[PATCH 1/3] drm/amdgpu: use ctx bytes_moved

[PATCH 1/3] drm/amdgpu: use ctx bytes_moved

[Christian König]

Instead of the global (inaccurate) counter.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index d7d7ce1507ec..de69ab12bb55 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -412,7 +412,6 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
 		struct amdgpu_bo_list_entry *candidate = p->evictable;
 		struct amdgpu_bo *bo = candidate->robj;
 		struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-		u64 initial_bytes_moved, bytes_moved;
 		bool update_bytes_moved_vis;
 		uint32_t other;
 
@@ -436,18 +435,15 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
 			continue;
 
 		/* Good we can try to move this BO somewhere else */
-		amdgpu_ttm_placement_from_domain(bo, other);
 		update_bytes_moved_vis =
 			adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
 			bo->tbo.mem.mem_type == TTM_PL_VRAM &&
 			bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT;
-		initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
+		amdgpu_ttm_placement_from_domain(bo, other);
 		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
-		bytes_moved = atomic64_read(&adev->num_bytes_moved) -
-			initial_bytes_moved;
-		p->bytes_moved += bytes_moved;
+		p->bytes_moved += ctx.bytes_moved;
 		if (update_bytes_moved_vis)
-			p->bytes_moved_vis += bytes_moved;
+			p->bytes_moved_vis += ctx.bytes_moved;
 
 		if (unlikely(r))
 			break;
-- 
2.14.1

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

[PATCH 2/3] drm/amdgpu: fix and cleanup cpu visible VRAM handling

[Christian König]

The detection if a BO was placed in CPU visible VRAM was incorrect.

Fix it and merge it with the correct detection in amdgpu_ttm.c

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c     |  6 ++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 21 +++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 19 +++----------------
 3 files changed, 26 insertions(+), 20 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index de69ab12bb55..68af2f878bc9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -382,8 +382,7 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
 
 	p->bytes_moved += ctx.bytes_moved;
 	if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
-	    bo->tbo.mem.mem_type == TTM_PL_VRAM &&
-	    bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT)
+	    amdgpu_bo_in_cpu_visible_vram(bo))
 		p->bytes_moved_vis += ctx.bytes_moved;
 
 	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains &&
@@ -437,8 +436,7 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
 		/* Good we can try to move this BO somewhere else */
 		update_bytes_moved_vis =
 			adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
-			bo->tbo.mem.mem_type == TTM_PL_VRAM &&
-			bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT;
+			amdgpu_bo_in_cpu_visible_vram(bo);
 		amdgpu_ttm_placement_from_domain(bo, other);
 		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
 		p->bytes_moved += ctx.bytes_moved;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index deb10ecd89a1..0e9555526950 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -193,6 +193,27 @@ static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
 	}
 }
 
+/**
+ * amdgpu_bo_in_cpu_visible_vram - check if BO is (partly) in visible VRAM
+ */
+static inline bool amdgpu_bo_in_cpu_visible_vram(struct amdgpu_bo *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	unsigned fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
+	struct drm_mm_node *node = bo->tbo.mem.mm_node;
+	unsigned long pages_left;
+
+	if (bo->tbo.mem.mem_type != TTM_PL_VRAM)
+		return false;
+
+	for (pages_left = bo->tbo.mem.num_pages; pages_left;
+	     pages_left -= node->size, node++)
+		if (node->start < fpfn)
+			return true;
+
+	return false;
+}
+
 /**
  * amdgpu_bo_explicit_sync - return whether the bo is explicitly synced
  */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index d2ab40494a4c..ab829fb3e840 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -222,20 +222,8 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
 		if (!adev->mman.buffer_funcs_enabled) {
 			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);
 		} else if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
-			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)) {
-			unsigned fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
-			struct drm_mm_node *node = bo->mem.mm_node;
-			unsigned long pages_left;
-
-			for (pages_left = bo->mem.num_pages;
-			     pages_left;
-			     pages_left -= node->size, node++) {
-				if (node->start < fpfn)
-					break;
-			}
-
-			if (!pages_left)
-				goto gtt;
+			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
+			   amdgpu_bo_in_cpu_visible_vram(abo)) {
 
 			/* Try evicting to the CPU inaccessible part of VRAM
 			 * first, but only set GTT as busy placement, so this
@@ -244,12 +232,11 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
 			 */
 			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM |
 							 AMDGPU_GEM_DOMAIN_GTT);
-			abo->placements[0].fpfn = fpfn;
+			abo->placements[0].fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
 			abo->placements[0].lpfn = 0;
 			abo->placement.busy_placement = &abo->placements[1];
 			abo->placement.num_busy_placement = 1;
 		} else {
-gtt:
 			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT);
 		}
 		break;
-- 
2.14.1

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

Re: [PATCH 2/3] drm/amdgpu: fix and cleanup cpu visible VRAM handling

[Michel Dänzer]

On 2018-04-09 12:18 PM, Christian König wrote:
> The detection if a BO was placed in CPU visible VRAM was incorrect.
> 
> Fix it and merge it with the correct detection in amdgpu_ttm.c
> 
> Signed-off-by: Christian König <christian.koenig@amd.com>

Good catch! All three patches are

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>


-- 
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 2/3] drm/amdgpu: fix and cleanup cpu visible VRAM handling

[zhoucm1]

On 2018年04月09日 18:18, Christian König wrote:
> The detection if a BO was placed in CPU visible VRAM was incorrect.
>
> Fix it and merge it with the correct detection in amdgpu_ttm.c
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c     |  6 ++----
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 21 +++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 19 +++----------------
>   3 files changed, 26 insertions(+), 20 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> index de69ab12bb55..68af2f878bc9 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> @@ -382,8 +382,7 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
>   
>   	p->bytes_moved += ctx.bytes_moved;
>   	if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
> -	    bo->tbo.mem.mem_type == TTM_PL_VRAM &&
> -	    bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT)
> +	    amdgpu_bo_in_cpu_visible_vram(bo))
>   		p->bytes_moved_vis += ctx.bytes_moved;
>   
>   	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains &&
> @@ -437,8 +436,7 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
>   		/* Good we can try to move this BO somewhere else */
>   		update_bytes_moved_vis =
>   			adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
> -			bo->tbo.mem.mem_type == TTM_PL_VRAM &&
> -			bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT;
> +			amdgpu_bo_in_cpu_visible_vram(bo);
>   		amdgpu_ttm_placement_from_domain(bo, other);
>   		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
>   		p->bytes_moved += ctx.bytes_moved;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> index deb10ecd89a1..0e9555526950 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> @@ -193,6 +193,27 @@ static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
>   	}
>   }
>   
> +/**
> + * amdgpu_bo_in_cpu_visible_vram - check if BO is (partly) in visible VRAM
> + */
> +static inline bool amdgpu_bo_in_cpu_visible_vram(struct amdgpu_bo *bo)
> +{
> +	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
> +	unsigned fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
> +	struct drm_mm_node *node = bo->tbo.mem.mm_node;
> +	unsigned long pages_left;
> +
> +	if (bo->tbo.mem.mem_type != TTM_PL_VRAM)
> +		return false;
> +
> +	for (pages_left = bo->tbo.mem.num_pages; pages_left;
> +	     pages_left -= node->size, node++)
> +		if (node->start < fpfn)
> +			return true;
> +
> +	return false;
> +}
> +
>   /**
>    * amdgpu_bo_explicit_sync - return whether the bo is explicitly synced
>    */
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index d2ab40494a4c..ab829fb3e840 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -222,20 +222,8 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
>   		if (!adev->mman.buffer_funcs_enabled) {
>   			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);
>   		} else if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
> -			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)) {
> -			unsigned fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
> -			struct drm_mm_node *node = bo->mem.mm_node;
> -			unsigned long pages_left;
> -
> -			for (pages_left = bo->mem.num_pages;
> -			     pages_left;
> -			     pages_left -= node->size, node++) {
> -				if (node->start < fpfn)
> -					break;
> -			}
> -
> -			if (!pages_left)
> -				goto gtt;
> +			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
> +			   amdgpu_bo_in_cpu_visible_vram(abo)) {
>   
>   			/* Try evicting to the CPU inaccessible part of VRAM
>   			 * first, but only set GTT as busy placement, so this
> @@ -244,12 +232,11 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
>   			 */
>   			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM |
>   							 AMDGPU_GEM_DOMAIN_GTT);
> -			abo->placements[0].fpfn = fpfn;
> +			abo->placements[0].fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
>   			abo->placements[0].lpfn = 0;
>   			abo->placement.busy_placement = &abo->placements[1];
>   			abo->placement.num_busy_placement = 1;
>   		} else {
> -gtt:
>   			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT);
>   		}
>   		break;

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

[PATCH 3/3] drm/amdgpu: remove AMDGPU_GEM_CREATE_NO_FALLBACK handling from CS again

[Christian König]

That should purely be handled by preferred/allowed domains.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 68af2f878bc9..e1756b68a17b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -385,8 +385,7 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
 	    amdgpu_bo_in_cpu_visible_vram(bo))
 		p->bytes_moved_vis += ctx.bytes_moved;
 
-	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains &&
-	    !(bo->flags & AMDGPU_GEM_CREATE_NO_FALLBACK)) {
+	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
 		domain = bo->allowed_domains;
 		goto retry;
 	}
-- 
2.14.1

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

Re: [PATCH 3/3] drm/amdgpu: remove AMDGPU_GEM_CREATE_NO_FALLBACK handling from CS again

[zhoucm1]

On 2018年04月09日 18:19, Christian König wrote:
> That should purely be handled by preferred/allowed domains.
Although this flag isn't exported to user space yet, I'm curious that 
how preferred/allowed domains handle no_fallback?
IIRC, currently, our driver will always add GTT fallback for VRAM bo.

Regards,
David Zhou
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +--
>   1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> index 68af2f878bc9..e1756b68a17b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> @@ -385,8 +385,7 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
>   	    amdgpu_bo_in_cpu_visible_vram(bo))
>   		p->bytes_moved_vis += ctx.bytes_moved;
>   
> -	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains &&
> -	    !(bo->flags & AMDGPU_GEM_CREATE_NO_FALLBACK)) {
> +	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
>   		domain = bo->allowed_domains;
>   		goto retry;
>   	}

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

Re: [PATCH 3/3] drm/amdgpu: remove AMDGPU_GEM_CREATE_NO_FALLBACK handling from CS again

[Christian König]

Am 10.04.2018 um 04:43 schrieb zhoucm1:
>
>
> On 2018年04月09日 18:19, Christian König wrote:
>> That should purely be handled by preferred/allowed domains.
> Although this flag isn't exported to user space yet, I'm curious that 
> how preferred/allowed domains handle no_fallback?
> IIRC, currently, our driver will always add GTT fallback for VRAM bo.

And that is intentional. Going a step further back I think moving the 
fallback handling into amdgpu_bo_do_create() and adding the flag was a 
mistake to begin with.

Going to send patches to revert all this and further clean the stuff up.

Regards,
Christian.

>
> Regards,
> David Zhou
>>
>> Signed-off-by: Christian König <christian.koenig@amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +--
>>   1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> index 68af2f878bc9..e1756b68a17b 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> @@ -385,8 +385,7 @@ static int amdgpu_cs_bo_validate(struct 
>> amdgpu_cs_parser *p,
>>           amdgpu_bo_in_cpu_visible_vram(bo))
>>           p->bytes_moved_vis += ctx.bytes_moved;
>>   -    if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains &&
>> -        !(bo->flags & AMDGPU_GEM_CREATE_NO_FALLBACK)) {
>> +    if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
>>           domain = bo->allowed_domains;
>>           goto retry;
>>       }
>

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

Re: [PATCH 3/3] drm/amdgpu: remove AMDGPU_GEM_CREATE_NO_FALLBACK handling from CS again

[zhoucm1]

On 2018年04月10日 17:00, Christian König wrote:
> Am 10.04.2018 um 04:43 schrieb zhoucm1:
>>
>>
>> On 2018年04月09日 18:19, Christian König wrote:
>>> That should purely be handled by preferred/allowed domains.
>> Although this flag isn't exported to user space yet, I'm curious that 
>> how preferred/allowed domains handle no_fallback?
>> IIRC, currently, our driver will always add GTT fallback for VRAM bo.
>
> And that is intentional. Going a step further back I think moving the 
> fallback handling into amdgpu_bo_do_create() and adding the flag was a 
> mistake to begin with.
>
> Going to send patches to revert all this and further clean the stuff up.
if you are able to not change the preferred domain when fallback, it's 
no problem to me.

David Zhou
>
> Regards,
> Christian.
>
>>
>> Regards,
>> David Zhou
>>>
>>> Signed-off-by: Christian König <christian.koenig@amd.com>
>>> ---
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +--
>>>   1 file changed, 1 insertion(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> index 68af2f878bc9..e1756b68a17b 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> @@ -385,8 +385,7 @@ static int amdgpu_cs_bo_validate(struct 
>>> amdgpu_cs_parser *p,
>>>           amdgpu_bo_in_cpu_visible_vram(bo))
>>>           p->bytes_moved_vis += ctx.bytes_moved;
>>>   -    if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains &&
>>> -        !(bo->flags & AMDGPU_GEM_CREATE_NO_FALLBACK)) {
>>> +    if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
>>>           domain = bo->allowed_domains;
>>>           goto retry;
>>>       }
>>
>

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

Re: [PATCH 1/3] drm/amdgpu: use ctx bytes_moved

[zhoucm1]

On 2018年04月09日 18:18, Christian König wrote:
> Instead of the global (inaccurate) counter.
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 10 +++-------
>   1 file changed, 3 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> index d7d7ce1507ec..de69ab12bb55 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> @@ -412,7 +412,6 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
>   		struct amdgpu_bo_list_entry *candidate = p->evictable;
>   		struct amdgpu_bo *bo = candidate->robj;
>   		struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
> -		u64 initial_bytes_moved, bytes_moved;
>   		bool update_bytes_moved_vis;
>   		uint32_t other;
>   
> @@ -436,18 +435,15 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
>   			continue;
>   
>   		/* Good we can try to move this BO somewhere else */
> -		amdgpu_ttm_placement_from_domain(bo, other);
>   		update_bytes_moved_vis =
>   			adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
>   			bo->tbo.mem.mem_type == TTM_PL_VRAM &&
>   			bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT;
> -		initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
> +		amdgpu_ttm_placement_from_domain(bo, other);
>   		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
> -		bytes_moved = atomic64_read(&adev->num_bytes_moved) -
> -			initial_bytes_moved;
> -		p->bytes_moved += bytes_moved;
> +		p->bytes_moved += ctx.bytes_moved;
>   		if (update_bytes_moved_vis)
> -			p->bytes_moved_vis += bytes_moved;
> +			p->bytes_moved_vis += ctx.bytes_moved;
>   
>   		if (unlikely(r))
>   			break;

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