On Thu, Apr 28, 2011 at 01:26:38PM +0200, Gerd Hoffmann wrote: > On 04/28/11 13:14, Alon Levy wrote: > >On Thu, Apr 28, 2011 at 12:13:07PM +0200, Gerd Hoffmann wrote: > >> Hi folks, > >> > >>Here is a alternative approach to fix the qemu locking issue. It simply > >>handles the updates in the display refresh timer instead of switching > >>back and forth between the threads using the pipe. > >> > >>comments? > > > >Looks good. I guess you are sure that the lock is less resource hungry then > >switching threads? Anyhow it's way less code, which is great. > > Yes, the main point of doing it this way is to avoid thread context > switches. > > Beside that even the pipe version needs a lock for the update > pointer variable (which is accessed in non-atomic ways from both > iothread and spice-server thread context), i.e. it is buggy as-is. > The race window is pretty small and probably hard to hit though, > guess thats why no issues showed up so far in practice.
I don't think it was, even without locking, but that's moot with your better solution. > > cheers, > Gerd > > _______________________________________________ > Spice-devel mailing list > Spice-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/spice-devel _______________________________________________ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
