thr3ads.net - Nouveau - [Nouveau] [PATCH v2 2/4] drm/amdkfd: Use cursor start instead of ttm resource start [Jan 2023]

If this information is useful, please help other people find it:
Share via:

Somalapuram Amaranath

2023-Jan-25 10:49 UTC

[Nouveau] [PATCH v2 1/4] drm/amdgpu: Use cursor start instead of ttm resource start

cleanup PAGE_SHIFT operation and replacing
ttm_resource resource->start with cursor start
using amdgpu_res_first API.
v1 -> v2: reorder patch sequence

Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 ++++++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 10 +++++++---
 2 files changed, 15 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 25a68d8888e0..2ab67ab204df 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1346,6 +1346,7 @@ vm_fault_t amdgpu_bo_fault_reserve_notify(struct
ttm_buffer_object *bo)
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
 	struct ttm_operation_ctx ctx = { false, false };
 	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
+	struct amdgpu_res_cursor cursor;
 	unsigned long offset;
 	int r;
 
@@ -1355,7 +1356,8 @@ vm_fault_t amdgpu_bo_fault_reserve_notify(struct
ttm_buffer_object *bo)
 	if (bo->resource->mem_type != TTM_PL_VRAM)
 		return 0;
 
-	offset = bo->resource->start << PAGE_SHIFT;
+	amdgpu_res_first(bo->resource, 0, bo->resource->size, &cursor);
+	offset = cursor.start;
 	if ((offset + bo->base.size) <= adev->gmc.visible_vram_size)
 		return 0;
 
@@ -1378,7 +1380,8 @@ vm_fault_t amdgpu_bo_fault_reserve_notify(struct
ttm_buffer_object *bo)
 	else if (unlikely(r))
 		return VM_FAULT_SIGBUS;
 
-	offset = bo->resource->start << PAGE_SHIFT;
+	amdgpu_res_first(bo->resource, 0, bo->resource->size, &cursor);
+	offset = cursor.start;
 	/* this should never happen */
 	if (bo->resource->mem_type == TTM_PL_VRAM &&
 	    (offset + bo->base.size) > adev->gmc.visible_vram_size)
@@ -1491,9 +1494,11 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
 u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct amdgpu_res_cursor cursor;
 	uint64_t offset;
 
-	offset = (bo->tbo.resource->start << PAGE_SHIFT) +
+	amdgpu_res_first(bo->tbo.resource, 0, bo->tbo.resource->size,
&cursor);
+	offset = cursor.start +
 		 amdgpu_ttm_domain_start(adev, bo->tbo.resource->mem_type);
 
 	return amdgpu_gmc_sign_extend(offset);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c5ef7f7bdc15..ffe6a1ab7f9a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -849,6 +849,7 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
 	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
+	struct amdgpu_res_cursor cursor;
 	uint64_t flags;
 	int r;
 
@@ -896,7 +897,8 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
 	flags = amdgpu_ttm_tt_pte_flags(adev, ttm, bo_mem);
 
 	/* bind pages into GART page tables */
-	gtt->offset = (u64)bo_mem->start << PAGE_SHIFT;
+	amdgpu_res_first(bo_mem, 0, bo_mem->size, &cursor);
+	gtt->offset = cursor.start;
 	amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
 			 gtt->ttm.dma_address, flags);
 	gtt->bound = true;
@@ -916,6 +918,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
 	struct ttm_operation_ctx ctx = { false, false };
 	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(bo->ttm);
+	struct amdgpu_res_cursor cursor;
 	struct ttm_placement placement;
 	struct ttm_place placements;
 	struct ttm_resource *tmp;
@@ -927,7 +930,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 
 	addr = amdgpu_gmc_agp_addr(bo);
 	if (addr != AMDGPU_BO_INVALID_OFFSET) {
-		bo->resource->start = addr >> PAGE_SHIFT;
+		bo->resource->start = addr;
 		return 0;
 	}
 
@@ -949,7 +952,8 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 	flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, tmp);
 
 	/* Bind pages */
-	gtt->offset = (u64)tmp->start << PAGE_SHIFT;
+	amdgpu_res_first(tmp, 0, tmp->size, &cursor);
+	gtt->offset = cursor.start;
 	amdgpu_ttm_gart_bind(adev, bo, flags);
 	amdgpu_gart_invalidate_tlb(adev);
 	ttm_resource_free(bo, &bo->resource);
-- 
2.32.0

Somalapuram Amaranath

2023-Jan-25 10:49 UTC

head link

[Nouveau] [PATCH v2 2/4] drm/amdkfd: Use cursor start instead of ttm resource start

cleanup PAGE_SHIFT operation and replacing
ttm_resource resource->start with cursor start
using amdgpu_res_first API
v1 -> v2: reorder patch sequence

Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram at amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index c06ada0844ba..f87ce4f1cb93 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -200,8 +200,11 @@ static int add_queue_mes(struct device_queue_manager *dqm,
struct queue *q,
 	queue_input.wptr_addr = (uint64_t)q->properties.write_ptr;
 
 	if (q->wptr_bo) {
+		struct amdgpu_res_cursor cursor;
 		wptr_addr_off = (uint64_t)q->properties.write_ptr & (PAGE_SIZE - 1);
-		queue_input.wptr_mc_addr =
((uint64_t)q->wptr_bo->tbo.resource->start << PAGE_SHIFT) +
wptr_addr_off;
+		amdgpu_res_first(q->wptr_bo->tbo.resource, 0,
+				 q->wptr_bo->tbo.resource->size, &cursor);
+		queue_input.wptr_mc_addr = cursor.start + wptr_addr_off;
 	}
 
 	queue_input.is_kfd_process = 1;
-- 
2.32.0

Somalapuram Amaranath

2023-Jan-25 10:49 UTC

head link

[Nouveau] [PATCH v2 3/4] drm/amdgpu: Movie the amdgpu_gtt_mgr start and size from pages to bytes

To support GTT manager amdgpu_res_first, amdgpu_res_next
from pages to bytes and clean up PAGE_SHIFT operation.
v1 -> v2: reorder patch sequence

Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
index 5c4f93ee0c57..5c78f0b09351 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
@@ -94,8 +94,8 @@ static inline void amdgpu_res_first(struct ttm_resource *res,
 		while (start >= node->size << PAGE_SHIFT)
 			start -= node++->size << PAGE_SHIFT;
 
-		cur->start = (node->start << PAGE_SHIFT) + start;
-		cur->size = min((node->size << PAGE_SHIFT) - start, size);
+		cur->start = node->start + start;
+		cur->size = min(node->size - start, size);
 		cur->remaining = size;
 		cur->node = node;
 		break;
@@ -155,8 +155,8 @@ static inline void amdgpu_res_next(struct amdgpu_res_cursor
*cur, uint64_t size)
 		node = cur->node;
 
 		cur->node = ++node;
-		cur->start = node->start << PAGE_SHIFT;
-		cur->size = min(node->size << PAGE_SHIFT, cur->remaining);
+		cur->start = node->start;
+		cur->size = min(node->size, cur->remaining);
 		break;
 	default:
 		return;
-- 
2.32.0

Somalapuram Amaranath

2023-Jan-25 10:49 UTC

head link

[Nouveau] [PATCH v2 4/4] drm/amdgpu: Support allocate of amdgpu_gtt_mgr from pages to bytes

Change the GTT manager init and allocate from pages to bytes
v1 -> v2: reorder patch sequence

Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 44367f03316f..a1fbfc5984d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -116,7 +116,6 @@ static int amdgpu_gtt_mgr_new(struct ttm_resource_manager
*man,
 			      struct ttm_resource **res)
 {
 	struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-	uint32_t num_pages = PFN_UP(tbo->base.size);
 	struct ttm_range_mgr_node *node;
 	int r;
 
@@ -134,8 +133,10 @@ static int amdgpu_gtt_mgr_new(struct ttm_resource_manager
*man,
 	if (place->lpfn) {
 		spin_lock(&mgr->lock);
 		r = drm_mm_insert_node_in_range(&mgr->mm, &node->mm_nodes[0],
-						num_pages, tbo->page_alignment,
-						0, place->fpfn, place->lpfn,
+						tbo->base.size,
+						tbo->page_alignment << PAGE_SHIFT, 0,
+						place->fpfn << PAGE_SHIFT,
+						place->lpfn << PAGE_SHIFT,
 						DRM_MM_INSERT_BEST);
 		spin_unlock(&mgr->lock);
 		if (unlikely(r))
@@ -144,7 +145,7 @@ static int amdgpu_gtt_mgr_new(struct ttm_resource_manager
*man,
 		node->base.start = node->mm_nodes[0].start;
 	} else {
 		node->mm_nodes[0].start = 0;
-		node->mm_nodes[0].size = PFN_UP(node->base.size);
+		node->mm_nodes[0].size = node->base.size;
 		node->base.start = AMDGPU_BO_INVALID_OFFSET;
 	}
 
@@ -285,8 +286,8 @@ int amdgpu_gtt_mgr_init(struct amdgpu_device *adev, uint64_t
gtt_size)
 
 	ttm_resource_manager_init(man, &adev->mman.bdev, gtt_size);
 
-	start = AMDGPU_GTT_MAX_TRANSFER_SIZE * AMDGPU_GTT_NUM_TRANSFER_WINDOWS;
-	size = (adev->gmc.gart_size >> PAGE_SHIFT) - start;
+	start = (AMDGPU_GTT_MAX_TRANSFER_SIZE * AMDGPU_GTT_NUM_TRANSFER_WINDOWS)
<< PAGE_SHIFT;
+	size = adev->gmc.gart_size - start;
 	drm_mm_init(&mgr->mm, start, size);
 	spin_lock_init(&mgr->lock);
 
-- 
2.32.0

Christian König

2023-Jan-25 11:41 UTC

head link

[Nouveau] [PATCH v2 1/4] drm/amdgpu: Use cursor start instead of ttm resource start

Am 25.01.23 um 11:48 schrieb Somalapuram Amaranath:> cleanup PAGE_SHIFT operation and replacing
> ttm_resource resource->start with cursor start
> using amdgpu_res_first API.
> v1 -> v2: reorder patch sequence
>
> Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram at
amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 ++++++++---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 10 +++++++---
>   2 files changed, 15 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index 25a68d8888e0..2ab67ab204df 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -1346,6 +1346,7 @@ vm_fault_t amdgpu_bo_fault_reserve_notify(struct
ttm_buffer_object *bo)
>   	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>   	struct ttm_operation_ctx ctx = { false, false };
>   	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
> +	struct amdgpu_res_cursor cursor;
>   	unsigned long offset;
>   	int r;
>   
> @@ -1355,7 +1356,8 @@ vm_fault_t amdgpu_bo_fault_reserve_notify(struct
ttm_buffer_object *bo)
>   	if (bo->resource->mem_type != TTM_PL_VRAM)
>   		return 0;
>   
> -	offset = bo->resource->start << PAGE_SHIFT;
> +	amdgpu_res_first(bo->resource, 0, bo->resource->size,
&cursor);
> +	offset = cursor.start;
That won't work like this. We use a virtual resource start for this 
check here.
>   	if ((offset + bo->base.size) <= adev->gmc.visible_vram_size)
>   		return 0;
Just replace this whole test with calling amdgpu_bo_in_cpu_visible_vram().
>   
> @@ -1378,7 +1380,8 @@ vm_fault_t amdgpu_bo_fault_reserve_notify(struct
ttm_buffer_object *bo)
>   	else if (unlikely(r))
>   		return VM_FAULT_SIGBUS;
>   
> -	offset = bo->resource->start << PAGE_SHIFT;
> +	amdgpu_res_first(bo->resource, 0, bo->resource->size,
&cursor);
> +	offset = cursor.start;
>   	/* this should never happen */
>   	if (bo->resource->mem_type == TTM_PL_VRAM &&
>   	    (offset + bo->base.size) > adev->gmc.visible_vram_size)
Same here, just call amdgpu_bo_in_cpu_visible_vram() instead.
> @@ -1491,9 +1494,11 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
>   u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo)
>   {
>   	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
> +	struct amdgpu_res_cursor cursor;
>   	uint64_t offset;
>   
> -	offset = (bo->tbo.resource->start << PAGE_SHIFT) +
> +	amdgpu_res_first(bo->tbo.resource, 0, bo->tbo.resource->size,
&cursor);
> +	offset = cursor.start +
>   		 amdgpu_ttm_domain_start(adev, bo->tbo.resource->mem_type);
>   
>   	return amdgpu_gmc_sign_extend(offset);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index c5ef7f7bdc15..ffe6a1ab7f9a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -849,6 +849,7 @@ static int amdgpu_ttm_backend_bind(struct ttm_device
*bdev,
>   {
>   	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
>   	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
> +	struct amdgpu_res_cursor cursor;
>   	uint64_t flags;
>   	int r;
>   
> @@ -896,7 +897,8 @@ static int amdgpu_ttm_backend_bind(struct ttm_device
*bdev,
>   	flags = amdgpu_ttm_tt_pte_flags(adev, ttm, bo_mem);
>   
>   	/* bind pages into GART page tables */
> -	gtt->offset = (u64)bo_mem->start << PAGE_SHIFT;
> +	amdgpu_res_first(bo_mem, 0, bo_mem->size, &cursor);
> +	gtt->offset = cursor.start;
>   	amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
>   			 gtt->ttm.dma_address, flags);
>   	gtt->bound = true;
> @@ -916,6 +918,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
>   	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>   	struct ttm_operation_ctx ctx = { false, false };
>   	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(bo->ttm);
> +	struct amdgpu_res_cursor cursor;
>   	struct ttm_placement placement;
>   	struct ttm_place placements;
>   	struct ttm_resource *tmp;
> @@ -927,7 +930,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
>   
>   	addr = amdgpu_gmc_agp_addr(bo);
>   	if (addr != AMDGPU_BO_INVALID_OFFSET) {
> -		bo->resource->start = addr >> PAGE_SHIFT;
> +		bo->resource->start = addr;
This doesn't belong into this patch and would break if only this patch 
is applied.

Please move it into a later patch.

Apart from those comments that looks good now,
Christian.
>   		return 0;
>   	}
>   
> @@ -949,7 +952,8 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
>   	flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, tmp);
>   
>   	/* Bind pages */
> -	gtt->offset = (u64)tmp->start << PAGE_SHIFT;
> +	amdgpu_res_first(tmp, 0, tmp->size, &cursor);
> +	gtt->offset = cursor.start;
>   	amdgpu_ttm_gart_bind(adev, bo, flags);
>   	amdgpu_gart_invalidate_tlb(adev);
>   	ttm_resource_free(bo, &bo->resource);

Christian König

2023-Jan-25 11:43 UTC

head link

[Nouveau] [PATCH v2 2/4] drm/amdkfd: Use cursor start instead of ttm resource start

Am 25.01.23 um 11:48 schrieb Somalapuram Amaranath:> cleanup PAGE_SHIFT operation and replacing
> ttm_resource resource->start with cursor start
> using amdgpu_res_first API
> v1 -> v2: reorder patch sequence
>
> Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram at
amd.com>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 5 ++++-
>   1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> index c06ada0844ba..f87ce4f1cb93 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> @@ -200,8 +200,11 @@ static int add_queue_mes(struct device_queue_manager
*dqm, struct queue *q,
>   	queue_input.wptr_addr = (uint64_t)q->properties.write_ptr;
>   
>   	if (q->wptr_bo) {
> +		struct amdgpu_res_cursor cursor;
>   		wptr_addr_off = (uint64_t)q->properties.write_ptr & (PAGE_SIZE -
1);
Add an empty line between declaration and code or otherwise the 
automated checkers will complain.

Apart from this nit pick the patch is Reviewed-by: Christian K?nig 
<christian.koenig at amd.com>

Regards,
Christian.
> -		queue_input.wptr_mc_addr =
((uint64_t)q->wptr_bo->tbo.resource->start << PAGE_SHIFT) +
wptr_addr_off;
> +		amdgpu_res_first(q->wptr_bo->tbo.resource, 0,
> +				 q->wptr_bo->tbo.resource->size, &cursor);
> +		queue_input.wptr_mc_addr = cursor.start + wptr_addr_off;
>   	}
>   
>   	queue_input.is_kfd_process = 1;

Nouveau - Jan 2023 - [PATCH v2 2/4] drm/amdkfd: Use cursor start instead of ttm resource start

[Nouveau] [PATCH v2 1/4] drm/amdgpu: Use cursor start instead of ttm resource start

[Nouveau] [PATCH v2 2/4] drm/amdkfd: Use cursor start instead of ttm resource start

[Nouveau] [PATCH v2 3/4] drm/amdgpu: Movie the amdgpu_gtt_mgr start and size from pages to bytes

[Nouveau] [PATCH v2 4/4] drm/amdgpu: Support allocate of amdgpu_gtt_mgr from pages to bytes

[Nouveau] [PATCH v2 1/4] drm/amdgpu: Use cursor start instead of ttm resource start

[Nouveau] [PATCH v2 2/4] drm/amdkfd: Use cursor start instead of ttm resource start