[CRIU] [PATCH] zdtm: Use SIGWINCH instead of SIGTRAP in pthread01

Cyrill Gorcunov gorcunov at openvz.org
Thu Nov 22 10:10:06 EST 2012


On Thu, Nov 22, 2012 at 07:04:49PM +0400, Pavel Emelyanov wrote:
> On 11/19/2012 04:59 PM, Cyrill Gorcunov wrote:
> > Otherwise test with -d option get fail (this is because
> > SIGTRAP get cleared by our parasizte engine).
> > 
> > The proper fix most likely will require kernel patching
> > or redesign of parasite code. Meanwhile to make test
> > working we switch to the signal which just known to work.
> > 
> > Also I've fixed a couple of s/sigprocmask/pthread_sigmask/
> > and reworked the procedure of showing the signals mask to
> > more readable format.
> > 
> > Reported-by: Andrey Vagin <avagin at openvz.org>
> > Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
> 
> Andrey, Cyrill, what about this one?

Until we change parasite code to use pipes instead of
int3 -- we can't do anything. It's a side effect of int3
usage as a 'sign' of 'we're done' in parasite.

But changing parasite design is a big slab of code which
I've in my TODO list. Thus I think it's worth to switch to
SIGWINCH for a while.

Note, that this problem appears only if we're continue working
after checkpoint, if we do complete c/r cycle -- the signals
restored fine, because we 'remember' original mask before
we yield first int3 instructions.


More information about the CRIU mailing list