gpu/cuda: set rte_errno

Set correct rte_errno variable in CUDA driver
and return -rte_errno in case of error.

rte_errno values are compliant with the gpudev library documentation.

Fixes: 1306a73b19 ("gpu/cuda: introduce CUDA driver")

Signed-off-by: Elena Agostini <eagostini@nvidia.com>
This commit is contained in:
Elena Agostini 2021-11-24 18:34:09 +00:00 committed by Thomas Monjalon
parent 1674c56dc3
commit 94220b3977

View File

@ -464,8 +464,10 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
CUcontext current_ctx;
CUcontext input_ctx;
if (dev == NULL)
return -ENODEV;
if (dev == NULL) {
rte_errno = ENODEV;
return -rte_errno;
}
/* Child initialization time probably called by rte_gpu_add_child() */
if (dev->mpshared->info.parent != RTE_GPU_ID_NONE &&
@ -476,7 +478,8 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxGetCurrent failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Set child ctx as current ctx */
@ -486,7 +489,8 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent input failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/*
@ -505,8 +509,10 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
(uint32_t)affinityPrm.param.smCount.val;
ret = rte_gpu_info_get(dev->mpshared->info.parent, &parent_info);
if (ret)
return -ENODEV;
if (ret) {
rte_errno = ENODEV;
return -rte_errno;
}
dev->mpshared->info.total_memory = parent_info.total_memory;
/*
@ -517,7 +523,8 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
RTE_CACHE_LINE_SIZE);
if (dev->mpshared->dev_private == NULL) {
rte_cuda_log(ERR, "Failed to allocate memory for GPU process private");
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
private = (struct cuda_info *)dev->mpshared->dev_private;
@ -527,7 +534,8 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxGetDevice failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
res = pfn_cuDeviceGetName(private->gpu_name,
@ -536,7 +544,8 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceGetName failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Restore original ctx as current ctx */
@ -545,7 +554,8 @@ cuda_dev_info_get(struct rte_gpu *dev, struct rte_gpu_info *info)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
}
@ -576,7 +586,8 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxGetCurrent failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Set child ctx as current ctx */
@ -586,13 +597,16 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent input failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Get next memory list item */
mem_alloc_list_tail = mem_list_add_item();
if (mem_alloc_list_tail == NULL)
return -ENOMEM;
if (mem_alloc_list_tail == NULL) {
rte_errno = EPERM;
return -rte_errno;
}
/* Allocate memory */
mem_alloc_list_tail->size = size;
@ -602,7 +616,8 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* GPUDirect RDMA attribute required */
@ -613,7 +628,8 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
rte_cuda_log(ERR, "Could not set SYNC MEMOP attribute for "
"GPU memory at %"PRIu32", err %d",
(uint32_t)mem_alloc_list_tail->ptr_d, res);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
mem_alloc_list_tail->pkey = get_hash_from_ptr((void *)mem_alloc_list_tail->ptr_d);
@ -629,7 +645,8 @@ cuda_mem_alloc(struct rte_gpu *dev, size_t size, void **ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
*ptr = (void *)mem_alloc_list_tail->ptr_d;
@ -656,7 +673,8 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxGetCurrent failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Set child ctx as current ctx */
@ -666,13 +684,16 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent input failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Get next memory list item */
mem_alloc_list_tail = mem_list_add_item();
if (mem_alloc_list_tail == NULL)
return -ENOMEM;
if (mem_alloc_list_tail == NULL) {
rte_errno = EPERM;
return -rte_errno;
}
/* Allocate memory */
mem_alloc_list_tail->size = size;
@ -688,7 +709,8 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
err_string,
mem_alloc_list_tail->ptr_h,
mem_alloc_list_tail->size);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
res = pfn_cuDeviceGetAttribute(&(use_ptr_h),
@ -698,7 +720,8 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceGetAttribute failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
if (use_ptr_h == 0) {
@ -708,13 +731,15 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuMemHostGetDevicePointer failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
if ((uintptr_t)mem_alloc_list_tail->ptr_d !=
(uintptr_t)mem_alloc_list_tail->ptr_h) {
rte_cuda_log(ERR, "Host input pointer is different wrt GPU registered pointer");
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
} else {
mem_alloc_list_tail->ptr_d = (CUdeviceptr)mem_alloc_list_tail->ptr_h;
@ -727,7 +752,8 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
if (res != 0) {
rte_cuda_log(ERR, "Could not set SYNC MEMOP attribute for GPU memory at %"PRIu32
", err %d", (uint32_t)mem_alloc_list_tail->ptr_d, res);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
mem_alloc_list_tail->pkey = get_hash_from_ptr((void *)mem_alloc_list_tail->ptr_h);
@ -742,7 +768,8 @@ cuda_mem_register(struct rte_gpu *dev, size_t size, void *ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
return 0;
@ -764,7 +791,8 @@ cuda_mem_free(struct rte_gpu *dev, void *ptr)
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);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
if (mem_item->mtype == GPU_MEM) {
@ -773,7 +801,8 @@ cuda_mem_free(struct rte_gpu *dev, void *ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuMemFree current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
return mem_list_del_item(hk);
@ -800,7 +829,8 @@ cuda_mem_unregister(struct rte_gpu *dev, void *ptr)
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);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
if (mem_item->mtype == CPU_REGISTERED) {
@ -809,7 +839,8 @@ cuda_mem_unregister(struct rte_gpu *dev, void *ptr)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuMemHostUnregister current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
return mem_list_del_item(hk);
@ -817,7 +848,8 @@ cuda_mem_unregister(struct rte_gpu *dev, void *ptr)
rte_cuda_log(ERR, "Memory type %d not supported", mem_item->mtype);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
static int
@ -840,8 +872,10 @@ cuda_wmb(struct rte_gpu *dev)
CUcontext input_ctx;
struct cuda_info *private;
if (dev == NULL)
return -ENODEV;
if (dev == NULL) {
rte_errno = ENODEV;
return -rte_errno;
}
private = (struct cuda_info *)dev->mpshared->dev_private;
@ -860,7 +894,9 @@ cuda_wmb(struct rte_gpu *dev)
*/
rte_cuda_log(WARNING, "Can't flush GDR writes with cuFlushGPUDirectRDMAWrites CUDA function."
"Application needs to use alternative methods.");
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
/* Store current ctx */
@ -869,7 +905,8 @@ cuda_wmb(struct rte_gpu *dev)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxGetCurrent failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Set child ctx as current ctx */
@ -879,7 +916,8 @@ cuda_wmb(struct rte_gpu *dev)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent input failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
res = pfn_cuFlushGPUDirectRDMAWrites(CU_FLUSH_GPU_DIRECT_RDMA_WRITES_TARGET_CURRENT_CTX,
@ -888,7 +926,8 @@ cuda_wmb(struct rte_gpu *dev)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuFlushGPUDirectRDMAWrites current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/* Restore original ctx as current ctx */
@ -897,7 +936,8 @@ cuda_wmb(struct rte_gpu *dev)
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuCtxSetCurrent current failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
return 0;
@ -917,15 +957,18 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
if (pci_dev == NULL) {
rte_cuda_log(ERR, "NULL PCI device");
return -EINVAL;
rte_errno = ENODEV;
return -rte_errno;
}
rte_pci_device_name(&pci_dev->addr, dev_name, sizeof(dev_name));
/* Allocate memory to be used privately by drivers */
dev = rte_gpu_allocate(pci_dev->device.name);
if (dev == NULL)
return -ENODEV;
if (dev == NULL) {
rte_errno = ENODEV;
return -rte_errno;
}
/* Initialize values only for the first CUDA driver call */
if (dev->mpshared->info.dev_id == 0) {
@ -936,13 +979,15 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
/* Load libcuda.so library */
if (cuda_loader()) {
rte_cuda_log(ERR, "CUDA Driver library not found");
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
/* Load initial CUDA functions */
if (cuda_sym_func_loader()) {
rte_cuda_log(ERR, "CUDA functions not found in library");
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
/*
@ -955,7 +1000,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
res = sym_cuDriverGetVersion(&cuda_driver_version);
if (res != 0) {
rte_cuda_log(ERR, "cuDriverGetVersion failed with %d", res);
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
if (cuda_driver_version < CUDA_DRIVER_MIN_VERSION) {
@ -963,12 +1009,14 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
"Minimum requirement is %d",
cuda_driver_version,
CUDA_DRIVER_MIN_VERSION);
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
if (cuda_pfn_func_loader()) {
rte_cuda_log(ERR, "CUDA PFN functions not found in library");
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
}
@ -982,7 +1030,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceGetByPCIBusId name %s failed with %d: %s",
dev->device->name, res, err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
res = pfn_cuDevicePrimaryCtxRetain(&pctx, cu_dev_id);
@ -990,19 +1039,22 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDevicePrimaryCtxRetain name %s failed with %d: %s",
dev->device->name, res, err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
res = pfn_cuCtxGetApiVersion(pctx, &cuda_api_version);
if (res != 0) {
rte_cuda_log(ERR, "cuCtxGetApiVersion failed with %d", res);
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
if (cuda_api_version < CUDA_API_MIN_VERSION) {
rte_cuda_log(ERR, "CUDA API version found is %d Minimum requirement is %d",
cuda_api_version, CUDA_API_MIN_VERSION);
return -ENOTSUP;
rte_errno = ENOTSUP;
return -rte_errno;
}
dev->mpshared->info.context = (uint64_t)pctx;
@ -1019,7 +1071,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceGetAttribute failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
dev->mpshared->info.processor_count = (uint32_t)processor_count;
@ -1029,7 +1082,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceTotalMem failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
/*
@ -1040,7 +1094,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
RTE_CACHE_LINE_SIZE);
if (dev->mpshared->dev_private == NULL) {
rte_cuda_log(ERR, "Failed to allocate memory for GPU process private");
return -ENOMEM;
rte_errno = EPERM;
return -rte_errno;
}
private = (struct cuda_info *)dev->mpshared->dev_private;
@ -1052,7 +1107,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceGetName failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
res = pfn_cuDeviceGetAttribute(&(private->gdr_supported),
@ -1062,7 +1118,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceGetAttribute failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
if (private->gdr_supported == 0)
@ -1077,7 +1134,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
rte_cuda_log(ERR,
"cuDeviceGetAttribute failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
if (private->gdr_write_ordering == CU_GPU_DIRECT_RDMA_WRITES_ORDERING_NONE) {
@ -1088,7 +1146,8 @@ cuda_gpu_probe(__rte_unused struct rte_pci_driver *pci_drv, struct rte_pci_devic
pfn_cuGetErrorString(res, &(err_string));
rte_cuda_log(ERR, "cuDeviceGetAttribute failed with %s",
err_string);
return -EPERM;
rte_errno = EPERM;
return -rte_errno;
}
if (private->gdr_flush_type != CU_FLUSH_GPU_DIRECT_RDMA_WRITES_OPTION_HOST)
@ -1118,14 +1177,17 @@ cuda_gpu_remove(struct rte_pci_device *pci_dev)
int ret;
uint8_t gpu_id;
if (pci_dev == NULL)
return -EINVAL;
if (pci_dev == NULL) {
rte_errno = ENODEV;
return -rte_errno;
}
dev = rte_gpu_get_by_name(pci_dev->device.name);
if (dev == NULL) {
rte_cuda_log(ERR, "Couldn't find HW dev \"%s\" to uninitialise it",
pci_dev->device.name);
return -ENODEV;
rte_errno = ENODEV;
return -rte_errno;
}
gpu_id = dev->mpshared->info.dev_id;