xserver 1.20 RC1 tomorrow

[Daniel Stone]

Hi,

On 27 February 2018 at 21:59, Adam Jackson <ajax at nwnk.net> wrote:
> 2) https://patchwork.freedesktop.org/series/27289/
> DRI3 v1.2. Has some ABI changes, but none that are essential, the old
> single-fd API can either be kept as is, or implemented in terms opf the
> new multiple-fd API.

Perhaps I'm being dense, but I can't see that this does have driver
ABI changes? The old single-plane hooks are still in place, and the
core has helpers to call them if possible (single-plane, no modifier,
driver doesn't have newer hook). We do change the helper to be
multi-FD rather than single FD, but that's just in dri3_priv.h, so
presumably not ABI?

> 3) https://patchwork.freedesktop.org/series/30946/
> DMA fences for DRI3. Does have at least one ABI change that looks
> essential, in misync.

I would certainly like to see this land, but there are some open
questions it'd be great to get more feedback: namely, should we use
the fences to schedule in-server (and not submit blits/flips until the
fence has passed), or should we pass the fences into the client APIs
(e.g. GPU command-stream wait on EGL-imported fences, KMS commit
wait). The former is more obvious, but also a behavioural change from
how we handled implicit fences previously.

> 4) https://patchwork.freedesktop.org/series/37274/
> Per-window flips for Xwayland. No xfree86 changes.

I'd quite like to give this another review pass, especially for
subsurfaces, but generally if Michel is happy than I am.

Cheers,
Daniel