On Wed, Aug 29, 2012 at 8:17 AM, Bo Yang <boy...@suse.com> wrote: > On 08/29/2012 02:45 PM, Stefan Hajnoczi wrote: >> BTW, guest driver out-of-memory is not the hardware's problem. The >> driver needs to handle it. For example, the Linux e100 driver has a >> watchdog timer which will raise an interrupt to refill the rx ring if >> allocation failed. > > You're absolutely right here. Can we do our best to work around broken > drivers? does it deserve to provide such a workaround?
Adding behavior that is not in the spec can introduce new bugs. Without a reproducible bug that shows we need this non-spec behavior, I think we shouldn't add it. Stefan
