drm/amdkfd: change system memory overcommit limit

It is to improve system limit by:
1. replacing userptrlimit with a total memory limit that
conunts TTM memory usage and userptr usage.
2. counting acc size for all BOs.
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -46,9 +46,9 @@
 /* Impose limit on how much memory KFD can use */
 static struct {
 	uint64_t max_system_mem_limit;
-	uint64_t max_userptr_mem_limit;
+	uint64_t max_ttm_mem_limit;
 	int64_t system_mem_used;
-	int64_t userptr_mem_used;
+	int64_t ttm_mem_used;
 	spinlock_t mem_limit_lock;
 } kfd_mem_limit;
 
@@ -90,8 +90,8 @@ static bool check_if_add_bo_to_vm(struct amdgpu_vm *avm,
 }
 
 /* Set memory usage limits. Current, limits are
- *  System (kernel) memory - 3/8th System RAM
- *  Userptr memory - 3/4th System RAM
+ *  System (TTM + userptr) memory - 3/4th System RAM
+ *  TTM memory - 3/8th System RAM
  */
 void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
 {
@@ -103,48 +103,54 @@ void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
 	mem *= si.mem_unit;
 
 	spin_lock_init(&kfd_mem_limit.mem_limit_lock);
-	kfd_mem_limit.max_system_mem_limit = (mem >> 1) - (mem >> 3);
-	kfd_mem_limit.max_userptr_mem_limit = mem - (mem >> 2);
-	pr_debug("Kernel memory limit %lluM, userptr limit %lluM\n",
+	kfd_mem_limit.max_system_mem_limit = (mem >> 1) + (mem >> 2);
+	kfd_mem_limit.max_ttm_mem_limit = (mem >> 1) - (mem >> 3);
+	pr_debug("Kernel memory limit %lluM, TTM limit %lluM\n",
 		(kfd_mem_limit.max_system_mem_limit >> 20),
-		(kfd_mem_limit.max_userptr_mem_limit >> 20));
+		(kfd_mem_limit.max_ttm_mem_limit >> 20));
 }
 
 static int amdgpu_amdkfd_reserve_system_mem_limit(struct amdgpu_device *adev,
-					      uint64_t size, u32 domain)
+		uint64_t size, u32 domain, bool sg)
 {
-	size_t acc_size;
+	size_t acc_size, system_mem_needed, ttm_mem_needed;
 	int ret = 0;
 
 	acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size,
 				       sizeof(struct amdgpu_bo));
 
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
+
 	if (domain == AMDGPU_GEM_DOMAIN_GTT) {
-		if (kfd_mem_limit.system_mem_used + (acc_size + size) >
-			kfd_mem_limit.max_system_mem_limit) {
-			ret = -ENOMEM;
-			goto err_no_mem;
-		}
-		kfd_mem_limit.system_mem_used += (acc_size + size);
-	} else if (domain == AMDGPU_GEM_DOMAIN_CPU) {
-		if ((kfd_mem_limit.system_mem_used + acc_size >
-			kfd_mem_limit.max_system_mem_limit) ||
-			(kfd_mem_limit.userptr_mem_used + (size + acc_size) >
-			kfd_mem_limit.max_userptr_mem_limit)) {
-			ret = -ENOMEM;
-			goto err_no_mem;
-		}
-		kfd_mem_limit.system_mem_used += acc_size;
-		kfd_mem_limit.userptr_mem_used += size;
+		/* TTM GTT memory */
+		system_mem_needed = acc_size + size;
+		ttm_mem_needed = acc_size + size;
+	} else if (domain == AMDGPU_GEM_DOMAIN_CPU && !sg) {
+		/* Userptr */
+		system_mem_needed = acc_size + size;
+		ttm_mem_needed = acc_size;
+	} else {
+		/* VRAM and SG */
+		system_mem_needed = acc_size;
+		ttm_mem_needed = acc_size;
+	}
+
+	if ((kfd_mem_limit.system_mem_used + system_mem_needed >
+		kfd_mem_limit.max_system_mem_limit) ||
+		(kfd_mem_limit.ttm_mem_used + ttm_mem_needed >
+		kfd_mem_limit.max_ttm_mem_limit))
+		ret = -ENOMEM;
+	else {
+		kfd_mem_limit.system_mem_used += system_mem_needed;
+		kfd_mem_limit.ttm_mem_used += ttm_mem_needed;
 	}
-err_no_mem:
+
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
 	return ret;
 }
 
 static void unreserve_system_mem_limit(struct amdgpu_device *adev,
-				       uint64_t size, u32 domain)
+		uint64_t size, u32 domain, bool sg)
 {
 	size_t acc_size;
 
@@ -154,14 +160,18 @@ static void unreserve_system_mem_limit(struct amdgpu_device *adev,
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
 	if (domain == AMDGPU_GEM_DOMAIN_GTT) {
 		kfd_mem_limit.system_mem_used -= (acc_size + size);
-	} else if (domain == AMDGPU_GEM_DOMAIN_CPU) {
+		kfd_mem_limit.ttm_mem_used -= (acc_size + size);
+	} else if (domain == AMDGPU_GEM_DOMAIN_CPU && !sg) {
+		kfd_mem_limit.system_mem_used -= (acc_size + size);
+		kfd_mem_limit.ttm_mem_used -= acc_size;
+	} else {
 		kfd_mem_limit.system_mem_used -= acc_size;
-		kfd_mem_limit.userptr_mem_used -= size;
+		kfd_mem_limit.ttm_mem_used -= acc_size;
 	}
 	WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
 		  "kfd system memory accounting unbalanced");
-	WARN_ONCE(kfd_mem_limit.userptr_mem_used < 0,
-		  "kfd userptr memory accounting unbalanced");
+	WARN_ONCE(kfd_mem_limit.ttm_mem_used < 0,
+		  "kfd TTM memory accounting unbalanced");
 
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
 }
@@ -171,16 +181,22 @@ void amdgpu_amdkfd_unreserve_system_memory_limit(struct amdgpu_bo *bo)
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
 
 	if (bo->flags & AMDGPU_AMDKFD_USERPTR_BO) {
-		kfd_mem_limit.system_mem_used -= bo->tbo.acc_size;
-		kfd_mem_limit.userptr_mem_used -= amdgpu_bo_size(bo);
+		kfd_mem_limit.system_mem_used -=
+			(bo->tbo.acc_size + amdgpu_bo_size(bo));
+		kfd_mem_limit.ttm_mem_used -= bo->tbo.acc_size;
 	} else if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT) {
 		kfd_mem_limit.system_mem_used -=
 			(bo->tbo.acc_size + amdgpu_bo_size(bo));
+		kfd_mem_limit.ttm_mem_used -=
+			(bo->tbo.acc_size + amdgpu_bo_size(bo));
+	} else {
+		kfd_mem_limit.system_mem_used -= bo->tbo.acc_size;
+		kfd_mem_limit.ttm_mem_used -= bo->tbo.acc_size;
 	}
 	WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
 		  "kfd system memory accounting unbalanced");
-	WARN_ONCE(kfd_mem_limit.userptr_mem_used < 0,
-		  "kfd userptr memory accounting unbalanced");
+	WARN_ONCE(kfd_mem_limit.ttm_mem_used < 0,
+		  "kfd TTM memory accounting unbalanced");
 
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
 }
@@ -1219,10 +1235,11 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 
 	amdgpu_sync_create(&(*mem)->sync);
 
-	ret = amdgpu_amdkfd_reserve_system_mem_limit(adev, size, alloc_domain);
+	ret = amdgpu_amdkfd_reserve_system_mem_limit(adev, size,
+						     alloc_domain, false);
 	if (ret) {
 		pr_debug("Insufficient system memory\n");
-		goto err_reserve_system_mem;
+		goto err_reserve_limit;
 	}
 
 	pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s\n",
@@ -1270,10 +1287,10 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 allocate_init_user_pages_failed:
 	amdgpu_bo_unref(&bo);
 	/* Don't unreserve system mem limit twice */
-	goto err_reserve_system_mem;
+	goto err_reserve_limit;
 err_bo_create:
-	unreserve_system_mem_limit(adev, size, alloc_domain);
-err_reserve_system_mem:
+	unreserve_system_mem_limit(adev, size, alloc_domain, false);
+err_reserve_limit:
 	mutex_destroy(&(*mem)->lock);
 	kfree(*mem);
 	return ret;