test mbuf attach
[dpdk.git] / app / test-compress-perf / main.c
index f52b98d..ed21605 100644 (file)
  * Copyright(c) 2018 Intel Corporation
  */
 
+#include <signal.h>
+#include <sys/types.h>
+#include <unistd.h>
+
 #include <rte_malloc.h>
 #include <rte_eal.h>
 #include <rte_log.h>
 #include <rte_compressdev.h>
 
+#include "comp_perf.h"
 #include "comp_perf_options.h"
+#include "comp_perf_test_common.h"
+#include "comp_perf_test_cyclecount.h"
+#include "comp_perf_test_throughput.h"
+#include "comp_perf_test_verify.h"
+
+#define NUM_MAX_XFORMS 16
+#define NUM_MAX_INFLIGHT_OPS 512
+
+__extension__
+const char *comp_perf_test_type_strs[] = {
+       [CPERF_TEST_TYPE_THROUGHPUT] = "throughput",
+       [CPERF_TEST_TYPE_VERIFY] = "verify",
+       [CPERF_TEST_TYPE_PMDCC] = "pmd-cyclecount"
+};
+
+__extension__
+static const struct cperf_test cperf_testmap[] = {
+       [CPERF_TEST_TYPE_THROUGHPUT] = {
+                       cperf_throughput_test_constructor,
+                       cperf_throughput_test_runner,
+                       cperf_throughput_test_destructor
+
+       },
+       [CPERF_TEST_TYPE_VERIFY] = {
+                       cperf_verify_test_constructor,
+                       cperf_verify_test_runner,
+                       cperf_verify_test_destructor
+       },
+
+       [CPERF_TEST_TYPE_PMDCC] = {
+                       cperf_cyclecount_test_constructor,
+                       cperf_cyclecount_test_runner,
+                       cperf_cyclecount_test_destructor
+       }
+};
+
+static struct comp_test_data *test_data;
+
+static int
+comp_perf_check_capabilities(struct comp_test_data *test_data, uint8_t cdev_id)
+{
+       const struct rte_compressdev_capabilities *cap;
+
+       cap = rte_compressdev_capability_get(cdev_id,
+                                            RTE_COMP_ALGO_DEFLATE);
+
+       if (cap == NULL) {
+               RTE_LOG(ERR, USER1,
+                       "Compress device does not support DEFLATE\n");
+               return -1;
+       }
+
+       uint64_t comp_flags = cap->comp_feature_flags;
+
+       /* Huffman enconding */
+       if (test_data->huffman_enc == RTE_COMP_HUFFMAN_FIXED &&
+                       (comp_flags & RTE_COMP_FF_HUFFMAN_FIXED) == 0) {
+               RTE_LOG(ERR, USER1,
+                       "Compress device does not supported Fixed Huffman\n");
+               return -1;
+       }
+
+       if (test_data->huffman_enc == RTE_COMP_HUFFMAN_DYNAMIC &&
+                       (comp_flags & RTE_COMP_FF_HUFFMAN_DYNAMIC) == 0) {
+               RTE_LOG(ERR, USER1,
+                       "Compress device does not supported Dynamic Huffman\n");
+               return -1;
+       }
+
+       /* Window size */
+       if (test_data->window_sz != -1) {
+               if (param_range_check(test_data->window_sz, &cap->window_size)
+                               < 0) {
+                       RTE_LOG(ERR, USER1,
+                               "Compress device does not support "
+                               "this window size\n");
+                       return -1;
+               }
+       } else
+               /* Set window size to PMD maximum if none was specified */
+               test_data->window_sz = cap->window_size.max;
+
+       /* Check if chained mbufs is supported */
+       if (test_data->max_sgl_segs > 1  &&
+                       (comp_flags & RTE_COMP_FF_OOP_SGL_IN_SGL_OUT) == 0) {
+               RTE_LOG(INFO, USER1, "Compress device does not support "
+                               "chained mbufs. Max SGL segments set to 1\n");
+               test_data->max_sgl_segs = 1;
+       }
+
+       /* Level 0 support */
+       if (test_data->level_lst.min == 0 &&
+                       (comp_flags & RTE_COMP_FF_NONCOMPRESSED_BLOCKS) == 0) {
+               RTE_LOG(ERR, USER1, "Compress device does not support "
+                               "level 0 (no compression)\n");
+               return -1;
+       }
+
+       return 0;
+}
+
+static int
+comp_perf_initialize_compressdev(struct comp_test_data *test_data,
+                                uint8_t *enabled_cdevs)
+{
+       uint8_t enabled_cdev_count, nb_lcores, cdev_id;
+       unsigned int i, j;
+       int ret;
+
+       enabled_cdev_count = rte_compressdev_devices_get(test_data->driver_name,
+                       enabled_cdevs, RTE_COMPRESS_MAX_DEVS);
+       if (enabled_cdev_count == 0) {
+               RTE_LOG(ERR, USER1, "No compress devices type %s available,"
+                                   " please check the list of specified devices in EAL section\n",
+                               test_data->driver_name);
+               return -EINVAL;
+       }
+
+       nb_lcores = rte_lcore_count() - 1;
+       /*
+        * Use fewer devices,
+        * if there are more available than cores.
+        */
+       if (enabled_cdev_count > nb_lcores) {
+               if (nb_lcores == 0) {
+                       RTE_LOG(ERR, USER1, "Cannot run with 0 cores! Increase the number of cores\n");
+                       return -EINVAL;
+               }
+               enabled_cdev_count = nb_lcores;
+               RTE_LOG(INFO, USER1,
+                       "There's more available devices than cores!"
+                       " The number of devices has been aligned to %d cores\n",
+                       nb_lcores);
+       }
+
+       /*
+        * Calculate number of needed queue pairs, based on the amount
+        * of available number of logical cores and compression devices.
+        * For instance, if there are 4 cores and 2 compression devices,
+        * 2 queue pairs will be set up per device.
+        * One queue pair per one core.
+        * if e.g.: there're 3 cores and 2 compression devices,
+        * 2 queue pairs will be set up per device but one queue pair
+        * will left unused in the last one device
+        */
+       test_data->nb_qps = (nb_lcores % enabled_cdev_count) ?
+                               (nb_lcores / enabled_cdev_count) + 1 :
+                               nb_lcores / enabled_cdev_count;
+
+       for (i = 0; i < enabled_cdev_count &&
+                       i < RTE_COMPRESS_MAX_DEVS; i++,
+                                       nb_lcores -= test_data->nb_qps) {
+               cdev_id = enabled_cdevs[i];
+
+               struct rte_compressdev_info cdev_info;
+               uint8_t socket_id = rte_compressdev_socket_id(cdev_id);
+
+               rte_compressdev_info_get(cdev_id, &cdev_info);
+               if (cdev_info.max_nb_queue_pairs &&
+                       test_data->nb_qps > cdev_info.max_nb_queue_pairs) {
+                       RTE_LOG(ERR, USER1,
+                               "Number of needed queue pairs is higher "
+                               "than the maximum number of queue pairs "
+                               "per device.\n");
+                       RTE_LOG(ERR, USER1,
+                               "Lower the number of cores or increase "
+                               "the number of crypto devices\n");
+                       return -EINVAL;
+               }
+
+               if (comp_perf_check_capabilities(test_data, cdev_id) < 0)
+                       return -EINVAL;
+
+               /* Configure compressdev */
+               struct rte_compressdev_config config = {
+                       .socket_id = socket_id,
+                       .nb_queue_pairs = nb_lcores > test_data->nb_qps
+                                       ? test_data->nb_qps : nb_lcores,
+                       .max_nb_priv_xforms = NUM_MAX_XFORMS,
+                       .max_nb_streams = 0
+               };
+
+               if (rte_compressdev_configure(cdev_id, &config) < 0) {
+                       RTE_LOG(ERR, USER1, "Device configuration failed\n");
+                       return -EINVAL;
+               }
+
+               for (j = 0; j < test_data->nb_qps; j++) {
+                       ret = rte_compressdev_queue_pair_setup(cdev_id, j,
+                                       NUM_MAX_INFLIGHT_OPS, socket_id);
+                       if (ret < 0) {
+                               RTE_LOG(ERR, USER1,
+                             "Failed to setup queue pair %u on compressdev %u",
+                                       j, cdev_id);
+                               return -EINVAL;
+                       }
+               }
+
+               ret = rte_compressdev_start(cdev_id);
+               if (ret < 0) {
+                       RTE_LOG(ERR, USER1,
+                               "Failed to start device %u: error %d\n",
+                               cdev_id, ret);
+                       return -EPERM;
+               }
+       }
+
+       return enabled_cdev_count;
+}
+
+static int
+comp_perf_dump_input_data(struct comp_test_data *test_data)
+{
+       FILE *f = fopen(test_data->input_file, "r");
+       int ret = -1;
+
+       if (f == NULL) {
+               RTE_LOG(ERR, USER1, "Input file could not be opened\n");
+               return -1;
+       }
+
+       if (fseek(f, 0, SEEK_END) != 0) {
+               RTE_LOG(ERR, USER1, "Size of input could not be calculated\n");
+               goto end;
+       }
+       size_t actual_file_sz = ftell(f);
+       /* If extended input data size has not been set,
+        * input data size = file size
+        */
+
+       if (test_data->input_data_sz == 0)
+               test_data->input_data_sz = actual_file_sz;
+
+       if (test_data->input_data_sz <= 0 || actual_file_sz <= 0 ||
+                       fseek(f, 0, SEEK_SET) != 0) {
+               RTE_LOG(ERR, USER1, "Size of input could not be calculated\n");
+               goto end;
+       }
+
+       test_data->input_data = rte_zmalloc_socket(NULL,
+                               test_data->input_data_sz, 0, rte_socket_id());
+
+       if (test_data->input_data == NULL) {
+               RTE_LOG(ERR, USER1, "Memory to hold the data from the input "
+                               "file could not be allocated\n");
+               goto end;
+       }
+
+       size_t remaining_data = test_data->input_data_sz;
+       uint8_t *data = test_data->input_data;
+
+       while (remaining_data > 0) {
+               size_t data_to_read = RTE_MIN(remaining_data, actual_file_sz);
+
+               if (fread(data, data_to_read, 1, f) != 1) {
+                       RTE_LOG(ERR, USER1, "Input file could not be read\n");
+                       goto end;
+               }
+               if (fseek(f, 0, SEEK_SET) != 0) {
+                       RTE_LOG(ERR, USER1,
+                               "Size of input could not be calculated\n");
+                       goto end;
+               }
+               remaining_data -= data_to_read;
+               data += data_to_read;
+       }
+
+       printf("\n");
+       if (test_data->input_data_sz > actual_file_sz)
+               RTE_LOG(INFO, USER1,
+                 "%zu bytes read from file %s, extending the file %.2f times\n",
+                       test_data->input_data_sz, test_data->input_file,
+                       (double)test_data->input_data_sz/actual_file_sz);
+       else
+               RTE_LOG(INFO, USER1,
+                       "%zu bytes read from file %s\n",
+                       test_data->input_data_sz, test_data->input_file);
+
+       ret = 0;
+
+end:
+       fclose(f);
+       return ret;
+}
+
+static void
+comp_perf_cleanup_on_signal(int signalNumber __rte_unused)
+{
+       test_data->perf_comp_force_stop = 1;
+}
+
+static void
+comp_perf_register_cleanup_on_signal(void)
+{
+       signal(SIGTERM, comp_perf_cleanup_on_signal);
+       signal(SIGINT, comp_perf_cleanup_on_signal);
+}
 
 int
 main(int argc, char **argv)
 {
-       int ret;
-       struct comp_test_data *test_data;
+       uint8_t level_idx = 0;
+       int ret, i;
+       void *ctx[RTE_MAX_LCORE] = {};
+       uint8_t enabled_cdevs[RTE_COMPRESS_MAX_DEVS];
+       int nb_compressdevs = 0;
+       uint16_t total_nb_qps = 0;
+       uint8_t cdev_id;
+       uint32_t lcore_id;
 
        /* Initialise DPDK EAL */
        ret = rte_eal_init(argc, argv);
@@ -29,24 +337,212 @@ main(int argc, char **argv)
                rte_exit(EXIT_FAILURE, "Cannot reserve memory in socket %d\n",
                                rte_socket_id());
 
+       comp_perf_register_cleanup_on_signal();
+
+       ret = EXIT_SUCCESS;
+       test_data->cleanup = ST_TEST_DATA;
        comp_perf_options_default(test_data);
 
        if (comp_perf_options_parse(test_data, argc, argv) < 0) {
                RTE_LOG(ERR, USER1,
                        "Parsing one or more user options failed\n");
                ret = EXIT_FAILURE;
-               goto err;
+               goto end;
        }
 
        if (comp_perf_options_check(test_data) < 0) {
                ret = EXIT_FAILURE;
-               goto err;
+               goto end;
        }
 
-       ret = EXIT_SUCCESS;
+       nb_compressdevs =
+               comp_perf_initialize_compressdev(test_data, enabled_cdevs);
+
+       if (nb_compressdevs < 1) {
+               ret = EXIT_FAILURE;
+               goto end;
+       }
+
+       test_data->cleanup = ST_COMPDEV;
+       if (comp_perf_dump_input_data(test_data) < 0) {
+               ret = EXIT_FAILURE;
+               goto end;
+       }
+
+       test_data->cleanup = ST_INPUT_DATA;
+
+       if (test_data->level_lst.inc != 0)
+               test_data->level = test_data->level_lst.min;
+       else
+               test_data->level = test_data->level_lst.list[0];
+
+       printf("\nApp uses socket: %u\n", rte_socket_id());
+       printf("Burst size = %u\n", test_data->burst_sz);
+       printf("Input data size = %zu\n", test_data->input_data_sz);
+       if (test_data->test == CPERF_TEST_TYPE_PMDCC)
+               printf("Cycle-count delay = %u [us]\n",
+                      test_data->cyclecount_delay);
+
+       test_data->cleanup = ST_DURING_TEST;
+       total_nb_qps = nb_compressdevs * test_data->nb_qps;
+
+       i = 0;
+       uint8_t qp_id = 0, cdev_index = 0;
+
+       RTE_LCORE_FOREACH_SLAVE(lcore_id) {
+
+               if (i == total_nb_qps)
+                       break;
+
+               cdev_id = enabled_cdevs[cdev_index];
+               ctx[i] = cperf_testmap[test_data->test].constructor(
+                                                       cdev_id, qp_id,
+                                                       test_data);
+               if (ctx[i] == NULL) {
+                       RTE_LOG(ERR, USER1, "Test run constructor failed\n");
+                       goto end;
+               }
+               qp_id = (qp_id + 1) % test_data->nb_qps;
+               if (qp_id == 0)
+                       cdev_index++;
+               i++;
+       }
+
+       print_test_dynamics(test_data);
+
+       while (test_data->level <= test_data->level_lst.max) {
+
+               i = 0;
+               RTE_LCORE_FOREACH_SLAVE(lcore_id) {
+
+                       if (i == total_nb_qps)
+                               break;
+
+                       rte_eal_remote_launch(
+                                       cperf_testmap[test_data->test].runner,
+                                       ctx[i], lcore_id);
+                       i++;
+               }
+               i = 0;
+               RTE_LCORE_FOREACH_SLAVE(lcore_id) {
+
+                       if (i == total_nb_qps)
+                               break;
+                       ret |= rte_eal_wait_lcore(lcore_id);
+                       i++;
+               }
 
-err:
-       rte_free(test_data);
+               if (ret != EXIT_SUCCESS)
+                       break;
 
+               if (test_data->level_lst.inc != 0)
+                       test_data->level += test_data->level_lst.inc;
+               else {
+                       if (++level_idx == test_data->level_lst.count)
+                               break;
+                       test_data->level = test_data->level_lst.list[level_idx];
+               }
+       }
+
+end:
+       switch (test_data->cleanup) {
+
+       case ST_DURING_TEST:
+               i = 0;
+               RTE_LCORE_FOREACH_SLAVE(lcore_id) {
+                       if (i == total_nb_qps)
+                               break;
+
+                       if (ctx[i] && cperf_testmap[test_data->test].destructor)
+                               cperf_testmap[test_data->test].destructor(
+                                                                       ctx[i]);
+                       i++;
+               }
+               /* fallthrough */
+       case ST_INPUT_DATA:
+               rte_free(test_data->input_data);
+               /* fallthrough */
+       case ST_COMPDEV:
+               for (i = 0; i < nb_compressdevs &&
+                    i < RTE_COMPRESS_MAX_DEVS; i++) {
+                       rte_compressdev_stop(enabled_cdevs[i]);
+                       rte_compressdev_close(enabled_cdevs[i]);
+               }
+               /* fallthrough */
+       case ST_TEST_DATA:
+               rte_free(test_data);
+               /* fallthrough */
+       case ST_CLEAR:
+       default:
+               i = rte_eal_cleanup();
+               if (i) {
+                       RTE_LOG(ERR, USER1,
+                               "Error from rte_eal_cleanup(), %d\n", i);
+                       ret = i;
+               }
+               break;
+       }
        return ret;
 }
+
+__rte_weak void *
+cperf_cyclecount_test_constructor(uint8_t dev_id __rte_unused,
+                                uint16_t qp_id __rte_unused,
+                                struct comp_test_data *options __rte_unused)
+{
+       RTE_LOG(INFO, USER1, "Cycle count test is not supported yet\n");
+       return NULL;
+}
+
+__rte_weak void
+cperf_cyclecount_test_destructor(void *arg __rte_unused)
+{
+       RTE_LOG(INFO, USER1, "Something wrong happened!!!\n");
+}
+
+__rte_weak int
+cperf_cyclecount_test_runner(void *test_ctx __rte_unused)
+{
+       return 0;
+}
+
+__rte_weak void *
+cperf_throughput_test_constructor(uint8_t dev_id __rte_unused,
+                                uint16_t qp_id __rte_unused,
+                                struct comp_test_data *options __rte_unused)
+{
+       RTE_LOG(INFO, USER1, "Benchmark test is not supported yet\n");
+       return NULL;
+}
+
+__rte_weak void
+cperf_throughput_test_destructor(void *arg __rte_unused)
+{
+
+}
+
+__rte_weak int
+cperf_throughput_test_runner(void *test_ctx __rte_unused)
+{
+       return 0;
+}
+__rte_weak void *
+cperf_verify_test_constructor(uint8_t dev_id __rte_unused,
+                                uint16_t qp_id __rte_unused,
+                                struct comp_test_data *options __rte_unused)
+{
+       RTE_LOG(INFO, USER1, "Verify test is not supported yet\n");
+       return NULL;
+}
+
+__rte_weak void
+cperf_verify_test_destructor(void *arg __rte_unused)
+{
+
+}
+
+__rte_weak int
+cperf_verify_test_runner(void *test_ctx __rte_unused)
+{
+       return 0;
+}