Re: [PATCH v3] perf bench sched pipe: Add -G/--cgroups option

[Athira Rajeev <atrajeev@linux.vnet.ibm.com>]

> On 16-Oct-2023, at 10:12 AM, Namhyung Kim <namhyung@kernel.org> wrote:
> 
> The -G/--cgroups option is to put sender and receiver in different
> cgroups in order to measure cgroup context switch overheads.
> 
> Users need to make sure the cgroups exist and accessible.
> 
>  # perf stat -e context-switches,cgroup-switches \
>> taskset -c 0 perf bench sched pipe -l 10000 > /dev/null
> 
>   Performance counter stats for 'taskset -c 0 perf bench sched pipe -l 10000':
> 
>              20,001      context-switches
>                   2      cgroup-switches
> 
>         0.053449651 seconds time elapsed
> 
>         0.011286000 seconds user
>         0.041869000 seconds sys
> 
>  # perf stat -e context-switches,cgroup-switches \
>> taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB > /dev/null
> 
>   Performance counter stats for 'taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB':
> 
>              20,001      context-switches
>              20,001      cgroup-switches
> 
>         0.052768627 seconds time elapsed
> 
>         0.006284000 seconds user
>         0.046266000 seconds sys
> 
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>

Hi Namhyung,

I tried V3 on top of perf-tools-next 

# ./perf stat -e context-switches,cgroup-switches taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB > /dev/null
no access to cgroup /sys/fs/cgroup/perf_event/AAA
cannot open sender cgroup: AAA
 Usage: perf bench sched pipe <options>

    -G, --cgroups <SEND,RECV>
                          Put sender and receivers in given cgroups

 Performance counter stats for 'taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB':

                 2      context-switches                                                      
                 0      cgroup-switches                                                       

       0.007291460 seconds time elapsed

       0.007438000 seconds user
       0.000000000 seconds sys


I created AAA and BBB
mkdir /sys/fs/cgroup/perf_event/AAA
mkdir /sys/fs/cgroup/perf_event/BBB

Got the results below:

./perf stat -e context-switches,cgroup-switches taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB > /dev/null

 Performance counter stats for 'taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB':

             20002      context-switches                                                      
             19999      cgroup-switches                                                       

       0.120063986 seconds time elapsed

       0.001716000 seconds user
       0.065995000 seconds sys


Thanks
Athira


> ---
> tools/perf/Documentation/perf-bench.txt |  19 ++++
> tools/perf/bench/sched-pipe.c           | 118 +++++++++++++++++++++++-
> 2 files changed, 134 insertions(+), 3 deletions(-)
> 
> diff --git a/tools/perf/Documentation/perf-bench.txt b/tools/perf/Documentation/perf-bench.txt
> index ca5789625cd2..8331bd28b10e 100644
> --- a/tools/perf/Documentation/perf-bench.txt
> +++ b/tools/perf/Documentation/perf-bench.txt
> @@ -124,6 +124,14 @@ Options of *pipe*
> --loop=::
> Specify number of loops.
> 
> +-G::
> +--cgroups=::
> +Names of cgroups for sender and receiver, separated by a comma.
> +This is useful to check cgroup context switching overhead.
> +Note that perf doesn't create nor delete the cgroups, so users should
> +make sure that the cgroups exist and are accessible before use.
> +
> +
> Example of *pipe*
> ^^^^^^^^^^^^^^^^^
> 
> @@ -141,6 +149,17 @@ Example of *pipe*
>         Total time:0.016 sec
>                 16.948000 usecs/op
>                 59004 ops/sec
> +
> +% perf bench sched pipe -G AAA,BBB
> +(executing 1000000 pipe operations between cgroups)
> +# Running 'sched/pipe' benchmark:
> +# Executed 1000000 pipe operations between two processes
> +
> +     Total time: 6.886 [sec]
> +
> +       6.886208 usecs/op
> +         145217 ops/sec
> +
> ---------------------
> 
> SUITES FOR 'syscall'
> diff --git a/tools/perf/bench/sched-pipe.c b/tools/perf/bench/sched-pipe.c
> index a960e7a93aec..88d20a34adb2 100644
> --- a/tools/perf/bench/sched-pipe.c
> +++ b/tools/perf/bench/sched-pipe.c
> @@ -10,7 +10,9 @@
>  * Ported to perf by Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
>  */
> #include <subcmd/parse-options.h>
> +#include <api/fs/fs.h>
> #include "bench.h"
> +#include "util/cgroup.h"
> 
> #include <unistd.h>
> #include <stdio.h>
> @@ -19,6 +21,7 @@
> #include <sys/wait.h>
> #include <string.h>
> #include <errno.h>
> +#include <fcntl.h>
> #include <assert.h>
> #include <sys/time.h>
> #include <sys/types.h>
> @@ -31,6 +34,7 @@ struct thread_data {
> int nr;
> int pipe_read;
> int pipe_write;
> + bool cgroup_failed;
> pthread_t pthread;
> };
> 
> @@ -40,9 +44,55 @@ static int loops = LOOPS_DEFAULT;
> /* Use processes by default: */
> static bool threaded;
> 
> +static struct cgroup *cgrp_send = NULL;
> +static struct cgroup *cgrp_recv = NULL;
> +
> +static int parse_two_cgroups(const struct option *opt __maybe_unused,
> +     const char *str, int unset __maybe_unused)
> +{
> + char *p = strdup(str);
> + char *q;
> + int ret = -1;
> +
> + if (p == NULL) {
> + fprintf(stderr, "memory allocation failure");
> + return -1;
> + }
> +
> + q = strchr(p, ',');
> + if (q == NULL) {
> + fprintf(stderr, "it should have two cgroup names: %s", p);
> + goto out;
> + }
> + *q = '\0';
> +
> + cgrp_send = cgroup__new(p, /*do_open=*/true);
> + if (cgrp_send == NULL) {
> + fprintf(stderr, "cannot open sender cgroup: %s", p);
> + goto out;
> + }
> +
> + /* skip ',' */
> + q++;
> +
> + cgrp_recv = cgroup__new(q, /*do_open=*/true);
> + if (cgrp_recv == NULL) {
> + fprintf(stderr, "cannot open receiver cgroup: %s", q);
> + goto out;
> + }
> + ret = 0;
> +
> +out:
> + free(p);
> + return ret;
> +}
> +
> static const struct option options[] = {
> OPT_INTEGER('l', "loop", &loops, "Specify number of loops"),
> OPT_BOOLEAN('T', "threaded", &threaded, "Specify threads/process based task setup"),
> + OPT_CALLBACK('G', "cgroups", NULL, "SEND,RECV",
> +     "Put sender and receivers in given cgroups",
> +     parse_two_cgroups),
> OPT_END()
> };
> 
> @@ -51,12 +101,71 @@ static const char * const bench_sched_pipe_usage[] = {
> NULL
> };
> 
> +static int enter_cgroup(struct cgroup *cgrp)
> +{
> + char buf[32];
> + int fd, len, ret;
> + pid_t pid;
> +
> + if (cgrp == NULL)
> + return 0;
> +
> + if (threaded)
> + pid = syscall(__NR_gettid);
> + else
> + pid = getpid();
> +
> + snprintf(buf, sizeof(buf), "%d\n", pid);
> + len = strlen(buf);
> +
> + /* try cgroup v2 interface first */
> + if (threaded)
> + fd = openat(cgrp->fd, "cgroup.threads", O_WRONLY);
> + else
> + fd = openat(cgrp->fd, "cgroup.procs", O_WRONLY);
> +
> + /* try cgroup v1 if failed */
> + if (fd < 0)
> + fd = openat(cgrp->fd, "tasks", O_WRONLY);
> +
> + if (fd < 0) {
> + char mnt[PATH_MAX];
> +
> + printf("Failed to open cgroup file in %s\n", cgrp->name);
> +
> + if (cgroupfs_find_mountpoint(mnt, sizeof(mnt), "perf_event") == 0)
> + printf(" Hint: create the cgroup first, like 'mkdir %s/%s'\n",
> +       mnt, cgrp->name);
> + return -1;
> + }
> +
> + ret = write(fd, buf, len);
> + close(fd);
> +
> + if (ret != len) {
> + printf("Cannot enter to cgroup: %s\n", cgrp->name);
> + return -1;
> + }
> + return 0;
> +}
> +
> static void *worker_thread(void *__tdata)
> {
> struct thread_data *td = __tdata;
> int m = 0, i;
> int ret;
> 
> + if (td->nr)
> + ret = enter_cgroup(cgrp_send);
> + else
> + ret = enter_cgroup(cgrp_recv);
> +
> + if (ret < 0) {
> + td->cgroup_failed = true;
> + return NULL;
> + }
> + td->cgroup_failed = false;
> +
> for (i = 0; i < loops; i++) {
> if (!td->nr) {
> ret = read(td->pipe_read, &m, sizeof(int));
> @@ -112,9 +221,7 @@ int bench_sched_pipe(int argc, const char **argv)
> }
> }
> 
> -
> if (threaded) {
> -
> for (t = 0; t < nr_threads; t++) {
> td = threads + t;
> 
> @@ -128,7 +235,6 @@ int bench_sched_pipe(int argc, const char **argv)
> ret = pthread_join(td->pthread, NULL);
> BUG_ON(ret);
> }
> -
> } else {
> pid = fork();
> assert(pid >= 0);
> @@ -147,6 +253,12 @@ int bench_sched_pipe(int argc, const char **argv)
> gettimeofday(&stop, NULL);
> timersub(&stop, &start, &diff);
> 
> + cgroup__put(cgrp_send);
> + cgroup__put(cgrp_recv);
> +
> + if (threads[0].cgroup_failed || threads[1].cgroup_failed)
> + return 0;
> +
> switch (bench_format) {
> case BENCH_FORMAT_DEFAULT:
> printf("# Executed %d pipe operations between two %s\n\n",
> -- 
> 2.42.0.655.g421f12c284-goog
>