1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright (c) 2021 NVIDIA Corporation & Affiliates
14 #include <rte_common.h>
15 #include <rte_malloc.h>
16 #include <rte_memory.h>
18 #include <rte_ether.h>
19 #include <rte_ethdev.h>
20 #include <rte_mempool.h>
23 #include <rte_gpudev.h>
31 usage(const char *prog_name)
33 printf("%s [EAL options] --\n",
38 args_parse(int argc, char **argv)
44 static struct option lgopts[] = {
45 { "help", 0, 0, ARG_HELP},
51 while ((opt = getopt_long(argc, argvopt, "",
52 lgopts, &opt_idx)) != EOF) {
59 rte_exit(EXIT_FAILURE, "Invalid option: %s\n", argv[optind]);
66 alloc_gpu_memory(uint16_t gpu_id)
70 size_t buf_bytes = 1024;
73 printf("\n=======> TEST: Allocate GPU memory\n");
75 /* Alloc memory on GPU 0 */
76 ptr_1 = rte_gpu_mem_alloc(gpu_id, buf_bytes);
78 fprintf(stderr, "rte_gpu_mem_alloc GPU memory returned error\n");
81 printf("GPU memory allocated at 0x%p %zdB\n", ptr_1, buf_bytes);
83 ptr_2 = rte_gpu_mem_alloc(gpu_id, buf_bytes);
85 fprintf(stderr, "rte_gpu_mem_alloc GPU memory returned error\n");
88 printf("GPU memory allocated at 0x%p %zdB\n", ptr_2, buf_bytes);
90 ret = rte_gpu_mem_free(gpu_id, (uint8_t *)(ptr_1)+0x700);
92 printf("GPU memory 0x%p + 0x700 NOT freed because of memory address not recognized by driver\n", ptr_1);
94 fprintf(stderr, "rte_gpu_mem_free erroneously freed GPU memory 0x%p + 0x700\n", ptr_1);
98 ret = rte_gpu_mem_free(gpu_id, ptr_2);
100 fprintf(stderr, "rte_gpu_mem_free returned error %d\n", ret);
103 printf("GPU memory 0x%p freed\n", ptr_2);
105 ret = rte_gpu_mem_free(gpu_id, ptr_1);
107 fprintf(stderr, "rte_gpu_mem_free returned error %d\n", ret);
110 printf("GPU memory 0x%p freed\n", ptr_1);
116 register_cpu_memory(uint16_t gpu_id)
119 size_t buf_bytes = 1024;
122 printf("\n=======> TEST: Register CPU memory\n");
124 /* Alloc memory on CPU visible from GPU 0 */
125 ptr = rte_zmalloc(NULL, buf_bytes, 0);
127 fprintf(stderr, "Failed to allocate CPU memory.\n");
131 ret = rte_gpu_mem_register(gpu_id, buf_bytes, ptr);
133 fprintf(stderr, "rte_gpu_mem_register CPU memory returned error %d\n", ret);
136 printf("CPU memory registered at 0x%p %zdB\n", ptr, buf_bytes);
138 ret = rte_gpu_mem_unregister(gpu_id, (uint8_t *)(ptr)+0x700);
140 printf("CPU memory 0x%p + 0x700 NOT unregistered because of memory address not recognized by driver\n", ptr);
142 fprintf(stderr, "rte_gpu_mem_free erroneously freed GPU memory 0x%p + 0x700\n", ptr);
145 printf("CPU memory 0x%p unregistered\n", ptr);
147 ret = rte_gpu_mem_unregister(gpu_id, ptr);
149 fprintf(stderr, "rte_gpu_mem_unregister returned error %d\n", ret);
152 printf("CPU memory 0x%p unregistered\n", ptr);
158 create_update_comm_flag(uint16_t gpu_id)
160 struct rte_gpu_comm_flag devflag;
165 printf("\n=======> TEST: Communication flag\n");
167 ret = rte_gpu_comm_create_flag(gpu_id, &devflag, RTE_GPU_COMM_FLAG_CPU);
169 fprintf(stderr, "rte_gpu_comm_create_flag returned error %d\n", ret);
174 ret = rte_gpu_comm_set_flag(&devflag, set_val);
176 fprintf(stderr, "rte_gpu_comm_set_flag returned error %d\n", ret);
180 ret = rte_gpu_comm_get_flag_value(&devflag, &get_val);
182 fprintf(stderr, "rte_gpu_comm_get_flag_value returned error %d\n", ret);
186 printf("Communication flag value at 0x%p was set to %d and current value is %d\n", devflag.ptr, set_val, get_val);
189 ret = rte_gpu_comm_set_flag(&devflag, set_val);
191 fprintf(stderr, "rte_gpu_comm_set_flag returned error %d\n", ret);
195 ret = rte_gpu_comm_get_flag_value(&devflag, &get_val);
197 fprintf(stderr, "rte_gpu_comm_get_flag_value returned error %d\n", ret);
201 printf("Communication flag value at 0x%p was set to %d and current value is %d\n", devflag.ptr, set_val, get_val);
203 ret = rte_gpu_comm_destroy_flag(&devflag);
205 fprintf(stderr, "rte_gpu_comm_destroy_flags returned error %d\n", ret);
213 simulate_gpu_task(struct rte_gpu_comm_list *comm_list_item, int num_pkts)
217 if (comm_list_item == NULL)
220 for (idx = 0; idx < num_pkts; idx++) {
222 * consume(comm_list_item->pkt_list[idx].addr);
225 comm_list_item->status = RTE_GPU_COMM_LIST_DONE;
231 create_update_comm_list(uint16_t gpu_id)
235 struct rte_gpu_comm_list *comm_list;
236 uint32_t num_comm_items = 1024;
237 struct rte_mbuf *mbufs[10];
239 printf("\n=======> TEST: Communication list\n");
241 comm_list = rte_gpu_comm_create_list(gpu_id, num_comm_items);
242 if (comm_list == NULL) {
243 fprintf(stderr, "rte_gpu_comm_create_list returned error %d\n", ret);
248 * Simulate DPDK receive functions like rte_eth_rx_burst()
250 for (i = 0; i < 10; i++) {
251 mbufs[i] = rte_zmalloc(NULL, sizeof(struct rte_mbuf), 0);
252 if (mbufs[i] == NULL) {
253 fprintf(stderr, "Failed to allocate fake mbufs in CPU memory.\n");
257 memset(mbufs[i], 0, sizeof(struct rte_mbuf));
261 * Populate just the first item of the list
263 ret = rte_gpu_comm_populate_list_pkts(&(comm_list[0]), mbufs, 10);
265 fprintf(stderr, "rte_gpu_comm_populate_list_pkts returned error %d\n", ret);
269 ret = rte_gpu_comm_cleanup_list(&(comm_list[0]));
271 fprintf(stderr, "rte_gpu_comm_cleanup_list erroneously cleaned the list even if packets have not been consumed yet\n");
274 fprintf(stderr, "rte_gpu_comm_cleanup_list correctly didn't clean up the packets because they have not been consumed yet\n");
277 * Simulate a GPU tasks going through the packet list to consume
278 * mbufs packets and release them
280 simulate_gpu_task(&(comm_list[0]), 10);
283 * Packets have been consumed, now the communication item
284 * and the related mbufs can be all released
286 ret = rte_gpu_comm_cleanup_list(&(comm_list[0]));
288 fprintf(stderr, "rte_gpu_comm_cleanup_list returned error %d\n", ret);
292 ret = rte_gpu_comm_destroy_list(comm_list, num_comm_items);
294 fprintf(stderr, "rte_gpu_comm_destroy_list returned error %d\n", ret);
298 for (i = 0; i < 10; i++)
301 printf("\nCommunication list test passed!\n");
306 main(int argc, char **argv)
311 struct rte_gpu_info ginfo;
314 ret = rte_eal_init(argc, argv);
316 rte_exit(EXIT_FAILURE, "EAL init failed\n");
320 args_parse(argc, argv);
324 nb_gpus = rte_gpu_count_avail();
325 printf("\n\nDPDK found %d GPUs:\n", nb_gpus);
326 RTE_GPU_FOREACH(gpu_id)
328 if (rte_gpu_info_get(gpu_id, &ginfo))
329 rte_exit(EXIT_FAILURE, "rte_gpu_info_get error - bye\n");
331 printf("\tGPU ID %d\n\t\tparent ID %d GPU Bus ID %s NUMA node %d Tot memory %.02f MB, Tot processors %d\n",
336 (((float)ginfo.total_memory)/(float)1024)/(float)1024,
337 ginfo.processor_count
343 fprintf(stderr, "Need at least one GPU on the system to run the example\n");
352 alloc_gpu_memory(gpu_id);
353 register_cpu_memory(gpu_id);
356 * Communication items test
358 create_update_comm_flag(gpu_id);
359 create_update_comm_list(gpu_id);
361 /* clean up the EAL */