Re: [PATCH] drm/amdgpu: consider user preference when pinning for SG display

From: "Christian König" <ckoenig.leichtzumerken-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
To: Alex Deucher
	<alexdeucher-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>,
	Samuel Li <samuel.li-5C7GfCeVMHo@public.gmane.org>
Cc: "Alex Deucher" <alexander.deucher-5C7GfCeVMHo@public.gmane.org>,
	"Michel Dänzer" <michel-otUistvHUpPR7s880joybQ@public.gmane.org>,
	"amd-gfx list"
	<amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>
Subject: Re: [PATCH] drm/amdgpu: consider user preference when pinning for SG display
Date: Sat, 19 May 2018 17:59:19 +0200	[thread overview]
Message-ID: <7d9abb21-8cd9-c860-8fd1-5e8a74278a01@gmail.com> (raw)
In-Reply-To: <CADnq5_O+y23n9wKdWNbTk8phx34Tt3Cj3NRQ2AqN=qAmYrxppQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>

Am 18.05.2018 um 21:16 schrieb Alex Deucher:
> On Fri, May 18, 2018 at 2:22 PM, Samuel Li <samuel.li@amd.com> wrote:
>>
>> On 2018-05-18 04:21 AM, Michel Dänzer wrote:
>>> On 2018-05-17 06:55 PM, Alex Deucher wrote:
>>>> If the pin domain is set to GTT | VRAM, look at the preferred domains
>>>> for the bo and respect that if it's been set explicitly.
>>>>
>>>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>>>> ---
>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 9 +++++++--
>>>>   1 file changed, 7 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>>> index 6a9e46ae7f0a..16192f17653e 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>>> @@ -704,9 +704,14 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
>>>>        * See function amdgpu_display_supported_domains()
>>>>        */
>>>>       if (domain == (AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT)) {
>>>> -            domain = AMDGPU_GEM_DOMAIN_VRAM;
>>>> -            if (adev->gmc.real_vram_size <= AMDGPU_SG_THRESHOLD)
>>>> +            if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_VRAM)
>>>> +                    domain = AMDGPU_GEM_DOMAIN_VRAM; /* if user really wants vram, respect it */
>>>> +            else if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT)
>>>> +                    domain = AMDGPU_GEM_DOMAIN_GTT; /* if user really wants gtt, respect it */
>>> I'd spell VRAM and GTT in capital letters in the comments.
>>>
>>>
>>>> +            else if (adev->gmc.real_vram_size <= AMDGPU_SG_THRESHOLD)
>>>>                       domain = AMDGPU_GEM_DOMAIN_GTT;
>>>> +            else
>>>> +                    domain = AMDGPU_GEM_DOMAIN_VRAM;
>>>>       }
>>> Is everything in place to deal with any issues that might occur when
>>> flipping between buffers in VRAM and GTT?
>>>
>> Besides this question, I am also wondering how to deal with the first display buffer created in kernel at amdgpufb_create_pinned_object()?
> amdgpufb_create_pinned_object() calls
> amdgpu_display_supported_domains() for the initial domain and
> amdgpu_gem_object_create() uses that for the preferred domain so it
> won't change the behavior there.

That was actually something I was not sure about when I initially 
created the patch.

See the FB emulation expects some pointer into a linear PCI resource 
(VRAM), I'm not sure if that can actually deal with something kmapped 
into the vmap (GTT).

Since this is seldom used and only with low resolution it might be 
better to do the conservative approach and use VRAM here.

And we do a full mode set when switching away from FB emulation AFAIK, 
so the move from VRAM to GTT shouldn't be a problem.

Christian.

>
> Alex
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx