drm/amdgpu: expose more memory stats in fdinfo

This will be used for performance investigations.
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c

@@ -60,12 +60,13 @@ void amdgpu_show_fdinfo(struct seq_file *m, struct file *f)
 	struct amdgpu_fpriv *fpriv = file->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
 
-	uint64_t vram_mem = 0, gtt_mem = 0, cpu_mem = 0;
+	struct amdgpu_mem_stats stats;
 	ktime_t usage[AMDGPU_HW_IP_NUM];
 	uint32_t bus, dev, fn, domain;
 	unsigned int hw_ip;
 	int ret;
 
+	memset(&stats, 0, sizeof(stats));
 	bus = adev->pdev->bus->number;
 	domain = pci_domain_nr(adev->pdev->bus);
 	dev = PCI_SLOT(adev->pdev->devfn);
@@ -75,7 +76,7 @@ void amdgpu_show_fdinfo(struct seq_file *m, struct file *f)
 	if (ret)
 		return;
 
-	amdgpu_vm_get_memory(vm, &vram_mem, &gtt_mem, &cpu_mem);
+	amdgpu_vm_get_memory(vm, &stats);
 	amdgpu_bo_unreserve(vm->root.bo);
 
 	amdgpu_ctx_mgr_usage(&fpriv->ctx_mgr, usage);
@@ -90,9 +91,22 @@ void amdgpu_show_fdinfo(struct seq_file *m, struct file *f)
 	seq_printf(m, "drm-driver:\t%s\n", file->minor->dev->driver->name);
 	seq_printf(m, "drm-pdev:\t%04x:%02x:%02x.%d\n", domain, bus, dev, fn);
 	seq_printf(m, "drm-client-id:\t%Lu\n", vm->immediate.fence_context);
-	seq_printf(m, "drm-memory-vram:\t%llu KiB\n", vram_mem/1024UL);
-	seq_printf(m, "drm-memory-gtt: \t%llu KiB\n", gtt_mem/1024UL);
-	seq_printf(m, "drm-memory-cpu: \t%llu KiB\n", cpu_mem/1024UL);
+	seq_printf(m, "drm-memory-vram:\t%llu KiB\n", stats.vram/1024UL);
+	seq_printf(m, "drm-memory-gtt: \t%llu KiB\n", stats.gtt/1024UL);
+	seq_printf(m, "drm-memory-cpu: \t%llu KiB\n", stats.cpu/1024UL);
+	seq_printf(m, "amd-memory-visible-vram:\t%llu KiB\n",
+		   stats.visible_vram/1024UL);
+	seq_printf(m, "amd-evicted-vram:\t%llu KiB\n",
+		   stats.evicted_vram/1024UL);
+	seq_printf(m, "amd-evicted-visible-vram:\t%llu KiB\n",
+		   stats.evicted_visible_vram/1024UL);
+	seq_printf(m, "amd-requested-vram:\t%llu KiB\n",
+		   stats.requested_vram/1024UL);
+	seq_printf(m, "amd-requested-visible-vram:\t%llu KiB\n",
+		   stats.requested_visible_vram/1024UL);
+	seq_printf(m, "amd-requested-gtt:\t%llu KiB\n",
+		   stats.requested_gtt/1024UL);
+
 	for (hw_ip = 0; hw_ip < AMDGPU_HW_IP_NUM; ++hw_ip) {
 		if (!usage[hw_ip])
 			continue;

drivers/gpu/drm/amd/amdgpu/amdgpu_object.c

@@ -1265,24 +1265,41 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 	trace_amdgpu_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
 }
 
-void amdgpu_bo_get_memory(struct amdgpu_bo *bo, uint64_t *vram_mem,
-				uint64_t *gtt_mem, uint64_t *cpu_mem)
+void amdgpu_bo_get_memory(struct amdgpu_bo *bo,
+			  struct amdgpu_mem_stats *stats)
 {
 	unsigned int domain;
+	uint64_t size = amdgpu_bo_size(bo);
 
 	domain = amdgpu_mem_type_to_domain(bo->tbo.resource->mem_type);
 	switch (domain) {
 	case AMDGPU_GEM_DOMAIN_VRAM:
-		*vram_mem += amdgpu_bo_size(bo);
+		stats->vram += size;
+		if (amdgpu_bo_in_cpu_visible_vram(bo))
+			stats->visible_vram += size;
 		break;
 	case AMDGPU_GEM_DOMAIN_GTT:
-		*gtt_mem += amdgpu_bo_size(bo);
+		stats->gtt += size;
 		break;
 	case AMDGPU_GEM_DOMAIN_CPU:
 	default:
-		*cpu_mem += amdgpu_bo_size(bo);
+		stats->cpu += size;
 		break;
 	}
+
+	if (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) {
+		stats->requested_vram += size;
+		if (bo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
+			stats->requested_visible_vram += size;
+
+		if (domain != AMDGPU_GEM_DOMAIN_VRAM) {
+			stats->evicted_vram += size;
+			if (bo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
+				stats->evicted_visible_vram += size;
+		}
+	} else if (bo->preferred_domains & AMDGPU_GEM_DOMAIN_GTT) {
+		stats->requested_gtt += size;
+	}
 }
 
 /**

drivers/gpu/drm/amd/amdgpu/amdgpu_object.h

@@ -126,6 +126,27 @@ struct amdgpu_bo_vm {
 	struct amdgpu_vm_bo_base        entries[];
 };
 
+struct amdgpu_mem_stats {
+	/* current VRAM usage, includes visible VRAM */
+	uint64_t vram;
+	/* current visible VRAM usage */
+	uint64_t visible_vram;
+	/* current GTT usage */
+	uint64_t gtt;
+	/* current system memory usage */
+	uint64_t cpu;
+	/* sum of evicted buffers, includes visible VRAM */
+	uint64_t evicted_vram;
+	/* sum of evicted buffers due to CPU access */
+	uint64_t evicted_visible_vram;
+	/* how much userspace asked for, includes vis.VRAM */
+	uint64_t requested_vram;
+	/* how much userspace asked for */
+	uint64_t requested_visible_vram;
+	/* how much userspace asked for */
+	uint64_t requested_gtt;
+};
+
 static inline struct amdgpu_bo *ttm_to_amdgpu_bo(struct ttm_buffer_object *tbo)
 {
 	return container_of(tbo, struct amdgpu_bo, tbo);
@@ -325,8 +346,8 @@ int amdgpu_bo_sync_wait_resv(struct amdgpu_device *adev, struct dma_resv *resv,
 int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr);
 u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
 u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
-void amdgpu_bo_get_memory(struct amdgpu_bo *bo, uint64_t *vram_mem,
-				uint64_t *gtt_mem, uint64_t *cpu_mem);
+void amdgpu_bo_get_memory(struct amdgpu_bo *bo,
+			  struct amdgpu_mem_stats *stats);
 void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo_vm *vmbo);
 int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow,
 			     struct dma_fence **fence);

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

@@ -920,8 +920,8 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	return r;
 }
 
-void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem,
-				uint64_t *gtt_mem, uint64_t *cpu_mem)
+void amdgpu_vm_get_memory(struct amdgpu_vm *vm,
+			  struct amdgpu_mem_stats *stats)
 {
 	struct amdgpu_bo_va *bo_va, *tmp;
 
@@ -929,41 +929,36 @@ void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem,
 	list_for_each_entry_safe(bo_va, tmp, &vm->idle, base.vm_status) {
 		if (!bo_va->base.bo)
 			continue;
-		amdgpu_bo_get_memory(bo_va->base.bo, vram_mem,
-				gtt_mem, cpu_mem);
+		amdgpu_bo_get_memory(bo_va->base.bo, stats);
 	}
 	list_for_each_entry_safe(bo_va, tmp, &vm->evicted, base.vm_status) {
 		if (!bo_va->base.bo)
 			continue;
-		amdgpu_bo_get_memory(bo_va->base.bo, vram_mem,
-				gtt_mem, cpu_mem);
+		amdgpu_bo_get_memory(bo_va->base.bo, stats);
 	}
 	list_for_each_entry_safe(bo_va, tmp, &vm->relocated, base.vm_status) {
 		if (!bo_va->base.bo)
 			continue;
-		amdgpu_bo_get_memory(bo_va->base.bo, vram_mem,
-				gtt_mem, cpu_mem);
+		amdgpu_bo_get_memory(bo_va->base.bo, stats);
 	}
 	list_for_each_entry_safe(bo_va, tmp, &vm->moved, base.vm_status) {
 		if (!bo_va->base.bo)
 			continue;
-		amdgpu_bo_get_memory(bo_va->base.bo, vram_mem,
-				gtt_mem, cpu_mem);
+		amdgpu_bo_get_memory(bo_va->base.bo, stats);
 	}
 	list_for_each_entry_safe(bo_va, tmp, &vm->invalidated, base.vm_status) {
 		if (!bo_va->base.bo)
 			continue;
-		amdgpu_bo_get_memory(bo_va->base.bo, vram_mem,
-				gtt_mem, cpu_mem);
+		amdgpu_bo_get_memory(bo_va->base.bo, stats);
 	}
 	list_for_each_entry_safe(bo_va, tmp, &vm->done, base.vm_status) {
 		if (!bo_va->base.bo)
 			continue;
-		amdgpu_bo_get_memory(bo_va->base.bo, vram_mem,
-				gtt_mem, cpu_mem);
+		amdgpu_bo_get_memory(bo_va->base.bo, stats);
 	}
 	spin_unlock(&vm->status_lock);
 }
+
 /**
  * amdgpu_vm_bo_update - update all BO mappings in the vm page table
  *

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h

@@ -40,6 +40,7 @@ struct amdgpu_bo_va;
 struct amdgpu_job;
 struct amdgpu_bo_list_entry;
 struct amdgpu_bo_vm;
+struct amdgpu_mem_stats;
 
 /*
  * GPUVM handling
@@ -457,8 +458,8 @@ void amdgpu_vm_set_task_info(struct amdgpu_vm *vm);
 
 void amdgpu_vm_move_to_lru_tail(struct amdgpu_device *adev,
 				struct amdgpu_vm *vm);
-void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem,
-				uint64_t *gtt_mem, uint64_t *cpu_mem);
+void amdgpu_vm_get_memory(struct amdgpu_vm *vm,
+			  struct amdgpu_mem_stats *stats);
 
 int amdgpu_vm_pt_clear(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		       struct amdgpu_bo_vm *vmbo, bool immediate);