Re: fio signal 11

[Jan Kara <jack@suse.cz>]

On Mon 25-07-16 09:21:00, Jens Axboe wrote:
> On 07/19/2016 11:08 PM, Jan Kara wrote:
> >On Thu 16-06-16 09:06:51, Jens Axboe wrote:
> >>On 06/15/2016 04:45 PM, Jan Kara wrote:
> >>>On Sat 11-06-16 21:30:00, Jens Axboe wrote:
> >>>>On 06/11/2016 08:56 PM, Jens Axboe wrote:
> >>>>>On 06/10/2016 12:42 PM, Jeff Furlong wrote:
> >>>>>>Good point.  Here is the trace:
> >>>>>>
> >>>>>>[New LWP 59231]
> >>>>>>[Thread debugging using libthread_db enabled]
> >>>>>>Using host libthread_db library "/lib64/libthread_db.so.1".
> >>>>>>Core was generated by `fio --name=test_job --ioengine=libaio
> >>>>>>--direct=1 --rw=write --iodepth=32'.
> >>>>>>Program terminated with signal 11, Segmentation fault.
> >>>>>>#0  0x0000000000421e39 in regrow_log (iolog=0x7f828c0c5ad0) at
> >>>>>>stat.c:1909
> >>>>>>1909        if (!cur_log) {
> >>>>>>
> >>>>>>(gdb) bt
> >>>>>>#0  0x0000000000421e39 in regrow_log (iolog=0x7f828c0c5ad0) at
> >>>>>>stat.c:1909
> >>>>>>#1  0x000000000042d4df in regrow_logs (td=td@entry=0x7f8277de0000) at
> >>>>>>stat.c:1965
> >>>>>>#2  0x000000000040ca90 in wait_for_completions
> >>>>>>(td=td@entry=0x7f8277de0000, time=time@entry=0x7fffcfb6b300) at
> >>>>>>backend.c:446
> >>>>>>#3  0x000000000045ade7 in do_io (bytes_done=<synthetic pointer>,
> >>>>>>td=0x7f8277de0000) at backend.c:991
> >>>>>>#4  thread_main (data=data@entry=0x264d450) at backend.c:1667
> >>>>>>#5  0x000000000045cfec in run_threads (sk_out=sk_out@entry=0x0) at
> >>>>>>backend.c:2217
> >>>>>>#6  0x000000000045d2cd in fio_backend (sk_out=sk_out@entry=0x0) at
> >>>>>>backend.c:2349
> >>>>>>#7  0x000000000040d09c in main (argc=22, argv=0x7fffcfb6f638,
> >>>>>>envp=<optimized out>) at fio.c:63
> >>>>>
> >>>>>That looks odd, thanks for reporting this. I'll see if I can get to this
> >>>>>on Monday, if not, it'll have to wait until after my vacation... So
> >>>>>while I appreciate people running -git and finding issues like these
> >>>>>before they show up in a release, might be best to revert back to 2.2.11
> >>>>>until I can get this debugged.
> >>>>
> >>>>I take that back - continue using -git! Just pull a fresh copy, should
> >>>>be fixed now.
> >>>>
> >>>>Jan, the reporter is right, 2.11 works and -git does not. So I just ran
> >>>>a quick bisect, changing the logging from every second to every 100ms to
> >>>>make it reproduce faster. I don't have time to look into why yet, so I
> >>>>just reverted the commit.
> >>>>
> >>>>commit d7982dd0ab2a1a315b5f9859c67a02414ce6274f
> >>>>Author: Jan Kara <jack@suse.cz>
> >>>>Date:   Tue May 24 17:03:21 2016 +0200
> >>>>
> >>>>    fio: Simplify forking of processes
> >>>
> >>>Hum, I've tried reproducing this but failed (I've tried using /dev/ram0 and
> >>>/dev/sda4 as devices for fio). Is it somehow dependent on the
> >>>device fio works with? I have used commit
> >>>54d0a3150d44adca3ee4047fabd85651c6ea2db1 (just before you reverted my
> >>>patch) for testing.
> >>
> >>On vacation right now, I'll check when I get back. It is possible that it
> >>was just a fluke, since there was another bug there related to shared
> >>memory, but it was predictably crashing at the same time for the bisect.
> >>
> >>It doesn't make a lot of sense, however.
> >
> >Did you have a chance to look into this?
> 
> I have not, unfortunately, but I'm suspecting the patch is fine and the
> later fix to allocate the cur_log out of the shared pool was the real
> fix and that the original patch was fine.

So that's what I'd suspect as well but I'm not able to reproduce even the
original crash so I cannot verify this theory... What's the plan going
forward? Will you re-apply the patch? Frankly, I don't care much, it was
just a small cleanup. I'm just curious whether it was really that other bug
or whether I miss something.

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR