On Wed, Jun 29, 2011 at 10:41:43AM +0200, Gerd Hoffmann wrote: > Hi, > > >>Shouldn't the mode change to QXL_MODE_UNDEFINED on DESTROY_SURFACES? > >>(I can see it currently doesn't, but it seems like a mistake) > > > >No, that is the point - this patch makes the lifetime of QXL_MODE_NATIVE be > >the lifetime of the driver in the guest, kinda. we never unload the driver > >in the guest, the only time it stops being operational is when we enter vga > >mode and that's explicitly done when we get any vga port write/read. > > I agree with yonit here, moving to UNDEFINED makes sense. Keeping > track of guest driver state in qxl is asking for trouble. It will > blow up when it comes to S4 support and we'll exit and restart the > qemu process.
Actually S4 support == S3 support. There is absolutely no difference. Since qemu still does a reset. But I already checked dropping this patch and using UNDEFINED instead and it works fine. I just had to additionally allow io when in UNDEFINED. I think that makes sense, since UNDEFINED is just the same as "driver working but primary destroyed". > > cheers, > Gerd > >