Re: [Qemu-devel] [PATCH 0/6] virtio: use ioeventfd in TCG and qtest mode

[Kevin Wolf <kwolf@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1724 bytes --]

Am 28.06.2017 um 21:38 hat Eric Blake geschrieben:
> On 06/28/2017 01:47 PM, Stefan Hajnoczi wrote:
> > This patch series fixes qemu-iotests 068.  Since commit
> > ea4f3cebc4e0224605ab9dd9724aa4e7768fe372 ("qemu-iotests: 068: test iothread
> > mode") the test case has attempted to use dataplane without -M accel=kvm.
> > Although QEMU is capable of running TCG or qtest with emulated ioeventfd/irqfd
> > we haven't enabled it yet.
> > 
> > Unfortunately the virtio test cases fail when ioeventfd is enabled in qtest
> > mode.  This is because they make assumptions about virtqueue ISR signalling.
> > They assume that a request is completed when ISR becomes 1.  However, the ISR
> > can be set to 1 even though no new request has completed since commit
> > 83d768b5640946b7da55ce8335509df297e2c7cd "virtio: set ISR on dataplane
> > notifications".
> > 
> > This issue is solved by introducing a proper qvirtqueue_get_buf() API (similar
> > to the Linux guest drivers) instead of making assumptions about the ISR.  Most
> > of the patches update the test cases to use the new API.
> > 
> > Stefan Hajnoczi (6):
> >   libqos: fix typo in virtio.h QVirtQueue->used comment
> >   libqos: add virtio used ring support
> >   tests: fix virtio-scsi-test ISR dependence
> >   tests: fix virtio-blk-test ISR dependence
> >   tests: fix virtio-net-test ISR dependence
> >   virtio-pci: use ioeventfd even when KVM is disabled
> 
> I'm less familiar with the code in question, so I'll let others review,
> but it did fix the failure of 068 for me.
> 
> Tested-by: Eric Blake <eblake@redhat.com>

Also passes the virtio-scsi qtest under valgrind for me now.

Tested-by: Kevin Wolf <kwolf@redhat.com>

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]