On Tue, 21 Oct 2014 15:10:22 +0300 "Michael S. Tsirkin" <m...@redhat.com> wrote:
> On Tue, Oct 21, 2014 at 01:35:39PM +0200, Alexander Graf wrote: > > > > > > On 21.10.14 12:16, Michael S. Tsirkin wrote: > > > On Mon, Oct 20, 2014 at 06:08:56PM +0200, Alexander Graf wrote: > > >> > > >> > > >> On 20.10.14 08:58, Michael S. Tsirkin wrote: > > >>> Current support for bus master (clearing OK bit) together with the need > > >>> to > > >>> support guests which do not enable PCI bus mastering, leads to extra > > >>> state in > > >>> VIRTIO_PCI_FLAG_BUS_MASTER_BUG bit, which isn't robust in case of > > >>> cross-version > > >>> migration for the case when guests use the device before setting > > >>> DRIVER_OK. > > >>> > > >>> Rip out this code, and replace it: > > >>> - Modern QEMU doesn't need VIRTIO_PCI_FLAG_BUS_MASTER_BUG > > >>> so just drop it for latest machine type. > > >>> - For compat machine types, set PCI_COMMAND if DRIVER_OK > > >>> is set. > > >>> > > >>> As this is needed for 2.1 for both pc and ppc, move PC_COMPAT macros > > >>> from pc.h > > >>> to a new common header. > > >>> > > >>> Reviewed-by: Greg Kurz <gk...@linux.vnet.ibm.com> > > >>> Tested-by: Greg Kurz <gk...@linux.vnet.ibm.com> > > >>> Signed-off-by: Michael S. Tsirkin <m...@redhat.com> > > >>> --- > > >>> > > >>> Alexander, could you pls ack me merging this? > > >>> Thanks! > > >> > > >> Have you tried whether this works with old kernel versions? We > > >> introduced the broken flag for very specific old kernel versions that > > >> got their flags wrong IIRC. > > >> > > >> > > >> Alex > > > > > > Yes. The point is that since 2.1 the broken flag isn't effective > > > anymore: when bus master is off, PCI blocks outgoing transactions and > > > setting internal flags isn't going to help. > > > > > > Instead, we detect a buggy guest and immediately enable bus master for > > > it - no need for a separate flag. > > > > Ok, works for me then. > > > > > > Alex > > cool. > Gerd tells me there's a conflict with your next tree: > would you like for me to wait until it's merged, > or go ahead and send pull request and rebase yours on top? > This patch also needs to be rebased on current master actually because PC_COMPAT_2_1 in include/hw/i386/pc.h got changed. Cheers. -- Greg