On Sun, 11 Aug 2024 21:29:57 +0530
Rakesh Kudurumalla <rkuduruma...@marvell.com> wrote:
> Race condition between jobstats and time metrics
> for forwarding and flushing is maintained using spinlock.
> Timer metrics are not displayed properly due to the
> frequent unavailability of the lock.This patch fixes
> the issue by introducing a delay before acquiring
> the lock in the loop. This delay allows for betteravailability
> of the lock, ensuring that show_lcore_stats() can
> periodically update the statistics even when forwarding
> jobs are running.
>
> Fixes: 204896f8d66c ("examples/l2fwd-jobstats: add new example")
> Cc: sta...@dpdk.org
>
> Signed-off-by: Rakesh Kudurumalla <rkuduruma...@marvell.com>
The original code is a mess here.
The whole idle job loop here is the problem.
It should use rte_timer_next_ticks() to know when next timer is
about to expire and use that.
And instead of fighting with spin lock, use ticket lock which will
cause in-order waiting.
Something like the following (untested):
diff --git a/examples/l2fwd-jobstats/main.c b/examples/l2fwd-jobstats/main.c
index 308b8edd20..9586d90ab6 100644
--- a/examples/l2fwd-jobstats/main.c
+++ b/examples/l2fwd-jobstats/main.c
@@ -27,7 +27,7 @@
#include <rte_ethdev.h>
#include <rte_mempool.h>
#include <rte_mbuf.h>
-#include <rte_spinlock.h>
+#include <rte_ticketlock.h>
#include <rte_errno.h>
#include <rte_jobstats.h>
@@ -80,8 +80,7 @@ struct __rte_cache_aligned lcore_queue_conf {
struct rte_jobstats idle_job;
struct rte_jobstats_context jobs_context;
- RTE_ATOMIC(uint16_t) stats_read_pending;
- rte_spinlock_t lock;
+ rte_ticketlock_t lock;
};
/* >8 End of list of queues to be polled for given lcore. */
struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
@@ -151,9 +150,7 @@ show_lcore_stats(unsigned lcore_id)
uint64_t collection_time = rte_get_timer_cycles();
/* Ask forwarding thread to give us stats. */
- rte_atomic_store_explicit(&qconf->stats_read_pending, 1,
rte_memory_order_relaxed);
- rte_spinlock_lock(&qconf->lock);
- rte_atomic_store_explicit(&qconf->stats_read_pending, 0,
rte_memory_order_relaxed);
+ rte_ticketlock_lock(&qconf->lock);
/* Collect context statistics. */
stats_period = ctx->state_time - ctx->start_time;
@@ -195,7 +192,7 @@ show_lcore_stats(unsigned lcore_id)
idle_exec_max = qconf->idle_job.max_exec_time;
rte_jobstats_reset(&qconf->idle_job);
- rte_spinlock_unlock(&qconf->lock);
+ rte_ticketlock_unlock(&qconf->lock);
exec -= idle_exec;
busy = exec + management;
@@ -478,11 +475,11 @@ l2fwd_main_loop(void)
unsigned lcore_id;
unsigned i, portid;
struct lcore_queue_conf *qconf;
- uint8_t stats_read_pending = 0;
- uint8_t need_manage;
+ uint64_t hz;
lcore_id = rte_lcore_id();
qconf = &lcore_queue_conf[lcore_id];
+ hz = rte_get_timer_hz();
if (qconf->n_rx_port == 0) {
RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
@@ -502,47 +499,22 @@ l2fwd_main_loop(void)
/* Minimize impact of stats reading. 8< */
for (;;) {
- rte_spinlock_lock(&qconf->lock);
-
- do {
- rte_jobstats_context_start(&qconf->jobs_context);
-
- /* Do the Idle job:
- * - Read stats_read_pending flag
- * - check if some real job need to be executed
- */
- rte_jobstats_start(&qconf->jobs_context,
&qconf->idle_job);
-
- uint64_t repeats = 0;
-
- do {
- uint8_t i;
- uint64_t now = rte_get_timer_cycles();
-
- repeats++;
- need_manage = qconf->flush_timer.expire < now;
- /* Check if we was esked to give a stats. */
- stats_read_pending = rte_atomic_load_explicit(
- &qconf->stats_read_pending,
- rte_memory_order_relaxed);
- need_manage |= stats_read_pending;
- for (i = 0; i < qconf->n_rx_port &&
!need_manage; i++)
- need_manage =
qconf->rx_timers[i].expire < now;
+ rte_ticketlock_lock(&qconf->lock);
- } while (!need_manage);
+ rte_jobstats_context_start(&qconf->jobs_context);
+ rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
+ rte_timer_manage();
+ rte_jobstats_context_finish(&qconf->jobs_context);
- if (likely(repeats != 1))
- rte_jobstats_finish(&qconf->idle_job,
qconf->idle_job.target);
- else
- rte_jobstats_abort(&qconf->idle_job);
+ int64_t next_ticks = rte_timer_next_ticks();
- rte_timer_manage();
- rte_jobstats_context_finish(&qconf->jobs_context);
- } while (likely(stats_read_pending == 0));
+ rte_ticketlock_unlock(&qconf->lock);
- rte_spinlock_unlock(&qconf->lock);
- rte_pause();
+ if (next_ticks > 0)
+ rte_delay_us((1000000 * next_ticks) / hz);
+ else
+ rte_pause();
}
/* >8 End of minimize impact of stats reading. */
}
@@ -972,7 +944,7 @@ main(int argc, char **argv)
RTE_LCORE_FOREACH(lcore_id) {
qconf = &lcore_queue_conf[lcore_id];
- rte_spinlock_init(&qconf->lock);
+ rte_ticketlock_init(&qconf->lock);
if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
rte_panic("Jobs stats context for core %u init
failed\n", lcore_id);
