[Qemu-devel] Re: [PULL] vhost,e1000 fixes

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Sep 20, 2010 at 07:35:36PM +0000, Blue Swirl wrote:
> On Mon, Sep 20, 2010 at 7:21 PM, Michael S. Tsirkin <mst@redhat.com> wrote:
> > On Mon, Sep 20, 2010 at 06:23:55PM +0000, Blue Swirl wrote:
> >> On Mon, Sep 20, 2010 at 6:08 PM, Michael S. Tsirkin <mst@redhat.com> wrote:
> >> > This fixes a bug in vhost error handling
> >> > (also triggers build warning with vhost enabled)
> >> > and fixes e1000 handling of short frames.
> >> >
> >> > Discussion on best ways to fix the e1000 issue
> >> > is still ongoing but the  bug is severe enough
> >> > for some guests and the fix is safe enough
> >> > that I feel we should have it fixed ASAP
> >> > and look for that perfect approach later.
> >> >
> >> > Both fixes are 0.13 material IMO.
> >> >
> >> > The following changes since commit 952afb719f3c965bae12b5bd5f0f0f7ed0251cb8:
> >> >
> >> >  mingw: use ASLR, no-SEH and DEP if available (2010-09-19 08:36:34 +0000)
> >> >
> >> > are available in the git repository at:
> >> >  git://git.kernel.org/pub/scm/linux/kernel/git/mst/qemu.git for_anthony
> >> >
> >> > Michael S. Tsirkin (1):
> >> >      vhost: fix infinite loop on error path
> >>
> >> I don't think your fix is correct either, it will call the ioctl()
> >> with file.index == -1.
> >
> > This is my patch:
> >
> > -    while (--file.index >= 0) {
> > +    while (file.index-- > 0) {
> >         int r = ioctl(net->dev.control, VHOST_NET_SET_BACKEND, &file);
> >         assert(r >= 0);
> >     }
> >
> > For ioctl to get called with -1, index needs to be 0
> > before the decrement, and while won't be entered ...
> > what am I missing?
> 
> Then ioctl() won't be called at all. Is that correct?

Yes, this means the first vq SET_BACKEND call failed so
nothing needs to be reverted.

-- 
MST