Am 11.04.2018 um 11:22 schrieb Christian König:
> Am 11.04.2018 um 04:38 schrieb zhoucm1:
>>
>>
>> On 2018å¹´04æ??10æ?¥ 21:42, Christian König wrote:
>>> When GEM needs to fallback to GTT for VRAM BOs we still want the
>>> preferred domain to be untouched so that the BO has a cance to move
>>> back
>>> to VRAM in the future.
>>>
>>> Signed-off-by: Christian König <christian.koenig at amd.com>
>>> ---
>>>  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c   | 14 +++++++++++---
>>> Â drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 13 ++-----------
>>> Â 2 files changed, 13 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>> index 46b9ea4e6103..9dc0a190413c 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>> @@ -47,6 +47,7 @@ int amdgpu_gem_object_create(struct amdgpu_device
>>> *adev, unsigned long size,
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct reservation_object *resv,
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_gem_object **obj)
>>> Â {
>>> +Â Â Â uint32_t domain = initial_domain;
>>> Â Â Â Â Â struct amdgpu_bo *bo;
>>> Â Â Â Â Â int r;
>>> Â @@ -57,7 +58,7 @@ int amdgpu_gem_object_create(struct
>>> amdgpu_device *adev, unsigned long size,
>>> Â Â Â Â Â }
>>> Â Â retry:
>>> -Â Â Â r = amdgpu_bo_create(adev, size, alignment, initial_domain,
>>> +Â Â Â r = amdgpu_bo_create(adev, size, alignment, domain,
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â flags, type, resv, &bo);
>>> Â Â Â Â Â if (r) {
>>> Â Â Â Â Â Â Â Â Â if (r != -ERESTARTSYS) {
>>> @@ -66,8 +67,8 @@ int amdgpu_gem_object_create(struct amdgpu_device
>>> *adev, unsigned long size,
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â goto retry;
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â }
>>> Â -Â Â Â Â Â Â Â Â Â Â Â if (initial_domain == AMDGPU_GEM_DOMAIN_VRAM) {
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â initial_domain |= AMDGPU_GEM_DOMAIN_GTT;
>>> +Â Â Â Â Â Â Â Â Â Â Â if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
>>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â domain |= AMDGPU_GEM_DOMAIN_GTT;
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â goto retry;
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â }
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â DRM_DEBUG("Failed to allocate GEM object (%ld, %d, %u,
>>> %d)\n",
>>> @@ -75,6 +76,13 @@ int amdgpu_gem_object_create(struct amdgpu_device
>>> *adev, unsigned long size,
>>> Â Â Â Â Â Â Â Â Â }
>>> Â Â Â Â Â Â Â Â Â return r;
>>> Â Â Â Â Â }
>>> +
>>> +Â Â Â bo->preferred_domains = initial_domain;
>>> +Â Â Â bo->allowed_domains = initial_domain;
>>> +Â Â Â if (type != ttm_bo_type_kernel &&
>>> +Â Â Â Â Â Â Â bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
>>> +Â Â Â Â Â Â Â bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
>> It's an ugly change back after bo creation! Do you think it's better
>> than previous?
>
> Well it's better than moving the fallback handling into the core
> functions and then adding a flag to disable it again because we don't
> want it in the core function.
>
>>
>> Alternative way, you can add one parameter to amdgpu_bo_create,
>> directly pass preferred_domains and allowed_domains to amdgpu_bo_create.
>
> Then we would need three parameters, one where to create the BO now
> and two for preferred/allowed domains.
>
> I think that in this case overwriting the placement from the initial
> value looks cleaner to me.
Ping? Any more opinions on how to proceed?
I agree that neither solution is perfect, but updating the placement
after we created the BO still sounds like the least painful to me.
Christian.
>
> Regards,
> Christian.
>
>>
>> Regards,
>> David Zhou
>>> +
>>> Â Â Â Â Â *obj = &bo->gem_base;
>>> Â Â Â Â Â Â return 0;
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> index 6d08cde8443c..854d18d5cff4 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> @@ -375,17 +375,8 @@ static int amdgpu_bo_do_create(struct
>>> amdgpu_device *adev, unsigned long size,
>>> Â Â Â Â Â drm_gem_private_object_init(adev->ddev, &bo->gem_base, size);
>>> Â Â Â Â Â INIT_LIST_HEAD(&bo->shadow_list);
>>> Â Â Â Â Â INIT_LIST_HEAD(&bo->va);
>>> -Â Â Â bo->preferred_domains = domain & (AMDGPU_GEM_DOMAIN_VRAM |
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â AMDGPU_GEM_DOMAIN_GTT |
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â AMDGPU_GEM_DOMAIN_CPU |
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â AMDGPU_GEM_DOMAIN_GDS |
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â AMDGPU_GEM_DOMAIN_GWS |
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â AMDGPU_GEM_DOMAIN_OA);
>>> -Â Â Â bo->allowed_domains = bo->preferred_domains;
>>> -Â Â Â if (type != ttm_bo_type_kernel &&
>>> -Â Â Â Â Â Â Â bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
>>> -Â Â Â Â Â Â Â bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
>>> -
>>> +Â Â Â bo->preferred_domains = domain;
>>> +Â Â Â bo->allowed_domains = domain;
>>> Â Â Â Â Â bo->flags = flags;
>>> Â Â #ifdef CONFIG_X86_32
>>
>
