| Message ID | 1655448052-12537-1-git-send-email-chensong_2000@189.cn |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v2] sched_deadline/cyclicdeadline add tracelimit | expand |
Hi, On Fri, Jun 17, 2022 at 02:40:52PM +0800, Song Chen wrote: > static void usage(int error) > @@ -689,6 +697,7 @@ static void usage(int error) > " (default 500us).\n" > "-t NUM --threads The number of threads to run as deadline (default 1).\n" > "-q --quiet print a summary only on exit\n" > + "-b USEC --breaktrace=USEC send break trace command when latency > USEC\n" Please also consider to update the man pages. Daniel
On Wed, 29 Jun 2022, Daniel Wagner wrote: > Hi, > > On Fri, Jun 17, 2022 at 02:40:52PM +0800, Song Chen wrote: > > static void usage(int error) > > @@ -689,6 +697,7 @@ static void usage(int error) > > " (default 500us).\n" > > "-t NUM --threads The number of threads to run as deadline (default 1).\n" > > "-q --quiet print a summary only on exit\n" > > + "-b USEC --breaktrace=USEC send break trace command when latency > USEC\n" > > Please also consider to update the man pages. @Song Chen, You can send that as an extra patch, but this is a good point, please update the man page as well. @Danel Wagner, is that just a comment or would you like a Reviewed-by ?
On Wed, Jun 29, 2022 at 11:34:40AM -0400, John Kacur wrote:
> @Danel Wagner, is that just a comment or would you like a Reviewed-by ?
I just noticed that several option have been added to various tools an
no one cared to add them to the manual patches. Makes a bit sad after
spending time syncing all man pages up.
On Wed, 29 Jun 2022, Daniel Wagner wrote: > On Wed, Jun 29, 2022 at 11:34:40AM -0400, John Kacur wrote: > > @Danel Wagner, is that just a comment or would you like a Reviewed-by ? > > I just noticed that several option have been added to various tools an > no one cared to add them to the manual patches. Makes a bit sad after > spending time syncing all man pages up. > Thanks Daniel I should start requiring that before I pull in patches. John
Hi, 在 2022/6/29 23:34, John Kacur 写道: > > > On Wed, 29 Jun 2022, Daniel Wagner wrote: > >> Hi, >> >> On Fri, Jun 17, 2022 at 02:40:52PM +0800, Song Chen wrote: >>> static void usage(int error) >>> @@ -689,6 +697,7 @@ static void usage(int error) >>> " (default 500us).\n" >>> "-t NUM --threads The number of threads to run as deadline (default 1).\n" >>> "-q --quiet print a summary only on exit\n" >>> + "-b USEC --breaktrace=USEC send break trace command when latency > USEC\n" >> >> Please also consider to update the man pages. > > @Song Chen, You can send that as an extra patch, but this is a good point, > please update the man page as well. > > @Danel Wagner, is that just a comment or would you like a Reviewed-by ? I will send v3 with man page update and "Reviewed-by" today, many thanks. Song > >
Hi, 在 2022/6/30 01:03, Daniel Wagner 写道: > On Wed, Jun 29, 2022 at 11:34:40AM -0400, John Kacur wrote: >> @Danel Wagner, is that just a comment or would you like a Reviewed-by ? > > I just noticed that several option have been added to various tools an > no one cared to add them to the manual patches. Makes a bit sad after > spending time syncing all man pages up. I totally agree with you in this point, in a lot of cases, document updates are behind the code implementation. people got confused if they are not synced. Thanks for pointing it out. BR Song >
diff --git a/src/sched_deadline/cyclicdeadline.c b/src/sched_deadline/cyclicdeadline.c index d865fa3..f0f4008 100644 --- a/src/sched_deadline/cyclicdeadline.c +++ b/src/sched_deadline/cyclicdeadline.c @@ -80,7 +80,11 @@ struct sched_data { }; static int shutdown; - +static int tracelimit = 0; +static int trace_marker = 0; +static pthread_mutex_t break_thread_id_lock = PTHREAD_MUTEX_INITIALIZER; +static pid_t break_thread_id = 0; +static uint64_t break_thread_value = 0; static pthread_barrier_t barrier; static int cpu_count; @@ -667,6 +671,10 @@ static void teardown(void) destroy_cpuset(CPUSET_ALL, 0); destroy_cpuset(CPUSET_LOCAL, 1); + + /* close any tracer file descriptors */ + disable_trace_mark(); + } static void usage(int error) @@ -689,6 +697,7 @@ static void usage(int error) " (default 500us).\n" "-t NUM --threads The number of threads to run as deadline (default 1).\n" "-q --quiet print a summary only on exit\n" + "-b USEC --breaktrace=USEC send break trace command when latency > USEC\n" ); exit(error); } @@ -825,6 +834,18 @@ void *run_deadline(void *data) while (!shutdown) { period = do_runtime(tid, sd, period); + if (tracelimit && (stat->max > tracelimit)) { + shutdown++; + pthread_mutex_lock(&break_thread_id_lock); + if (break_thread_id == 0) { + break_thread_id = stat->tid; + break_thread_value = stat->max; + tracemark( "hit latency threshold (%lld > %d)", + (unsigned long long) stat->max, tracelimit); + } + pthread_mutex_unlock(&break_thread_id_lock); + break; + } sched_yield(); } ret = sched_getattr(0, &attr, sizeof(attr), 0); @@ -1063,9 +1084,10 @@ static void write_stats(FILE *f, void *data) fprintf(f, " }\n"); } -enum options_valud { +enum options_values { OPT_AFFINITY=1, OPT_DURATION, OPT_HELP, OPT_INTERVAL, - OPT_JSON, OPT_STEP, OPT_THREADS, OPT_QUIET + OPT_JSON, OPT_STEP, OPT_THREADS, OPT_QUIET, + OPT_BREAKTRACE, OPT_TRACEMARK, }; int main(int argc, char **argv) @@ -1104,9 +1126,11 @@ int main(int argc, char **argv) { "step", required_argument, NULL, OPT_STEP }, { "threads", required_argument, NULL, OPT_THREADS }, { "quiet", no_argument, NULL, OPT_QUIET }, + { "breaktrace", required_argument, NULL, OPT_BREAKTRACE }, + { "tracemark", no_argument, NULL, OPT_TRACEMARK }, { NULL, 0, NULL, 0 }, }; - c = getopt_long(argc, argv, "a::c:D:hi:s:t:q", options, NULL); + c = getopt_long(argc, argv, "a::c:D:hi:s:t:b:q", options, NULL); if (c == -1) break; switch (c) { @@ -1141,6 +1165,10 @@ int main(int argc, char **argv) case 'D': duration = parse_time_string(optarg); break; + case 'b': + case OPT_BREAKTRACE: + tracelimit = atoi(optarg); + break; case OPT_QUIET: case 'q': quiet = 1; @@ -1149,6 +1177,9 @@ int main(int argc, char **argv) case 'h': usage(0); break; + case OPT_TRACEMARK: + trace_marker = 1; + break; default: usage(1); } @@ -1186,6 +1217,8 @@ int main(int argc, char **argv) warn("mlockall"); setup_ftrace_marker(); + if (tracelimit && trace_marker) + enable_trace_mark(); thread = calloc(nr_threads, sizeof(*thread)); sched_data = calloc(nr_threads, sizeof(*sched_data)); @@ -1294,6 +1327,13 @@ int main(int argc, char **argv) loop(sched_data, nr_threads); + if (tracelimit) { + if (break_thread_id) { + printf("# Break thread: %d\n", break_thread_id); + printf("# Break value: %llu\n", (unsigned long long)break_thread_value); + } + } + for (i = 0; i < nr_threads; i++) { sd = &sched_data[i];
cyclictest has tracelimit to stop running when max latency is higher than threshold but cyclicdeadline doesn't. Therefore, tracelimit is introduced to cyclicdeadline in this commit, once max latency is over, test stops and log prints at the bottom of the ftrace file. Signed-off-by: Song Chen <chensong_2000@189.cn> --- v2: 1, disable ftrace when hitting thresold --- src/sched_deadline/cyclicdeadline.c | 48 ++++++++++++++++++++++++++--- 1 file changed, 44 insertions(+), 4 deletions(-)