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
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri, 29 Mar 2024 17:25:15 +0000
From: "Wang, Weilin" <weilin.wang@...el.com>
To: Namhyung Kim <namhyung@...nel.org>
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>, "Hunter, Adrian" <adrian.hunter@...el.com>, Kan
 Liang <kan.liang@...ux.intel.com>, "linux-perf-users@...r.kernel.org"
	<linux-perf-users@...r.kernel.org>, "linux-kernel@...r.kernel.org"
	<linux-kernel@...r.kernel.org>, "Taylor, Perry" <perry.taylor@...el.com>,
	"Alt, Samantha" <samantha.alt@...el.com>, "Biggers, Caleb"
	<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.



> -----Original Message-----
> From: Namhyung Kim <namhyung@...nel.org>
> Sent: Thursday, March 28, 2024 11:43 PM
> To: Wang, Weilin <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>; Hunter,
> Adrian <adrian.hunter@...el.com>; Kan Liang <kan.liang@...ux.intel.com>;
> linux-perf-users@...r.kernel.org; linux-kernel@...r.kernel.org; Taylor, Perry
> <perry.taylor@...el.com>; Alt, Samantha <samantha.alt@...el.com>; Biggers,
> Caleb <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.

This will require a little code change in function check_pipe() in data.c. 
Do you want me to update that?

Thanks,
Weilin

> 
> 
> >         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

Powered by Openwall GNU/*/Linux Powered by OpenVZ