On 02.03.2010, at 11:09, Michael S. Tsirkin wrote: > On Mon, Mar 01, 2010 at 01:26:37PM +0200, Gleb Natapov wrote: >> On Mon, Mar 01, 2010 at 12:14:43PM +0100, Alexander Graf wrote: >>> >>> On 14.09.2009, at 15:31, Yan Vugenfirer wrote: >>> >>>> Signed-off-by: Yan Vugenfirer <yvuge...@redhat.com> >>>> >>>> --- >>>> hw/virtio-pci.c | 14 ++++++++++++-- >>>> 1 files changed, 12 insertions(+), 2 deletions(-) >>>> >>>> diff --git a/hw/virtio-pci.c b/hw/virtio-pci.c >>>> index f812ab7..a0a22c4 100644 >>>> --- a/hw/virtio-pci.c >>>> +++ b/hw/virtio-pci.c >>>> @@ -364,8 +364,17 @@ static void virtio_map(PCIDevice *pci_dev, int >>>> region_num, >>>> static void virtio_write_config(PCIDevice *pci_dev, uint32_t address, >>>> uint32_t val, int len) >>>> { >>>> + VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev); >>>> + >>>> + if (PCI_COMMAND == address) { >>>> + if (!(val & PCI_COMMAND_MASTER)) { >>>> + proxy->vdev->status &= ~VIRTIO_CONFIG_S_DRIVER_OK; >>> >>> This part breaks PCI hotplug with Linux guests. >>> >>> What happens is the following: >> Looks like something is broken even without this patch. >> >>> >>> (qemu) pci_add auto nic model=virtio,vlan=0 >>> - Virtio dev 1 -> write config (0) >>> - Virtio dev 1 -> write config (0x3) >>> - Virtio dev 1 -> set status explicitly to OK >> Why Linux doesn't enable bus mastering on this device? > > I fixed this, and Rusty said he has applied the patch, > but it seems the patch got lost later: > > http://lkml.org/lkml/2009/11/29/73 > > Alexander, could you please check whether applying > this patch fixes hotplug for you? > If yes I will queue it up for 2.6.34.
So we're looking at a guest bug? It's still a qemu regression then. This configuration used to work with 0.10. There should at least be a feature bit from the guest saying "I'm good with bus mastering", so we can enable the check only for those. Alex