lkml.org 
[lkml]   [2019]   [Apr]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH] perf/report: [RFC] Handling OOM in perf report
On Mon, Apr 01, 2019 at 04:20:00PM +0200, Thomas Richter wrote:

SNIP

> perf_session__process_event() returns to its caller, where -ENOMEM is
> changed to -EINVAL and processing stops:
>
> if ((skip = perf_session__process_event(session, event, head)) < 0) {
> pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
> head, event->header.size, event->header.type);
> err = -EINVAL;
> goto out_err;
> }
>
> This occured in the FINISHED_ROUND event when it has to process some
> 10000 entries and ran out of memory.
>
> I understand that my perf.data file might just be too big, but I would
> like to see some error message indicating OOM error instead of
> processing failure of a unrelated event.

you can limit the size of the report queue via ~/.perfconfig file:

[report]
queue-size=1M

above should use only 1M for the queue management data the data
for sample still get allocated thought.. but it could help

>
> However this patch just does what the pr_debug() statement indicates,
> the event is skipped and processing continues.
> But at least the root cause is indicated and also shows up in the
> GUI.
>
> Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
> ---
> tools/perf/builtin-report.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index 4054eb1f98ac..7a27b815f7a8 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -283,8 +283,13 @@ static int process_sample_event(struct perf_tool *tool,
> al.map->dso->hit = 1;
>
> ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
> - if (ret < 0)
> + if (ret < 0) {
> pr_debug("problem adding hist entry, skipping event\n");
> + if (ret == -ENOMEM) {
> + pr_err("Running out of memory\n");
> + ret = 0;
> + }
> + }


I think we can propagate the error completely (like below),
and even warn about ENOMEM specificaly

but I think we should bail out in case of ENOMEM, because
the data are incomplete and misleading

jirka


---
diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
index b17f1c9bc965..eea247a26ad8 100644
--- a/tools/perf/util/session.c
+++ b/tools/perf/util/session.c
@@ -1933,7 +1933,7 @@ reader__process_events(struct reader *rd, struct perf_session *session,
pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
file_offset + head, event->header.size,
event->header.type);
- err = -EINVAL;
+ err = skip;
goto out;
}

\
 
 \ /
  Last update: 2019-04-09 12:42    [W:0.080 / U:0.844 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site