Re: [PATCH 1/3] perf tools: Use /proc/<PID>/task/<TID>/status for synthesis

[Jiri Olsa <jolsa@redhat.com>]

On Fri, Jan 29, 2021 at 02:48:59PM +0900, Namhyung Kim wrote:
> To save memory usage, it needs to reduce number of entries in the proc
> filesystem.  It's using /proc/<PID>/task directory to traverse threads
> in the process and then kernel creates /proc/<PID>/task/<TID> entries.
> 
> After that it checks the thread info using the /proc/<TID>/status file
> rather than /proc/<PID>/task/<TID>/status.  As far as I can see, they
> are the same and contain all the info we need.
> 
> Using the latter eliminates the unnecessary /proc/<TID> entry.  This
> can be useful especially a large number of threads are used in the
> system.  In my experiment around 1KB of memory on average was saved
> for each thread (which is not a thread group leader).
> 
> To do this, pass both pid and tid to perf_event_prepare_comm() if it
> knows them.  In case it doesn't know, passing 0 as pid will do the old
> way.
> 
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> ---
>  tools/perf/util/synthetic-events.c | 30 +++++++++++++++++++-----------
>  1 file changed, 19 insertions(+), 11 deletions(-)
> 
> diff --git a/tools/perf/util/synthetic-events.c b/tools/perf/util/synthetic-events.c
> index 3a898520f05c..800522591dde 100644
> --- a/tools/perf/util/synthetic-events.c
> +++ b/tools/perf/util/synthetic-events.c
> @@ -69,7 +69,7 @@ int perf_tool__process_synth_event(struct perf_tool *tool,
>   * Assumes that the first 4095 bytes of /proc/pid/stat contains
>   * the comm, tgid and ppid.
>   */
> -static int perf_event__get_comm_ids(pid_t pid, char *comm, size_t len,
> +static int perf_event__get_comm_ids(pid_t pid, pid_t tid, char *comm, size_t len,
>  				    pid_t *tgid, pid_t *ppid)
>  {
>  	char bf[4096];
> @@ -81,7 +81,10 @@ static int perf_event__get_comm_ids(pid_t pid, char *comm, size_t len,
>  	*tgid = -1;
>  	*ppid = -1;
>  
> -	snprintf(bf, sizeof(bf), "/proc/%d/status", pid);
> +	if (pid)
> +		snprintf(bf, sizeof(bf), "/proc/%d/task/%d/status", pid, tid);
> +	else
> +		snprintf(bf, sizeof(bf), "/proc/%d/status", tid);
>  
>  	fd = open(bf, O_RDONLY);
>  	if (fd < 0) {
> @@ -93,7 +96,7 @@ static int perf_event__get_comm_ids(pid_t pid, char *comm, size_t len,
>  	close(fd);
>  	if (n <= 0) {
>  		pr_warning("Couldn't get COMM, tigd and ppid for pid %d\n",
> -			   pid);
> +			   tid);
>  		return -1;
>  	}
>  	bf[n] = '\0';
> @@ -116,27 +119,32 @@ static int perf_event__get_comm_ids(pid_t pid, char *comm, size_t len,
>  		memcpy(comm, name, size);
>  		comm[size] = '\0';
>  	} else {
> -		pr_debug("Name: string not found for pid %d\n", pid);
> +		pr_debug("Name: string not found for pid %d\n", tid);
>  	}
>  
>  	if (tgids) {
>  		tgids += 5;  /* strlen("Tgid:") */
>  		*tgid = atoi(tgids);
> +
> +		if (pid && pid != *tgid) {
> +			pr_debug("Tgid: not match to given pid: %d vs %d\n",
> +				 pid, *tgid);

hm, could this actually happen in our case?

jirka