If the perf record command is interrupted in record__mmap_read_all
function, the 'done' is set and err has the latest poll return
value, which is most likely positive number (= number of pollfds
ready to read).
This 'positive err' is then propagated to the exit code, resulting
in not finishing the perf.data header properly, causing following
error in report:
# perf record -F 50000 -a
---
make the system real busy, so there's more chance
to interrupt perf in event writing code
---
^C[ perf record: Woken up 16 times to write data ]
[ perf record: Captured and wrote 30.292 MB perf.data (~1323468 samples) ]
# perf report --stdio > /dev/null
WARNING: The perf.data file's data size field is 0 which is unexpected.
Was the 'perf record' command properly terminated?
Fixing this by checking for positive poll return value
and setting err to 0.
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Corey Ashford <[email protected]>
Cc: David Ahern <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Signed-off-by: Jiri Olsa <[email protected]>
---
tools/perf/builtin-record.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index e4c85b8..ce2cfec 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -454,7 +454,11 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
if (done)
break;
err = poll(rec->evlist->pollfd, rec->evlist->nr_fds, -1);
- if (err < 0 && errno == EINTR)
+ /*
+ * Propagate error, only if there's any. Ignore positive
+ * number of returned events and interrupt error.
+ */
+ if (err > 0 || (err < 0 && errno == EINTR))
err = 0;
waking++;
}
--
1.8.3.1
Em Mon, Jun 02, 2014 at 08:02:06PM +0200, Jiri Olsa escreveu:
> If the perf record command is interrupted in record__mmap_read_all
> function, the 'done' is set and err has the latest poll return
> value, which is most likely positive number (= number of pollfds
> ready to read).
>
> This 'positive err' is then propagated to the exit code, resulting
> in not finishing the perf.data header properly, causing following
> error in report:
>
> # perf record -F 50000 -a
Acked-by: Arnaldo Carvalho de Melo <[email protected]>
On Mon, 2 Jun 2014 20:02:06 +0200, Jiri Olsa wrote:
> If the perf record command is interrupted in record__mmap_read_all
> function, the 'done' is set and err has the latest poll return
> value, which is most likely positive number (= number of pollfds
> ready to read).
>
> This 'positive err' is then propagated to the exit code, resulting
> in not finishing the perf.data header properly, causing following
> error in report:
>
> # perf record -F 50000 -a
>
> ---
> make the system real busy, so there's more chance
> to interrupt perf in event writing code
> ---
>
> ^C[ perf record: Woken up 16 times to write data ]
> [ perf record: Captured and wrote 30.292 MB perf.data (~1323468 samples) ]
>
> # perf report --stdio > /dev/null
> WARNING: The perf.data file's data size field is 0 which is unexpected.
> Was the 'perf record' command properly terminated?
>
> Fixing this by checking for positive poll return value
> and setting err to 0.
Acked-by: Namhyung Kim <[email protected]>
Thanks,
Namhyung
Commit-ID: a515114fa3cff8f1da10cd68914d55c10879c3e0
Gitweb: http://git.kernel.org/tip/a515114fa3cff8f1da10cd68914d55c10879c3e0
Author: Jiri Olsa <[email protected]>
AuthorDate: Mon, 2 Jun 2014 13:44:23 -0400
Committer: Jiri Olsa <[email protected]>
CommitDate: Tue, 3 Jun 2014 21:35:05 +0200
perf record: Fix poll return value propagation
If the perf record command is interrupted in record__mmap_read_all
function, the 'done' is set and err has the latest poll return
value, which is most likely positive number (= number of pollfds
ready to read).
This 'positive err' is then propagated to the exit code, resulting
in not finishing the perf.data header properly, causing following
error in report:
# perf record -F 50000 -a
---
make the system real busy, so there's more chance
to interrupt perf in event writing code
---
^C[ perf record: Woken up 16 times to write data ]
[ perf record: Captured and wrote 30.292 MB perf.data (~1323468 samples) ]
# perf report --stdio > /dev/null
WARNING: The perf.data file's data size field is 0 which is unexpected.
Was the 'perf record' command properly terminated?
Fixing this by checking for positive poll return value
and setting err to 0.
Acked-by: Arnaldo Carvalho de Melo <[email protected]>
Acked-by: Namhyung Kim <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Corey Ashford <[email protected]>
Cc: David Ahern <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Jiri Olsa <[email protected]>
---
tools/perf/builtin-record.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index e4c85b8..ce2cfec 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -454,7 +454,11 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
if (done)
break;
err = poll(rec->evlist->pollfd, rec->evlist->nr_fds, -1);
- if (err < 0 && errno == EINTR)
+ /*
+ * Propagate error, only if there's any. Ignore positive
+ * number of returned events and interrupt error.
+ */
+ if (err > 0 || (err < 0 && errno == EINTR))
err = 0;
waking++;
}