Merge tag 'drm-misc-next-2025-01-06' of https://gitlab.freedesktop.org/drm/misc/kernel into drm-next

drm-misc-next for 6.14:

UAPI Changes:
- Clarify drm memory stats documentation

Cross-subsystem Changes:

Core Changes:
 - sched: Documentation fixes,

Driver Changes:
 - amdgpu: Track BO memory stats at runtime
 - amdxdna: Various fixes
 - hisilicon: New HIBMC driver
 - bridges:
   - Provide default implementation of atomic_check for HDMI bridges
   - it605: HDCP improvements, MCCS Support

Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Maxime Ripard <mripard@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250106-augmented-kakapo-of-action-0cf000@houat

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

@@ -1105,7 +1105,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 	 * We can't use gang submit on with reserved VMIDs when the VM changes
 	 * can't be invalidated by more than one engine at the same time.
 	 */
-	if (p->gang_size > 1 && !p->adev->vm_manager.concurrent_flush) {
+	if (p->gang_size > 1 && !adev->vm_manager.concurrent_flush) {
 		for (i = 0; i < p->gang_size; ++i) {
 			struct drm_sched_entity *entity = p->entities[i];
 			struct drm_gpu_scheduler *sched = entity->rq->sched;
@@ -1189,7 +1189,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 			if (!bo)
 				continue;
 
-			amdgpu_vm_bo_invalidate(adev, bo, false);
+			amdgpu_vm_bo_invalidate(bo, false);
 		}
 	}
 

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

@@ -36,6 +36,7 @@
 #include "amdgpu_gem.h"
 #include "amdgpu_dma_buf.h"
 #include "amdgpu_xgmi.h"
+#include "amdgpu_vm.h"
 #include <drm/amdgpu_drm.h>
 #include <drm/ttm/ttm_tt.h>
 #include <linux/dma-buf.h>
@@ -60,6 +61,8 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf,
 	if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0)
 		attach->peer2peer = false;
 
+	amdgpu_vm_bo_update_shared(bo);
+
 	return 0;
 }
 
@@ -345,7 +348,7 @@ amdgpu_dma_buf_move_notify(struct dma_buf_attachment *attach)
 	/* FIXME: This should be after the "if", but needs a fix to make sure
 	 * DMABuf imports are initialized in the right VM list.
 	 */
-	amdgpu_vm_bo_invalidate(adev, bo, false);
+	amdgpu_vm_bo_invalidate(bo, false);
 	if (!bo->tbo.resource || bo->tbo.resource->mem_type == TTM_PL_SYSTEM)
 		return;
 

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

@@ -60,7 +60,7 @@ void amdgpu_show_fdinfo(struct drm_printer *p, struct drm_file *file)
 	struct amdgpu_fpriv *fpriv = file->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
 
-	struct amdgpu_mem_stats stats[__AMDGPU_PL_LAST + 1] = { };
+	struct amdgpu_mem_stats stats[__AMDGPU_PL_NUM];
 	ktime_t usage[AMDGPU_HW_IP_NUM];
 	const char *pl_name[] = {
 		[TTM_PL_VRAM] = "vram",
@@ -72,15 +72,8 @@ void amdgpu_show_fdinfo(struct drm_printer *p, struct drm_file *file)
 		[AMDGPU_PL_DOORBELL] = "doorbell",
 	};
 	unsigned int hw_ip, i;
-	int ret;
-
-	ret = amdgpu_bo_reserve(vm->root.bo, false);
-	if (ret)
-		return;
-
-	amdgpu_vm_get_memory(vm, stats, ARRAY_SIZE(stats));
-	amdgpu_bo_unreserve(vm->root.bo);
 
+	amdgpu_vm_get_memory(vm, stats);
 	amdgpu_ctx_mgr_usage(&fpriv->ctx_mgr, usage);
 
 	/*
@@ -114,9 +107,11 @@ void amdgpu_show_fdinfo(struct drm_printer *p, struct drm_file *file)
 	drm_printf(p, "amd-evicted-vram:\t%llu KiB\n",
 		   stats[TTM_PL_VRAM].evicted/1024UL);
 	drm_printf(p, "amd-requested-vram:\t%llu KiB\n",
-		   stats[TTM_PL_VRAM].requested/1024UL);
+		   (stats[TTM_PL_VRAM].drm.shared +
+		    stats[TTM_PL_VRAM].drm.private) / 1024UL);
 	drm_printf(p, "amd-requested-gtt:\t%llu KiB\n",
-		   stats[TTM_PL_TT].requested/1024UL);
+		   (stats[TTM_PL_TT].drm.shared +
+		    stats[TTM_PL_TT].drm.private) / 1024UL);
 
 	for (hw_ip = 0; hw_ip < AMDGPU_HW_IP_NUM; ++hw_ip) {
 		if (!usage[hw_ip])

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

@@ -42,6 +42,7 @@
 #include "amdgpu_dma_buf.h"
 #include "amdgpu_hmm.h"
 #include "amdgpu_xgmi.h"
+#include "amdgpu_vm.h"
 
 static vm_fault_t amdgpu_gem_fault(struct vm_fault *vmf)
 {
@@ -179,6 +180,7 @@ static int amdgpu_gem_object_open(struct drm_gem_object *obj,
 	if (r)
 		return r;
 
+	amdgpu_vm_bo_update_shared(abo);
 	bo_va = amdgpu_vm_bo_find(vm, abo);
 	if (!bo_va)
 		bo_va = amdgpu_vm_bo_add(adev, vm, abo);
@@ -252,6 +254,7 @@ static void amdgpu_gem_object_close(struct drm_gem_object *obj,
 		goto out_unlock;
 
 	amdgpu_vm_bo_del(adev, bo_va);
+	amdgpu_vm_bo_update_shared(bo);
 	if (!amdgpu_vm_ready(vm))
 		goto out_unlock;
 
@@ -839,7 +842,6 @@ error:
 int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 			struct drm_file *filp)
 {
-	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct drm_amdgpu_gem_op *args = data;
 	struct drm_gem_object *gobj;
 	struct amdgpu_vm_bo_base *base;
@@ -899,7 +901,7 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 			robj->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
 
 		if (robj->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID)
-			amdgpu_vm_bo_invalidate(adev, robj, true);
+			amdgpu_vm_bo_invalidate(robj, true);
 
 		amdgpu_bo_unreserve(robj);
 		break;

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

@@ -1251,7 +1251,6 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 			   bool evict,
 			   struct ttm_resource *new_mem)
 {
-	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
 	struct ttm_resource *old_mem = bo->resource;
 	struct amdgpu_bo *abo;
 
@@ -1259,7 +1258,7 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 		return;
 
 	abo = ttm_to_amdgpu_bo(bo);
-	amdgpu_vm_bo_invalidate(adev, abo, evict);
+	amdgpu_vm_bo_move(abo, new_mem, evict);
 
 	amdgpu_bo_kunmap(abo);
 
@@ -1272,75 +1271,6 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 			     old_mem ? old_mem->mem_type : -1);
 }
 
-void amdgpu_bo_get_memory(struct amdgpu_bo *bo,
-			  struct amdgpu_mem_stats *stats,
-			  unsigned int sz)
-{
-	const unsigned int domain_to_pl[] = {
-		[ilog2(AMDGPU_GEM_DOMAIN_CPU)]	    = TTM_PL_SYSTEM,
-		[ilog2(AMDGPU_GEM_DOMAIN_GTT)]	    = TTM_PL_TT,
-		[ilog2(AMDGPU_GEM_DOMAIN_VRAM)]	    = TTM_PL_VRAM,
-		[ilog2(AMDGPU_GEM_DOMAIN_GDS)]	    = AMDGPU_PL_GDS,
-		[ilog2(AMDGPU_GEM_DOMAIN_GWS)]	    = AMDGPU_PL_GWS,
-		[ilog2(AMDGPU_GEM_DOMAIN_OA)]	    = AMDGPU_PL_OA,
-		[ilog2(AMDGPU_GEM_DOMAIN_DOORBELL)] = AMDGPU_PL_DOORBELL,
-	};
-	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	struct ttm_resource *res = bo->tbo.resource;
-	struct drm_gem_object *obj = &bo->tbo.base;
-	uint64_t size = amdgpu_bo_size(bo);
-	unsigned int type;
-
-	if (!res) {
-		/*
-		 * If no backing store use one of the preferred domain for basic
-		 * stats. We take the MSB since that should give a reasonable
-		 * view.
-		 */
-		BUILD_BUG_ON(TTM_PL_VRAM < TTM_PL_TT ||
-			     TTM_PL_VRAM < TTM_PL_SYSTEM);
-		type = fls(bo->preferred_domains & AMDGPU_GEM_DOMAIN_MASK);
-		if (!type)
-			return;
-		type--;
-		if (drm_WARN_ON_ONCE(&adev->ddev,
-				     type >= ARRAY_SIZE(domain_to_pl)))
-			return;
-		type = domain_to_pl[type];
-	} else {
-		type = res->mem_type;
-	}
-
-	if (drm_WARN_ON_ONCE(&adev->ddev, type >= sz))
-		return;
-
-	/* DRM stats common fields: */
-
-	if (drm_gem_object_is_shared_for_memory_stats(obj))
-		stats[type].drm.shared += size;
-	else
-		stats[type].drm.private += size;
-
-	if (res) {
-		stats[type].drm.resident += size;
-
-		if (!dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_BOOKKEEP))
-			stats[type].drm.active += size;
-		else if (bo->flags & AMDGPU_GEM_CREATE_DISCARDABLE)
-			stats[type].drm.purgeable += size;
-	}
-
-	/* amdgpu specific stats: */
-
-	if (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) {
-		stats[TTM_PL_VRAM].requested += size;
-		if (type != TTM_PL_VRAM)
-			stats[TTM_PL_VRAM].evicted += size;
-	} else if (bo->preferred_domains & AMDGPU_GEM_DOMAIN_GTT) {
-		stats[TTM_PL_TT].requested += size;
-	}
-}
-
 /**
  * amdgpu_bo_release_notify - notification about a BO being released
  * @bo: pointer to a buffer object
@@ -1555,6 +1485,45 @@ u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo)
 	return amdgpu_gmc_sign_extend(offset);
 }
 
+/**
+ * amdgpu_bo_mem_stats_placement - bo placement for memory accounting
+ * @bo:	the buffer object we should look at
+ *
+ * BO can have multiple preferred placements, to avoid double counting we want
+ * to file it under a single placement for memory stats.
+ * Luckily, if we take the highest set bit in preferred_domains the result is
+ * quite sensible.
+ *
+ * Returns:
+ * Which of the placements should the BO be accounted under.
+ */
+uint32_t amdgpu_bo_mem_stats_placement(struct amdgpu_bo *bo)
+{
+	uint32_t domain = bo->preferred_domains & AMDGPU_GEM_DOMAIN_MASK;
+
+	if (!domain)
+		return TTM_PL_SYSTEM;
+
+	switch (rounddown_pow_of_two(domain)) {
+	case AMDGPU_GEM_DOMAIN_CPU:
+		return TTM_PL_SYSTEM;
+	case AMDGPU_GEM_DOMAIN_GTT:
+		return TTM_PL_TT;
+	case AMDGPU_GEM_DOMAIN_VRAM:
+		return TTM_PL_VRAM;
+	case AMDGPU_GEM_DOMAIN_GDS:
+		return AMDGPU_PL_GDS;
+	case AMDGPU_GEM_DOMAIN_GWS:
+		return AMDGPU_PL_GWS;
+	case AMDGPU_GEM_DOMAIN_OA:
+		return AMDGPU_PL_OA;
+	case AMDGPU_GEM_DOMAIN_DOORBELL:
+		return AMDGPU_PL_DOORBELL;
+	default:
+		return TTM_PL_SYSTEM;
+	}
+}
+
 /**
  * amdgpu_bo_get_preferred_domain - get preferred domain
  * @adev: amdgpu device object

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

@@ -305,9 +305,7 @@ 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,
-			  struct amdgpu_mem_stats *stats,
-			  unsigned int size);
+uint32_t amdgpu_bo_mem_stats_placement(struct amdgpu_bo *bo);
 uint32_t amdgpu_bo_get_preferred_domain(struct amdgpu_device *adev,
 					    uint32_t domain);
 

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

@@ -26,15 +26,15 @@
 
 #include <linux/dma-direction.h>
 #include <drm/gpu_scheduler.h>
+#include <drm/ttm/ttm_placement.h>
 #include "amdgpu_vram_mgr.h"
-#include "amdgpu.h"
 
 #define AMDGPU_PL_GDS		(TTM_PL_PRIV + 0)
 #define AMDGPU_PL_GWS		(TTM_PL_PRIV + 1)
 #define AMDGPU_PL_OA		(TTM_PL_PRIV + 2)
 #define AMDGPU_PL_PREEMPT	(TTM_PL_PRIV + 3)
 #define AMDGPU_PL_DOORBELL	(TTM_PL_PRIV + 4)
-#define __AMDGPU_PL_LAST	(TTM_PL_PRIV + 4)
+#define __AMDGPU_PL_NUM	(TTM_PL_PRIV + 5)
 
 #define AMDGPU_GTT_MAX_TRANSFER_SIZE	512
 #define AMDGPU_GTT_NUM_TRANSFER_WINDOWS	2

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

@@ -36,6 +36,7 @@
 #include <drm/ttm/ttm_tt.h>
 #include <drm/drm_exec.h>
 #include "amdgpu.h"
+#include "amdgpu_vm.h"
 #include "amdgpu_trace.h"
 #include "amdgpu_amdkfd.h"
 #include "amdgpu_gmc.h"
@@ -310,6 +311,111 @@ static void amdgpu_vm_bo_reset_state_machine(struct amdgpu_vm *vm)
 	spin_unlock(&vm->status_lock);
 }
 
+/**
+ * amdgpu_vm_update_shared - helper to update shared memory stat
+ * @base: base structure for tracking BO usage in a VM
+ *
+ * Takes the vm status_lock and updates the shared memory stat. If the basic
+ * stat changed (e.g. buffer was moved) amdgpu_vm_update_stats need to be called
+ * as well.
+ */
+static void amdgpu_vm_update_shared(struct amdgpu_vm_bo_base *base)
+{
+	struct amdgpu_vm *vm = base->vm;
+	struct amdgpu_bo *bo = base->bo;
+	uint64_t size = amdgpu_bo_size(bo);
+	uint32_t bo_memtype = amdgpu_bo_mem_stats_placement(bo);
+	bool shared;
+
+	spin_lock(&vm->status_lock);
+	shared = drm_gem_object_is_shared_for_memory_stats(&bo->tbo.base);
+	if (base->shared != shared) {
+		base->shared = shared;
+		if (shared) {
+			vm->stats[bo_memtype].drm.shared += size;
+			vm->stats[bo_memtype].drm.private -= size;
+		} else {
+			vm->stats[bo_memtype].drm.shared -= size;
+			vm->stats[bo_memtype].drm.private += size;
+		}
+	}
+	spin_unlock(&vm->status_lock);
+}
+
+/**
+ * amdgpu_vm_bo_update_shared - callback when bo gets shared/unshared
+ * @bo: amdgpu buffer object
+ *
+ * Update the per VM stats for all the vm if needed from private to shared or
+ * vice versa.
+ */
+void amdgpu_vm_bo_update_shared(struct amdgpu_bo *bo)
+{
+	struct amdgpu_vm_bo_base *base;
+
+	for (base = bo->vm_bo; base; base = base->next)
+		amdgpu_vm_update_shared(base);
+}
+
+/**
+ * amdgpu_vm_update_stats_locked - helper to update normal memory stat
+ * @base: base structure for tracking BO usage in a VM
+ * @res:  the ttm_resource to use for the purpose of accounting, may or may not
+ *        be bo->tbo.resource
+ * @sign: if we should add (+1) or subtract (-1) from the stat
+ *
+ * Caller need to have the vm status_lock held. Useful for when multiple update
+ * need to happen at the same time.
+ */
+static void amdgpu_vm_update_stats_locked(struct amdgpu_vm_bo_base *base,
+			    struct ttm_resource *res, int sign)
+{
+	struct amdgpu_vm *vm = base->vm;
+	struct amdgpu_bo *bo = base->bo;
+	int64_t size = sign * amdgpu_bo_size(bo);
+	uint32_t bo_memtype = amdgpu_bo_mem_stats_placement(bo);
+
+	/* For drm-total- and drm-shared-, BO are accounted by their preferred
+	 * placement, see also amdgpu_bo_mem_stats_placement.
+	 */
+	if (base->shared)
+		vm->stats[bo_memtype].drm.shared += size;
+	else
+		vm->stats[bo_memtype].drm.private += size;
+
+	if (res && res->mem_type < __AMDGPU_PL_NUM) {
+		uint32_t res_memtype = res->mem_type;
+
+		vm->stats[res_memtype].drm.resident += size;
+		/* BO only count as purgeable if it is resident,
+		 * since otherwise there's nothing to purge.
+		 */
+		if (bo->flags & AMDGPU_GEM_CREATE_DISCARDABLE)
+			vm->stats[res_memtype].drm.purgeable += size;
+		if (!(bo->preferred_domains & amdgpu_mem_type_to_domain(res_memtype)))
+			vm->stats[bo_memtype].evicted += size;
+	}
+}
+
+/**
+ * amdgpu_vm_update_stats - helper to update normal memory stat
+ * @base: base structure for tracking BO usage in a VM
+ * @res:  the ttm_resource to use for the purpose of accounting, may or may not
+ *        be bo->tbo.resource
+ * @sign: if we should add (+1) or subtract (-1) from the stat
+ *
+ * Updates the basic memory stat when bo is added/deleted/moved.
+ */
+void amdgpu_vm_update_stats(struct amdgpu_vm_bo_base *base,
+			    struct ttm_resource *res, int sign)
+{
+	struct amdgpu_vm *vm = base->vm;
+
+	spin_lock(&vm->status_lock);
+	amdgpu_vm_update_stats_locked(base, res, sign);
+	spin_unlock(&vm->status_lock);
+}
+
 /**
  * amdgpu_vm_bo_base_init - Adds bo to the list of bos associated with the vm
  *
@@ -333,6 +439,11 @@ void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
 	base->next = bo->vm_bo;
 	bo->vm_bo = base;
 
+	spin_lock(&vm->status_lock);
+	base->shared = drm_gem_object_is_shared_for_memory_stats(&bo->tbo.base);
+	amdgpu_vm_update_stats_locked(base, bo->tbo.resource, +1);
+	spin_unlock(&vm->status_lock);
+
 	if (!amdgpu_vm_is_bo_always_valid(vm, bo))
 		return;
 
@@ -1083,53 +1194,11 @@ error_free:
 	return r;
 }
 
-static void amdgpu_vm_bo_get_memory(struct amdgpu_bo_va *bo_va,
-				    struct amdgpu_mem_stats *stats,
-				    unsigned int size)
-{
-	struct amdgpu_vm *vm = bo_va->base.vm;
-	struct amdgpu_bo *bo = bo_va->base.bo;
-
-	if (!bo)
-		return;
-
-	/*
-	 * For now ignore BOs which are currently locked and potentially
-	 * changing their location.
-	 */
-	if (!amdgpu_vm_is_bo_always_valid(vm, bo) &&
-	    !dma_resv_trylock(bo->tbo.base.resv))
-		return;
-
-	amdgpu_bo_get_memory(bo, stats, size);
-	if (!amdgpu_vm_is_bo_always_valid(vm, bo))
-		dma_resv_unlock(bo->tbo.base.resv);
-}
-
 void amdgpu_vm_get_memory(struct amdgpu_vm *vm,
-			  struct amdgpu_mem_stats *stats,
-			  unsigned int size)
+			  struct amdgpu_mem_stats stats[__AMDGPU_PL_NUM])
 {
-	struct amdgpu_bo_va *bo_va, *tmp;
-
 	spin_lock(&vm->status_lock);
-	list_for_each_entry_safe(bo_va, tmp, &vm->idle, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats, size);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->evicted, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats, size);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->relocated, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats, size);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->moved, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats, size);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->invalidated, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats, size);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->done, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats, size);
+	memcpy(stats, vm->stats, sizeof(*stats) * __AMDGPU_PL_NUM);
 	spin_unlock(&vm->status_lock);
 }
 
@@ -2075,6 +2144,7 @@ void amdgpu_vm_bo_del(struct amdgpu_device *adev,
 			if (*base != &bo_va->base)
 				continue;
 
+			amdgpu_vm_update_stats(*base, bo->tbo.resource, -1);
 			*base = bo_va->base.next;
 			break;
 		}
@@ -2143,14 +2213,12 @@ bool amdgpu_vm_evictable(struct amdgpu_bo *bo)
 /**
  * amdgpu_vm_bo_invalidate - mark the bo as invalid
  *
- * @adev: amdgpu_device pointer
  * @bo: amdgpu buffer object
  * @evicted: is the BO evicted
  *
  * Mark @bo as invalid.
  */
-void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
-			     struct amdgpu_bo *bo, bool evicted)
+void amdgpu_vm_bo_invalidate(struct amdgpu_bo *bo, bool evicted)
 {
 	struct amdgpu_vm_bo_base *bo_base;
 
@@ -2175,6 +2243,32 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
 	}
 }
 
+/**
+ * amdgpu_vm_bo_move - handle BO move
+ *
+ * @bo: amdgpu buffer object
+ * @new_mem: the new placement of the BO move
+ * @evicted: is the BO evicted
+ *
+ * Update the memory stats for the new placement and mark @bo as invalid.
+ */
+void amdgpu_vm_bo_move(struct amdgpu_bo *bo, struct ttm_resource *new_mem,
+		       bool evicted)
+{
+	struct amdgpu_vm_bo_base *bo_base;
+
+	for (bo_base = bo->vm_bo; bo_base; bo_base = bo_base->next) {
+		struct amdgpu_vm *vm = bo_base->vm;
+
+		spin_lock(&vm->status_lock);
+		amdgpu_vm_update_stats_locked(bo_base, bo->tbo.resource, -1);
+		amdgpu_vm_update_stats_locked(bo_base, new_mem, +1);
+		spin_unlock(&vm->status_lock);
+	}
+
+	amdgpu_vm_bo_invalidate(bo, evicted);
+}
+
 /**
  * amdgpu_vm_get_block_size - calculate VM page table size as power of two
  *
@@ -2594,6 +2688,16 @@ void amdgpu_vm_release_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	vm->is_compute_context = false;
 }
 
+static int amdgpu_vm_stats_is_zero(struct amdgpu_vm *vm)
+{
+	for (int i = 0; i < __AMDGPU_PL_NUM; ++i) {
+		if (!(drm_memory_stats_is_zero(&vm->stats[i].drm) &&
+		      vm->stats[i].evicted == 0))
+			return false;
+	}
+	return true;
+}
+
 /**
  * amdgpu_vm_fini - tear down a vm instance
  *
@@ -2617,7 +2721,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 
 	root = amdgpu_bo_ref(vm->root.bo);
 	amdgpu_bo_reserve(root, true);
-	amdgpu_vm_put_task_info(vm->task_info);
 	amdgpu_vm_set_pasid(adev, vm, 0);
 	dma_fence_wait(vm->last_unlocked, false);
 	dma_fence_put(vm->last_unlocked);
@@ -2666,6 +2769,16 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	}
 
 	ttm_lru_bulk_move_fini(&adev->mman.bdev, &vm->lru_bulk_move);
+
+	if (!amdgpu_vm_stats_is_zero(vm)) {
+		struct amdgpu_task_info *ti = vm->task_info;
+
+		dev_warn(adev->dev,
+			 "VM memory stats for proc %s(%d) task %s(%d) is non-zero when fini\n",
+			 ti->process_name, ti->pid, ti->task_name, ti->tgid);
+	}
+
+	amdgpu_vm_put_task_info(vm->task_info);
 }
 
 /**

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

@@ -35,6 +35,7 @@
 #include "amdgpu_sync.h"
 #include "amdgpu_ring.h"
 #include "amdgpu_ids.h"
+#include "amdgpu_ttm.h"
 
 struct drm_exec;
 
@@ -202,9 +203,13 @@ struct amdgpu_vm_bo_base {
 	/* protected by bo being reserved */
 	struct amdgpu_vm_bo_base	*next;
 
-	/* protected by spinlock */
+	/* protected by vm status_lock */
 	struct list_head		vm_status;
 
+	/* if the bo is counted as shared in mem stats
+	 * protected by vm status_lock */
+	bool				shared;
+
 	/* protected by the BO being reserved */
 	bool				moved;
 };
@@ -324,10 +329,7 @@ struct amdgpu_vm_fault_info {
 struct amdgpu_mem_stats {
 	struct drm_memory_stats drm;
 
-	/* buffers that requested this placement */
-	uint64_t requested;
-	/* buffers that requested this placement
-	 * but are currently evicted */
+	/* buffers that requested this placement but are currently evicted */
 	uint64_t evicted;
 };
 
@@ -345,6 +347,9 @@ struct amdgpu_vm {
 	/* Lock to protect vm_bo add/del/move on all lists of vm */
 	spinlock_t		status_lock;
 
+	/* Memory statistics for this vm, protected by status_lock */
+	struct amdgpu_mem_stats stats[__AMDGPU_PL_NUM];
+
 	/* Per-VM and PT BOs who needs a validation */
 	struct list_head	evicted;
 
@@ -524,8 +529,12 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
 			struct amdgpu_bo_va *bo_va,
 			bool clear);
 bool amdgpu_vm_evictable(struct amdgpu_bo *bo);
-void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
-			     struct amdgpu_bo *bo, bool evicted);
+void amdgpu_vm_bo_invalidate(struct amdgpu_bo *bo, bool evicted);
+void amdgpu_vm_update_stats(struct amdgpu_vm_bo_base *base,
+			    struct ttm_resource *new_res, int sign);
+void amdgpu_vm_bo_update_shared(struct amdgpu_bo *bo);
+void amdgpu_vm_bo_move(struct amdgpu_bo *bo, struct ttm_resource *new_mem,
+		       bool evicted);
 uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr);
 struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm,
 				       struct amdgpu_bo *bo);
@@ -576,8 +585,7 @@ 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,
-			  struct amdgpu_mem_stats *stats,
-			  unsigned int size);
+			  struct amdgpu_mem_stats stats[__AMDGPU_PL_NUM]);
 
 int amdgpu_vm_pt_clear(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		       struct amdgpu_bo_vm *vmbo, bool immediate);

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

@@ -537,6 +537,7 @@ static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry)
 	if (!entry->bo)
 		return;
 
+	amdgpu_vm_update_stats(entry, entry->bo->tbo.resource, -1);
 	entry->bo->vm_bo = NULL;
 	ttm_bo_set_bulk_move(&entry->bo->tbo, NULL);