#include <dlfcn.h>
-#include <rte_common.h>
-#include <rte_log.h>
#include <rte_malloc.h>
-#include <rte_errno.h>
#include <rte_pci.h>
#include <rte_bus_pci.h>
#include <rte_byteorder.h>
#include <rte_dev.h>
#include <gpudev_driver.h>
+
#include <cuda.h>
#include <cudaTypedefs.h>
+#include "common.h"
+
#define CUDA_DRIVER_MIN_VERSION 11040
#define CUDA_API_MIN_VERSION 3020
static void *cudalib;
static unsigned int cuda_api_version;
static int cuda_driver_version;
+static gdr_t gdrc_h;
/* NVIDIA GPU vendor */
#define NVIDIA_GPU_VENDOR_ID (0x10de)
/* 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)
-#define NVIDIA_GPU_V100_32GB_DEVICE_ID (0x1db6)
+#define NVIDIA_GPU_V100_32GB_SXM_DEVICE_ID (0x1db5)
+#define NVIDIA_GPU_V100_32GB_PCIE_DEVICE_ID (0x1db6)
#define NVIDIA_GPU_V100_16GB_DEVICE_ID (0x1db4)
#define NVIDIA_GPU_T4_16GB_DEVICE_ID (0x1eb8)
#define GPU_PAGE_SHIFT 16
#define GPU_PAGE_SIZE (1UL << GPU_PAGE_SHIFT)
-static RTE_LOG_REGISTER_DEFAULT(cuda_logtype, NOTICE);
-
-/* Helper macro for logging */
-#define rte_cuda_log(level, fmt, ...) \
- rte_log(RTE_LOG_ ## level, cuda_logtype, fmt "\n", ##__VA_ARGS__)
-
-#define rte_cuda_debug(fmt, ...) \
- rte_cuda_log(DEBUG, RTE_STR(__LINE__) ":%s() " fmt, __func__, \
- ##__VA_ARGS__)
+RTE_LOG_REGISTER_DEFAULT(cuda_logtype, NOTICE);
/* NVIDIA GPU address map */
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)
},
{
RTE_PCI_DEVICE(NVIDIA_GPU_VENDOR_ID,
- NVIDIA_GPU_V100_32GB_DEVICE_ID)
+ NVIDIA_GPU_V100_32GB_SXM_DEVICE_ID)
+ },
+ {
+ RTE_PCI_DEVICE(NVIDIA_GPU_VENDOR_ID,
+ NVIDIA_GPU_V100_32GB_PCIE_DEVICE_ID)
},
{
RTE_PCI_DEVICE(NVIDIA_GPU_VENDOR_ID,
/* 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;
enum mem_type mtype;
+ gdr_mh_t mh;
struct mem_entry *prev;
struct mem_entry *next;
};
*/
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;
/* 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",
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,
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;
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);
return 0;
}
+static int
+cuda_mem_cpu_map(struct rte_gpu *dev, __rte_unused size_t size, void *ptr_in, void **ptr_out)
+{
+ struct mem_entry *mem_item;
+ cuda_ptr_key hk;
+
+ if (dev == NULL)
+ return -ENODEV;
+
+ hk = get_hash_from_ptr((void *)ptr_in);
+
+ mem_item = mem_list_find_item(hk);
+ if (mem_item == NULL) {
+ rte_cuda_log(ERR, "Memory address 0x%p not found in driver memory.", ptr_in);
+ rte_errno = EPERM;
+ return -rte_errno;
+ }
+
+ if (mem_item->mtype != GPU_MEM) {
+ rte_cuda_log(ERR, "Memory address 0x%p is not GPU memory type.", ptr_in);
+ rte_errno = EPERM;
+ return -rte_errno;
+ }
+
+ if (mem_item->size != size)
+ rte_cuda_log(WARNING,
+ "Can't expose memory area with size (%zd) different from original size (%zd).",
+ size, mem_item->size);
+
+ if (gdrcopy_pin(&gdrc_h, &(mem_item->mh), (uint64_t)mem_item->ptr_d,
+ mem_item->size, &(mem_item->ptr_h))) {
+ rte_cuda_log(ERR, "Error exposing GPU memory address 0x%p.", ptr_in);
+ rte_errno = EPERM;
+ return -rte_errno;
+ }
+
+ *ptr_out = mem_item->ptr_h;
+
+ return 0;
+}
+
static int
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",
return -rte_errno;
}
+static int
+cuda_mem_cpu_unmap(struct rte_gpu *dev, void *ptr_in)
+{
+ struct mem_entry *mem_item;
+ cuda_ptr_key hk;
+
+ if (dev == NULL)
+ return -ENODEV;
+
+ hk = get_hash_from_ptr((void *)ptr_in);
+
+ mem_item = mem_list_find_item(hk);
+ if (mem_item == NULL) {
+ rte_cuda_log(ERR, "Memory address 0x%p not found in driver memory.", ptr_in);
+ rte_errno = EPERM;
+ return -rte_errno;
+ }
+
+ if (gdrcopy_unpin(gdrc_h, mem_item->mh, (void *)mem_item->ptr_d,
+ mem_item->size)) {
+ rte_cuda_log(ERR, "Error unexposing GPU memory address 0x%p.", ptr_in);
+ rte_errno = EPERM;
+ return -rte_errno;
+ }
+
+ return 0;
+}
+
static int
cuda_dev_close(struct rte_gpu *dev)
{
rte_errno = ENOTSUP;
return -rte_errno;
}
+
+ gdrc_h = NULL;
}
/* Fill HW specific part of device structure */
dev->ops.mem_free = cuda_mem_free;
dev->ops.mem_register = cuda_mem_register;
dev->ops.mem_unregister = cuda_mem_unregister;
+ dev->ops.mem_cpu_map = cuda_mem_cpu_map;
+ dev->ops.mem_cpu_unmap = cuda_mem_cpu_unmap;
dev->ops.wmb = cuda_wmb;
rte_gpu_complete_new(dev);