Xen devel - Dec 2008 - [PATCH] ioemu-remote: check shutdown_requested timely.

After a recent merge of the ioemu-remote tree several weeks ago, I find in
main() -> termsig_seup(), ioemu sets a SIGHUP handler that sets a global
variable shutdown_requested to 1, but ioemu doesn''t check the variable
and as a result, ioemu wouldn''t terminate when xend sends a SIGHUP to
it; finally xend has to send a SIGKILL to it in 10 seconds. The small patch adds
a check to terminate ioemu timely.

Thanks,
-- Dexuan



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Cui, Dexuan writes ("[Xen-devel] [PATCH]
ioemu-remote:
> After a recent merge of the ioemu-remote tree several weeks ago, I
> find in main() -> termsig_seup(), ioemu sets a SIGHUP handler that
> sets a global variable shutdown_requested to 1, but ioemu doesn''t
> check the variable and as a result, ioemu wouldn''t terminate when
> xend sends a SIGHUP to it; finally xend has to send a SIGKILL to it
> in 10 seconds. The small patch adds a check to terminate ioemu
> timely.
I can reproduce the problem you report here - the symptoms I see are
log messages in xend.log and that `xm destroy'' is slow.

But while it''s true that the main loop in vl.c doesn''t check
qemu_shutdown_requested, that main loop is not compiled in in qemu-dm.
Did you test your fix ?  Your fix would also seem to make qemu-dm exit
immediately if the domain powers off even if on_poweroff=preserve.

The root cause of the problem seems to be that upstream qemu installs
signal handlers for the common terminating signals, which simply
request shutdown.  That''s not appropriate for us because xend sends us
a signal when it wants us to really die.  So I have disabled that and
the fix should be in the usual places soon.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Ian Jackson wrote:
> Cui, Dexuan writes ("[Xen-devel] [PATCH] ioemu-remote:
>> After a recent merge of the ioemu-remote tree several weeks ago, I
>> find in main() -> termsig_seup(), ioemu sets a SIGHUP handler that
>> sets a global variable shutdown_requested to 1, but ioemu
doesn''t
>> check the variable and as a result, ioemu wouldn''t terminate
when
>> xend sends a SIGHUP to it; finally xend has to send a SIGKILL to it
>> in 10 seconds. The small patch adds a check to terminate ioemu
>> timely.
> 
> I can reproduce the problem you report here - the symptoms I see are
> log messages in xend.log and that `xm destroy'' is slow.
> 
> But while it''s true that the main loop in vl.c doesn''t
check
> qemu_shutdown_requested, that main loop is not compiled in in qemu-dm.
The vl.c: main_loop() checks qemu_shutdown_requested but we don''t use
it; what we use is helper2.c:main_loop() which doesn''t exit() when the
variable is 1.
> Did you test your fix ?  Your fix would also seem to make qemu-dm exit
> immediately if the domain powers off even if on_poweroff=preserve.
Yes, I tested it and don''t meet with the issue you mention. I think
when we specify ''on_poweroff=preserve'' in hvm config file,
xend wouldn''t send SIGHUP to ioemu at all when guest itself powers off
normally.
> The root cause of the problem seems to be that upstream qemu installs
> signal handlers for the common terminating signals, which simply
> request shutdown.  That''s not appropriate for us because xend
sends us
> a signal when it wants us to really die.  So I have disabled that and
> the fix should be in the usual places soon.
I see you fix in the qemu-xen-unstable tree. I think it''s a better fix.

Thanks,
 -- Dexuan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel