RE: [RFC PATCH 6/6] drm/i915/gvt: support QEMU getting the dmabuf

[Chen, Xiaoguang]

Hi Gerd and Alex,

>-----Original Message-----
>From: Gerd Hoffmann [mailto:kraxel@xxxxxxxxxx]
>Sent: Friday, May 19, 2017 4:05 PM
>To: Alex Williamson <alex.williamson@xxxxxxxxxx>
>Cc: Chen, Xiaoguang <xiaoguang.chen@xxxxxxxxx>; Tian, Kevin
><kevin.tian@xxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; zhenyuw@xxxxxxxxxxxxxxx;
>Lv, Zhiyuan <zhiyuan.lv@xxxxxxxxx>; intel-gvt-dev@xxxxxxxxxxxxxxxxxxxxx; Wang,
>Zhi A <zhi.a.wang@xxxxxxxxx>
>Subject: Re: [RFC PATCH 6/6] drm/i915/gvt: support QEMU getting the dmabuf
>
>  Hi,
>
>> > User space need to check whether there's a dmabuf for the plane(user space
>usually cached two or three dmabuf to handle double buffer or triple buffer
>situation) only there's no dmabuf for the plane we will create a dmabuf for
>it(another ioctl).
>>
>> If our ioctls are "Query current plane" and "Give me a dmabuf for
>> current plane", isn't that racey?  The current plane could have
>> changed between those two calls so the user doesn't absolutely know
>> which plane the dmabuf retrieved is for.  The "Give me a dmabuf"
>> therefore needs to take some sort of plane index so the user can
>> request a specific plane.
>
>The "give me a dmabuf" ioctl returns the plane description too, so userspace can
>at least figure it did hit the race window.
>
>We could also do it the other way around:  Instead of having the kernel returning
>the plane description userspace could pass it in, and the kernel throws -EINVAL in
>case it doesn't match due to things having changed meanwhile.
Or just return a dmabuf  based on the current plane ?
Because if user got -EINVAL while ioctl "Give me a dmabuf" user should do the ioctl again.

>
>cheers,
>  Gerd