[Mesa-dev] About drm_hwcomposer (Was Re: About the PixelFormat mappings in drm_gralloc)

[Rob Clark]

On Wed, Jan 20, 2016 at 4:59 PM, Rob Herring <robh at kernel.org> wrote:
> Hi Sean,
>
> On Thu, Jan 14, 2016 at 1:15 PM, Sean Paul <seanpaul at chromium.org> wrote:
>>
>>
>> On Thu, Jan 14, 2016 at 9:47 AM, Chih-Wei Huang <cwhuang at android-x86.org>
>> wrote:
>>>
>>> Hello Sean,
>>> My last try of drm_hwcomposer failed due to
>>> my limited time and knowledge.
>>>
>>> Fortunately more developers join us
>>> to work on this topic now, including
>>> Rob Herring (kernel drm developer),
>>> Rob Clark and Emil Velikov (Mesa developers)
>>> We are working on drm_hwcomposer
>>> for virtio-gpu, the emulated GPU for QEMU.
>>>
>>> If you don't mind, I would invite you to join
>>> our devel group for the discussion.
>>
>>
>> [adding Emil, Rob, and Rob]
>>
>> Hi,
>> Ok, I've signed up for the android-x86 list. I don't anticipate seeing all
>> posts, so please cc me on threads needing feedback.
>>
>> We also just today open-sourced drm_hwcomposer on the chromium gerrit
>> instance, such that we'll be doing all of our development in the open on
>> that repo.
>>
>> The repo is here:
>> https://chromium.googlesource.com/chromiumos/drm_hwcomposer
>>
>> Contributing instructions are here:
>> https://sites.google.com/a/chromium.org/dev/contributing-to-drm_hwcomposer
>
> Thanks for the pointer. This is essentially what I'm based on.
>
> Is there a testing target we need to use for submitting patches? Pixel
> C? The first problem is the use of the pre mainline atomic functions,
> so using mainline kernel and libdrm are a problem. What is your plan
> to handle that?
>
> It also no longer works with the drm_gralloc in AOSP due to missing
> GRALLOC_MODULE_PERFORM_GET_USAGE support, so is there any plan to also
> host drm_gralloc?

Just my own $0.02, but I would really like to move drm_gralloc into
mesa.. or at least the gallium pipe-driver part.

Having gralloc essentially being a sort of state-tracker is great for
a lot of reasons.. ie. don't have to add new gralloc specific code for
new drivers like freedreno/vc4/etnaviv, and also by sharing a common
pipe_screen we can avoid some refcnt'ing / handle lifecycle issues.

But that also means using API's which aren't really intended to be
exposed outside of mesa.  Ie. gallium is not intended to be consumed
as a stable API from external state trackers.

I know that chromium had started adding support for some other
non-mesa drivers.  I'm not sure how best to handle that.  Maybe what
makes most sense is to just copy/paste some code into mesa, and leave
external drm_gralloc just for blob drivers..

(I was kinda hoping Emil would someday tackle moving the code into
mesa, since whenever I mess with mesa build system I make a mess of
things :-P)

BR,
-R

> Anything you can share about planned changes/features you are working
> on so we can better coordinate work Linaro is doing here.
>
> Rob