Re: [Qemu-devel] [PULL 12/23] virtio: use VRingMemoryRegionCaches for avail and used rings

[Cédric Le Goater]

Hello,

On 02/23/2017 01:32 AM, Alex Williamson wrote:
> On Wed, 22 Feb 2017 10:03:56 +0100
> Paolo Bonzini <address@hidden> wrote:
> 
>> On 21/02/2017 18:54, Laszlo Ersek wrote:
>>> Actually, QEMU segfaults. From the dmesg:
>>>
>>> [Tue Feb 21 18:47:28 2017] CPU 0/KVM[8298]: segfault at 48 ip
>>> 00007fcb5dd02105 sp 00007fcb49efc270 error 4 in
>>> qemu-system-x86_64[7fcb5dae3000+905000]
>>>
>>> Complete backtrace below. (Thread 11 seems to be the one
>>> segfaulting.)  
>>
>> Indeed.  It's the virtio-blk device that is segfaulting, please try
>> this one liner (haven't reproduced it, but it seems obvious with the
>> backtrace and some sleep).
> 
> I hit this as well, the patch below resolves it.

I still have migration issues with this patch on ppc, which fails
on the target guest with : 

qemu-system-ppc64: VQ 0 size 0x80 < last_avail_idx 0xe1e - used_idx 0x0
qemu-system-ppc64: Failed to load virtio-blk:virtio
qemu-system-ppc64: error while loading state for instance 0x0 of device 
'address@hidden:01.0/virtio-blk'
qemu-system-ppc64: load of migration failed: Operation not permitted

Thanks,

C. 

> Tested-by: Alex Williamson <address@hidden>
>  
>> commit c1aa478c7181c543606ca81404c59e126d66213d
>> Author: Paolo Bonzini <address@hidden>
>> Date:   Wed Feb 22 10:02:37 2017 +0100
>>
>>     virtio: check for vring setup in virtio_queue_empty
>>     
>>     If the vring has not been set up, there is nothing in the virtqueue.
>>     virtio_queue_host_notifier_aio_poll calls virtio_queue_empty even in
>>     this case; we have to filter it out just like virtio_queue_notify_aio_vq.
>>     
>>     Reported-by: Gerd Hoffmann <address@hidden>
>>     Signed-off-by: Paolo Bonzini <address@hidden>
>>
>> diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
>> index 23483c7..e487e36 100644
>> --- a/hw/virtio/virtio.c
>> +++ b/hw/virtio/virtio.c
>> @@ -2291,7 +2291,7 @@ static bool virtio_queue_host_notifier_aio_poll(void 
>> *opaque)
>>      VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
>>      bool progress;
>>  
>> -    if (virtio_queue_empty(vq)) {
>> +    if (!vq->vring.desc || virtio_queue_empty(vq)) {
>>          return false;
>>      }
>>  
>>
> 
>