Skip to contentSkip to conversation

drm/ttm: make ttm_mem_type_manager_func debug more usfull

PATCH In review Christian König 2 participants 8 messages Aug 7, 2017

Subsystems (8)
lore
Share & tools
lore
#0 lore raw

PATCH 1/7 drm/ttm: make ttm_mem_type_manager_func debug more usfull

From: Christian König <[email protected]>

Provide the drm printer directly instead of just the callback.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c   | 7 +++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c  | 7 +++----
 drivers/gpu/drm/nouveau/nouveau_ttm.c         | 6 ++++--
 drivers/gpu/drm/ttm/ttm_bo.c                  | 3 ++-
 drivers/gpu/drm/ttm/ttm_bo_manager.c          | 5 ++---
 drivers/gpu/drm/virtio/virtgpu_ttm.c          | 2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c | 4 ++--
 include/drm/ttm/ttm_bo_driver.h               | 5 +++--
 8 files changed, 20 insertions(+), 19 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c +3 -4
index 5e6b90c..26818b0 100644
@@ -253,18 +253,17 @@ static void amdgpu_gtt_mgr_del(struct ttm_mem_type_manager *man,
  * amdgpu_gtt_mgr_debug - dump VRAM table
  *
  * @man: TTM memory type manager
- * @prefix: text prefix
+ * @printer: DRM printer to use
  *
  * Dump the table content using printk.
  */
 static void amdgpu_gtt_mgr_debug(struct ttm_mem_type_manager *man,
-				  const char *prefix)
+				 struct drm_printer *printer)
 {
 	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_printer p = drm_debug_printer(prefix);
 
 	spin_lock(&mgr->lock);
-	drm_mm_print(&mgr->mm, &p);
+	drm_mm_print(&mgr->mm, printer);
 	spin_unlock(&mgr->lock);
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +3 -4
index a2c59a0..3f86b47 100644
@@ -204,18 +204,17 @@ static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
  * amdgpu_vram_mgr_debug - dump VRAM table
  *
  * @man: TTM memory type manager
- * @prefix: text prefix
+ * @printer: DRM printer to use
  *
  * Dump the table content using printk.
  */
 static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
-				  const char *prefix)
+				  struct drm_printer *printer)
 {
 	struct amdgpu_vram_mgr *mgr = man->priv;
-	struct drm_printer p = drm_debug_printer(prefix);
 
 	spin_lock(&mgr->lock);
-	drm_mm_print(&mgr->mm, &p);
+	drm_mm_print(&mgr->mm, printer);
 	spin_unlock(&mgr->lock);
 }
 
drivers/gpu/drm/nouveau/nouveau_ttm.c +4 -2
index 13e5cc5..9142068 100644
@@ -179,7 +179,8 @@ nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
 }
 
 static void
-nouveau_gart_manager_debug(struct ttm_mem_type_manager *man, const char *prefix)
+nouveau_gart_manager_debug(struct ttm_mem_type_manager *man,
+			   struct drm_printer *printer)
 {
 }
 
@@ -252,7 +253,8 @@ nv04_gart_manager_new(struct ttm_mem_type_manager *man,
 }
 
 static void
-nv04_gart_manager_debug(struct ttm_mem_type_manager *man, const char *prefix)
+nv04_gart_manager_debug(struct ttm_mem_type_manager *man,
+			struct drm_printer *printer)
 {
 }
 
drivers/gpu/drm/ttm/ttm_bo.c +2 -1
index d3463eb..58e7fce 100644
@@ -70,6 +70,7 @@ static inline int ttm_mem_type_from_place(const struct ttm_place *place,
 static void ttm_mem_type_debug(struct ttm_bo_device *bdev, int mem_type)
 {
 	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
+	struct drm_printer p = drm_debug_printer(TTM_PFX);
 
 	pr_err("    has_type: %d\n", man->has_type);
 	pr_err("    use_type: %d\n", man->use_type);
@@ -79,7 +80,7 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, int mem_type)
 	pr_err("    available_caching: 0x%08X\n", man->available_caching);
 	pr_err("    default_caching: 0x%08X\n", man->default_caching);
 	if (mem_type != TTM_PL_SYSTEM)
-		(*man->func->debug)(man, TTM_PFX);
+		(*man->func->debug)(man, &p);
 }
 
 static void ttm_bo_mem_space_debug(struct ttm_buffer_object *bo,
drivers/gpu/drm/ttm/ttm_bo_manager.c +2 -3
index 90a6c0b..a7c232d 100644
@@ -136,13 +136,12 @@ static int ttm_bo_man_takedown(struct ttm_mem_type_manager *man)
 }
 
 static void ttm_bo_man_debug(struct ttm_mem_type_manager *man,
-			     const char *prefix)
+			     struct drm_printer *printer)
 {
 	struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv;
-	struct drm_printer p = drm_debug_printer(prefix);
 
 	spin_lock(&rman->lock);
-	drm_mm_print(&rman->mm, &p);
+	drm_mm_print(&rman->mm, printer);
 	spin_unlock(&rman->lock);
 }
 
drivers/gpu/drm/virtio/virtgpu_ttm.c +1 -1
index 4e8e27d..ed2a6d2 100644
@@ -192,7 +192,7 @@ static int ttm_bo_man_takedown(struct ttm_mem_type_manager *man)
 }
 
 static void ttm_bo_man_debug(struct ttm_mem_type_manager *man,
-			     const char *prefix)
+			     struct drm_printer *printer)
 {
 }
 
drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c +2 -2
index c1900f4..5b82c3f 100644
@@ -157,9 +157,9 @@ static int vmw_gmrid_man_takedown(struct ttm_mem_type_manager *man)
 }
 
 static void vmw_gmrid_man_debug(struct ttm_mem_type_manager *man,
-				const char *prefix)
+				struct drm_printer *printer)
 {
-	pr_info("%s: No debug info available for the GMR id manager\n", prefix);
+	drm_printf(printer, "No debug info available for the GMR id manager\n");
 }
 
 const struct ttm_mem_type_manager_func vmw_gmrid_manager_func = {
include/drm/ttm/ttm_bo_driver.h +3 -2
index 04380ba..c92bf07 100644
@@ -228,13 +228,14 @@ struct ttm_mem_type_manager_func {
 	 * struct ttm_mem_type_manager member debug
 	 *
 	 * @man: Pointer to a memory type manager.
-	 * @prefix: Prefix to be used in printout to identify the caller.
+	 * @printer: Prefix to be used in printout to identify the caller.
 	 *
 	 * This function is called to print out the state of the memory
 	 * type manager to aid debugging of out-of-memory conditions.
 	 * It may not be called from within atomic context.
 	 */
-	void (*debug)(struct ttm_mem_type_manager *man, const char *prefix);
+	void (*debug)(struct ttm_mem_type_manager *man,
+		      struct drm_printer *printer);
 };
 
 /**
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH 2/7 drm/qxl: fix incorrect use of the lru_lock

From: Christian König <[email protected]>

The BO manager has its own lock and doesn't use the lru_lock.

---
 drivers/gpu/drm/qxl/qxl_ttm.c | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)
drivers/gpu/drm/qxl/qxl_ttm.c +4 -9
index 0fdedee..07dc08d 100644
@@ -454,15 +454,10 @@ void qxl_ttm_fini(struct qxl_device *qdev)
 static int qxl_mm_dump_table(struct seq_file *m, void *data)
 {
 	struct drm_info_node *node = (struct drm_info_node *)m->private;
-	struct drm_mm *mm = (struct drm_mm *)node->info_ent->data;
-	struct drm_device *dev = node->minor->dev;
-	struct qxl_device *rdev = dev->dev_private;
-	struct ttm_bo_global *glob = rdev->mman.bdev.glob;
+	struct ttm_mem_type_manager *man = node->info_ent->data;
 	struct drm_printer p = drm_seq_file_printer(m);
 
-	spin_lock(&glob->lru_lock);
-	drm_mm_print(mm, &p);
-	spin_unlock(&glob->lru_lock);
+	man->func->debug(man, &p);
 	return 0;
 }
 #endif
@@ -483,9 +478,9 @@ int qxl_ttm_debugfs_init(struct qxl_device *qdev)
 		qxl_mem_types_list[i].show = &qxl_mm_dump_table;
 		qxl_mem_types_list[i].driver_features = 0;
 		if (i == 0)
-			qxl_mem_types_list[i].data = qdev->mman.bdev.man[TTM_PL_VRAM].priv;
+			qxl_mem_types_list[i].data = &qdev->mman.bdev.man[TTM_PL_VRAM];
 		else
-			qxl_mem_types_list[i].data = qdev->mman.bdev.man[TTM_PL_PRIV].priv;
+			qxl_mem_types_list[i].data = &qdev->mman.bdev.man[TTM_PL_PRIV];
 
 	}
 	return qxl_debugfs_add_files(qdev, qxl_mem_types_list, i);
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#2 lore raw

PATCH 3/7 drm/radeon: fix incorrect use of the lru_lock

In reply to Christian König

From: Christian König <[email protected]>

The BO manager has its own lock and doesn't use the lru_lock.

---
 drivers/gpu/drm/radeon/radeon_ttm.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
drivers/gpu/drm/radeon/radeon_ttm.c +4 -6
index 6499832..78be75a 100644
@@ -1030,19 +1030,17 @@ int radeon_mmap(struct file *filp, struct vm_area_struct *vma)
 static int radeon_mm_dump_table(struct seq_file *m, void *data)
 {
 	struct drm_info_node *node = (struct drm_info_node *)m->private;
-	unsigned ttm_pl = *(int *)node->info_ent->data;
+	unsigned ttm_pl = *(int*)node->info_ent->data;
 	struct drm_device *dev = node->minor->dev;
 	struct radeon_device *rdev = dev->dev_private;
-	struct drm_mm *mm = (struct drm_mm *)rdev->mman.bdev.man[ttm_pl].priv;
-	struct ttm_bo_global *glob = rdev->mman.bdev.glob;
+	struct ttm_mem_type_manager *man = &rdev->mman.bdev.man[ttm_pl];
 	struct drm_printer p = drm_seq_file_printer(m);
 
-	spin_lock(&glob->lru_lock);
-	drm_mm_print(mm, &p);
-	spin_unlock(&glob->lru_lock);
+	man->func->debug(man, &p);
 	return 0;
 }
 
+
 static int ttm_pl_vram = TTM_PL_VRAM;
 static int ttm_pl_tt = TTM_PL_TT;
 
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#3 lore raw

PATCH 4/7 drm/amdgpu: fix incorrect use of the lru_lock

In reply to Christian König

From: Christian König <[email protected]>

The BO manager has its own lock and doesn't use the lru_lock.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +2 -5
index f6bc4ae..d26ec55 100644
@@ -1605,13 +1605,10 @@ static int amdgpu_mm_dump_table(struct seq_file *m, void *data)
 	unsigned ttm_pl = *(int *)node->info_ent->data;
 	struct drm_device *dev = node->minor->dev;
 	struct amdgpu_device *adev = dev->dev_private;
-	struct drm_mm *mm = (struct drm_mm *)adev->mman.bdev.man[ttm_pl].priv;
-	struct ttm_bo_global *glob = adev->mman.bdev.glob;
+	struct ttm_mem_type_manager *man = &adev->mman.bdev.man[ttm_pl];
 	struct drm_printer p = drm_seq_file_printer(m);
 
-	spin_lock(&glob->lru_lock);
-	drm_mm_print(mm, &p);
-	spin_unlock(&glob->lru_lock);
+	man->func->debug(man, &p);
 	switch (ttm_pl) {
 	case TTM_PL_VRAM:
 		seq_printf(m, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#4 lore raw

PATCH 5/7 drm/amdgpu: move debug print into the MM managers

In reply to Christian König

From: Christian König <[email protected]>

Instead of the separate switch/case in the calling function.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c  | 14 +++++---------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c      | 13 -------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c |  6 ++++++
 3 files changed, 11 insertions(+), 22 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c +5 -9
index 26818b0..97c63ee 100644
@@ -153,15 +153,6 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 	return r;
 }
 
-void amdgpu_gtt_mgr_print(struct seq_file *m, struct ttm_mem_type_manager *man)
-{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
-	struct amdgpu_gtt_mgr *mgr = man->priv;
-
-	seq_printf(m, "man size:%llu pages, gtt available:%llu pages, usage:%lluMB\n",
-		   man->size, mgr->available, (u64)atomic64_read(&adev->gtt_usage) >> 20);
-
-}
 /**
  * amdgpu_gtt_mgr_new - allocate a new node
  *
@@ -260,11 +251,16 @@ static void amdgpu_gtt_mgr_del(struct ttm_mem_type_manager *man,
 static void amdgpu_gtt_mgr_debug(struct ttm_mem_type_manager *man,
 				 struct drm_printer *printer)
 {
+	struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
 	struct amdgpu_gtt_mgr *mgr = man->priv;
 
 	spin_lock(&mgr->lock);
 	drm_mm_print(&mgr->mm, printer);
 	spin_unlock(&mgr->lock);
+
+	drm_printf(printer, "man size:%llu pages, gtt available:%llu pages, usage:%lluMB\n",
+		   man->size, mgr->available,
+		   (u64)atomic64_read(&adev->gtt_usage) >> 20);
 }
 
 const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func = {
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -13
index d26ec55..076edf8 100644
@@ -1597,8 +1597,6 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
 
 #if defined(CONFIG_DEBUG_FS)
 
-extern void amdgpu_gtt_mgr_print(struct seq_file *m, struct ttm_mem_type_manager
-				 *man);
 static int amdgpu_mm_dump_table(struct seq_file *m, void *data)
 {
 	struct drm_info_node *node = (struct drm_info_node *)m->private;
@@ -1609,17 +1607,6 @@ static int amdgpu_mm_dump_table(struct seq_file *m, void *data)
 	struct drm_printer p = drm_seq_file_printer(m);
 
 	man->func->debug(man, &p);
-	switch (ttm_pl) {
-	case TTM_PL_VRAM:
-		seq_printf(m, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
-			   adev->mman.bdev.man[ttm_pl].size,
-			   (u64)atomic64_read(&adev->vram_usage) >> 20,
-			   (u64)atomic64_read(&adev->vram_vis_usage) >> 20);
-		break;
-	case TTM_PL_TT:
-		amdgpu_gtt_mgr_print(m, &adev->mman.bdev.man[TTM_PL_TT]);
-		break;
-	}
 	return 0;
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +6 -0
index 3f86b47..1eb8d5d 100644
@@ -211,11 +211,17 @@ static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
 static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
 				  struct drm_printer *printer)
 {
+	struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
 	struct amdgpu_vram_mgr *mgr = man->priv;
 
 	spin_lock(&mgr->lock);
 	drm_mm_print(&mgr->mm, printer);
 	spin_unlock(&mgr->lock);
+
+	drm_printf(printer, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
+		   adev->mman.bdev.man[TTM_PL_VRAM].size,
+		   (u64)atomic64_read(&adev->vram_usage) >> 20,
+		   (u64)atomic64_read(&adev->vram_vis_usage) >> 20);
 }
 
 const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#5 lore raw

PATCH 6/7 drm/amdgpu: move gtt usage tracking into the gtt manager

In reply to Christian König

From: Christian König <[email protected]>

It doesn't make much sense to count those numbers twice.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h         |  1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 18 +++++++++++++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c     |  5 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  |  2 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
 5 files changed, 19 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +0 -1
index 67c5286..aff89d7 100644
@@ -1486,7 +1486,6 @@ struct amdgpu_device {
 	struct amdgpu_wb		wb;
 	atomic64_t			vram_usage;
 	atomic64_t			vram_vis_usage;
-	atomic64_t			gtt_usage;
 	atomic64_t			num_bytes_moved;
 	atomic64_t			num_evictions;
 	atomic64_t			num_vram_cpu_page_faults;
drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c +15 -3
index 97c63ee..0e47c20 100644
@@ -241,6 +241,20 @@ static void amdgpu_gtt_mgr_del(struct ttm_mem_type_manager *man,
 }
 
 /**
+ * amdgpu_gtt_mgr_usage - return usage of GTT domain
+ *
+ * @man: TTM memory type manager
+ *
+ * Return how many bytes are used in the GTT domain
+ */
+uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man)
+{
+	struct amdgpu_gtt_mgr *mgr = man->priv;
+
+	return (u64)(man->size - READ_ONCE(mgr->available)) * PAGE_SIZE;
+}
+
+/**
  * amdgpu_gtt_mgr_debug - dump VRAM table
  *
  * @man: TTM memory type manager
@@ -251,7 +265,6 @@ static void amdgpu_gtt_mgr_del(struct ttm_mem_type_manager *man,
 static void amdgpu_gtt_mgr_debug(struct ttm_mem_type_manager *man,
 				 struct drm_printer *printer)
 {
-	struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
 	struct amdgpu_gtt_mgr *mgr = man->priv;
 
 	spin_lock(&mgr->lock);
@@ -259,8 +272,7 @@ static void amdgpu_gtt_mgr_debug(struct ttm_mem_type_manager *man,
 	spin_unlock(&mgr->lock);
 
 	drm_printf(printer, "man size:%llu pages, gtt available:%llu pages, usage:%lluMB\n",
-		   man->size, mgr->available,
-		   (u64)atomic64_read(&adev->gtt_usage) >> 20);
+		   man->size, mgr->available, amdgpu_gtt_mgr_usage(man) >> 20);
 }
 
 const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func = {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +3 -2
index a96c703..3209198 100644
@@ -461,7 +461,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		ui64 = atomic64_read(&adev->vram_vis_usage);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_GTT_USAGE:
-		ui64 = atomic64_read(&adev->gtt_usage);
+		ui64 = amdgpu_gtt_mgr_usage(&adev->mman.bdev.man[TTM_PL_TT]);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_GDS_CONFIG: {
 		struct drm_amdgpu_info_gds gds_info;
@@ -514,7 +514,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		mem.gtt.total_heap_size *= PAGE_SIZE;
 		mem.gtt.usable_heap_size = mem.gtt.total_heap_size
 			- adev->gart_pin_size;
-		mem.gtt.heap_usage = atomic64_read(&adev->gtt_usage);
+		mem.gtt.heap_usage =
+			amdgpu_gtt_mgr_usage(&adev->mman.bdev.man[TTM_PL_TT]);
 		mem.gtt.max_allocation = mem.gtt.usable_heap_size * 3 / 4;
 
 		return copy_to_user(out, &mem,
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +0 -2
index 16f31cb..d67a997 100644
@@ -62,7 +62,6 @@ static void amdgpu_update_memory_usage(struct amdgpu_device *adev,
 	if (new_mem) {
 		switch (new_mem->mem_type) {
 		case TTM_PL_TT:
-			atomic64_add(new_mem->size, &adev->gtt_usage);
 			break;
 		case TTM_PL_VRAM:
 			atomic64_add(new_mem->size, &adev->vram_usage);
@@ -75,7 +74,6 @@ static void amdgpu_update_memory_usage(struct amdgpu_device *adev,
 	if (old_mem) {
 		switch (old_mem->mem_type) {
 		case TTM_PL_TT:
-			atomic64_sub(old_mem->size, &adev->gtt_usage);
 			break;
 		case TTM_PL_VRAM:
 			atomic64_sub(old_mem->size, &adev->vram_usage);
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +1 -0
index 0e2399f..1307ab5 100644
@@ -66,6 +66,7 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 			 struct ttm_buffer_object *tbo,
 			 const struct ttm_place *place,
 			 struct ttm_mem_reg *mem);
+uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man);
 
 int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
 		       uint64_t dst_offset, uint32_t byte_count,
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#6 lore raw

PATCH 7/7 drm/amdgpu: move vram usage tracking into the vram manager

In reply to Christian König

From: Christian König <[email protected]>

Looks like a better place for this.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h          |  2 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c       |  5 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c      |  9 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c   | 50 ----------------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h      |  3 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 64 ++++++++++++++++++++++++++--
 6 files changed, 71 insertions(+), 62 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +0 -2
index aff89d7..03d6342 100644
@@ -1484,8 +1484,6 @@ struct amdgpu_device {
 	struct amdgpu_mman		mman;
 	struct amdgpu_vram_scratch	vram_scratch;
 	struct amdgpu_wb		wb;
-	atomic64_t			vram_usage;
-	atomic64_t			vram_vis_usage;
 	atomic64_t			num_bytes_moved;
 	atomic64_t			num_evictions;
 	atomic64_t			num_vram_cpu_page_faults;
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +3 -2
index fe974f7..6741deb 100644
@@ -243,7 +243,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev,
 	}
 
 	total_vram = adev->mc.real_vram_size - adev->vram_pin_size;
-	used_vram = atomic64_read(&adev->vram_usage);
+	used_vram = amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 	free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;
 
 	spin_lock(&adev->mm_stats.lock);
@@ -289,7 +289,8 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev,
 	/* Do the same for visible VRAM if half of it is free */
 	if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
 		u64 total_vis_vram = adev->mc.visible_vram_size;
-		u64 used_vis_vram = atomic64_read(&adev->vram_vis_usage);
+		u64 used_vis_vram =
+			amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 
 		if (used_vis_vram < total_vis_vram) {
 			u64 free_vis_vram = total_vis_vram - used_vis_vram;
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +5 -4
index 3209198..4a6407d 100644
@@ -455,10 +455,10 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		ui64 = atomic64_read(&adev->num_vram_cpu_page_faults);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_VRAM_USAGE:
-		ui64 = atomic64_read(&adev->vram_usage);
+		ui64 = amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_VIS_VRAM_USAGE:
-		ui64 = atomic64_read(&adev->vram_vis_usage);
+		ui64 = amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_GTT_USAGE:
 		ui64 = amdgpu_gtt_mgr_usage(&adev->mman.bdev.man[TTM_PL_TT]);
@@ -497,7 +497,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		mem.vram.total_heap_size = adev->mc.real_vram_size;
 		mem.vram.usable_heap_size =
 			adev->mc.real_vram_size - adev->vram_pin_size;
-		mem.vram.heap_usage = atomic64_read(&adev->vram_usage);
+		mem.vram.heap_usage =
+			amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 		mem.vram.max_allocation = mem.vram.usable_heap_size * 3 / 4;
 
 		mem.cpu_accessible_vram.total_heap_size =
@@ -506,7 +507,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			adev->mc.visible_vram_size -
 			(adev->vram_pin_size - adev->invisible_pin_size);
 		mem.cpu_accessible_vram.heap_usage =
-			atomic64_read(&adev->vram_vis_usage);
+			amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 		mem.cpu_accessible_vram.max_allocation =
 			mem.cpu_accessible_vram.usable_heap_size * 3 / 4;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +0 -50

68 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +3 -0
index 1307ab5..f22a475 100644
@@ -68,6 +68,9 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 			 struct ttm_mem_reg *mem);
 uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man);
 
+uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man);
+uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man);
+
 int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
 		       uint64_t dst_offset, uint32_t byte_count,
 		       struct reservation_object *resv,
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +60 -4

118 lines not on this page · raw message

Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Reviewed-by on v1 7/7
#7 lore raw

Re: [PATCH 7/7] drm/amdgpu: move vram usage tracking into the vram manager

Quoted 5 lines · Christian König

From: Christian König <[email protected]>

Looks like a better place for this.

Series is:

Quoted 304 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h          |  2 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c       |  5 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c      |  9 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c   | 50 ----------------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h      |  3 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 64 ++++++++++++++++++++++++++--
 6 files changed, 71 insertions(+), 62 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +0 -2
index aff89d7..03d6342 100644
@@ -1484,8 +1484,6 @@ struct amdgpu_device {
        struct amdgpu_mman              mman;
        struct amdgpu_vram_scratch      vram_scratch;
        struct amdgpu_wb                wb;
-       atomic64_t                      vram_usage;
-       atomic64_t                      vram_vis_usage;
        atomic64_t                      num_bytes_moved;
        atomic64_t                      num_evictions;
        atomic64_t                      num_vram_cpu_page_faults;
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +3 -2
index fe974f7..6741deb 100644
@@ -243,7 +243,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev,
        }
 
        total_vram = adev->mc.real_vram_size - adev->vram_pin_size;
-       used_vram = atomic64_read(&adev->vram_usage);
+       used_vram = amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
        free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;
 
        spin_lock(&adev->mm_stats.lock);
@@ -289,7 +289,8 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev,
        /* Do the same for visible VRAM if half of it is free */
        if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
                u64 total_vis_vram = adev->mc.visible_vram_size;
-               u64 used_vis_vram = atomic64_read(&adev->vram_vis_usage);
+               u64 used_vis_vram =
+                       amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 
                if (used_vis_vram < total_vis_vram) {
                        u64 free_vis_vram = total_vis_vram - used_vis_vram;
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +5 -4
index 3209198..4a6407d 100644
@@ -455,10 +455,10 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
                ui64 = atomic64_read(&adev->num_vram_cpu_page_faults);
                return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
        case AMDGPU_INFO_VRAM_USAGE:
-               ui64 = atomic64_read(&adev->vram_usage);
+               ui64 = amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
                return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
        case AMDGPU_INFO_VIS_VRAM_USAGE:
-               ui64 = atomic64_read(&adev->vram_vis_usage);
+               ui64 = amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
                return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
        case AMDGPU_INFO_GTT_USAGE:
                ui64 = amdgpu_gtt_mgr_usage(&adev->mman.bdev.man[TTM_PL_TT]);
@@ -497,7 +497,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
                mem.vram.total_heap_size = adev->mc.real_vram_size;
                mem.vram.usable_heap_size =
                        adev->mc.real_vram_size - adev->vram_pin_size;
-               mem.vram.heap_usage = atomic64_read(&adev->vram_usage);
+               mem.vram.heap_usage =
+                       amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
                mem.vram.max_allocation = mem.vram.usable_heap_size * 3 / 4;
 
                mem.cpu_accessible_vram.total_heap_size =
@@ -506,7 +507,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
                        adev->mc.visible_vram_size -
                        (adev->vram_pin_size - adev->invisible_pin_size);
                mem.cpu_accessible_vram.heap_usage =
-                       atomic64_read(&adev->vram_vis_usage);
+                       amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
                mem.cpu_accessible_vram.max_allocation =
                        mem.cpu_accessible_vram.usable_heap_size * 3 / 4;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +0 -50
index d67a997..01a125b 100644
@@ -37,53 +37,6 @@
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
 
-
-
-static u64 amdgpu_get_vis_part_size(struct amdgpu_device *adev,
-                                               struct ttm_mem_reg *mem)
-{
-       if (mem->start << PAGE_SHIFT >= adev->mc.visible_vram_size)
-               return 0;
-
-       return ((mem->start << PAGE_SHIFT) + mem->size) >
-               adev->mc.visible_vram_size ?
-               adev->mc.visible_vram_size - (mem->start << PAGE_SHIFT) :
-               mem->size;
-}
-
-static void amdgpu_update_memory_usage(struct amdgpu_device *adev,
-                      struct ttm_mem_reg *old_mem,
-                      struct ttm_mem_reg *new_mem)
-{
-       u64 vis_size;
-       if (!adev)
-               return;
-
-       if (new_mem) {
-               switch (new_mem->mem_type) {
-               case TTM_PL_TT:
-                       break;
-               case TTM_PL_VRAM:
-                       atomic64_add(new_mem->size, &adev->vram_usage);
-                       vis_size = amdgpu_get_vis_part_size(adev, new_mem);
-                       atomic64_add(vis_size, &adev->vram_vis_usage);
-                       break;
-               }
-       }
-
-       if (old_mem) {
-               switch (old_mem->mem_type) {
-               case TTM_PL_TT:
-                       break;
-               case TTM_PL_VRAM:
-                       atomic64_sub(old_mem->size, &adev->vram_usage);
-                       vis_size = amdgpu_get_vis_part_size(adev, old_mem);
-                       atomic64_sub(vis_size, &adev->vram_vis_usage);
-                       break;
-               }
-       }
-}
-
 static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo)
 {
        struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
@@ -92,7 +45,6 @@ static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo)
        bo = container_of(tbo, struct amdgpu_bo, tbo);
 
        amdgpu_bo_kunmap(bo);
-       amdgpu_update_memory_usage(adev, &bo->tbo.mem, NULL);
 
        drm_gem_object_release(&bo->gem_base);
        amdgpu_bo_unref(&bo->parent);
@@ -990,8 +942,6 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
                return;
 
        /* move_notify is called before move happens */
-       amdgpu_update_memory_usage(adev, &bo->mem, new_mem);
-
        trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +3 -0
index 1307ab5..f22a475 100644
@@ -68,6 +68,9 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
                         struct ttm_mem_reg *mem);
 uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man);
 
+uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man);
+uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man);
+
 int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
                       uint64_t dst_offset, uint32_t byte_count,
                       struct reservation_object *resv,
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +60 -5
index 1eb8d5d..49b0aa1 100644
@@ -28,6 +28,8 @@
 struct amdgpu_vram_mgr {
        struct drm_mm mm;
        spinlock_t lock;
+       uint64_t usage;
+       uint64_t vis_usage;
 };
 
 /**
@@ -79,6 +81,27 @@ static int amdgpu_vram_mgr_fini(struct ttm_mem_type_manager *man)
 }
 
 /**
+ * amdgpu_vram_mgr_vis_size - Calculate visible node size
+ *
+ * @adev: amdgpu device structure
+ * @node: MM node structure
+ *
+ * Calculate how many bytes of the MM node are inside visible VRAM
+ */
+static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev,
+                                   struct drm_mm_node *node)
+{
+       uint64_t start = node->start << PAGE_SHIFT;
+       uint64_t end = (node->size + node->start) << PAGE_SHIFT;
+
+       if (start >= adev->mc.visible_vram_size)
+               return 0;
+
+       return (end > adev->mc.visible_vram_size ?
+               adev->mc.visible_vram_size : end) - start;
+}
+
+/**
  * amdgpu_vram_mgr_new - allocate new ranges
  *
  * @man: TTM memory type manager
@@ -93,6 +116,7 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
                               const struct ttm_place *place,
                               struct ttm_mem_reg *mem)
 {
+       struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
        struct amdgpu_vram_mgr *mgr = man->priv;
        struct drm_mm *mm = &mgr->mm;
        struct drm_mm_node *nodes;
@@ -142,6 +166,9 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
                if (unlikely(r))
                        goto error;
 
+               mgr->usage += nodes[i].size << PAGE_SHIFT;
+               mgr->vis_usage += amdgpu_vram_mgr_vis_size(adev, &nodes[i]);
+
                /* Calculate a virtual BO start address to easily check if
                 * everything is CPU accessible.
                 */
@@ -181,6 +208,7 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
 static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
                                struct ttm_mem_reg *mem)
 {
+       struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
        struct amdgpu_vram_mgr *mgr = man->priv;
        struct drm_mm_node *nodes = mem->mm_node;
        unsigned pages = mem->num_pages;
@@ -192,6 +220,8 @@ static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
        while (pages) {
                pages -= nodes->size;
                drm_mm_remove_node(nodes);
+               mgr->usage -= nodes->size << PAGE_SHIFT;
+               mgr->vis_usage -= amdgpu_vram_mgr_vis_size(adev, nodes);
                ++nodes;
        }
        spin_unlock(&mgr->lock);
@@ -201,6 +231,34 @@ static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
 }
 
 /**
+ * amdgpu_vram_mgr_usage - how many bytes are used in this domain
+ *
+ * @man: TTM memory type manager
+ *
+ * Returns how many bytes are used in this domain.
+ */
+uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man)
+{
+       struct amdgpu_vram_mgr *mgr = man->priv;
+
+       return READ_ONCE(mgr->usage);
+}
+
+/**
+ * amdgpu_vram_mgr_vis_usage - how many bytes are used in the visible part
+ *
+ * @man: TTM memory type manager
+ *
+ * Returns how many bytes are used in the visible part of VRAM
+ */
+uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man)
+{
+       struct amdgpu_vram_mgr *mgr = man->priv;
+
+       return READ_ONCE(mgr->vis_usage);
+}
+
+/**
  * amdgpu_vram_mgr_debug - dump VRAM table
  *
  * @man: TTM memory type manager
@@ -211,7 +269,6 @@ static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
 static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
                                  struct drm_printer *printer)
 {
-       struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
        struct amdgpu_vram_mgr *mgr = man->priv;
 
        spin_lock(&mgr->lock);
@@ -219,9 +276,8 @@ static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
        spin_unlock(&mgr->lock);
 
        drm_printf(printer, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
-                  adev->mman.bdev.man[TTM_PL_VRAM].size,
-                  (u64)atomic64_read(&adev->vram_usage) >> 20,
-                  (u64)atomic64_read(&adev->vram_vis_usage) >> 20);
+                  man->size, amdgpu_vram_mgr_usage(man) >> 20,
+                  amdgpu_vram_mgr_vis_usage(man) >> 20);
 }
 
 const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
--

2.7.4

Quoted 4 lines

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel