Merge drm/drm-next into drm-misc-next

Backmerging for v5.16-rc5. Resolves a conflict between drm-misc-next
and drm-misc-fixes in the vc4 driver.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>

1 files changed, 149 insertions, 75 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 71acd577803e..5df89a295177 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -60,12 +60,6 @@ static const char * const domain_bit_to_string[] = {
 
 static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work);
 
-
-static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
-{
-	return (struct amdgpu_device *)kgd;
-}
-
 static bool kfd_mem_is_attached(struct amdgpu_vm *avm,
 		struct kgd_mem *mem)
 {
@@ -126,8 +120,19 @@ static size_t amdgpu_amdkfd_acc_size(uint64_t size)
 		PAGE_ALIGN(size);
 }
 
+/**
+ * @amdgpu_amdkfd_reserve_mem_limit() - Decrease available memory by size
+ * of buffer including any reserved for control structures
+ *
+ * @adev: Device to which allocated BO belongs to
+ * @size: Size of buffer, in bytes, encapsulated by B0. This should be
+ * equivalent to amdgpu_bo_size(BO)
+ * @alloc_flag: Flag used in allocating a BO as noted above
+ *
+ * Return: returns -ENOMEM in case of error, ZERO otherwise
+ */
 static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
-		uint64_t size, u32 domain, bool sg)
+		uint64_t size, u32 alloc_flag)
 {
 	uint64_t reserved_for_pt =
 		ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
@@ -137,20 +142,24 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 	acc_size = amdgpu_amdkfd_acc_size(size);
 
 	vram_needed = 0;
-	if (domain == AMDGPU_GEM_DOMAIN_GTT) {
-		/* TTM GTT memory */
+	if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
 		system_mem_needed = acc_size + size;
 		ttm_mem_needed = acc_size + size;
-	} else if (domain == AMDGPU_GEM_DOMAIN_CPU && !sg) {
-		/* Userptr */
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
+		system_mem_needed = acc_size;
+		ttm_mem_needed = acc_size;
+		vram_needed = size;
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
 		system_mem_needed = acc_size + size;
 		ttm_mem_needed = acc_size;
-	} else {
-		/* VRAM and SG */
+	} else if (alloc_flag &
+		   (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+		    KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
 		system_mem_needed = acc_size;
 		ttm_mem_needed = acc_size;
-		if (domain == AMDGPU_GEM_DOMAIN_VRAM)
-			vram_needed = size;
+	} else {
+		pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag);
+		return -ENOMEM;
 	}
 
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
@@ -166,64 +175,72 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 	    (adev->kfd.vram_used + vram_needed >
 	     adev->gmc.real_vram_size - reserved_for_pt)) {
 		ret = -ENOMEM;
-	} else {
-		kfd_mem_limit.system_mem_used += system_mem_needed;
-		kfd_mem_limit.ttm_mem_used += ttm_mem_needed;
-		adev->kfd.vram_used += vram_needed;
+		goto release;
 	}
 
+	/* Update memory accounting by decreasing available system
+	 * memory, TTM memory and GPU memory as computed above
+	 */
+	adev->kfd.vram_used += vram_needed;
+	kfd_mem_limit.system_mem_used += system_mem_needed;
+	kfd_mem_limit.ttm_mem_used += ttm_mem_needed;
+
+release:
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
 	return ret;
 }
 
 static void unreserve_mem_limit(struct amdgpu_device *adev,
-		uint64_t size, u32 domain, bool sg)
+		uint64_t size, u32 alloc_flag)
 {
 	size_t acc_size;
 
 	acc_size = amdgpu_amdkfd_acc_size(size);
 
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
-	if (domain == AMDGPU_GEM_DOMAIN_GTT) {
+
+	if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
 		kfd_mem_limit.system_mem_used -= (acc_size + size);
 		kfd_mem_limit.ttm_mem_used -= (acc_size + size);
-	} else if (domain == AMDGPU_GEM_DOMAIN_CPU && !sg) {
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
+		kfd_mem_limit.system_mem_used -= acc_size;
+		kfd_mem_limit.ttm_mem_used -= acc_size;
+		adev->kfd.vram_used -= size;
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
 		kfd_mem_limit.system_mem_used -= (acc_size + size);
 		kfd_mem_limit.ttm_mem_used -= acc_size;
-	} else {
+	} else if (alloc_flag &
+		   (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+		    KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
 		kfd_mem_limit.system_mem_used -= acc_size;
 		kfd_mem_limit.ttm_mem_used -= acc_size;
-		if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
-			adev->kfd.vram_used -= size;
-			WARN_ONCE(adev->kfd.vram_used < 0,
-				  "kfd VRAM memory accounting unbalanced");
-		}
+	} else {
+		pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag);
+		goto release;
 	}
-	WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
-		  "kfd system memory accounting unbalanced");
+
+	WARN_ONCE(adev->kfd.vram_used < 0,
+		  "KFD VRAM memory accounting unbalanced");
 	WARN_ONCE(kfd_mem_limit.ttm_mem_used < 0,
-		  "kfd TTM memory accounting unbalanced");
+		  "KFD TTM memory accounting unbalanced");
+	WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
+		  "KFD system memory accounting unbalanced");
 
+release:
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
 }
 
 void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	u32 domain = bo->preferred_domains;
-	bool sg = (bo->preferred_domains == AMDGPU_GEM_DOMAIN_CPU);
-
-	if (bo->flags & AMDGPU_AMDKFD_CREATE_USERPTR_BO) {
-		domain = AMDGPU_GEM_DOMAIN_CPU;
-		sg = false;
-	}
+	u32 alloc_flags = bo->kfd_bo->alloc_flags;
+	u64 size = amdgpu_bo_size(bo);
 
-	unreserve_mem_limit(adev, amdgpu_bo_size(bo), domain, sg);
+	unreserve_mem_limit(adev, size, alloc_flags);
 
 	kfree(bo->kfd_bo);
 }
 
-
 /* amdgpu_amdkfd_remove_eviction_fence - Removes eviction fence from BO's
  *  reservation object.
  *
@@ -646,12 +663,6 @@ kfd_mem_attach_dmabuf(struct amdgpu_device *adev, struct kgd_mem *mem,
 	if (IS_ERR(gobj))
 		return PTR_ERR(gobj);
 
-	/* Import takes an extra reference on the dmabuf. Drop it now to
-	 * avoid leaking it. We only need the one reference in
-	 * kgd_mem->dmabuf.
-	 */
-	dma_buf_put(mem->dmabuf);
-
 	*bo = gem_to_amdgpu_bo(gobj);
 	(*bo)->flags |= AMDGPU_GEM_CREATE_PREEMPTIBLE;
 	(*bo)->parent = amdgpu_bo_ref(mem->bo);
@@ -1278,12 +1289,60 @@ create_evict_fence_fail:
 	return ret;
 }
 
-int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct kgd_dev *kgd,
+/**
+ * amdgpu_amdkfd_gpuvm_pin_bo() - Pins a BO using following criteria
+ * @bo: Handle of buffer object being pinned
+ * @domain: Domain into which BO should be pinned
+ *
+ *   - USERPTR BOs are UNPINNABLE and will return error
+ *   - All other BO types (GTT, VRAM, MMIO and DOORBELL) will have their
+ *     PIN count incremented. It is valid to PIN a BO multiple times
+ *
+ * Return: ZERO if successful in pinning, Non-Zero in case of error.
+ */
+static int amdgpu_amdkfd_gpuvm_pin_bo(struct amdgpu_bo *bo, u32 domain)
+{
+	int ret = 0;
+
+	ret = amdgpu_bo_reserve(bo, false);
+	if (unlikely(ret))
+		return ret;
+
+	ret = amdgpu_bo_pin_restricted(bo, domain, 0, 0);
+	if (ret)
+		pr_err("Error in Pinning BO to domain: %d\n", domain);
+
+	amdgpu_bo_sync_wait(bo, AMDGPU_FENCE_OWNER_KFD, false);
+	amdgpu_bo_unreserve(bo);
+
+	return ret;
+}
+
+/**
+ * amdgpu_amdkfd_gpuvm_unpin_bo() - Unpins BO using following criteria
+ * @bo: Handle of buffer object being unpinned
+ *
+ *   - Is a illegal request for USERPTR BOs and is ignored
+ *   - All other BO types (GTT, VRAM, MMIO and DOORBELL) will have their
+ *     PIN count decremented. Calls to UNPIN must balance calls to PIN
+ */
+static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo)
+{
+	int ret = 0;
+
+	ret = amdgpu_bo_reserve(bo, false);
+	if (unlikely(ret))
+		return;
+
+	amdgpu_bo_unpin(bo);
+	amdgpu_bo_unreserve(bo);
+}
+
+int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,
 					   struct file *filp, u32 pasid,
 					   void **process_info,
 					   struct dma_fence **ef)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct amdgpu_fpriv *drv_priv;
 	struct amdgpu_vm *avm;
 	int ret;
@@ -1359,12 +1418,12 @@ void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
 	}
 }
 
-void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *drm_priv)
+void amdgpu_amdkfd_gpuvm_release_process_vm(struct amdgpu_device *adev,
+					    void *drm_priv)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct amdgpu_vm *avm;
 
-	if (WARN_ON(!kgd || !drm_priv))
+	if (WARN_ON(!adev || !drm_priv))
 		return;
 
 	avm = drm_priv_to_vm(drm_priv);
@@ -1392,17 +1451,16 @@ uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv)
 }
 
 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
-		struct kgd_dev *kgd, uint64_t va, uint64_t size,
+		struct amdgpu_device *adev, uint64_t va, uint64_t size,
 		void *drm_priv, struct kgd_mem **mem,
 		uint64_t *offset, uint32_t flags)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	enum ttm_bo_type bo_type = ttm_bo_type_device;
 	struct sg_table *sg = NULL;
 	uint64_t user_addr = 0;
 	struct amdgpu_bo *bo;
-	struct drm_gem_object *gobj;
+	struct drm_gem_object *gobj = NULL;
 	u32 domain, alloc_domain;
 	u64 alloc_flags;
 	int ret;
@@ -1460,7 +1518,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 
 	amdgpu_sync_create(&(*mem)->sync);
 
-	ret = amdgpu_amdkfd_reserve_mem_limit(adev, size, alloc_domain, !!sg);
+	ret = amdgpu_amdkfd_reserve_mem_limit(adev, size, flags);
 	if (ret) {
 		pr_debug("Insufficient memory\n");
 		goto err_reserve_limit;
@@ -1506,20 +1564,34 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 	if (offset)
 		*offset = amdgpu_bo_mmap_offset(bo);
 
+	if (flags & (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+			KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
+		ret = amdgpu_amdkfd_gpuvm_pin_bo(bo, AMDGPU_GEM_DOMAIN_GTT);
+		if (ret) {
+			pr_err("Pinning MMIO/DOORBELL BO during ALLOC FAILED\n");
+			goto err_pin_bo;
+		}
+		bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
+		bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
+	}
+
 	return 0;
 
 allocate_init_user_pages_failed:
 	remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info);
+err_pin_bo:
 	drm_vma_node_revoke(&gobj->vma_node, drm_priv);
 err_node_allow:
-	drm_gem_object_put(gobj);
 	/* Don't unreserve system mem limit twice */
 	goto err_reserve_limit;
 err_bo_create:
-	unreserve_mem_limit(adev, size, alloc_domain, !!sg);
+	unreserve_mem_limit(adev, size, flags);
 err_reserve_limit:
 	mutex_destroy(&(*mem)->lock);
-	kfree(*mem);
+	if (gobj)
+		drm_gem_object_put(gobj);
+	else
+		kfree(*mem);
 err:
 	if (sg) {
 		sg_free_table(sg);
@@ -1529,7 +1601,7 @@ err:
 }
 
 int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
-		struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv,
+		struct amdgpu_device *adev, struct kgd_mem *mem, void *drm_priv,
 		uint64_t *size)
 {
 	struct amdkfd_process_info *process_info = mem->process_info;
@@ -1542,6 +1614,14 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 	bool is_imported = false;
 
 	mutex_lock(&mem->lock);
+
+	/* Unpin MMIO/DOORBELL BO's that were pinnned during allocation */
+	if (mem->alloc_flags &
+	    (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+	     KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
+		amdgpu_amdkfd_gpuvm_unpin_bo(mem->bo);
+	}
+
 	mapped_to_gpu_memory = mem->mapped_to_gpu_memory;
 	is_imported = mem->is_imported;
 	mutex_unlock(&mem->lock);
@@ -1621,10 +1701,9 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 }
 
 int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
-		struct kgd_dev *kgd, struct kgd_mem *mem,
+		struct amdgpu_device *adev, struct kgd_mem *mem,
 		void *drm_priv, bool *table_freed)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	int ret;
 	struct amdgpu_bo *bo;
@@ -1751,7 +1830,7 @@ out:
 }
 
 int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
-		struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv)
+		struct amdgpu_device *adev, struct kgd_mem *mem, void *drm_priv)
 {
 	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	struct amdkfd_process_info *process_info = avm->process_info;
@@ -1812,7 +1891,7 @@ out:
 }
 
 int amdgpu_amdkfd_gpuvm_sync_memory(
-		struct kgd_dev *kgd, struct kgd_mem *mem, bool intr)
+		struct amdgpu_device *adev, struct kgd_mem *mem, bool intr)
 {
 	struct amdgpu_sync sync;
 	int ret;
@@ -1828,7 +1907,7 @@ int amdgpu_amdkfd_gpuvm_sync_memory(
 	return ret;
 }
 
-int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_dev *kgd,
+int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct amdgpu_device *adev,
 		struct kgd_mem *mem, void **kptr, uint64_t *size)
 {
 	int ret;
@@ -1884,7 +1963,8 @@ bo_reserve_failed:
 	return ret;
 }
 
-void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_dev *kgd, struct kgd_mem *mem)
+void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct amdgpu_device *adev,
+						  struct kgd_mem *mem)
 {
 	struct amdgpu_bo *bo = mem->bo;
 
@@ -1894,12 +1974,9 @@ void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_dev *kgd, struct kg
 	amdgpu_bo_unreserve(bo);
 }
 
-int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd,
-					      struct kfd_vm_fault_info *mem)
+int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
+					  struct kfd_vm_fault_info *mem)
 {
-	struct amdgpu_device *adev;
-
-	adev = (struct amdgpu_device *)kgd;
 	if (atomic_read(&adev->gmc.vm_fault_info_updated) == 1) {
 		*mem = *adev->gmc.vm_fault_info;
 		mb();
@@ -1908,13 +1985,12 @@ int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd,
 	return 0;
 }
 
-int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd,
+int amdgpu_amdkfd_gpuvm_import_dmabuf(struct amdgpu_device *adev,
 				      struct dma_buf *dma_buf,
 				      uint64_t va, void *drm_priv,
 				      struct kgd_mem **mem, uint64_t *size,
 				      uint64_t *mmap_offset)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
 	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	struct drm_gem_object *obj;
 	struct amdgpu_bo *bo;
@@ -2541,11 +2617,9 @@ int amdgpu_amdkfd_remove_gws_from_process(void *info, void *mem)
 }
 
 /* Returns GPU-specific tiling mode information */
-int amdgpu_amdkfd_get_tile_config(struct kgd_dev *kgd,
+int amdgpu_amdkfd_get_tile_config(struct amdgpu_device *adev,
 				struct tile_config *config)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
-
 	config->gb_addr_config = adev->gfx.config.gb_addr_config;
 	config->tile_config_ptr = adev->gfx.config.tile_mode_array;
 	config->num_tile_configs =