test/ring: enhance mp/mc coverage
authorJoyce Kong <joyce.kong@arm.com>
Mon, 9 Sep 2019 05:19:00 +0000 (13:19 +0800)
committerDavid Marchand <david.marchand@redhat.com>
Thu, 24 Oct 2019 07:57:31 +0000 (09:57 +0200)
Run ring perf test on all available cores to really verify MPMC operations.
The old way of running on a pair of cores is not enough for MPMC rings.

Suggested-by: Gavin Hu <gavin.hu@arm.com>
Signed-off-by: Joyce Kong <joyce.kong@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Ola Liljedahl <ola.liljedahl@arm.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
app/test/test_ring_perf.c

index b6ad703..70ee46f 100644 (file)
@@ -1,5 +1,6 @@
 /* SPDX-License-Identifier: BSD-3-Clause
  * Copyright(c) 2010-2014 Intel Corporation
+ * Copyright(c) 2019 Arm Limited
  */
 
 
@@ -9,6 +10,7 @@
 #include <rte_cycles.h>
 #include <rte_launch.h>
 #include <rte_pause.h>
+#include <string.h>
 
 #include "test.h"
 
@@ -20,6 +22,7 @@
  *  * Empty ring dequeue
  *  * Enqueue/dequeue of bursts in 1 threads
  *  * Enqueue/dequeue of bursts in 2 threads
+ *  * Enqueue/dequeue of bursts in all available threads
  */
 
 #define RING_NAME "RING_PERF"
@@ -258,6 +261,76 @@ run_on_core_pair(struct lcore_pair *cores, struct rte_ring *r,
        }
 }
 
+static rte_atomic32_t synchro;
+static uint64_t queue_count[RTE_MAX_LCORE];
+
+#define TIME_MS 100
+
+static int
+load_loop_fn(void *p)
+{
+       uint64_t time_diff = 0;
+       uint64_t begin = 0;
+       uint64_t hz = rte_get_timer_hz();
+       uint64_t lcount = 0;
+       const unsigned int lcore = rte_lcore_id();
+       struct thread_params *params = p;
+       void *burst[MAX_BURST] = {0};
+
+       /* wait synchro for slaves */
+       if (lcore != rte_get_master_lcore())
+               while (rte_atomic32_read(&synchro) == 0)
+                       rte_pause();
+
+       begin = rte_get_timer_cycles();
+       while (time_diff < hz * TIME_MS / 1000) {
+               rte_ring_mp_enqueue_bulk(params->r, burst, params->size, NULL);
+               rte_ring_mc_dequeue_bulk(params->r, burst, params->size, NULL);
+               lcount++;
+               time_diff = rte_get_timer_cycles() - begin;
+       }
+       queue_count[lcore] = lcount;
+       return 0;
+}
+
+static int
+run_on_all_cores(struct rte_ring *r)
+{
+       uint64_t total = 0;
+       struct thread_params param;
+       unsigned int i, c;
+
+       memset(&param, 0, sizeof(struct thread_params));
+       for (i = 0; i < RTE_DIM(bulk_sizes); i++) {
+               printf("\nBulk enq/dequeue count on size %u\n", bulk_sizes[i]);
+               param.size = bulk_sizes[i];
+               param.r = r;
+
+               /* clear synchro and start slaves */
+               rte_atomic32_set(&synchro, 0);
+               if (rte_eal_mp_remote_launch(load_loop_fn, &param,
+                       SKIP_MASTER) < 0)
+                       return -1;
+
+               /* start synchro and launch test on master */
+               rte_atomic32_set(&synchro, 1);
+               load_loop_fn(&param);
+
+               rte_eal_mp_wait_lcore();
+
+               RTE_LCORE_FOREACH(c) {
+                       printf("Core [%u] count = %"PRIu64"\n",
+                                       c, queue_count[c]);
+                       total += queue_count[c];
+               }
+
+               printf("Total count (size: %u): %"PRIu64"\n",
+                               bulk_sizes[i], total);
+       }
+
+       return 0;
+}
+
 /*
  * Test function that determines how long an enqueue + dequeue of a single item
  * takes on a single lcore. Result is for comparison with the bulk enq+deq.
@@ -404,6 +477,10 @@ test_ring_perf(void)
                printf("\n### Testing using two NUMA nodes ###\n");
                run_on_core_pair(&cores, r, enqueue_bulk, dequeue_bulk);
        }
+
+       printf("\n### Testing using all slave nodes ###\n");
+       run_on_all_cores(r);
+
        rte_ring_free(r);
        return 0;
 }