Re: [PATCH v8 09/11] virtio-gpu: Resource UUID

2024-04-24 Thread Dmitry Osipenko
On 4/18/24 22:00, Dmitry Osipenko wrote:
> @@ -1405,6 +1408,8 @@ static int virtio_gpu_blob_load(QEMUFile *f, void 
> *opaque, size_t size,
>  res->iov[i].iov_len = qemu_get_be32(f);
>  }
>  
> +qemu_get_buffer(f, res->uuid.data, sizeof(res->uuid.data));

Save/loading uuid without changing vm version was a bad idea. Will drop
it in v9, we don't need to save/load uuid for virgl anyways.

-- 
Best regards,
Dmitry




Re: [PATCH v8 09/11] virtio-gpu: Resource UUID

2024-04-23 Thread Dmitry Osipenko
On 4/19/24 12:29, Akihiko Odaki wrote:
> On 2024/04/19 4:00, Dmitry Osipenko wrote:
>> From: Antonio Caggiano 
>>
>> Enable resource UUID feature and implement command resource assign UUID.
>> UUID feature availability is mandatory for Vulkan Venus context.
>>
>> UUID is intended for sharing dmabufs between virtio devices on host. Qemu
>> doesn't have second virtio device for sharing, thus a simple stub UUID
>> implementation is enough. More complete implementation using global UUID
>> resource table might become interesting for a multi-gpu cases.
> 
> Isn't it possible to add two virtio-gpu devices even now?

We can add two virtio-gpu devices, but these devices can't interact with
each other efficiently. They won't be able to share host blob resources
without proper UUID implementation.

> A new subsection should also be added for migration compatibility; see:
> docs/devel/migration/main.rst

Will update the docs, thanks.

-- 
Best regards,
Dmitry




Re: [PATCH v8 09/11] virtio-gpu: Resource UUID

2024-04-19 Thread Akihiko Odaki

On 2024/04/19 4:00, Dmitry Osipenko wrote:

From: Antonio Caggiano 

Enable resource UUID feature and implement command resource assign UUID.
UUID feature availability is mandatory for Vulkan Venus context.

UUID is intended for sharing dmabufs between virtio devices on host. Qemu
doesn't have second virtio device for sharing, thus a simple stub UUID
implementation is enough. More complete implementation using global UUID
resource table might become interesting for a multi-gpu cases.


Isn't it possible to add two virtio-gpu devices even now?

A new subsection should also be added for migration compatibility; see: 
docs/devel/migration/main.rst