Linux Virtualization - Mar 2017 - [BUG] virtio-net linux driver fails to probe on MIPS Malta since 'hw/virtio-pci: fix virtio behaviour'

On Mon, Mar 20, 2017 at 05:21:22PM +0200, Marcel Apfelbaum
wrote:
> On 03/17/2017 11:57 PM, James Hogan wrote:
> > Hi,
> > 
> > I've bisected the following failure of the virtio_net linux v4.10
driver
> > to probe in QEMU v2.9.0-rc1 emulating a MIPS Malta machine:
> > 
> > virtio_net virtio0: virtio: device uses modern interface but does not
have VIRTIO_F_VERSION_1
> > virtio_net: probe of virtio0 failed with error -22
> > 
> > To QEMU commit 9a4c0e220d8a ("hw/virtio-pci: fix virtio
behaviour").
> > 
> > It appears that adding
",disable-modern=on,disable-legacy=off" to the
> > virtio-net -device makes it work again.
> > 
> > I presume this should really just work out of the box. Any ideas why
it
> > isn't?
> > 
> 
> Hi,
> 
> 
> This is strange. This commit changes virtio devices from legacy to virtio
"transitional".
> (your command line changes it to legacy)
> Linux 4.10 supports virtio modern/transitional (as far as I know) and on
QEMU side
> there is nothing new.
> 
> Michael, do you have any idea?
> 
> Thanks,
> Marcel
My guess would be firmware mishandling 64 bit BARs - we saw such
a case on sparc previously. As a result you are probably reading
all zeroes from features register or something like that.
Marcel, could you send a patch making the bar 32 bit?
If that helps we know what the issue is.
> > Cheers
> > James
> >

On 03/20/2017 05:43 PM, Michael S. Tsirkin wrote:
> On Mon, Mar 20, 2017 at 05:21:22PM +0200, Marcel Apfelbaum wrote:
>> On 03/17/2017 11:57 PM, James Hogan wrote:
>>> Hi,
>>>
>>> I've bisected the following failure of the virtio_net linux
v4.10 driver
>>> to probe in QEMU v2.9.0-rc1 emulating a MIPS Malta machine:
>>>
>>> virtio_net virtio0: virtio: device uses modern interface but does
not have VIRTIO_F_VERSION_1
>>> virtio_net: probe of virtio0 failed with error -22
>>>
>>> To QEMU commit 9a4c0e220d8a ("hw/virtio-pci: fix virtio
behaviour").
>>>
>>> It appears that adding
",disable-modern=on,disable-legacy=off" to the
>>> virtio-net -device makes it work again.
>>>
>>> I presume this should really just work out of the box. Any ideas
why it
>>> isn't?
>>>
>>
>> Hi,
>>
>>
>> This is strange. This commit changes virtio devices from legacy to
virtio "transitional".
>> (your command line changes it to legacy)
>> Linux 4.10 supports virtio modern/transitional (as far as I know) and
on QEMU side
>> there is nothing new.
>>
>> Michael, do you have any idea?
>>
>> Thanks,
>> Marcel
>
> My guess would be firmware mishandling 64 bit BARs - we saw such
> a case on sparc previously. As a result you are probably reading
> all zeroes from features register or something like that.
> Marcel, could you send a patch making the bar 32 bit?
> If that helps we know what the issue is.
>
Sure,

Thanks,
Marcel
>>> Cheers
>>> James
>>>

On 03/20/2017 05:43 PM, Michael S. Tsirkin wrote:
> On Mon, Mar 20, 2017 at 05:21:22PM +0200, Marcel Apfelbaum wrote:
>> On 03/17/2017 11:57 PM, James Hogan wrote:
>>> Hi,
>>>
>>> I've bisected the following failure of the virtio_net linux
v4.10 driver
>>> to probe in QEMU v2.9.0-rc1 emulating a MIPS Malta machine:
>>>
>>> virtio_net virtio0: virtio: device uses modern interface but does
not have VIRTIO_F_VERSION_1
>>> virtio_net: probe of virtio0 failed with error -22
>>>
>>> To QEMU commit 9a4c0e220d8a ("hw/virtio-pci: fix virtio
behaviour").
>>>
>>> It appears that adding
",disable-modern=on,disable-legacy=off" to the
>>> virtio-net -device makes it work again.
>>>
>>> I presume this should really just work out of the box. Any ideas
why it
>>> isn't?
>>>
>>
>> Hi,
>>
>>
>> This is strange. This commit changes virtio devices from legacy to
virtio "transitional".
>> (your command line changes it to legacy)
>> Linux 4.10 supports virtio modern/transitional (as far as I know) and
on QEMU side
>> there is nothing new.
>>
>> Michael, do you have any idea?
>>
>> Thanks,
>> Marcel
>
> My guess would be firmware mishandling 64 bit BARs - we saw such
> a case on sparc previously. As a result you are probably reading
> all zeroes from features register or something like that.
> Marcel, could you send a patch making the bar 32 bit?
> If that helps we know what the issue is.
>
Hi James,

Can you please check if the below patch fixes the problem?
Please note it is not a solution.

diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
index f9b7244..5b4d429 100644
--- a/hw/virtio/virtio-pci.c
+++ b/hw/virtio/virtio-pci.c
@@ -1671,9 +1671,7 @@ static void virtio_pci_device_plugged(DeviceState *d,
Error **errp)
          }

          pci_register_bar(&proxy->pci_dev, proxy->modern_mem_bar_idx,
-                         PCI_BASE_ADDRESS_SPACE_MEMORY |
-                         PCI_BASE_ADDRESS_MEM_PREFETCH |
-                         PCI_BASE_ADDRESS_MEM_TYPE_64,
+                         PCI_BASE_ADDRESS_SPACE_MEMORY,
                           &proxy->modern_bar);

          proxy->config_cap = virtio_pci_add_mem_cap(proxy, &cfg.cap);


Thanks,
Marcel

Hi Marcel,

On Tue, Mar 21, 2017 at 04:16:58PM +0200, Marcel Apfelbaum
wrote:
> Can you please check if the below patch fixes the problem?
> Please note it is not a solution.
> 
> diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
> index f9b7244..5b4d429 100644
> --- a/hw/virtio/virtio-pci.c
> +++ b/hw/virtio/virtio-pci.c
> @@ -1671,9 +1671,7 @@ static void virtio_pci_device_plugged(DeviceState *d,
Error **errp)
>           }
> 
>           pci_register_bar(&proxy->pci_dev,
proxy->modern_mem_bar_idx,
> -                         PCI_BASE_ADDRESS_SPACE_MEMORY |
> -                         PCI_BASE_ADDRESS_MEM_PREFETCH |
> -                         PCI_BASE_ADDRESS_MEM_TYPE_64,
> +                         PCI_BASE_ADDRESS_SPACE_MEMORY,
>                            &proxy->modern_bar);
> 
>           proxy->config_cap = virtio_pci_add_mem_cap(proxy,
&cfg.cap);
Sorry for the delay trying this, I was away last week.

No, it doesn't seem to make any difference.

Thanks
James
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: Digital signature
URL:
<http://lists.linuxfoundation.org/pipermail/virtualization/attachments/20170327/d6b94f63/attachment.sig>