[PATCH wayland-protocols, weston] Extend linux-dmabuf to bypass roundtrip during dmabuf import

[Daniel Stone]

Hi,

On 11 November 2016 at 11:40, Varad Gautam <varadgautam at gmail.com> wrote:
> This patchset provides an immediate path to import dmabufs into a
> wl_buffer to avoid the round-trip wait a client must face to receive
> the buffer handle.
>
> Previous discussion: https://phabricator.freedesktop.org/T7570

Copy and pasting from here, this was requested for ChromeOS:
'The usecase that convinced me was ARC++: thanks to the way that
Android works, they can't actually create a static buffer pool which
gets reused on startup, so ultimately they have to roundtrip once per
buffer per frame. This is a dealbreaker for them in terms of
latency/power/etc, and given it's a closed system, they know that
allocation would succeed, so don't need the roundtrip.

I'd expected that closed systems would look like an STB where you have
static buffer pools, so the roundtrip at init is actually fine, but
this case of a closed system which cannot do static upfront pools,
didn't occur to me at the time.'

Having discussed it for a few others as well, it does seem like there
are a bunch of systems which can guarantee up front that they won't
fail import, but also can't maintain static wl_buffer pools. Oh well.

Cheers,
Daniel