Re: [Qemu-devel] [PATCH RFC for-2.2] virtio-blk: force 1st s/g to match header

[Stefan Hajnoczi <stefanha@gmail.com>]

On Fri, Nov 28, 2014 at 7:05 AM, Jason Wang <jasowang@redhat.com> wrote:
>
>
> On Fri, Nov 28, 2014 at 9:16 AM, Fam Zheng <famz@redhat.com> wrote:
>>
>> On Thu, 11/27 23:13, Michael S. Tsirkin wrote:
>>>
>>>  On Thu, Nov 27, 2014 at 07:21:35PM +0000, Stefan Hajnoczi wrote:
>>>  > On Thu, Nov 27, 2014 at 4:33 PM, Michael S. Tsirkin <mst@redhat.com>
>>> wrote:
>>>  > > We leak cpu mappings when 1st s/g is not exactly the
>>>  > > header. As we don't set ANY_LAYOUT, we can at this point
>>>  > > simply assert the correct length.
>>>  > >
>>>  > > This will have to be fixed once ANY_LAYOUT is set.
>>>  > >
>>>  > > Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
>>>  > > ---
>>>  > >
>>>  > > Untested: posting for early feedback.
>>>  >  > Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
>>>   Stefan, you are going to merge this for 2.2?
>>>  Peter, if Stefan is merging this one, we can
>>>  take Jason's patch for -net and that should be
>>>  enough for 2.2.
>>
>>
>> My test bot saw a failure of "make check":
>>
>> qemu-system-x86_64: virtio-blk request outhdr too long
>> Broken pipe
>> GTester: last random seed: R02S44c5a09d74c0603f0091ed20d1395121
>> qemu-system-x86_64: virtio-blk request outhdr too long
>> Broken pipe
>> GTester: last random seed: R02Sfdf270c8e1ed75c1f2047e3f0fb89b88
>> qemu-system-x86_64: virtio-blk request outhdr too long
>> Broken pipe
>> GTester: last random seed: R02Sd292c540929b963ed9d7d155f3db5452
>> qemu-system-x86_64: virtio-blk request outhdr too long
>> Broken pipe
>> GTester: last random seed: R02Sfc775a34a844c39f376aa478eda7573f
>> [vmxnet3][WR][vmxnet3_peer_has_vnet_hdr]: Peer has no virtio extension.
>> Task offloads will be emulated.
>> make: *** [check-qtest-x86_64] Error 1
>>
>> Fam
>
>
> This probably because of the test itself.
>
> But anyway all kinds of guests (especially Windows drivers) need to be
> tested.

Right, the test case explicitly tests different descriptor layouts,
even though virtio-blk-pci does not set the ANY_LAYOUT feature bit.

Either the test case needs to check ANY_LAYOUT before using the
2-descriptor layout or it needs to expect QEMU to refuse (in this case
exit(1), which is not very graceful).

The quick fix is to skip the 2-descriptor layout tests and re-enable
them once virtio-blk actually supports ANY_LAYOUT.  Any objections?

Stefan