| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAM9d7cjXE_XgZAGnr9byL17KbgKrum+kyozR5yTVkKvcUd8wow@mail.gmail.com>
Date: Thu, 28 Mar 2024 23:43:22 -0700
From: Namhyung Kim <namhyung@...nel.org>
To: weilin.wang@...el.com
Cc: Ian Rogers <irogers@...gle.com>, Arnaldo Carvalho de Melo <acme@...nel.org>,
Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>, Jiri Olsa <jolsa@...nel.org>,
Adrian Hunter <adrian.hunter@...el.com>, Kan Liang <kan.liang@...ux.intel.com>,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
Perry Taylor <perry.taylor@...el.com>, Samantha Alt <samantha.alt@...el.com>,
Caleb Biggers <caleb.biggers@...el.com>
Subject: Re: [RFC PATCH v5 2/6] perf stat: Fork and launch perf record when
perf stat needs to get retire latency value for a metric.
On Tue, Mar 26, 2024 at 1:29 PM <weilin.wang@...el.com> wrote:
>
> From: Weilin Wang <weilin.wang@...el.com>
>
> When retire_latency value is used in a metric formula, perf stat would fork a
> perf record process with "-e" and "-W" options. Perf record will collect
> required retire_latency values in parallel while perf stat is collecting
> counting values.
>
> At the point of time that perf stat stops counting, it would send sigterm signal
> to perf record process and receiving sampling data back from perf record from a
> pipe. Perf stat will then process the received data to get retire latency data
> and calculate metric result.
I'm afraid this requirement is too Intel specific. Maybe we need to move
the code to somewhere under the arch directory.
>
> Signed-off-by: Weilin Wang <weilin.wang@...el.com>
> Reviewed-by: Ian Rogers <irogers@...gle.com>
> ---
> tools/perf/builtin-stat.c | 165 +++++++++++++++++++++++++++++++++-
> tools/perf/util/data.c | 3 +
> tools/perf/util/data.h | 5 ++
> tools/perf/util/metricgroup.h | 8 ++
> tools/perf/util/stat.h | 2 +
> 5 files changed, 181 insertions(+), 2 deletions(-)
>
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index 6291e1e24535..fb08cef42d95 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -94,8 +94,13 @@
> #include <perf/evlist.h>
> #include <internal/threadmap.h>
>
> +#include "util/sample.h"
> +#include <sys/param.h>
> +#include <subcmd/run-command.h>
> +
> #define DEFAULT_SEPARATOR " "
> #define FREEZE_ON_SMI_PATH "devices/cpu/freeze_on_smi"
> +#define PERF_DATA "-"
>
> static void print_counters(struct timespec *ts, int argc, const char **argv);
>
> @@ -163,6 +168,8 @@ static struct perf_stat_config stat_config = {
> .ctl_fd_ack = -1,
> .iostat_run = false,
> .tpebs_events = LIST_HEAD_INIT(stat_config.tpebs_events),
> + .tpebs_results = LIST_HEAD_INIT(stat_config.tpebs_results),
> + .tpebs_pid = -1,
> };
>
> static bool cpus_map_matched(struct evsel *a, struct evsel *b)
> @@ -687,12 +694,155 @@ static enum counter_recovery stat_handle_error(struct evsel *counter)
> return COUNTER_FATAL;
> }
>
> -static int __run_perf_record(void)
> +static int __run_perf_record(const char **record_argv)
> {
> + int i = 0;
> + struct tpebs_event *e;
> +
> pr_debug("Prepare perf record for retire_latency\n");
> +
> + record_argv[i++] = "perf";
> + record_argv[i++] = "record";
> + record_argv[i++] = "-W";
> + record_argv[i++] = "--synth=no";
> +
> + if (stat_config.user_requested_cpu_list) {
> + record_argv[i++] = "-C";
> + record_argv[i++] = stat_config.user_requested_cpu_list;
> + }
> +
> + if (stat_config.system_wide)
> + record_argv[i++] = "-a";
What do you want to do if neither of -a nor -C is used?
Maybe you can disable perf record in background.
> +
> + list_for_each_entry(e, &stat_config.tpebs_events, nd) {
> + record_argv[i++] = "-e";
> + record_argv[i++] = e->name;
> + }
> +
> + record_argv[i++] = "-o";
> + record_argv[i++] = PERF_DATA;
> +
> return 0;
> }
>
> +static void prepare_run_command(struct child_process *cmd,
> + const char **argv)
> +{
> + memset(cmd, 0, sizeof(*cmd));
> + cmd->argv = argv;
> + cmd->out = -1;
> +}
> +
> +static int prepare_perf_record(struct child_process *cmd)
> +{
> + const char **record_argv;
> +
> + record_argv = calloc(10 + 2 * stat_config.tpebs_event_size, sizeof(char *));
> + if (!record_argv)
> + return -1;
> + __run_perf_record(record_argv);
> +
> + prepare_run_command(cmd, record_argv);
> + return start_command(cmd);
> +}
> +
> +struct perf_script {
> + struct perf_tool tool;
> + struct perf_session *session;
> +};
> +
> +static void tpebs_data__delete(void)
> +{
> + struct tpebs_retire_lat *r, *rtmp;
> + struct tpebs_event *e, *etmp;
A blank line please.
> + list_for_each_entry_safe(r, rtmp, &stat_config.tpebs_results, event.nd) {
> + list_del_init(&r->event.nd);
> + free(r);
> + }
> + list_for_each_entry_safe(e, etmp, &stat_config.tpebs_events, nd) {
> + list_del_init(&e->nd);
> + free(e);
> + }
> +}
> +
> +static int process_sample_event(struct perf_tool *tool __maybe_unused,
> + union perf_event *event __maybe_unused,
> + struct perf_sample *sample,
> + struct evsel *evsel,
> + struct machine *machine __maybe_unused)
> +{
> + int ret = 0;
> + const char *evname;
> + struct tpebs_retire_lat *t;
> +
> + evname = evsel__name(evsel);
> +
> + /*
> + * Need to handle per core results? We are assuming average retire
> + * latency value will be used. Save the number of samples and the sum of
> + * retire latency value for each event.
> + */
> + list_for_each_entry(t, &stat_config.tpebs_results, event.nd) {
> + if (!strcmp(evname, t->event.name)) {
> + t->count += 1;
> + t->sum += sample->retire_lat;
> + break;
> + }
> + }
> +
> + return ret;
> +}
> +
> +static int process_feature_event(struct perf_session *session,
> + union perf_event *event)
> +{
> + if (event->feat.feat_id < HEADER_LAST_FEATURE)
> + return perf_event__process_feature(session, event);
> + return 0;
> +}
> +
> +static int __cmd_script(struct child_process *cmd __maybe_unused)
> +{
> + int err = 0;
> + struct perf_session *session;
> + struct perf_data data = {
> + .mode = PERF_DATA_MODE_READ,
> + .path = PERF_DATA,
> + .fd = cmd->out,
I don't like this part as it added the fd field unnecessarily.
I think we need a new session API that can pass a fd directly.
> + };
> + struct perf_script script = {
> + .tool = {
> + .sample = process_sample_event,
> + .ordering_requires_timestamps = true,
Is this really needed? I don't think it needs ordering.
> + .feature = process_feature_event,
> + .attr = perf_event__process_attr,
> + },
> + };
> + struct tpebs_event *e;
> +
> + list_for_each_entry(e, &stat_config.tpebs_events, nd) {
> + struct tpebs_retire_lat *new = malloc(sizeof(struct tpebs_retire_lat));
> +
> + if (!new)
> + return -1;
> + new->event.name = strdup(e->name);
> + new->event.tpebs_name = strdup(e->tpebs_name);
> + new->count = 0;
> + new->sum = 0;
> + list_add_tail(&new->event.nd, &stat_config.tpebs_results);
> + }
> +
> + kill(cmd->pid, SIGTERM);
Is this the only place to stop the perf record?
What if perf stat exits before calling this function?
> + session = perf_session__new(&data, &script.tool);
> + if (IS_ERR(session))
> + return PTR_ERR(session);
> + script.session = session;
> + err = perf_session__process_events(session);
> + perf_session__delete(session);
> +
> + return err;
> +}
> +
> static int __run_perf_stat(int argc, const char **argv, int run_idx)
> {
> int interval = stat_config.interval;
> @@ -709,13 +859,15 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
> struct affinity saved_affinity, *affinity = NULL;
> int err;
> bool second_pass = false;
> + struct child_process cmd;
>
> /* Prepare perf record for sampling event retire_latency before fork and
> * prepare workload */
> if (stat_config.tpebs_event_size > 0) {
> int ret;
>
> - ret = __run_perf_record();
> + pr_debug("perf stat pid = %d\n", getpid());
> + ret = prepare_perf_record(&cmd);
> if (ret)
> return ret;
> }
> @@ -925,6 +1077,13 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
>
> t1 = rdclock();
>
> + if (stat_config.tpebs_event_size > 0) {
> + int ret;
> +
> + ret = __cmd_script(&cmd);
> + close(cmd.out);
> + }
> +
> if (stat_config.walltime_run_table)
> stat_config.walltime_run[run_idx] = t1 - t0;
>
> @@ -2972,5 +3131,7 @@ int cmd_stat(int argc, const char **argv)
> metricgroup__rblist_exit(&stat_config.metric_events);
> evlist__close_control(stat_config.ctl_fd, stat_config.ctl_fd_ack, &stat_config.ctl_fd_close);
>
> + tpebs_data__delete();
> +
> return status;
> }
> diff --git a/tools/perf/util/data.c b/tools/perf/util/data.c
> index 08c4bfbd817f..f6cc0ccdec5e 100644
> --- a/tools/perf/util/data.c
> +++ b/tools/perf/util/data.c
> @@ -185,6 +185,9 @@ static bool check_pipe(struct perf_data *data)
> int fd = perf_data__is_read(data) ?
> STDIN_FILENO : STDOUT_FILENO;
>
> + if (data->fd > 0)
> + fd = data->fd;
> +
> if (!data->path) {
> if (!fstat(fd, &st) && S_ISFIFO(st.st_mode))
> is_pipe = true;
> diff --git a/tools/perf/util/data.h b/tools/perf/util/data.h
> index 110f3ebde30f..f5e7479f1892 100644
> --- a/tools/perf/util/data.h
> +++ b/tools/perf/util/data.h
> @@ -28,6 +28,11 @@ struct perf_data_file {
>
> struct perf_data {
> const char *path;
> + /*
> + * When fd is given, use PIPE instead of file, use given fd instead of
> + * STDIN_FILENO or STDOUT_FILENO
> + */
> + int fd;
Maybe users can set data.file.fd directly instead of this.
> struct perf_data_file file;
> bool is_pipe;
> bool is_dir;
> diff --git a/tools/perf/util/metricgroup.h b/tools/perf/util/metricgroup.h
> index 7c24ed768ff3..e5bafb45d4d9 100644
> --- a/tools/perf/util/metricgroup.h
> +++ b/tools/perf/util/metricgroup.h
> @@ -68,10 +68,18 @@ struct metric_expr {
>
> struct tpebs_event {
> struct list_head nd;
> + /* Event name */
> const char *name;
> + /* Event name with TPEBS modifier */
What is the TPEBS modifier?
Thanks,
Namhyung
> const char *tpebs_name;
> };
>
> +struct tpebs_retire_lat {
> + struct tpebs_event event;
> + size_t count;
> + int sum;
> +};
> +
> struct metric_event *metricgroup__lookup(struct rblist *metric_events,
> struct evsel *evsel,
> bool create);
> diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
> index b987960df3c5..0726bdc06681 100644
> --- a/tools/perf/util/stat.h
> +++ b/tools/perf/util/stat.h
> @@ -111,6 +111,8 @@ struct perf_stat_config {
> struct rblist metric_events;
> struct list_head tpebs_events;
> size_t tpebs_event_size;
> + struct list_head tpebs_results;
> + pid_t tpebs_pid;
> int ctl_fd;
> int ctl_fd_ack;
> bool ctl_fd_close;
> --
> 2.43.0
>
Powered by blists - more mailing lists