test/distributor: fix freeing mbufs
authorLukasz Wojciechowski <l.wojciechow@partner.samsung.com>
Sat, 17 Oct 2020 03:06:51 +0000 (05:06 +0200)
committerDavid Marchand <david.marchand@redhat.com>
Mon, 19 Oct 2020 08:57:17 +0000 (10:57 +0200)
Sanity tests with mbuf alloc and shutdown tests assume that
mbufs passed to worker cores are freed in handlers.
Such packets should not be returned to the distributor's main
core. The only packets that should be returned are the packets
send after completion of the tests in quit_workers function.

This patch stops returning mbufs to distributor's core.
In case of shutdown tests it is impossible to determine
how worker and distributor threads would synchronize.
Packets used by tests should be freed and packets used during
quit_workers() shouldn't. That's why returning mbufs to mempool
is moved to test procedure run on distributor thread
from worker threads.

Additionally this patch cleans up unused variables.

Fixes: c0de0eb82e40 ("distributor: switch over to new API")
Cc: stable@dpdk.org
Signed-off-by: Lukasz Wojciechowski <l.wojciechow@partner.samsung.com>
Acked-by: David Hunt <david.hunt@intel.com>
app/test/test_distributor.c

index 6cd7a2e..ec1fe34 100644 (file)
@@ -63,20 +63,18 @@ handle_work(void *arg)
        struct rte_mbuf *buf[8] __rte_cache_aligned;
        struct worker_params *wp = arg;
        struct rte_distributor *db = wp->dist;
-       unsigned int count = 0, num;
+       unsigned int num;
        unsigned int id = __atomic_fetch_add(&worker_idx, 1, __ATOMIC_RELAXED);
 
        num = rte_distributor_get_pkt(db, id, buf, NULL, 0);
        while (!quit) {
                __atomic_fetch_add(&worker_stats[id].handled_packets, num,
                                __ATOMIC_RELAXED);
-               count += num;
                num = rte_distributor_get_pkt(db, id,
                                buf, buf, num);
        }
        __atomic_fetch_add(&worker_stats[id].handled_packets, num,
                        __ATOMIC_RELAXED);
-       count += num;
        rte_distributor_return_pkt(db, id, buf, num);
        return 0;
 }
@@ -268,7 +266,6 @@ handle_work_with_free_mbufs(void *arg)
        struct rte_mbuf *buf[8] __rte_cache_aligned;
        struct worker_params *wp = arg;
        struct rte_distributor *d = wp->dist;
-       unsigned int count = 0;
        unsigned int i;
        unsigned int num;
        unsigned int id = __atomic_fetch_add(&worker_idx, 1, __ATOMIC_RELAXED);
@@ -276,13 +273,11 @@ handle_work_with_free_mbufs(void *arg)
        num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
        while (!quit) {
                worker_stats[id].handled_packets += num;
-               count += num;
                for (i = 0; i < num; i++)
                        rte_pktmbuf_free(buf[i]);
                num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
        }
        worker_stats[id].handled_packets += num;
-       count += num;
        rte_distributor_return_pkt(d, id, buf, num);
        return 0;
 }
@@ -308,7 +303,6 @@ sanity_test_with_mbuf_alloc(struct worker_params *wp, struct rte_mempool *p)
                        rte_distributor_process(d, NULL, 0);
                for (j = 0; j < BURST; j++) {
                        bufs[j]->hash.usr = (i+j) << 1;
-                       rte_mbuf_refcnt_set(bufs[j], 1);
                }
 
                rte_distributor_process(d, bufs, BURST);
@@ -332,15 +326,10 @@ sanity_test_with_mbuf_alloc(struct worker_params *wp, struct rte_mempool *p)
 static int
 handle_work_for_shutdown_test(void *arg)
 {
-       struct rte_mbuf *pkt = NULL;
        struct rte_mbuf *buf[8] __rte_cache_aligned;
        struct worker_params *wp = arg;
        struct rte_distributor *d = wp->dist;
-       unsigned int count = 0;
        unsigned int num;
-       unsigned int total = 0;
-       unsigned int i;
-       unsigned int returned = 0;
        unsigned int zero_id = 0;
        unsigned int zero_unset;
        const unsigned int id = __atomic_fetch_add(&worker_idx, 1,
@@ -359,9 +348,6 @@ handle_work_for_shutdown_test(void *arg)
         * for zero_quit */
        while (!quit && !(id == zero_id && zero_quit)) {
                worker_stats[id].handled_packets += num;
-               count += num;
-               for (i = 0; i < num; i++)
-                       rte_pktmbuf_free(buf[i]);
                num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
 
                if (num > 0) {
@@ -370,14 +356,12 @@ handle_work_for_shutdown_test(void *arg)
                                false, __ATOMIC_ACQ_REL, __ATOMIC_ACQUIRE);
                }
                zero_id = __atomic_load_n(&zero_idx, __ATOMIC_ACQUIRE);
-
-               total += num;
        }
        worker_stats[id].handled_packets += num;
-       count += num;
-       returned = rte_distributor_return_pkt(d, id, buf, num);
 
        if (id == zero_id) {
+               rte_distributor_return_pkt(d, id, NULL, 0);
+
                /* for worker zero, allow it to restart to pick up last packet
                 * when all workers are shutting down.
                 */
@@ -388,14 +372,10 @@ handle_work_for_shutdown_test(void *arg)
 
                while (!quit) {
                        worker_stats[id].handled_packets += num;
-                       count += num;
-                       rte_pktmbuf_free(pkt);
                        num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
                }
-               returned = rte_distributor_return_pkt(d,
-                               id, buf, num);
-               printf("Num returned = %d\n", returned);
        }
+       rte_distributor_return_pkt(d, id, buf, num);
        return 0;
 }
 
@@ -411,7 +391,9 @@ sanity_test_with_worker_shutdown(struct worker_params *wp,
 {
        struct rte_distributor *d = wp->dist;
        struct rte_mbuf *bufs[BURST];
-       unsigned i;
+       struct rte_mbuf *bufs2[BURST];
+       unsigned int i;
+       unsigned int failed = 0;
 
        printf("=== Sanity test of worker shutdown ===\n");
 
@@ -437,16 +419,17 @@ sanity_test_with_worker_shutdown(struct worker_params *wp,
         */
 
        /* get more buffers to queue up, again setting them to the same flow */
-       if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
+       if (rte_mempool_get_bulk(p, (void *)bufs2, BURST) != 0) {
                printf("line %d: Error getting mbufs from pool\n", __LINE__);
+               rte_mempool_put_bulk(p, (void *)bufs, BURST);
                return -1;
        }
        for (i = 0; i < BURST; i++)
-               bufs[i]->hash.usr = 1;
+               bufs2[i]->hash.usr = 1;
 
        /* get worker zero to quit */
        zero_quit = 1;
-       rte_distributor_process(d, bufs, BURST);
+       rte_distributor_process(d, bufs2, BURST);
 
        /* flush the distributor */
        rte_distributor_flush(d);
@@ -460,9 +443,15 @@ sanity_test_with_worker_shutdown(struct worker_params *wp,
                printf("Line %d: Error, not all packets flushed. "
                                "Expected %u, got %u\n",
                                __LINE__, BURST * 2, total_packet_count());
-               return -1;
+               failed = 1;
        }
 
+       rte_mempool_put_bulk(p, (void *)bufs, BURST);
+       rte_mempool_put_bulk(p, (void *)bufs2, BURST);
+
+       if (failed)
+               return -1;
+
        printf("Sanity test with worker shutdown passed\n\n");
        return 0;
 }
@@ -476,7 +465,8 @@ test_flush_with_worker_shutdown(struct worker_params *wp,
 {
        struct rte_distributor *d = wp->dist;
        struct rte_mbuf *bufs[BURST];
-       unsigned i;
+       unsigned int i;
+       unsigned int failed = 0;
 
        printf("=== Test flush fn with worker shutdown (%s) ===\n", wp->name);
 
@@ -513,9 +503,14 @@ test_flush_with_worker_shutdown(struct worker_params *wp,
                printf("Line %d: Error, not all packets flushed. "
                                "Expected %u, got %u\n",
                                __LINE__, BURST, total_packet_count());
-               return -1;
+               failed = 1;
        }
 
+       rte_mempool_put_bulk(p, (void *)bufs, BURST);
+
+       if (failed)
+               return -1;
+
        printf("Flush test with worker shutdown passed\n\n");
        return 0;
 }
@@ -581,7 +576,10 @@ quit_workers(struct worker_params *wp, struct rte_mempool *p)
        const unsigned num_workers = rte_lcore_count() - 1;
        unsigned i;
        struct rte_mbuf *bufs[RTE_MAX_LCORE];
-       rte_mempool_get_bulk(p, (void *)bufs, num_workers);
+       if (rte_mempool_get_bulk(p, (void *)bufs, num_workers) != 0) {
+               printf("line %d: Error getting mbufs from pool\n", __LINE__);
+               return;
+       }
 
        zero_quit = 0;
        quit = 1;
@@ -589,11 +587,12 @@ quit_workers(struct worker_params *wp, struct rte_mempool *p)
                bufs[i]->hash.usr = i << 1;
        rte_distributor_process(d, bufs, num_workers);
 
-       rte_mempool_put_bulk(p, (void *)bufs, num_workers);
-
        rte_distributor_process(d, NULL, 0);
        rte_distributor_flush(d);
        rte_eal_mp_wait_lcore();
+
+       rte_mempool_put_bulk(p, (void *)bufs, num_workers);
+
        quit = 0;
        worker_idx = 0;
        zero_idx = RTE_MAX_LCORE;