gpudev: add alignment for memory allocation

[dpdk.git] / drivers / gpu / cuda / cuda.c

diff --git a/drivers/gpu/cuda/cuda.c b/drivers/gpu/cuda/cuda.c
index 882df08..0ece1bb 100644 (file)
--- a/drivers/gpu/cuda/cuda.c
+++ b/drivers/gpu/cuda/cuda.c
@@ -58,6 +58,7 @@ static int cuda_driver_version;
 /* NVIDIA GPU device IDs */
 #define NVIDIA_GPU_A100_40GB_DEVICE_ID (0x20f1)
 #define NVIDIA_GPU_A100_80GB_DEVICE_ID (0x20b5)
+#define NVIDIA_GPU_A100_80GB_DPU_DEVICE_ID (0x20b8)
 
 #define NVIDIA_GPU_A30_24GB_DEVICE_ID (0x20b7)
 #define NVIDIA_GPU_A10_24GB_DEVICE_ID (0x2236)
@@ -92,6 +93,10 @@ static const struct rte_pci_id pci_id_cuda_map[] = {
                RTE_PCI_DEVICE(NVIDIA_GPU_VENDOR_ID,
                                NVIDIA_GPU_A100_80GB_DEVICE_ID)
        },
+       {
+               RTE_PCI_DEVICE(NVIDIA_GPU_VENDOR_ID,
+                               NVIDIA_GPU_A100_80GB_DPU_DEVICE_ID)
+       },
        {
                RTE_PCI_DEVICE(NVIDIA_GPU_VENDOR_ID,
                                NVIDIA_GPU_A30_24GB_DEVICE_ID)
@@ -139,8 +144,10 @@ typedef uintptr_t cuda_ptr_key;
 /* Single entry of the memory list */
 struct mem_entry {
        CUdeviceptr ptr_d;
+       CUdeviceptr ptr_orig_d;
        void *ptr_h;
        size_t size;
+       size_t size_orig;
        struct rte_gpu *dev;
        CUcontext ctx;
        cuda_ptr_key pkey;
@@ -437,9 +444,11 @@ mem_list_del_item(cuda_ptr_key pk)
                return -EINVAL;
 
        /* if key is in head */
-       if (mem_alloc_list_cur->prev == NULL)
+       if (mem_alloc_list_cur->prev == NULL) {
                mem_alloc_list_head = mem_alloc_list_cur->next;
-       else {
+               if (mem_alloc_list_head != NULL)
+                       mem_alloc_list_head->prev = NULL;
+       } else {
                mem_alloc_list_cur->prev->next = mem_alloc_list_cur->next;
                if (mem_alloc_list_cur->next != NULL)
                        mem_alloc_list_cur->next->prev = mem_alloc_list_cur->prev;
@@ -569,7 +578,7 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
  */
 
 static int
-cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
+cuda_mem_alloc(struct rte_gpu *dev, size_t size, unsigned int align, void **ptr)
 {
        CUresult res;
        const char *err_string;
@@ -610,8 +619,10 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
 
        /* Allocate memory */
        mem_alloc_list_tail->size = size;
-       res = pfn_cuMemAlloc(&(mem_alloc_list_tail->ptr_d),
-                       mem_alloc_list_tail->size);
+       mem_alloc_list_tail->size_orig = size + align;
+
+       res = pfn_cuMemAlloc(&(mem_alloc_list_tail->ptr_orig_d),
+                       mem_alloc_list_tail->size_orig);
        if (res != 0) {
                pfn_cuGetErrorString(res, &(err_string));
                rte_cuda_log(ERR, "cuCtxSetCurrent current failed with %s",
@@ -620,6 +631,12 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
                return -rte_errno;
        }
 
+       /* Align memory address */
+       mem_alloc_list_tail->ptr_d = mem_alloc_list_tail->ptr_orig_d;
+       if (align && ((uintptr_t)mem_alloc_list_tail->ptr_d) % align)
+               mem_alloc_list_tail->ptr_d += (align -
+                               (((uintptr_t)mem_alloc_list_tail->ptr_d) % align));
+
        /* GPUDirect RDMA attribute required */
        res = pfn_cuPointerSetAttribute(&flag,
                        CU_POINTER_ATTRIBUTE_SYNC_MEMOPS,
@@ -634,7 +651,6 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
 
        mem_alloc_list_tail->pkey = get_hash_from_ptr((void *)mem_alloc_list_tail->ptr_d);
        mem_alloc_list_tail->ptr_h = NULL;
-       mem_alloc_list_tail->size = size;
        mem_alloc_list_tail->dev = dev;
        mem_alloc_list_tail->ctx = (CUcontext)((uintptr_t)dev->mpshared->info.context);
        mem_alloc_list_tail->mtype = GPU_MEM;
@@ -761,6 +777,7 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
        mem_alloc_list_tail->dev = dev;
        mem_alloc_list_tail->ctx = (CUcontext)((uintptr_t)dev->mpshared->info.context);
        mem_alloc_list_tail->mtype = CPU_REGISTERED;
+       mem_alloc_list_tail->ptr_orig_d = mem_alloc_list_tail->ptr_d;
 
        /* Restore original ctx as current ctx */
        res = pfn_cuCtxSetCurrent(current_ctx);
@@ -796,7 +813,7 @@ cuda_mem_free(struct rte_gpu *dev, void *ptr)
        }
 
        if (mem_item->mtype == GPU_MEM) {
-               res = pfn_cuMemFree(mem_item->ptr_d);
+               res = pfn_cuMemFree(mem_item->ptr_orig_d);
                if (res != 0) {
                        pfn_cuGetErrorString(res, &(err_string));
                        rte_cuda_log(ERR, "cuMemFree current failed with %s",