Re: [Qemu-devel] qemu-nbd vs 'simple' trace backend vs iotest 147

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1199 bytes --]

On Fri, Jul 13, 2018 at 08:40:19AM +0200, Paolo Bonzini wrote:
> On 12/07/2018 18:30, Stefan Hajnoczi wrote:
> > On Wed, Jul 11, 2018 at 03:33:21PM +0200, Cornelia Huck wrote:
> >> The other qemu-nbds (the inet and the unix socket ones from the first
> >> run, the second inet one from the second run) have a single thread with
> >> the same backtrace I posted above.
> > 
> > We just discussed this on IRC, but for the record:
> > 
> > qemu-nbd --fork will fork the process after the simpletrace write-out
> > thread has been spawned.  The child process lacks this thread (due to
> > how fork(2) handles multithreading).  Either qemu-nbd needs to
> > initialize tracing later (but that means we cannot trace early init) or
> > simpletrace needs a way to respawn the write-out thread.
> 
> You can use pthread_atfork for this.

Thanks.  The man page says:

  The intent of  pthread_atfork()  was  to provide  a  mechanism
  whereby  the application (or a library) could ensure that mutexes and
  other process and thread state would be restored to a consistent
  state.  In practice, this task is generally too difficult to be
  practicable.

Challenge accepted!

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]