Am Donnerstag, den 20.09.2012, 15:56 +0200 schrieb Estanislao Gonzalez:
> Am Donnerstag, den 20.09.2012, 09:29 -0400 schrieb Benjamin Smedberg:
> > On 9/20/2012 9:14 AM, Estanislao Gonzalez wrote:
> > > Hi,
> > >
> > > This is probably nothing... but I was testing something else and I
> > > wondered why there are so many calls to getrlimit about the size of the
> > > stack...
> > >
> > > These are the "top 10" calls from one second dump of strace for the
> > > xulrunner process (first field is the count):
> > >
> > > $ (strace -p 10762 2>&1 & sleep 1 && pkill strace) | awk '{a[$0]+=1} END
> > > {for (k in a) {print a[k],k}}' | sort -k 1 -rn | head
> > >
> > > 58498 getrlimit(RLIMIT_STACK, {rlim_cur=8192*1024,
> > > rlim_max=RLIM_INFINITY}) = 0
> > > 53 read(13, 0xb4d65058, 4096) = -1 EAGAIN (Resource
> > > temporarily unavailable)
> > > 26 poll([{fd=4, events=POLLIN}, {fd=13, events=POLLIN}, {fd=6,
> > > events=POLLIN}], 3, 0) = 0 (Timeout)
> > > 16 poll([{fd=4, events=POLLIN}, {fd=13, events=POLLIN}], 2, 0) = 0
> > > (Timeout)
> > > 12 semop(131073, {{0, -1, SEM_UNDO}}, 1) = 0
> > > 12 semop(131073, {{0, 1, SEM_UNDO}}, 1) = 0
> > > 6 poll([{fd=4, events=POLLIN}, {fd=13, events=POLLIN}, {fd=6,
> > > events=POLLIN}], 3, 9) = 0 (Timeout)
> > > 5 write(7, "!", 1) = 1
> > > 5 read(6, "!", 1) = 1
> > > 5 poll([{fd=4, events=POLLIN}, {fd=13, events=POLLIN}, {fd=6,
> > > events=POLLIN}], 3, 0) = 1 ([{fd=6, revents=POLLIN}])
> > >
> > > so there are almost 60.000 calls per second (min:53.000 max:64.000) to
> > > RLIMIT_STACK ...
> > This is while running Firefox? Or some other XULRunner-based app?
> >
> > The ps list below shows seems to indicate that you're seeing the
> > plugin-container process for Flash player. If this is correct, and
> > you're looking at a Flash-based video, it is *probably* the Flash player
> > code which is making these getrlimit calls. It would be good to get
> > some basic stack traces from the calls to verify my hunch, of course. If
> > they walk directly into libflashplayer.so then this is something you
> > should report to Adobe.
> >
> > Does anyone know how expensive getrlimit is?
> >
> > --BDS
> >
>
> I did asume this was caused by the flash player library (to some extent). It
> is while running iceweasel (firefox debian version).
>
> And you were right:
>
> #0 0xb65dbd7a in pthread_cond_wait@@GLIBC_2.3.2 () from /lib/libpthread.so.0
> #1 0xb320659f in ?? () from /usr/lib/flashplayer-mozilla/libflashplayer.so
> #2 0xb2dce2e5 in ?? () from /usr/lib/flashplayer-mozilla/libflashplayer.so
> #3 0xb320685c in ?? () from /usr/lib/flashplayer-mozilla/libflashplayer.so
> #4 0xb3206f06 in ?? () from /usr/lib/flashplayer-mozilla/libflashplayer.so
> #5 0xb65d77b0 in start_thread () from /lib/libpthread.so.0
> #6 0xb653b0be in clone () from /lib/libc.so.6
>
> So, wrong list :-)
>
> getrlimit is not expensive at all... but that doesn't mean you can call it
> 60.000 per second without incurring in any penalty...
>
> Thanks,
> Estani
>
>
>
I might have answered too fast...
#0 0xb653b8ec in epoll_wait () from /lib/libc.so.6
#1 0xb5b5ec97 in ?? () from /usr/lib/libevent-1.4.so.2
#2 0xb5b51c5a in event_base_loop () from /usr/lib/libevent-1.4.so.2
#3 0xb7330e80 in ?? () from /usr/lib/xulrunner-10.0/libxul.so
#4 0xb7702ea2 in moz_free () from /usr/lib/xulrunner-10.0/libmozalloc.so
#5 0xb771a580 in ?? () from /lib/ld-linux.so.2
#6 0xb7323c3b in ?? () from /usr/lib/xulrunner-10.0/libxul.so
#7 0xb653b0be in clone () from /lib/libc.so.6
How can I check how many calls come from which library to see who's the
relevant culprit?
(a step by step shows (now) one from each...)
Thanks,
Estani
PS: feel free to dismiss this subject by not answering.
--
Estanislao Gonzalez
Institute of Meteorology
Freie Universität Berlin
Climate System Modeling Working Group
Carl-Heinrich-Becker Weg 6-10, Room 079 AB/1
D-12165 Berlin
Phone: +49 30 838-71116
Mail: estanislao.gonza...@met.fu-berlin.de
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
