qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] Re: [Qemu-devel] [PATCH 01/13] Handle terminating signal


From: Anthony Liguori
Subject: Re: [Xen-devel] Re: [Qemu-devel] [PATCH 01/13] Handle terminating signals.
Date: Tue, 26 Aug 2008 10:04:30 -0500
User-agent: Thunderbird 2.0.0.16 (X11/20080723)

Ian Jackson wrote:
Anthony Liguori writes ("[Xen-devel] Re: [Qemu-devel] [PATCH 01/13] Handle 
terminating signals."):
The race I know of is that you may get an aio signal completion before select but after you've already qemu_aio_poll()'d. In practice, we only sleep for 10ms at a time in select() so the race is handled by that. If we wanted to increase the amount of time we slept, we would have to handle this race.

Yes.  And, 10ms is too long anyway for reasonable performance.  During
my merge with upstream I found that the qemu aio functionality (which
was done quite differently to the old xen ioemu) caused a severe
performance regression under some conditions because of this race.

Yeah, noticed that too, especially with qcow2.

In KVM, we sleep for 1s in select() and use signalfd() to receive the aio notifications. For older hosts, we emulate signalfd using a thread and the pipe-to-self trick.

Why does it need a thread ?  You can just write to the pipe in the
signal handler.  I'll post my code.

It's a little less perfect. Your signal handler can write() to the pipe but what happens if you get EAGAIN?

So what we do in KVM is use sigwait() within a separate thread. We don't set O_NONBLOCK so the thread blocks if the pipe fills up which is exactly the semantics you would want.

Below is our implementation. I'll queue up to push this change into QEMU after I finish with the migration patches.

Regards,

Anthony Liguori

#include <sys/syscall.h>
#include <pthread.h>

struct sigfd_compat_info
{
   sigset_t mask;
   int fd;
};

static void *sigwait_compat(void *opaque)
{
   struct sigfd_compat_info *info = opaque;
   int err;
   sigset_t all;

   sigfillset(&all);
   sigprocmask(SIG_BLOCK, &all, NULL);

   do {
   siginfo_t siginfo;

   err = sigwaitinfo(&info->mask, &siginfo);
   if (err == -1 && errno == EINTR) {
           err = 0;
           continue;
       }

   if (err > 0) {
       char buffer[128];
       size_t offset = 0;

       memcpy(buffer, &err, sizeof(err));
       while (offset < sizeof(buffer)) {
       ssize_t len;

       len = write(info->fd, buffer + offset,
               sizeof(buffer) - offset);
       if (len == -1 && errno == EINTR)
           continue;

       if (len <= 0) {
           err = -1;
           break;
       }

       offset += len;
       }
   }
   } while (err >= 0);

   return NULL;
}

static int kvm_signalfd_compat(const sigset_t *mask)
{
   pthread_attr_t attr;
   pthread_t tid;
   struct sigfd_compat_info *info;
   int fds[2];

   info = malloc(sizeof(*info));
   if (info == NULL) {
   errno = ENOMEM;
   return -1;
   }

   if (pipe(fds) == -1) {
   free(info);
   return -1;
   }

   memcpy(&info->mask, mask, sizeof(*mask));
   info->fd = fds[1];

   pthread_attr_init(&attr);
   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);

   pthread_create(&tid, &attr, sigwait_compat, info);

   pthread_attr_destroy(&attr);

   return fds[0];
}

int kvm_signalfd(const sigset_t *mask)
{
#if defined(SYS_signalfd)
   int ret;

   ret = syscall(SYS_signalfd, -1, mask, _NSIG / 8);
   if (!(ret == -1 && errno == ENOSYS))
   return ret;
#endif

   return kvm_signalfd_compat(mask);
}


Ian.





reply via email to

[Prev in Thread] Current Thread [Next in Thread]