[PATCH v6] Add udmabuf misc device

Dave Airlie airlied at gmail.com
Fri Jul 6 04:00:55 UTC 2018


On 4 July 2018 at 18:00, Gerd Hoffmann <kraxel at redhat.com> wrote:
> On Wed, Jul 04, 2018 at 09:26:39AM +0200, Tomeu Vizoso wrote:
>> On 07/04/2018 07:53 AM, Gerd Hoffmann wrote:
>> > On Tue, Jul 03, 2018 at 10:37:57AM +0200, Daniel Vetter wrote:
>> > > On Tue, Jul 03, 2018 at 09:53:58AM +0200, Gerd Hoffmann wrote:
>> > > > A driver to let userspace turn memfd regions into dma-bufs.
>> > > >
>> > > > Use case:  Allows qemu create dmabufs for the vga framebuffer or
>> > > > virtio-gpu ressources.  Then they can be passed around to display
>> > > > those guest things on the host.  To spice client for classic full
>> > > > framebuffer display, and hopefully some day to wayland server for
>> > > > seamless guest window display.
>> > > >
>> > > > qemu test branch:
>> > > >    https://git.kraxel.org/cgit/qemu/log/?h=sirius/udmabuf
>> > > >
>> > > > Cc: David Airlie <airlied at linux.ie>
>> > > > Cc: Tomeu Vizoso <tomeu.vizoso at collabora.com>
>> > > > Cc: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
>> > > > Cc: Daniel Vetter <daniel at ffwll.ch>
>> > > > Signed-off-by: Gerd Hoffmann <kraxel at redhat.com>
>> > >
>> > > I think some ack for a 2nd use-case, like virtio-wl or whatever would be
>> > > really cool. To give us some assurance that this is generically useful.
>> >
>> > Tomeu?  Laurent?
>>
>> Sorry, but I think I will need some help to understand how this could help
>> in the virtio-wl case [adding Zach Reizner to CC].
>>
>> Any graphics buffers that are allocated with memfd will be shared with the
>> compositor via wl_shm, without need for dmabufs.
>
> Within one machine, yes.  Once virtualization is added to the mix things
> become more complicated ...
>
> When using virtio-gpu the guest will allocate graphics buffers from
> normal (guest) ram, then register these buffers (which are allowed to be
> scattered) with the host as resource.
>
> qemu can use memfd to allocate guest ram.  Now, with the help of
> udmabuf, qemu can create a *host* dma-buf for the *guest* graphics
> buffer.
>
> That dma-buf can be used by qemu internally (mmap it to get a linear
> mapping of the resource, to avoid copying).  It can be passed on to
> spice-client, to display the guest framebuffer.
>
> And I think it could also be quite useful to pass guest wayland windows
> to the host compositor, without mapping host-allocated buffers into the
> guest, so we don't have do deal with the "find some address space for
> the mapping" issue in the first place.  There are more things needed to
> complete this of course, but it's a building block ...

There is a use case where I think we have to deal with the "find some address
space" problem. For GL4.4 ARB_buffer_storage and Vulkan memory mangement
there is the concept of coherent buffers between GPU and CPU. From the
virgl point of view, we'd create a host buffer in GL, and then create
a mapping from
it on the host that we'd need to present in the guest userspace as a
linear buffer.

Just in case we think this can solve all our problems :-)

Dave.


More information about the dri-devel mailing list