* [dpdk-stable] [RFC PATCH v1 1/6] app/eventdev: fix SMP barrier bugs for perf test
[not found] <20201222050728.41000-1-feifei.wang2@arm.com>
@ 2020-12-22 5:07 ` Feifei Wang
2020-12-22 5:07 ` [dpdk-stable] [RFC PATCH v1 4/6] app/eventdev: add release barriers for pipeline test Feifei Wang
1 sibling, 0 replies; 3+ messages in thread
From: Feifei Wang @ 2020-12-22 5:07 UTC (permalink / raw)
To: Jerin Jacob, Harry van Haaren
Cc: dev, nd, Honnappa.Nagarahalli, Feifei Wang, stable
This patch fixes RTE SMP barrier bugs for the perf test of eventdev.
For the "perf_process_last_stage" function, wmb after storing
processed_pkts should be moved before it. This is because the worker
lcore should ensure it has really finished data processing, e.g. event
stored into buffers, before the shared variables "w->processed_pkts"are
stored.
For the "perf_process_last_stage_latency", on the one hand, the wmb
should be moved before storing into "w->processed_pkts". The reason is
the same as above. But on the other hand, for "w->latency", wmb is
unnecessary due to data dependency.
Fixes: 2369f73329f8 ("app/testeventdev: add perf queue worker functions")
Cc: jerinj@marvell.com
Cc: stable@dpdk.org
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
---
app/test-eventdev/test_perf_common.h | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
diff --git a/app/test-eventdev/test_perf_common.h b/app/test-eventdev/test_perf_common.h
index ff9705df8..e7233e5a5 100644
--- a/app/test-eventdev/test_perf_common.h
+++ b/app/test-eventdev/test_perf_common.h
@@ -97,8 +97,13 @@ perf_process_last_stage(struct rte_mempool *const pool,
void *bufs[], int const buf_sz, uint8_t count)
{
bufs[count++] = ev->event_ptr;
- w->processed_pkts++;
+
+ /* wmb here ensures event_prt is stored before
+ * updating the number of processed packets
+ * for worker lcores
+ */
rte_smp_wmb();
+ w->processed_pkts++;
if (unlikely(count == buf_sz)) {
count = 0;
@@ -116,6 +121,12 @@ perf_process_last_stage_latency(struct rte_mempool *const pool,
struct perf_elt *const m = ev->event_ptr;
bufs[count++] = ev->event_ptr;
+
+ /* wmb here ensures event_prt is stored before
+ * updating the number of processed packets
+ * for worker lcores
+ */
+ rte_smp_wmb();
w->processed_pkts++;
if (unlikely(count == buf_sz)) {
@@ -127,7 +138,6 @@ perf_process_last_stage_latency(struct rte_mempool *const pool,
}
w->latency += latency;
- rte_smp_wmb();
return count;
}
--
2.17.1
^ permalink raw reply [flat|nested] 3+ messages in thread* [dpdk-stable] [RFC PATCH v1 4/6] app/eventdev: add release barriers for pipeline test
[not found] <20201222050728.41000-1-feifei.wang2@arm.com>
2020-12-22 5:07 ` [dpdk-stable] [RFC PATCH v1 1/6] app/eventdev: fix SMP barrier bugs for perf test Feifei Wang
@ 2020-12-22 5:07 ` Feifei Wang
1 sibling, 0 replies; 3+ messages in thread
From: Feifei Wang @ 2020-12-22 5:07 UTC (permalink / raw)
To: Jerin Jacob, Harry van Haaren, Pavan Nikhilesh
Cc: dev, nd, Honnappa.Nagarahalli, Feifei Wang, pbhagavatula, stable,
Phil Yang
Add release barriers before updating the processed packets for worker
lcores to ensure the worker lcore has really finished data processing
and then it can update the processed packets number.
Fixes: 314bcf58ca8f ("app/eventdev: add pipeline queue worker functions")
Cc: pbhagavatula@marvell.com
Cc: stable@dpdk.org
Signed-off-by: Phil Yang <phil.yang@arm.com>
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
---
app/test-eventdev/test_pipeline_queue.c | 64 +++++++++++++++++++++----
1 file changed, 56 insertions(+), 8 deletions(-)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 7bebac34f..0c0ec0ceb 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -30,7 +30,13 @@ pipeline_queue_worker_single_stage_tx(void *arg)
if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
pipeline_event_tx(dev, port, &ev);
- w->processed_pkts++;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), 1,
+ __ATOMIC_RELEASE);
} else {
ev.queue_id++;
pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
@@ -59,7 +65,13 @@ pipeline_queue_worker_single_stage_fwd(void *arg)
rte_event_eth_tx_adapter_txq_set(ev.mbuf, 0);
pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
pipeline_event_enqueue(dev, port, &ev);
- w->processed_pkts++;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), 1,
+ __ATOMIC_RELEASE);
}
return 0;
@@ -84,7 +96,13 @@ pipeline_queue_worker_single_stage_burst_tx(void *arg)
if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
pipeline_event_tx(dev, port, &ev[i]);
ev[i].op = RTE_EVENT_OP_RELEASE;
- w->processed_pkts++;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), 1,
+ __ATOMIC_RELEASE);
} else {
ev[i].queue_id++;
pipeline_fwd_event(&ev[i],
@@ -121,7 +139,13 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
}
pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
- w->processed_pkts += nb_rx;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), nb_rx,
+ __ATOMIC_RELEASE);
}
return 0;
@@ -146,7 +170,13 @@ pipeline_queue_worker_multi_stage_tx(void *arg)
if (ev.queue_id == tx_queue[ev.mbuf->port]) {
pipeline_event_tx(dev, port, &ev);
- w->processed_pkts++;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), 1,
+ __ATOMIC_RELEASE);
continue;
}
@@ -180,7 +210,13 @@ pipeline_queue_worker_multi_stage_fwd(void *arg)
ev.queue_id = tx_queue[ev.mbuf->port];
rte_event_eth_tx_adapter_txq_set(ev.mbuf, 0);
pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
- w->processed_pkts++;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), 1,
+ __ATOMIC_RELEASE);
} else {
ev.queue_id++;
pipeline_fwd_event(&ev, sched_type_list[cq_id]);
@@ -214,7 +250,13 @@ pipeline_queue_worker_multi_stage_burst_tx(void *arg)
if (ev[i].queue_id == tx_queue[ev[i].mbuf->port]) {
pipeline_event_tx(dev, port, &ev[i]);
ev[i].op = RTE_EVENT_OP_RELEASE;
- w->processed_pkts++;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), 1,
+ __ATOMIC_RELEASE);
continue;
}
@@ -254,7 +296,13 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
rte_event_eth_tx_adapter_txq_set(ev[i].mbuf, 0);
pipeline_fwd_event(&ev[i],
RTE_SCHED_TYPE_ATOMIC);
- w->processed_pkts++;
+
+ /* release barrier here ensures stored operation
+ * of the event completes before the number of
+ * processed pkts is visible to the main core
+ */
+ __atomic_fetch_add(&(w->processed_pkts), 1,
+ __ATOMIC_RELEASE);
} else {
ev[i].queue_id++;
pipeline_fwd_event(&ev[i],
--
2.17.1
^ permalink raw reply [flat|nested] 3+ messages in thread