Re: [Qemu-devel] [PATCH v11 0/7] trace: Generic event state description

[Stefan Hajnoczi <stefanha@gmail.com>]

On Mon, Mar 04, 2013 at 10:01:43PM +0100, Lluís Vilanova wrote:
> Stefan Hajnoczi writes:
> 
> > On Fri, Mar 01, 2013 at 04:29:35PM +0100, Lluís Vilanova wrote:
> >> Provides a generic event state description structure (TraceEvent) and a more
> >> detailed event control and query interface.
> >> 
> >> This is achieved by creating a new "non-public" tracing backend (i.e., not
> >> selectable by the user at configure time) that will generate the appropriate
> >> event description information.
> >> 
> >> Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
> >> ---
> >> 
> >> Changes in v11:
> >> 
> >> * Rebase on a4bcea3 from master.
> 
> > Hi Lluís,
> > Thanks for rebasing.  Unfortunately I still hit the hang on shutdown
> > with the simple trace backend.
> 
> > Steps to reproduce:
> > $ ./configure --target-list=x86_64-softmmu --enable-trace-backend=simple
> > $ make
> > $ cat my-events
> > bdrv_open_common
> > $ x86_64-softmmu/qemu-system-x86_64 \
> >       -enable-kvm -m 1024 \
> >       -trace events=my-events \
> >       -drive if=virtio,cache=none,file=test.img
> > Ctrl-Alt-2
> > (qemu) quit
> > ...hang...
> 
> > Please let me know if you are able to reproduce it.  qemu.git/master
> > does not behave this way.
> 
> I've been unable to reproduce it with v11, but was able to using v10 (although
> only a couple of times).
> 
> What I saw in v10 looked like a deadlock between 'flush_trace_file' and
> 'wait_for_trace_records_available' (both use 'trace_lock'), but I only was able
> to reproduce it a couple of times, so I couldn't haunt the problem down.
> 
> After a quick re-read of the patches, the changes should have no effect on how
> the mutexes in the "simple" backend are handled, so maybe the problem is not
> related to this change (it just makes it more frequent in your build).

I'll take a look here since it reproduces every time on this host.

> Could it be possible that you exit QEMU before calling 'trace_backend_init'?
> (as unlikely as it might be)

No, the guest is running at the BIOS prompt.

Stefan