drm/amdkfd: Decouple CRAT parsing from device list update

Currently, CRAT parsing is intertwined with topology_device_list and
hence repeated calls to kfd_parse_crat_table() will fail. Decouple
kfd_parse_crat_table() and topology_device_list.

kfd_parse_crat_table() will parse CRAT and add topology devices to a
temporary list temp_topology_device_list and then
kfd_topology_update_device_list will move contents from temporary list to
master list.
Signed-off-by: NHarish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: NKent Russell <kent.russell@amd.com>
Signed-off-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: NOded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: NOded Gabbay <oded.gabbay@gmail.com>

drivers/gpu/drm/amd/amdkfd/kfd_crat.c

@@ -23,8 +23,6 @@
 #include "kfd_crat.h"
 #include "kfd_topology.h"
 
-static int topology_crat_parsed;
-extern struct list_head topology_device_list;
 extern struct kfd_system_properties sys_props;
 
 static void kfd_populated_cu_info_cpu(struct kfd_topology_device *dev,
@@ -57,16 +55,18 @@ static void kfd_populated_cu_info_gpu(struct kfd_topology_device *dev,
 	pr_info("CU GPU: id_base=%d\n", cu->processor_id_low);
 }
 
-/* kfd_parse_subtype_cu is called when the topology mutex is already acquired */
-static int kfd_parse_subtype_cu(struct crat_subtype_computeunit *cu)
+/* kfd_parse_subtype_cu - parse compute unit subtypes and attach it to correct
+ * topology device present in the device_list
+ */
+static int kfd_parse_subtype_cu(struct crat_subtype_computeunit *cu,
+				struct list_head *device_list)
 {
 	struct kfd_topology_device *dev;
-	int i = 0;
 
 	pr_info("Found CU entry in CRAT table with proximity_domain=%d caps=%x\n",
 			cu->proximity_domain, cu->hsa_capability);
-	list_for_each_entry(dev, &topology_device_list, list) {
-		if (cu->proximity_domain == i) {
+	list_for_each_entry(dev, device_list, list) {
+		if (cu->proximity_domain == dev->proximity_domain) {
 			if (cu->flags & CRAT_CU_FLAGS_CPU_PRESENT)
 				kfd_populated_cu_info_cpu(dev, cu);
 
@@ -74,26 +74,24 @@ static int kfd_parse_subtype_cu(struct crat_subtype_computeunit *cu)
 				kfd_populated_cu_info_gpu(dev, cu);
 			break;
 		}
-		i++;
 	}
 
 	return 0;
 }
 
-/*
- * kfd_parse_subtype_mem is called when the topology mutex is
- * already acquired
+/* kfd_parse_subtype_mem - parse memory subtypes and attach it to correct
+ * topology device present in the device_list
  */
-static int kfd_parse_subtype_mem(struct crat_subtype_memory *mem)
+static int kfd_parse_subtype_mem(struct crat_subtype_memory *mem,
+				struct list_head *device_list)
 {
 	struct kfd_mem_properties *props;
 	struct kfd_topology_device *dev;
-	int i = 0;
 
 	pr_info("Found memory entry in CRAT table with proximity_domain=%d\n",
 			mem->proximity_domain);
-	list_for_each_entry(dev, &topology_device_list, list) {
-		if (mem->proximity_domain == i) {
+	list_for_each_entry(dev, device_list, list) {
+		if (mem->proximity_domain == dev->proximity_domain) {
 			props = kfd_alloc_struct(props);
 			if (!props)
 				return -ENOMEM;
@@ -118,17 +116,16 @@ static int kfd_parse_subtype_mem(struct crat_subtype_memory *mem)
 
 			break;
 		}
-		i++;
 	}
 
 	return 0;
 }
 
-/*
- * kfd_parse_subtype_cache is called when the topology mutex
- * is already acquired
+/* kfd_parse_subtype_cache - parse cache subtypes and attach it to correct
+ * topology device present in the device_list
  */
-static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache)
+static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache,
+			struct list_head *device_list)
 {
 	struct kfd_cache_properties *props;
 	struct kfd_topology_device *dev;
@@ -137,7 +134,7 @@ static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache)
 	id = cache->processor_id_low;
 
 	pr_info("Found cache entry in CRAT table with processor_id=%d\n", id);
-	list_for_each_entry(dev, &topology_device_list, list)
+	list_for_each_entry(dev, device_list, list)
 		if (id == dev->node_props.cpu_core_id_base ||
 		    id == dev->node_props.simd_id_base) {
 			props = kfd_alloc_struct(props);
@@ -171,15 +168,14 @@ static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache)
 	return 0;
 }
 
-/*
- * kfd_parse_subtype_iolink is called when the topology mutex
- * is already acquired
+/* kfd_parse_subtype_iolink - parse iolink subtypes and attach it to correct
+ * topology device present in the device_list
  */
-static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink)
+static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink,
+					struct list_head *device_list)
 {
 	struct kfd_iolink_properties *props;
 	struct kfd_topology_device *dev;
-	uint32_t i = 0;
 	uint32_t id_from;
 	uint32_t id_to;
 
@@ -187,8 +183,8 @@ static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink)
 	id_to = iolink->proximity_domain_to;
 
 	pr_info("Found IO link entry in CRAT table with id_from=%d\n", id_from);
-	list_for_each_entry(dev, &topology_device_list, list) {
-		if (id_from == i) {
+	list_for_each_entry(dev, device_list, list) {
+		if (id_from == dev->proximity_domain) {
 			props = kfd_alloc_struct(props);
 			if (!props)
 				return -ENOMEM;
@@ -216,13 +212,18 @@ static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink)
 
 			break;
 		}
-		i++;
 	}
 
 	return 0;
 }
 
-static int kfd_parse_subtype(struct crat_subtype_generic *sub_type_hdr)
+/* kfd_parse_subtype - parse subtypes and attach it to correct topology device
+ * present in the device_list
+ *	@sub_type_hdr - subtype section of crat_image
+ *	@device_list - list of topology devices present in this crat_image
+ */
+static int kfd_parse_subtype(struct crat_subtype_generic *sub_type_hdr,
+				struct list_head *device_list)
 {
 	struct crat_subtype_computeunit *cu;
 	struct crat_subtype_memory *mem;
@@ -233,15 +234,15 @@ static int kfd_parse_subtype(struct crat_subtype_generic *sub_type_hdr)
 	switch (sub_type_hdr->type) {
 	case CRAT_SUBTYPE_COMPUTEUNIT_AFFINITY:
 		cu = (struct crat_subtype_computeunit *)sub_type_hdr;
-		ret = kfd_parse_subtype_cu(cu);
+		ret = kfd_parse_subtype_cu(cu, device_list);
 		break;
 	case CRAT_SUBTYPE_MEMORY_AFFINITY:
 		mem = (struct crat_subtype_memory *)sub_type_hdr;
-		ret = kfd_parse_subtype_mem(mem);
+		ret = kfd_parse_subtype_mem(mem, device_list);
 		break;
 	case CRAT_SUBTYPE_CACHE_AFFINITY:
 		cache = (struct crat_subtype_cache *)sub_type_hdr;
-		ret = kfd_parse_subtype_cache(cache);
+		ret = kfd_parse_subtype_cache(cache, device_list);
 		break;
 	case CRAT_SUBTYPE_TLB_AFFINITY:
 		/*
@@ -257,7 +258,7 @@ static int kfd_parse_subtype(struct crat_subtype_generic *sub_type_hdr)
 		break;
 	case CRAT_SUBTYPE_IOLINK_AFFINITY:
 		iolink = (struct crat_subtype_iolink *)sub_type_hdr;
-		ret = kfd_parse_subtype_iolink(iolink);
+		ret = kfd_parse_subtype_iolink(iolink, device_list);
 		break;
 	default:
 		pr_warn("Unknown subtype %d in CRAT\n",
@@ -267,12 +268,23 @@ static int kfd_parse_subtype(struct crat_subtype_generic *sub_type_hdr)
 	return ret;
 }
 
-int kfd_parse_crat_table(void *crat_image)
+/* kfd_parse_crat_table - parse CRAT table. For each node present in CRAT
+ * create a kfd_topology_device and add in to device_list. Also parse
+ * CRAT subtypes and attach it to appropriate kfd_topology_device
+ *	@crat_image - input image containing CRAT
+ *	@device_list - [OUT] list of kfd_topology_device generated after
+ *		       parsing crat_image
+ *	@proximity_domain - Proximity domain of the first device in the table
+ *
+ *	Return - 0 if successful else -ve value
+ */
+int kfd_parse_crat_table(void *crat_image, struct list_head *device_list,
+			 uint32_t proximity_domain)
 {
 	struct kfd_topology_device *top_dev;
 	struct crat_subtype_generic *sub_type_hdr;
 	uint16_t node_id;
-	int ret;
+	int ret = 0;
 	struct crat_header *crat_table = (struct crat_header *)crat_image;
 	uint16_t num_nodes;
 	uint32_t image_len;
@@ -280,17 +292,26 @@ int kfd_parse_crat_table(void *crat_image)
 	if (!crat_image)
 		return -EINVAL;
 
+	if (!list_empty(device_list)) {
+		pr_warn("Error device list should be empty\n");
+		return -EINVAL;
+	}
+
 	num_nodes = crat_table->num_domains;
 	image_len = crat_table->length;
 
 	pr_info("Parsing CRAT table with %d nodes\n", num_nodes);
 
 	for (node_id = 0; node_id < num_nodes; node_id++) {
-		top_dev = kfd_create_topology_device();
-		if (!top_dev) {
-			kfd_release_live_view();
-			return -ENOMEM;
-		}
+		top_dev = kfd_create_topology_device(device_list);
+		if (!top_dev)
+			break;
+		top_dev->proximity_domain = proximity_domain++;
+	}
+
+	if (!top_dev) {
+		ret = -ENOMEM;
+		goto err;
 	}
 
 	sys_props.platform_id =
@@ -302,21 +323,20 @@ int kfd_parse_crat_table(void *crat_image)
 	while ((char *)sub_type_hdr + sizeof(struct crat_subtype_generic) <
 			((char *)crat_image) + image_len) {
 		if (sub_type_hdr->flags & CRAT_SUBTYPE_FLAGS_ENABLED) {
-			ret = kfd_parse_subtype(sub_type_hdr);
-			if (ret != 0) {
-				kfd_release_live_view();
-				return ret;
-			}
+			ret = kfd_parse_subtype(sub_type_hdr, device_list);
+			if (ret)
+				break;
 		}
 
 		sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr +
 				sub_type_hdr->length);
 	}
 
-	sys_props.generation_count++;
-	topology_crat_parsed = 1;
+err:
+	if (ret)
+		kfd_release_topology_device_list(device_list);
 
-	return 0;
+	return ret;
 }
 
 /*

drivers/gpu/drm/amd/amdkfd/kfd_crat.h

@@ -293,6 +293,7 @@ struct cdit_header {
 
 int kfd_create_crat_image_acpi(void **crat_image, size_t *size);
 void kfd_destroy_crat_image(void *crat_image);
-int kfd_parse_crat_table(void *crat_image);
+int kfd_parse_crat_table(void *crat_image, struct list_head *device_list,
+			 uint32_t proximity_domain);
 
 #endif /* KFD_CRAT_H_INCLUDED */

drivers/gpu/drm/amd/amdkfd/kfd_topology.c

@@ -34,7 +34,8 @@
 #include "kfd_topology.h"
 #include "kfd_device_queue_manager.h"
 
-struct list_head topology_device_list;
+/* topology_device_list - Master list of all topology devices */
+static struct list_head topology_device_list;
 struct kfd_system_properties sys_props;
 
 static DECLARE_RWSEM(topology_lock);
@@ -105,24 +106,27 @@ static void kfd_release_topology_device(struct kfd_topology_device *dev)
 	}
 
 	kfree(dev);
-
-	sys_props.num_devices--;
 }
 
-void kfd_release_live_view(void)
+void kfd_release_topology_device_list(struct list_head *device_list)
 {
 	struct kfd_topology_device *dev;
 
-	while (topology_device_list.next != &topology_device_list) {
-		dev = container_of(topology_device_list.next,
-				 struct kfd_topology_device, list);
+	while (!list_empty(device_list)) {
+		dev = list_first_entry(device_list,
+				       struct kfd_topology_device, list);
 		kfd_release_topology_device(dev);
+	}
 }
 
+static void kfd_release_live_view(void)
+{
+	kfd_release_topology_device_list(&topology_device_list);
 	memset(&sys_props, 0, sizeof(sys_props));
 }
 
-struct kfd_topology_device *kfd_create_topology_device(void)
+struct kfd_topology_device *kfd_create_topology_device(
+				struct list_head *device_list)
 {
 	struct kfd_topology_device *dev;
 
@@ -136,8 +140,7 @@ struct kfd_topology_device *kfd_create_topology_device(void)
 	INIT_LIST_HEAD(&dev->cache_props);
 	INIT_LIST_HEAD(&dev->io_link_props);
 
-	list_add_tail(&dev->list, &topology_device_list);
-	sys_props.num_devices++;
+	list_add_tail(&dev->list, device_list);
 
 	return dev;
 }
@@ -682,16 +685,32 @@ static void kfd_topology_release_sysfs(void)
 	}
 }
 
+/* Called with write topology_lock acquired */
+static void kfd_topology_update_device_list(struct list_head *temp_list,
+					struct list_head *master_list)
+{
+	while (!list_empty(temp_list)) {
+		list_move_tail(temp_list->next, master_list);
+		sys_props.num_devices++;
+	}
+}
+
 int kfd_topology_init(void)
 {
 	void *crat_image = NULL;
 	size_t image_size = 0;
 	int ret;
+	struct list_head temp_topology_device_list;
 
-	/*
-	 * Initialize the head for the topology device list
+	/* topology_device_list - Master list of all topology devices
+	 * temp_topology_device_list - temporary list created while parsing CRAT
+	 * or VCRAT. Once parsing is complete the contents of list is moved to
+	 * topology_device_list
 	 */
+
+	/* Initialize the head for the both the lists */
 	INIT_LIST_HEAD(&topology_device_list);
+	INIT_LIST_HEAD(&temp_topology_device_list);
 	init_rwsem(&topology_lock);
 
 	memset(&sys_props, 0, sizeof(sys_props));
@@ -701,7 +720,8 @@ int kfd_topology_init(void)
 	 */
 	ret = kfd_create_crat_image_acpi(&crat_image, &image_size);
 	if (!ret) {
-		ret = kfd_parse_crat_table(crat_image);
+		ret = kfd_parse_crat_table(crat_image,
+					   &temp_topology_device_list, 0);
 		if (ret)
 			goto err;
 	} else if (ret == -ENODATA) {
@@ -714,12 +734,15 @@ int kfd_topology_init(void)
 	}
 
 	down_write(&topology_lock);
+	kfd_topology_update_device_list(&temp_topology_device_list,
+					&topology_device_list);
 	ret = kfd_topology_update_sysfs();
 	up_write(&topology_lock);
 
-	if (!ret)
+	if (!ret) {
+		sys_props.generation_count++;
 		pr_info("Finished initializing topology\n");
-	else
+	} else
 		pr_err("Failed to update topology in sysfs ret=%d\n", ret);
 
 err:
@@ -729,8 +752,10 @@ int kfd_topology_init(void)
 
 void kfd_topology_shutdown(void)
 {
+	down_write(&topology_lock);
 	kfd_topology_release_sysfs();
 	kfd_release_live_view();
+	up_write(&topology_lock);
 }
 
 static void kfd_debug_print_topology(void)
@@ -806,13 +831,15 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	uint32_t gpu_id;
 	struct kfd_topology_device *dev;
 	struct kfd_cu_info cu_info;
-	int res;
+	int res = 0;
+	struct list_head temp_topology_device_list;
+
+	INIT_LIST_HEAD(&temp_topology_device_list);
 
 	gpu_id = kfd_generate_gpu_id(gpu);
 
 	pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id);
 
-	down_write(&topology_lock);
 	/*
 	 * Try to assign the GPU to existing topology device (generated from
 	 * CRAT table
@@ -821,11 +848,12 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	if (!dev) {
 		pr_info("GPU was not found in the current topology. Extending.\n");
 		kfd_debug_print_topology();
-		dev = kfd_create_topology_device();
+		dev = kfd_create_topology_device(&temp_topology_device_list);
 		if (!dev) {
 			res = -ENOMEM;
 			goto err;
 		}
+
 		dev->gpu = gpu;
 
 		/*
@@ -833,12 +861,18 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 		 * GPU vBIOS
 		 */
 
+		down_write(&topology_lock);
+		kfd_topology_update_device_list(&temp_topology_device_list,
+			&topology_device_list);
+
 		/* Update the SYSFS tree, since we added another topology
 		 * device
 		 */
 		if (kfd_topology_update_sysfs() < 0)
 			kfd_topology_release_sysfs();
 
+		up_write(&topology_lock);
+
 	}
 
 	dev->gpu_id = gpu_id;
@@ -859,30 +893,26 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 		pr_info("Adding doorbell packet type capability\n");
 	}
 
-	res = 0;
-
-err:
-	up_write(&topology_lock);
-
-	if (res == 0)
+	if (!res)
 		kfd_notify_gpu_change(gpu_id, 1);
-
+err:
 	return res;
 }
 
 int kfd_topology_remove_device(struct kfd_dev *gpu)
 {
-	struct kfd_topology_device *dev;
+	struct kfd_topology_device *dev, *tmp;
 	uint32_t gpu_id;
 	int res = -ENODEV;
 
 	down_write(&topology_lock);
 
-	list_for_each_entry(dev, &topology_device_list, list)
+	list_for_each_entry_safe(dev, tmp, &topology_device_list, list)
 		if (dev->gpu == gpu) {
 			gpu_id = dev->gpu_id;
 			kfd_remove_sysfs_node_entry(dev);
 			kfd_release_topology_device(dev);
+			sys_props.num_devices--;
 			res = 0;
 			if (kfd_topology_update_sysfs() < 0)
 				kfd_topology_release_sysfs();

drivers/gpu/drm/amd/amdkfd/kfd_topology.h

@@ -135,6 +135,7 @@ struct kfd_iolink_properties {
 struct kfd_topology_device {
 	struct list_head		list;
 	uint32_t			gpu_id;
+	uint32_t			proximity_domain;
 	struct kfd_node_properties	node_props;
 	uint32_t			mem_bank_count;
 	struct list_head		mem_props;
@@ -164,7 +165,8 @@ struct kfd_system_properties {
 	struct attribute	attr_props;
 };
 
-struct kfd_topology_device *kfd_create_topology_device(void);
-void kfd_release_live_view(void);
+struct kfd_topology_device *kfd_create_topology_device(
+		struct list_head *device_list);
+void kfd_release_topology_device_list(struct list_head *device_list);
 
 #endif /* __KFD_TOPOLOGY_H__ */