Phase out ttm_exec

[Christian König]

Am 22.11.24 um 05:22 schrieb Zack Rusin:
> On Thu, Nov 21, 2024 at 5:27 AM Christian König
> <ckoenig.leichtzumerken at gmail.com> wrote:
>> Hey Zack,
>>
>> vmwgfx is the only driver left using this component.
>>
>> So I though I go ahead and complete switching over to drm_exec.
>> The patches are only compile tested, but relatively staright forward.
> Hi, Christian.
>
> Thank you so much for tackling this! It's on our todo.
>
> I just applied the patches and they break rather badly,

Crap, but yeah that was kind of expected.

The changes are rather complicated and not as straight forward as it was 
on the other drivers, I probably just missed a detail.

>   i.e. I don't
> see any command submission with the changes. I won't have the time to
> look at it today, I'll try to carve out some time over the weekend to
> debug this. Two other observations:
> 1) The patches break the build, I think they're missing:
> diff --git a/drivers/gpu/drm/vmwgfx/Kconfig b/drivers/gpu/drm/vmwgfx/Kconfig
> index 6c3c2922ae8b..f7aff3fdfb1a 100644
> --- a/drivers/gpu/drm/vmwgfx/Kconfig
> +++ b/drivers/gpu/drm/vmwgfx/Kconfig
> @@ -4,6 +4,7 @@ config DRM_VMWGFX
>          depends on DRM && PCI && MMU
>          depends on (X86 && HYPERVISOR_GUEST) || ARM64
>          select DRM_CLIENT_SELECTION
> +       select DRM_EXEC
>          select DRM_TTM
>          select DRM_TTM_HELPER
>          select MAPPING_DIRTY_HELPERS

Oh, good point. I missed that on radeon as well because other drivers 
selected that dependency.

> 2) qxl would need to be patched as well. At least it uses
> ttm_eu_reserve_buffers which you're removing.

I pushed the qxl change the day before yesterday. Could only compile 
test them, but their were so straight forward that I had enough 
confidence that they will work. If not we can trivially revert them.

Thanks,
Christian.

>
> z