Re: capturing SIGKILL in DomU

[Michal Novotny <minovotn@redhat.com>]

Hi Srujan,
basically this would be the trigger-like mechanism using the xenstore 
since it's using the xenstore watch facility itself if you implement it 
as I recommend. I'm not suggesting using the polling via xs_read() but 
you may study how to establish the xs_watch. This is being done, for 
example, by qemu-dm for looking for pci_ins/pci_del  and USB hotplug 
stuff using the xenstore_process_dm_command_event() function being 
called from the xenstore_process_event for the XS_WATCH_TOKEN vector - 
see tools/ioemu-dir/xenstore.c for reference.

I think this method is OK since it's already used for watching the 
hot-plug/hot-unplug stuff by the device model itself. Nevertheless I 
don't know what exactly do you want to do and maybe it's better to 
modify the hypervisor for your purpose but like I say, I don't know the 
purpose why you need this. All I'm saying it that this communication can 
be achieved using the xenstore watch implementation of the user-space 
stack which is being used for hot-plugging/hot-unplugging stuff already.

Regards,
Michal


On 10/05/2010 03:36 PM, Srujan D. Kotikela wrote:
> Hi Michal,
>
> I would prefer an trigger-like mechanism for communication rather a a 
> process where I have to constantly poll or look for an event in DomU. 
> How computing intensive would be xenstore lookup compared to 
> event-channels with 100s of DomUs?
>
> How about modifying hypervisor to do an upcall for the event handler 
> in Dom0 when a specific interrupt occurs in DomU? If this can be done 
> can you suggest any pointers for the same?
>
> Thanks for your help.
>
> --
> Srujan D. Kotikela
>
>
> On Tue, Oct 5, 2010 at 8:22 AM, Michal Novotny <minovotn@redhat.com 
> <mailto:minovotn@redhat.com>> wrote:
>
>     Hi Srujan,
>
>     I'm not that familiar with event channels themselves however I was
>     thinking about using xenstore. You can modify the device model
>     (qemu-dm) to be watching some entry in the xenstore and the
>     communication could be both way since if you establish a xenstore
>     watch in both Dom0 and DomU you could intercept the changes on
>     both sides.
>
>     If you would like to use interrupts instead you may have to modify
>     the HVMLoader source codes at tools/firmware/hvmloader of the
>     user-space stack but I think using the xenstore could do the job
>     since this is how it's working with PV drivers AFAIK since PV
>     drivers themselves implement xenbus to connect to host's xenstore
>     facility.
>
>     Hope this helps!
>
>     Cheers,
>     Michal
>
>
>     On 10/05/2010 03:14 PM, Srujan D. Kotikela wrote:
>
>         Hi Michal,
>
>         I have no special interest in SIGKILL. All I want to do is
>         notify Dom0 about an event in DomU (I don't need to pass any
>         data). I am trying to indicate events by signals or
>         interrupts. It means, if a "particular" interrupt has occurred
>         in DomU, the Dom0 should be notified. Is there any other way
>         of doing the same other than using event channels?
>
>         I am successful in establishing the event channel. But I am
>         not quite sure how to send a notification that an event
>         occurred in DomU to Dom0. Any pointers for the same would be
>         appreciated.
>
>         --
>         Srujan D. Kotikela
>
>
>         On Tue, Oct 5, 2010 at 3:36 AM, Michal Novotny
>         <minovotn@redhat.com <mailto:minovotn@redhat.com>
>         <mailto:minovotn@redhat.com <mailto:minovotn@redhat.com>>> wrote:
>
>            Hi Srujan,
>            what about adding a signal handler to qemu-dm in the
>            tools/ioemu-dir of the user-space tools? Using the signal()
>         API?
>            Nevertheless why would you like to catch SIGKILL? This one
>         (as can
>            be seen using included program source and killing it using
>         kill -9
>            pid or kill -SIGKILL pid) is not being caught at all
>         nevertheless
>            most of the other signals can be caught.
>
>            This is the source of the example mentioned:
>            #include <stdio.h>
>            #include <signal.h>
>            #include <stdlib.h>
>
>            void sig_handler(int sig) {
>               fprintf(stderr, "Signal %d caught.\n", sig);
>               exit(sig);
>            }
>
>            int main()
>            {
>               signal(SIGINT, sig_handler);
>               signal(SIGKILL, sig_handler);
>
>               sleep(10000);
>               return 0;
>            }
>
>            When I did try SIGINT (Ctrl + C or kill -2 pid) it caught the
>            signal well but when I did try kill -9 pid (or kill
>         -SIGKILL pid
>            respectively) it was not working at all since it killed the
>            process instead of going to the signal handler. When you
>         need to
>            catch signals like interruption signal (Ctrl + C one) this will
>            work fine.
>
>            Michal
>
>
>            On 10/04/2010 09:03 PM, Srujan D. Kotikela wrote:
>
>                Hi,
>
>                I am trying to capture SIGKILL through event channel.
>
>                On my Dom0, the following process is running (remaining
>         code
>                in attachment).
>
>
>                   int main(void){
>
>                       int ret, dom, remote_dom;
>
>                       //initialize domains
>                       dom=0;
>                       remote_dom=2;
>
>                       //create the event channel
>                       ret = create_channel(dom, remote_dom);
>
>
>                       if (0 == ret) {
>                           printf("\n Event Channel established
>         successfully \n");
>                       } else {
>                           return -1;    //EVENT_CHANNEL_CREATION_FAILED
>                       }
>
>                       //wait 20 seconds for an event to occur in DomU
>                       wait_for_event(20);
>
>                       //close the opened interfaces
>                       close_channel();
>
>                       return 0;
>
>                   }
>
>
>                While this process is running; I killed a process in DomU
>                using `*kill SIGKILL pid*`
>
>                How can I capture this event (occured in DomU) at the
>         Dom0. I
>                watched /dev/xen/evtchn, but no notification.
>
>
>                --
>                Srujan D. Kotikela
>
>
>                _______________________________________________
>                Xen-devel mailing list
>         Xen-devel@lists.xensource.com
>         <mailto:Xen-devel@lists.xensource.com>
>         <mailto:Xen-devel@lists.xensource.com
>         <mailto:Xen-devel@lists.xensource.com>>
>
>         http://lists.xensource.com/xen-devel
>
>
>
>            --     Michal Novotny<minovotn@redhat.com
>         <mailto:minovotn@redhat.com> <mailto:minovotn@redhat.com
>         <mailto:minovotn@redhat.com>>>, RHCE
>
>            Virtualization Team (xen userspace), Red Hat
>
>
>
>
>     -- 
>     Michal Novotny<minovotn@redhat.com <mailto:minovotn@redhat.com>>, RHCE
>     Virtualization Team (xen userspace), Red Hat
>
>


-- 
Michal Novotny<minovotn@redhat.com>, RHCE
Virtualization Team (xen userspace), Red Hat