe_finder *pf = data; + int ret; + + if (lineno != pf->lno || strtailcmp(fname, pf->fname) != 0) + return 0; + + pf->addr = addr; + ret = call_probe_finder(NULL, pf); + + /* Continue if no error, because the line will be in inline function */ + return ret < 0 ?: 0; +} + /* Find probe point from its line number */ static int find_probe_point_by_line(struct probe_finder *pf) { - Dwarf_Lines *lines; - Dwarf_Line *line; - size_t nlines, i; - Dwarf_Addr addr; - int lineno; - int ret = 0; - - if (dwarf_getsrclines(&pf->cu_die, &lines, &nlines) != 0) { - pr_warning("No source lines found.\n"); - return -ENOENT; - } - - for (i = 0; i < nlines && ret == 0; i++) { - line = dwarf_onesrcline(lines, i); - if (dwarf_lineno(line, &lineno) != 0 || - lineno != pf->lno) - continue; - - /* TODO: Get fileno from line, but how? */ - if (strtailcmp(dwarf_linesrc(line, NULL, NULL), pf->fname) != 0) - continue; - - if (dwarf_lineaddr(line, &addr) != 0) { - pr_warning("Failed to get the address of the line.\n"); - return -ENOENT; - } - pr_debug("Probe line found: line[%d]:%d addr:0x%jx\n", - (int)i, lineno, (uintmax_t)addr); - pf->addr = addr; - - ret = call_probe_finder(NULL, pf); - /* Continuing, because target line might be inlined. */ - } - return ret; + return die_walk_lines(&pf->cu_die, probe_point_line_walker, pf); } /* Find lines which match lazy pattern */ @@ -1140,15 +1241,31 @@ out_close: return nlines; } +static int probe_point_lazy_walker(const char *fname, int lineno, + Dwarf_Addr addr, void *data) +{ + struct probe_finder *pf = data; + int ret; + + if (!line_list__has_line(&pf->lcache, lineno) || + strtailcmp(fname, pf->fname) != 0) + return 0; + + pr_debug("Probe line found: line:%d addr:0x%llx\n", + lineno, (unsigned long long)addr); + pf->addr = addr; + ret = call_probe_finder(NULL, pf); + + /* + * Continue if no error, because the lazy pattern will match + * to other lines + */ + return ret < 0 ?: 0; +} + /* Find probe points from lazy pattern */ static int find_probe_point_lazy(Dwarf_Die *sp_die, struct probe_finder *pf) { - Dwarf_Lines *lines; - Dwarf_Line *line; - size_t nlines, i; - Dwarf_Addr addr; - Dwarf_Die die_mem; - int lineno; int ret = 0; if (list_empty(&pf->lcache)) { @@ -1162,45 +1279,7 @@ static int find_probe_point_lazy(Dwarf_Die *sp_die, struct probe_finder *pf) return ret; } - if (dwarf_getsrclines(&pf->cu_die, &lines, &nlines) != 0) { - pr_warning("No source lines found.\n"); - return -ENOENT; - } - - for (i = 0; i < nlines && ret >= 0; i++) { - line = dwarf_onesrcline(lines, i); - - if (dwarf_lineno(line, &lineno) != 0 || - !line_list__has_line(&pf->lcache, lineno)) - continue; - - /* TODO: Get fileno from line, but how? */ - if (strtailcmp(dwarf_linesrc(line, NULL, NULL), pf->fname) != 0) - continue; - - if (dwarf_lineaddr(line, &addr) != 0) { - pr_debug("Failed to get the address of line %d.\n", - lineno); - continue; - } - if (sp_die) { - /* Address filtering 1: does sp_die include addr? */ - if (!dwarf_haspc(sp_die, addr)) - continue; - /* Address filtering 2: No child include addr? */ - if (die_find_inlinefunc(sp_die, addr, &die_mem)) - continue; - } - - pr_debug("Probe line found: line[%d]:%d addr:0x%llx\n", - (int)i, lineno, (unsigned long long)addr); - pf->addr = addr; - - ret = call_probe_finder(sp_die, pf); - /* Continuing, because target line might be inlined. */ - } - /* TODO: deallocate lines, but how? */ - return ret; + return die_walk_lines(sp_die, probe_point_lazy_walker, pf); } /* Callback parameter with return value */ @@ -1644,91 +1723,28 @@ static int line_range_add_line(const char *src, unsigned int lineno, return line_list__add_line(&lr->line_list, lineno); } -/* Search function declaration lines */ -static int line_range_funcdecl_cb(Dwarf_Die *sp_die, void *data) +static int line_range_walk_cb(const char *fname, int lineno, + Dwarf_Addr addr __used, + void *data) { - struct dwarf_callback_param *param = data; - struct line_finder *lf = param->data; - const char *src; - int lineno; + struct line_finder *lf = data; - src = dwarf_decl_file(sp_die); - if (src && strtailcmp(src, lf->fname) != 0) - return DWARF_CB_OK; - - if (dwarf_decl_line(sp_die, &lineno) != 0 || + if ((strtailcmp(fname, lf->fname) != 0) || (lf->lno_s > lineno || lf->lno_e < lineno)) - return DWARF_CB_OK; + return 0; - param->retval = line_range_add_line(src, lineno, lf->lr); - if (param->retval < 0) - return DWARF_CB_ABORT; - return DWARF_CB_OK; -} + if (line_range_add_line(fname, lineno, lf->lr) < 0) + return -EINVAL; -static int find_line_range_func_decl_lines(struct line_finder *lf) -{ - struct dwarf_callback_param param = {.data = (void *)lf, .retval = 0}; - dwarf_getfuncs(&lf->cu_die, line_range_funcdecl_cb, ¶m, 0); - return param.retval; + return 0; } /* Find line range from its line number */ static int find_line_range_by_line(Dwarf_Die *sp_die, struct line_finder *lf) { - Dwarf_Lines *lines; - Dwarf_Line *line; - size_t nlines, i; - Dwarf_Addr addr; - int lineno, ret = 0; - const char *src; - Dwarf_Die die_mem; + int ret; - line_list__init(&lf->lr->line_list); - if (dwarf_getsrclines(&lf->cu_die, &lines, &nlines) != 0) { - pr_warning("No source lines found.\n"); - return -ENOENT; - } - - /* Search probable lines on lines list */ - for (i = 0; i < nlines; i++) { - line = dwarf_onesrcline(lines, i); - if (dwarf_lineno(line, &lineno) != 0 || - (lf->lno_s > lineno || lf->lno_e < lineno)) - continue; - - if (sp_die) { - /* Address filtering 1: does sp_die include addr? */ - if (dwarf_lineaddr(line, &addr) != 0 || - !dwarf_haspc(sp_die, addr)) - continue; - - /* Address filtering 2: No child include addr? */ - if (die_find_inlinefunc(sp_die, addr, &die_mem)) - continue; - } - - /* TODO: Get fileno from line, but how? */ - src = dwarf_linesrc(line, NULL, NULL); - if (strtailcmp(src, lf->fname) != 0) - continue; - - ret = line_range_add_line(src, lineno, lf->lr); - if (ret < 0) - return ret; - } - - /* - * Dwarf lines doesn't include function declarations. We have to - * check functions list or given function. - */ - if (sp_die) { - src = dwarf_decl_file(sp_die); - if (src && dwarf_decl_line(sp_die, &lineno) == 0 && - (lf->lno_s <= lineno && lf->lno_e >= lineno)) - ret = line_range_add_line(src, lineno, lf->lr); - } else - ret = find_line_range_func_decl_lines(lf); + ret = die_walk_lines(sp_die ?: &lf->cu_die, line_range_walk_cb, lf); /* Update status */ if (ret >= 0) @@ -1758,9 +1774,6 @@ static int line_range_search_cb(Dwarf_Die *sp_die, void *data) struct line_finder *lf = param->data; struct line_range *lr = lf->lr; - pr_debug("find (%llx) %s\n", - (unsigned long long)dwarf_dieoffset(sp_die), - dwarf_diename(sp_die)); if (dwarf_tag(sp_die) == DW_TAG_subprogram && die_compare_name(sp_die, lr->function)) { lf->fname = dwarf_decl_file(sp_die); From 5069ed86be3c2f28bcdf7fae1374ec0c325aafba Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Thu, 13 Jan 2011 21:46:05 +0900 Subject: [PATCH 23/29] perf probe: Enable to put probe inline function call site Enable to put probe inline function call site. This will increase line-based probe-ability. $ ./perf probe -L schedule:48 pre_schedule(rq, prev); 50 if (unlikely(!rq->nr_running)) idle_balance(cpu, rq); put_prev_task(rq, prev); next = pick_next_task(rq); 56 if (likely(prev != next)) { sched_info_switch(prev, next); trace_sched_switch_out(prev, next); perf_event_task_sched_out(prev, next); $ ./perf probe -L schedule:48 48 pre_schedule(rq, prev); 50 if (unlikely(!rq->nr_running)) 51 idle_balance(cpu, rq); 53 put_prev_task(rq, prev); 54 next = pick_next_task(rq); 56 if (likely(prev != next)) { 57 sched_info_switch(prev, next); 58 trace_sched_switch_out(prev, next); 59 perf_event_task_sched_out(prev, next); Cc: 2nddept-manager@sdl.hitachi.co.jp Cc: Franck Bui-Huu Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Srikar Dronamraju Cc: Steven Rostedt LKML-Reference: <20110113124604.22426.48873.stgit@ltc236.sdl.hitachi.co.jp> Signed-off-by: Masami Hiramatsu Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/probe-finder.c | 56 +++++++++++++++++++++++++++++----- 1 file changed, 48 insertions(+), 8 deletions(-) diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c index 508c017f566a..69215bff17e9 100644 --- a/tools/perf/util/probe-finder.c +++ b/tools/perf/util/probe-finder.c @@ -280,6 +280,19 @@ static bool die_compare_name(Dwarf_Die *dw_die, const char *tname) return name ? (strcmp(tname, name) == 0) : false; } +/* Get callsite line number of inline-function instance */ +static int die_get_call_lineno(Dwarf_Die *in_die) +{ + Dwarf_Attribute attr; + Dwarf_Word ret; + + if (!dwarf_attr(in_die, DW_AT_call_line, &attr)) + return -ENOENT; + + dwarf_formudata(&attr, &ret); + return (int)ret; +} + /* Get type die */ static Dwarf_Die *die_get_type(Dwarf_Die *vr_die, Dwarf_Die *die_mem) { @@ -463,27 +476,54 @@ typedef int (* line_walk_handler_t) (const char *fname, int lineno, Dwarf_Addr addr, void *data); struct __line_walk_param { + const char *fname; line_walk_handler_t handler; void *data; int retval; }; -/* Walk on decl lines in given DIE */ +static int __die_walk_funclines_cb(Dwarf_Die *in_die, void *data) +{ + struct __line_walk_param *lw = data; + Dwarf_Addr addr; + int lineno; + + if (dwarf_tag(in_die) == DW_TAG_inlined_subroutine) { + lineno = die_get_call_lineno(in_die); + if (lineno > 0 && dwarf_entrypc(in_die, &addr) == 0) { + lw->retval = lw->handler(lw->fname, lineno, addr, + lw->data); + if (lw->retval != 0) + return DIE_FIND_CB_FOUND; + } + } + return DIE_FIND_CB_SIBLING; +} + +/* Walk on lines of blocks included in given DIE */ static int __die_walk_funclines(Dwarf_Die *sp_die, line_walk_handler_t handler, void *data) { - const char *fname; + struct __line_walk_param lw = { + .handler = handler, + .data = data, + .retval = 0, + }; + Dwarf_Die die_mem; Dwarf_Addr addr; - int lineno, ret = 0; + int lineno; /* Handle function declaration line */ - fname = dwarf_decl_file(sp_die); - if (fname && dwarf_decl_line(sp_die, &lineno) == 0 && + lw.fname = dwarf_decl_file(sp_die); + if (lw.fname && dwarf_decl_line(sp_die, &lineno) == 0 && dwarf_entrypc(sp_die, &addr) == 0) { - ret = handler(fname, lineno, addr, data); + lw.retval = handler(lw.fname, lineno, addr, data); + if (lw.retval != 0) + goto done; } - - return ret; + die_find_child(sp_die, __die_walk_funclines_cb, &lw, &die_mem); +done: + return lw.retval; } static int __die_walk_culines_cb(Dwarf_Die *sp_die, void *data) From e80711ca8512c8586da0c3e18e2f1caf73c88731 Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Thu, 13 Jan 2011 21:46:11 +0900 Subject: [PATCH 24/29] perf probe: Add --funcs to show available functions in symtab Add --funcs to show available functions in symtab. Originally this feature came from Srikar's uprobes patches ( http://lkml.org/lkml/2010/8/27/244 ) e.g. ... __ablkcipher_walk_complete __absent_pages_in_range __account_scheduler_latency __add_pages __alloc_pages_nodemask __alloc_percpu __alloc_reserved_percpu __alloc_skb __alloc_workqueue_key __any_online_cpu __ata_ehi_push_desc ... This also supports symbols in module, e.g. ... cleanup_module cpuid_maxphyaddr emulate_clts emulate_instruction emulate_int_real emulate_invlpg emulator_get_dr emulator_set_dr emulator_task_switch emulator_write_emulated emulator_write_phys fx_init ... Original-patch-from: Srikar Dronamraju Cc: 2nddept-manager@sdl.hitachi.co.jp Cc: Franck Bui-Huu Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Srikar Dronamraju Cc: Steven Rostedt LKML-Reference: <20110113124611.22426.10835.stgit@ltc236.sdl.hitachi.co.jp> Signed-off-by: Masami Hiramatsu [ committer note: Add missing elf.h for STB_GLOBAL that broke a RHEL4 build ] Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Documentation/perf-probe.txt | 4 ++ tools/perf/builtin-probe.c | 29 ++++++++++- tools/perf/util/probe-event.c | 68 ++++++++++++++++++++++++- tools/perf/util/probe-event.h | 1 + 4 files changed, 99 insertions(+), 3 deletions(-) diff --git a/tools/perf/Documentation/perf-probe.txt b/tools/perf/Documentation/perf-probe.txt index 86b797a35aa6..fcc51fe0195c 100644 --- a/tools/perf/Documentation/perf-probe.txt +++ b/tools/perf/Documentation/perf-probe.txt @@ -73,6 +73,10 @@ OPTIONS (Only for --vars) Show external defined variables in addition to local variables. +-F:: +--funcs:: + Show available functions in given module or kernel. + -f:: --force:: Forcibly add events with existing name. diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c index add163c9f0e7..6cf708aba7c9 100644 --- a/tools/perf/builtin-probe.c +++ b/tools/perf/builtin-probe.c @@ -52,6 +52,7 @@ static struct { bool show_lines; bool show_vars; bool show_ext_vars; + bool show_funcs; bool mod_events; int nevents; struct perf_probe_event events[MAX_PROBES]; @@ -221,6 +222,8 @@ static const struct option options[] = { OPT__DRY_RUN(&probe_event_dry_run), OPT_INTEGER('\0', "max-probes", ¶ms.max_probe_points, "Set how many probe points can be found for a probe."), + OPT_BOOLEAN('F', "funcs", ¶ms.show_funcs, + "Show potential probe-able functions."), OPT_END() }; @@ -246,7 +249,7 @@ int cmd_probe(int argc, const char **argv, const char *prefix __used) params.max_probe_points = MAX_PROBES; if ((!params.nevents && !params.dellist && !params.list_events && - !params.show_lines)) + !params.show_lines && !params.show_funcs)) usage_with_options(probe_usage, options); /* @@ -267,12 +270,36 @@ int cmd_probe(int argc, const char **argv, const char *prefix __used) pr_err(" Error: Don't use --list with --vars.\n"); usage_with_options(probe_usage, options); } + if (params.show_funcs) { + pr_err(" Error: Don't use --list with --funcs.\n"); + usage_with_options(probe_usage, options); + } ret = show_perf_probe_events(); if (ret < 0) pr_err(" Error: Failed to show event list. (%d)\n", ret); return ret; } + if (params.show_funcs) { + if (params.nevents != 0 || params.dellist) { + pr_err(" Error: Don't use --funcs with" + " --add/--del.\n"); + usage_with_options(probe_usage, options); + } + if (params.show_lines) { + pr_err(" Error: Don't use --funcs with --line.\n"); + usage_with_options(probe_usage, options); + } + if (params.show_vars) { + pr_err(" Error: Don't use --funcs with --vars.\n"); + usage_with_options(probe_usage, options); + } + ret = show_available_funcs(params.target_module); + if (ret < 0) + pr_err(" Error: Failed to show functions." + " (%d)\n", ret); + return ret; + } #ifdef DWARF_SUPPORT if (params.show_lines) { diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c index 6e29d9c9dccc..859d377a3df3 100644 --- a/tools/perf/util/probe-event.c +++ b/tools/perf/util/probe-event.c @@ -31,6 +31,7 @@ #include #include #include +#include #undef _GNU_SOURCE #include "util.h" @@ -111,7 +112,25 @@ static struct symbol *__find_kernel_function_by_name(const char *name, NULL); } -const char *kernel_get_module_path(const char *module) +static struct map *kernel_get_module_map(const char *module) +{ + struct rb_node *nd; + struct map_groups *grp = &machine.kmaps; + + if (!module) + module = "kernel"; + + for (nd = rb_first(&grp->maps[MAP__FUNCTION]); nd; nd = rb_next(nd)) { + struct map *pos = rb_entry(nd, struct map, rb_node); + if (strncmp(pos->dso->short_name + 1, module, + pos->dso->short_name_len - 2) == 0) { + return pos; + } + } + return NULL; +} + +static struct dso *kernel_get_module_dso(const char *module) { struct dso *dso; struct map *map; @@ -141,7 +160,13 @@ const char *kernel_get_module_path(const char *module) } } found: - return dso->long_name; + return dso; +} + +const char *kernel_get_module_path(const char *module) +{ + struct dso *dso = kernel_get_module_dso(module); + return (dso) ? dso->long_name : NULL; } #ifdef DWARF_SUPPORT @@ -1913,3 +1938,42 @@ int del_perf_probe_events(struct strlist *dellist) return ret; } +/* + * If a symbol corresponds to a function with global binding return 0. + * For all others return 1. + */ +static int filter_non_global_functions(struct map *map __unused, + struct symbol *sym) +{ + if (sym->binding != STB_GLOBAL) + return 1; + + return 0; +} + +int show_available_funcs(const char *module) +{ + struct map *map; + int ret; + + setup_pager(); + + ret = init_vmlinux(); + if (ret < 0) + return ret; + + map = kernel_get_module_map(module); + if (!map) { + pr_err("Failed to find %s map.\n", (module) ? : "kernel"); + return -EINVAL; + } + if (map__load(map, filter_non_global_functions)) { + pr_err("Failed to load map.\n"); + return -EINVAL; + } + if (!dso__sorted_by_name(map->dso, map->type)) + dso__sort_by_name(map->dso, map->type); + + dso__fprintf_symbols_by_name(map->dso, map->type, stdout); + return 0; +} diff --git a/tools/perf/util/probe-event.h b/tools/perf/util/probe-event.h index 5accbedfea37..1fb4f18337d3 100644 --- a/tools/perf/util/probe-event.h +++ b/tools/perf/util/probe-event.h @@ -127,6 +127,7 @@ extern int show_line_range(struct line_range *lr, const char *module); extern int show_available_vars(struct perf_probe_event *pevs, int npevs, int max_probe_points, const char *module, bool externs); +extern int show_available_funcs(const char *module); /* Maximum index number of event-name postfix */ From d7065adb9b4f3384c2615f0a3dbdb6c3aae1eb18 Mon Sep 17 00:00:00 2001 From: Franck Bui-Huu Date: Sun, 16 Jan 2011 17:14:45 +0100 Subject: [PATCH 25/29] perf record: auto detect when stdout is a pipe This patch gives the ability to 'perf record' to detect when its stdout has been redirected to a pipe. There's now no more need to add '-o -' switch in this case. However '-o ' option has always precedence, that is if specified and stdout has been connected via a pipe then the output will go into the specified output. LKML-Reference: Signed-off-by: Franck Bui-Huu Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 32 ++++++++++++++++++++------------ 1 file changed, 20 insertions(+), 12 deletions(-) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 45a3689f9ed6..1346d4230bc0 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -48,7 +48,7 @@ static unsigned int user_freq = UINT_MAX; static int freq = 1000; static int output; static int pipe_output = 0; -static const char *output_name = "perf.data"; +static const char *output_name = NULL; static int group = 0; static int realtime_prio = 0; static bool nodelay = false; @@ -497,18 +497,26 @@ static int __cmd_record(int argc, const char **argv) exit(-1); } - if (!strcmp(output_name, "-")) - pipe_output = 1; - else if (!stat(output_name, &st) && st.st_size) { - if (write_mode == WRITE_FORCE) { - char oldname[PATH_MAX]; - snprintf(oldname, sizeof(oldname), "%s.old", - output_name); - unlink(oldname); - rename(output_name, oldname); + if (!output_name) { + if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode)) + pipe_output = 1; + else + output_name = "perf.data"; + } + if (output_name) { + if (!strcmp(output_name, "-")) + pipe_output = 1; + else if (!stat(output_name, &st) && st.st_size) { + if (write_mode == WRITE_FORCE) { + char oldname[PATH_MAX]; + snprintf(oldname, sizeof(oldname), "%s.old", + output_name); + unlink(oldname); + rename(output_name, oldname); + } + } else if (write_mode == WRITE_APPEND) { + write_mode = WRITE_FORCE; } - } else if (write_mode == WRITE_APPEND) { - write_mode = WRITE_FORCE; } flags = O_CREAT|O_RDWR; From 17ea1b70a87e28457821318341bead2b45563092 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Mon, 17 Jan 2011 14:40:46 -0200 Subject: [PATCH 26/29] perf tools: Pass the struct opt to the wildcard parsing routine It is needed because it will call parse_event for each tracepoint name that matches, and we pass the perf_evlist via opt->value. Problem introduced in 4503fdd where my assumption about opt being always non NULL made me not look at callers of parse_events outside builtin-*.c. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/parse-events.c | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-) diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index d3086cecd2dd..cf082daa43e3 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -446,8 +446,8 @@ parse_single_tracepoint_event(char *sys_name, /* sys + ':' + event + ':' + flags*/ #define MAX_EVOPT_LEN (MAX_EVENT_LENGTH * 2 + 2 + 128) static enum event_result -parse_multiple_tracepoint_event(char *sys_name, const char *evt_exp, - char *flags) +parse_multiple_tracepoint_event(const struct option *opt, char *sys_name, + const char *evt_exp, char *flags) { char evt_path[MAXPATHLEN]; struct dirent *evt_ent; @@ -480,15 +480,16 @@ parse_multiple_tracepoint_event(char *sys_name, const char *evt_exp, if (len < 0) return EVT_FAILED; - if (parse_events(NULL, event_opt, 0)) + if (parse_events(opt, event_opt, 0)) return EVT_FAILED; } return EVT_HANDLED_ALL; } -static enum event_result parse_tracepoint_event(const char **strp, - struct perf_event_attr *attr) +static enum event_result +parse_tracepoint_event(const struct option *opt, const char **strp, + struct perf_event_attr *attr) { const char *evt_name; char *flags = NULL, *comma_loc; @@ -527,7 +528,7 @@ static enum event_result parse_tracepoint_event(const char **strp, return EVT_FAILED; if (strpbrk(evt_name, "*?")) { *strp += strlen(sys_name) + evt_length + 1; /* 1 == the ':' */ - return parse_multiple_tracepoint_event(sys_name, evt_name, + return parse_multiple_tracepoint_event(opt, sys_name, evt_name, flags); } else { return parse_single_tracepoint_event(sys_name, evt_name, @@ -737,11 +738,12 @@ parse_event_modifier(const char **strp, struct perf_event_attr *attr) * Symbolic names are (almost) exactly matched. */ static enum event_result -parse_event_symbols(const char **str, struct perf_event_attr *attr) +parse_event_symbols(const struct option *opt, const char **str, + struct perf_event_attr *attr) { enum event_result ret; - ret = parse_tracepoint_event(str, attr); + ret = parse_tracepoint_event(opt, str, attr); if (ret != EVT_FAILED) goto modifier; @@ -783,7 +785,7 @@ int parse_events(const struct option *opt, const char *str, int unset __used) for (;;) { memset(&attr, 0, sizeof(attr)); - ret = parse_event_symbols(&str, &attr); + ret = parse_event_symbols(opt, &str, &attr); if (ret == EVT_FAILED) return -1; From fd78260b5376173faeb17127bd63b3c99a8e8bfb Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 18 Jan 2011 15:15:24 -0200 Subject: [PATCH 27/29] perf threads: Move thread_map to separate file To untangle it from struct thread handling, that is tied to symbols, etc. Right now in the python bindings I'm working on I need just a subset of the util/ files, untangling it allows me to do that. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Makefile | 2 ++ tools/perf/builtin-record.c | 1 + tools/perf/builtin-stat.c | 1 + tools/perf/builtin-test.c | 2 +- tools/perf/builtin-top.c | 1 + tools/perf/util/evsel.c | 2 +- tools/perf/util/thread.c | 55 ------------------------------- tools/perf/util/thread.h | 14 -------- tools/perf/util/thread_map.c | 64 ++++++++++++++++++++++++++++++++++++ tools/perf/util/thread_map.h | 15 +++++++++ 10 files changed, 86 insertions(+), 71 deletions(-) create mode 100644 tools/perf/util/thread_map.c create mode 100644 tools/perf/util/thread_map.h diff --git a/tools/perf/Makefile b/tools/perf/Makefile index f20bc6f85611..638e8e146bb9 100644 --- a/tools/perf/Makefile +++ b/tools/perf/Makefile @@ -426,6 +426,7 @@ LIB_H += util/values.h LIB_H += util/sort.h LIB_H += util/hist.h LIB_H += util/thread.h +LIB_H += util/thread_map.h LIB_H += util/trace-event.h LIB_H += util/probe-finder.h LIB_H += util/probe-event.h @@ -471,6 +472,7 @@ LIB_OBJS += $(OUTPUT)util/map.o LIB_OBJS += $(OUTPUT)util/pstack.o LIB_OBJS += $(OUTPUT)util/session.o LIB_OBJS += $(OUTPUT)util/thread.o +LIB_OBJS += $(OUTPUT)util/thread_map.o LIB_OBJS += $(OUTPUT)util/trace-event-parse.o LIB_OBJS += $(OUTPUT)util/trace-event-read.o LIB_OBJS += $(OUTPUT)util/trace-event-info.o diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 1346d4230bc0..d7886307f6f4 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -24,6 +24,7 @@ #include "util/session.h" #include "util/symbol.h" #include "util/cpumap.h" +#include "util/thread_map.h" #include #include diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index e2a2d02c5c43..8906adfdbd8e 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -49,6 +49,7 @@ #include "util/header.h" #include "util/cpumap.h" #include "util/thread.h" +#include "util/thread_map.h" #include #include diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index 4fd34537c01d..dc91ee06a37c 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -12,7 +12,7 @@ #include "util/parse-events.h" #include "util/session.h" #include "util/symbol.h" -#include "util/thread.h" +#include "util/thread_map.h" static long page_size; diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 31fbaf38d9c1..d0b16d905405 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -26,6 +26,7 @@ #include "util/session.h" #include "util/symbol.h" #include "util/thread.h" +#include "util/thread_map.h" #include "util/util.h" #include #include "util/parse-options.h" diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index ee490356c817..9a6d94299ab8 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -3,7 +3,7 @@ #include "../perf.h" #include "util.h" #include "cpumap.h" -#include "thread.h" +#include "thread_map.h" #include #include diff --git a/tools/perf/util/thread.c b/tools/perf/util/thread.c index 00f4eade2e3e..d5d3b22250f3 100644 --- a/tools/perf/util/thread.c +++ b/tools/perf/util/thread.c @@ -7,61 +7,6 @@ #include "util.h" #include "debug.h" -/* Skip "." and ".." directories */ -static int filter(const struct dirent *dir) -{ - if (dir->d_name[0] == '.') - return 0; - else - return 1; -} - -struct thread_map *thread_map__new_by_pid(pid_t pid) -{ - struct thread_map *threads; - char name[256]; - int items; - struct dirent **namelist = NULL; - int i; - - sprintf(name, "/proc/%d/task", pid); - items = scandir(name, &namelist, filter, NULL); - if (items <= 0) - return NULL; - - threads = malloc(sizeof(*threads) + sizeof(pid_t) * items); - if (threads != NULL) { - for (i = 0; i < items; i++) - threads->map[i] = atoi(namelist[i]->d_name); - threads->nr = items; - } - - for (i=0; imap[0] = tid; - threads->nr = 1; - } - - return threads; -} - -struct thread_map *thread_map__new(pid_t pid, pid_t tid) -{ - if (pid != -1) - return thread_map__new_by_pid(pid); - return thread_map__new_by_tid(tid); -} - static struct thread *thread__new(pid_t pid) { struct thread *self = zalloc(sizeof(*self)); diff --git a/tools/perf/util/thread.h b/tools/perf/util/thread.h index d7574101054a..e5f2401c1b5e 100644 --- a/tools/perf/util/thread.h +++ b/tools/perf/util/thread.h @@ -18,24 +18,10 @@ struct thread { int comm_len; }; -struct thread_map { - int nr; - int map[]; -}; - struct perf_session; void thread__delete(struct thread *self); -struct thread_map *thread_map__new_by_pid(pid_t pid); -struct thread_map *thread_map__new_by_tid(pid_t tid); -struct thread_map *thread_map__new(pid_t pid, pid_t tid); - -static inline void thread_map__delete(struct thread_map *threads) -{ - free(threads); -} - int thread__set_comm(struct thread *self, const char *comm); int thread__comm_len(struct thread *self); struct thread *perf_session__findnew(struct perf_session *self, pid_t pid); diff --git a/tools/perf/util/thread_map.c b/tools/perf/util/thread_map.c new file mode 100644 index 000000000000..a5df131b77c3 --- /dev/null +++ b/tools/perf/util/thread_map.c @@ -0,0 +1,64 @@ +#include +#include +#include +#include "thread_map.h" + +/* Skip "." and ".." directories */ +static int filter(const struct dirent *dir) +{ + if (dir->d_name[0] == '.') + return 0; + else + return 1; +} + +struct thread_map *thread_map__new_by_pid(pid_t pid) +{ + struct thread_map *threads; + char name[256]; + int items; + struct dirent **namelist = NULL; + int i; + + sprintf(name, "/proc/%d/task", pid); + items = scandir(name, &namelist, filter, NULL); + if (items <= 0) + return NULL; + + threads = malloc(sizeof(*threads) + sizeof(pid_t) * items); + if (threads != NULL) { + for (i = 0; i < items; i++) + threads->map[i] = atoi(namelist[i]->d_name); + threads->nr = items; + } + + for (i=0; imap[0] = tid; + threads->nr = 1; + } + + return threads; +} + +struct thread_map *thread_map__new(pid_t pid, pid_t tid) +{ + if (pid != -1) + return thread_map__new_by_pid(pid); + return thread_map__new_by_tid(tid); +} + +void thread_map__delete(struct thread_map *threads) +{ + free(threads); +} diff --git a/tools/perf/util/thread_map.h b/tools/perf/util/thread_map.h new file mode 100644 index 000000000000..3cb907311409 --- /dev/null +++ b/tools/perf/util/thread_map.h @@ -0,0 +1,15 @@ +#ifndef __PERF_THREAD_MAP_H +#define __PERF_THREAD_MAP_H + +#include + +struct thread_map { + int nr; + int map[]; +}; + +struct thread_map *thread_map__new_by_pid(pid_t pid); +struct thread_map *thread_map__new_by_tid(pid_t tid); +struct thread_map *thread_map__new(pid_t pid, pid_t tid); +void thread_map__delete(struct thread_map *threads); +#endif /* __PERF_THREAD_MAP_H */ From d0dd74e853a0a6f37e8061d6d50be41c7034c54c Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Fri, 21 Jan 2011 13:46:41 -0200 Subject: [PATCH 28/29] perf tools: Move event__parse_sample to evsel.c To avoid linking more stuff in the python binding I'm working on, future csets will make the sample type be taken from the evsel itself, but for that we need to first have one file per cpu and per sample_type, not a single perf.data file. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-test.c | 11 +--- tools/perf/builtin-top.c | 2 +- tools/perf/util/event.c | 125 -------------------------------------- tools/perf/util/event.h | 5 +- tools/perf/util/evsel.c | 118 +++++++++++++++++++++++++++++++++++ tools/perf/util/session.c | 4 +- tools/perf/util/session.h | 9 +++ 7 files changed, 134 insertions(+), 140 deletions(-) diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c index dc91ee06a37c..231e3e21810c 100644 --- a/tools/perf/builtin-test.c +++ b/tools/perf/builtin-test.c @@ -10,7 +10,6 @@ #include "util/evlist.h" #include "util/parse-options.h" #include "util/parse-events.h" -#include "util/session.h" #include "util/symbol.h" #include "util/thread_map.h" @@ -457,7 +456,6 @@ static int test__basic_mmap(void) int err = -1; event_t *event; struct thread_map *threads; - struct perf_session session; struct cpu_map *cpus; struct perf_evlist *evlist; struct perf_event_attr attr = { @@ -521,13 +519,6 @@ static int test__basic_mmap(void) attr.wakeup_events = 1; attr.sample_period = 1; - /* - * FIXME: use evsel->attr.sample_type in event__parse_sample. - * This will nicely remove the requirement that we have - * all the events with the same sample_type. - */ - session.sample_type = attr.sample_type; - for (i = 0; i < nsyscalls; ++i) { attr.config = ids[i]; evsels[i] = perf_evsel__new(&attr, i); @@ -567,7 +558,7 @@ static int test__basic_mmap(void) goto out_munmap; } - event__parse_sample(event, &session, &sample); + event__parse_sample(event, attr.sample_type, false, &sample); evsel = perf_evlist__id2evsel(evlist, sample.id); if (evsel == NULL) { pr_debug("event with id %" PRIu64 diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index d0b16d905405..ce2e50c891c7 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -1106,7 +1106,7 @@ static void perf_session__mmap_read_cpu(struct perf_session *self, int cpu) event_t *event; while ((event = perf_evlist__read_on_cpu(evsel_list, cpu)) != NULL) { - event__parse_sample(event, self, &sample); + perf_session__parse_sample(self, event, &sample); if (event->header.type == PERF_RECORD_SAMPLE) event__process_sample(event, &sample, self); diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c index 1478ab4ee222..e4db8b888546 100644 --- a/tools/perf/util/event.c +++ b/tools/perf/util/event.c @@ -826,128 +826,3 @@ out_filtered: al->filtered = true; return 0; } - -static int event__parse_id_sample(const event_t *event, - struct perf_session *session, - struct sample_data *sample) -{ - const u64 *array; - u64 type; - - sample->cpu = sample->pid = sample->tid = -1; - sample->stream_id = sample->id = sample->time = -1ULL; - - if (!session->sample_id_all) - return 0; - - array = event->sample.array; - array += ((event->header.size - - sizeof(event->header)) / sizeof(u64)) - 1; - type = session->sample_type; - - if (type & PERF_SAMPLE_CPU) { - u32 *p = (u32 *)array; - sample->cpu = *p; - array--; - } - - if (type & PERF_SAMPLE_STREAM_ID) { - sample->stream_id = *array; - array--; - } - - if (type & PERF_SAMPLE_ID) { - sample->id = *array; - array--; - } - - if (type & PERF_SAMPLE_TIME) { - sample->time = *array; - array--; - } - - if (type & PERF_SAMPLE_TID) { - u32 *p = (u32 *)array; - sample->pid = p[0]; - sample->tid = p[1]; - } - - return 0; -} - -int event__parse_sample(const event_t *event, struct perf_session *session, - struct sample_data *data) -{ - const u64 *array; - u64 type; - - if (event->header.type != PERF_RECORD_SAMPLE) - return event__parse_id_sample(event, session, data); - - array = event->sample.array; - type = session->sample_type; - - if (type & PERF_SAMPLE_IP) { - data->ip = event->ip.ip; - array++; - } - - if (type & PERF_SAMPLE_TID) { - u32 *p = (u32 *)array; - data->pid = p[0]; - data->tid = p[1]; - array++; - } - - if (type & PERF_SAMPLE_TIME) { - data->time = *array; - array++; - } - - if (type & PERF_SAMPLE_ADDR) { - data->addr = *array; - array++; - } - - data->id = -1ULL; - if (type & PERF_SAMPLE_ID) { - data->id = *array; - array++; - } - - if (type & PERF_SAMPLE_STREAM_ID) { - data->stream_id = *array; - array++; - } - - if (type & PERF_SAMPLE_CPU) { - u32 *p = (u32 *)array; - data->cpu = *p; - array++; - } else - data->cpu = -1; - - if (type & PERF_SAMPLE_PERIOD) { - data->period = *array; - array++; - } - - if (type & PERF_SAMPLE_READ) { - pr_debug("PERF_SAMPLE_READ is unsuported for now\n"); - return -1; - } - - if (type & PERF_SAMPLE_CALLCHAIN) { - data->callchain = (struct ip_callchain *)array; - array += 1 + data->callchain->nr; - } - - if (type & PERF_SAMPLE_RAW) { - u32 *p = (u32 *)array; - data->raw_size = *p; - p++; - data->raw_data = p; - } - - return 0; -} diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h index 2b7e91902f10..d79e4edd82f9 100644 --- a/tools/perf/util/event.h +++ b/tools/perf/util/event.h @@ -169,9 +169,10 @@ struct addr_location; int event__preprocess_sample(const event_t *self, struct perf_session *session, struct addr_location *al, struct sample_data *data, symbol_filter_t filter); -int event__parse_sample(const event_t *event, struct perf_session *session, - struct sample_data *sample); const char *event__get_event_name(unsigned int id); +int event__parse_sample(const event_t *event, u64 type, bool sample_id_all, + struct sample_data *sample); + #endif /* __PERF_RECORD_H */ diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 9a6d94299ab8..a85ae12845ea 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -355,3 +355,121 @@ out_unmap: } return -1; } + +static int event__parse_id_sample(const event_t *event, u64 type, + struct sample_data *sample) +{ + const u64 *array = event->sample.array; + + array += ((event->header.size - + sizeof(event->header)) / sizeof(u64)) - 1; + + if (type & PERF_SAMPLE_CPU) { + u32 *p = (u32 *)array; + sample->cpu = *p; + array--; + } + + if (type & PERF_SAMPLE_STREAM_ID) { + sample->stream_id = *array; + array--; + } + + if (type & PERF_SAMPLE_ID) { + sample->id = *array; + array--; + } + + if (type & PERF_SAMPLE_TIME) { + sample->time = *array; + array--; + } + + if (type & PERF_SAMPLE_TID) { + u32 *p = (u32 *)array; + sample->pid = p[0]; + sample->tid = p[1]; + } + + return 0; +} + +int event__parse_sample(const event_t *event, u64 type, bool sample_id_all, + struct sample_data *data) +{ + const u64 *array; + + data->cpu = data->pid = data->tid = -1; + data->stream_id = data->id = data->time = -1ULL; + + if (event->header.type != PERF_RECORD_SAMPLE) { + if (!sample_id_all) + return 0; + return event__parse_id_sample(event, type, data); + } + + array = event->sample.array; + + if (type & PERF_SAMPLE_IP) { + data->ip = event->ip.ip; + array++; + } + + if (type & PERF_SAMPLE_TID) { + u32 *p = (u32 *)array; + data->pid = p[0]; + data->tid = p[1]; + array++; + } + + if (type & PERF_SAMPLE_TIME) { + data->time = *array; + array++; + } + + if (type & PERF_SAMPLE_ADDR) { + data->addr = *array; + array++; + } + + data->id = -1ULL; + if (type & PERF_SAMPLE_ID) { + data->id = *array; + array++; + } + + if (type & PERF_SAMPLE_STREAM_ID) { + data->stream_id = *array; + array++; + } + + if (type & PERF_SAMPLE_CPU) { + u32 *p = (u32 *)array; + data->cpu = *p; + array++; + } + + if (type & PERF_SAMPLE_PERIOD) { + data->period = *array; + array++; + } + + if (type & PERF_SAMPLE_READ) { + fprintf(stderr, "PERF_SAMPLE_READ is unsuported for now\n"); + return -1; + } + + if (type & PERF_SAMPLE_CALLCHAIN) { + data->callchain = (struct ip_callchain *)array; + array += 1 + data->callchain->nr; + } + + if (type & PERF_SAMPLE_RAW) { + u32 *p = (u32 *)array; + data->raw_size = *p; + p++; + data->raw_data = p; + } + + return 0; +} diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c index b58a48a5e5a9..e6a07408669e 100644 --- a/tools/perf/util/session.c +++ b/tools/perf/util/session.c @@ -496,7 +496,7 @@ static void flush_sample_queue(struct perf_session *s, if (iter->timestamp > limit) break; - event__parse_sample(iter->event, s, &sample); + perf_session__parse_sample(s, iter->event, &sample); perf_session_deliver_event(s, iter->event, &sample, ops, iter->file_offset); @@ -806,7 +806,7 @@ static int perf_session__process_event(struct perf_session *session, /* * For all kernel events we get the sample data */ - event__parse_sample(event, session, &sample); + perf_session__parse_sample(session, event, &sample); /* Preprocess sample records - precheck callchains */ if (perf_session__preprocess_sample(session, event, &sample)) diff --git a/tools/perf/util/session.h b/tools/perf/util/session.h index e815468eb888..78239767011e 100644 --- a/tools/perf/util/session.h +++ b/tools/perf/util/session.h @@ -155,4 +155,13 @@ size_t perf_session__fprintf_nr_events(struct perf_session *self, FILE *fp) { return hists__fprintf_nr_events(&self->hists, fp); } + +static inline int perf_session__parse_sample(struct perf_session *session, + const event_t *event, + struct sample_data *sample) +{ + return event__parse_sample(event, session->sample_type, + session->sample_id_all, sample); +} + #endif /* __PERF_SESSION_H */ From ef1d1af28ca37fdbc2745da040529cd2953c1af5 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 18 Jan 2011 21:41:45 -0200 Subject: [PATCH 29/29] perf evsel: Introduce perf_evsel__{in,ex}it Out of the {con,des}structor, as in interpreted language bindings we will need to go back from the wrapper object to the real thing. In that case using container_of will save us to have an extra pointer in the perf_evsel struct. Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Mike Galbraith Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Zanussi LKML-Reference: Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/evlist.c | 29 ++++++++++++++++++++--------- tools/perf/util/evlist.h | 2 ++ tools/perf/util/evsel.c | 22 ++++++++++++++++------ tools/perf/util/evsel.h | 3 +++ 4 files changed, 41 insertions(+), 15 deletions(-) diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c index 4b3b84cd71a1..df0610e9c61b 100644 --- a/tools/perf/util/evlist.c +++ b/tools/perf/util/evlist.c @@ -6,17 +6,21 @@ #include #include +void perf_evlist__init(struct perf_evlist *evlist) +{ + int i; + + for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) + INIT_HLIST_HEAD(&evlist->heads[i]); + INIT_LIST_HEAD(&evlist->entries); +} + struct perf_evlist *perf_evlist__new(void) { struct perf_evlist *evlist = zalloc(sizeof(*evlist)); - if (evlist != NULL) { - int i; - - for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) - INIT_HLIST_HEAD(&evlist->heads[i]); - INIT_LIST_HEAD(&evlist->entries); - } + if (evlist != NULL) + perf_evlist__init(evlist); return evlist; } @@ -33,11 +37,18 @@ static void perf_evlist__purge(struct perf_evlist *evlist) evlist->nr_entries = 0; } +void perf_evlist__exit(struct perf_evlist *evlist) +{ + free(evlist->mmap); + free(evlist->pollfd); + evlist->mmap = NULL; + evlist->pollfd = NULL; +} + void perf_evlist__delete(struct perf_evlist *evlist) { perf_evlist__purge(evlist); - free(evlist->mmap); - free(evlist->pollfd); + perf_evlist__exit(evlist); free(evlist); } diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h index 28712063db97..acbe48eac608 100644 --- a/tools/perf/util/evlist.h +++ b/tools/perf/util/evlist.h @@ -24,6 +24,8 @@ struct perf_evlist { struct perf_evsel; struct perf_evlist *perf_evlist__new(void); +void perf_evlist__init(struct perf_evlist *evlist); +void perf_evlist__exit(struct perf_evlist *evlist); void perf_evlist__delete(struct perf_evlist *evlist); void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry); diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index a85ae12845ea..76ab553637d6 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -14,15 +14,20 @@ #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) #define SID(e, x, y) xyarray__entry(e->id, x, y) +void perf_evsel__init(struct perf_evsel *evsel, + struct perf_event_attr *attr, int idx) +{ + evsel->idx = idx; + evsel->attr = *attr; + INIT_LIST_HEAD(&evsel->node); +} + struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx) { struct perf_evsel *evsel = zalloc(sizeof(*evsel)); - if (evsel != NULL) { - evsel->idx = idx; - evsel->attr = *attr; - INIT_LIST_HEAD(&evsel->node); - } + if (evsel != NULL) + perf_evsel__init(evsel, attr, idx); return evsel; } @@ -87,11 +92,16 @@ int perf_evlist__alloc_mmap(struct perf_evlist *evlist, int ncpus) return evlist->mmap != NULL ? 0 : -ENOMEM; } -void perf_evsel__delete(struct perf_evsel *evsel) +void perf_evsel__exit(struct perf_evsel *evsel) { assert(list_empty(&evsel->node)); xyarray__delete(evsel->fd); xyarray__delete(evsel->id); +} + +void perf_evsel__delete(struct perf_evsel *evsel) +{ + perf_evsel__exit(evsel); free(evsel); } diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 667ee4e2e35e..7962e7587dea 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -52,6 +52,9 @@ struct thread_map; struct perf_evlist; struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx); +void perf_evsel__init(struct perf_evsel *evsel, + struct perf_event_attr *attr, int idx); +void perf_evsel__exit(struct perf_evsel *evsel); void perf_evsel__delete(struct perf_evsel *evsel); int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads);