amdgpu: fix performance drop

amdgpu: fix performance drop

[Christian König]

Hi Roger,

as promised the proper fix for the performance drop during command submission.

Please test the attached set of patches with your OpenCL test case.

Regards,
Christian.

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

[PATCH 1/6] drm/ttm: remove allow_errors parameter from ttm_bo_force_list_clean

[Christian König]

From: Christian König <christian.koenig@amd.com>

Not allowing errors here is completely pointless and actually dangerous
cause trying to continue on an error can cause an endless loop.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c | 28 +++++++++++-----------------
 1 file changed, 11 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index b5038c5..6683399 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1291,7 +1291,7 @@ int ttm_bo_create(struct ttm_bo_device *bdev,
 EXPORT_SYMBOL(ttm_bo_create);
 
 static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
-					unsigned mem_type, bool allow_errors)
+				   unsigned mem_type)
 {
 	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
 	struct ttm_bo_global *glob = bdev->glob;
@@ -1306,13 +1306,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 	while (!list_empty(&man->lru)) {
 		spin_unlock(&glob->lru_lock);
 		ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
-		if (ret) {
-			if (allow_errors) {
-				return ret;
-			} else {
-				pr_err("Cleanup eviction failed\n");
-			}
-		}
+		if (ret)
+			return ret;
 		spin_lock(&glob->lru_lock);
 	}
 	spin_unlock(&glob->lru_lock);
@@ -1324,13 +1319,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 	if (fence) {
 		ret = fence_wait(fence, false);
 		fence_put(fence);
-		if (ret) {
-			if (allow_errors) {
-				return ret;
-			} else {
-				pr_err("Cleanup eviction failed\n");
-			}
-		}
+		if (ret)
+			return ret;
 	}
 
 	return 0;
@@ -1359,7 +1349,11 @@ int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type)
 
 	ret = 0;
 	if (mem_type > 0) {
-		ttm_bo_force_list_clean(bdev, mem_type, false);
+		ret = ttm_bo_force_list_clean(bdev, mem_type);
+		if (ret) {
+			pr_err("Cleanup eviction failed\n");
+			return ret;
+		}
 
 		ret = (*man->func->takedown)(man);
 	}
@@ -1382,7 +1376,7 @@ int ttm_bo_evict_mm(struct ttm_bo_device *bdev, unsigned mem_type)
 		return 0;
 	}
 
-	return ttm_bo_force_list_clean(bdev, mem_type, true);
+	return ttm_bo_force_list_clean(bdev, mem_type);
 }
 EXPORT_SYMBOL(ttm_bo_evict_mm);
 
-- 
2.7.4

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

RE: [PATCH 1/6] drm/ttm: remove allow_errors parameter from ttm_bo_force_list_clean

[He, Hongbo]

Looks good to me, patch is Reviewed-by: Roger.He <Hongbo.He@amd.com>


-----Original Message-----
From: amd-gfx [mailto:amd-gfx-bounces@lists.freedesktop.org] On Behalf Of Christian K?nig
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <Hongbo.He@amd.com>
Cc: dri-devel@lists.freedesktop.org; amd-gfx@lists.freedesktop.org
Subject: [PATCH 1/6] drm/ttm: remove allow_errors parameter from ttm_bo_force_list_clean

From: Christian König <christian.koenig@amd.com>

Not allowing errors here is completely pointless and actually dangerous cause trying to continue on an error can cause an endless loop.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c | 28 +++++++++++-----------------
 1 file changed, 11 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index b5038c5..6683399 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1291,7 +1291,7 @@ int ttm_bo_create(struct ttm_bo_device *bdev,  EXPORT_SYMBOL(ttm_bo_create);
 
 static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
-					unsigned mem_type, bool allow_errors)
+				   unsigned mem_type)
 {
 	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
 	struct ttm_bo_global *glob = bdev->glob; @@ -1306,13 +1306,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 	while (!list_empty(&man->lru)) {
 		spin_unlock(&glob->lru_lock);
 		ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
-		if (ret) {
-			if (allow_errors) {
-				return ret;
-			} else {
-				pr_err("Cleanup eviction failed\n");
-			}
-		}
+		if (ret)
+			return ret;
 		spin_lock(&glob->lru_lock);
 	}
 	spin_unlock(&glob->lru_lock);
@@ -1324,13 +1319,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 	if (fence) {
 		ret = fence_wait(fence, false);
 		fence_put(fence);
-		if (ret) {
-			if (allow_errors) {
-				return ret;
-			} else {
-				pr_err("Cleanup eviction failed\n");
-			}
-		}
+		if (ret)
+			return ret;
 	}
 
 	return 0;
@@ -1359,7 +1349,11 @@ int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type)
 
 	ret = 0;
 	if (mem_type > 0) {
-		ttm_bo_force_list_clean(bdev, mem_type, false);
+		ret = ttm_bo_force_list_clean(bdev, mem_type);
+		if (ret) {
+			pr_err("Cleanup eviction failed\n");
+			return ret;
+		}
 
 		ret = (*man->func->takedown)(man);
 	}
@@ -1382,7 +1376,7 @@ int ttm_bo_evict_mm(struct ttm_bo_device *bdev, unsigned mem_type)
 		return 0;
 	}
 
-	return ttm_bo_force_list_clean(bdev, mem_type, true);
+	return ttm_bo_force_list_clean(bdev, mem_type);
 }
 EXPORT_SYMBOL(ttm_bo_evict_mm);
 
--
2.7.4

_______________________________________________
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

[PATCH 2/6] drm/ttm: add BO priorities for the LRUs

[Christian König]

From: Christian König <christian.koenig@amd.com>

This way the driver can specify a priority for a BO which has the effect that
a BO is only evicted when all other BOs with a lower priority are evicted
first.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  4 +-
 drivers/gpu/drm/ttm/ttm_bo.c            | 67 ++++++++++++++++++++++-----------
 include/drm/ttm/ttm_bo_api.h            |  2 +
 include/drm/ttm/ttm_bo_driver.h         |  6 ++-
 4 files changed, 52 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 0717dd1..0a61930 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1166,8 +1166,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 		struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
 
 		for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-			lru->lru[j] = &adev->mman.bdev.man[j].lru;
-		lru->swap_lru = &adev->mman.bdev.glob->swap_lru;
+			lru->lru[j] = &adev->mman.bdev.man[j].lru[0];
+		lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0];
 	}
 
 	for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 6683399..f078b43 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -242,13 +242,13 @@ EXPORT_SYMBOL(ttm_bo_move_to_lru_tail);
 
 struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo)
 {
-	return bo->bdev->man[bo->mem.mem_type].lru.prev;
+	return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev;
 }
 EXPORT_SYMBOL(ttm_bo_default_lru_tail);
 
 struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo)
 {
-	return bo->glob->swap_lru.prev;
+	return bo->glob->swap_lru[bo->priority].prev;
 }
 EXPORT_SYMBOL(ttm_bo_default_swap_lru_tail);
 
@@ -741,20 +741,27 @@ static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
 	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
 	struct ttm_buffer_object *bo;
 	int ret = -EBUSY, put_count;
+	unsigned i;
 
 	spin_lock(&glob->lru_lock);
-	list_for_each_entry(bo, &man->lru, lru) {
-		ret = __ttm_bo_reserve(bo, false, true, NULL);
-		if (ret)
-			continue;
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+		list_for_each_entry(bo, &man->lru[i], lru) {
+			ret = __ttm_bo_reserve(bo, false, true, NULL);
+			if (ret)
+				continue;
 
-		if (place && !bdev->driver->eviction_valuable(bo, place)) {
-			__ttm_bo_unreserve(bo);
-			ret = -EBUSY;
-			continue;
+			if (place && !bdev->driver->eviction_valuable(bo,
+								      place)) {
+				__ttm_bo_unreserve(bo);
+				ret = -EBUSY;
+				continue;
+			}
+
+			break;
 		}
 
-		break;
+		if (!ret)
+			break;
 	}
 
 	if (ret) {
@@ -1197,6 +1204,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
 	}
 	atomic_inc(&bo->glob->bo_count);
 	drm_vma_node_reset(&bo->vma_node);
+	bo->priority = 0;
 
 	/*
 	 * For ttm_bo_type_device buffers, allocate
@@ -1297,18 +1305,21 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 	struct ttm_bo_global *glob = bdev->glob;
 	struct fence *fence;
 	int ret;
+	unsigned i;
 
 	/*
 	 * Can't use standard list traversal since we're unlocking.
 	 */
 
 	spin_lock(&glob->lru_lock);
-	while (!list_empty(&man->lru)) {
-		spin_unlock(&glob->lru_lock);
-		ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
-		if (ret)
-			return ret;
-		spin_lock(&glob->lru_lock);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+		while (!list_empty(&man->lru[i])) {
+			spin_unlock(&glob->lru_lock);
+			ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
+			if (ret)
+				return ret;
+			spin_lock(&glob->lru_lock);
+		}
 	}
 	spin_unlock(&glob->lru_lock);
 
@@ -1385,6 +1396,7 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
 {
 	int ret = -EINVAL;
 	struct ttm_mem_type_manager *man;
+	unsigned i;
 
 	BUG_ON(type >= TTM_NUM_MEM_TYPES);
 	man = &bdev->man[type];
@@ -1410,7 +1422,8 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
 	man->use_type = true;
 	man->size = p_size;
 
-	INIT_LIST_HEAD(&man->lru);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+		INIT_LIST_HEAD(&man->lru[i]);
 	man->move = NULL;
 
 	return 0;
@@ -1442,6 +1455,7 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
 		container_of(ref, struct ttm_bo_global_ref, ref);
 	struct ttm_bo_global *glob = ref->object;
 	int ret;
+	unsigned i;
 
 	mutex_init(&glob->device_list_mutex);
 	spin_lock_init(&glob->lru_lock);
@@ -1453,7 +1467,8 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
 		goto out_no_drp;
 	}
 
-	INIT_LIST_HEAD(&glob->swap_lru);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+		INIT_LIST_HEAD(&glob->swap_lru[i]);
 	INIT_LIST_HEAD(&glob->device_list);
 
 	ttm_mem_init_shrink(&glob->shrink, ttm_bo_swapout);
@@ -1512,8 +1527,9 @@ int ttm_bo_device_release(struct ttm_bo_device *bdev)
 	if (list_empty(&bdev->ddestroy))
 		TTM_DEBUG("Delayed destroy list was clean\n");
 
-	if (list_empty(&bdev->man[0].lru))
-		TTM_DEBUG("Swap list was clean\n");
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+		if (list_empty(&bdev->man[0].lru[0]))
+			TTM_DEBUG("Swap list %d was clean\n", i);
 	spin_unlock(&glob->lru_lock);
 
 	drm_vma_offset_manager_destroy(&bdev->vma_manager);
@@ -1665,10 +1681,15 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
 	int ret = -EBUSY;
 	int put_count;
 	uint32_t swap_placement = (TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM);
+	unsigned i;
 
 	spin_lock(&glob->lru_lock);
-	list_for_each_entry(bo, &glob->swap_lru, swap) {
-		ret = __ttm_bo_reserve(bo, false, true, NULL);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+		list_for_each_entry(bo, &glob->swap_lru[i], swap) {
+			ret = __ttm_bo_reserve(bo, false, true, NULL);
+			if (!ret)
+				break;
+		}
 		if (!ret)
 			break;
 	}
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index bb6a335..f195899 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -215,6 +215,8 @@ struct ttm_buffer_object {
 
 	struct drm_vma_offset_node vma_node;
 
+	unsigned priority;
+
 	/**
 	 * Special members that are protected by the reserve lock
 	 * and the bo::lock when written to. Can be read with
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 12aa155..90dd5e9 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -42,6 +42,8 @@
 #include <linux/spinlock.h>
 #include <linux/reservation.h>
 
+#define TTM_MAX_BO_PRIORITY	16
+
 struct ttm_backend_func {
 	/**
 	 * struct ttm_backend_func member bind
@@ -298,7 +300,7 @@ struct ttm_mem_type_manager {
 	 * Protected by the global->lru_lock.
 	 */
 
-	struct list_head lru;
+	struct list_head lru[TTM_MAX_BO_PRIORITY];
 
 	/*
 	 * Protected by @move_lock.
@@ -518,7 +520,7 @@ struct ttm_bo_global {
 	/**
 	 * Protected by the lru_lock.
 	 */
-	struct list_head swap_lru;
+	struct list_head swap_lru[TTM_MAX_BO_PRIORITY];
 
 	/**
 	 * Internal protection.
-- 
2.7.4

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

RE: [PATCH 2/6] drm/ttm: add BO priorities for the LRUs

[He, Hongbo]

Reviewed-by: Roger.He <Hongbo.He@amd.com> 

-----Original Message-----
From: amd-gfx [mailto:amd-gfx-bounces@lists.freedesktop.org] On Behalf Of Christian K?nig
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <Hongbo.He@amd.com>
Cc: dri-devel@lists.freedesktop.org; amd-gfx@lists.freedesktop.org
Subject: [PATCH 2/6] drm/ttm: add BO priorities for the LRUs

From: Christian König <christian.koenig@amd.com>

This way the driver can specify a priority for a BO which has the effect that a BO is only evicted when all other BOs with a lower priority are evicted first.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  4 +-
 drivers/gpu/drm/ttm/ttm_bo.c            | 67 ++++++++++++++++++++++-----------
 include/drm/ttm/ttm_bo_api.h            |  2 +
 include/drm/ttm/ttm_bo_driver.h         |  6 ++-
 4 files changed, 52 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 0717dd1..0a61930 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1166,8 +1166,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 		struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
 
 		for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-			lru->lru[j] = &adev->mman.bdev.man[j].lru;
-		lru->swap_lru = &adev->mman.bdev.glob->swap_lru;
+			lru->lru[j] = &adev->mman.bdev.man[j].lru[0];
+		lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0];
 	}
 
 	for (j = 0; j < TTM_NUM_MEM_TYPES; ++j) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 6683399..f078b43 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -242,13 +242,13 @@ EXPORT_SYMBOL(ttm_bo_move_to_lru_tail);
 
 struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo)  {
-	return bo->bdev->man[bo->mem.mem_type].lru.prev;
+	return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev;
 }
 EXPORT_SYMBOL(ttm_bo_default_lru_tail);
 
 struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo)  {
-	return bo->glob->swap_lru.prev;
+	return bo->glob->swap_lru[bo->priority].prev;
 }
 EXPORT_SYMBOL(ttm_bo_default_swap_lru_tail);
 
@@ -741,20 +741,27 @@ static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
 	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
 	struct ttm_buffer_object *bo;
 	int ret = -EBUSY, put_count;
+	unsigned i;
 
 	spin_lock(&glob->lru_lock);
-	list_for_each_entry(bo, &man->lru, lru) {
-		ret = __ttm_bo_reserve(bo, false, true, NULL);
-		if (ret)
-			continue;
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+		list_for_each_entry(bo, &man->lru[i], lru) {
+			ret = __ttm_bo_reserve(bo, false, true, NULL);
+			if (ret)
+				continue;
 
-		if (place && !bdev->driver->eviction_valuable(bo, place)) {
-			__ttm_bo_unreserve(bo);
-			ret = -EBUSY;
-			continue;
+			if (place && !bdev->driver->eviction_valuable(bo,
+								      place)) {
+				__ttm_bo_unreserve(bo);
+				ret = -EBUSY;
+				continue;
+			}
+
+			break;
 		}
 
-		break;
+		if (!ret)
+			break;
 	}
 
 	if (ret) {
@@ -1197,6 +1204,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
 	}
 	atomic_inc(&bo->glob->bo_count);
 	drm_vma_node_reset(&bo->vma_node);
+	bo->priority = 0;
 
 	/*
 	 * For ttm_bo_type_device buffers, allocate @@ -1297,18 +1305,21 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 	struct ttm_bo_global *glob = bdev->glob;
 	struct fence *fence;
 	int ret;
+	unsigned i;
 
 	/*
 	 * Can't use standard list traversal since we're unlocking.
 	 */
 
 	spin_lock(&glob->lru_lock);
-	while (!list_empty(&man->lru)) {
-		spin_unlock(&glob->lru_lock);
-		ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
-		if (ret)
-			return ret;
-		spin_lock(&glob->lru_lock);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+		while (!list_empty(&man->lru[i])) {
+			spin_unlock(&glob->lru_lock);
+			ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
+			if (ret)
+				return ret;
+			spin_lock(&glob->lru_lock);
+		}
 	}
 	spin_unlock(&glob->lru_lock);
 
@@ -1385,6 +1396,7 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,  {
 	int ret = -EINVAL;
 	struct ttm_mem_type_manager *man;
+	unsigned i;
 
 	BUG_ON(type >= TTM_NUM_MEM_TYPES);
 	man = &bdev->man[type];
@@ -1410,7 +1422,8 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
 	man->use_type = true;
 	man->size = p_size;
 
-	INIT_LIST_HEAD(&man->lru);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+		INIT_LIST_HEAD(&man->lru[i]);
 	man->move = NULL;
 
 	return 0;
@@ -1442,6 +1455,7 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
 		container_of(ref, struct ttm_bo_global_ref, ref);
 	struct ttm_bo_global *glob = ref->object;
 	int ret;
+	unsigned i;
 
 	mutex_init(&glob->device_list_mutex);
 	spin_lock_init(&glob->lru_lock);
@@ -1453,7 +1467,8 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
 		goto out_no_drp;
 	}
 
-	INIT_LIST_HEAD(&glob->swap_lru);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+		INIT_LIST_HEAD(&glob->swap_lru[i]);
 	INIT_LIST_HEAD(&glob->device_list);
 
 	ttm_mem_init_shrink(&glob->shrink, ttm_bo_swapout); @@ -1512,8 +1527,9 @@ int ttm_bo_device_release(struct ttm_bo_device *bdev)
 	if (list_empty(&bdev->ddestroy))
 		TTM_DEBUG("Delayed destroy list was clean\n");
 
-	if (list_empty(&bdev->man[0].lru))
-		TTM_DEBUG("Swap list was clean\n");
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+		if (list_empty(&bdev->man[0].lru[0]))
+			TTM_DEBUG("Swap list %d was clean\n", i);
 	spin_unlock(&glob->lru_lock);
 
 	drm_vma_offset_manager_destroy(&bdev->vma_manager);
@@ -1665,10 +1681,15 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
 	int ret = -EBUSY;
 	int put_count;
 	uint32_t swap_placement = (TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM);
+	unsigned i;
 
 	spin_lock(&glob->lru_lock);
-	list_for_each_entry(bo, &glob->swap_lru, swap) {
-		ret = __ttm_bo_reserve(bo, false, true, NULL);
+	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+		list_for_each_entry(bo, &glob->swap_lru[i], swap) {
+			ret = __ttm_bo_reserve(bo, false, true, NULL);
+			if (!ret)
+				break;
+		}
 		if (!ret)
 			break;
 	}
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index bb6a335..f195899 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -215,6 +215,8 @@ struct ttm_buffer_object {
 
 	struct drm_vma_offset_node vma_node;
 
+	unsigned priority;
+
 	/**
 	 * Special members that are protected by the reserve lock
 	 * and the bo::lock when written to. Can be read with diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index 12aa155..90dd5e9 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -42,6 +42,8 @@
 #include <linux/spinlock.h>
 #include <linux/reservation.h>
 
+#define TTM_MAX_BO_PRIORITY	16
+
 struct ttm_backend_func {
 	/**
 	 * struct ttm_backend_func member bind @@ -298,7 +300,7 @@ struct ttm_mem_type_manager {
 	 * Protected by the global->lru_lock.
 	 */
 
-	struct list_head lru;
+	struct list_head lru[TTM_MAX_BO_PRIORITY];
 
 	/*
 	 * Protected by @move_lock.
@@ -518,7 +520,7 @@ struct ttm_bo_global {
 	/**
 	 * Protected by the lru_lock.
 	 */
-	struct list_head swap_lru;
+	struct list_head swap_lru[TTM_MAX_BO_PRIORITY];
 
 	/**
 	 * Internal protection.
--
2.7.4

_______________________________________________
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

[PATCH 5/6] drm/ttm: revert "implement LRU add callbacks v2"

[Christian König]

From: Christian König <christian.koenig@amd.com>

The additional housekeeping had to much CPU overhead,
let's use the BO priorities instead.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  2 --
 drivers/gpu/drm/ast/ast_ttm.c           |  2 --
 drivers/gpu/drm/bochs/bochs_mm.c        |  2 --
 drivers/gpu/drm/cirrus/cirrus_ttm.c     |  2 --
 drivers/gpu/drm/mgag200/mgag200_ttm.c   |  2 --
 drivers/gpu/drm/nouveau/nouveau_bo.c    |  2 --
 drivers/gpu/drm/qxl/qxl_ttm.c           |  2 --
 drivers/gpu/drm/radeon/radeon_ttm.c     |  2 --
 drivers/gpu/drm/ttm/ttm_bo.c            | 19 +++++--------------
 drivers/gpu/drm/virtio/virtgpu_ttm.c    |  2 --
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c  |  2 --
 include/drm/ttm/ttm_bo_driver.h         |  9 ---------
 12 files changed, 5 insertions(+), 43 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index dd086d8..0402a11 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1086,8 +1086,6 @@ static struct ttm_bo_driver amdgpu_bo_driver = {
 	.fault_reserve_notify = &amdgpu_bo_fault_reserve_notify,
 	.io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
 	.io_mem_free = &amdgpu_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int amdgpu_ttm_init(struct amdgpu_device *adev)
diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c
index 2a1368f..50c910e 100644
--- a/drivers/gpu/drm/ast/ast_ttm.c
+++ b/drivers/gpu/drm/ast/ast_ttm.c
@@ -236,8 +236,6 @@ struct ttm_bo_driver ast_bo_driver = {
 	.verify_access = ast_bo_verify_access,
 	.io_mem_reserve = &ast_ttm_io_mem_reserve,
 	.io_mem_free = &ast_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int ast_mm_init(struct ast_private *ast)
diff --git a/drivers/gpu/drm/bochs/bochs_mm.c b/drivers/gpu/drm/bochs/bochs_mm.c
index 099a3c6..e4c1125 100644
--- a/drivers/gpu/drm/bochs/bochs_mm.c
+++ b/drivers/gpu/drm/bochs/bochs_mm.c
@@ -205,8 +205,6 @@ struct ttm_bo_driver bochs_bo_driver = {
 	.verify_access = bochs_bo_verify_access,
 	.io_mem_reserve = &bochs_ttm_io_mem_reserve,
 	.io_mem_free = &bochs_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int bochs_mm_init(struct bochs_device *bochs)
diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c
index d6da848..f53aa8f 100644
--- a/drivers/gpu/drm/cirrus/cirrus_ttm.c
+++ b/drivers/gpu/drm/cirrus/cirrus_ttm.c
@@ -236,8 +236,6 @@ struct ttm_bo_driver cirrus_bo_driver = {
 	.verify_access = cirrus_bo_verify_access,
 	.io_mem_reserve = &cirrus_ttm_io_mem_reserve,
 	.io_mem_free = &cirrus_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int cirrus_mm_init(struct cirrus_device *cirrus)
diff --git a/drivers/gpu/drm/mgag200/mgag200_ttm.c b/drivers/gpu/drm/mgag200/mgag200_ttm.c
index 5e20220..657598b 100644
--- a/drivers/gpu/drm/mgag200/mgag200_ttm.c
+++ b/drivers/gpu/drm/mgag200/mgag200_ttm.c
@@ -236,8 +236,6 @@ struct ttm_bo_driver mgag200_bo_driver = {
 	.verify_access = mgag200_bo_verify_access,
 	.io_mem_reserve = &mgag200_ttm_io_mem_reserve,
 	.io_mem_free = &mgag200_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int mgag200_mm_init(struct mga_device *mdev)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 5b0a28b..3949a74 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1570,8 +1570,6 @@ struct ttm_bo_driver nouveau_bo_driver = {
 	.fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
 	.io_mem_reserve = &nouveau_ttm_io_mem_reserve,
 	.io_mem_free = &nouveau_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 struct nvkm_vma *
diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
index f3939a9..2955f91 100644
--- a/drivers/gpu/drm/qxl/qxl_ttm.c
+++ b/drivers/gpu/drm/qxl/qxl_ttm.c
@@ -395,8 +395,6 @@ static struct ttm_bo_driver qxl_bo_driver = {
 	.io_mem_reserve = &qxl_ttm_io_mem_reserve,
 	.io_mem_free = &qxl_ttm_io_mem_free,
 	.move_notify = &qxl_bo_move_notify,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int qxl_ttm_init(struct qxl_device *qdev)
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 0cf03cc..d610481 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -871,8 +871,6 @@ static struct ttm_bo_driver radeon_bo_driver = {
 	.fault_reserve_notify = &radeon_bo_fault_reserve_notify,
 	.io_mem_reserve = &radeon_ttm_io_mem_reserve,
 	.io_mem_free = &radeon_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int radeon_ttm_init(struct radeon_device *rdev)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index f078b43..59fac2f 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -163,6 +163,7 @@ static void ttm_bo_release_list(struct kref *list_kref)
 void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
+	struct ttm_mem_type_manager *man;
 
 	lockdep_assert_held(&bo->resv->lock.base);
 
@@ -170,11 +171,13 @@ void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
 
 		BUG_ON(!list_empty(&bo->lru));
 
-		list_add(&bo->lru, bdev->driver->lru_tail(bo));
+		man = &bdev->man[bo->mem.mem_type];
+		list_add_tail(&bo->lru, &man->lru[bo->priority]);
 		kref_get(&bo->list_kref);
 
 		if (bo->ttm && !(bo->ttm->page_flags & TTM_PAGE_FLAG_SG)) {
-			list_add(&bo->swap, bdev->driver->swap_lru_tail(bo));
+			list_add_tail(&bo->swap,
+				      &bo->glob->swap_lru[bo->priority]);
 			kref_get(&bo->list_kref);
 		}
 	}
@@ -240,18 +243,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)
 }
 EXPORT_SYMBOL(ttm_bo_move_to_lru_tail);
 
-struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo)
-{
-	return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev;
-}
-EXPORT_SYMBOL(ttm_bo_default_lru_tail);
-
-struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo)
-{
-	return bo->glob->swap_lru[bo->priority].prev;
-}
-EXPORT_SYMBOL(ttm_bo_default_swap_lru_tail);
-
 /*
  * Call bo->mutex locked.
  */
diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
index 10387d7..a8875a7 100644
--- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
+++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
@@ -434,8 +434,6 @@ static struct ttm_bo_driver virtio_gpu_bo_driver = {
 	.io_mem_free = &virtio_gpu_ttm_io_mem_free,
 	.move_notify = &virtio_gpu_bo_move_notify,
 	.swap_notify = &virtio_gpu_bo_swap_notify,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int virtio_gpu_ttm_init(struct virtio_gpu_device *vgdev)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index caa279b..1de9669 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -859,6 +859,4 @@ struct ttm_bo_driver vmw_bo_driver = {
 	.fault_reserve_notify = &vmw_ttm_fault_reserve_notify,
 	.io_mem_reserve = &vmw_ttm_io_mem_reserve,
 	.io_mem_free = &vmw_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 90dd5e9..a1235fa 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -468,12 +468,6 @@ struct ttm_bo_driver {
 	 * Called with LRU lock held immediately before the removal.
 	 */
 	void (*lru_removal)(struct ttm_buffer_object *bo);
-
-	/**
-	 * Return the list_head after which a BO should be inserted in the LRU.
-	 */
-	struct list_head *(*lru_tail)(struct ttm_buffer_object *bo);
-	struct list_head *(*swap_lru_tail)(struct ttm_buffer_object *bo);
 };
 
 /**
@@ -788,9 +782,6 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);
 extern void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo);
 extern void ttm_bo_add_to_lru(struct ttm_buffer_object *bo);
 
-struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo);
-struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo);
-
 /**
  * __ttm_bo_reserve:
  *
-- 
2.7.4

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

RE: [PATCH 5/6] drm/ttm: revert "implement LRU add callbacks v2"

[He, Hongbo]

Series patches:
 Reviewed-by: Roger.He <Hongbo.He@amd.com> 
and Test-by Roger.He

-----Original Message-----
From: Christian König [mailto:deathsimple@vodafone.de] 
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <Hongbo.He@amd.com>
Cc: amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Subject: [PATCH 5/6] drm/ttm: revert "implement LRU add callbacks v2"

From: Christian König <christian.koenig@amd.com>

The additional housekeeping had to much CPU overhead, let's use the BO priorities instead.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  2 --
 drivers/gpu/drm/ast/ast_ttm.c           |  2 --
 drivers/gpu/drm/bochs/bochs_mm.c        |  2 --
 drivers/gpu/drm/cirrus/cirrus_ttm.c     |  2 --
 drivers/gpu/drm/mgag200/mgag200_ttm.c   |  2 --
 drivers/gpu/drm/nouveau/nouveau_bo.c    |  2 --
 drivers/gpu/drm/qxl/qxl_ttm.c           |  2 --
 drivers/gpu/drm/radeon/radeon_ttm.c     |  2 --
 drivers/gpu/drm/ttm/ttm_bo.c            | 19 +++++--------------
 drivers/gpu/drm/virtio/virtgpu_ttm.c    |  2 --
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c  |  2 --
 include/drm/ttm/ttm_bo_driver.h         |  9 ---------
 12 files changed, 5 insertions(+), 43 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index dd086d8..0402a11 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1086,8 +1086,6 @@ static struct ttm_bo_driver amdgpu_bo_driver = {
 	.fault_reserve_notify = &amdgpu_bo_fault_reserve_notify,
 	.io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
 	.io_mem_free = &amdgpu_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int amdgpu_ttm_init(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c index 2a1368f..50c910e 100644
--- a/drivers/gpu/drm/ast/ast_ttm.c
+++ b/drivers/gpu/drm/ast/ast_ttm.c
@@ -236,8 +236,6 @@ struct ttm_bo_driver ast_bo_driver = {
 	.verify_access = ast_bo_verify_access,
 	.io_mem_reserve = &ast_ttm_io_mem_reserve,
 	.io_mem_free = &ast_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int ast_mm_init(struct ast_private *ast) diff --git a/drivers/gpu/drm/bochs/bochs_mm.c b/drivers/gpu/drm/bochs/bochs_mm.c
index 099a3c6..e4c1125 100644
--- a/drivers/gpu/drm/bochs/bochs_mm.c
+++ b/drivers/gpu/drm/bochs/bochs_mm.c
@@ -205,8 +205,6 @@ struct ttm_bo_driver bochs_bo_driver = {
 	.verify_access = bochs_bo_verify_access,
 	.io_mem_reserve = &bochs_ttm_io_mem_reserve,
 	.io_mem_free = &bochs_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int bochs_mm_init(struct bochs_device *bochs) diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c
index d6da848..f53aa8f 100644
--- a/drivers/gpu/drm/cirrus/cirrus_ttm.c
+++ b/drivers/gpu/drm/cirrus/cirrus_ttm.c
@@ -236,8 +236,6 @@ struct ttm_bo_driver cirrus_bo_driver = {
 	.verify_access = cirrus_bo_verify_access,
 	.io_mem_reserve = &cirrus_ttm_io_mem_reserve,
 	.io_mem_free = &cirrus_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int cirrus_mm_init(struct cirrus_device *cirrus) diff --git a/drivers/gpu/drm/mgag200/mgag200_ttm.c b/drivers/gpu/drm/mgag200/mgag200_ttm.c
index 5e20220..657598b 100644
--- a/drivers/gpu/drm/mgag200/mgag200_ttm.c
+++ b/drivers/gpu/drm/mgag200/mgag200_ttm.c
@@ -236,8 +236,6 @@ struct ttm_bo_driver mgag200_bo_driver = {
 	.verify_access = mgag200_bo_verify_access,
 	.io_mem_reserve = &mgag200_ttm_io_mem_reserve,
 	.io_mem_free = &mgag200_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int mgag200_mm_init(struct mga_device *mdev) diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 5b0a28b..3949a74 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1570,8 +1570,6 @@ struct ttm_bo_driver nouveau_bo_driver = {
 	.fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
 	.io_mem_reserve = &nouveau_ttm_io_mem_reserve,
 	.io_mem_free = &nouveau_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 struct nvkm_vma *
diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c index f3939a9..2955f91 100644
--- a/drivers/gpu/drm/qxl/qxl_ttm.c
+++ b/drivers/gpu/drm/qxl/qxl_ttm.c
@@ -395,8 +395,6 @@ static struct ttm_bo_driver qxl_bo_driver = {
 	.io_mem_reserve = &qxl_ttm_io_mem_reserve,
 	.io_mem_free = &qxl_ttm_io_mem_free,
 	.move_notify = &qxl_bo_move_notify,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int qxl_ttm_init(struct qxl_device *qdev) diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 0cf03cc..d610481 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -871,8 +871,6 @@ static struct ttm_bo_driver radeon_bo_driver = {
 	.fault_reserve_notify = &radeon_bo_fault_reserve_notify,
 	.io_mem_reserve = &radeon_ttm_io_mem_reserve,
 	.io_mem_free = &radeon_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int radeon_ttm_init(struct radeon_device *rdev) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index f078b43..59fac2f 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -163,6 +163,7 @@ static void ttm_bo_release_list(struct kref *list_kref)  void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)  {
 	struct ttm_bo_device *bdev = bo->bdev;
+	struct ttm_mem_type_manager *man;
 
 	lockdep_assert_held(&bo->resv->lock.base);
 
@@ -170,11 +171,13 @@ void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
 
 		BUG_ON(!list_empty(&bo->lru));
 
-		list_add(&bo->lru, bdev->driver->lru_tail(bo));
+		man = &bdev->man[bo->mem.mem_type];
+		list_add_tail(&bo->lru, &man->lru[bo->priority]);
 		kref_get(&bo->list_kref);
 
 		if (bo->ttm && !(bo->ttm->page_flags & TTM_PAGE_FLAG_SG)) {
-			list_add(&bo->swap, bdev->driver->swap_lru_tail(bo));
+			list_add_tail(&bo->swap,
+				      &bo->glob->swap_lru[bo->priority]);
 			kref_get(&bo->list_kref);
 		}
 	}
@@ -240,18 +243,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)  }  EXPORT_SYMBOL(ttm_bo_move_to_lru_tail);
 
-struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo) -{
-	return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev;
-}
-EXPORT_SYMBOL(ttm_bo_default_lru_tail);
-
-struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo) -{
-	return bo->glob->swap_lru[bo->priority].prev;
-}
-EXPORT_SYMBOL(ttm_bo_default_swap_lru_tail);
-
 /*
  * Call bo->mutex locked.
  */
diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
index 10387d7..a8875a7 100644
--- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
+++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
@@ -434,8 +434,6 @@ static struct ttm_bo_driver virtio_gpu_bo_driver = {
 	.io_mem_free = &virtio_gpu_ttm_io_mem_free,
 	.move_notify = &virtio_gpu_bo_move_notify,
 	.swap_notify = &virtio_gpu_bo_swap_notify,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int virtio_gpu_ttm_init(struct virtio_gpu_device *vgdev) diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index caa279b..1de9669 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -859,6 +859,4 @@ struct ttm_bo_driver vmw_bo_driver = {
 	.fault_reserve_notify = &vmw_ttm_fault_reserve_notify,
 	.io_mem_reserve = &vmw_ttm_io_mem_reserve,
 	.io_mem_free = &vmw_ttm_io_mem_free,
-	.lru_tail = &ttm_bo_default_lru_tail,
-	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index 90dd5e9..a1235fa 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -468,12 +468,6 @@ struct ttm_bo_driver {
 	 * Called with LRU lock held immediately before the removal.
 	 */
 	void (*lru_removal)(struct ttm_buffer_object *bo);
-
-	/**
-	 * Return the list_head after which a BO should be inserted in the LRU.
-	 */
-	struct list_head *(*lru_tail)(struct ttm_buffer_object *bo);
-	struct list_head *(*swap_lru_tail)(struct ttm_buffer_object *bo);
 };
 
 /**
@@ -788,9 +782,6 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);  extern void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo);  extern void ttm_bo_add_to_lru(struct ttm_buffer_object *bo);
 
-struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo); -struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo);
-
 /**
  * __ttm_bo_reserve:
  *
--
2.7.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 5/6] drm/ttm: revert "implement LRU add callbacks v2"

[Sinclair Yeh]

Minor typo

On Fri, Jan 13, 2017 at 10:51:10AM +0100, Christian König wrote:
> From: Christian König <christian.koenig@amd.com>
> 
> The additional housekeeping had to much CPU overhead,
                                   ^


> let's use the BO priorities instead.
> 
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  2 --
>  drivers/gpu/drm/ast/ast_ttm.c           |  2 --
>  drivers/gpu/drm/bochs/bochs_mm.c        |  2 --
>  drivers/gpu/drm/cirrus/cirrus_ttm.c     |  2 --
>  drivers/gpu/drm/mgag200/mgag200_ttm.c   |  2 --
>  drivers/gpu/drm/nouveau/nouveau_bo.c    |  2 --
>  drivers/gpu/drm/qxl/qxl_ttm.c           |  2 --
>  drivers/gpu/drm/radeon/radeon_ttm.c     |  2 --
>  drivers/gpu/drm/ttm/ttm_bo.c            | 19 +++++--------------
>  drivers/gpu/drm/virtio/virtgpu_ttm.c    |  2 --
>  drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c  |  2 --
>  include/drm/ttm/ttm_bo_driver.h         |  9 ---------
>  12 files changed, 5 insertions(+), 43 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index dd086d8..0402a11 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -1086,8 +1086,6 @@ static struct ttm_bo_driver amdgpu_bo_driver = {
>  	.fault_reserve_notify = &amdgpu_bo_fault_reserve_notify,
>  	.io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
>  	.io_mem_free = &amdgpu_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int amdgpu_ttm_init(struct amdgpu_device *adev)
> diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c
> index 2a1368f..50c910e 100644
> --- a/drivers/gpu/drm/ast/ast_ttm.c
> +++ b/drivers/gpu/drm/ast/ast_ttm.c
> @@ -236,8 +236,6 @@ struct ttm_bo_driver ast_bo_driver = {
>  	.verify_access = ast_bo_verify_access,
>  	.io_mem_reserve = &ast_ttm_io_mem_reserve,
>  	.io_mem_free = &ast_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int ast_mm_init(struct ast_private *ast)
> diff --git a/drivers/gpu/drm/bochs/bochs_mm.c b/drivers/gpu/drm/bochs/bochs_mm.c
> index 099a3c6..e4c1125 100644
> --- a/drivers/gpu/drm/bochs/bochs_mm.c
> +++ b/drivers/gpu/drm/bochs/bochs_mm.c
> @@ -205,8 +205,6 @@ struct ttm_bo_driver bochs_bo_driver = {
>  	.verify_access = bochs_bo_verify_access,
>  	.io_mem_reserve = &bochs_ttm_io_mem_reserve,
>  	.io_mem_free = &bochs_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int bochs_mm_init(struct bochs_device *bochs)
> diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c
> index d6da848..f53aa8f 100644
> --- a/drivers/gpu/drm/cirrus/cirrus_ttm.c
> +++ b/drivers/gpu/drm/cirrus/cirrus_ttm.c
> @@ -236,8 +236,6 @@ struct ttm_bo_driver cirrus_bo_driver = {
>  	.verify_access = cirrus_bo_verify_access,
>  	.io_mem_reserve = &cirrus_ttm_io_mem_reserve,
>  	.io_mem_free = &cirrus_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int cirrus_mm_init(struct cirrus_device *cirrus)
> diff --git a/drivers/gpu/drm/mgag200/mgag200_ttm.c b/drivers/gpu/drm/mgag200/mgag200_ttm.c
> index 5e20220..657598b 100644
> --- a/drivers/gpu/drm/mgag200/mgag200_ttm.c
> +++ b/drivers/gpu/drm/mgag200/mgag200_ttm.c
> @@ -236,8 +236,6 @@ struct ttm_bo_driver mgag200_bo_driver = {
>  	.verify_access = mgag200_bo_verify_access,
>  	.io_mem_reserve = &mgag200_ttm_io_mem_reserve,
>  	.io_mem_free = &mgag200_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int mgag200_mm_init(struct mga_device *mdev)
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index 5b0a28b..3949a74 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1570,8 +1570,6 @@ struct ttm_bo_driver nouveau_bo_driver = {
>  	.fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
>  	.io_mem_reserve = &nouveau_ttm_io_mem_reserve,
>  	.io_mem_free = &nouveau_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  struct nvkm_vma *
> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
> index f3939a9..2955f91 100644
> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
> @@ -395,8 +395,6 @@ static struct ttm_bo_driver qxl_bo_driver = {
>  	.io_mem_reserve = &qxl_ttm_io_mem_reserve,
>  	.io_mem_free = &qxl_ttm_io_mem_free,
>  	.move_notify = &qxl_bo_move_notify,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int qxl_ttm_init(struct qxl_device *qdev)
> diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
> index 0cf03cc..d610481 100644
> --- a/drivers/gpu/drm/radeon/radeon_ttm.c
> +++ b/drivers/gpu/drm/radeon/radeon_ttm.c
> @@ -871,8 +871,6 @@ static struct ttm_bo_driver radeon_bo_driver = {
>  	.fault_reserve_notify = &radeon_bo_fault_reserve_notify,
>  	.io_mem_reserve = &radeon_ttm_io_mem_reserve,
>  	.io_mem_free = &radeon_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int radeon_ttm_init(struct radeon_device *rdev)
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index f078b43..59fac2f 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -163,6 +163,7 @@ static void ttm_bo_release_list(struct kref *list_kref)
>  void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
>  {
>  	struct ttm_bo_device *bdev = bo->bdev;
> +	struct ttm_mem_type_manager *man;
>  
>  	lockdep_assert_held(&bo->resv->lock.base);
>  
> @@ -170,11 +171,13 @@ void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
>  
>  		BUG_ON(!list_empty(&bo->lru));
>  
> -		list_add(&bo->lru, bdev->driver->lru_tail(bo));
> +		man = &bdev->man[bo->mem.mem_type];
> +		list_add_tail(&bo->lru, &man->lru[bo->priority]);
>  		kref_get(&bo->list_kref);
>  
>  		if (bo->ttm && !(bo->ttm->page_flags & TTM_PAGE_FLAG_SG)) {
> -			list_add(&bo->swap, bdev->driver->swap_lru_tail(bo));
> +			list_add_tail(&bo->swap,
> +				      &bo->glob->swap_lru[bo->priority]);
>  			kref_get(&bo->list_kref);
>  		}
>  	}
> @@ -240,18 +243,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)
>  }
>  EXPORT_SYMBOL(ttm_bo_move_to_lru_tail);
>  
> -struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo)
> -{
> -	return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev;
> -}
> -EXPORT_SYMBOL(ttm_bo_default_lru_tail);
> -
> -struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo)
> -{
> -	return bo->glob->swap_lru[bo->priority].prev;
> -}
> -EXPORT_SYMBOL(ttm_bo_default_swap_lru_tail);
> -
>  /*
>   * Call bo->mutex locked.
>   */
> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> index 10387d7..a8875a7 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> @@ -434,8 +434,6 @@ static struct ttm_bo_driver virtio_gpu_bo_driver = {
>  	.io_mem_free = &virtio_gpu_ttm_io_mem_free,
>  	.move_notify = &virtio_gpu_bo_move_notify,
>  	.swap_notify = &virtio_gpu_bo_swap_notify,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
>  
>  int virtio_gpu_ttm_init(struct virtio_gpu_device *vgdev)
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index caa279b..1de9669 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -859,6 +859,4 @@ struct ttm_bo_driver vmw_bo_driver = {
>  	.fault_reserve_notify = &vmw_ttm_fault_reserve_notify,
>  	.io_mem_reserve = &vmw_ttm_io_mem_reserve,
>  	.io_mem_free = &vmw_ttm_io_mem_free,
> -	.lru_tail = &ttm_bo_default_lru_tail,
> -	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
>  };
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index 90dd5e9..a1235fa 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -468,12 +468,6 @@ struct ttm_bo_driver {
>  	 * Called with LRU lock held immediately before the removal.
>  	 */
>  	void (*lru_removal)(struct ttm_buffer_object *bo);
> -
> -	/**
> -	 * Return the list_head after which a BO should be inserted in the LRU.
> -	 */
> -	struct list_head *(*lru_tail)(struct ttm_buffer_object *bo);
> -	struct list_head *(*swap_lru_tail)(struct ttm_buffer_object *bo);
>  };
>  
>  /**
> @@ -788,9 +782,6 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);
>  extern void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo);
>  extern void ttm_bo_add_to_lru(struct ttm_buffer_object *bo);
>  
> -struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo);
> -struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo);
> -
>  /**
>   * __ttm_bo_reserve:
>   *
> -- 
> 2.7.4
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 6/6] drm/ttm: revert "add optional LRU removal callback v2"

[Christian König]

From: Christian König <christian.koenig@amd.com>

Without the custom LRU management the callback is not used any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 12 +++++-------
 include/drm/ttm/ttm_bo_driver.h |  6 ------
 2 files changed, 5 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 59fac2f..dfaeac4 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -186,12 +186,8 @@ EXPORT_SYMBOL(ttm_bo_add_to_lru);
 
 int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
 {
-	struct ttm_bo_device *bdev = bo->bdev;
 	int put_count = 0;
 
-	if (bdev->driver->lru_removal)
-		bdev->driver->lru_removal(bo);
-
 	if (!list_empty(&bo->swap)) {
 		list_del_init(&bo->swap);
 		++put_count;
@@ -201,6 +197,11 @@ int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
 		++put_count;
 	}
 
+	/*
+	 * TODO: Add a driver hook to delete from
+	 * driver-specific LRU's here.
+	 */
+
 	return put_count;
 }
 
@@ -234,9 +235,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)
 
 	lockdep_assert_held(&bo->resv->lock.base);
 
-	if (bdev->driver->lru_removal)
-		bdev->driver->lru_removal(bo);
-
 	put_count = ttm_bo_del_from_lru(bo);
 	ttm_bo_list_ref_sub(bo, put_count, true);
 	ttm_bo_add_to_lru(bo);
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index a1235fa..4395db1 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -462,12 +462,6 @@ struct ttm_bo_driver {
 			      struct ttm_mem_reg *mem);
 	void (*io_mem_free)(struct ttm_bo_device *bdev,
 			    struct ttm_mem_reg *mem);
-
-	/**
-	 * Optional driver callback for when BO is removed from the LRU.
-	 * Called with LRU lock held immediately before the removal.
-	 */
-	void (*lru_removal)(struct ttm_buffer_object *bo);
 };
 
 /**
-- 
2.7.4

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

RE: [PATCH 6/6] drm/ttm: revert "add optional LRU removal callback v2"

[He, Hongbo]

Good way to keep original idea and remove CPU overhead.
Reviewed-by: Roger.He <Hongbo.He@amd.com> and Test-by Roger.He <Hongbo.He@amd.com>

-----Original Message-----
From: Christian König [mailto:deathsimple@vodafone.de] 
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <Hongbo.He@amd.com>
Cc: amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Subject: [PATCH 6/6] drm/ttm: revert "add optional LRU removal callback v2"

From: Christian König <christian.koenig@amd.com>

Without the custom LRU management the callback is not used any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 12 +++++-------
 include/drm/ttm/ttm_bo_driver.h |  6 ------
 2 files changed, 5 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 59fac2f..dfaeac4 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -186,12 +186,8 @@ EXPORT_SYMBOL(ttm_bo_add_to_lru);
 
 int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)  {
-	struct ttm_bo_device *bdev = bo->bdev;
 	int put_count = 0;
 
-	if (bdev->driver->lru_removal)
-		bdev->driver->lru_removal(bo);
-
 	if (!list_empty(&bo->swap)) {
 		list_del_init(&bo->swap);
 		++put_count;
@@ -201,6 +197,11 @@ int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
 		++put_count;
 	}
 
+	/*
+	 * TODO: Add a driver hook to delete from
+	 * driver-specific LRU's here.
+	 */
+
 	return put_count;
 }
 
@@ -234,9 +235,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)
 
 	lockdep_assert_held(&bo->resv->lock.base);
 
-	if (bdev->driver->lru_removal)
-		bdev->driver->lru_removal(bo);
-
 	put_count = ttm_bo_del_from_lru(bo);
 	ttm_bo_list_ref_sub(bo, put_count, true);
 	ttm_bo_add_to_lru(bo);
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index a1235fa..4395db1 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -462,12 +462,6 @@ struct ttm_bo_driver {
 			      struct ttm_mem_reg *mem);
 	void (*io_mem_free)(struct ttm_bo_device *bdev,
 			    struct ttm_mem_reg *mem);
-
-	/**
-	 * Optional driver callback for when BO is removed from the LRU.
-	 * Called with LRU lock held immediately before the removal.
-	 */
-	void (*lru_removal)(struct ttm_buffer_object *bo);
 };
 
 /**
--
2.7.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/6] drm/ttm: revert "add optional LRU removal callback v2"

[Sinclair Yeh]

Pathces 1, 2, 5, 6:
Reviewed-by: Sinclair Yeh <syeh@vmware.com>

On Fri, Jan 13, 2017 at 10:51:11AM +0100, Christian König wrote:
> From: Christian König <christian.koenig@amd.com>
> 
> Without the custom LRU management the callback is not used any more.
> 
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
>  drivers/gpu/drm/ttm/ttm_bo.c    | 12 +++++-------
>  include/drm/ttm/ttm_bo_driver.h |  6 ------
>  2 files changed, 5 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 59fac2f..dfaeac4 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -186,12 +186,8 @@ EXPORT_SYMBOL(ttm_bo_add_to_lru);
>  
>  int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
>  {
> -	struct ttm_bo_device *bdev = bo->bdev;
>  	int put_count = 0;
>  
> -	if (bdev->driver->lru_removal)
> -		bdev->driver->lru_removal(bo);
> -
>  	if (!list_empty(&bo->swap)) {
>  		list_del_init(&bo->swap);
>  		++put_count;
> @@ -201,6 +197,11 @@ int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
>  		++put_count;
>  	}
>  
> +	/*
> +	 * TODO: Add a driver hook to delete from
> +	 * driver-specific LRU's here.
> +	 */
> +
>  	return put_count;
>  }
>  
> @@ -234,9 +235,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)
>  
>  	lockdep_assert_held(&bo->resv->lock.base);
>  
> -	if (bdev->driver->lru_removal)
> -		bdev->driver->lru_removal(bo);
> -
>  	put_count = ttm_bo_del_from_lru(bo);
>  	ttm_bo_list_ref_sub(bo, put_count, true);
>  	ttm_bo_add_to_lru(bo);
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index a1235fa..4395db1 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -462,12 +462,6 @@ struct ttm_bo_driver {
>  			      struct ttm_mem_reg *mem);
>  	void (*io_mem_free)(struct ttm_bo_device *bdev,
>  			    struct ttm_mem_reg *mem);
> -
> -	/**
> -	 * Optional driver callback for when BO is removed from the LRU.
> -	 * Called with LRU lock held immediately before the removal.
> -	 */
> -	void (*lru_removal)(struct ttm_buffer_object *bo);
>  };
>  
>  /**
> -- 
> 2.7.4
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: amdgpu: fix performance drop

[Christian König]

Hi Sinclair,

you probably have missed this set of patches I've send to the dri-devel 
mailing list a while ago.

It basically replaces some work for TTM I've did back in April last year 
and you reviewed with a different implementation.

The original callback design turned out to have to much overhead which 
made this change necessary. Could you take a look at the new 
implementation as well?

Thanks in advance,
Christian.
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/6] drm/amdgpu: user BO priority instead of self coding it

[Christian König]

From: Christian König <christian.koenig@amd.com>

Keeping groups of BOs on the LRU is to time consuming on command submission.

Instead use the newly added BO priority to give a certain eviction order.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |  3 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 69 +-----------------------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    | 12 ------
 3 files changed, 5 insertions(+), 79 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 5076bc2..f399d98 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -386,6 +386,9 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 	if (unlikely(r != 0))
 		return r;
 
+	bo->tbo.priority = ilog2(bo->tbo.num_pages);
+	bo->tbo.priority = min(bo->tbo.priority, TTM_MAX_BO_PRIORITY - 1);
+
 	if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
 	    bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) {
 		struct fence *fence;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 0a61930..dd086d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1048,56 +1048,6 @@ uint32_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
 	return flags;
 }
 
-static void amdgpu_ttm_lru_removal(struct ttm_buffer_object *tbo)
-{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-	unsigned i, j;
-
-	for (i = 0; i < AMDGPU_TTM_LRU_SIZE; ++i) {
-		struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
-
-		for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-			if (&tbo->lru == lru->lru[j])
-				lru->lru[j] = tbo->lru.prev;
-
-		if (&tbo->swap == lru->swap_lru)
-			lru->swap_lru = tbo->swap.prev;
-	}
-}
-
-static struct amdgpu_mman_lru *amdgpu_ttm_lru(struct ttm_buffer_object *tbo)
-{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-	unsigned log2_size = min(ilog2(tbo->num_pages),
-				 AMDGPU_TTM_LRU_SIZE - 1);
-
-	return &adev->mman.log2_size[log2_size];
-}
-
-static struct list_head *amdgpu_ttm_lru_tail(struct ttm_buffer_object *tbo)
-{
-	struct amdgpu_mman_lru *lru = amdgpu_ttm_lru(tbo);
-	struct list_head *res = lru->lru[tbo->mem.mem_type];
-
-	lru->lru[tbo->mem.mem_type] = &tbo->lru;
-	while ((++lru)->lru[tbo->mem.mem_type] == res)
-		lru->lru[tbo->mem.mem_type] = &tbo->lru;
-
-	return res;
-}
-
-static struct list_head *amdgpu_ttm_swap_lru_tail(struct ttm_buffer_object *tbo)
-{
-	struct amdgpu_mman_lru *lru = amdgpu_ttm_lru(tbo);
-	struct list_head *res = lru->swap_lru;
-
-	lru->swap_lru = &tbo->swap;
-	while ((++lru)->swap_lru == res)
-		lru->swap_lru = &tbo->swap;
-
-	return res;
-}
-
 static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
 					    const struct ttm_place *place)
 {
@@ -1136,14 +1086,12 @@ static struct ttm_bo_driver amdgpu_bo_driver = {
 	.fault_reserve_notify = &amdgpu_bo_fault_reserve_notify,
 	.io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
 	.io_mem_free = &amdgpu_ttm_io_mem_free,
-	.lru_removal = &amdgpu_ttm_lru_removal,
-	.lru_tail = &amdgpu_ttm_lru_tail,
-	.swap_lru_tail = &amdgpu_ttm_swap_lru_tail,
+	.lru_tail = &ttm_bo_default_lru_tail,
+	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int amdgpu_ttm_init(struct amdgpu_device *adev)
 {
-	unsigned i, j;
 	int r;
 
 	r = amdgpu_ttm_global_init(adev);
@@ -1161,19 +1109,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 		DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
 		return r;
 	}
-
-	for (i = 0; i < AMDGPU_TTM_LRU_SIZE; ++i) {
-		struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
-
-		for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-			lru->lru[j] = &adev->mman.bdev.man[j].lru[0];
-		lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0];
-	}
-
-	for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-		adev->mman.guard.lru[j] = NULL;
-	adev->mman.guard.swap_lru = NULL;
-
 	adev->mman.initialized = true;
 	r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM,
 				adev->mc.real_vram_size >> PAGE_SHIFT);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index d1c00c0..0eabbb2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -34,13 +34,6 @@
 #define AMDGPU_PL_FLAG_GWS		(TTM_PL_FLAG_PRIV << 1)
 #define AMDGPU_PL_FLAG_OA		(TTM_PL_FLAG_PRIV << 2)
 
-#define AMDGPU_TTM_LRU_SIZE	20
-
-struct amdgpu_mman_lru {
-	struct list_head		*lru[TTM_NUM_MEM_TYPES];
-	struct list_head		*swap_lru;
-};
-
 struct amdgpu_mman {
 	struct ttm_bo_global_ref        bo_global_ref;
 	struct drm_global_reference	mem_global_ref;
@@ -58,11 +51,6 @@ struct amdgpu_mman {
 	struct amdgpu_ring			*buffer_funcs_ring;
 	/* Scheduler entity for buffer moves */
 	struct amd_sched_entity			entity;
-
-	/* custom LRU management */
-	struct amdgpu_mman_lru			log2_size[AMDGPU_TTM_LRU_SIZE];
-	/* guard for log2_size array, don't add anything in between */
-	struct amdgpu_mman_lru			guard;
 };
 
 extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
-- 
2.7.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

RE: [PATCH 3/6] drm/amdgpu: user BO priority instead of self coding it

[He, Hongbo]

Reviewed-by: Roger.He <Hongbo.He@amd.com> 

-----Original Message-----
From: Christian König [mailto:deathsimple@vodafone.de] 
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <Hongbo.He@amd.com>
Cc: amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Subject: [PATCH 3/6] drm/amdgpu: user BO priority instead of self coding it

From: Christian König <christian.koenig@amd.com>

Keeping groups of BOs on the LRU is to time consuming on command submission.

Instead use the newly added BO priority to give a certain eviction order.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |  3 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 69 +-----------------------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    | 12 ------
 3 files changed, 5 insertions(+), 79 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 5076bc2..f399d98 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -386,6 +386,9 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 	if (unlikely(r != 0))
 		return r;
 
+	bo->tbo.priority = ilog2(bo->tbo.num_pages);
+	bo->tbo.priority = min(bo->tbo.priority, TTM_MAX_BO_PRIORITY - 1);
+
 	if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
 	    bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) {
 		struct fence *fence;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 0a61930..dd086d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1048,56 +1048,6 @@ uint32_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
 	return flags;
 }
 
-static void amdgpu_ttm_lru_removal(struct ttm_buffer_object *tbo) -{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-	unsigned i, j;
-
-	for (i = 0; i < AMDGPU_TTM_LRU_SIZE; ++i) {
-		struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
-
-		for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-			if (&tbo->lru == lru->lru[j])
-				lru->lru[j] = tbo->lru.prev;
-
-		if (&tbo->swap == lru->swap_lru)
-			lru->swap_lru = tbo->swap.prev;
-	}
-}
-
-static struct amdgpu_mman_lru *amdgpu_ttm_lru(struct ttm_buffer_object *tbo) -{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-	unsigned log2_size = min(ilog2(tbo->num_pages),
-				 AMDGPU_TTM_LRU_SIZE - 1);
-
-	return &adev->mman.log2_size[log2_size];
-}
-
-static struct list_head *amdgpu_ttm_lru_tail(struct ttm_buffer_object *tbo) -{
-	struct amdgpu_mman_lru *lru = amdgpu_ttm_lru(tbo);
-	struct list_head *res = lru->lru[tbo->mem.mem_type];
-
-	lru->lru[tbo->mem.mem_type] = &tbo->lru;
-	while ((++lru)->lru[tbo->mem.mem_type] == res)
-		lru->lru[tbo->mem.mem_type] = &tbo->lru;
-
-	return res;
-}
-
-static struct list_head *amdgpu_ttm_swap_lru_tail(struct ttm_buffer_object *tbo) -{
-	struct amdgpu_mman_lru *lru = amdgpu_ttm_lru(tbo);
-	struct list_head *res = lru->swap_lru;
-
-	lru->swap_lru = &tbo->swap;
-	while ((++lru)->swap_lru == res)
-		lru->swap_lru = &tbo->swap;
-
-	return res;
-}
-
 static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
 					    const struct ttm_place *place)  { @@ -1136,14 +1086,12 @@ static struct ttm_bo_driver amdgpu_bo_driver = {
 	.fault_reserve_notify = &amdgpu_bo_fault_reserve_notify,
 	.io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
 	.io_mem_free = &amdgpu_ttm_io_mem_free,
-	.lru_removal = &amdgpu_ttm_lru_removal,
-	.lru_tail = &amdgpu_ttm_lru_tail,
-	.swap_lru_tail = &amdgpu_ttm_swap_lru_tail,
+	.lru_tail = &ttm_bo_default_lru_tail,
+	.swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int amdgpu_ttm_init(struct amdgpu_device *adev)  {
-	unsigned i, j;
 	int r;
 
 	r = amdgpu_ttm_global_init(adev);
@@ -1161,19 +1109,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 		DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
 		return r;
 	}
-
-	for (i = 0; i < AMDGPU_TTM_LRU_SIZE; ++i) {
-		struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
-
-		for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-			lru->lru[j] = &adev->mman.bdev.man[j].lru[0];
-		lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0];
-	}
-
-	for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-		adev->mman.guard.lru[j] = NULL;
-	adev->mman.guard.swap_lru = NULL;
-
 	adev->mman.initialized = true;
 	r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM,
 				adev->mc.real_vram_size >> PAGE_SHIFT); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index d1c00c0..0eabbb2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -34,13 +34,6 @@
 #define AMDGPU_PL_FLAG_GWS		(TTM_PL_FLAG_PRIV << 1)
 #define AMDGPU_PL_FLAG_OA		(TTM_PL_FLAG_PRIV << 2)
 
-#define AMDGPU_TTM_LRU_SIZE	20
-
-struct amdgpu_mman_lru {
-	struct list_head		*lru[TTM_NUM_MEM_TYPES];
-	struct list_head		*swap_lru;
-};
-
 struct amdgpu_mman {
 	struct ttm_bo_global_ref        bo_global_ref;
 	struct drm_global_reference	mem_global_ref;
@@ -58,11 +51,6 @@ struct amdgpu_mman {
 	struct amdgpu_ring			*buffer_funcs_ring;
 	/* Scheduler entity for buffer moves */
 	struct amd_sched_entity			entity;
-
-	/* custom LRU management */
-	struct amdgpu_mman_lru			log2_size[AMDGPU_TTM_LRU_SIZE];
-	/* guard for log2_size array, don't add anything in between */
-	struct amdgpu_mman_lru			guard;
 };
 
 extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
--
2.7.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 4/6] drm/amdgpu: double the priority of kernel allocations

[Christian König]

From: Christian König <christian.koenig@amd.com>

Give kernel allocations a higher priority cause it is often
more work to swap them back in.

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

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index f399d98..2de1dda 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -387,6 +387,8 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 		return r;
 
 	bo->tbo.priority = ilog2(bo->tbo.num_pages);
+	if (kernel)
+		bo->tbo.priority *= 2;
 	bo->tbo.priority = min(bo->tbo.priority, TTM_MAX_BO_PRIORITY - 1);
 
 	if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
-- 
2.7.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

RE: [PATCH 4/6] drm/amdgpu: double the priority of kernel allocations

[He, Hongbo]

Reviewed-by: Roger.He <Hongbo.He@amd.com>

-----Original Message-----
From: Christian König [mailto:deathsimple@vodafone.de] 
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <Hongbo.He@amd.com>
Cc: amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Subject: [PATCH 4/6] drm/amdgpu: double the priority of kernel allocations

From: Christian König <christian.koenig@amd.com>

Give kernel allocations a higher priority cause it is often more work to swap them back in.

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

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index f399d98..2de1dda 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -387,6 +387,8 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 		return r;
 
 	bo->tbo.priority = ilog2(bo->tbo.num_pages);
+	if (kernel)
+		bo->tbo.priority *= 2;
 	bo->tbo.priority = min(bo->tbo.priority, TTM_MAX_BO_PRIORITY - 1);
 
 	if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
--
2.7.4

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