[Mesa-dev] [android-x86-devel] Re: gralloc_drm_pipe

[Rob Herring]

On Wed, Apr 6, 2016 at 11:12 AM, Chih-Wei Huang <cwhuang at android-x86.org> wrote:
> 2016-04-04 6:25 GMT+08:00 Rob Herring <robherring2 at gmail.com>:
>> On Sun, Apr 3, 2016 at 12:29 PM, Chih-Wei Huang <cwhuang at android-x86.org> wrote:
>>> Besides, the module name is still gralloc.drm.
>>> How about call it gralloc.gbm?
>>
>> Eventually yes, but for now it is more convenient for my development
>> to keep the name the same.
>>
>>> That means it can coexist with the current
>>> gralloc.drm module so the transition to it
>>> will be easier. Agree?
>>
>> It's not binary compatible, only source compatible ATM, so they can't
>> really coexist yet. The gralloc implementation specific dependencies
>
> The co-existence means to put the two gralloc
> implementations in the same image and
> select which one to be used by GPU at runtime.
> This is carried out by our init.sh like
>
>         case "$(cat /proc/fb | head -1)" in
>                 *virtiodrmfb)
>                         set_property ro.hardware.gralloc gbm
>                         set_property ro.hardware.hwcomposer drm
>                         ;;
>                 0*inteldrmfb|0*radeondrmfb|0*nouveaufb|0*svgadrmfb)
>                         set_property ro.hardware.gralloc drm
>                         set_drm_mode
>                         ;;
>                 ...
>         esac
>
> They don't need to be binary or source compatible, I think.

They have to be binary compatible with mesa and hwc as those are
dependent on the specific gralloc implementation.

> I guess the first supported GPU is virgl. Right?

Yes. Any gallium driver really. The classic mesa drivers will need
their own additions for GBM map/unmap.

> When could we expect it's ready to test?

Not sure. Definitely not until the GBM interface is set. There's not
really much reason for android-x86 to move to it until it gets flushed
out.

Rob