Re: [Qemu-devel] [PATCH 2/2] e1000: unconditionally enable bus mastering

[Stefan Hajnoczi <stefanha@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2313 bytes --]

On Wed, Jan 07, 2015 at 04:08:29PM +0000, Stefan Hajnoczi wrote:
> On Thu, Dec 18, 2014 at 05:22:19PM +0800, Amos Kong wrote:
> > After enabled network debug of e1000 in Win2012-64r2 guest,
> > Bus mastering of e1000 can't be enabled by e1000 driver. It
> > caused guest can't get IP address.
> > 
> >   # bcdedit /debug on
> >   # bcdedit /dbgsettings net hostip:192.168.122.100 port:50000
> >   (We can use non-existed IP here, it's just used to pass the
> >    setup, not really use it)
> > 
> > If we disable debug function, e1000 driver can enable bus
> > mastering bit successfully, guest network is fine.
> > 
> > This patch changed e1000 backend to enalbe bus mastering
> > unconditionally as a workaround.
> > 
> > Signed-off-by: Amos Kong <akong@redhat.com>
> > ---
> >  hw/net/e1000.c | 11 +++++++++--
> >  1 file changed, 9 insertions(+), 2 deletions(-)
> > 
> > diff --git a/hw/net/e1000.c b/hw/net/e1000.c
> > index ec9224b..82829ae 100644
> > --- a/hw/net/e1000.c
> > +++ b/hw/net/e1000.c
> > @@ -1544,8 +1544,15 @@ static void e1000_write_config(PCIDevice *pci_dev, uint32_t address,
> >  
> >      pci_default_write_config(pci_dev, address, val, len);
> >  
> > -    if (range_covers_byte(address, len, PCI_COMMAND) &&
> > -        (pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
> > +    if (range_covers_byte(address, len, PCI_COMMAND)) {
> > +        /*
> > +         * Some guest (eg: Win2012-64r2) doesn't enable bus mastering
> > +         * correctly, it caused guest network down. So we unconditionally
> > +         * enable PCI bus mastering and BM memory region for e1000 as
> > +         * a workaround.
> > +         */
> > +        pci_dev->config[PCI_COMMAND] |= PCI_COMMAND_MASTER;
> > +        memory_region_set_enabled(&pci_dev->bus_master_enable_region, true);
> >          qemu_flush_queued_packets(qemu_get_queue(s->nic));
> >          start_xmit(s);
> >      }
> 
> This is weird.
> 
> Are you sure there's not some guest behavior missing like the NIC option
> ROM leaving bus mastering enabled after the BIOS/EFI has booted, causing
> Windows debug to work on physical machines?
> 
> Before we merge a hack like this we should understand the problem 100%.

Any new insights into what is going on here?

Stefan

[-- Attachment #2: Type: application/pgp-signature, Size: 473 bytes --]