Adding new memory allocation IOCTL

Change-Id: I0eb1924811a2e1e436296ebe632d8f112a61637d
Signed-off-by: Ben Goz <ben.goz@amd.com>


[ROCm/ROCR-Runtime commit: 692e004047]

projects/rocr-runtime/include/linux/kfd_ioctl.h

@@ -279,6 +279,24 @@ struct kfd_ioctl_set_process_dgpu_aperture_args {
 	uint64_t dgpu_limit;
 };
 
+/*
+ * GPU Memory allocation flags
+ */
+#define KFD_IOC_ALLOC_MEM_FLAGS_DGPU_HOST 	(1 << 0)
+#define KFD_IOC_ALLOC_MEM_FLAGS_DGPU_DEVICE	(1 << 1)
+#define KFD_IOC_ALLOC_MEM_FLAGS_DGPU_SCRATCH	(1 << 2)
+#define KFD_IOC_ALLOC_MEM_FLAGS_APU_DEVICE	(1 << 3)
+#define KFD_IOC_ALLOC_MEM_FLAGS_APU_SCRATCH	(1 << 4)
+
+struct kfd_ioctl_alloc_memory_of_gpu_new_args {
+	uint64_t va_addr;	/* to KFD */
+	uint64_t size;		/* to KFD */
+	uint64_t handle;	/* from KFD */
+	uint32_t gpu_id;	/* to KFD */
+	uint64_t mmap_offset;   /* from KFD */
+	uint32_t flags;
+};
+
 #define AMDKFD_IOCTL_BASE 'K'
 #define AMDKFD_IO(nr)			_IO(AMDKFD_IOCTL_BASE, nr)
 #define AMDKFD_IOR(nr, type)		_IOR(AMDKFD_IOCTL_BASE, nr, type)
@@ -357,8 +375,11 @@ struct kfd_ioctl_set_process_dgpu_aperture_args {
 #define AMDKFD_IOC_SET_PROCESS_DGPU_APERTURE   \
 		AMDKFD_IOW(0x18, struct kfd_ioctl_set_process_dgpu_aperture_args)
 
+#define AMDKFD_IOC_ALLOC_MEMORY_OF_GPU_NEW		\
+		AMDKFD_IOWR(0x19, struct kfd_ioctl_alloc_memory_of_gpu_new_args)
+
 
 #define AMDKFD_COMMAND_START		0x01
-#define AMDKFD_COMMAND_END		0x19
+#define AMDKFD_COMMAND_END		0x20
 
 #endif

projects/rocr-runtime/src/fmm.c

@@ -382,9 +382,10 @@ static int32_t gpu_mem_find_by_gpu_id(uint32_t gpu_id)
 static int fmm_allocate_memory_in_device(uint32_t gpu_id, void *mem,
 						uint64_t MemorySizeInBytes,
 						manageble_aperture_t *aperture,
-						uint64_t *mmap_offset)
+						uint64_t *mmap_offset,
+						uint32_t flags)
 {
-	struct kfd_ioctl_alloc_memory_of_gpu_args args;
+	struct kfd_ioctl_alloc_memory_of_gpu_new_args args;
 	struct kfd_ioctl_free_memory_of_gpu_args free_args;
 
 	if (!mem)
@@ -394,11 +395,12 @@ static int fmm_allocate_memory_in_device(uint32_t gpu_id, void *mem,
 	args.gpu_id = gpu_id;
 	args.size = MemorySizeInBytes;
 
+	args.flags = flags;
 	args.va_addr = (uint64_t)mem;
-	if (!mmap_offset)
+	if (!mmap_offset && !is_dgpu)
 		args.va_addr = VOID_PTRS_SUB(mem, aperture->base);
 
-	if (kmtIoctl(kfd_fd, AMDKFD_IOC_ALLOC_MEMORY_OF_GPU, &args))
+	if (kmtIoctl(kfd_fd, AMDKFD_IOC_ALLOC_MEMORY_OF_GPU_NEW, &args))
 		return -1;
 
 	/* Allocate object */
@@ -530,7 +532,8 @@ void *fmm_allocate_scratch(uint32_t gpu_id, uint64_t MemorySizeInBytes)
 }
 
 static void* __fmm_allocate_device(uint32_t gpu_id, uint64_t MemorySizeInBytes,
-		manageble_aperture_t *aperture, uint64_t offset, uint64_t *mmap_offset)
+		manageble_aperture_t *aperture, uint64_t offset, uint64_t *mmap_offset,
+		uint32_t flags)
 {
 	void *mem = NULL;
 	/* Check that aperture is properly initialized/supported */
@@ -548,7 +551,7 @@ static void* __fmm_allocate_device(uint32_t gpu_id, uint64_t MemorySizeInBytes,
 	 * itself
 	 */
 	if (fmm_allocate_memory_in_device(gpu_id, mem,
-			MemorySizeInBytes, aperture, mmap_offset)) {
+			MemorySizeInBytes, aperture, mmap_offset, flags)) {
 		/*
 		 * allocation of memory in device failed.
 		 * Release region in aperture
@@ -582,7 +585,8 @@ void *fmm_allocate_device(uint32_t gpu_id, uint64_t MemorySizeInBytes)
 	aperture = &gpu_mem[gpu_mem_id].gpuvm_aperture;
 
 	return __fmm_allocate_device(gpu_id, MemorySizeInBytes,
-			aperture, GPUVM_APP_OFFSET, NULL);
+			aperture, GPUVM_APP_OFFSET, NULL,
+			KFD_IOC_ALLOC_MEM_FLAGS_DGPU_DEVICE);
 }
 
 static void* fmm_allocate_host_cpu(uint32_t gpu_id,
@@ -627,7 +631,8 @@ static void* fmm_allocate_host_gpu(uint32_t gpu_id,
 	MemorySizeInBytes += 0x8000 - (MemorySizeInBytes % 0x8000);
 
 	mem =  __fmm_allocate_device(gpu_id, MemorySizeInBytes,
-			aperture, 0, &mmap_offset);
+			aperture, 0, &mmap_offset,
+			KFD_IOC_ALLOC_MEM_FLAGS_DGPU_HOST);
 
 	void *ret = mmap(mem, MemorySizeInBytes,
 			PROT_READ | PROT_WRITE | PROT_EXEC,