On Fri, 2013-11-08 at 17:12 +0100, Paolo Bonzini wrote: > Il 08/11/2013 16:08, Marcel Apfelbaum ha scritto: > > Actually, as I see, the default behavior of "system" memory region > > is to use unassigned_mem_ops that has valid.accepts method returning > > false (no read/write methods). I don't see that read all-ones/ignore > > writes is implemented. > > Right, it's read-all-zeroes (see unassigned_mem_read). It was > read-all-ones for a brief time, then it got changed back to read-all-zeroes. I remember something, Jan's patches got reverted because the modification was system wide and not only for pci address space. > > > This was the main reason I submitted this patch. I had *no* clue that > > it would impact so much the system... > > Yeah, it's not your fault. But it should have been held back until 1.8. > Do you agree with reverting it? Given the actual impact on version's stability, I agree it is the right thing to do.
> > > I still think the patch is needed ans the guests will benefit from > > more accurate PCI spec emulation. > > Only buggy guests :) but yes, I agree it's a good thing to have. Yes, it may be a driver there making some a wrong decision and "reading-all-ones" may give it a clue. (the same for writes) Thanks, Marcel > > Paolo