From 823254edc66eb44bf612b1dfa4829afa9840f691 Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Thu, 29 Nov 2012 15:38:30 +0900 Subject: perf evsel: Convert to _is_group_leader method Convert perf_evsel__is_group_member to perf_evsel__is_group_leader. This is because the most usecases are using negative form to check whether the given evsel is a leader or not and it's IMHO somewhat ambiguous - leader also *is* a member of the group. Signed-off-by: Namhyung Kim Acked-by: Jiri Olsa Cc: Andi Kleen Cc: Ingo Molnar Cc: Jiri Olsa Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/r/1354171126-14387-4-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-stat.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'tools/perf/builtin-stat.c') diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index c247faca7127..c12655af2b88 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -153,7 +153,7 @@ retry: } if (!perf_target__has_task(&target) && - !perf_evsel__is_group_member(evsel)) { + perf_evsel__is_group_leader(evsel)) { attr->disabled = 1; attr->enable_on_exec = 1; } -- cgit v1.2.3 From 594ac61ad3be9c80c738a9fe3bb95c05d8d1bae1 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 13 Dec 2012 13:13:07 -0300 Subject: perf evsel: Do missing feature fallbacks in just one place Instead of doing it in stat, top, record or any other tool that opens event descriptors. Cc: David Ahern Cc: Frederic Weisbecker Cc: Jiri Olsa Cc: Mike Galbraith Cc: Namhyung Kim Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/n/tip-vr8hzph83d5t2mdlkf565h84@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 36 ------------------------------------ tools/perf/builtin-stat.c | 30 +++--------------------------- tools/perf/builtin-top.c | 20 -------------------- tools/perf/perf.h | 2 -- tools/perf/util/evsel.c | 31 ++++++++++++++++++++++++++++--- tools/perf/util/top.h | 2 -- 6 files changed, 31 insertions(+), 90 deletions(-) (limited to 'tools/perf/builtin-stat.c') diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 028de726b832..a4b97269cfc6 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -234,25 +234,6 @@ static int perf_record__open(struct perf_record *rec) 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; - -fallback_missing_features: - if (opts->exclude_guest_missing) - attr->exclude_guest = attr->exclude_host = 0; -retry_sample_id: - attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1; try_again: if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) { int err = errno; @@ -266,23 +247,6 @@ try_again: " an out-of-range profile CPU?\n"); rc = -err; goto out; - } else if (err == EINVAL) { - if (!opts->exclude_guest_missing && - (attr->exclude_guest || attr->exclude_host)) { - pr_debug("Old kernel, cannot exclude " - "guest or host samples.\n"); - opts->exclude_guest_missing = true; - goto fallback_missing_features; - } else if (!opts->sample_id_all_missing) { - /* - * Old kernel, no attr->sample_id_type_all field - */ - opts->sample_id_all_missing = true; - if (!opts->sample_time && !opts->raw_samples && !time_needed) - perf_evsel__reset_sample_bit(pos, TIME); - - goto retry_sample_id; - } } /* diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index c12655af2b88..ef067c193f93 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -132,8 +132,6 @@ static struct stats walltime_nsecs_stats; static int create_perf_stat_counter(struct perf_evsel *evsel) { struct perf_event_attr *attr = &evsel->attr; - bool exclude_guest_missing = false; - int ret; if (scale) attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | @@ -141,16 +139,8 @@ static int create_perf_stat_counter(struct perf_evsel *evsel) attr->inherit = !no_inherit; -retry: - if (exclude_guest_missing) - evsel->attr.exclude_guest = evsel->attr.exclude_host = 0; - - if (perf_target__has_cpu(&target)) { - ret = perf_evsel__open_per_cpu(evsel, perf_evsel__cpus(evsel)); - if (ret) - goto check_ret; - return 0; - } + if (perf_target__has_cpu(&target)) + return perf_evsel__open_per_cpu(evsel, perf_evsel__cpus(evsel)); if (!perf_target__has_task(&target) && perf_evsel__is_group_leader(evsel)) { @@ -158,21 +148,7 @@ retry: attr->enable_on_exec = 1; } - ret = perf_evsel__open_per_thread(evsel, evsel_list->threads); - if (!ret) - return 0; - /* fall through */ -check_ret: - if (ret && errno == EINVAL) { - if (!exclude_guest_missing && - (evsel->attr.exclude_guest || evsel->attr.exclude_host)) { - pr_debug("Old kernel, cannot exclude " - "guest or host samples.\n"); - exclude_guest_missing = true; - goto retry; - } - } - return ret; + return perf_evsel__open_per_thread(evsel, evsel_list->threads); } /* diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index b7d2ea62dbc6..74fca619fc4e 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -900,12 +900,6 @@ static void perf_top__start_counters(struct perf_top *top) list_for_each_entry(counter, &evlist->entries, node) { struct perf_event_attr *attr = &counter->attr; - -fallback_missing_features: - if (top->exclude_guest_missing) - attr->exclude_guest = attr->exclude_host = 0; -retry_sample_id: - attr->sample_id_all = top->sample_id_all_missing ? 0 : 1; try_again: if (perf_evsel__open(counter, top->evlist->cpus, top->evlist->threads) < 0) { @@ -914,20 +908,6 @@ try_again: if (err == EPERM || err == EACCES) { ui__error_paranoid(); goto out_err; - } else if (err == EINVAL) { - if (!top->exclude_guest_missing && - (attr->exclude_guest || attr->exclude_host)) { - pr_debug("Old kernel, cannot exclude " - "guest or host samples.\n"); - top->exclude_guest_missing = true; - goto fallback_missing_features; - } else if (!top->sample_id_all_missing) { - /* - * Old kernel, no attr->sample_id_type_all field - */ - top->sample_id_all_missing = true; - goto retry_sample_id; - } } /* * If it's cycles then fall back to hrtimer diff --git a/tools/perf/perf.h b/tools/perf/perf.h index 2c340e7da458..7622f15bcfb7 100644 --- a/tools/perf/perf.h +++ b/tools/perf/perf.h @@ -237,8 +237,6 @@ struct perf_record_opts { bool raw_samples; bool sample_address; bool sample_time; - bool sample_id_all_missing; - bool exclude_guest_missing; bool period; unsigned int freq; unsigned int mmap_pages; diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 7a2a3dc3ff03..ee6ee3f45c2a 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -22,6 +22,11 @@ #include #include "perf_regs.h" +static struct { + bool sample_id_all; + bool exclude_guest; +} perf_missing_features; + #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) static int __perf_evsel__sample_size(u64 sample_type) @@ -463,7 +468,7 @@ void perf_evsel__config(struct perf_evsel *evsel, struct perf_event_attr *attr = &evsel->attr; int track = !evsel->idx; /* only the first counter needs these */ - attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1; + attr->sample_id_all = perf_missing_features.sample_id_all ? 0 : 1; attr->inherit = !opts->no_inherit; perf_evsel__set_sample_bit(evsel, IP); @@ -513,7 +518,7 @@ void perf_evsel__config(struct perf_evsel *evsel, if (opts->period) perf_evsel__set_sample_bit(evsel, PERIOD); - if (!opts->sample_id_all_missing && + if (!perf_missing_features.sample_id_all && (opts->sample_time || !opts->no_inherit || perf_target__has_cpu(&opts->target))) perf_evsel__set_sample_bit(evsel, TIME); @@ -761,6 +766,13 @@ static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, pid = evsel->cgrp->fd; } +fallback_missing_features: + if (perf_missing_features.exclude_guest) + evsel->attr.exclude_guest = evsel->attr.exclude_host = 0; +retry_sample_id: + if (perf_missing_features.sample_id_all) + evsel->attr.sample_id_all = 0; + for (cpu = 0; cpu < cpus->nr; cpu++) { for (thread = 0; thread < threads->nr; thread++) { @@ -777,13 +789,26 @@ static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, group_fd, flags); if (FD(evsel, cpu, thread) < 0) { err = -errno; - goto out_close; + goto try_fallback; } } } return 0; +try_fallback: + if (err != -EINVAL || cpu > 0 || thread > 0) + goto out_close; + + if (!perf_missing_features.exclude_guest && + (evsel->attr.exclude_guest || evsel->attr.exclude_host)) { + perf_missing_features.exclude_guest = true; + goto fallback_missing_features; + } else if (!perf_missing_features.sample_id_all) { + perf_missing_features.sample_id_all = true; + goto retry_sample_id; + } + out_close: do { while (--thread >= 0) { diff --git a/tools/perf/util/top.h b/tools/perf/util/top.h index 927c229c2d9a..7ebf357dc9e1 100644 --- a/tools/perf/util/top.h +++ b/tools/perf/util/top.h @@ -29,8 +29,6 @@ struct perf_top { bool sort_has_symbols; bool kptr_restrict_warned; bool vmlinux_warned; - bool sample_id_all_missing; - bool exclude_guest_missing; bool dump_symtab; struct hist_entry *sym_filter_entry; struct perf_evsel *sym_evsel; -- cgit v1.2.3 From 56e52e85366717481cde16b3480f015c7eb32ba3 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 13 Dec 2012 15:10:58 -0300 Subject: perf evsel: Introduce perf_evsel__open_strerror method That consolidates the error messages in 'record', 'stat' and 'top', that now get a consistent set of messages and allow other tools to use the new method to report problems using whatever UI toolkit. Cc: David Ahern Cc: Frederic Weisbecker Cc: Jiri Olsa Cc: Mike Galbraith Cc: Namhyung Kim Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/n/tip-1cudb7wl996kz7ilz83ctvhr@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-record.c | 54 +++++---------------------------------------- tools/perf/builtin-stat.c | 16 +++++--------- tools/perf/builtin-top.c | 34 +++++----------------------- tools/perf/ui/util.c | 11 --------- tools/perf/util/debug.h | 1 - tools/perf/util/evsel.c | 49 ++++++++++++++++++++++++++++++++++++++++ tools/perf/util/evsel.h | 3 +++ 7 files changed, 68 insertions(+), 100 deletions(-) (limited to 'tools/perf/builtin-stat.c') diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index b8d0a39b484e..de60dd4ee5d3 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -224,7 +224,7 @@ static bool perf_evlist__equal(struct perf_evlist *evlist, static int perf_record__open(struct perf_record *rec) { - char msg[128]; + char msg[512]; struct perf_evsel *pos; struct perf_evlist *evlist = rec->evlist; struct perf_session *session = rec->session; @@ -234,60 +234,18 @@ static int perf_record__open(struct perf_record *rec) perf_evlist__config(evlist, opts); list_for_each_entry(pos, &evlist->entries, node) { - struct perf_event_attr *attr = &pos->attr; try_again: if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) { - int err = errno; - - if (err == EPERM || err == EACCES) { - ui__error_paranoid(); - rc = -err; - goto out; - } else if (err == ENODEV && opts->target.cpu_list) { - pr_err("No such device - did you specify" - " an out-of-range profile CPU?\n"); - rc = -err; - goto out; - } - - if (perf_evsel__fallback(pos, err, msg, sizeof(msg))) { + if (perf_evsel__fallback(pos, errno, msg, sizeof(msg))) { if (verbose) ui__warning("%s\n", msg); goto try_again; } - if (err == ENOENT) { - ui__error("The %s event is not supported.\n", - perf_evsel__name(pos)); - rc = -err; - goto out; - } else if ((err == EOPNOTSUPP) && (attr->precise_ip)) { - ui__error("\'precise\' request may not be supported. " - "Try removing 'p' modifier\n"); - rc = -err; - goto out; - } - - printf("\n"); - error("sys_perf_event_open() syscall returned with %d " - "(%s) for event %s. /bin/dmesg may provide " - "additional information.\n", - err, strerror(err), perf_evsel__name(pos)); - -#if defined(__i386__) || defined(__x86_64__) - if (attr->type == PERF_TYPE_HARDWARE && - err == EOPNOTSUPP) { - pr_err("No hardware sampling interrupt available." - " No APIC? If so then you can boot the kernel" - " with the \"lapic\" boot parameter to" - " force-enable it.\n"); - rc = -err; - goto out; - } -#endif - - pr_err("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); - rc = -err; + rc = -errno; + perf_evsel__open_strerror(pos, &opts->target, + errno, msg, sizeof(msg)); + ui__error("%s\n", msg); goto out; } } diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index ef067c193f93..1c2ac148a7d5 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -247,6 +247,7 @@ static int read_counter(struct perf_evsel *counter) static int __run_perf_stat(int argc __maybe_unused, const char **argv) { + char msg[512]; unsigned long long t0, t1; struct perf_evsel *counter; int status = 0; @@ -324,20 +325,13 @@ static int __run_perf_stat(int argc __maybe_unused, const char **argv) continue; } - if (errno == EPERM || errno == EACCES) { - error("You may not have permission to collect %sstats.\n" - "\t Consider tweaking" - " /proc/sys/kernel/perf_event_paranoid or running as root.", - target.system_wide ? "system-wide " : ""); - } else { - error("open_counter returned with %d (%s). " - "/bin/dmesg may provide additional information.\n", - errno, strerror(errno)); - } + perf_evsel__open_strerror(counter, &target, + errno, msg, sizeof(msg)); + ui__error("%s\n", msg); + if (child_pid != -1) kill(child_pid, SIGTERM); - pr_err("Not all events could be opened.\n"); return -1; } counter->supported = true; diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 8d41d0b58956..f5fd260f7b20 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -892,7 +892,7 @@ static void perf_top__mmap_read(struct perf_top *top) static void perf_top__start_counters(struct perf_top *top) { - char msg[128]; + char msg[512]; struct perf_evsel *counter; struct perf_evlist *evlist = top->evlist; struct perf_record_opts *opts = &top->record_opts; @@ -900,42 +900,18 @@ static void perf_top__start_counters(struct perf_top *top) perf_evlist__config(evlist, opts); list_for_each_entry(counter, &evlist->entries, node) { - struct perf_event_attr *attr = &counter->attr; try_again: if (perf_evsel__open(counter, top->evlist->cpus, top->evlist->threads) < 0) { - int err = errno; - - if (err == EPERM || err == EACCES) { - ui__error_paranoid(); - goto out_err; - } - - if (perf_evsel__fallback(counter, err, msg, sizeof(msg))) { + if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) { if (verbose) ui__warning("%s\n", msg); goto try_again; } - if (err == ENOENT) { - ui__error("The %s event is not supported.\n", - perf_evsel__name(counter)); - goto out_err; - } else if (err == EMFILE) { - ui__error("Too many events are opened.\n" - "Try again after reducing the number of events\n"); - goto out_err; - } else if ((err == EOPNOTSUPP) && (attr->precise_ip)) { - ui__error("\'precise\' request may not be supported. " - "Try removing 'p' modifier\n"); - goto out_err; - } - - ui__error("The sys_perf_event_open() syscall " - "returned with %d (%s). /bin/dmesg " - "may provide additional information.\n" - "No CONFIG_PERF_EVENTS=y kernel support " - "configured?\n", err, strerror(err)); + perf_evsel__open_strerror(counter, &opts->target, + errno, msg, sizeof(msg)); + ui__error("%s\n", msg); goto out_err; } } diff --git a/tools/perf/ui/util.c b/tools/perf/ui/util.c index 3014a7cd5271..e3e0a963d03a 100644 --- a/tools/perf/ui/util.c +++ b/tools/perf/ui/util.c @@ -52,17 +52,6 @@ int ui__warning(const char *format, ...) return ret; } -int ui__error_paranoid(void) -{ - return ui__error("Permission error - are you root?\n" - "Consider tweaking /proc/sys/kernel/perf_event_paranoid:\n" - " -1 - Not paranoid at all\n" - " 0 - Disallow raw tracepoint access for unpriv\n" - " 1 - Disallow cpu events for unpriv\n" - " 2 - Disallow kernel profiling for unpriv\n"); -} - - /** * perf_error__register - Register error logging functions * @eops: The pointer to error logging function struct diff --git a/tools/perf/util/debug.h b/tools/perf/util/debug.h index 6e2667fb8211..efbd98805ad0 100644 --- a/tools/perf/util/debug.h +++ b/tools/perf/util/debug.h @@ -16,6 +16,5 @@ void trace_event(union perf_event *event); int ui__error(const char *format, ...) __attribute__((format(printf, 1, 2))); int ui__warning(const char *format, ...) __attribute__((format(printf, 1, 2))); -int ui__error_paranoid(void); #endif /* __PERF_DEBUG_H */ diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 0c88e5c12dab..e45332d08a58 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -1406,3 +1406,52 @@ bool perf_evsel__fallback(struct perf_evsel *evsel, int err, return false; } + +int perf_evsel__open_strerror(struct perf_evsel *evsel, + struct perf_target *target, + int err, char *msg, size_t size) +{ + switch (err) { + case EPERM: + case EACCES: + return scnprintf(msg, size, "%s", + "You may not have permission to collect %sstats.\n" + "Consider tweaking /proc/sys/kernel/perf_event_paranoid:\n" + " -1 - Not paranoid at all\n" + " 0 - Disallow raw tracepoint access for unpriv\n" + " 1 - Disallow cpu events for unpriv\n" + " 2 - Disallow kernel profiling for unpriv", + target->system_wide ? "system-wide " : ""); + case ENOENT: + return scnprintf(msg, size, "The %s event is not supported.", + perf_evsel__name(evsel)); + case EMFILE: + return scnprintf(msg, size, "%s", + "Too many events are opened.\n" + "Try again after reducing the number of events."); + case ENODEV: + if (target->cpu_list) + return scnprintf(msg, size, "%s", + "No such device - did you specify an out-of-range profile CPU?\n"); + break; + case EOPNOTSUPP: + if (evsel->attr.precise_ip) + return scnprintf(msg, size, "%s", + "\'precise\' request may not be supported. Try removing 'p' modifier."); +#if defined(__i386__) || defined(__x86_64__) + if (evsel->attr.type == PERF_TYPE_HARDWARE) + return scnprintf(msg, size, "%s", + "No hardware sampling interrupt available.\n" + "No APIC? If so then you can boot the kernel with the \"lapic\" boot parameter to force-enable it."); +#endif + break; + default: + break; + } + + return scnprintf(msg, size, + "The sys_perf_event_open() syscall returned with %d (%s) for event (%s). \n" + "/bin/dmesg may provide additional information.\n" + "No CONFIG_PERF_EVENTS=y kernel support configured?\n", + err, strerror(err), perf_evsel__name(evsel)); +} diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 26d8cfd86e0a..c68d1b82e843 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -254,4 +254,7 @@ int perf_evsel__fprintf(struct perf_evsel *evsel, bool perf_evsel__fallback(struct perf_evsel *evsel, int err, char *msg, size_t msgsize); +int perf_evsel__open_strerror(struct perf_evsel *evsel, + struct perf_target *target, + int err, char *msg, size_t size); #endif /* __PERF_EVSEL_H */ -- cgit v1.2.3 From 13370a9b5bb88f7aa90e5be68972d95096b20a6d Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Tue, 29 Jan 2013 12:47:44 +0100 Subject: perf stat: Add interval printing This patch adds a new printing mode for perf stat. It allows interval printing. That means perf stat can now print event deltas at regular time interval. This is useful to detect phases in programs. The -I option enables interval printing. It expects an interval duration in milliseconds. Minimum is 100ms. Once, activated perf stat prints events deltas since last printout. All modes are supported. $ perf stat -I 1000 -e cycles noploop 10 noploop for 10 seconds # time counts events 1.000109853 2,388,560,546 cycles 2.000262846 2,393,332,358 cycles 3.000354131 2,393,176,537 cycles 4.000439503 2,393,203,790 cycles 5.000527075 2,393,167,675 cycles 6.000609052 2,393,203,670 cycles 7.000691082 2,393,175,678 cycles The output format makes it easy to feed into a plotting program such as gnuplot when the -I option is used in combination with the -x option: $ perf stat -x, -I 1000 -e cycles noploop 10 noploop for 10 seconds 1.000084113,2378775498,cycles 2.000245798,2391056897,cycles 3.000354445,2392089414,cycles 4.000459115,2390936603,cycles 5.000565341,2392108173,cycles Signed-off-by: Stephane Eranian Cc: Andi Kleen Cc: Ingo Molnar Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1359460064-3060-3-git-send-email-eranian@google.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Documentation/perf-stat.txt | 4 + tools/perf/builtin-stat.c | 157 +++++++++++++++++++++++++++++---- 2 files changed, 146 insertions(+), 15 deletions(-) (limited to 'tools/perf/builtin-stat.c') diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt index cf0c3107e06e..5289da3344e9 100644 --- a/tools/perf/Documentation/perf-stat.txt +++ b/tools/perf/Documentation/perf-stat.txt @@ -114,6 +114,10 @@ with it. --append may be used here. Examples: perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage +-I msecs:: +--interval-print msecs:: + print count deltas every N milliseconds (minimum: 100ms) + example: perf stat -I 1000 -e cycles -a sleep 5 EXAMPLES -------- diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 1c2ac148a7d5..493043abd164 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -65,6 +65,10 @@ #define CNTR_NOT_SUPPORTED "" #define CNTR_NOT_COUNTED "" +static void print_stat(int argc, const char **argv); +static void print_counter_aggr(struct perf_evsel *counter, char *prefix); +static void print_counter(struct perf_evsel *counter, char *prefix); + static struct perf_evlist *evsel_list; static struct perf_target target = { @@ -87,6 +91,8 @@ static FILE *output = NULL; static const char *pre_cmd = NULL; static const char *post_cmd = NULL; static bool sync_run = false; +static unsigned int interval = 0; +static struct timespec ref_time; static volatile int done = 0; @@ -94,6 +100,28 @@ struct perf_stat { struct stats res_stats[3]; }; +static inline void diff_timespec(struct timespec *r, struct timespec *a, + struct timespec *b) +{ + r->tv_sec = a->tv_sec - b->tv_sec; + if (a->tv_nsec < b->tv_nsec) { + r->tv_nsec = a->tv_nsec + 1000000000L - b->tv_nsec; + r->tv_sec--; + } else { + r->tv_nsec = a->tv_nsec - b->tv_nsec ; + } +} + +static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel) +{ + return (evsel->cpus && !target.cpu_list) ? evsel->cpus : evsel_list->cpus; +} + +static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel) +{ + return perf_evsel__cpus(evsel)->nr; +} + static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel) { evsel->priv = zalloc(sizeof(struct perf_stat)); @@ -106,14 +134,27 @@ static void perf_evsel__free_stat_priv(struct perf_evsel *evsel) evsel->priv = NULL; } -static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel) +static int perf_evsel__alloc_prev_raw_counts(struct perf_evsel *evsel) { - return (evsel->cpus && !target.cpu_list) ? evsel->cpus : evsel_list->cpus; + void *addr; + size_t sz; + + sz = sizeof(*evsel->counts) + + (perf_evsel__nr_cpus(evsel) * sizeof(struct perf_counts_values)); + + addr = zalloc(sz); + if (!addr) + return -ENOMEM; + + evsel->prev_raw_counts = addr; + + return 0; } -static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel) +static void perf_evsel__free_prev_raw_counts(struct perf_evsel *evsel) { - return perf_evsel__cpus(evsel)->nr; + free(evsel->prev_raw_counts); + evsel->prev_raw_counts = NULL; } static struct stats runtime_nsecs_stats[MAX_NR_CPUS]; @@ -245,16 +286,69 @@ static int read_counter(struct perf_evsel *counter) return 0; } +static void print_interval(void) +{ + static int num_print_interval; + struct perf_evsel *counter; + struct perf_stat *ps; + struct timespec ts, rs; + char prefix[64]; + + if (no_aggr) { + list_for_each_entry(counter, &evsel_list->entries, node) { + ps = counter->priv; + memset(ps->res_stats, 0, sizeof(ps->res_stats)); + read_counter(counter); + } + } else { + list_for_each_entry(counter, &evsel_list->entries, node) { + ps = counter->priv; + memset(ps->res_stats, 0, sizeof(ps->res_stats)); + read_counter_aggr(counter); + } + } + clock_gettime(CLOCK_MONOTONIC, &ts); + diff_timespec(&rs, &ts, &ref_time); + sprintf(prefix, "%6lu.%09lu%s", rs.tv_sec, rs.tv_nsec, csv_sep); + + if (num_print_interval == 0 && !csv_output) { + if (no_aggr) + fprintf(output, "# time CPU counts events\n"); + else + fprintf(output, "# time counts events\n"); + } + + if (++num_print_interval == 25) + num_print_interval = 0; + + if (no_aggr) { + list_for_each_entry(counter, &evsel_list->entries, node) + print_counter(counter, prefix); + } else { + list_for_each_entry(counter, &evsel_list->entries, node) + print_counter_aggr(counter, prefix); + } +} + static int __run_perf_stat(int argc __maybe_unused, const char **argv) { char msg[512]; unsigned long long t0, t1; struct perf_evsel *counter; + struct timespec ts; int status = 0; int child_ready_pipe[2], go_pipe[2]; const bool forks = (argc > 0); char buf; + if (interval) { + ts.tv_sec = interval / 1000; + ts.tv_nsec = (interval % 1000) * 1000000; + } else { + ts.tv_sec = 1; + ts.tv_nsec = 0; + } + if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) { perror("failed to create pipes"); return -1; @@ -347,14 +441,25 @@ static int __run_perf_stat(int argc __maybe_unused, const char **argv) * Enable counters and exec the command: */ t0 = rdclock(); + clock_gettime(CLOCK_MONOTONIC, &ref_time); if (forks) { close(go_pipe[1]); + if (interval) { + while (!waitpid(child_pid, &status, WNOHANG)) { + nanosleep(&ts, NULL); + print_interval(); + } + } wait(&status); if (WIFSIGNALED(status)) psignal(WTERMSIG(status), argv[0]); } else { - while(!done) sleep(1); + while (!done) { + nanosleep(&ts, NULL); + if (interval) + print_interval(); + } } t1 = rdclock(); @@ -440,7 +545,7 @@ static void nsec_printout(int cpu, struct perf_evsel *evsel, double avg) if (evsel->cgrp) fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); - if (csv_output) + if (csv_output || interval) return; if (perf_evsel__match(evsel, SOFTWARE, SW_TASK_CLOCK)) @@ -654,12 +759,11 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg) if (evsel->cgrp) fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); - if (csv_output) + if (csv_output || interval) return; if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) { total = avg_stats(&runtime_cycles_stats[cpu]); - if (total) ratio = avg / total; @@ -753,12 +857,15 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg) * Print out the results of a single counter: * aggregated counts in system-wide mode */ -static void print_counter_aggr(struct perf_evsel *counter) +static void print_counter_aggr(struct perf_evsel *counter, char *prefix) { struct perf_stat *ps = counter->priv; double avg = avg_stats(&ps->res_stats[0]); int scaled = counter->counts->scaled; + if (prefix) + fprintf(output, "%s", prefix); + if (scaled == -1) { fprintf(output, "%*s%s%*s", csv_output ? 0 : 18, @@ -801,7 +908,7 @@ static void print_counter_aggr(struct perf_evsel *counter) * Print out the results of a single counter: * does not use aggregated count in system-wide */ -static void print_counter(struct perf_evsel *counter) +static void print_counter(struct perf_evsel *counter, char *prefix) { u64 ena, run, val; int cpu; @@ -810,6 +917,10 @@ static void print_counter(struct perf_evsel *counter) val = counter->counts->cpu[cpu].val; ena = counter->counts->cpu[cpu].ena; run = counter->counts->cpu[cpu].run; + + if (prefix) + fprintf(output, "%s", prefix); + if (run == 0 || ena == 0) { fprintf(output, "CPU%*d%s%*s%s%*s", csv_output ? 0 : -4, @@ -871,10 +982,10 @@ static void print_stat(int argc, const char **argv) if (no_aggr) { list_for_each_entry(counter, &evsel_list->entries, node) - print_counter(counter); + print_counter(counter, NULL); } else { list_for_each_entry(counter, &evsel_list->entries, node) - print_counter_aggr(counter); + print_counter_aggr(counter, NULL); } if (!csv_output) { @@ -895,7 +1006,7 @@ static volatile int signr = -1; static void skip_signal(int signo) { - if(child_pid == -1) + if ((child_pid == -1) || interval) done = 1; signr = signo; @@ -1115,6 +1226,8 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) "command to run prior to the measured command"), OPT_STRING(0, "post", &post_cmd, "command", "command to run after to the measured command"), + OPT_UINTEGER('I', "interval-print", &interval, + "print counts at regular interval in ms (>= 100)"), OPT_END() }; const char * const stat_usage[] = { @@ -1215,12 +1328,23 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) usage_with_options(stat_usage, options); return -1; } + if (interval && interval < 100) { + pr_err("print interval must be >= 100ms\n"); + usage_with_options(stat_usage, options); + return -1; + } list_for_each_entry(pos, &evsel_list->entries, node) { if (perf_evsel__alloc_stat_priv(pos) < 0 || perf_evsel__alloc_counts(pos, perf_evsel__nr_cpus(pos)) < 0) goto out_free_fd; } + if (interval) { + list_for_each_entry(pos, &evsel_list->entries, node) { + if (perf_evsel__alloc_prev_raw_counts(pos) < 0) + goto out_free_fd; + } + } /* * We dont want to block the signals - that would cause @@ -1230,6 +1354,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) */ atexit(sig_atexit); signal(SIGINT, skip_signal); + signal(SIGCHLD, skip_signal); signal(SIGALRM, skip_signal); signal(SIGABRT, skip_signal); @@ -1242,11 +1367,13 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) status = run_perf_stat(argc, argv); } - if (status != -1) + if (status != -1 && !interval) print_stat(argc, argv); out_free_fd: - list_for_each_entry(pos, &evsel_list->entries, node) + list_for_each_entry(pos, &evsel_list->entries, node) { perf_evsel__free_stat_priv(pos); + perf_evsel__free_prev_raw_counts(pos); + } perf_evlist__delete_maps(evsel_list); out: perf_evlist__delete(evsel_list); -- cgit v1.2.3 From 43f8e76e6b96eb1327cff62ac1cc733a51f31068 Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Fri, 25 Jan 2013 10:44:44 +0900 Subject: perf evsel: Fix memory leaks on evsel->counts The ->counts field was never freed in the current code. Add perf_evsel__free_counts() function to free it properly. Signed-off-by: Namhyung Kim Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/r/1359078284-32080-1-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-stat.c | 1 + tools/perf/tests/open-syscall-all-cpus.c | 1 + tools/perf/util/evsel.c | 5 +++++ tools/perf/util/evsel.h | 1 + 4 files changed, 8 insertions(+) (limited to 'tools/perf/builtin-stat.c') diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 493043abd164..0368a1036ad6 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -1372,6 +1372,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) out_free_fd: list_for_each_entry(pos, &evsel_list->entries, node) { perf_evsel__free_stat_priv(pos); + perf_evsel__free_counts(pos); perf_evsel__free_prev_raw_counts(pos); } perf_evlist__delete_maps(evsel_list); diff --git a/tools/perf/tests/open-syscall-all-cpus.c b/tools/perf/tests/open-syscall-all-cpus.c index 9b920a0cce79..b0657a9ccda6 100644 --- a/tools/perf/tests/open-syscall-all-cpus.c +++ b/tools/perf/tests/open-syscall-all-cpus.c @@ -98,6 +98,7 @@ int test__open_syscall_event_on_all_cpus(void) } } + perf_evsel__free_counts(evsel); out_close_fd: perf_evsel__close_fd(evsel, 1, threads->nr); out_evsel_delete: diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index dbdcca43cac6..baa26ddbcc7b 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -640,6 +640,11 @@ void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads) } } +void perf_evsel__free_counts(struct perf_evsel *evsel) +{ + free(evsel->counts); +} + void perf_evsel__exit(struct perf_evsel *evsel) { assert(list_empty(&evsel->node)); diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 3a4cd60044ea..cbf42322a27e 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -117,6 +117,7 @@ int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads); int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus); void perf_evsel__free_fd(struct perf_evsel *evsel); void perf_evsel__free_id(struct perf_evsel *evsel); +void perf_evsel__free_counts(struct perf_evsel *evsel); void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads); void __perf_evsel__set_sample_bit(struct perf_evsel *evsel, -- cgit v1.2.3 From d7e7a451c13e784f497c054f1bd083d77be87498 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Wed, 6 Feb 2013 15:46:02 +0100 Subject: perf stat: Add per processor socket count aggregation This patch adds per-processor socket count aggregation for system-wide mode measurements. This is a useful mode to detect imbalance between sockets. To enable this mode, use --aggr-socket in addition to -a. (system-wide). The output includes the socket number and the number of online processors on that socket. This is useful to gauge the amount of aggregation. # ./perf stat -I 1000 -a --aggr-socket -e cycles sleep 2 # time socket cpus counts events 1.000097680 S0 4 5,788,785 cycles 2.000379943 S0 4 27,361,546 cycles 2.001167808 S0 4 818,275 cycles Signed-off-by: Stephane Eranian Cc: Andi Kleen Cc: Ingo Molnar Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1360161962-9675-3-git-send-email-eranian@google.com [ committer note: Added missing man page entry based on above comments ] Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/Documentation/perf-stat.txt | 9 ++- tools/perf/builtin-stat.c | 126 ++++++++++++++++++++++++++++++--- 2 files changed, 123 insertions(+), 12 deletions(-) (limited to 'tools/perf/builtin-stat.c') diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt index 5289da3344e9..faf4f4feebcc 100644 --- a/tools/perf/Documentation/perf-stat.txt +++ b/tools/perf/Documentation/perf-stat.txt @@ -116,9 +116,16 @@ perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- m -I msecs:: --interval-print msecs:: - print count deltas every N milliseconds (minimum: 100ms) + Print count deltas every N milliseconds (minimum: 100ms) example: perf stat -I 1000 -e cycles -a sleep 5 +--aggr-socket:: +Aggregate counts per processor socket for system-wide mode measurements. This +is a useful mode to detect imbalance between sockets. To enable this mode, +use --aggr-socket in addition to -a. (system-wide). The output includes the +socket number and the number of online processors on that socket. This is +useful to gauge the amount of aggregation. + EXAMPLES -------- diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 0368a1036ad6..99848761f573 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -68,6 +68,7 @@ static void print_stat(int argc, const char **argv); static void print_counter_aggr(struct perf_evsel *counter, char *prefix); static void print_counter(struct perf_evsel *counter, char *prefix); +static void print_aggr_socket(char *prefix); static struct perf_evlist *evsel_list; @@ -79,6 +80,7 @@ static int run_count = 1; static bool no_inherit = false; static bool scale = true; static bool no_aggr = false; +static bool aggr_socket = false; static pid_t child_pid = -1; static bool null_run = false; static int detailed_run = 0; @@ -93,6 +95,7 @@ static const char *post_cmd = NULL; static bool sync_run = false; static unsigned int interval = 0; static struct timespec ref_time; +static struct cpu_map *sock_map; static volatile int done = 0; @@ -312,7 +315,9 @@ static void print_interval(void) sprintf(prefix, "%6lu.%09lu%s", rs.tv_sec, rs.tv_nsec, csv_sep); if (num_print_interval == 0 && !csv_output) { - if (no_aggr) + if (aggr_socket) + fprintf(output, "# time socket cpus counts events\n"); + else if (no_aggr) fprintf(output, "# time CPU counts events\n"); else fprintf(output, "# time counts events\n"); @@ -321,7 +326,9 @@ static void print_interval(void) if (++num_print_interval == 25) num_print_interval = 0; - if (no_aggr) { + if (aggr_socket) + print_aggr_socket(prefix); + else if (no_aggr) { list_for_each_entry(counter, &evsel_list->entries, node) print_counter(counter, prefix); } else { @@ -349,6 +356,12 @@ static int __run_perf_stat(int argc __maybe_unused, const char **argv) ts.tv_nsec = 0; } + if (aggr_socket + && cpu_map__build_socket_map(evsel_list->cpus, &sock_map)) { + perror("cannot build socket map"); + return -1; + } + if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) { perror("failed to create pipes"); return -1; @@ -529,13 +542,21 @@ static void print_noise(struct perf_evsel *evsel, double avg) print_noise_pct(stddev_stats(&ps->res_stats[0]), avg); } -static void nsec_printout(int cpu, struct perf_evsel *evsel, double avg) +static void nsec_printout(int cpu, int nr, struct perf_evsel *evsel, double avg) { double msecs = avg / 1e6; char cpustr[16] = { '\0', }; const char *fmt = csv_output ? "%s%.6f%s%s" : "%s%18.6f%s%-25s"; - if (no_aggr) + if (aggr_socket) + sprintf(cpustr, "S%*d%s%*d%s", + csv_output ? 0 : -5, + cpu, + csv_sep, + csv_output ? 0 : 4, + nr, + csv_sep); + else if (no_aggr) sprintf(cpustr, "CPU%*d%s", csv_output ? 0 : -4, perf_evsel__cpus(evsel)->map[cpu], csv_sep); @@ -734,7 +755,7 @@ static void print_ll_cache_misses(int cpu, fprintf(output, " of all LL-cache hits "); } -static void abs_printout(int cpu, struct perf_evsel *evsel, double avg) +static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg) { double total, ratio = 0.0; char cpustr[16] = { '\0', }; @@ -747,7 +768,15 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg) else fmt = "%s%18.0f%s%-25s"; - if (no_aggr) + if (aggr_socket) + sprintf(cpustr, "S%*d%s%*d%s", + csv_output ? 0 : -5, + cpu, + csv_sep, + csv_output ? 0 : 4, + nr, + csv_sep); + else if (no_aggr) sprintf(cpustr, "CPU%*d%s", csv_output ? 0 : -4, perf_evsel__cpus(evsel)->map[cpu], csv_sep); @@ -853,6 +882,70 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg) } } +static void print_aggr_socket(char *prefix) +{ + struct perf_evsel *counter; + u64 ena, run, val; + int cpu, s, s2, sock, nr; + + if (!sock_map) + return; + + for (s = 0; s < sock_map->nr; s++) { + sock = cpu_map__socket(sock_map, s); + list_for_each_entry(counter, &evsel_list->entries, node) { + val = ena = run = 0; + nr = 0; + for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) { + s2 = cpu_map__get_socket(evsel_list->cpus, cpu); + if (s2 != sock) + continue; + val += counter->counts->cpu[cpu].val; + ena += counter->counts->cpu[cpu].ena; + run += counter->counts->cpu[cpu].run; + nr++; + } + if (prefix) + fprintf(output, "%s", prefix); + + if (run == 0 || ena == 0) { + fprintf(output, "S%*d%s%*d%s%*s%s%*s", + csv_output ? 0 : -5, + s, + csv_sep, + csv_output ? 0 : 4, + nr, + csv_sep, + csv_output ? 0 : 18, + counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED, + csv_sep, + csv_output ? 0 : -24, + perf_evsel__name(counter)); + if (counter->cgrp) + fprintf(output, "%s%s", + csv_sep, counter->cgrp->name); + + fputc('\n', output); + continue; + } + + if (nsec_counter(counter)) + nsec_printout(sock, nr, counter, val); + else + abs_printout(sock, nr, counter, val); + + if (!csv_output) { + print_noise(counter, 1.0); + + if (run != ena) + fprintf(output, " (%.2f%%)", + 100.0 * run / ena); + } + fputc('\n', output); + } + } +} + /* * Print out the results of a single counter: * aggregated counts in system-wide mode @@ -882,9 +975,9 @@ static void print_counter_aggr(struct perf_evsel *counter, char *prefix) } if (nsec_counter(counter)) - nsec_printout(-1, counter, avg); + nsec_printout(-1, 0, counter, avg); else - abs_printout(-1, counter, avg); + abs_printout(-1, 0, counter, avg); print_noise(counter, avg); @@ -940,9 +1033,9 @@ static void print_counter(struct perf_evsel *counter, char *prefix) } if (nsec_counter(counter)) - nsec_printout(cpu, counter, val); + nsec_printout(cpu, 0, counter, val); else - abs_printout(cpu, counter, val); + abs_printout(cpu, 0, counter, val); if (!csv_output) { print_noise(counter, 1.0); @@ -980,7 +1073,9 @@ static void print_stat(int argc, const char **argv) fprintf(output, ":\n\n"); } - if (no_aggr) { + if (aggr_socket) + print_aggr_socket(NULL); + else if (no_aggr) { list_for_each_entry(counter, &evsel_list->entries, node) print_counter(counter, NULL); } else { @@ -1228,6 +1323,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) "command to run after to the measured command"), OPT_UINTEGER('I', "interval-print", &interval, "print counts at regular interval in ms (>= 100)"), + OPT_BOOLEAN(0, "aggr-socket", &aggr_socket, "aggregate counts per processor socket"), OPT_END() }; const char * const stat_usage[] = { @@ -1314,6 +1410,14 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) usage_with_options(stat_usage, options); } + if (aggr_socket) { + if (!perf_target__has_cpu(&target)) { + fprintf(stderr, "--aggr-socket only available in system-wide mode (-a)\n"); + usage_with_options(stat_usage, options); + } + no_aggr = true; + } + if (add_default_attributes()) goto out; -- cgit v1.2.3