Re: [PATCH 07/23] drm/vmwgfx: stop using dma_resv_excl_fence

["Christian König" <christian.koenig@amd.com>]

Am 21.03.22 um 16:11 schrieb Zack Rusin:
> On Mon, 2022-03-21 at 15:12 +0100, Christian König wrote:
>> Am 21.03.22 um 15:02 schrieb Zack Rusin:
>>> On Mon, 2022-03-21 at 14:58 +0100, Christian König wrote:
>>>> ⚠ External Email: This email originated from outside of the
>>>> organization. Do not click links or open attachments unless you
>>>> recognize the sender.
>>>>
>>>> Instead use the new dma_resv_get_singleton function.
>>>>
>>>> Signed-off-by: Christian König <christian.koenig@amd.com>
>>>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>>> Cc: VMware Graphics <linux-graphics-maintainer@vmware.com>
>>>> Cc: Zack Rusin <zackr@vmware.com>
>>>> ---
>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 6 ++++--
>>>>    1 file changed, 4 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
>>>> b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
>>>> index 708899ba2102..36c3b5db7e69 100644
>>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
>>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
>>>> @@ -1165,8 +1165,10 @@ int vmw_resources_clean(struct
>>>> vmw_buffer_object *vbo, pgoff_t start,
>>>>                   vmw_bo_fence_single(bo, NULL);
>>>>                   if (bo->moving)
>>>>                           dma_fence_put(bo->moving);
>>>> -               bo->moving = dma_fence_get
>>>> -                       (dma_resv_excl_fence(bo->base.resv));
>>>> +
>>>> +               /* TODO: This is actually a memory management
>>>> dependency */
>>>> +               return dma_resv_get_singleton(bo->base.resv,
>>>> false,
>>>> +                                             &bo->moving);
>>>>           }
>>>>
>>>>           return 0;
>>>> --
>>>> 2.25.1
>>>>
>>> Sorry, I haven't had the time to go over the entire series, the
>>> patch
>>> looks good, but what's the memory management dependency the todo
>>> mentions?
>> Previously the function installed only the exclusive fence as moving
>> fence into the BO.
>> Now it grabs all fences and installs them as moving fence into the
>> BO.
>>
>> But what we really need is tracking if a fence in the reservation
>> object
>> is a kernel memory management dependency or not.
>>
>> Patch #19 adds that and patch #23 then finally gets rid of the whole
>> bo->moving handling here because it becomes completely unnecessary.
>>
>> I can drop the comment if you want or just note that it is only
>> temporary until the follow up patches are merged.
> Ah, yes, if you could remove it that'd be great. The patch will never
> be backported anywhere without the rest of the series, so it shouldn't
> be a problem.

Sure, can I then have your rb for this one?

Thanks,
Christian.

>
> z