Silence Valgrind warnings

Change-Id: I8803f3d310fccd69d0d04b2464b00dccc40270e3
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>


[ROCm/ROCR-Runtime commit: 056ddbbc82]

projects/rocr-runtime/include/hsakmttypes.h

@@ -1094,8 +1094,8 @@ typedef struct _HsaPmcTraceRoot
 
 typedef struct _HsaGpuTileConfig
 {
-    const HSAuint32 *TileConfig;
-    const HSAuint32 *MacroTileConfig;
+    HSAuint32 *TileConfig;
+    HSAuint32 *MacroTileConfig;
     HSAuint32 NumTileConfigs;
     HSAuint32 NumMacroTileConfigs;
 

projects/rocr-runtime/src/debug.c

@@ -67,9 +67,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtDbgRegister(HSAuint32 NodeId)
 	if (result != HSAKMT_STATUS_SUCCESS)
 		return result;
 
-	struct kfd_ioctl_dbg_register_args args;
+	struct kfd_ioctl_dbg_register_args args = {0};
 
-	memset(&args, 0, sizeof(args));
 	args.gpu_id = gpu_id;
 
 	long err = kmtIoctl(kfd_fd, AMDKFD_IOC_DBG_REGISTER, &args);
@@ -96,9 +95,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtDbgUnregister(HSAuint32 NodeId)
 	if (result != HSAKMT_STATUS_SUCCESS)
 		return result;
 
-	struct kfd_ioctl_dbg_unregister_args args;
+	struct kfd_ioctl_dbg_unregister_args args = {0};
 
-	memset(&args, 0, sizeof(args));
 	args.gpu_id = gpu_id;
 	long err = kmtIoctl(kfd_fd, AMDKFD_IOC_DBG_UNREGISTER, &args);
 

projects/rocr-runtime/src/events.c

@@ -65,9 +65,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtCreateEvent(HsaEventDescriptor *EventDesc,
 
 	memset(e, 0, sizeof(*e));
 
-	struct kfd_ioctl_create_event_args args;
-
-	memset(&args, 0, sizeof(args));
+	struct kfd_ioctl_create_event_args args = {0};
 
 	args.event_type = EventDesc->EventType;
 	args.node_id = EventDesc->NodeId;
@@ -127,9 +125,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtCreateEvent(HsaEventDescriptor *EventDesc,
 		EventDesc->SyncVar.SyncVarSize;
 
 	if (IsSignaled && !IsSystemEventType(e->EventData.EventType)) {
-		struct kfd_ioctl_set_event_args set_args;
+		struct kfd_ioctl_set_event_args set_args = {0};
 
-		memset(&set_args, 0, sizeof(set_args));
 		set_args.event_id = args.event_id;
 
 		kmtIoctl(kfd_fd, AMDKFD_IOC_SET_EVENT, &set_args);
@@ -147,9 +144,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtDestroyEvent(HsaEvent *Event)
 	if (!Event)
 		return HSAKMT_STATUS_INVALID_HANDLE;
 
-	struct kfd_ioctl_destroy_event_args args;
-
-	memset(&args, 0, sizeof(args));
+	struct kfd_ioctl_destroy_event_args args = {0};
 
 	args.event_id = Event->EventId;
 
@@ -173,9 +168,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtSetEvent(HsaEvent *Event)
 	if (IsSystemEventType(Event->EventData.EventType))
 		return HSAKMT_STATUS_ERROR;
 
-	struct kfd_ioctl_set_event_args args;
-
-	memset(&args, 0, sizeof(args));
+	struct kfd_ioctl_set_event_args args = {0};
 
 	args.event_id = Event->EventId;
 
@@ -198,9 +191,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtResetEvent(HsaEvent *Event)
 	if (IsSystemEventType(Event->EventData.EventType))
 		return HSAKMT_STATUS_ERROR;
 
-	struct kfd_ioctl_reset_event_args args;
-
-	memset(&args, 0, sizeof(args));
+	struct kfd_ioctl_reset_event_args args = {0};
 
 	args.event_id = Event->EventId;
 
@@ -306,9 +297,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtWaitOnMultipleEvents(HsaEvent *Events[],
 		event_data[i].kfd_event_data_ext = (uint64_t)(uintptr_t)NULL;
 	}
 
-	struct kfd_ioctl_wait_events_args args;
-
-	memset(&args, 0, sizeof(args));
+	struct kfd_ioctl_wait_events_args args = {0};
 
 	args.wait_for_all = WaitOnAll;
 	args.timeout = Milliseconds;

projects/rocr-runtime/src/fmm.c

@@ -728,8 +728,8 @@ static vm_object_t *fmm_allocate_memory_object(uint32_t gpu_id, void *mem,
 						uint64_t *mmap_offset,
 						uint32_t flags)
 {
-	struct kfd_ioctl_alloc_memory_of_gpu_args args;
-	struct kfd_ioctl_free_memory_of_gpu_args free_args;
+	struct kfd_ioctl_alloc_memory_of_gpu_args args = {0};
+	struct kfd_ioctl_free_memory_of_gpu_args free_args = {0};
 	vm_object_t *vm_obj = NULL;
 
 	if (!mem)
@@ -891,7 +891,7 @@ static uint32_t fmm_translate_hsa_to_ioc_flags(HsaMemFlags flags)
 void *fmm_allocate_scratch(uint32_t gpu_id, uint64_t MemorySizeInBytes)
 {
 	manageable_aperture_t *aperture_phy;
-	struct kfd_ioctl_alloc_memory_of_scratch_args args;
+	struct kfd_ioctl_alloc_memory_of_scratch_args args = {0};
 	int32_t gpu_mem_id;
 	void *mem = NULL;
 	uint64_t aligned_size = ALIGN_UP(MemorySizeInBytes, SCRATCH_ALIGN);
@@ -1298,7 +1298,7 @@ void *fmm_allocate_host(uint32_t node_id, uint64_t MemorySizeInBytes,
 
 static void __fmm_release(void *address, manageable_aperture_t *aperture)
 {
-	struct kfd_ioctl_free_memory_of_gpu_args args;
+	struct kfd_ioctl_free_memory_of_gpu_args args = {0};
 	vm_object_t *object;
 
 	if (!address)
@@ -1394,7 +1394,7 @@ void fmm_release(void *address)
 static int fmm_set_memory_policy(uint32_t gpu_id, int default_policy, int alt_policy,
 				 uintptr_t alt_base, uint64_t alt_size)
 {
-	struct kfd_ioctl_set_memory_policy_args args;
+	struct kfd_ioctl_set_memory_policy_args args = {0};
 
 	args.gpu_id = gpu_id;
 	args.default_policy = default_policy;
@@ -1421,12 +1421,11 @@ static HSAKMT_STATUS get_process_apertures(
 	struct kfd_process_device_apertures *process_apertures,
 	uint32_t *num_of_nodes)
 {
-	struct kfd_ioctl_get_process_apertures_new_args args_new = {
-		.kfd_process_device_apertures_ptr = (uintptr_t)process_apertures,
-		.num_of_nodes = *num_of_nodes
-	};
+	struct kfd_ioctl_get_process_apertures_new_args args_new = {0};
 	struct kfd_ioctl_get_process_apertures_args args_old;
 
+	args_new.kfd_process_device_apertures_ptr = (uintptr_t)process_apertures;
+	args_new.num_of_nodes = *num_of_nodes;
 	if (!kmtIoctl(kfd_fd, AMDKFD_IOC_GET_PROCESS_APERTURES_NEW,
 		      (void *)&args_new)) {
 		*num_of_nodes = args_new.num_of_nodes;
@@ -1435,6 +1434,8 @@ static HSAKMT_STATUS get_process_apertures(
 
 	/* New IOCTL failed, try the old one in case we're running on
 	 * a really old kernel */
+	memset(&args_old, 0, sizeof(args_old));
+
 	if (kmtIoctl(kfd_fd, AMDKFD_IOC_GET_PROCESS_APERTURES,
 		     (void *)&args_old))
 		return HSAKMT_STATUS_ERROR;
@@ -1530,7 +1531,7 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
 	 * the apertures in kfd_process_device_apertures_ptr
 	 */
 	num_of_nodes = gpu_mem_count;
-	process_apertures = malloc(num_of_nodes * sizeof(struct kfd_process_device_apertures));
+	process_apertures = calloc(num_of_nodes, sizeof(struct kfd_process_device_apertures));
 	if (!process_apertures) {
 		ret = HSAKMT_STATUS_NO_MEMORY;
 		goto sysfs_parse_failed;
@@ -1782,7 +1783,7 @@ static int _fmm_map_to_gpu(manageable_aperture_t *aperture,
 			void *address, uint64_t size, vm_object_t *obj,
 			uint32_t *nodes_to_map, uint32_t nodes_array_size)
 {
-	struct kfd_ioctl_map_memory_to_gpu_args args;
+	struct kfd_ioctl_map_memory_to_gpu_args args = {0};
 	vm_object_t *object;
 
 	if (!obj)
@@ -2057,7 +2058,7 @@ static int _fmm_unmap_from_gpu(manageable_aperture_t *aperture, void *address,
 {
 	vm_object_t *object;
 	int ret = 0;
-	struct kfd_ioctl_unmap_memory_from_gpu_args args;
+	struct kfd_ioctl_unmap_memory_from_gpu_args args = {0};
 	HSAuint32 page_offset = (HSAint64)address & (PAGE_SIZE - 1);
 
 	if (!obj)
@@ -2124,7 +2125,7 @@ static int _fmm_unmap_from_gpu_scratch(uint32_t gpu_id,
 {
 	int32_t gpu_mem_id;
 	vm_object_t *object;
-	struct kfd_ioctl_unmap_memory_from_gpu_args args;
+	struct kfd_ioctl_unmap_memory_from_gpu_args args = {0};
 
 	/* Retrieve gpu_mem id according to gpu_id */
 	gpu_mem_id = gpu_mem_find_by_gpu_id(gpu_id);
@@ -2245,7 +2246,7 @@ static bool is_dgpu_mem_init = false;
 
 static int set_dgpu_aperture(uint32_t gpu_id, uint64_t base, uint64_t limit)
 {
-	struct kfd_ioctl_set_process_dgpu_aperture_args args;
+	struct kfd_ioctl_set_process_dgpu_aperture_args args = {0};
 
 	args.gpu_id = gpu_id;
 	args.dgpu_base = base;
@@ -2578,9 +2579,9 @@ HSAKMT_STATUS fmm_register_graphics_handle(HSAuint64 GraphicsResourceHandle,
 					   uint32_t *gpu_id_array,
 					   uint32_t gpu_id_array_size)
 {
-	struct kfd_ioctl_get_dmabuf_info_args infoArgs;
-	struct kfd_ioctl_import_dmabuf_args importArgs;
-	struct kfd_ioctl_free_memory_of_gpu_args freeArgs;
+	struct kfd_ioctl_get_dmabuf_info_args infoArgs = {0};
+	struct kfd_ioctl_import_dmabuf_args importArgs = {0};
+	struct kfd_ioctl_free_memory_of_gpu_args freeArgs = {0};
 	manageable_aperture_t *aperture;
 	vm_object_t *obj;
 	void *metadata;
@@ -2684,7 +2685,7 @@ HSAKMT_STATUS fmm_share_memory(void *MemoryAddress,
 	HSAuint32 gpu_id = 0;
 	vm_object_t *obj = NULL;
 	manageable_aperture_t *aperture = NULL;
-	struct kfd_ioctl_ipc_export_handle_args exportArgs;
+	struct kfd_ioctl_ipc_export_handle_args exportArgs = {0};
 	HsaApertureInfo ApeInfo;
 	HsaSharedMemoryStruct *SharedMemoryStruct =
 		to_hsa_shared_memory_struct(SharedMemoryHandle);
@@ -2741,8 +2742,8 @@ HSAKMT_STATUS fmm_register_shared_memory(const HsaSharedMemoryHandle *SharedMemo
 	vm_object_t *obj = NULL;
 	void *reservedMem = NULL;
 	manageable_aperture_t *aperture;
-	struct kfd_ioctl_ipc_import_handle_args importArgs;
-	struct kfd_ioctl_free_memory_of_gpu_args freeArgs;
+	struct kfd_ioctl_ipc_import_handle_args importArgs = {0};
+	struct kfd_ioctl_free_memory_of_gpu_args freeArgs = {0};
 	const HsaSharedMemoryStruct *SharedMemoryStruct =
 		to_const_hsa_shared_memory_struct(SharedMemoryHandle);
 

projects/rocr-runtime/src/memory.c

@@ -40,7 +40,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtSetMemoryPolicy(HSAuint32 Node,
 					      void *MemoryAddressAlternate,
 					      HSAuint64 MemorySizeInBytes)
 {
-	struct kfd_ioctl_set_memory_policy_args args;
+	struct kfd_ioctl_set_memory_policy_args args = {0};
 	HSAKMT_STATUS result;
 	uint32_t gpu_id;
 
@@ -74,8 +74,6 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtSetMemoryPolicy(HSAuint32 Node,
 	CHECK_PAGE_MULTIPLE(MemoryAddressAlternate);
 	CHECK_PAGE_MULTIPLE(MemorySizeInBytes);
 
-	memset(&args, 0, sizeof(args));
-
 	args.gpu_id = gpu_id;
 	args.default_policy = (DefaultPolicy == HSA_CACHING_CACHED) ?
 					KFD_IOC_CACHE_POLICY_COHERENT :
@@ -356,7 +354,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtProcessVMRead(HSAuint32 Pid,
 					    HSAuint64 RemoteMemoryArrayCount,
 					    HSAuint64 *SizeCopied)
 {
-	struct kfd_ioctl_cross_memory_copy_args args;
+	struct kfd_ioctl_cross_memory_copy_args args = {0};
 
 	pr_debug("[%s]\n", __func__);
 
@@ -391,7 +389,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtProcessVMWrite(HSAuint32 Pid,
 					     HSAuint64 RemoteMemoryArrayCount,
 					     HSAuint64 *SizeCopied)
 {
-	struct kfd_ioctl_cross_memory_copy_args args;
+	struct kfd_ioctl_cross_memory_copy_args args = {0};
 
 	pr_debug("[%s]\n", __func__);
 
@@ -530,7 +528,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtUnmapGraphicHandle(HSAuint32 NodeId,
 
 HSAKMT_STATUS HSAKMTAPI hsaKmtGetTileConfig(HSAuint32 NodeId, HsaGpuTileConfig *config)
 {
-	struct kfd_ioctl_get_tile_config_args args;
+	struct kfd_ioctl_get_tile_config_args args = {0};
 	uint32_t gpu_id;
 	HSAKMT_STATUS result;
 
@@ -540,6 +538,12 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtGetTileConfig(HSAuint32 NodeId, HsaGpuTileConfig *
 	if (result != HSAKMT_STATUS_SUCCESS)
 		return result;
 
+	/* Avoid Valgrind warnings about uninitialized data. Valgrind doesn't
+	 * know that KFD writes this.
+	 */
+	memset(config->TileConfig, 0, sizeof(*config->TileConfig) * config->NumTileConfigs);
+	memset(config->MacroTileConfig, 0, sizeof(*config->MacroTileConfig) * config->NumMacroTileConfigs);
+
 	args.gpu_id = gpu_id;
 	args.tile_config_ptr = (uint64_t)config->TileConfig;
 	args.macro_tile_config_ptr = (uint64_t)config->MacroTileConfig;

projects/rocr-runtime/src/queues.c

@@ -528,9 +528,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtCreateQueue(HSAuint32 NodeId,
 			q->cu_mask[i/32] |= (1 << (i % 32));
 	}
 
-	struct kfd_ioctl_create_queue_args args;
-
-	memset(&args, 0, sizeof(args));
+	struct kfd_ioctl_create_queue_args args = {0};
 
 	args.gpu_id = gpu_id;
 
@@ -618,7 +616,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtUpdateQueue(HSA_QUEUEID QueueId,
 					  HSAuint64 QueueSize,
 					  HsaEvent *Event)
 {
-	struct kfd_ioctl_update_queue_args arg;
+	struct kfd_ioctl_update_queue_args arg = {0};
 	struct queue *q = PORT_UINT64_TO_VPTR(QueueId);
 
 	CHECK_KFD_OPEN();
@@ -648,13 +646,11 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtDestroyQueue(HSA_QUEUEID QueueId)
 	CHECK_KFD_OPEN();
 
 	struct queue *q = PORT_UINT64_TO_VPTR(QueueId);
-	struct kfd_ioctl_destroy_queue_args args;
+	struct kfd_ioctl_destroy_queue_args args = {0};
 
 	if (!q)
 		return HSAKMT_STATUS_INVALID_PARAMETER;
 
-	memset(&args, 0, sizeof(args));
-
 	args.queue_id = q->queue_id;
 
 	int err = kmtIoctl(kfd_fd, AMDKFD_IOC_DESTROY_QUEUE, &args);
@@ -671,14 +667,13 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtSetQueueCUMask(HSA_QUEUEID QueueId,
 					     HSAuint32 *QueueCUMask)
 {
 	struct queue *q = PORT_UINT64_TO_VPTR(QueueId);
-	struct kfd_ioctl_set_cu_mask_args args;
+	struct kfd_ioctl_set_cu_mask_args args = {0};
 
 	CHECK_KFD_OPEN();
 
 	if (CUMaskCount == 0 || !QueueCUMask || ((CUMaskCount % 32) != 0))
 		return HSAKMT_STATUS_INVALID_PARAMETER;
 
-	memset(&args, 0, sizeof(args));
 	args.queue_id = q->queue_id;
 	args.num_cu_mask = CUMaskCount;
 	args.cu_mask_ptr = (uintptr_t)QueueCUMask;
@@ -702,7 +697,7 @@ hsaKmtGetQueueInfo(
 )
 {
 	struct queue *q = PORT_UINT64_TO_VPTR(QueueId);
-	struct kfd_ioctl_get_queue_wave_state_args args;
+	struct kfd_ioctl_get_queue_wave_state_args args = {0};
 
 	CHECK_KFD_OPEN();
 
@@ -712,7 +707,6 @@ hsaKmtGetQueueInfo(
 	if (q->ctx_save_restore == NULL)
 		return HSAKMT_STATUS_ERROR;
 
-	memset(&args, 0, sizeof(args));
 	args.queue_id = q->queue_id;
 	args.ctl_stack_address = (uintptr_t)q->ctx_save_restore;
 
@@ -739,7 +733,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtSetTrapHandler(HSAuint32 Node,
 					     void *TrapBufferBaseAddress,
 					     HSAuint64 TrapBufferSizeInBytes)
 {
-	struct kfd_ioctl_set_trap_handler_args args;
+	struct kfd_ioctl_set_trap_handler_args args = {0};
 	HSAKMT_STATUS result;
 	uint32_t gpu_id;
 
@@ -749,8 +743,6 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtSetTrapHandler(HSAuint32 Node,
 	if (result != HSAKMT_STATUS_SUCCESS)
 		return result;
 
-	memset(&args, 0, sizeof(args));
-
 	args.gpu_id = gpu_id;
 	args.tba_addr = (uintptr_t)TrapHandlerBaseAddress;
 	args.tma_addr = (uintptr_t)TrapBufferBaseAddress;

projects/rocr-runtime/src/time.c

@@ -31,7 +31,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtGetClockCounters(HSAuint32 NodeId,
 {
 	HSAKMT_STATUS result;
 	uint32_t gpu_id;
-	struct kfd_ioctl_get_clock_counters_args args;
+	struct kfd_ioctl_get_clock_counters_args args = {0};
 	int err;
 
 	CHECK_KFD_OPEN();

projects/rocr-runtime/src/version.c

@@ -32,9 +32,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtGetVersion(HsaVersionInfo *VersionInfo)
 {
 	CHECK_KFD_OPEN();
 
-	struct kfd_ioctl_get_version_args args;
-
-	memset(&args, 0, sizeof(args));
+	struct kfd_ioctl_get_version_args args = {0};
 
 	if (kmtIoctl(kfd_fd, AMDKFD_IOC_GET_VERSION, &args) == -1)
 		return HSAKMT_STATUS_ERROR;