[PATCH 1/7] drm/ttm: call ttm_bo_swapout directly when ttm shrink

[PATCH 1/7] drm/ttm: call ttm_bo_swapout directly when ttm shrink

[Roger He]

then remove superfluous functions

Change-Id: Iea020f0e30a239e0265e7a1500168c7d7f819bd9
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c     | 21 +++---------
 drivers/gpu/drm/ttm/ttm_memory.c | 12 ++-----
 include/drm/ttm/ttm_bo_api.h     |  1 +
 include/drm/ttm/ttm_bo_driver.h  |  1 -
 include/drm/ttm/ttm_memory.h     | 69 +---------------------------------------
 5 files changed, 9 insertions(+), 95 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 60bb5c1..fa57aa8 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -42,7 +42,6 @@
 #include <linux/atomic.h>
 #include <linux/reservation.h>
 
-static int ttm_bo_swapout(struct ttm_mem_shrink *shrink);
 static void ttm_bo_global_kobj_release(struct kobject *kobj);
 
 static struct attribute ttm_bo_count = {
@@ -1454,7 +1453,6 @@ static void ttm_bo_global_kobj_release(struct kobject *kobj)
 	struct ttm_bo_global *glob =
 		container_of(kobj, struct ttm_bo_global, kobj);
 
-	ttm_mem_unregister_shrink(glob->mem_glob, &glob->shrink);
 	__free_page(glob->dummy_read_page);
 	kfree(glob);
 }
@@ -1479,6 +1477,7 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
 	mutex_init(&glob->device_list_mutex);
 	spin_lock_init(&glob->lru_lock);
 	glob->mem_glob = bo_ref->mem_glob;
+	glob->mem_glob->bo_glob = glob;
 	glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32);
 
 	if (unlikely(glob->dummy_read_page == NULL)) {
@@ -1489,14 +1488,6 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
 	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);
-	ret = ttm_mem_register_shrink(glob->mem_glob, &glob->shrink);
-	if (unlikely(ret != 0)) {
-		pr_err("Could not register buffer object swapout\n");
-		goto out_no_shrink;
-	}
-
 	atomic_set(&glob->bo_count, 0);
 
 	ret = kobject_init_and_add(
@@ -1504,8 +1495,6 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
 	if (unlikely(ret != 0))
 		kobject_put(&glob->kobj);
 	return ret;
-out_no_shrink:
-	__free_page(glob->dummy_read_page);
 out_no_drp:
 	kfree(glob);
 	return ret;
@@ -1688,11 +1677,8 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
  * A buffer object shrink method that tries to swap out the first
  * buffer object on the bo_global::swap_lru list.
  */
-
-static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
+int ttm_bo_swapout(struct ttm_bo_global *glob)
 {
-	struct ttm_bo_global *glob =
-	    container_of(shrink, struct ttm_bo_global, shrink);
 	struct ttm_buffer_object *bo;
 	int ret = -EBUSY;
 	unsigned i;
@@ -1774,10 +1760,11 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
 	kref_put(&bo->list_kref, ttm_bo_release_list);
 	return ret;
 }
+EXPORT_SYMBOL(ttm_bo_swapout);
 
 void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
 {
-	while (ttm_bo_swapout(&bdev->glob->shrink) == 0)
+	while (ttm_bo_swapout(bdev->glob) == 0)
 		;
 }
 EXPORT_SYMBOL(ttm_bo_swapout_all);
diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
index e963749..9130bdf 100644
--- a/drivers/gpu/drm/ttm/ttm_memory.c
+++ b/drivers/gpu/drm/ttm/ttm_memory.c
@@ -214,26 +214,20 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
 		       uint64_t extra)
 {
 	int ret;
-	struct ttm_mem_shrink *shrink;
 
 	spin_lock(&glob->lock);
-	if (glob->shrink == NULL)
-		goto out;
 
 	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
-		shrink = glob->shrink;
 		spin_unlock(&glob->lock);
-		ret = shrink->do_shrink(shrink);
+		ret = ttm_bo_swapout(glob->bo_glob);
 		spin_lock(&glob->lock);
 		if (unlikely(ret != 0))
-			goto out;
+			break;
 	}
-out:
+
 	spin_unlock(&glob->lock);
 }
 
-
-
 static void ttm_shrink_work(struct work_struct *work)
 {
 	struct ttm_mem_global *glob =
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index c126330..24a8db7 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -752,6 +752,7 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
 		  const char __user *wbuf, char __user *rbuf,
 		  size_t count, loff_t *f_pos, bool write);
 
+int ttm_bo_swapout(struct ttm_bo_global *glob);
 void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
 int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
 #endif
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 5115718..934fecf 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -522,7 +522,6 @@ struct ttm_bo_global {
 	struct kobject kobj;
 	struct ttm_mem_global *mem_glob;
 	struct page *dummy_read_page;
-	struct ttm_mem_shrink shrink;
 	struct mutex device_list_mutex;
 	spinlock_t lru_lock;
 
diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
index 2c1e359..85f3ad6 100644
--- a/include/drm/ttm/ttm_memory.h
+++ b/include/drm/ttm/ttm_memory.h
@@ -37,20 +37,6 @@
 #include <linux/mm.h>
 
 /**
- * struct ttm_mem_shrink - callback to shrink TTM memory usage.
- *
- * @do_shrink: The callback function.
- *
- * Arguments to the do_shrink functions are intended to be passed using
- * inheritance. That is, the argument class derives from struct ttm_mem_shrink,
- * and can be accessed using container_of().
- */
-
-struct ttm_mem_shrink {
-	int (*do_shrink) (struct ttm_mem_shrink *);
-};
-
-/**
  * struct ttm_mem_global - Global memory accounting structure.
  *
  * @shrink: A single callback to shrink TTM memory usage. Extend this
@@ -76,7 +62,7 @@ struct ttm_mem_shrink {
 struct ttm_mem_zone;
 struct ttm_mem_global {
 	struct kobject kobj;
-	struct ttm_mem_shrink *shrink;
+	struct ttm_bo_global *bo_glob;
 	struct workqueue_struct *swap_queue;
 	struct work_struct work;
 	spinlock_t lock;
@@ -90,59 +76,6 @@ struct ttm_mem_global {
 #endif
 };
 
-/**
- * ttm_mem_init_shrink - initialize a struct ttm_mem_shrink object
- *
- * @shrink: The object to initialize.
- * @func: The callback function.
- */
-
-static inline void ttm_mem_init_shrink(struct ttm_mem_shrink *shrink,
-				       int (*func) (struct ttm_mem_shrink *))
-{
-	shrink->do_shrink = func;
-}
-
-/**
- * ttm_mem_register_shrink - register a struct ttm_mem_shrink object.
- *
- * @glob: The struct ttm_mem_global object to register with.
- * @shrink: An initialized struct ttm_mem_shrink object to register.
- *
- * Returns:
- * -EBUSY: There's already a callback registered. (May change).
- */
-
-static inline int ttm_mem_register_shrink(struct ttm_mem_global *glob,
-					  struct ttm_mem_shrink *shrink)
-{
-	spin_lock(&glob->lock);
-	if (glob->shrink != NULL) {
-		spin_unlock(&glob->lock);
-		return -EBUSY;
-	}
-	glob->shrink = shrink;
-	spin_unlock(&glob->lock);
-	return 0;
-}
-
-/**
- * ttm_mem_unregister_shrink - unregister a struct ttm_mem_shrink object.
- *
- * @glob: The struct ttm_mem_global object to unregister from.
- * @shrink: A previously registert struct ttm_mem_shrink object.
- *
- */
-
-static inline void ttm_mem_unregister_shrink(struct ttm_mem_global *glob,
-					     struct ttm_mem_shrink *shrink)
-{
-	spin_lock(&glob->lock);
-	BUG_ON(glob->shrink != shrink);
-	glob->shrink = NULL;
-	spin_unlock(&glob->lock);
-}
-
 extern int ttm_mem_global_init(struct ttm_mem_global *glob);
 extern void ttm_mem_global_release(struct ttm_mem_global *glob);
 extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
-- 
2.7.4

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

[PATCH 2/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc

[Roger He]

Change-Id: I5279b5cd3560c4082b00f822219575a5f9c3808a
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c                    |  2 +-
 drivers/gpu/drm/ttm/ttm_memory.c                | 15 +++++++++------
 drivers/gpu/drm/ttm/ttm_object.c                | 13 ++++++++++---
 drivers/gpu/drm/vmwgfx/vmwgfx_binding.c         |  6 +++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c          | 13 ++++++++++---
 drivers/gpu/drm/vmwgfx/vmwgfx_context.c         |  6 +++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c         |  6 +++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c           |  6 +++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_shader.c          | 18 +++++++++++++++---
 drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c |  6 +++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_so.c              |  6 +++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_surface.c         | 12 ++++++++++--
 include/drm/ttm/ttm_memory.h                    |  3 ++-
 13 files changed, 87 insertions(+), 25 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index fa57aa8..c59f572 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1133,7 +1133,7 @@ int ttm_bo_init_reserved(struct ttm_bo_device *bdev,
 	struct ttm_mem_global *mem_glob = bdev->glob->mem_glob;
 	bool locked;
 
-	ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false);
+	ret = ttm_mem_global_alloc(mem_glob, acc_size, ctx);
 	if (ret) {
 		pr_err("Out of kernel memory\n");
 		if (destroy)
diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
index 9130bdf..525d3b6 100644
--- a/drivers/gpu/drm/ttm/ttm_memory.c
+++ b/drivers/gpu/drm/ttm/ttm_memory.c
@@ -508,7 +508,7 @@ static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
 static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
 				     struct ttm_mem_zone *single_zone,
 				     uint64_t memory,
-				     bool no_wait, bool interruptible)
+				     struct ttm_operation_ctx *ctx)
 {
 	int count = TTM_MEMORY_ALLOC_RETRIES;
 
@@ -516,7 +516,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
 					       single_zone,
 					       memory, true)
 			!= 0)) {
-		if (no_wait)
+		if (ctx->no_wait_gpu)
 			return -ENOMEM;
 		if (unlikely(count-- == 0))
 			return -ENOMEM;
@@ -527,15 +527,14 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
 }
 
 int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
-			 bool no_wait, bool interruptible)
+			 struct ttm_operation_ctx *ctx)
 {
 	/**
 	 * Normal allocations of kernel memory are registered in
 	 * all zones.
 	 */
 
-	return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
-					 interruptible);
+	return ttm_mem_global_alloc_zone(glob, NULL, memory, ctx);
 }
 EXPORT_SYMBOL(ttm_mem_global_alloc);
 
@@ -544,6 +543,10 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
 {
 
 	struct ttm_mem_zone *zone = NULL;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 
 	/**
 	 * Page allocations may be registed in a single zone
@@ -557,7 +560,7 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
 	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
 		zone = glob->zone_kernel;
 #endif
-	return ttm_mem_global_alloc_zone(glob, zone, size, false, false);
+	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
 }
 
 void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
diff --git a/drivers/gpu/drm/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c
index 26a7ad0..1aa2baa 100644
--- a/drivers/gpu/drm/ttm/ttm_object.c
+++ b/drivers/gpu/drm/ttm/ttm_object.c
@@ -325,6 +325,10 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
 	struct ttm_ref_object *ref;
 	struct drm_hash_item *hash;
 	struct ttm_mem_global *mem_glob = tfile->tdev->mem_glob;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	int ret = -EINVAL;
 
 	if (base->tfile != tfile && !base->shareable)
@@ -350,7 +354,7 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
 			return -EPERM;
 
 		ret = ttm_mem_global_alloc(mem_glob, sizeof(*ref),
-					   false, false);
+					   &ctx);
 		if (unlikely(ret != 0))
 			return ret;
 		ref = kmalloc(sizeof(*ref), GFP_KERNEL);
@@ -686,7 +690,10 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
 	dma_buf = prime->dma_buf;
 	if (!dma_buf || !get_dma_buf_unless_doomed(dma_buf)) {
 		DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
-
+		struct ttm_operation_ctx ctx = {
+			.interruptible = true,
+			.no_wait_gpu = false
+		};
 		exp_info.ops = &tdev->ops;
 		exp_info.size = prime->size;
 		exp_info.flags = flags;
@@ -696,7 +703,7 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
 		 * Need to create a new dma_buf, with memory accounting.
 		 */
 		ret = ttm_mem_global_alloc(tdev->mem_glob, tdev->dma_buf_size,
-					   false, true);
+					   &ctx);
 		if (unlikely(ret != 0)) {
 			mutex_unlock(&prime->mutex);
 			goto out_unref;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
index 9c42e96..55d32ae 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
@@ -1202,10 +1202,14 @@ struct vmw_ctx_binding_state *
 vmw_binding_state_alloc(struct vmw_private *dev_priv)
 {
 	struct vmw_ctx_binding_state *cbs;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	int ret;
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), sizeof(*cbs),
-				   false, false);
+				&ctx);
 	if (ret)
 		return ERR_PTR(ret);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index c705632..ef97542 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -394,6 +394,10 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
 	struct vmw_private *dev_priv = vmw_tt->dev_priv;
 	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
 	struct vmw_sg_table *vsgt = &vmw_tt->vsgt;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	struct vmw_piter iter;
 	dma_addr_t old;
 	int ret = 0;
@@ -417,8 +421,7 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
 			sgt_size = ttm_round_pot(sizeof(struct sg_table));
 		}
 		vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages;
-		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false,
-					   true);
+		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, &ctx);
 		if (unlikely(ret != 0))
 			return ret;
 
@@ -638,6 +641,10 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
 		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
 	struct vmw_private *dev_priv = vmw_tt->dev_priv;
 	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	int ret;
 
 	if (ttm->state != tt_unpopulated)
@@ -646,7 +653,7 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
 	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
 		size_t size =
 			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
-		ret = ttm_mem_global_alloc(glob, size, false, true);
+		ret = ttm_mem_global_alloc(glob, size, &ctx);
 		if (unlikely(ret != 0))
 			return ret;
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
index 4212b3e..3767ac3 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
@@ -746,6 +746,10 @@ static int vmw_context_define(struct drm_device *dev, void *data,
 	struct vmw_resource *tmp;
 	struct drm_vmw_context_arg *arg = (struct drm_vmw_context_arg *)data;
 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct ttm_operation_ctx ttm_opt_ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	int ret;
 
 	if (!dev_priv->has_dx && dx) {
@@ -768,7 +772,7 @@ static int vmw_context_define(struct drm_device *dev, void *data,
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
 				   vmw_user_context_size,
-				   false, true);
+				   &ttm_opt_ctx);
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for context"
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
index 92df0b0..cbf54ea 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
@@ -573,6 +573,10 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
 				       u32 type)
 {
 	struct vmw_cotable *vcotbl;
+	struct ttm_operation_ctx ttm_opt_ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	int ret;
 	u32 num_entries;
 
@@ -580,7 +584,7 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
 		cotable_acc_size = ttm_round_pot(sizeof(struct vmw_cotable));
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
-				   cotable_acc_size, false, true);
+				   cotable_acc_size, &ttm_opt_ctx);
 	if (unlikely(ret))
 		return ERR_PTR(ret);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
index d6b1c50..6c5c75c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
@@ -588,6 +588,10 @@ int vmw_user_fence_create(struct drm_file *file_priv,
 	struct vmw_user_fence *ufence;
 	struct vmw_fence_obj *tmp;
 	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	int ret;
 
 	/*
@@ -596,7 +600,7 @@ int vmw_user_fence_create(struct drm_file *file_priv,
 	 */
 
 	ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size,
-				   false, false);
+				   &ctx);
 	if (unlikely(ret != 0))
 		return ret;
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
index 004e18b..73b8e9a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
@@ -607,6 +607,10 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
 	struct vmw_dx_shader *shader;
 	struct vmw_resource *res;
 	struct vmw_private *dev_priv = ctx->dev_priv;
+	struct ttm_operation_ctx ttm_opt_ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	int ret;
 
 	if (!vmw_shader_dx_size)
@@ -616,7 +620,7 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
 		return -EINVAL;
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), vmw_shader_dx_size,
-				   false, true);
+				   &ttm_opt_ctx);
 	if (ret) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for shader "
@@ -730,6 +734,10 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
 {
 	struct vmw_user_shader *ushader;
 	struct vmw_resource *res, *tmp;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	int ret;
 
 	/*
@@ -742,7 +750,7 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
 				   vmw_user_shader_size,
-				   false, true);
+				   &ctx);
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for shader "
@@ -800,6 +808,10 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
 {
 	struct vmw_shader *shader;
 	struct vmw_resource *res;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	int ret;
 
 	/*
@@ -812,7 +824,7 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
 				   vmw_shader_size,
-				   false, true);
+				   &ctx);
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for shader "
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
index 051d3b3..a0cb310 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
@@ -149,6 +149,10 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
 	struct vmw_resource *res;
 	struct vmw_resource *tmp;
 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	size_t alloc_size;
 	size_t account_size;
 	int ret;
@@ -162,7 +166,7 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
 		return ret;
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), account_size,
-				   false, true);
+				   &ctx);
 	ttm_read_unlock(&dev_priv->reservation_sem);
 	if (ret) {
 		if (ret != -ERESTARTSYS)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
index 5a73eeb..d3573c3 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
@@ -329,6 +329,10 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
 	struct vmw_private *dev_priv = ctx->dev_priv;
 	struct vmw_resource *res;
 	struct vmw_view *view;
+	struct ttm_operation_ctx ttm_opt_ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	size_t size;
 	int ret;
 
@@ -345,7 +349,7 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
 
 	size = offsetof(struct vmw_view, cmd) + cmd_size;
 
-	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, false, true);
+	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, &ttm_opt_ctx);
 	if (ret) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for view"
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
index 6ac094e..db1bb16 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
@@ -700,6 +700,10 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
 	struct drm_vmw_surface_create_req *req = &arg->req;
 	struct drm_vmw_surface_arg *rep = &arg->rep;
 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	int ret;
 	int i, j;
 	uint32_t cur_bo_offset;
@@ -741,7 +745,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
 		return ret;
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
-				   size, false, true);
+				   size, &ctx);
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for surface"
@@ -1479,6 +1483,10 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
 {
 	struct vmw_private *dev_priv = vmw_priv(dev);
 	struct vmw_user_surface *user_srf;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
 	struct vmw_surface *srf;
 	int ret;
 	u32 num_layers;
@@ -1525,7 +1533,7 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
 		return ret;
 
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
-				   user_accounting_size, false, true);
+				   user_accounting_size, &ctx);
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for surface"
diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
index 85f3ad6..755c107 100644
--- a/include/drm/ttm/ttm_memory.h
+++ b/include/drm/ttm/ttm_memory.h
@@ -35,6 +35,7 @@
 #include <linux/errno.h>
 #include <linux/kobject.h>
 #include <linux/mm.h>
+#include "ttm_bo_api.h"
 
 /**
  * struct ttm_mem_global - Global memory accounting structure.
@@ -79,7 +80,7 @@ struct ttm_mem_global {
 extern int ttm_mem_global_init(struct ttm_mem_global *glob);
 extern void ttm_mem_global_release(struct ttm_mem_global *glob);
 extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
-				bool no_wait, bool interruptible);
+				struct ttm_operation_ctx *ctx);
 extern void ttm_mem_global_free(struct ttm_mem_global *glob,
 				uint64_t amount);
 extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
-- 
2.7.4

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

[PATCH 3/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc_page

[Roger He]

Change-Id: I4104a12e09a374b6477a0dd5a8fce26dce27a746
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_memory.c         | 15 ++++++++-------
 drivers/gpu/drm/ttm/ttm_page_alloc.c     |  6 +++++-
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 ++++++--
 include/drm/ttm/ttm_memory.h             |  3 ++-
 4 files changed, 21 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
index 525d3b6..8df0755 100644
--- a/drivers/gpu/drm/ttm/ttm_memory.c
+++ b/drivers/gpu/drm/ttm/ttm_memory.c
@@ -539,15 +539,14 @@ int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
 EXPORT_SYMBOL(ttm_mem_global_alloc);
 
 int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
-			      struct page *page, uint64_t size)
+			      struct page *page, uint64_t size,
+			      struct ttm_operation_ctx *ctx)
 {
-
+	int ret;
 	struct ttm_mem_zone *zone = NULL;
-	struct ttm_operation_ctx ctx = {
-		.interruptible = false,
-		.no_wait_gpu = false
-	};
+	bool tmp_no_wait_gpu = ctx->no_wait_gpu;
 
+	ctx->no_wait_gpu = false;
 	/**
 	 * Page allocations may be registed in a single zone
 	 * only if highmem or !dma32.
@@ -560,7 +559,9 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
 	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
 		zone = glob->zone_kernel;
 #endif
-	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
+	ret = ttm_mem_global_alloc_zone(glob, zone, size, ctx);
+	ctx->no_wait_gpu = tmp_no_wait_gpu;
+	return ret;
 }
 
 void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index b5ba644..8f93ff3 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -1061,6 +1061,10 @@ void ttm_page_alloc_fini(void)
 int ttm_pool_populate(struct ttm_tt *ttm)
 {
 	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	unsigned i;
 	int ret;
 
@@ -1076,7 +1080,7 @@ int ttm_pool_populate(struct ttm_tt *ttm)
 
 	for (i = 0; i < ttm->num_pages; ++i) {
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
-						PAGE_SIZE);
+						PAGE_SIZE, &ctx);
 		if (unlikely(ret != 0)) {
 			ttm_pool_unpopulate(ttm);
 			return -ENOMEM;
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
index bda00b2..8aac86a 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
@@ -927,6 +927,10 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 {
 	struct ttm_tt *ttm = &ttm_dma->ttm;
 	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	unsigned long num_pages = ttm->num_pages;
 	struct dma_pool *pool;
 	enum pool_type type;
@@ -962,7 +966,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 			break;
 
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
-						pool->size);
+						pool->size, &ctx);
 		if (unlikely(ret != 0)) {
 			ttm_dma_unpopulate(ttm_dma, dev);
 			return -ENOMEM;
@@ -998,7 +1002,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 		}
 
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
-						pool->size);
+						pool->size, &ctx);
 		if (unlikely(ret != 0)) {
 			ttm_dma_unpopulate(ttm_dma, dev);
 			return -ENOMEM;
diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
index 755c107..8936285 100644
--- a/include/drm/ttm/ttm_memory.h
+++ b/include/drm/ttm/ttm_memory.h
@@ -84,7 +84,8 @@ extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
 extern void ttm_mem_global_free(struct ttm_mem_global *glob,
 				uint64_t amount);
 extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
-				     struct page *page, uint64_t size);
+				     struct page *page, uint64_t size,
+				     struct ttm_operation_ctx *ctx);
 extern void ttm_mem_global_free_page(struct ttm_mem_global *glob,
 				     struct page *page, uint64_t size);
 extern size_t ttm_round_pot(size_t size);
-- 
2.7.4

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

[PATCH 4/7] drm/ttm: use an operation ctx for ttm_tt_populate in ttm_bo_driver

[Roger He]

Change-Id: I803ea52d11e5c06add0dffab836c3aecc00b56dd
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  7 ++++---
 drivers/gpu/drm/ast/ast_ttm.c            |  5 +++--
 drivers/gpu/drm/cirrus/cirrus_ttm.c      |  5 +++--
 drivers/gpu/drm/nouveau/nouveau_bo.c     |  8 ++++----
 drivers/gpu/drm/qxl/qxl_ttm.c            |  5 +++--
 drivers/gpu/drm/radeon/radeon_ttm.c      |  9 +++++----
 drivers/gpu/drm/ttm/ttm_agp_backend.c    |  4 ++--
 drivers/gpu/drm/ttm/ttm_bo_util.c        | 11 ++++++++---
 drivers/gpu/drm/ttm/ttm_bo_vm.c          |  7 ++++++-
 drivers/gpu/drm/ttm/ttm_page_alloc.c     | 13 +++++--------
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 11 ++++-------
 drivers/gpu/drm/ttm/ttm_tt.c             |  6 +++++-
 drivers/gpu/drm/virtio/virtgpu_object.c  |  6 +++++-
 drivers/gpu/drm/virtio/virtgpu_ttm.c     |  5 +++--
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c   | 13 +++++--------
 drivers/gpu/drm/vmwgfx/vmwgfx_mob.c      | 13 +++++++++++--
 include/drm/ttm/ttm_bo_driver.h          |  5 +++--
 include/drm/ttm/ttm_page_alloc.h         | 11 +++++++----
 18 files changed, 86 insertions(+), 58 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index f1b7d98..52aab9d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -990,7 +990,8 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_bo_device *bdev,
 	return &gtt->ttm.ttm;
 }
 
-static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm)
+static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
+					struct ttm_operation_ctx *ctx)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
@@ -1018,11 +1019,11 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm)
 
 #ifdef CONFIG_SWIOTLB
 	if (swiotlb_nr_tbl()) {
-		return ttm_dma_populate(&gtt->ttm, adev->dev);
+		return ttm_dma_populate(&gtt->ttm, adev->dev, ctx);
 	}
 #endif
 
-	return ttm_populate_and_map_pages(adev->dev, &gtt->ttm);
+	return ttm_populate_and_map_pages(adev->dev, &gtt->ttm, ctx);
 }
 
 static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c
index 28da7c2..1413e94 100644
--- a/drivers/gpu/drm/ast/ast_ttm.c
+++ b/drivers/gpu/drm/ast/ast_ttm.c
@@ -216,9 +216,10 @@ static struct ttm_tt *ast_ttm_tt_create(struct ttm_bo_device *bdev,
 	return tt;
 }
 
-static int ast_ttm_tt_populate(struct ttm_tt *ttm)
+static int ast_ttm_tt_populate(struct ttm_tt *ttm,
+			struct ttm_operation_ctx *ctx)
 {
-	return ttm_pool_populate(ttm);
+	return ttm_pool_populate(ttm, ctx);
 }
 
 static void ast_ttm_tt_unpopulate(struct ttm_tt *ttm)
diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c
index 2a5b54d..95e2d40 100644
--- a/drivers/gpu/drm/cirrus/cirrus_ttm.c
+++ b/drivers/gpu/drm/cirrus/cirrus_ttm.c
@@ -216,9 +216,10 @@ static struct ttm_tt *cirrus_ttm_tt_create(struct ttm_bo_device *bdev,
 	return tt;
 }
 
-static int cirrus_ttm_tt_populate(struct ttm_tt *ttm)
+static int cirrus_ttm_tt_populate(struct ttm_tt *ttm,
+				struct ttm_operation_ctx *ctx)
 {
-	return ttm_pool_populate(ttm);
+	return ttm_pool_populate(ttm, ctx);
 }
 
 static void cirrus_ttm_tt_unpopulate(struct ttm_tt *ttm)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 6b6fb20..b141c27 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1547,7 +1547,7 @@ nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
 }
 
 static int
-nouveau_ttm_tt_populate(struct ttm_tt *ttm)
+nouveau_ttm_tt_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
 {
 	struct ttm_dma_tt *ttm_dma = (void *)ttm;
 	struct nouveau_drm *drm;
@@ -1572,17 +1572,17 @@ nouveau_ttm_tt_populate(struct ttm_tt *ttm)
 
 #if IS_ENABLED(CONFIG_AGP)
 	if (drm->agp.bridge) {
-		return ttm_agp_tt_populate(ttm);
+		return ttm_agp_tt_populate(ttm, ctx);
 	}
 #endif
 
 #if IS_ENABLED(CONFIG_SWIOTLB) && IS_ENABLED(CONFIG_X86)
 	if (swiotlb_nr_tbl()) {
-		return ttm_dma_populate((void *)ttm, dev);
+		return ttm_dma_populate((void *)ttm, dev, ctx);
 	}
 #endif
 
-	r = ttm_pool_populate(ttm);
+	r = ttm_pool_populate(ttm, ctx);
 	if (r) {
 		return r;
 	}
diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
index 78ce118..989645c 100644
--- a/drivers/gpu/drm/qxl/qxl_ttm.c
+++ b/drivers/gpu/drm/qxl/qxl_ttm.c
@@ -291,14 +291,15 @@ static struct ttm_backend_func qxl_backend_func = {
 	.destroy = &qxl_ttm_backend_destroy,
 };
 
-static int qxl_ttm_tt_populate(struct ttm_tt *ttm)
+static int qxl_ttm_tt_populate(struct ttm_tt *ttm,
+				struct ttm_operation_ctx *ctx)
 {
 	int r;
 
 	if (ttm->state != tt_unpopulated)
 		return 0;
 
-	r = ttm_pool_populate(ttm);
+	r = ttm_pool_populate(ttm, ctx);
 	if (r)
 		return r;
 
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 557fd79..634fc6f 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -721,7 +721,8 @@ static struct radeon_ttm_tt *radeon_ttm_tt_to_gtt(struct ttm_tt *ttm)
 	return (struct radeon_ttm_tt *)ttm;
 }
 
-static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
+static int radeon_ttm_tt_populate(struct ttm_tt *ttm,
+				struct ttm_operation_ctx *ctx)
 {
 	struct radeon_ttm_tt *gtt = radeon_ttm_tt_to_gtt(ttm);
 	struct radeon_device *rdev;
@@ -750,17 +751,17 @@ static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
 	rdev = radeon_get_rdev(ttm->bdev);
 #if IS_ENABLED(CONFIG_AGP)
 	if (rdev->flags & RADEON_IS_AGP) {
-		return ttm_agp_tt_populate(ttm);
+		return ttm_agp_tt_populate(ttm, ctx);
 	}
 #endif
 
 #ifdef CONFIG_SWIOTLB
 	if (swiotlb_nr_tbl()) {
-		return ttm_dma_populate(&gtt->ttm, rdev->dev);
+		return ttm_dma_populate(&gtt->ttm, rdev->dev, ctx);
 	}
 #endif
 
-	return ttm_populate_and_map_pages(rdev->dev, &gtt->ttm);
+	return ttm_populate_and_map_pages(rdev->dev, &gtt->ttm, ctx);
 }
 
 static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
diff --git a/drivers/gpu/drm/ttm/ttm_agp_backend.c b/drivers/gpu/drm/ttm/ttm_agp_backend.c
index 028ab60..3e795a0 100644
--- a/drivers/gpu/drm/ttm/ttm_agp_backend.c
+++ b/drivers/gpu/drm/ttm/ttm_agp_backend.c
@@ -133,12 +133,12 @@ struct ttm_tt *ttm_agp_tt_create(struct ttm_bo_device *bdev,
 }
 EXPORT_SYMBOL(ttm_agp_tt_create);
 
-int ttm_agp_tt_populate(struct ttm_tt *ttm)
+int ttm_agp_tt_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
 {
 	if (ttm->state != tt_unpopulated)
 		return 0;
 
-	return ttm_pool_populate(ttm);
+	return ttm_pool_populate(ttm, ctx);
 }
 EXPORT_SYMBOL(ttm_agp_tt_populate);
 
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 6e353df..b7eb507 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -376,7 +376,7 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
 	 * TTM might be null for moves within the same region.
 	 */
 	if (ttm && ttm->state == tt_unpopulated) {
-		ret = ttm->bdev->driver->ttm_tt_populate(ttm);
+		ret = ttm->bdev->driver->ttm_tt_populate(ttm, ctx);
 		if (ret)
 			goto out1;
 	}
@@ -545,14 +545,19 @@ static int ttm_bo_kmap_ttm(struct ttm_buffer_object *bo,
 			   unsigned long num_pages,
 			   struct ttm_bo_kmap_obj *map)
 {
-	struct ttm_mem_reg *mem = &bo->mem; pgprot_t prot;
+	struct ttm_mem_reg *mem = &bo->mem;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	struct ttm_tt *ttm = bo->ttm;
+	pgprot_t prot;
 	int ret;
 
 	BUG_ON(!ttm);
 
 	if (ttm->state == tt_unpopulated) {
-		ret = ttm->bdev->driver->ttm_tt_populate(ttm);
+		ret = ttm->bdev->driver->ttm_tt_populate(ttm, &ctx);
 		if (ret)
 			return ret;
 	}
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index c8ebb75..65dfcdd 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -215,12 +215,17 @@ static int ttm_bo_vm_fault(struct vm_fault *vmf)
 		cvma.vm_page_prot = ttm_io_prot(bo->mem.placement,
 						cvma.vm_page_prot);
 	} else {
+		struct ttm_operation_ctx ctx = {
+			.interruptible = false,
+			.no_wait_gpu = false
+		};
+
 		ttm = bo->ttm;
 		cvma.vm_page_prot = ttm_io_prot(bo->mem.placement,
 						cvma.vm_page_prot);
 
 		/* Allocate all page at once, most common usage */
-		if (ttm->bdev->driver->ttm_tt_populate(ttm)) {
+		if (ttm->bdev->driver->ttm_tt_populate(ttm, &ctx)) {
 			retval = VM_FAULT_OOM;
 			goto out_io_unlock;
 		}
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index 8f93ff3..f1a3d55 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -1058,13 +1058,9 @@ void ttm_page_alloc_fini(void)
 	_manager = NULL;
 }
 
-int ttm_pool_populate(struct ttm_tt *ttm)
+int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
 {
 	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
-	struct ttm_operation_ctx ctx = {
-		.interruptible = false,
-		.no_wait_gpu = false
-	};
 	unsigned i;
 	int ret;
 
@@ -1080,7 +1076,7 @@ int ttm_pool_populate(struct ttm_tt *ttm)
 
 	for (i = 0; i < ttm->num_pages; ++i) {
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
-						PAGE_SIZE, &ctx);
+						PAGE_SIZE, ctx);
 		if (unlikely(ret != 0)) {
 			ttm_pool_unpopulate(ttm);
 			return -ENOMEM;
@@ -1117,12 +1113,13 @@ void ttm_pool_unpopulate(struct ttm_tt *ttm)
 }
 EXPORT_SYMBOL(ttm_pool_unpopulate);
 
-int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt)
+int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt,
+					struct ttm_operation_ctx *ctx)
 {
 	unsigned i, j;
 	int r;
 
-	r = ttm_pool_populate(&tt->ttm);
+	r = ttm_pool_populate(&tt->ttm, ctx);
 	if (r)
 		return r;
 
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
index 8aac86a..3ac5391 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
@@ -923,14 +923,11 @@ static gfp_t ttm_dma_pool_gfp_flags(struct ttm_dma_tt *ttm_dma, bool huge)
  * On success pages list will hold count number of correctly
  * cached pages. On failure will hold the negative return value (-ENOMEM, etc).
  */
-int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
+int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
+			struct ttm_operation_ctx *ctx)
 {
 	struct ttm_tt *ttm = &ttm_dma->ttm;
 	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
-	struct ttm_operation_ctx ctx = {
-		.interruptible = false,
-		.no_wait_gpu = false
-	};
 	unsigned long num_pages = ttm->num_pages;
 	struct dma_pool *pool;
 	enum pool_type type;
@@ -966,7 +963,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 			break;
 
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
-						pool->size, &ctx);
+						pool->size, ctx);
 		if (unlikely(ret != 0)) {
 			ttm_dma_unpopulate(ttm_dma, dev);
 			return -ENOMEM;
@@ -1002,7 +999,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 		}
 
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
-						pool->size, &ctx);
+						pool->size, ctx);
 		if (unlikely(ret != 0)) {
 			ttm_dma_unpopulate(ttm_dma, dev);
 			return -ENOMEM;
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c
index 8ebc8d3..b48d7a0 100644
--- a/drivers/gpu/drm/ttm/ttm_tt.c
+++ b/drivers/gpu/drm/ttm/ttm_tt.c
@@ -263,6 +263,10 @@ void ttm_tt_unbind(struct ttm_tt *ttm)
 
 int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
 {
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	int ret = 0;
 
 	if (!ttm)
@@ -271,7 +275,7 @@ int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
 	if (ttm->state == tt_bound)
 		return 0;
 
-	ret = ttm->bdev->driver->ttm_tt_populate(ttm);
+	ret = ttm->bdev->driver->ttm_tt_populate(ttm, &ctx);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/virtio/virtgpu_object.c b/drivers/gpu/drm/virtio/virtgpu_object.c
index 6f66b73..0b90cdb 100644
--- a/drivers/gpu/drm/virtio/virtgpu_object.c
+++ b/drivers/gpu/drm/virtio/virtgpu_object.c
@@ -124,13 +124,17 @@ int virtio_gpu_object_get_sg_table(struct virtio_gpu_device *qdev,
 	int ret;
 	struct page **pages = bo->tbo.ttm->pages;
 	int nr_pages = bo->tbo.num_pages;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 
 	/* wtf swapping */
 	if (bo->pages)
 		return 0;
 
 	if (bo->tbo.ttm->state == tt_unpopulated)
-		bo->tbo.ttm->bdev->driver->ttm_tt_populate(bo->tbo.ttm);
+		bo->tbo.ttm->bdev->driver->ttm_tt_populate(bo->tbo.ttm, &ctx);
 	bo->pages = kmalloc(sizeof(struct sg_table), GFP_KERNEL);
 	if (!bo->pages)
 		goto out;
diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
index 488c6bd..72eb417 100644
--- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
+++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
@@ -324,12 +324,13 @@ static struct ttm_backend_func virtio_gpu_backend_func = {
 	.destroy = &virtio_gpu_ttm_backend_destroy,
 };
 
-static int virtio_gpu_ttm_tt_populate(struct ttm_tt *ttm)
+static int virtio_gpu_ttm_tt_populate(struct ttm_tt *ttm,
+					struct ttm_operation_ctx *ctx)
 {
 	if (ttm->state != tt_unpopulated)
 		return 0;
 
-	return ttm_pool_populate(ttm);
+	return ttm_pool_populate(ttm, ctx);
 }
 
 static void virtio_gpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index ef97542..90b0d6b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -635,16 +635,12 @@ static void vmw_ttm_destroy(struct ttm_tt *ttm)
 }
 
 
-static int vmw_ttm_populate(struct ttm_tt *ttm)
+static int vmw_ttm_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
 {
 	struct vmw_ttm_tt *vmw_tt =
 		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
 	struct vmw_private *dev_priv = vmw_tt->dev_priv;
 	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
-	struct ttm_operation_ctx ctx = {
-		.interruptible = true,
-		.no_wait_gpu = false
-	};
 	int ret;
 
 	if (ttm->state != tt_unpopulated)
@@ -653,15 +649,16 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
 	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
 		size_t size =
 			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
-		ret = ttm_mem_global_alloc(glob, size, &ctx);
+		ret = ttm_mem_global_alloc(glob, size, ctx);
 		if (unlikely(ret != 0))
 			return ret;
 
-		ret = ttm_dma_populate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
+		ret = ttm_dma_populate(&vmw_tt->dma_ttm, dev_priv->dev->dev,
+					ctx);
 		if (unlikely(ret != 0))
 			ttm_mem_global_free(glob, size);
 	} else
-		ret = ttm_pool_populate(ttm);
+		ret = ttm_pool_populate(ttm, ctx);
 
 	return ret;
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
index b17f08f..736ca47 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
@@ -240,6 +240,10 @@ static int vmw_otable_batch_setup(struct vmw_private *dev_priv,
 	unsigned long offset;
 	unsigned long bo_size;
 	struct vmw_otable *otables = batch->otables;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	SVGAOTableType i;
 	int ret;
 
@@ -264,7 +268,7 @@ static int vmw_otable_batch_setup(struct vmw_private *dev_priv,
 
 	ret = ttm_bo_reserve(batch->otable_bo, false, true, NULL);
 	BUG_ON(ret != 0);
-	ret = vmw_bo_driver.ttm_tt_populate(batch->otable_bo->ttm);
+	ret = vmw_bo_driver.ttm_tt_populate(batch->otable_bo->ttm, &ctx);
 	if (unlikely(ret != 0))
 		goto out_unreserve;
 	ret = vmw_bo_map_dma(batch->otable_bo);
@@ -430,6 +434,11 @@ static int vmw_mob_pt_populate(struct vmw_private *dev_priv,
 			       struct vmw_mob *mob)
 {
 	int ret;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
+
 	BUG_ON(mob->pt_bo != NULL);
 
 	ret = ttm_bo_create(&dev_priv->bdev, mob->num_pages * PAGE_SIZE,
@@ -442,7 +451,7 @@ static int vmw_mob_pt_populate(struct vmw_private *dev_priv,
 	ret = ttm_bo_reserve(mob->pt_bo, false, true, NULL);
 
 	BUG_ON(ret != 0);
-	ret = vmw_bo_driver.ttm_tt_populate(mob->pt_bo->ttm);
+	ret = vmw_bo_driver.ttm_tt_populate(mob->pt_bo->ttm, &ctx);
 	if (unlikely(ret != 0))
 		goto out_unreserve;
 	ret = vmw_bo_map_dma(mob->pt_bo);
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 934fecf..84860ec 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -352,7 +352,8 @@ struct ttm_bo_driver {
 	 * Returns:
 	 * -ENOMEM: Out of memory.
 	 */
-	int (*ttm_tt_populate)(struct ttm_tt *ttm);
+	int (*ttm_tt_populate)(struct ttm_tt *ttm,
+			struct ttm_operation_ctx *ctx);
 
 	/**
 	 * ttm_tt_unpopulate
@@ -1077,7 +1078,7 @@ struct ttm_tt *ttm_agp_tt_create(struct ttm_bo_device *bdev,
 				 struct agp_bridge_data *bridge,
 				 unsigned long size, uint32_t page_flags,
 				 struct page *dummy_read_page);
-int ttm_agp_tt_populate(struct ttm_tt *ttm);
+int ttm_agp_tt_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx);
 void ttm_agp_tt_unpopulate(struct ttm_tt *ttm);
 #endif
 
diff --git a/include/drm/ttm/ttm_page_alloc.h b/include/drm/ttm/ttm_page_alloc.h
index 5938113..f8395dd 100644
--- a/include/drm/ttm/ttm_page_alloc.h
+++ b/include/drm/ttm/ttm_page_alloc.h
@@ -47,7 +47,7 @@ void ttm_page_alloc_fini(void);
  *
  * Add backing pages to all of @ttm
  */
-int ttm_pool_populate(struct ttm_tt *ttm);
+int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx);
 
 /**
  * ttm_pool_unpopulate:
@@ -61,7 +61,8 @@ void ttm_pool_unpopulate(struct ttm_tt *ttm);
 /**
  * Populates and DMA maps pages to fullfil a ttm_dma_populate() request
  */
-int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt);
+int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt,
+				struct ttm_operation_ctx *ctx);
 
 /**
  * Unpopulates and DMA unmaps pages as part of a
@@ -89,7 +90,8 @@ void ttm_dma_page_alloc_fini(void);
  */
 int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data);
 
-int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev);
+int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
+			struct ttm_operation_ctx *ctx);
 void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev);
 
 #else
@@ -106,7 +108,8 @@ static inline int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data)
 	return 0;
 }
 static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma,
-				   struct device *dev)
+				   struct device *dev,
+				   struct ttm_operation_ctx *ctx)
 {
 	return -ENOMEM;
 }
-- 
2.7.4

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

[PATCH 5/7] drm/ttm: use an operation ctx for ttm_tt_bind

[Roger He]

Change-Id: I42a7df8c50e1ce3b527ee9cb78809f8e58136f07
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
 drivers/gpu/drm/nouveau/nouveau_bo.c    | 2 +-
 drivers/gpu/drm/radeon/radeon_ttm.c     | 2 +-
 drivers/gpu/drm/ttm/ttm_bo.c            | 2 +-
 drivers/gpu/drm/ttm/ttm_bo_util.c       | 2 +-
 drivers/gpu/drm/ttm/ttm_tt.c            | 9 +++------
 include/drm/ttm/ttm_bo_driver.h         | 3 ++-
 7 files changed, 10 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 52aab9d..ed235b7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -497,7 +497,7 @@ static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
 		goto out_cleanup;
 	}
 
-	r = ttm_tt_bind(bo->ttm, &tmp_mem);
+	r = ttm_tt_bind(bo->ttm, &tmp_mem, ctx);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index b141c27..a83ba6a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1218,7 +1218,7 @@ nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
 	if (ret)
 		return ret;
 
-	ret = ttm_tt_bind(bo->ttm, &tmp_reg);
+	ret = ttm_tt_bind(bo->ttm, &tmp_reg, &ctx);
 	if (ret)
 		goto out;
 
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 634fc6f..0f6fdc2 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -339,7 +339,7 @@ static int radeon_move_vram_ram(struct ttm_buffer_object *bo,
 		goto out_cleanup;
 	}
 
-	r = ttm_tt_bind(bo->ttm, &tmp_mem);
+	r = ttm_tt_bind(bo->ttm, &tmp_mem, &ctx);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index c59f572..e7595b4 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -300,7 +300,7 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 			goto out_err;
 
 		if (mem->mem_type != TTM_PL_SYSTEM) {
-			ret = ttm_tt_bind(bo->ttm, mem);
+			ret = ttm_tt_bind(bo->ttm, mem, ctx);
 			if (ret)
 				goto out_err;
 		}
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index b7eb507..153de1b 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -73,7 +73,7 @@ int ttm_bo_move_ttm(struct ttm_buffer_object *bo,
 		return ret;
 
 	if (new_mem->mem_type != TTM_PL_SYSTEM) {
-		ret = ttm_tt_bind(ttm, new_mem);
+		ret = ttm_tt_bind(ttm, new_mem, ctx);
 		if (unlikely(ret != 0))
 			return ret;
 	}
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c
index b48d7a0..5a046a3 100644
--- a/drivers/gpu/drm/ttm/ttm_tt.c
+++ b/drivers/gpu/drm/ttm/ttm_tt.c
@@ -261,12 +261,9 @@ void ttm_tt_unbind(struct ttm_tt *ttm)
 	}
 }
 
-int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
+int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem,
+		struct ttm_operation_ctx *ctx)
 {
-	struct ttm_operation_ctx ctx = {
-		.interruptible = false,
-		.no_wait_gpu = false
-	};
 	int ret = 0;
 
 	if (!ttm)
@@ -275,7 +272,7 @@ int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
 	if (ttm->state == tt_bound)
 		return 0;
 
-	ret = ttm->bdev->driver->ttm_tt_populate(ttm, &ctx);
+	ret = ttm->bdev->driver->ttm_tt_populate(ttm, ctx);
 	if (ret)
 		return ret;
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 84860ec..94064b1 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -650,7 +650,8 @@ void ttm_dma_tt_fini(struct ttm_dma_tt *ttm_dma);
  *
  * Bind the pages of @ttm to an aperture location identified by @bo_mem
  */
-int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem);
+int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem,
+		struct ttm_operation_ctx *ctx);
 
 /**
  * ttm_ttm_destroy:
-- 
2.7.4

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

[PATCH 6/7] drm/ttm: add ttm_bo_evict_swapout_allowable to check bo is allowable to evict or swapout

[Roger He]

extract this function since eviction and swapout share same logic

Change-Id: I80a475a93fceed8d66d74a1832c815a0756341ac
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c | 29 +++++++++++++++++++----------
 1 file changed, 19 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index e7595b4..313925c 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -706,6 +706,23 @@ bool ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
 }
 EXPORT_SYMBOL(ttm_bo_eviction_valuable);
 
+static bool ttm_bo_evict_swapout_allowable(struct ttm_buffer_object *bo,
+			struct ttm_operation_ctx *ctx, bool *locked)
+{
+	bool ret = false;
+
+	*locked = false;
+	if (bo->resv == ctx->resv) {
+		if (ctx->allow_reserved_eviction || !list_empty(&bo->ddestroy))
+			ret = true;
+	} else {
+		*locked = reservation_object_trylock(bo->resv);
+		ret = *locked;
+	}
+
+	return ret;
+}
+
 static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
 			       uint32_t mem_type,
 			       const struct ttm_place *place,
@@ -721,21 +738,13 @@ static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
 	spin_lock(&glob->lru_lock);
 	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
 		list_for_each_entry(bo, &man->lru[i], lru) {
-			if (bo->resv == ctx->resv) {
-				if (!ctx->allow_reserved_eviction &&
-				    list_empty(&bo->ddestroy))
-					continue;
-			} else {
-				locked = reservation_object_trylock(bo->resv);
-				if (!locked)
-					continue;
-			}
+			if (!ttm_bo_evict_swapout_allowable(bo, ctx, &locked))
+				continue;
 
 			if (place && !bdev->driver->eviction_valuable(bo,
 								      place)) {
 				if (locked)
 					reservation_object_unlock(bo->resv);
-				locked = false;
 				continue;
 			}
 			break;
-- 
2.7.4

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

[PATCH 7/7] drm/ttm: enable swapout of per VM BOs during allocation and allows reaping of deleted BOs

[Roger He]

Change-Id: I1e87954564f38ad298bf6e4ff88c9f26f291a62d
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c     | 15 +++++++++++----
 drivers/gpu/drm/ttm/ttm_memory.c | 12 ++++++++----
 include/drm/ttm/ttm_bo_api.h     |  3 ++-
 3 files changed, 21 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 313925c..ecb8916 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1686,18 +1686,20 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
  * A buffer object shrink method that tries to swap out the first
  * buffer object on the bo_global::swap_lru list.
  */
-int ttm_bo_swapout(struct ttm_bo_global *glob)
+int ttm_bo_swapout(struct ttm_bo_global *glob, struct ttm_operation_ctx *ctx)
 {
 	struct ttm_buffer_object *bo;
 	int ret = -EBUSY;
+	bool locked;
 	unsigned i;
 
 	spin_lock(&glob->lru_lock);
 	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
 		list_for_each_entry(bo, &glob->swap_lru[i], swap) {
-			ret = reservation_object_trylock(bo->resv) ? 0 : -EBUSY;
-			if (!ret)
+			if (ttm_bo_evict_swapout_allowable(bo, ctx, &locked)) {
+				ret = 0;
 				break;
+			}
 		}
 		if (!ret)
 			break;
@@ -1773,7 +1775,12 @@ EXPORT_SYMBOL(ttm_bo_swapout);
 
 void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
 {
-	while (ttm_bo_swapout(bdev->glob) == 0)
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
+
+	while (ttm_bo_swapout(bdev->glob, &ctx) == 0)
 		;
 }
 EXPORT_SYMBOL(ttm_bo_swapout_all);
diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
index 8df0755..8817b86 100644
--- a/drivers/gpu/drm/ttm/ttm_memory.c
+++ b/drivers/gpu/drm/ttm/ttm_memory.c
@@ -211,7 +211,7 @@ static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
  */
 
 static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
-		       uint64_t extra)
+			uint64_t extra, struct ttm_operation_ctx *ctx)
 {
 	int ret;
 
@@ -219,7 +219,7 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
 
 	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
 		spin_unlock(&glob->lock);
-		ret = ttm_bo_swapout(glob->bo_glob);
+		ret = ttm_bo_swapout(glob->bo_glob, ctx);
 		spin_lock(&glob->lock);
 		if (unlikely(ret != 0))
 			break;
@@ -230,10 +230,14 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
 
 static void ttm_shrink_work(struct work_struct *work)
 {
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	struct ttm_mem_global *glob =
 	    container_of(work, struct ttm_mem_global, work);
 
-	ttm_shrink(glob, true, 0ULL);
+	ttm_shrink(glob, true, 0ULL, &ctx);
 }
 
 static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
@@ -520,7 +524,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
 			return -ENOMEM;
 		if (unlikely(count-- == 0))
 			return -ENOMEM;
-		ttm_shrink(glob, false, memory + (memory >> 2) + 16);
+		ttm_shrink(glob, false, memory + (memory >> 2) + 16, ctx);
 	}
 
 	return 0;
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 24a8db7..f1c74c2 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -752,7 +752,8 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
 		  const char __user *wbuf, char __user *rbuf,
 		  size_t count, loff_t *f_pos, bool write);
 
-int ttm_bo_swapout(struct ttm_bo_global *glob);
+int ttm_bo_swapout(struct ttm_bo_global *glob,
+			struct ttm_operation_ctx *ctx);
 void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
 int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
 #endif
-- 
2.7.4

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

Re: [PATCH 1/7] drm/ttm: call ttm_bo_swapout directly when ttm shrink

[Christian König]

Am 20.12.2017 um 11:34 schrieb Roger He:
> then remove superfluous functions

We need a better commit message. Something like:

Remove the extra indirection, cause we have only one implementation anyway.

>
> Change-Id: Iea020f0e30a239e0265e7a1500168c7d7f819bd9
> Signed-off-by: Roger He <Hongbo.He@amd.com>

With the commit message fixed Reviewed-by: Christian König 
<christian.koenig@amd.com>.

Regards,
Christian.

> ---
>   drivers/gpu/drm/ttm/ttm_bo.c     | 21 +++---------
>   drivers/gpu/drm/ttm/ttm_memory.c | 12 ++-----
>   include/drm/ttm/ttm_bo_api.h     |  1 +
>   include/drm/ttm/ttm_bo_driver.h  |  1 -
>   include/drm/ttm/ttm_memory.h     | 69 +---------------------------------------
>   5 files changed, 9 insertions(+), 95 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 60bb5c1..fa57aa8 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -42,7 +42,6 @@
>   #include <linux/atomic.h>
>   #include <linux/reservation.h>
>   
> -static int ttm_bo_swapout(struct ttm_mem_shrink *shrink);
>   static void ttm_bo_global_kobj_release(struct kobject *kobj);
>   
>   static struct attribute ttm_bo_count = {
> @@ -1454,7 +1453,6 @@ static void ttm_bo_global_kobj_release(struct kobject *kobj)
>   	struct ttm_bo_global *glob =
>   		container_of(kobj, struct ttm_bo_global, kobj);
>   
> -	ttm_mem_unregister_shrink(glob->mem_glob, &glob->shrink);
>   	__free_page(glob->dummy_read_page);
>   	kfree(glob);
>   }
> @@ -1479,6 +1477,7 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	mutex_init(&glob->device_list_mutex);
>   	spin_lock_init(&glob->lru_lock);
>   	glob->mem_glob = bo_ref->mem_glob;
> +	glob->mem_glob->bo_glob = glob;
>   	glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32);
>   
>   	if (unlikely(glob->dummy_read_page == NULL)) {
> @@ -1489,14 +1488,6 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	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);
> -	ret = ttm_mem_register_shrink(glob->mem_glob, &glob->shrink);
> -	if (unlikely(ret != 0)) {
> -		pr_err("Could not register buffer object swapout\n");
> -		goto out_no_shrink;
> -	}
> -
>   	atomic_set(&glob->bo_count, 0);
>   
>   	ret = kobject_init_and_add(
> @@ -1504,8 +1495,6 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	if (unlikely(ret != 0))
>   		kobject_put(&glob->kobj);
>   	return ret;
> -out_no_shrink:
> -	__free_page(glob->dummy_read_page);
>   out_no_drp:
>   	kfree(glob);
>   	return ret;
> @@ -1688,11 +1677,8 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
>    * A buffer object shrink method that tries to swap out the first
>    * buffer object on the bo_global::swap_lru list.
>    */
> -
> -static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
> +int ttm_bo_swapout(struct ttm_bo_global *glob)
>   {
> -	struct ttm_bo_global *glob =
> -	    container_of(shrink, struct ttm_bo_global, shrink);
>   	struct ttm_buffer_object *bo;
>   	int ret = -EBUSY;
>   	unsigned i;
> @@ -1774,10 +1760,11 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
>   	kref_put(&bo->list_kref, ttm_bo_release_list);
>   	return ret;
>   }
> +EXPORT_SYMBOL(ttm_bo_swapout);
>   
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
>   {
> -	while (ttm_bo_swapout(&bdev->glob->shrink) == 0)
> +	while (ttm_bo_swapout(bdev->glob) == 0)
>   		;
>   }
>   EXPORT_SYMBOL(ttm_bo_swapout_all);
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
> index e963749..9130bdf 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -214,26 +214,20 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
>   		       uint64_t extra)
>   {
>   	int ret;
> -	struct ttm_mem_shrink *shrink;
>   
>   	spin_lock(&glob->lock);
> -	if (glob->shrink == NULL)
> -		goto out;
>   
>   	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
> -		shrink = glob->shrink;
>   		spin_unlock(&glob->lock);
> -		ret = shrink->do_shrink(shrink);
> +		ret = ttm_bo_swapout(glob->bo_glob);
>   		spin_lock(&glob->lock);
>   		if (unlikely(ret != 0))
> -			goto out;
> +			break;
>   	}
> -out:
> +
>   	spin_unlock(&glob->lock);
>   }
>   
> -
> -
>   static void ttm_shrink_work(struct work_struct *work)
>   {
>   	struct ttm_mem_global *glob =
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index c126330..24a8db7 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -752,6 +752,7 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
>   		  const char __user *wbuf, char __user *rbuf,
>   		  size_t count, loff_t *f_pos, bool write);
>   
> +int ttm_bo_swapout(struct ttm_bo_global *glob);
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
>   int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
>   #endif
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index 5115718..934fecf 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -522,7 +522,6 @@ struct ttm_bo_global {
>   	struct kobject kobj;
>   	struct ttm_mem_global *mem_glob;
>   	struct page *dummy_read_page;
> -	struct ttm_mem_shrink shrink;
>   	struct mutex device_list_mutex;
>   	spinlock_t lru_lock;
>   
> diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
> index 2c1e359..85f3ad6 100644
> --- a/include/drm/ttm/ttm_memory.h
> +++ b/include/drm/ttm/ttm_memory.h
> @@ -37,20 +37,6 @@
>   #include <linux/mm.h>
>   
>   /**
> - * struct ttm_mem_shrink - callback to shrink TTM memory usage.
> - *
> - * @do_shrink: The callback function.
> - *
> - * Arguments to the do_shrink functions are intended to be passed using
> - * inheritance. That is, the argument class derives from struct ttm_mem_shrink,
> - * and can be accessed using container_of().
> - */
> -
> -struct ttm_mem_shrink {
> -	int (*do_shrink) (struct ttm_mem_shrink *);
> -};
> -
> -/**
>    * struct ttm_mem_global - Global memory accounting structure.
>    *
>    * @shrink: A single callback to shrink TTM memory usage. Extend this
> @@ -76,7 +62,7 @@ struct ttm_mem_shrink {
>   struct ttm_mem_zone;
>   struct ttm_mem_global {
>   	struct kobject kobj;
> -	struct ttm_mem_shrink *shrink;
> +	struct ttm_bo_global *bo_glob;
>   	struct workqueue_struct *swap_queue;
>   	struct work_struct work;
>   	spinlock_t lock;
> @@ -90,59 +76,6 @@ struct ttm_mem_global {
>   #endif
>   };
>   
> -/**
> - * ttm_mem_init_shrink - initialize a struct ttm_mem_shrink object
> - *
> - * @shrink: The object to initialize.
> - * @func: The callback function.
> - */
> -
> -static inline void ttm_mem_init_shrink(struct ttm_mem_shrink *shrink,
> -				       int (*func) (struct ttm_mem_shrink *))
> -{
> -	shrink->do_shrink = func;
> -}
> -
> -/**
> - * ttm_mem_register_shrink - register a struct ttm_mem_shrink object.
> - *
> - * @glob: The struct ttm_mem_global object to register with.
> - * @shrink: An initialized struct ttm_mem_shrink object to register.
> - *
> - * Returns:
> - * -EBUSY: There's already a callback registered. (May change).
> - */
> -
> -static inline int ttm_mem_register_shrink(struct ttm_mem_global *glob,
> -					  struct ttm_mem_shrink *shrink)
> -{
> -	spin_lock(&glob->lock);
> -	if (glob->shrink != NULL) {
> -		spin_unlock(&glob->lock);
> -		return -EBUSY;
> -	}
> -	glob->shrink = shrink;
> -	spin_unlock(&glob->lock);
> -	return 0;
> -}
> -
> -/**
> - * ttm_mem_unregister_shrink - unregister a struct ttm_mem_shrink object.
> - *
> - * @glob: The struct ttm_mem_global object to unregister from.
> - * @shrink: A previously registert struct ttm_mem_shrink object.
> - *
> - */
> -
> -static inline void ttm_mem_unregister_shrink(struct ttm_mem_global *glob,
> -					     struct ttm_mem_shrink *shrink)
> -{
> -	spin_lock(&glob->lock);
> -	BUG_ON(glob->shrink != shrink);
> -	glob->shrink = NULL;
> -	spin_unlock(&glob->lock);
> -}
> -
>   extern int ttm_mem_global_init(struct ttm_mem_global *glob);
>   extern void ttm_mem_global_release(struct ttm_mem_global *glob);
>   extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,

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

Re: [PATCH 2/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc

[Christian König]

Commit message needed! Something like:

Forward the operation context to ttm_mem_global_alloc as well.

Am 20.12.2017 um 11:34 schrieb Roger He:
> Change-Id: I5279b5cd3560c4082b00f822219575a5f9c3808a
> Signed-off-by: Roger He <Hongbo.He@amd.com>

With the commit message fixed, patch is Reviewed-by: Christian König 
<christian.koenig@amd.com>.

I would like to get an rb or ab from Thomas and/or Sinclair as well, 
since this is touching a lot of vmwgfx code.

Regards,
Christian.

> ---
>   drivers/gpu/drm/ttm/ttm_bo.c                    |  2 +-
>   drivers/gpu/drm/ttm/ttm_memory.c                | 15 +++++++++------
>   drivers/gpu/drm/ttm/ttm_object.c                | 13 ++++++++++---
>   drivers/gpu/drm/vmwgfx/vmwgfx_binding.c         |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c          | 13 ++++++++++---
>   drivers/gpu/drm/vmwgfx/vmwgfx_context.c         |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c         |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_fence.c           |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_shader.c          | 18 +++++++++++++++---
>   drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_so.c              |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_surface.c         | 12 ++++++++++--
>   include/drm/ttm/ttm_memory.h                    |  3 ++-
>   13 files changed, 87 insertions(+), 25 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index fa57aa8..c59f572 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -1133,7 +1133,7 @@ int ttm_bo_init_reserved(struct ttm_bo_device *bdev,
>   	struct ttm_mem_global *mem_glob = bdev->glob->mem_glob;
>   	bool locked;
>   
> -	ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false);
> +	ret = ttm_mem_global_alloc(mem_glob, acc_size, ctx);
>   	if (ret) {
>   		pr_err("Out of kernel memory\n");
>   		if (destroy)
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
> index 9130bdf..525d3b6 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -508,7 +508,7 @@ static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
>   static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>   				     struct ttm_mem_zone *single_zone,
>   				     uint64_t memory,
> -				     bool no_wait, bool interruptible)
> +				     struct ttm_operation_ctx *ctx)
>   {
>   	int count = TTM_MEMORY_ALLOC_RETRIES;
>   
> @@ -516,7 +516,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>   					       single_zone,
>   					       memory, true)
>   			!= 0)) {
> -		if (no_wait)
> +		if (ctx->no_wait_gpu)
>   			return -ENOMEM;
>   		if (unlikely(count-- == 0))
>   			return -ENOMEM;
> @@ -527,15 +527,14 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>   }
>   
>   int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
> -			 bool no_wait, bool interruptible)
> +			 struct ttm_operation_ctx *ctx)
>   {
>   	/**
>   	 * Normal allocations of kernel memory are registered in
>   	 * all zones.
>   	 */
>   
> -	return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
> -					 interruptible);
> +	return ttm_mem_global_alloc_zone(glob, NULL, memory, ctx);
>   }
>   EXPORT_SYMBOL(ttm_mem_global_alloc);
>   
> @@ -544,6 +543,10 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>   {
>   
>   	struct ttm_mem_zone *zone = NULL;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   
>   	/**
>   	 * Page allocations may be registed in a single zone
> @@ -557,7 +560,7 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>   	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
>   		zone = glob->zone_kernel;
>   #endif
> -	return ttm_mem_global_alloc_zone(glob, zone, size, false, false);
> +	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
>   }
>   
>   void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
> diff --git a/drivers/gpu/drm/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c
> index 26a7ad0..1aa2baa 100644
> --- a/drivers/gpu/drm/ttm/ttm_object.c
> +++ b/drivers/gpu/drm/ttm/ttm_object.c
> @@ -325,6 +325,10 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
>   	struct ttm_ref_object *ref;
>   	struct drm_hash_item *hash;
>   	struct ttm_mem_global *mem_glob = tfile->tdev->mem_glob;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	int ret = -EINVAL;
>   
>   	if (base->tfile != tfile && !base->shareable)
> @@ -350,7 +354,7 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
>   			return -EPERM;
>   
>   		ret = ttm_mem_global_alloc(mem_glob, sizeof(*ref),
> -					   false, false);
> +					   &ctx);
>   		if (unlikely(ret != 0))
>   			return ret;
>   		ref = kmalloc(sizeof(*ref), GFP_KERNEL);
> @@ -686,7 +690,10 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
>   	dma_buf = prime->dma_buf;
>   	if (!dma_buf || !get_dma_buf_unless_doomed(dma_buf)) {
>   		DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
> -
> +		struct ttm_operation_ctx ctx = {
> +			.interruptible = true,
> +			.no_wait_gpu = false
> +		};
>   		exp_info.ops = &tdev->ops;
>   		exp_info.size = prime->size;
>   		exp_info.flags = flags;
> @@ -696,7 +703,7 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
>   		 * Need to create a new dma_buf, with memory accounting.
>   		 */
>   		ret = ttm_mem_global_alloc(tdev->mem_glob, tdev->dma_buf_size,
> -					   false, true);
> +					   &ctx);
>   		if (unlikely(ret != 0)) {
>   			mutex_unlock(&prime->mutex);
>   			goto out_unref;
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> index 9c42e96..55d32ae 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> @@ -1202,10 +1202,14 @@ struct vmw_ctx_binding_state *
>   vmw_binding_state_alloc(struct vmw_private *dev_priv)
>   {
>   	struct vmw_ctx_binding_state *cbs;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), sizeof(*cbs),
> -				   false, false);
> +				&ctx);
>   	if (ret)
>   		return ERR_PTR(ret);
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index c705632..ef97542 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -394,6 +394,10 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
>   	struct vmw_private *dev_priv = vmw_tt->dev_priv;
>   	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
>   	struct vmw_sg_table *vsgt = &vmw_tt->vsgt;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	struct vmw_piter iter;
>   	dma_addr_t old;
>   	int ret = 0;
> @@ -417,8 +421,7 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
>   			sgt_size = ttm_round_pot(sizeof(struct sg_table));
>   		}
>   		vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages;
> -		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false,
> -					   true);
> +		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, &ctx);
>   		if (unlikely(ret != 0))
>   			return ret;
>   
> @@ -638,6 +641,10 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
>   		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
>   	struct vmw_private *dev_priv = vmw_tt->dev_priv;
>   	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	if (ttm->state != tt_unpopulated)
> @@ -646,7 +653,7 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
>   	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
>   		size_t size =
>   			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
> -		ret = ttm_mem_global_alloc(glob, size, false, true);
> +		ret = ttm_mem_global_alloc(glob, size, &ctx);
>   		if (unlikely(ret != 0))
>   			return ret;
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> index 4212b3e..3767ac3 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> @@ -746,6 +746,10 @@ static int vmw_context_define(struct drm_device *dev, void *data,
>   	struct vmw_resource *tmp;
>   	struct drm_vmw_context_arg *arg = (struct drm_vmw_context_arg *)data;
>   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	if (!dev_priv->has_dx && dx) {
> @@ -768,7 +772,7 @@ static int vmw_context_define(struct drm_device *dev, void *data,
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>   				   vmw_user_context_size,
> -				   false, true);
> +				   &ttm_opt_ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for context"
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> index 92df0b0..cbf54ea 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> @@ -573,6 +573,10 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
>   				       u32 type)
>   {
>   	struct vmw_cotable *vcotbl;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   	u32 num_entries;
>   
> @@ -580,7 +584,7 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
>   		cotable_acc_size = ttm_round_pot(sizeof(struct vmw_cotable));
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> -				   cotable_acc_size, false, true);
> +				   cotable_acc_size, &ttm_opt_ctx);
>   	if (unlikely(ret))
>   		return ERR_PTR(ret);
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> index d6b1c50..6c5c75c 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> @@ -588,6 +588,10 @@ int vmw_user_fence_create(struct drm_file *file_priv,
>   	struct vmw_user_fence *ufence;
>   	struct vmw_fence_obj *tmp;
>   	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	/*
> @@ -596,7 +600,7 @@ int vmw_user_fence_create(struct drm_file *file_priv,
>   	 */
>   
>   	ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size,
> -				   false, false);
> +				   &ctx);
>   	if (unlikely(ret != 0))
>   		return ret;
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> index 004e18b..73b8e9a 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> @@ -607,6 +607,10 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
>   	struct vmw_dx_shader *shader;
>   	struct vmw_resource *res;
>   	struct vmw_private *dev_priv = ctx->dev_priv;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	if (!vmw_shader_dx_size)
> @@ -616,7 +620,7 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
>   		return -EINVAL;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), vmw_shader_dx_size,
> -				   false, true);
> +				   &ttm_opt_ctx);
>   	if (ret) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for shader "
> @@ -730,6 +734,10 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
>   {
>   	struct vmw_user_shader *ushader;
>   	struct vmw_resource *res, *tmp;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	/*
> @@ -742,7 +750,7 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>   				   vmw_user_shader_size,
> -				   false, true);
> +				   &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for shader "
> @@ -800,6 +808,10 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
>   {
>   	struct vmw_shader *shader;
>   	struct vmw_resource *res;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	/*
> @@ -812,7 +824,7 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>   				   vmw_shader_size,
> -				   false, true);
> +				   &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for shader "
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> index 051d3b3..a0cb310 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> @@ -149,6 +149,10 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
>   	struct vmw_resource *res;
>   	struct vmw_resource *tmp;
>   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	size_t alloc_size;
>   	size_t account_size;
>   	int ret;
> @@ -162,7 +166,7 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
>   		return ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), account_size,
> -				   false, true);
> +				   &ctx);
>   	ttm_read_unlock(&dev_priv->reservation_sem);
>   	if (ret) {
>   		if (ret != -ERESTARTSYS)
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> index 5a73eeb..d3573c3 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> @@ -329,6 +329,10 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
>   	struct vmw_private *dev_priv = ctx->dev_priv;
>   	struct vmw_resource *res;
>   	struct vmw_view *view;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	size_t size;
>   	int ret;
>   
> @@ -345,7 +349,7 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
>   
>   	size = offsetof(struct vmw_view, cmd) + cmd_size;
>   
> -	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, false, true);
> +	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, &ttm_opt_ctx);
>   	if (ret) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for view"
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> index 6ac094e..db1bb16 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> @@ -700,6 +700,10 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
>   	struct drm_vmw_surface_create_req *req = &arg->req;
>   	struct drm_vmw_surface_arg *rep = &arg->rep;
>   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   	int i, j;
>   	uint32_t cur_bo_offset;
> @@ -741,7 +745,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
>   		return ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> -				   size, false, true);
> +				   size, &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for surface"
> @@ -1479,6 +1483,10 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
>   {
>   	struct vmw_private *dev_priv = vmw_priv(dev);
>   	struct vmw_user_surface *user_srf;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	struct vmw_surface *srf;
>   	int ret;
>   	u32 num_layers;
> @@ -1525,7 +1533,7 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
>   		return ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> -				   user_accounting_size, false, true);
> +				   user_accounting_size, &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for surface"
> diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
> index 85f3ad6..755c107 100644
> --- a/include/drm/ttm/ttm_memory.h
> +++ b/include/drm/ttm/ttm_memory.h
> @@ -35,6 +35,7 @@
>   #include <linux/errno.h>
>   #include <linux/kobject.h>
>   #include <linux/mm.h>
> +#include "ttm_bo_api.h"
>   
>   /**
>    * struct ttm_mem_global - Global memory accounting structure.
> @@ -79,7 +80,7 @@ struct ttm_mem_global {
>   extern int ttm_mem_global_init(struct ttm_mem_global *glob);
>   extern void ttm_mem_global_release(struct ttm_mem_global *glob);
>   extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
> -				bool no_wait, bool interruptible);
> +				struct ttm_operation_ctx *ctx);
>   extern void ttm_mem_global_free(struct ttm_mem_global *glob,
>   				uint64_t amount);
>   extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,

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

Re: [PATCH 3/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc_page

[Christian König]

Commit message!

Am 20.12.2017 um 11:34 schrieb Roger He:
> Change-Id: I4104a12e09a374b6477a0dd5a8fce26dce27a746
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_memory.c         | 15 ++++++++-------
>   drivers/gpu/drm/ttm/ttm_page_alloc.c     |  6 +++++-
>   drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 ++++++--
>   include/drm/ttm/ttm_memory.h             |  3 ++-
>   4 files changed, 21 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
> index 525d3b6..8df0755 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -539,15 +539,14 @@ int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
>   EXPORT_SYMBOL(ttm_mem_global_alloc);
>   
>   int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
> -			      struct page *page, uint64_t size)
> +			      struct page *page, uint64_t size,
> +			      struct ttm_operation_ctx *ctx)
>   {
> -
> +	int ret;
>   	struct ttm_mem_zone *zone = NULL;
> -	struct ttm_operation_ctx ctx = {
> -		.interruptible = false,
> -		.no_wait_gpu = false
> -	};
> +	bool tmp_no_wait_gpu = ctx->no_wait_gpu;

Mhm, please drop that. That the function might wait for the GPU even 
when the caller requested not to do so sounds like a bug to me.

Christian.

>   
> +	ctx->no_wait_gpu = false;
>   	/**
>   	 * Page allocations may be registed in a single zone
>   	 * only if highmem or !dma32.
> @@ -560,7 +559,9 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>   	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
>   		zone = glob->zone_kernel;
>   #endif
> -	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
> +	ret = ttm_mem_global_alloc_zone(glob, zone, size, ctx);
> +	ctx->no_wait_gpu = tmp_no_wait_gpu;
> +	return ret;
>   }
>   
>   void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index b5ba644..8f93ff3 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -1061,6 +1061,10 @@ void ttm_page_alloc_fini(void)
>   int ttm_pool_populate(struct ttm_tt *ttm)
>   {
>   	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	unsigned i;
>   	int ret;
>   
> @@ -1076,7 +1080,7 @@ int ttm_pool_populate(struct ttm_tt *ttm)
>   
>   	for (i = 0; i < ttm->num_pages; ++i) {
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						PAGE_SIZE);
> +						PAGE_SIZE, &ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_pool_unpopulate(ttm);
>   			return -ENOMEM;
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> index bda00b2..8aac86a 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> @@ -927,6 +927,10 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   {
>   	struct ttm_tt *ttm = &ttm_dma->ttm;
>   	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	unsigned long num_pages = ttm->num_pages;
>   	struct dma_pool *pool;
>   	enum pool_type type;
> @@ -962,7 +966,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   			break;
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						pool->size);
> +						pool->size, &ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_dma_unpopulate(ttm_dma, dev);
>   			return -ENOMEM;
> @@ -998,7 +1002,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   		}
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						pool->size);
> +						pool->size, &ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_dma_unpopulate(ttm_dma, dev);
>   			return -ENOMEM;
> diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
> index 755c107..8936285 100644
> --- a/include/drm/ttm/ttm_memory.h
> +++ b/include/drm/ttm/ttm_memory.h
> @@ -84,7 +84,8 @@ extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
>   extern void ttm_mem_global_free(struct ttm_mem_global *glob,
>   				uint64_t amount);
>   extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
> -				     struct page *page, uint64_t size);
> +				     struct page *page, uint64_t size,
> +				     struct ttm_operation_ctx *ctx);
>   extern void ttm_mem_global_free_page(struct ttm_mem_global *glob,
>   				     struct page *page, uint64_t size);
>   extern size_t ttm_round_pot(size_t size);

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

Re: [PATCH 4/7] drm/ttm: use an operation ctx for ttm_tt_populate in ttm_bo_driver

[Christian König]

Am 20.12.2017 um 11:34 schrieb Roger He:
> Change-Id: I803ea52d11e5c06add0dffab836c3aecc00b56dd
> Signed-off-by: Roger He <Hongbo.He@amd.com>

Commit message! And please double check the coding style of 
ast_ttm_tt_populate.

With that fixed that patch is Reviewed-by: Christian König 
<christian.koenig@amd.com>.

Regards,
Christian.

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  7 ++++---
>   drivers/gpu/drm/ast/ast_ttm.c            |  5 +++--
>   drivers/gpu/drm/cirrus/cirrus_ttm.c      |  5 +++--
>   drivers/gpu/drm/nouveau/nouveau_bo.c     |  8 ++++----
>   drivers/gpu/drm/qxl/qxl_ttm.c            |  5 +++--
>   drivers/gpu/drm/radeon/radeon_ttm.c      |  9 +++++----
>   drivers/gpu/drm/ttm/ttm_agp_backend.c    |  4 ++--
>   drivers/gpu/drm/ttm/ttm_bo_util.c        | 11 ++++++++---
>   drivers/gpu/drm/ttm/ttm_bo_vm.c          |  7 ++++++-
>   drivers/gpu/drm/ttm/ttm_page_alloc.c     | 13 +++++--------
>   drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 11 ++++-------
>   drivers/gpu/drm/ttm/ttm_tt.c             |  6 +++++-
>   drivers/gpu/drm/virtio/virtgpu_object.c  |  6 +++++-
>   drivers/gpu/drm/virtio/virtgpu_ttm.c     |  5 +++--
>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c   | 13 +++++--------
>   drivers/gpu/drm/vmwgfx/vmwgfx_mob.c      | 13 +++++++++++--
>   include/drm/ttm/ttm_bo_driver.h          |  5 +++--
>   include/drm/ttm/ttm_page_alloc.h         | 11 +++++++----
>   18 files changed, 86 insertions(+), 58 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index f1b7d98..52aab9d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -990,7 +990,8 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_bo_device *bdev,
>   	return &gtt->ttm.ttm;
>   }
>   
> -static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm)
> +static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
> +					struct ttm_operation_ctx *ctx)
>   {
>   	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
>   	struct amdgpu_ttm_tt *gtt = (void *)ttm;
> @@ -1018,11 +1019,11 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm)
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		return ttm_dma_populate(&gtt->ttm, adev->dev);
> +		return ttm_dma_populate(&gtt->ttm, adev->dev, ctx);
>   	}
>   #endif
>   
> -	return ttm_populate_and_map_pages(adev->dev, &gtt->ttm);
> +	return ttm_populate_and_map_pages(adev->dev, &gtt->ttm, ctx);
>   }
>   
>   static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
> diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c
> index 28da7c2..1413e94 100644
> --- a/drivers/gpu/drm/ast/ast_ttm.c
> +++ b/drivers/gpu/drm/ast/ast_ttm.c
> @@ -216,9 +216,10 @@ static struct ttm_tt *ast_ttm_tt_create(struct ttm_bo_device *bdev,
>   	return tt;
>   }
>   
> -static int ast_ttm_tt_populate(struct ttm_tt *ttm)
> +static int ast_ttm_tt_populate(struct ttm_tt *ttm,
> +			struct ttm_operation_ctx *ctx)
>   {
> -	return ttm_pool_populate(ttm);
> +	return ttm_pool_populate(ttm, ctx);
>   }
>   
>   static void ast_ttm_tt_unpopulate(struct ttm_tt *ttm)
> diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c
> index 2a5b54d..95e2d40 100644
> --- a/drivers/gpu/drm/cirrus/cirrus_ttm.c
> +++ b/drivers/gpu/drm/cirrus/cirrus_ttm.c
> @@ -216,9 +216,10 @@ static struct ttm_tt *cirrus_ttm_tt_create(struct ttm_bo_device *bdev,
>   	return tt;
>   }
>   
> -static int cirrus_ttm_tt_populate(struct ttm_tt *ttm)
> +static int cirrus_ttm_tt_populate(struct ttm_tt *ttm,
> +				struct ttm_operation_ctx *ctx)
>   {
> -	return ttm_pool_populate(ttm);
> +	return ttm_pool_populate(ttm, ctx);
>   }
>   
>   static void cirrus_ttm_tt_unpopulate(struct ttm_tt *ttm)
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index 6b6fb20..b141c27 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1547,7 +1547,7 @@ nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
>   }
>   
>   static int
> -nouveau_ttm_tt_populate(struct ttm_tt *ttm)
> +nouveau_ttm_tt_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   {
>   	struct ttm_dma_tt *ttm_dma = (void *)ttm;
>   	struct nouveau_drm *drm;
> @@ -1572,17 +1572,17 @@ nouveau_ttm_tt_populate(struct ttm_tt *ttm)
>   
>   #if IS_ENABLED(CONFIG_AGP)
>   	if (drm->agp.bridge) {
> -		return ttm_agp_tt_populate(ttm);
> +		return ttm_agp_tt_populate(ttm, ctx);
>   	}
>   #endif
>   
>   #if IS_ENABLED(CONFIG_SWIOTLB) && IS_ENABLED(CONFIG_X86)
>   	if (swiotlb_nr_tbl()) {
> -		return ttm_dma_populate((void *)ttm, dev);
> +		return ttm_dma_populate((void *)ttm, dev, ctx);
>   	}
>   #endif
>   
> -	r = ttm_pool_populate(ttm);
> +	r = ttm_pool_populate(ttm, ctx);
>   	if (r) {
>   		return r;
>   	}
> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
> index 78ce118..989645c 100644
> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
> @@ -291,14 +291,15 @@ static struct ttm_backend_func qxl_backend_func = {
>   	.destroy = &qxl_ttm_backend_destroy,
>   };
>   
> -static int qxl_ttm_tt_populate(struct ttm_tt *ttm)
> +static int qxl_ttm_tt_populate(struct ttm_tt *ttm,
> +				struct ttm_operation_ctx *ctx)
>   {
>   	int r;
>   
>   	if (ttm->state != tt_unpopulated)
>   		return 0;
>   
> -	r = ttm_pool_populate(ttm);
> +	r = ttm_pool_populate(ttm, ctx);
>   	if (r)
>   		return r;
>   
> diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
> index 557fd79..634fc6f 100644
> --- a/drivers/gpu/drm/radeon/radeon_ttm.c
> +++ b/drivers/gpu/drm/radeon/radeon_ttm.c
> @@ -721,7 +721,8 @@ static struct radeon_ttm_tt *radeon_ttm_tt_to_gtt(struct ttm_tt *ttm)
>   	return (struct radeon_ttm_tt *)ttm;
>   }
>   
> -static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
> +static int radeon_ttm_tt_populate(struct ttm_tt *ttm,
> +				struct ttm_operation_ctx *ctx)
>   {
>   	struct radeon_ttm_tt *gtt = radeon_ttm_tt_to_gtt(ttm);
>   	struct radeon_device *rdev;
> @@ -750,17 +751,17 @@ static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
>   	rdev = radeon_get_rdev(ttm->bdev);
>   #if IS_ENABLED(CONFIG_AGP)
>   	if (rdev->flags & RADEON_IS_AGP) {
> -		return ttm_agp_tt_populate(ttm);
> +		return ttm_agp_tt_populate(ttm, ctx);
>   	}
>   #endif
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		return ttm_dma_populate(&gtt->ttm, rdev->dev);
> +		return ttm_dma_populate(&gtt->ttm, rdev->dev, ctx);
>   	}
>   #endif
>   
> -	return ttm_populate_and_map_pages(rdev->dev, &gtt->ttm);
> +	return ttm_populate_and_map_pages(rdev->dev, &gtt->ttm, ctx);
>   }
>   
>   static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
> diff --git a/drivers/gpu/drm/ttm/ttm_agp_backend.c b/drivers/gpu/drm/ttm/ttm_agp_backend.c
> index 028ab60..3e795a0 100644
> --- a/drivers/gpu/drm/ttm/ttm_agp_backend.c
> +++ b/drivers/gpu/drm/ttm/ttm_agp_backend.c
> @@ -133,12 +133,12 @@ struct ttm_tt *ttm_agp_tt_create(struct ttm_bo_device *bdev,
>   }
>   EXPORT_SYMBOL(ttm_agp_tt_create);
>   
> -int ttm_agp_tt_populate(struct ttm_tt *ttm)
> +int ttm_agp_tt_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   {
>   	if (ttm->state != tt_unpopulated)
>   		return 0;
>   
> -	return ttm_pool_populate(ttm);
> +	return ttm_pool_populate(ttm, ctx);
>   }
>   EXPORT_SYMBOL(ttm_agp_tt_populate);
>   
> diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
> index 6e353df..b7eb507 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo_util.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
> @@ -376,7 +376,7 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
>   	 * TTM might be null for moves within the same region.
>   	 */
>   	if (ttm && ttm->state == tt_unpopulated) {
> -		ret = ttm->bdev->driver->ttm_tt_populate(ttm);
> +		ret = ttm->bdev->driver->ttm_tt_populate(ttm, ctx);
>   		if (ret)
>   			goto out1;
>   	}
> @@ -545,14 +545,19 @@ static int ttm_bo_kmap_ttm(struct ttm_buffer_object *bo,
>   			   unsigned long num_pages,
>   			   struct ttm_bo_kmap_obj *map)
>   {
> -	struct ttm_mem_reg *mem = &bo->mem; pgprot_t prot;
> +	struct ttm_mem_reg *mem = &bo->mem;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	struct ttm_tt *ttm = bo->ttm;
> +	pgprot_t prot;
>   	int ret;
>   
>   	BUG_ON(!ttm);
>   
>   	if (ttm->state == tt_unpopulated) {
> -		ret = ttm->bdev->driver->ttm_tt_populate(ttm);
> +		ret = ttm->bdev->driver->ttm_tt_populate(ttm, &ctx);
>   		if (ret)
>   			return ret;
>   	}
> diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
> index c8ebb75..65dfcdd 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
> @@ -215,12 +215,17 @@ static int ttm_bo_vm_fault(struct vm_fault *vmf)
>   		cvma.vm_page_prot = ttm_io_prot(bo->mem.placement,
>   						cvma.vm_page_prot);
>   	} else {
> +		struct ttm_operation_ctx ctx = {
> +			.interruptible = false,
> +			.no_wait_gpu = false
> +		};
> +
>   		ttm = bo->ttm;
>   		cvma.vm_page_prot = ttm_io_prot(bo->mem.placement,
>   						cvma.vm_page_prot);
>   
>   		/* Allocate all page at once, most common usage */
> -		if (ttm->bdev->driver->ttm_tt_populate(ttm)) {
> +		if (ttm->bdev->driver->ttm_tt_populate(ttm, &ctx)) {
>   			retval = VM_FAULT_OOM;
>   			goto out_io_unlock;
>   		}
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 8f93ff3..f1a3d55 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -1058,13 +1058,9 @@ void ttm_page_alloc_fini(void)
>   	_manager = NULL;
>   }
>   
> -int ttm_pool_populate(struct ttm_tt *ttm)
> +int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   {
>   	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
> -	struct ttm_operation_ctx ctx = {
> -		.interruptible = false,
> -		.no_wait_gpu = false
> -	};
>   	unsigned i;
>   	int ret;
>   
> @@ -1080,7 +1076,7 @@ int ttm_pool_populate(struct ttm_tt *ttm)
>   
>   	for (i = 0; i < ttm->num_pages; ++i) {
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						PAGE_SIZE, &ctx);
> +						PAGE_SIZE, ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_pool_unpopulate(ttm);
>   			return -ENOMEM;
> @@ -1117,12 +1113,13 @@ void ttm_pool_unpopulate(struct ttm_tt *ttm)
>   }
>   EXPORT_SYMBOL(ttm_pool_unpopulate);
>   
> -int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt)
> +int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt,
> +					struct ttm_operation_ctx *ctx)
>   {
>   	unsigned i, j;
>   	int r;
>   
> -	r = ttm_pool_populate(&tt->ttm);
> +	r = ttm_pool_populate(&tt->ttm, ctx);
>   	if (r)
>   		return r;
>   
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> index 8aac86a..3ac5391 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> @@ -923,14 +923,11 @@ static gfp_t ttm_dma_pool_gfp_flags(struct ttm_dma_tt *ttm_dma, bool huge)
>    * On success pages list will hold count number of correctly
>    * cached pages. On failure will hold the negative return value (-ENOMEM, etc).
>    */
> -int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
> +int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			struct ttm_operation_ctx *ctx)
>   {
>   	struct ttm_tt *ttm = &ttm_dma->ttm;
>   	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
> -	struct ttm_operation_ctx ctx = {
> -		.interruptible = false,
> -		.no_wait_gpu = false
> -	};
>   	unsigned long num_pages = ttm->num_pages;
>   	struct dma_pool *pool;
>   	enum pool_type type;
> @@ -966,7 +963,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   			break;
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						pool->size, &ctx);
> +						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_dma_unpopulate(ttm_dma, dev);
>   			return -ENOMEM;
> @@ -1002,7 +999,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   		}
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						pool->size, &ctx);
> +						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_dma_unpopulate(ttm_dma, dev);
>   			return -ENOMEM;
> diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c
> index 8ebc8d3..b48d7a0 100644
> --- a/drivers/gpu/drm/ttm/ttm_tt.c
> +++ b/drivers/gpu/drm/ttm/ttm_tt.c
> @@ -263,6 +263,10 @@ void ttm_tt_unbind(struct ttm_tt *ttm)
>   
>   int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
>   {
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	int ret = 0;
>   
>   	if (!ttm)
> @@ -271,7 +275,7 @@ int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
>   	if (ttm->state == tt_bound)
>   		return 0;
>   
> -	ret = ttm->bdev->driver->ttm_tt_populate(ttm);
> +	ret = ttm->bdev->driver->ttm_tt_populate(ttm, &ctx);
>   	if (ret)
>   		return ret;
>   
> diff --git a/drivers/gpu/drm/virtio/virtgpu_object.c b/drivers/gpu/drm/virtio/virtgpu_object.c
> index 6f66b73..0b90cdb 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_object.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_object.c
> @@ -124,13 +124,17 @@ int virtio_gpu_object_get_sg_table(struct virtio_gpu_device *qdev,
>   	int ret;
>   	struct page **pages = bo->tbo.ttm->pages;
>   	int nr_pages = bo->tbo.num_pages;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   
>   	/* wtf swapping */
>   	if (bo->pages)
>   		return 0;
>   
>   	if (bo->tbo.ttm->state == tt_unpopulated)
> -		bo->tbo.ttm->bdev->driver->ttm_tt_populate(bo->tbo.ttm);
> +		bo->tbo.ttm->bdev->driver->ttm_tt_populate(bo->tbo.ttm, &ctx);
>   	bo->pages = kmalloc(sizeof(struct sg_table), GFP_KERNEL);
>   	if (!bo->pages)
>   		goto out;
> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> index 488c6bd..72eb417 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> @@ -324,12 +324,13 @@ static struct ttm_backend_func virtio_gpu_backend_func = {
>   	.destroy = &virtio_gpu_ttm_backend_destroy,
>   };
>   
> -static int virtio_gpu_ttm_tt_populate(struct ttm_tt *ttm)
> +static int virtio_gpu_ttm_tt_populate(struct ttm_tt *ttm,
> +					struct ttm_operation_ctx *ctx)
>   {
>   	if (ttm->state != tt_unpopulated)
>   		return 0;
>   
> -	return ttm_pool_populate(ttm);
> +	return ttm_pool_populate(ttm, ctx);
>   }
>   
>   static void virtio_gpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index ef97542..90b0d6b 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -635,16 +635,12 @@ static void vmw_ttm_destroy(struct ttm_tt *ttm)
>   }
>   
>   
> -static int vmw_ttm_populate(struct ttm_tt *ttm)
> +static int vmw_ttm_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   {
>   	struct vmw_ttm_tt *vmw_tt =
>   		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
>   	struct vmw_private *dev_priv = vmw_tt->dev_priv;
>   	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
> -	struct ttm_operation_ctx ctx = {
> -		.interruptible = true,
> -		.no_wait_gpu = false
> -	};
>   	int ret;
>   
>   	if (ttm->state != tt_unpopulated)
> @@ -653,15 +649,16 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
>   	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
>   		size_t size =
>   			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
> -		ret = ttm_mem_global_alloc(glob, size, &ctx);
> +		ret = ttm_mem_global_alloc(glob, size, ctx);
>   		if (unlikely(ret != 0))
>   			return ret;
>   
> -		ret = ttm_dma_populate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
> +		ret = ttm_dma_populate(&vmw_tt->dma_ttm, dev_priv->dev->dev,
> +					ctx);
>   		if (unlikely(ret != 0))
>   			ttm_mem_global_free(glob, size);
>   	} else
> -		ret = ttm_pool_populate(ttm);
> +		ret = ttm_pool_populate(ttm, ctx);
>   
>   	return ret;
>   }
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
> index b17f08f..736ca47 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
> @@ -240,6 +240,10 @@ static int vmw_otable_batch_setup(struct vmw_private *dev_priv,
>   	unsigned long offset;
>   	unsigned long bo_size;
>   	struct vmw_otable *otables = batch->otables;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	SVGAOTableType i;
>   	int ret;
>   
> @@ -264,7 +268,7 @@ static int vmw_otable_batch_setup(struct vmw_private *dev_priv,
>   
>   	ret = ttm_bo_reserve(batch->otable_bo, false, true, NULL);
>   	BUG_ON(ret != 0);
> -	ret = vmw_bo_driver.ttm_tt_populate(batch->otable_bo->ttm);
> +	ret = vmw_bo_driver.ttm_tt_populate(batch->otable_bo->ttm, &ctx);
>   	if (unlikely(ret != 0))
>   		goto out_unreserve;
>   	ret = vmw_bo_map_dma(batch->otable_bo);
> @@ -430,6 +434,11 @@ static int vmw_mob_pt_populate(struct vmw_private *dev_priv,
>   			       struct vmw_mob *mob)
>   {
>   	int ret;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
> +
>   	BUG_ON(mob->pt_bo != NULL);
>   
>   	ret = ttm_bo_create(&dev_priv->bdev, mob->num_pages * PAGE_SIZE,
> @@ -442,7 +451,7 @@ static int vmw_mob_pt_populate(struct vmw_private *dev_priv,
>   	ret = ttm_bo_reserve(mob->pt_bo, false, true, NULL);
>   
>   	BUG_ON(ret != 0);
> -	ret = vmw_bo_driver.ttm_tt_populate(mob->pt_bo->ttm);
> +	ret = vmw_bo_driver.ttm_tt_populate(mob->pt_bo->ttm, &ctx);
>   	if (unlikely(ret != 0))
>   		goto out_unreserve;
>   	ret = vmw_bo_map_dma(mob->pt_bo);
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index 934fecf..84860ec 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -352,7 +352,8 @@ struct ttm_bo_driver {
>   	 * Returns:
>   	 * -ENOMEM: Out of memory.
>   	 */
> -	int (*ttm_tt_populate)(struct ttm_tt *ttm);
> +	int (*ttm_tt_populate)(struct ttm_tt *ttm,
> +			struct ttm_operation_ctx *ctx);
>   
>   	/**
>   	 * ttm_tt_unpopulate
> @@ -1077,7 +1078,7 @@ struct ttm_tt *ttm_agp_tt_create(struct ttm_bo_device *bdev,
>   				 struct agp_bridge_data *bridge,
>   				 unsigned long size, uint32_t page_flags,
>   				 struct page *dummy_read_page);
> -int ttm_agp_tt_populate(struct ttm_tt *ttm);
> +int ttm_agp_tt_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx);
>   void ttm_agp_tt_unpopulate(struct ttm_tt *ttm);
>   #endif
>   
> diff --git a/include/drm/ttm/ttm_page_alloc.h b/include/drm/ttm/ttm_page_alloc.h
> index 5938113..f8395dd 100644
> --- a/include/drm/ttm/ttm_page_alloc.h
> +++ b/include/drm/ttm/ttm_page_alloc.h
> @@ -47,7 +47,7 @@ void ttm_page_alloc_fini(void);
>    *
>    * Add backing pages to all of @ttm
>    */
> -int ttm_pool_populate(struct ttm_tt *ttm);
> +int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx);
>   
>   /**
>    * ttm_pool_unpopulate:
> @@ -61,7 +61,8 @@ void ttm_pool_unpopulate(struct ttm_tt *ttm);
>   /**
>    * Populates and DMA maps pages to fullfil a ttm_dma_populate() request
>    */
> -int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt);
> +int ttm_populate_and_map_pages(struct device *dev, struct ttm_dma_tt *tt,
> +				struct ttm_operation_ctx *ctx);
>   
>   /**
>    * Unpopulates and DMA unmaps pages as part of a
> @@ -89,7 +90,8 @@ void ttm_dma_page_alloc_fini(void);
>    */
>   int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data);
>   
> -int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev);
> +int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			struct ttm_operation_ctx *ctx);
>   void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev);
>   
>   #else
> @@ -106,7 +108,8 @@ static inline int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data)
>   	return 0;
>   }
>   static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma,
> -				   struct device *dev)
> +				   struct device *dev,
> +				   struct ttm_operation_ctx *ctx)
>   {
>   	return -ENOMEM;
>   }

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

RE: [PATCH 3/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc_page

[He, Roger]

-----Original Message-----
From: Christian König [mailto:ckoenig.leichtzumerken@gmail.com] 
Sent: Wednesday, December 20, 2017 9:36 PM
To: He, Roger <Hongbo.He@amd.com>; amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Subject: Re: [PATCH 3/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc_page

Commit message!

Am 20.12.2017 um 11:34 schrieb Roger He:
> Change-Id: I4104a12e09a374b6477a0dd5a8fce26dce27a746
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_memory.c         | 15 ++++++++-------
>   drivers/gpu/drm/ttm/ttm_page_alloc.c     |  6 +++++-
>   drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 ++++++--
>   include/drm/ttm/ttm_memory.h             |  3 ++-
>   4 files changed, 21 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c 
> b/drivers/gpu/drm/ttm/ttm_memory.c
> index 525d3b6..8df0755 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -539,15 +539,14 @@ int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
>   EXPORT_SYMBOL(ttm_mem_global_alloc);
>   
>   int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
> -			      struct page *page, uint64_t size)
> +			      struct page *page, uint64_t size,
> +			      struct ttm_operation_ctx *ctx)
>   {
> -
> +	int ret;
>   	struct ttm_mem_zone *zone = NULL;
> -	struct ttm_operation_ctx ctx = {
> -		.interruptible = false,
> -		.no_wait_gpu = false
> -	};
> +	bool tmp_no_wait_gpu = ctx->no_wait_gpu;

	Mhm, please drop that. That the function might wait for the GPU even when the caller requested not to do so sounds like a bug to me.

Yes. I will remove this. Later I will send new patches. Will appreciate If Thomas can help to verify that.

Thanks
Roger(Hongbo.He)  
>   
> +	ctx->no_wait_gpu = false;
>   	/**
>   	 * Page allocations may be registed in a single zone
>   	 * only if highmem or !dma32.
> @@ -560,7 +559,9 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>   	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
>   		zone = glob->zone_kernel;
>   #endif
> -	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
> +	ret = ttm_mem_global_alloc_zone(glob, zone, size, ctx);
> +	ctx->no_wait_gpu = tmp_no_wait_gpu;
> +	return ret;
>   }
>   
>   void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct 
> page *page, diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c 
> b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index b5ba644..8f93ff3 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -1061,6 +1061,10 @@ void ttm_page_alloc_fini(void)
>   int ttm_pool_populate(struct ttm_tt *ttm)
>   {
>   	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	unsigned i;
>   	int ret;
>   
> @@ -1076,7 +1080,7 @@ int ttm_pool_populate(struct ttm_tt *ttm)
>   
>   	for (i = 0; i < ttm->num_pages; ++i) {
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						PAGE_SIZE);
> +						PAGE_SIZE, &ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_pool_unpopulate(ttm);
>   			return -ENOMEM;
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c 
> b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> index bda00b2..8aac86a 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> @@ -927,6 +927,10 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   {
>   	struct ttm_tt *ttm = &ttm_dma->ttm;
>   	struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	unsigned long num_pages = ttm->num_pages;
>   	struct dma_pool *pool;
>   	enum pool_type type;
> @@ -962,7 +966,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   			break;
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						pool->size);
> +						pool->size, &ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_dma_unpopulate(ttm_dma, dev);
>   			return -ENOMEM;
> @@ -998,7 +1002,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   		}
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> -						pool->size);
> +						pool->size, &ctx);
>   		if (unlikely(ret != 0)) {
>   			ttm_dma_unpopulate(ttm_dma, dev);
>   			return -ENOMEM;
> diff --git a/include/drm/ttm/ttm_memory.h 
> b/include/drm/ttm/ttm_memory.h index 755c107..8936285 100644
> --- a/include/drm/ttm/ttm_memory.h
> +++ b/include/drm/ttm/ttm_memory.h
> @@ -84,7 +84,8 @@ extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
>   extern void ttm_mem_global_free(struct ttm_mem_global *glob,
>   				uint64_t amount);
>   extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
> -				     struct page *page, uint64_t size);
> +				     struct page *page, uint64_t size,
> +				     struct ttm_operation_ctx *ctx);
>   extern void ttm_mem_global_free_page(struct ttm_mem_global *glob,
>   				     struct page *page, uint64_t size);
>   extern size_t ttm_round_pot(size_t size);

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

Re: [PATCH 1/7] drm/ttm: call ttm_bo_swapout directly when ttm shrink

[Thomas Hellstrom]

Roger,

5 out of 7 patches in this series are completely lacking a commit log 
message, and thats not OK. Really.

https://www.kernel.org/doc/html/v4.12/process/submitting-patches.html#describe-your-changes

I'll review these, but IIRC the no_wait in the memory accounting code is 
different in that it doesn't allow sleeping,
whereas in the bo code we had originally had no_wait_gpu and also 
no_wait. (IIRC Maarten or Jerome removed the latter due to lack of 
users.) Seems like these patches confuse the to. At the very least that 
requires some form of motivation.

Also I wonder what testing is being performed on these changes prior to 
submission?
We have pretty high number of critical deployments out there, that we 
need to support.

/Thomas


On 12/20/2017 11:34 AM, Roger He wrote:
> then remove superfluous functions
>
> Change-Id: Iea020f0e30a239e0265e7a1500168c7d7f819bd9
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c     | 21 +++---------
>   drivers/gpu/drm/ttm/ttm_memory.c | 12 ++-----
>   include/drm/ttm/ttm_bo_api.h     |  1 +
>   include/drm/ttm/ttm_bo_driver.h  |  1 -
>   include/drm/ttm/ttm_memory.h     | 69 +---------------------------------------
>   5 files changed, 9 insertions(+), 95 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 60bb5c1..fa57aa8 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -42,7 +42,6 @@
>   #include <linux/atomic.h>
>   #include <linux/reservation.h>
>   
> -static int ttm_bo_swapout(struct ttm_mem_shrink *shrink);
>   static void ttm_bo_global_kobj_release(struct kobject *kobj);
>   
>   static struct attribute ttm_bo_count = {
> @@ -1454,7 +1453,6 @@ static void ttm_bo_global_kobj_release(struct kobject *kobj)
>   	struct ttm_bo_global *glob =
>   		container_of(kobj, struct ttm_bo_global, kobj);
>   
> -	ttm_mem_unregister_shrink(glob->mem_glob, &glob->shrink);
>   	__free_page(glob->dummy_read_page);
>   	kfree(glob);
>   }
> @@ -1479,6 +1477,7 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	mutex_init(&glob->device_list_mutex);
>   	spin_lock_init(&glob->lru_lock);
>   	glob->mem_glob = bo_ref->mem_glob;
> +	glob->mem_glob->bo_glob = glob;
>   	glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32);
>   
>   	if (unlikely(glob->dummy_read_page == NULL)) {
> @@ -1489,14 +1488,6 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	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);
> -	ret = ttm_mem_register_shrink(glob->mem_glob, &glob->shrink);
> -	if (unlikely(ret != 0)) {
> -		pr_err("Could not register buffer object swapout\n");
> -		goto out_no_shrink;
> -	}
> -
>   	atomic_set(&glob->bo_count, 0);
>   
>   	ret = kobject_init_and_add(
> @@ -1504,8 +1495,6 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	if (unlikely(ret != 0))
>   		kobject_put(&glob->kobj);
>   	return ret;
> -out_no_shrink:
> -	__free_page(glob->dummy_read_page);
>   out_no_drp:
>   	kfree(glob);
>   	return ret;
> @@ -1688,11 +1677,8 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
>    * A buffer object shrink method that tries to swap out the first
>    * buffer object on the bo_global::swap_lru list.
>    */
> -
> -static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
> +int ttm_bo_swapout(struct ttm_bo_global *glob)
>   {
> -	struct ttm_bo_global *glob =
> -	    container_of(shrink, struct ttm_bo_global, shrink);
>   	struct ttm_buffer_object *bo;
>   	int ret = -EBUSY;
>   	unsigned i;
> @@ -1774,10 +1760,11 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
>   	kref_put(&bo->list_kref, ttm_bo_release_list);
>   	return ret;
>   }
> +EXPORT_SYMBOL(ttm_bo_swapout);
>   
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
>   {
> -	while (ttm_bo_swapout(&bdev->glob->shrink) == 0)
> +	while (ttm_bo_swapout(bdev->glob) == 0)
>   		;
>   }
>   EXPORT_SYMBOL(ttm_bo_swapout_all);
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
> index e963749..9130bdf 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -214,26 +214,20 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
>   		       uint64_t extra)
>   {
>   	int ret;
> -	struct ttm_mem_shrink *shrink;
>   
>   	spin_lock(&glob->lock);
> -	if (glob->shrink == NULL)
> -		goto out;
>   
>   	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
> -		shrink = glob->shrink;
>   		spin_unlock(&glob->lock);
> -		ret = shrink->do_shrink(shrink);
> +		ret = ttm_bo_swapout(glob->bo_glob);
>   		spin_lock(&glob->lock);
>   		if (unlikely(ret != 0))
> -			goto out;
> +			break;
>   	}
> -out:
> +
>   	spin_unlock(&glob->lock);
>   }
>   
> -
> -
>   static void ttm_shrink_work(struct work_struct *work)
>   {
>   	struct ttm_mem_global *glob =
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index c126330..24a8db7 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -752,6 +752,7 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
>   		  const char __user *wbuf, char __user *rbuf,
>   		  size_t count, loff_t *f_pos, bool write);
>   
> +int ttm_bo_swapout(struct ttm_bo_global *glob);
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
>   int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
>   #endif
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index 5115718..934fecf 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -522,7 +522,6 @@ struct ttm_bo_global {
>   	struct kobject kobj;
>   	struct ttm_mem_global *mem_glob;
>   	struct page *dummy_read_page;
> -	struct ttm_mem_shrink shrink;
>   	struct mutex device_list_mutex;
>   	spinlock_t lru_lock;
>   
> diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
> index 2c1e359..85f3ad6 100644
> --- a/include/drm/ttm/ttm_memory.h
> +++ b/include/drm/ttm/ttm_memory.h
> @@ -37,20 +37,6 @@
>   #include <linux/mm.h>
>   
>   /**
> - * struct ttm_mem_shrink - callback to shrink TTM memory usage.
> - *
> - * @do_shrink: The callback function.
> - *
> - * Arguments to the do_shrink functions are intended to be passed using
> - * inheritance. That is, the argument class derives from struct ttm_mem_shrink,
> - * and can be accessed using container_of().
> - */
> -
> -struct ttm_mem_shrink {
> -	int (*do_shrink) (struct ttm_mem_shrink *);
> -};
> -
> -/**
>    * struct ttm_mem_global - Global memory accounting structure.
>    *
>    * @shrink: A single callback to shrink TTM memory usage. Extend this
> @@ -76,7 +62,7 @@ struct ttm_mem_shrink {
>   struct ttm_mem_zone;
>   struct ttm_mem_global {
>   	struct kobject kobj;
> -	struct ttm_mem_shrink *shrink;
> +	struct ttm_bo_global *bo_glob;
>   	struct workqueue_struct *swap_queue;
>   	struct work_struct work;
>   	spinlock_t lock;
> @@ -90,59 +76,6 @@ struct ttm_mem_global {
>   #endif
>   };
>   
> -/**
> - * ttm_mem_init_shrink - initialize a struct ttm_mem_shrink object
> - *
> - * @shrink: The object to initialize.
> - * @func: The callback function.
> - */
> -
> -static inline void ttm_mem_init_shrink(struct ttm_mem_shrink *shrink,
> -				       int (*func) (struct ttm_mem_shrink *))
> -{
> -	shrink->do_shrink = func;
> -}
> -
> -/**
> - * ttm_mem_register_shrink - register a struct ttm_mem_shrink object.
> - *
> - * @glob: The struct ttm_mem_global object to register with.
> - * @shrink: An initialized struct ttm_mem_shrink object to register.
> - *
> - * Returns:
> - * -EBUSY: There's already a callback registered. (May change).
> - */
> -
> -static inline int ttm_mem_register_shrink(struct ttm_mem_global *glob,
> -					  struct ttm_mem_shrink *shrink)
> -{
> -	spin_lock(&glob->lock);
> -	if (glob->shrink != NULL) {
> -		spin_unlock(&glob->lock);
> -		return -EBUSY;
> -	}
> -	glob->shrink = shrink;
> -	spin_unlock(&glob->lock);
> -	return 0;
> -}
> -
> -/**
> - * ttm_mem_unregister_shrink - unregister a struct ttm_mem_shrink object.
> - *
> - * @glob: The struct ttm_mem_global object to unregister from.
> - * @shrink: A previously registert struct ttm_mem_shrink object.
> - *
> - */
> -
> -static inline void ttm_mem_unregister_shrink(struct ttm_mem_global *glob,
> -					     struct ttm_mem_shrink *shrink)
> -{
> -	spin_lock(&glob->lock);
> -	BUG_ON(glob->shrink != shrink);
> -	glob->shrink = NULL;
> -	spin_unlock(&glob->lock);
> -}
> -
>   extern int ttm_mem_global_init(struct ttm_mem_global *glob);
>   extern void ttm_mem_global_release(struct ttm_mem_global *glob);
>   extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,


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

Re: [PATCH 6/7] drm/ttm: add ttm_bo_evict_swapout_allowable to check bo is allowable to evict or swapout

[Thomas Hellstrom]

On 12/20/2017 11:35 AM, Roger He wrote:
> extract this function since eviction and swapout share same logic

But it's only used in eviction?

>
> Change-Id: I80a475a93fceed8d66d74a1832c815a0756341ac
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c | 29 +++++++++++++++++++----------
>   1 file changed, 19 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index e7595b4..313925c 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -706,6 +706,23 @@ bool ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
>   }
>   EXPORT_SYMBOL(ttm_bo_eviction_valuable);
>   
> +static bool ttm_bo_evict_swapout_allowable(struct ttm_buffer_object *bo,
> +			struct ttm_operation_ctx *ctx, bool *locked)

Could we have a description in the code what this function actually 
does? admittedly
it checks whether a swapout or eviction is allowable, but it aso performs a
recursive tryreserve(), and that is of value to any future code reader.?

/Thomas


> +{
> +	bool ret = false;
> +
> +	*locked = false;
> +	if (bo->resv == ctx->resv) {
> +		if (ctx->allow_reserved_eviction || !list_empty(&bo->ddestroy))
> +			ret = true;
> +	} else {
> +		*locked = reservation_object_trylock(bo->resv);
> +		ret = *locked;
> +	}
> +
> +	return ret;
> +}
> +
>   static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
>   			       uint32_t mem_type,
>   			       const struct ttm_place *place,
> @@ -721,21 +738,13 @@ static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
>   	spin_lock(&glob->lru_lock);
>   	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
>   		list_for_each_entry(bo, &man->lru[i], lru) {
> -			if (bo->resv == ctx->resv) {
> -				if (!ctx->allow_reserved_eviction &&
> -				    list_empty(&bo->ddestroy))
> -					continue;
> -			} else {
> -				locked = reservation_object_trylock(bo->resv);
> -				if (!locked)
> -					continue;
> -			}
> +			if (!ttm_bo_evict_swapout_allowable(bo, ctx, &locked))
> +				continue;
>   
>   			if (place && !bdev->driver->eviction_valuable(bo,
>   								      place)) {
>   				if (locked)
>   					reservation_object_unlock(bo->resv);
> -				locked = false;
>   				continue;
>   			}
>   			break;


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

Re: [PATCH 7/7] drm/ttm: enable swapout of per VM BOs during allocation and allows reaping of deleted BOs

[Thomas Hellstrom]

What about

"Enable recursive locking at swapout time to make it possible to swap 
out BOs that share the same reservation object."

Is "per VM BOs" an AMD specific name?  In that case, I'd avoid using it 
in the TTM code since most people have no idea what they are and why the 
need specific treatment in TTM.

/Thomas


On 12/20/2017 11:35 AM, Roger He wrote:
> Change-Id: I1e87954564f38ad298bf6e4ff88c9f26f291a62d
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c     | 15 +++++++++++----
>   drivers/gpu/drm/ttm/ttm_memory.c | 12 ++++++++----
>   include/drm/ttm/ttm_bo_api.h     |  3 ++-
>   3 files changed, 21 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 313925c..ecb8916 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -1686,18 +1686,20 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
>    * A buffer object shrink method that tries to swap out the first
>    * buffer object on the bo_global::swap_lru list.
>    */
> -int ttm_bo_swapout(struct ttm_bo_global *glob)
> +int ttm_bo_swapout(struct ttm_bo_global *glob, struct ttm_operation_ctx *ctx)
>   {
>   	struct ttm_buffer_object *bo;
>   	int ret = -EBUSY;
> +	bool locked;
>   	unsigned i;
>   
>   	spin_lock(&glob->lru_lock);
>   	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
>   		list_for_each_entry(bo, &glob->swap_lru[i], swap) {
> -			ret = reservation_object_trylock(bo->resv) ? 0 : -EBUSY;
> -			if (!ret)
> +			if (ttm_bo_evict_swapout_allowable(bo, ctx, &locked)) {
> +				ret = 0;
>   				break;
> +			}
>   		}
>   		if (!ret)
>   			break;
> @@ -1773,7 +1775,12 @@ EXPORT_SYMBOL(ttm_bo_swapout);
>   
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
>   {
> -	while (ttm_bo_swapout(bdev->glob) == 0)
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
> +
> +	while (ttm_bo_swapout(bdev->glob, &ctx) == 0)
>   		;
>   }
>   EXPORT_SYMBOL(ttm_bo_swapout_all);
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
> index 8df0755..8817b86 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -211,7 +211,7 @@ static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
>    */
>   
>   static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
> -		       uint64_t extra)
> +			uint64_t extra, struct ttm_operation_ctx *ctx)
>   {
>   	int ret;
>   
> @@ -219,7 +219,7 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
>   
>   	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
>   		spin_unlock(&glob->lock);
> -		ret = ttm_bo_swapout(glob->bo_glob);
> +		ret = ttm_bo_swapout(glob->bo_glob, ctx);
>   		spin_lock(&glob->lock);
>   		if (unlikely(ret != 0))
>   			break;
> @@ -230,10 +230,14 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
>   
>   static void ttm_shrink_work(struct work_struct *work)
>   {
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	struct ttm_mem_global *glob =
>   	    container_of(work, struct ttm_mem_global, work);
>   
> -	ttm_shrink(glob, true, 0ULL);
> +	ttm_shrink(glob, true, 0ULL, &ctx);
>   }
>   
>   static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
> @@ -520,7 +524,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>   			return -ENOMEM;
>   		if (unlikely(count-- == 0))
>   			return -ENOMEM;
> -		ttm_shrink(glob, false, memory + (memory >> 2) + 16);
> +		ttm_shrink(glob, false, memory + (memory >> 2) + 16, ctx);
>   	}
>   
>   	return 0;
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index 24a8db7..f1c74c2 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -752,7 +752,8 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
>   		  const char __user *wbuf, char __user *rbuf,
>   		  size_t count, loff_t *f_pos, bool write);
>   
> -int ttm_bo_swapout(struct ttm_bo_global *glob);
> +int ttm_bo_swapout(struct ttm_bo_global *glob,
> +			struct ttm_operation_ctx *ctx);
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
>   int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
>   #endif


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

Re: [PATCH 2/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc

[Thomas Hellstrom]

With a suitable commit log, LGTM.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>


On 12/20/2017 11:34 AM, Roger He wrote:
> Change-Id: I5279b5cd3560c4082b00f822219575a5f9c3808a
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c                    |  2 +-
>   drivers/gpu/drm/ttm/ttm_memory.c                | 15 +++++++++------
>   drivers/gpu/drm/ttm/ttm_object.c                | 13 ++++++++++---
>   drivers/gpu/drm/vmwgfx/vmwgfx_binding.c         |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c          | 13 ++++++++++---
>   drivers/gpu/drm/vmwgfx/vmwgfx_context.c         |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c         |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_fence.c           |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_shader.c          | 18 +++++++++++++++---
>   drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_so.c              |  6 +++++-
>   drivers/gpu/drm/vmwgfx/vmwgfx_surface.c         | 12 ++++++++++--
>   include/drm/ttm/ttm_memory.h                    |  3 ++-
>   13 files changed, 87 insertions(+), 25 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index fa57aa8..c59f572 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -1133,7 +1133,7 @@ int ttm_bo_init_reserved(struct ttm_bo_device *bdev,
>   	struct ttm_mem_global *mem_glob = bdev->glob->mem_glob;
>   	bool locked;
>   
> -	ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false);
> +	ret = ttm_mem_global_alloc(mem_glob, acc_size, ctx);
>   	if (ret) {
>   		pr_err("Out of kernel memory\n");
>   		if (destroy)
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
> index 9130bdf..525d3b6 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -508,7 +508,7 @@ static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
>   static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>   				     struct ttm_mem_zone *single_zone,
>   				     uint64_t memory,
> -				     bool no_wait, bool interruptible)
> +				     struct ttm_operation_ctx *ctx)
>   {
>   	int count = TTM_MEMORY_ALLOC_RETRIES;
>   
> @@ -516,7 +516,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>   					       single_zone,
>   					       memory, true)
>   			!= 0)) {
> -		if (no_wait)
> +		if (ctx->no_wait_gpu)
>   			return -ENOMEM;
>   		if (unlikely(count-- == 0))
>   			return -ENOMEM;
> @@ -527,15 +527,14 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>   }
>   
>   int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
> -			 bool no_wait, bool interruptible)
> +			 struct ttm_operation_ctx *ctx)
>   {
>   	/**
>   	 * Normal allocations of kernel memory are registered in
>   	 * all zones.
>   	 */
>   
> -	return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
> -					 interruptible);
> +	return ttm_mem_global_alloc_zone(glob, NULL, memory, ctx);
>   }
>   EXPORT_SYMBOL(ttm_mem_global_alloc);
>   
> @@ -544,6 +543,10 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>   {
>   
>   	struct ttm_mem_zone *zone = NULL;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   
>   	/**
>   	 * Page allocations may be registed in a single zone
> @@ -557,7 +560,7 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>   	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
>   		zone = glob->zone_kernel;
>   #endif
> -	return ttm_mem_global_alloc_zone(glob, zone, size, false, false);
> +	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
>   }
>   
>   void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
> diff --git a/drivers/gpu/drm/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c
> index 26a7ad0..1aa2baa 100644
> --- a/drivers/gpu/drm/ttm/ttm_object.c
> +++ b/drivers/gpu/drm/ttm/ttm_object.c
> @@ -325,6 +325,10 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
>   	struct ttm_ref_object *ref;
>   	struct drm_hash_item *hash;
>   	struct ttm_mem_global *mem_glob = tfile->tdev->mem_glob;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	int ret = -EINVAL;
>   
>   	if (base->tfile != tfile && !base->shareable)
> @@ -350,7 +354,7 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
>   			return -EPERM;
>   
>   		ret = ttm_mem_global_alloc(mem_glob, sizeof(*ref),
> -					   false, false);
> +					   &ctx);
>   		if (unlikely(ret != 0))
>   			return ret;
>   		ref = kmalloc(sizeof(*ref), GFP_KERNEL);
> @@ -686,7 +690,10 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
>   	dma_buf = prime->dma_buf;
>   	if (!dma_buf || !get_dma_buf_unless_doomed(dma_buf)) {
>   		DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
> -
> +		struct ttm_operation_ctx ctx = {
> +			.interruptible = true,
> +			.no_wait_gpu = false
> +		};
>   		exp_info.ops = &tdev->ops;
>   		exp_info.size = prime->size;
>   		exp_info.flags = flags;
> @@ -696,7 +703,7 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
>   		 * Need to create a new dma_buf, with memory accounting.
>   		 */
>   		ret = ttm_mem_global_alloc(tdev->mem_glob, tdev->dma_buf_size,
> -					   false, true);
> +					   &ctx);
>   		if (unlikely(ret != 0)) {
>   			mutex_unlock(&prime->mutex);
>   			goto out_unref;
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> index 9c42e96..55d32ae 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> @@ -1202,10 +1202,14 @@ struct vmw_ctx_binding_state *
>   vmw_binding_state_alloc(struct vmw_private *dev_priv)
>   {
>   	struct vmw_ctx_binding_state *cbs;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), sizeof(*cbs),
> -				   false, false);
> +				&ctx);
>   	if (ret)
>   		return ERR_PTR(ret);
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index c705632..ef97542 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -394,6 +394,10 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
>   	struct vmw_private *dev_priv = vmw_tt->dev_priv;
>   	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
>   	struct vmw_sg_table *vsgt = &vmw_tt->vsgt;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	struct vmw_piter iter;
>   	dma_addr_t old;
>   	int ret = 0;
> @@ -417,8 +421,7 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
>   			sgt_size = ttm_round_pot(sizeof(struct sg_table));
>   		}
>   		vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages;
> -		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false,
> -					   true);
> +		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, &ctx);
>   		if (unlikely(ret != 0))
>   			return ret;
>   
> @@ -638,6 +641,10 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
>   		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
>   	struct vmw_private *dev_priv = vmw_tt->dev_priv;
>   	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	if (ttm->state != tt_unpopulated)
> @@ -646,7 +653,7 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
>   	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
>   		size_t size =
>   			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
> -		ret = ttm_mem_global_alloc(glob, size, false, true);
> +		ret = ttm_mem_global_alloc(glob, size, &ctx);
>   		if (unlikely(ret != 0))
>   			return ret;
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> index 4212b3e..3767ac3 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> @@ -746,6 +746,10 @@ static int vmw_context_define(struct drm_device *dev, void *data,
>   	struct vmw_resource *tmp;
>   	struct drm_vmw_context_arg *arg = (struct drm_vmw_context_arg *)data;
>   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	if (!dev_priv->has_dx && dx) {
> @@ -768,7 +772,7 @@ static int vmw_context_define(struct drm_device *dev, void *data,
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>   				   vmw_user_context_size,
> -				   false, true);
> +				   &ttm_opt_ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for context"
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> index 92df0b0..cbf54ea 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> @@ -573,6 +573,10 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
>   				       u32 type)
>   {
>   	struct vmw_cotable *vcotbl;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   	u32 num_entries;
>   
> @@ -580,7 +584,7 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
>   		cotable_acc_size = ttm_round_pot(sizeof(struct vmw_cotable));
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> -				   cotable_acc_size, false, true);
> +				   cotable_acc_size, &ttm_opt_ctx);
>   	if (unlikely(ret))
>   		return ERR_PTR(ret);
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> index d6b1c50..6c5c75c 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> @@ -588,6 +588,10 @@ int vmw_user_fence_create(struct drm_file *file_priv,
>   	struct vmw_user_fence *ufence;
>   	struct vmw_fence_obj *tmp;
>   	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = false,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	/*
> @@ -596,7 +600,7 @@ int vmw_user_fence_create(struct drm_file *file_priv,
>   	 */
>   
>   	ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size,
> -				   false, false);
> +				   &ctx);
>   	if (unlikely(ret != 0))
>   		return ret;
>   
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> index 004e18b..73b8e9a 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> @@ -607,6 +607,10 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
>   	struct vmw_dx_shader *shader;
>   	struct vmw_resource *res;
>   	struct vmw_private *dev_priv = ctx->dev_priv;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	if (!vmw_shader_dx_size)
> @@ -616,7 +620,7 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
>   		return -EINVAL;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), vmw_shader_dx_size,
> -				   false, true);
> +				   &ttm_opt_ctx);
>   	if (ret) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for shader "
> @@ -730,6 +734,10 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
>   {
>   	struct vmw_user_shader *ushader;
>   	struct vmw_resource *res, *tmp;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	/*
> @@ -742,7 +750,7 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>   				   vmw_user_shader_size,
> -				   false, true);
> +				   &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for shader "
> @@ -800,6 +808,10 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
>   {
>   	struct vmw_shader *shader;
>   	struct vmw_resource *res;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   
>   	/*
> @@ -812,7 +824,7 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>   				   vmw_shader_size,
> -				   false, true);
> +				   &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for shader "
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> index 051d3b3..a0cb310 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> @@ -149,6 +149,10 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
>   	struct vmw_resource *res;
>   	struct vmw_resource *tmp;
>   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	size_t alloc_size;
>   	size_t account_size;
>   	int ret;
> @@ -162,7 +166,7 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
>   		return ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), account_size,
> -				   false, true);
> +				   &ctx);
>   	ttm_read_unlock(&dev_priv->reservation_sem);
>   	if (ret) {
>   		if (ret != -ERESTARTSYS)
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> index 5a73eeb..d3573c3 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> @@ -329,6 +329,10 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
>   	struct vmw_private *dev_priv = ctx->dev_priv;
>   	struct vmw_resource *res;
>   	struct vmw_view *view;
> +	struct ttm_operation_ctx ttm_opt_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	size_t size;
>   	int ret;
>   
> @@ -345,7 +349,7 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
>   
>   	size = offsetof(struct vmw_view, cmd) + cmd_size;
>   
> -	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, false, true);
> +	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, &ttm_opt_ctx);
>   	if (ret) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for view"
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> index 6ac094e..db1bb16 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> @@ -700,6 +700,10 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
>   	struct drm_vmw_surface_create_req *req = &arg->req;
>   	struct drm_vmw_surface_arg *rep = &arg->rep;
>   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	int ret;
>   	int i, j;
>   	uint32_t cur_bo_offset;
> @@ -741,7 +745,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
>   		return ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> -				   size, false, true);
> +				   size, &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for surface"
> @@ -1479,6 +1483,10 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
>   {
>   	struct vmw_private *dev_priv = vmw_priv(dev);
>   	struct vmw_user_surface *user_srf;
> +	struct ttm_operation_ctx ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false
> +	};
>   	struct vmw_surface *srf;
>   	int ret;
>   	u32 num_layers;
> @@ -1525,7 +1533,7 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
>   		return ret;
>   
>   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> -				   user_accounting_size, false, true);
> +				   user_accounting_size, &ctx);
>   	if (unlikely(ret != 0)) {
>   		if (ret != -ERESTARTSYS)
>   			DRM_ERROR("Out of graphics memory for surface"
> diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
> index 85f3ad6..755c107 100644
> --- a/include/drm/ttm/ttm_memory.h
> +++ b/include/drm/ttm/ttm_memory.h
> @@ -35,6 +35,7 @@
>   #include <linux/errno.h>
>   #include <linux/kobject.h>
>   #include <linux/mm.h>
> +#include "ttm_bo_api.h"
>   
>   /**
>    * struct ttm_mem_global - Global memory accounting structure.
> @@ -79,7 +80,7 @@ struct ttm_mem_global {
>   extern int ttm_mem_global_init(struct ttm_mem_global *glob);
>   extern void ttm_mem_global_release(struct ttm_mem_global *glob);
>   extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
> -				bool no_wait, bool interruptible);
> +				struct ttm_operation_ctx *ctx);
>   extern void ttm_mem_global_free(struct ttm_mem_global *glob,
>   				uint64_t amount);
>   extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,


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

Re: [PATCH 3/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc_page

[Thomas Hellstrom]

On 12/21/2017 07:05 AM, He, Roger wrote:
>
> -----Original Message-----
> From: Christian König [mailto:ckoenig.leichtzumerken@gmail.com]
> Sent: Wednesday, December 20, 2017 9:36 PM
> To: He, Roger <Hongbo.He@amd.com>; amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
> Subject: Re: [PATCH 3/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc_page
>
> Commit message!
>
> Am 20.12.2017 um 11:34 schrieb Roger He:
>> Change-Id: I4104a12e09a374b6477a0dd5a8fce26dce27a746
>> Signed-off-by: Roger He <Hongbo.He@amd.com>
>> ---
>>    drivers/gpu/drm/ttm/ttm_memory.c         | 15 ++++++++-------
>>    drivers/gpu/drm/ttm/ttm_page_alloc.c     |  6 +++++-
>>    drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 ++++++--
>>    include/drm/ttm/ttm_memory.h             |  3 ++-
>>    4 files changed, 21 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c
>> b/drivers/gpu/drm/ttm/ttm_memory.c
>> index 525d3b6..8df0755 100644
>> --- a/drivers/gpu/drm/ttm/ttm_memory.c
>> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
>> @@ -539,15 +539,14 @@ int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
>>    EXPORT_SYMBOL(ttm_mem_global_alloc);
>>    
>>    int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>> -			      struct page *page, uint64_t size)
>> +			      struct page *page, uint64_t size,
>> +			      struct ttm_operation_ctx *ctx)
>>    {
>> -
>> +	int ret;
>>    	struct ttm_mem_zone *zone = NULL;
>> -	struct ttm_operation_ctx ctx = {
>> -		.interruptible = false,
>> -		.no_wait_gpu = false
>> -	};
>> +	bool tmp_no_wait_gpu = ctx->no_wait_gpu;
> 	Mhm, please drop that. That the function might wait for the GPU even when the caller requested not to do so sounds like a bug to

Yes, I agree.

/Thomas


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

Re: [PATCH 7/7] drm/ttm: enable swapout of per VM BOs during allocation and allows reaping of deleted BOs

[Christian König]

Am 21.12.2017 um 08:58 schrieb Thomas Hellstrom:
> What about
>
> "Enable recursive locking at swapout time to make it possible to swap 
> out BOs that share the same reservation object."
>
> Is "per VM BOs" an AMD specific name?

Yes, absolutely. It's even amdgpu specific, radeon uses the same 
functionality but a bit different and not for user mode allocations.

> In that case, I'd avoid using it in the TTM code since most people 
> have no idea what they are and why the need specific treatment in TTM.

Good point, going to keep that in mind when reviewing the patch set.

Christian.

>
> /Thomas
>
>
> On 12/20/2017 11:35 AM, Roger He wrote:
>> Change-Id: I1e87954564f38ad298bf6e4ff88c9f26f291a62d
>> Signed-off-by: Roger He <Hongbo.He@amd.com>
>> ---
>>   drivers/gpu/drm/ttm/ttm_bo.c     | 15 +++++++++++----
>>   drivers/gpu/drm/ttm/ttm_memory.c | 12 ++++++++----
>>   include/drm/ttm/ttm_bo_api.h     |  3 ++-
>>   3 files changed, 21 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
>> index 313925c..ecb8916 100644
>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>> @@ -1686,18 +1686,20 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
>>    * A buffer object shrink method that tries to swap out the first
>>    * buffer object on the bo_global::swap_lru list.
>>    */
>> -int ttm_bo_swapout(struct ttm_bo_global *glob)
>> +int ttm_bo_swapout(struct ttm_bo_global *glob, struct 
>> ttm_operation_ctx *ctx)
>>   {
>>       struct ttm_buffer_object *bo;
>>       int ret = -EBUSY;
>> +    bool locked;
>>       unsigned i;
>>         spin_lock(&glob->lru_lock);
>>       for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
>>           list_for_each_entry(bo, &glob->swap_lru[i], swap) {
>> -            ret = reservation_object_trylock(bo->resv) ? 0 : -EBUSY;
>> -            if (!ret)
>> +            if (ttm_bo_evict_swapout_allowable(bo, ctx, &locked)) {
>> +                ret = 0;
>>                   break;
>> +            }
>>           }
>>           if (!ret)
>>               break;
>> @@ -1773,7 +1775,12 @@ EXPORT_SYMBOL(ttm_bo_swapout);
>>     void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
>>   {
>> -    while (ttm_bo_swapout(bdev->glob) == 0)
>> +    struct ttm_operation_ctx ctx = {
>> +        .interruptible = false,
>> +        .no_wait_gpu = false
>> +    };
>> +
>> +    while (ttm_bo_swapout(bdev->glob, &ctx) == 0)
>>           ;
>>   }
>>   EXPORT_SYMBOL(ttm_bo_swapout_all);
>> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c 
>> b/drivers/gpu/drm/ttm/ttm_memory.c
>> index 8df0755..8817b86 100644
>> --- a/drivers/gpu/drm/ttm/ttm_memory.c
>> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
>> @@ -211,7 +211,7 @@ static bool ttm_zones_above_swap_target(struct 
>> ttm_mem_global *glob,
>>    */
>>     static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
>> -               uint64_t extra)
>> +            uint64_t extra, struct ttm_operation_ctx *ctx)
>>   {
>>       int ret;
>>   @@ -219,7 +219,7 @@ static void ttm_shrink(struct ttm_mem_global 
>> *glob, bool from_wq,
>>         while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
>>           spin_unlock(&glob->lock);
>> -        ret = ttm_bo_swapout(glob->bo_glob);
>> +        ret = ttm_bo_swapout(glob->bo_glob, ctx);
>>           spin_lock(&glob->lock);
>>           if (unlikely(ret != 0))
>>               break;
>> @@ -230,10 +230,14 @@ static void ttm_shrink(struct ttm_mem_global 
>> *glob, bool from_wq,
>>     static void ttm_shrink_work(struct work_struct *work)
>>   {
>> +    struct ttm_operation_ctx ctx = {
>> +        .interruptible = false,
>> +        .no_wait_gpu = false
>> +    };
>>       struct ttm_mem_global *glob =
>>           container_of(work, struct ttm_mem_global, work);
>>   -    ttm_shrink(glob, true, 0ULL);
>> +    ttm_shrink(glob, true, 0ULL, &ctx);
>>   }
>>     static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
>> @@ -520,7 +524,7 @@ static int ttm_mem_global_alloc_zone(struct 
>> ttm_mem_global *glob,
>>               return -ENOMEM;
>>           if (unlikely(count-- == 0))
>>               return -ENOMEM;
>> -        ttm_shrink(glob, false, memory + (memory >> 2) + 16);
>> +        ttm_shrink(glob, false, memory + (memory >> 2) + 16, ctx);
>>       }
>>         return 0;
>> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
>> index 24a8db7..f1c74c2 100644
>> --- a/include/drm/ttm/ttm_bo_api.h
>> +++ b/include/drm/ttm/ttm_bo_api.h
>> @@ -752,7 +752,8 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, 
>> struct file *filp,
>>             const char __user *wbuf, char __user *rbuf,
>>             size_t count, loff_t *f_pos, bool write);
>>   -int ttm_bo_swapout(struct ttm_bo_global *glob);
>> +int ttm_bo_swapout(struct ttm_bo_global *glob,
>> +            struct ttm_operation_ctx *ctx);
>>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
>>   int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
>>   #endif
>
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx

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

RE: [PATCH 1/7] drm/ttm: call ttm_bo_swapout directly when ttm shrink

[He, Roger]

Hi Thomas:

	Also I wonder what testing is being performed on these changes prior to submission?

Vulkan CTS test with Per VM BO enabled.
After that, Command submit  will not need to provide BO list it will use. It is helpful for performance to CPU bound games.

The reason why we enable eviction and swapout here is the test always overalloction.

Thanks
Roger(Hongbo.He)
-----Original Message-----
From: Thomas Hellstrom [mailto:thomas@shipmail.org] 
Sent: Thursday, December 21, 2017 3:34 PM
To: He, Roger <Hongbo.He@amd.com>; amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Subject: Re: [PATCH 1/7] drm/ttm: call ttm_bo_swapout directly when ttm shrink

Roger,

5 out of 7 patches in this series are completely lacking a commit log message, and thats not OK. Really.

https://www.kernel.org/doc/html/v4.12/process/submitting-patches.html#describe-your-changes

I'll review these, but IIRC the no_wait in the memory accounting code is different in that it doesn't allow sleeping, whereas in the bo code we had originally had no_wait_gpu and also no_wait. (IIRC Maarten or Jerome removed the latter due to lack of
users.) Seems like these patches confuse the to. At the very least that requires some form of motivation.

Also I wonder what testing is being performed on these changes prior to submission?
We have pretty high number of critical deployments out there, that we need to support.

/Thomas


On 12/20/2017 11:34 AM, Roger He wrote:
> then remove superfluous functions
>
> Change-Id: Iea020f0e30a239e0265e7a1500168c7d7f819bd9
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c     | 21 +++---------
>   drivers/gpu/drm/ttm/ttm_memory.c | 12 ++-----
>   include/drm/ttm/ttm_bo_api.h     |  1 +
>   include/drm/ttm/ttm_bo_driver.h  |  1 -
>   include/drm/ttm/ttm_memory.h     | 69 +---------------------------------------
>   5 files changed, 9 insertions(+), 95 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
> b/drivers/gpu/drm/ttm/ttm_bo.c index 60bb5c1..fa57aa8 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -42,7 +42,6 @@
>   #include <linux/atomic.h>
>   #include <linux/reservation.h>
>   
> -static int ttm_bo_swapout(struct ttm_mem_shrink *shrink);
>   static void ttm_bo_global_kobj_release(struct kobject *kobj);
>   
>   static struct attribute ttm_bo_count = { @@ -1454,7 +1453,6 @@ 
> static void ttm_bo_global_kobj_release(struct kobject *kobj)
>   	struct ttm_bo_global *glob =
>   		container_of(kobj, struct ttm_bo_global, kobj);
>   
> -	ttm_mem_unregister_shrink(glob->mem_glob, &glob->shrink);
>   	__free_page(glob->dummy_read_page);
>   	kfree(glob);
>   }
> @@ -1479,6 +1477,7 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	mutex_init(&glob->device_list_mutex);
>   	spin_lock_init(&glob->lru_lock);
>   	glob->mem_glob = bo_ref->mem_glob;
> +	glob->mem_glob->bo_glob = glob;
>   	glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32);
>   
>   	if (unlikely(glob->dummy_read_page == NULL)) { @@ -1489,14 +1488,6 
> @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	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);
> -	ret = ttm_mem_register_shrink(glob->mem_glob, &glob->shrink);
> -	if (unlikely(ret != 0)) {
> -		pr_err("Could not register buffer object swapout\n");
> -		goto out_no_shrink;
> -	}
> -
>   	atomic_set(&glob->bo_count, 0);
>   
>   	ret = kobject_init_and_add(
> @@ -1504,8 +1495,6 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
>   	if (unlikely(ret != 0))
>   		kobject_put(&glob->kobj);
>   	return ret;
> -out_no_shrink:
> -	__free_page(glob->dummy_read_page);
>   out_no_drp:
>   	kfree(glob);
>   	return ret;
> @@ -1688,11 +1677,8 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
>    * A buffer object shrink method that tries to swap out the first
>    * buffer object on the bo_global::swap_lru list.
>    */
> -
> -static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
> +int ttm_bo_swapout(struct ttm_bo_global *glob)
>   {
> -	struct ttm_bo_global *glob =
> -	    container_of(shrink, struct ttm_bo_global, shrink);
>   	struct ttm_buffer_object *bo;
>   	int ret = -EBUSY;
>   	unsigned i;
> @@ -1774,10 +1760,11 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
>   	kref_put(&bo->list_kref, ttm_bo_release_list);
>   	return ret;
>   }
> +EXPORT_SYMBOL(ttm_bo_swapout);
>   
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
>   {
> -	while (ttm_bo_swapout(&bdev->glob->shrink) == 0)
> +	while (ttm_bo_swapout(bdev->glob) == 0)
>   		;
>   }
>   EXPORT_SYMBOL(ttm_bo_swapout_all);
> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c 
> b/drivers/gpu/drm/ttm/ttm_memory.c
> index e963749..9130bdf 100644
> --- a/drivers/gpu/drm/ttm/ttm_memory.c
> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> @@ -214,26 +214,20 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
>   		       uint64_t extra)
>   {
>   	int ret;
> -	struct ttm_mem_shrink *shrink;
>   
>   	spin_lock(&glob->lock);
> -	if (glob->shrink == NULL)
> -		goto out;
>   
>   	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
> -		shrink = glob->shrink;
>   		spin_unlock(&glob->lock);
> -		ret = shrink->do_shrink(shrink);
> +		ret = ttm_bo_swapout(glob->bo_glob);
>   		spin_lock(&glob->lock);
>   		if (unlikely(ret != 0))
> -			goto out;
> +			break;
>   	}
> -out:
> +
>   	spin_unlock(&glob->lock);
>   }
>   
> -
> -
>   static void ttm_shrink_work(struct work_struct *work)
>   {
>   	struct ttm_mem_global *glob =
> diff --git a/include/drm/ttm/ttm_bo_api.h 
> b/include/drm/ttm/ttm_bo_api.h index c126330..24a8db7 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -752,6 +752,7 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
>   		  const char __user *wbuf, char __user *rbuf,
>   		  size_t count, loff_t *f_pos, bool write);
>   
> +int ttm_bo_swapout(struct ttm_bo_global *glob);
>   void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
>   int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
>   #endif
> diff --git a/include/drm/ttm/ttm_bo_driver.h 
> b/include/drm/ttm/ttm_bo_driver.h index 5115718..934fecf 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -522,7 +522,6 @@ struct ttm_bo_global {
>   	struct kobject kobj;
>   	struct ttm_mem_global *mem_glob;
>   	struct page *dummy_read_page;
> -	struct ttm_mem_shrink shrink;
>   	struct mutex device_list_mutex;
>   	spinlock_t lru_lock;
>   
> diff --git a/include/drm/ttm/ttm_memory.h 
> b/include/drm/ttm/ttm_memory.h index 2c1e359..85f3ad6 100644
> --- a/include/drm/ttm/ttm_memory.h
> +++ b/include/drm/ttm/ttm_memory.h
> @@ -37,20 +37,6 @@
>   #include <linux/mm.h>
>   
>   /**
> - * struct ttm_mem_shrink - callback to shrink TTM memory usage.
> - *
> - * @do_shrink: The callback function.
> - *
> - * Arguments to the do_shrink functions are intended to be passed 
> using
> - * inheritance. That is, the argument class derives from struct 
> ttm_mem_shrink,
> - * and can be accessed using container_of().
> - */
> -
> -struct ttm_mem_shrink {
> -	int (*do_shrink) (struct ttm_mem_shrink *);
> -};
> -
> -/**
>    * struct ttm_mem_global - Global memory accounting structure.
>    *
>    * @shrink: A single callback to shrink TTM memory usage. Extend 
> this @@ -76,7 +62,7 @@ struct ttm_mem_shrink {
>   struct ttm_mem_zone;
>   struct ttm_mem_global {
>   	struct kobject kobj;
> -	struct ttm_mem_shrink *shrink;
> +	struct ttm_bo_global *bo_glob;
>   	struct workqueue_struct *swap_queue;
>   	struct work_struct work;
>   	spinlock_t lock;
> @@ -90,59 +76,6 @@ struct ttm_mem_global {
>   #endif
>   };
>   
> -/**
> - * ttm_mem_init_shrink - initialize a struct ttm_mem_shrink object
> - *
> - * @shrink: The object to initialize.
> - * @func: The callback function.
> - */
> -
> -static inline void ttm_mem_init_shrink(struct ttm_mem_shrink *shrink,
> -				       int (*func) (struct ttm_mem_shrink *))
> -{
> -	shrink->do_shrink = func;
> -}
> -
> -/**
> - * ttm_mem_register_shrink - register a struct ttm_mem_shrink object.
> - *
> - * @glob: The struct ttm_mem_global object to register with.
> - * @shrink: An initialized struct ttm_mem_shrink object to register.
> - *
> - * Returns:
> - * -EBUSY: There's already a callback registered. (May change).
> - */
> -
> -static inline int ttm_mem_register_shrink(struct ttm_mem_global *glob,
> -					  struct ttm_mem_shrink *shrink)
> -{
> -	spin_lock(&glob->lock);
> -	if (glob->shrink != NULL) {
> -		spin_unlock(&glob->lock);
> -		return -EBUSY;
> -	}
> -	glob->shrink = shrink;
> -	spin_unlock(&glob->lock);
> -	return 0;
> -}
> -
> -/**
> - * ttm_mem_unregister_shrink - unregister a struct ttm_mem_shrink object.
> - *
> - * @glob: The struct ttm_mem_global object to unregister from.
> - * @shrink: A previously registert struct ttm_mem_shrink object.
> - *
> - */
> -
> -static inline void ttm_mem_unregister_shrink(struct ttm_mem_global *glob,
> -					     struct ttm_mem_shrink *shrink)
> -{
> -	spin_lock(&glob->lock);
> -	BUG_ON(glob->shrink != shrink);
> -	glob->shrink = NULL;
> -	spin_unlock(&glob->lock);
> -}
> -
>   extern int ttm_mem_global_init(struct ttm_mem_global *glob);
>   extern void ttm_mem_global_release(struct ttm_mem_global *glob);
>   extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, 
> uint64_t memory,


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

RE: [PATCH 6/7] drm/ttm: add ttm_bo_evict_swapout_allowable to check bo is allowable to evict or swapout

[He, Roger]

-----Original Message-----
From: dri-devel [mailto:dri-devel-bounces@lists.freedesktop.org] On Behalf Of Thomas Hellstrom
Sent: Thursday, December 21, 2017 3:51 PM
To: dri-devel@lists.freedesktop.org
Subject: Re: [PATCH 6/7] drm/ttm: add ttm_bo_evict_swapout_allowable to check bo is allowable to evict or swapout

On 12/20/2017 11:35 AM, Roger He wrote:
> extract this function since eviction and swapout share same logic

	But it's only used in eviction?

Yes, currently it only used in eviction. And will be used in Swapout.

>
> Change-Id: I80a475a93fceed8d66d74a1832c815a0756341ac
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c | 29 +++++++++++++++++++----------
>   1 file changed, 19 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
> b/drivers/gpu/drm/ttm/ttm_bo.c index e7595b4..313925c 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -706,6 +706,23 @@ bool ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
>   }
>   EXPORT_SYMBOL(ttm_bo_eviction_valuable);
>   
> +static bool ttm_bo_evict_swapout_allowable(struct ttm_buffer_object *bo,
> +			struct ttm_operation_ctx *ctx, bool *locked)

	Could we have a description in the code what this function actually does? admittedly it checks whether a swapout or eviction is 	allowable, but it aso performs a recursive tryreserve(), and that is of value to any future code reader.?

Ok.

Thanks
Roger(Hongbo.He)

/Thomas


> +{
> +	bool ret = false;
> +
> +	*locked = false;
> +	if (bo->resv == ctx->resv) {
> +		if (ctx->allow_reserved_eviction || !list_empty(&bo->ddestroy))
> +			ret = true;
> +	} else {
> +		*locked = reservation_object_trylock(bo->resv);
> +		ret = *locked;
> +	}
> +
> +	return ret;
> +}
> +
>   static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
>   			       uint32_t mem_type,
>   			       const struct ttm_place *place, @@ -721,21 +738,13 @@ 
> static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
>   	spin_lock(&glob->lru_lock);
>   	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
>   		list_for_each_entry(bo, &man->lru[i], lru) {
> -			if (bo->resv == ctx->resv) {
> -				if (!ctx->allow_reserved_eviction &&
> -				    list_empty(&bo->ddestroy))
> -					continue;
> -			} else {
> -				locked = reservation_object_trylock(bo->resv);
> -				if (!locked)
> -					continue;
> -			}
> +			if (!ttm_bo_evict_swapout_allowable(bo, ctx, &locked))
> +				continue;
>   
>   			if (place && !bdev->driver->eviction_valuable(bo,
>   								      place)) {
>   				if (locked)
>   					reservation_object_unlock(bo->resv);
> -				locked = false;
>   				continue;
>   			}
>   			break;


_______________________________________________
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: [PATCH 2/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc

[Sinclair Yeh]

Hi,

Sorry, I've been on a leave.  Catching up on emails.

In case this is still not yet merged, my comment is
be consist with the new variable name.  In some cases
it's "ttm_opt_ctx", in others it's "ctx".

I prefer ttm_opt_ctx because it's more descriptive.

And yes, some description on why this change is necessary in the
commit message is nice, too.

Sinclair

On Wed, Dec 20, 2017 at 02:33:37PM +0100, Christian König wrote:
> Commit message needed! Something like:
> 
> Forward the operation context to ttm_mem_global_alloc as well.
> 
> Am 20.12.2017 um 11:34 schrieb Roger He:
> > Change-Id: I5279b5cd3560c4082b00f822219575a5f9c3808a
> > Signed-off-by: Roger He <Hongbo.He@amd.com>
> 
> With the commit message fixed, patch is Reviewed-by: Christian König
> <christian.koenig@amd.com>.
> 
> I would like to get an rb or ab from Thomas and/or Sinclair as well, since
> this is touching a lot of vmwgfx code.
> 
> Regards,
> Christian.
> 
> > ---
> >   drivers/gpu/drm/ttm/ttm_bo.c                    |  2 +-
> >   drivers/gpu/drm/ttm/ttm_memory.c                | 15 +++++++++------
> >   drivers/gpu/drm/ttm/ttm_object.c                | 13 ++++++++++---
> >   drivers/gpu/drm/vmwgfx/vmwgfx_binding.c         |  6 +++++-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c          | 13 ++++++++++---
> >   drivers/gpu/drm/vmwgfx/vmwgfx_context.c         |  6 +++++-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c         |  6 +++++-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_fence.c           |  6 +++++-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_shader.c          | 18 +++++++++++++++---
> >   drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c |  6 +++++-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_so.c              |  6 +++++-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_surface.c         | 12 ++++++++++--
> >   include/drm/ttm/ttm_memory.h                    |  3 ++-
> >   13 files changed, 87 insertions(+), 25 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> > index fa57aa8..c59f572 100644
> > --- a/drivers/gpu/drm/ttm/ttm_bo.c
> > +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> > @@ -1133,7 +1133,7 @@ int ttm_bo_init_reserved(struct ttm_bo_device *bdev,
> >   	struct ttm_mem_global *mem_glob = bdev->glob->mem_glob;
> >   	bool locked;
> > -	ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false);
> > +	ret = ttm_mem_global_alloc(mem_glob, acc_size, ctx);
> >   	if (ret) {
> >   		pr_err("Out of kernel memory\n");
> >   		if (destroy)
> > diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
> > index 9130bdf..525d3b6 100644
> > --- a/drivers/gpu/drm/ttm/ttm_memory.c
> > +++ b/drivers/gpu/drm/ttm/ttm_memory.c
> > @@ -508,7 +508,7 @@ static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
> >   static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
> >   				     struct ttm_mem_zone *single_zone,
> >   				     uint64_t memory,
> > -				     bool no_wait, bool interruptible)
> > +				     struct ttm_operation_ctx *ctx)
> >   {
> >   	int count = TTM_MEMORY_ALLOC_RETRIES;
> > @@ -516,7 +516,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
> >   					       single_zone,
> >   					       memory, true)
> >   			!= 0)) {
> > -		if (no_wait)
> > +		if (ctx->no_wait_gpu)
> >   			return -ENOMEM;
> >   		if (unlikely(count-- == 0))
> >   			return -ENOMEM;
> > @@ -527,15 +527,14 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
> >   }
> >   int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
> > -			 bool no_wait, bool interruptible)
> > +			 struct ttm_operation_ctx *ctx)
> >   {
> >   	/**
> >   	 * Normal allocations of kernel memory are registered in
> >   	 * all zones.
> >   	 */
> > -	return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
> > -					 interruptible);
> > +	return ttm_mem_global_alloc_zone(glob, NULL, memory, ctx);
> >   }
> >   EXPORT_SYMBOL(ttm_mem_global_alloc);
> > @@ -544,6 +543,10 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
> >   {
> >   	struct ttm_mem_zone *zone = NULL;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = false,
> > +		.no_wait_gpu = false
> > +	};
> >   	/**
> >   	 * Page allocations may be registed in a single zone
> > @@ -557,7 +560,7 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
> >   	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
> >   		zone = glob->zone_kernel;
> >   #endif
> > -	return ttm_mem_global_alloc_zone(glob, zone, size, false, false);
> > +	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
> >   }
> >   void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
> > diff --git a/drivers/gpu/drm/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c
> > index 26a7ad0..1aa2baa 100644
> > --- a/drivers/gpu/drm/ttm/ttm_object.c
> > +++ b/drivers/gpu/drm/ttm/ttm_object.c
> > @@ -325,6 +325,10 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
> >   	struct ttm_ref_object *ref;
> >   	struct drm_hash_item *hash;
> >   	struct ttm_mem_global *mem_glob = tfile->tdev->mem_glob;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = false,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret = -EINVAL;
> >   	if (base->tfile != tfile && !base->shareable)
> > @@ -350,7 +354,7 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
> >   			return -EPERM;
> >   		ret = ttm_mem_global_alloc(mem_glob, sizeof(*ref),
> > -					   false, false);
> > +					   &ctx);
> >   		if (unlikely(ret != 0))
> >   			return ret;
> >   		ref = kmalloc(sizeof(*ref), GFP_KERNEL);
> > @@ -686,7 +690,10 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
> >   	dma_buf = prime->dma_buf;
> >   	if (!dma_buf || !get_dma_buf_unless_doomed(dma_buf)) {
> >   		DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
> > -
> > +		struct ttm_operation_ctx ctx = {
> > +			.interruptible = true,
> > +			.no_wait_gpu = false
> > +		};
> >   		exp_info.ops = &tdev->ops;
> >   		exp_info.size = prime->size;
> >   		exp_info.flags = flags;
> > @@ -696,7 +703,7 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
> >   		 * Need to create a new dma_buf, with memory accounting.
> >   		 */
> >   		ret = ttm_mem_global_alloc(tdev->mem_glob, tdev->dma_buf_size,
> > -					   false, true);
> > +					   &ctx);
> >   		if (unlikely(ret != 0)) {
> >   			mutex_unlock(&prime->mutex);
> >   			goto out_unref;
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> > index 9c42e96..55d32ae 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
> > @@ -1202,10 +1202,14 @@ struct vmw_ctx_binding_state *
> >   vmw_binding_state_alloc(struct vmw_private *dev_priv)
> >   {
> >   	struct vmw_ctx_binding_state *cbs;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = false,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), sizeof(*cbs),
> > -				   false, false);
> > +				&ctx);
> >   	if (ret)
> >   		return ERR_PTR(ret);
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> > index c705632..ef97542 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> > @@ -394,6 +394,10 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
> >   	struct vmw_private *dev_priv = vmw_tt->dev_priv;
> >   	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
> >   	struct vmw_sg_table *vsgt = &vmw_tt->vsgt;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	struct vmw_piter iter;
> >   	dma_addr_t old;
> >   	int ret = 0;
> > @@ -417,8 +421,7 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
> >   			sgt_size = ttm_round_pot(sizeof(struct sg_table));
> >   		}
> >   		vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages;
> > -		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false,
> > -					   true);
> > +		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, &ctx);
> >   		if (unlikely(ret != 0))
> >   			return ret;
> > @@ -638,6 +641,10 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
> >   		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
> >   	struct vmw_private *dev_priv = vmw_tt->dev_priv;
> >   	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	if (ttm->state != tt_unpopulated)
> > @@ -646,7 +653,7 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
> >   	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
> >   		size_t size =
> >   			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
> > -		ret = ttm_mem_global_alloc(glob, size, false, true);
> > +		ret = ttm_mem_global_alloc(glob, size, &ctx);
> >   		if (unlikely(ret != 0))
> >   			return ret;
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> > index 4212b3e..3767ac3 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
> > @@ -746,6 +746,10 @@ static int vmw_context_define(struct drm_device *dev, void *data,
> >   	struct vmw_resource *tmp;
> >   	struct drm_vmw_context_arg *arg = (struct drm_vmw_context_arg *)data;
> >   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> > +	struct ttm_operation_ctx ttm_opt_ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	if (!dev_priv->has_dx && dx) {
> > @@ -768,7 +772,7 @@ static int vmw_context_define(struct drm_device *dev, void *data,
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> >   				   vmw_user_context_size,
> > -				   false, true);
> > +				   &ttm_opt_ctx);
> >   	if (unlikely(ret != 0)) {
> >   		if (ret != -ERESTARTSYS)
> >   			DRM_ERROR("Out of graphics memory for context"
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> > index 92df0b0..cbf54ea 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
> > @@ -573,6 +573,10 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
> >   				       u32 type)
> >   {
> >   	struct vmw_cotable *vcotbl;
> > +	struct ttm_operation_ctx ttm_opt_ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	u32 num_entries;
> > @@ -580,7 +584,7 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
> >   		cotable_acc_size = ttm_round_pot(sizeof(struct vmw_cotable));
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> > -				   cotable_acc_size, false, true);
> > +				   cotable_acc_size, &ttm_opt_ctx);
> >   	if (unlikely(ret))
> >   		return ERR_PTR(ret);
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> > index d6b1c50..6c5c75c 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> > @@ -588,6 +588,10 @@ int vmw_user_fence_create(struct drm_file *file_priv,
> >   	struct vmw_user_fence *ufence;
> >   	struct vmw_fence_obj *tmp;
> >   	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = false,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	/*
> > @@ -596,7 +600,7 @@ int vmw_user_fence_create(struct drm_file *file_priv,
> >   	 */
> >   	ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size,
> > -				   false, false);
> > +				   &ctx);
> >   	if (unlikely(ret != 0))
> >   		return ret;
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> > index 004e18b..73b8e9a 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
> > @@ -607,6 +607,10 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
> >   	struct vmw_dx_shader *shader;
> >   	struct vmw_resource *res;
> >   	struct vmw_private *dev_priv = ctx->dev_priv;
> > +	struct ttm_operation_ctx ttm_opt_ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	if (!vmw_shader_dx_size)
> > @@ -616,7 +620,7 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
> >   		return -EINVAL;
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), vmw_shader_dx_size,
> > -				   false, true);
> > +				   &ttm_opt_ctx);
> >   	if (ret) {
> >   		if (ret != -ERESTARTSYS)
> >   			DRM_ERROR("Out of graphics memory for shader "
> > @@ -730,6 +734,10 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
> >   {
> >   	struct vmw_user_shader *ushader;
> >   	struct vmw_resource *res, *tmp;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	/*
> > @@ -742,7 +750,7 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> >   				   vmw_user_shader_size,
> > -				   false, true);
> > +				   &ctx);
> >   	if (unlikely(ret != 0)) {
> >   		if (ret != -ERESTARTSYS)
> >   			DRM_ERROR("Out of graphics memory for shader "
> > @@ -800,6 +808,10 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
> >   {
> >   	struct vmw_shader *shader;
> >   	struct vmw_resource *res;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	/*
> > @@ -812,7 +824,7 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> >   				   vmw_shader_size,
> > -				   false, true);
> > +				   &ctx);
> >   	if (unlikely(ret != 0)) {
> >   		if (ret != -ERESTARTSYS)
> >   			DRM_ERROR("Out of graphics memory for shader "
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> > index 051d3b3..a0cb310 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
> > @@ -149,6 +149,10 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
> >   	struct vmw_resource *res;
> >   	struct vmw_resource *tmp;
> >   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	size_t alloc_size;
> >   	size_t account_size;
> >   	int ret;
> > @@ -162,7 +166,7 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
> >   		return ret;
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), account_size,
> > -				   false, true);
> > +				   &ctx);
> >   	ttm_read_unlock(&dev_priv->reservation_sem);
> >   	if (ret) {
> >   		if (ret != -ERESTARTSYS)
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> > index 5a73eeb..d3573c3 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
> > @@ -329,6 +329,10 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
> >   	struct vmw_private *dev_priv = ctx->dev_priv;
> >   	struct vmw_resource *res;
> >   	struct vmw_view *view;
> > +	struct ttm_operation_ctx ttm_opt_ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	size_t size;
> >   	int ret;
> > @@ -345,7 +349,7 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
> >   	size = offsetof(struct vmw_view, cmd) + cmd_size;
> > -	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, false, true);
> > +	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, &ttm_opt_ctx);
> >   	if (ret) {
> >   		if (ret != -ERESTARTSYS)
> >   			DRM_ERROR("Out of graphics memory for view"
> > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> > index 6ac094e..db1bb16 100644
> > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
> > @@ -700,6 +700,10 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
> >   	struct drm_vmw_surface_create_req *req = &arg->req;
> >   	struct drm_vmw_surface_arg *rep = &arg->rep;
> >   	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	int ret;
> >   	int i, j;
> >   	uint32_t cur_bo_offset;
> > @@ -741,7 +745,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
> >   		return ret;
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> > -				   size, false, true);
> > +				   size, &ctx);
> >   	if (unlikely(ret != 0)) {
> >   		if (ret != -ERESTARTSYS)
> >   			DRM_ERROR("Out of graphics memory for surface"
> > @@ -1479,6 +1483,10 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
> >   {
> >   	struct vmw_private *dev_priv = vmw_priv(dev);
> >   	struct vmw_user_surface *user_srf;
> > +	struct ttm_operation_ctx ctx = {
> > +		.interruptible = true,
> > +		.no_wait_gpu = false
> > +	};
> >   	struct vmw_surface *srf;
> >   	int ret;
> >   	u32 num_layers;
> > @@ -1525,7 +1533,7 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
> >   		return ret;
> >   	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
> > -				   user_accounting_size, false, true);
> > +				   user_accounting_size, &ctx);
> >   	if (unlikely(ret != 0)) {
> >   		if (ret != -ERESTARTSYS)
> >   			DRM_ERROR("Out of graphics memory for surface"
> > diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
> > index 85f3ad6..755c107 100644
> > --- a/include/drm/ttm/ttm_memory.h
> > +++ b/include/drm/ttm/ttm_memory.h
> > @@ -35,6 +35,7 @@
> >   #include <linux/errno.h>
> >   #include <linux/kobject.h>
> >   #include <linux/mm.h>
> > +#include "ttm_bo_api.h"
> >   /**
> >    * struct ttm_mem_global - Global memory accounting structure.
> > @@ -79,7 +80,7 @@ struct ttm_mem_global {
> >   extern int ttm_mem_global_init(struct ttm_mem_global *glob);
> >   extern void ttm_mem_global_release(struct ttm_mem_global *glob);
> >   extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
> > -				bool no_wait, bool interruptible);
> > +				struct ttm_operation_ctx *ctx);
> >   extern void ttm_mem_global_free(struct ttm_mem_global *glob,
> >   				uint64_t amount);
> >   extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
> 
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 2/7] drm/ttm: use an operation ctx for ttm_mem_global_alloc

[Christian König]

Unfortunately the patches are already comitted (with a better commit 
message IIRC).

But the variable name sounds just like a style nit pick to me we can fix 
in a separate patch as well.

Thanks,
Christian.

Am 25.01.2018 um 00:24 schrieb Sinclair Yeh:
> Hi,
>
> Sorry, I've been on a leave.  Catching up on emails.
>
> In case this is still not yet merged, my comment is
> be consist with the new variable name.  In some cases
> it's "ttm_opt_ctx", in others it's "ctx".
>
> I prefer ttm_opt_ctx because it's more descriptive.
>
> And yes, some description on why this change is necessary in the
> commit message is nice, too.
>
> Sinclair
>
> On Wed, Dec 20, 2017 at 02:33:37PM +0100, Christian König wrote:
>> Commit message needed! Something like:
>>
>> Forward the operation context to ttm_mem_global_alloc as well.
>>
>> Am 20.12.2017 um 11:34 schrieb Roger He:
>>> Change-Id: I5279b5cd3560c4082b00f822219575a5f9c3808a
>>> Signed-off-by: Roger He <Hongbo.He@amd.com>
>> With the commit message fixed, patch is Reviewed-by: Christian König
>> <christian.koenig@amd.com>.
>>
>> I would like to get an rb or ab from Thomas and/or Sinclair as well, since
>> this is touching a lot of vmwgfx code.
>>
>> Regards,
>> Christian.
>>
>>> ---
>>>    drivers/gpu/drm/ttm/ttm_bo.c                    |  2 +-
>>>    drivers/gpu/drm/ttm/ttm_memory.c                | 15 +++++++++------
>>>    drivers/gpu/drm/ttm/ttm_object.c                | 13 ++++++++++---
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_binding.c         |  6 +++++-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c          | 13 ++++++++++---
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_context.c         |  6 +++++-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c         |  6 +++++-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_fence.c           |  6 +++++-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_shader.c          | 18 +++++++++++++++---
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c |  6 +++++-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_so.c              |  6 +++++-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_surface.c         | 12 ++++++++++--
>>>    include/drm/ttm/ttm_memory.h                    |  3 ++-
>>>    13 files changed, 87 insertions(+), 25 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index fa57aa8..c59f572 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -1133,7 +1133,7 @@ int ttm_bo_init_reserved(struct ttm_bo_device *bdev,
>>>    	struct ttm_mem_global *mem_glob = bdev->glob->mem_glob;
>>>    	bool locked;
>>> -	ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false);
>>> +	ret = ttm_mem_global_alloc(mem_glob, acc_size, ctx);
>>>    	if (ret) {
>>>    		pr_err("Out of kernel memory\n");
>>>    		if (destroy)
>>> diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
>>> index 9130bdf..525d3b6 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_memory.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_memory.c
>>> @@ -508,7 +508,7 @@ static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
>>>    static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>>>    				     struct ttm_mem_zone *single_zone,
>>>    				     uint64_t memory,
>>> -				     bool no_wait, bool interruptible)
>>> +				     struct ttm_operation_ctx *ctx)
>>>    {
>>>    	int count = TTM_MEMORY_ALLOC_RETRIES;
>>> @@ -516,7 +516,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>>>    					       single_zone,
>>>    					       memory, true)
>>>    			!= 0)) {
>>> -		if (no_wait)
>>> +		if (ctx->no_wait_gpu)
>>>    			return -ENOMEM;
>>>    		if (unlikely(count-- == 0))
>>>    			return -ENOMEM;
>>> @@ -527,15 +527,14 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
>>>    }
>>>    int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
>>> -			 bool no_wait, bool interruptible)
>>> +			 struct ttm_operation_ctx *ctx)
>>>    {
>>>    	/**
>>>    	 * Normal allocations of kernel memory are registered in
>>>    	 * all zones.
>>>    	 */
>>> -	return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
>>> -					 interruptible);
>>> +	return ttm_mem_global_alloc_zone(glob, NULL, memory, ctx);
>>>    }
>>>    EXPORT_SYMBOL(ttm_mem_global_alloc);
>>> @@ -544,6 +543,10 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>>>    {
>>>    	struct ttm_mem_zone *zone = NULL;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = false,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	/**
>>>    	 * Page allocations may be registed in a single zone
>>> @@ -557,7 +560,7 @@ int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
>>>    	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
>>>    		zone = glob->zone_kernel;
>>>    #endif
>>> -	return ttm_mem_global_alloc_zone(glob, zone, size, false, false);
>>> +	return ttm_mem_global_alloc_zone(glob, zone, size, &ctx);
>>>    }
>>>    void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
>>> diff --git a/drivers/gpu/drm/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c
>>> index 26a7ad0..1aa2baa 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_object.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_object.c
>>> @@ -325,6 +325,10 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
>>>    	struct ttm_ref_object *ref;
>>>    	struct drm_hash_item *hash;
>>>    	struct ttm_mem_global *mem_glob = tfile->tdev->mem_glob;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = false,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret = -EINVAL;
>>>    	if (base->tfile != tfile && !base->shareable)
>>> @@ -350,7 +354,7 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
>>>    			return -EPERM;
>>>    		ret = ttm_mem_global_alloc(mem_glob, sizeof(*ref),
>>> -					   false, false);
>>> +					   &ctx);
>>>    		if (unlikely(ret != 0))
>>>    			return ret;
>>>    		ref = kmalloc(sizeof(*ref), GFP_KERNEL);
>>> @@ -686,7 +690,10 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
>>>    	dma_buf = prime->dma_buf;
>>>    	if (!dma_buf || !get_dma_buf_unless_doomed(dma_buf)) {
>>>    		DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
>>> -
>>> +		struct ttm_operation_ctx ctx = {
>>> +			.interruptible = true,
>>> +			.no_wait_gpu = false
>>> +		};
>>>    		exp_info.ops = &tdev->ops;
>>>    		exp_info.size = prime->size;
>>>    		exp_info.flags = flags;
>>> @@ -696,7 +703,7 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
>>>    		 * Need to create a new dma_buf, with memory accounting.
>>>    		 */
>>>    		ret = ttm_mem_global_alloc(tdev->mem_glob, tdev->dma_buf_size,
>>> -					   false, true);
>>> +					   &ctx);
>>>    		if (unlikely(ret != 0)) {
>>>    			mutex_unlock(&prime->mutex);
>>>    			goto out_unref;
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
>>> index 9c42e96..55d32ae 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
>>> @@ -1202,10 +1202,14 @@ struct vmw_ctx_binding_state *
>>>    vmw_binding_state_alloc(struct vmw_private *dev_priv)
>>>    {
>>>    	struct vmw_ctx_binding_state *cbs;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = false,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), sizeof(*cbs),
>>> -				   false, false);
>>> +				&ctx);
>>>    	if (ret)
>>>    		return ERR_PTR(ret);
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> index c705632..ef97542 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> @@ -394,6 +394,10 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
>>>    	struct vmw_private *dev_priv = vmw_tt->dev_priv;
>>>    	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
>>>    	struct vmw_sg_table *vsgt = &vmw_tt->vsgt;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	struct vmw_piter iter;
>>>    	dma_addr_t old;
>>>    	int ret = 0;
>>> @@ -417,8 +421,7 @@ static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
>>>    			sgt_size = ttm_round_pot(sizeof(struct sg_table));
>>>    		}
>>>    		vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages;
>>> -		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false,
>>> -					   true);
>>> +		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, &ctx);
>>>    		if (unlikely(ret != 0))
>>>    			return ret;
>>> @@ -638,6 +641,10 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
>>>    		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
>>>    	struct vmw_private *dev_priv = vmw_tt->dev_priv;
>>>    	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	if (ttm->state != tt_unpopulated)
>>> @@ -646,7 +653,7 @@ static int vmw_ttm_populate(struct ttm_tt *ttm)
>>>    	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
>>>    		size_t size =
>>>    			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
>>> -		ret = ttm_mem_global_alloc(glob, size, false, true);
>>> +		ret = ttm_mem_global_alloc(glob, size, &ctx);
>>>    		if (unlikely(ret != 0))
>>>    			return ret;
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
>>> index 4212b3e..3767ac3 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
>>> @@ -746,6 +746,10 @@ static int vmw_context_define(struct drm_device *dev, void *data,
>>>    	struct vmw_resource *tmp;
>>>    	struct drm_vmw_context_arg *arg = (struct drm_vmw_context_arg *)data;
>>>    	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
>>> +	struct ttm_operation_ctx ttm_opt_ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	if (!dev_priv->has_dx && dx) {
>>> @@ -768,7 +772,7 @@ static int vmw_context_define(struct drm_device *dev, void *data,
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>>>    				   vmw_user_context_size,
>>> -				   false, true);
>>> +				   &ttm_opt_ctx);
>>>    	if (unlikely(ret != 0)) {
>>>    		if (ret != -ERESTARTSYS)
>>>    			DRM_ERROR("Out of graphics memory for context"
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
>>> index 92df0b0..cbf54ea 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
>>> @@ -573,6 +573,10 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
>>>    				       u32 type)
>>>    {
>>>    	struct vmw_cotable *vcotbl;
>>> +	struct ttm_operation_ctx ttm_opt_ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	u32 num_entries;
>>> @@ -580,7 +584,7 @@ struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
>>>    		cotable_acc_size = ttm_round_pot(sizeof(struct vmw_cotable));
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>>> -				   cotable_acc_size, false, true);
>>> +				   cotable_acc_size, &ttm_opt_ctx);
>>>    	if (unlikely(ret))
>>>    		return ERR_PTR(ret);
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
>>> index d6b1c50..6c5c75c 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
>>> @@ -588,6 +588,10 @@ int vmw_user_fence_create(struct drm_file *file_priv,
>>>    	struct vmw_user_fence *ufence;
>>>    	struct vmw_fence_obj *tmp;
>>>    	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = false,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	/*
>>> @@ -596,7 +600,7 @@ int vmw_user_fence_create(struct drm_file *file_priv,
>>>    	 */
>>>    	ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size,
>>> -				   false, false);
>>> +				   &ctx);
>>>    	if (unlikely(ret != 0))
>>>    		return ret;
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
>>> index 004e18b..73b8e9a 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
>>> @@ -607,6 +607,10 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
>>>    	struct vmw_dx_shader *shader;
>>>    	struct vmw_resource *res;
>>>    	struct vmw_private *dev_priv = ctx->dev_priv;
>>> +	struct ttm_operation_ctx ttm_opt_ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	if (!vmw_shader_dx_size)
>>> @@ -616,7 +620,7 @@ int vmw_dx_shader_add(struct vmw_cmdbuf_res_manager *man,
>>>    		return -EINVAL;
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), vmw_shader_dx_size,
>>> -				   false, true);
>>> +				   &ttm_opt_ctx);
>>>    	if (ret) {
>>>    		if (ret != -ERESTARTSYS)
>>>    			DRM_ERROR("Out of graphics memory for shader "
>>> @@ -730,6 +734,10 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
>>>    {
>>>    	struct vmw_user_shader *ushader;
>>>    	struct vmw_resource *res, *tmp;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	/*
>>> @@ -742,7 +750,7 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>>>    				   vmw_user_shader_size,
>>> -				   false, true);
>>> +				   &ctx);
>>>    	if (unlikely(ret != 0)) {
>>>    		if (ret != -ERESTARTSYS)
>>>    			DRM_ERROR("Out of graphics memory for shader "
>>> @@ -800,6 +808,10 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
>>>    {
>>>    	struct vmw_shader *shader;
>>>    	struct vmw_resource *res;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	/*
>>> @@ -812,7 +824,7 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>>>    				   vmw_shader_size,
>>> -				   false, true);
>>> +				   &ctx);
>>>    	if (unlikely(ret != 0)) {
>>>    		if (ret != -ERESTARTSYS)
>>>    			DRM_ERROR("Out of graphics memory for shader "
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
>>> index 051d3b3..a0cb310 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
>>> @@ -149,6 +149,10 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
>>>    	struct vmw_resource *res;
>>>    	struct vmw_resource *tmp;
>>>    	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	size_t alloc_size;
>>>    	size_t account_size;
>>>    	int ret;
>>> @@ -162,7 +166,7 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
>>>    		return ret;
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), account_size,
>>> -				   false, true);
>>> +				   &ctx);
>>>    	ttm_read_unlock(&dev_priv->reservation_sem);
>>>    	if (ret) {
>>>    		if (ret != -ERESTARTSYS)
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
>>> index 5a73eeb..d3573c3 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
>>> @@ -329,6 +329,10 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
>>>    	struct vmw_private *dev_priv = ctx->dev_priv;
>>>    	struct vmw_resource *res;
>>>    	struct vmw_view *view;
>>> +	struct ttm_operation_ctx ttm_opt_ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	size_t size;
>>>    	int ret;
>>> @@ -345,7 +349,7 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
>>>    	size = offsetof(struct vmw_view, cmd) + cmd_size;
>>> -	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, false, true);
>>> +	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, &ttm_opt_ctx);
>>>    	if (ret) {
>>>    		if (ret != -ERESTARTSYS)
>>>    			DRM_ERROR("Out of graphics memory for view"
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
>>> index 6ac094e..db1bb16 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
>>> @@ -700,6 +700,10 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
>>>    	struct drm_vmw_surface_create_req *req = &arg->req;
>>>    	struct drm_vmw_surface_arg *rep = &arg->rep;
>>>    	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	int ret;
>>>    	int i, j;
>>>    	uint32_t cur_bo_offset;
>>> @@ -741,7 +745,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
>>>    		return ret;
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>>> -				   size, false, true);
>>> +				   size, &ctx);
>>>    	if (unlikely(ret != 0)) {
>>>    		if (ret != -ERESTARTSYS)
>>>    			DRM_ERROR("Out of graphics memory for surface"
>>> @@ -1479,6 +1483,10 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
>>>    {
>>>    	struct vmw_private *dev_priv = vmw_priv(dev);
>>>    	struct vmw_user_surface *user_srf;
>>> +	struct ttm_operation_ctx ctx = {
>>> +		.interruptible = true,
>>> +		.no_wait_gpu = false
>>> +	};
>>>    	struct vmw_surface *srf;
>>>    	int ret;
>>>    	u32 num_layers;
>>> @@ -1525,7 +1533,7 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
>>>    		return ret;
>>>    	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
>>> -				   user_accounting_size, false, true);
>>> +				   user_accounting_size, &ctx);
>>>    	if (unlikely(ret != 0)) {
>>>    		if (ret != -ERESTARTSYS)
>>>    			DRM_ERROR("Out of graphics memory for surface"
>>> diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h
>>> index 85f3ad6..755c107 100644
>>> --- a/include/drm/ttm/ttm_memory.h
>>> +++ b/include/drm/ttm/ttm_memory.h
>>> @@ -35,6 +35,7 @@
>>>    #include <linux/errno.h>
>>>    #include <linux/kobject.h>
>>>    #include <linux/mm.h>
>>> +#include "ttm_bo_api.h"
>>>    /**
>>>     * struct ttm_mem_global - Global memory accounting structure.
>>> @@ -79,7 +80,7 @@ struct ttm_mem_global {
>>>    extern int ttm_mem_global_init(struct ttm_mem_global *glob);
>>>    extern void ttm_mem_global_release(struct ttm_mem_global *glob);
>>>    extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
>>> -				bool no_wait, bool interruptible);
>>> +				struct ttm_operation_ctx *ctx);
>>>    extern void ttm_mem_global_free(struct ttm_mem_global *glob,
>>>    				uint64_t amount);
>>>    extern int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,

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