DPDK patches and discussions
 help / color / mirror / Atom feed
* [dpdk-dev] [PATCH 1/1] timer: fix race condition
@ 2018-11-29 18:53 Erik Gabriel Carrillo
  2018-12-19  3:36 ` Thomas Monjalon
  2018-12-19 16:09 ` [dpdk-dev] [PATCH v2] " Erik Gabriel Carrillo
  0 siblings, 2 replies; 6+ messages in thread
From: Erik Gabriel Carrillo @ 2018-11-29 18:53 UTC (permalink / raw)
  To: rsanford; +Cc: dev

rte_timer_manage() adds expired timers to a "run list", and walks the
list, transitioning each timer from the PENDING to the RUNNING state.
If another lcore resets or stops the timer at precisely this
moment, the timer state would instead be set to CONFIG by that other
lcore, which would cause timer_manage() to skip over it. This is
expected behavior.

However, if a timer expires quickly enough, there exists the
following race condition that causes the timer_manage() routine to
misinterpret a timer in CONFIG state, resulting in lost timers:

- Thread A:
  - starts a timer with rte_timer_reset()
  - the timer is moved to CONFIG state
  - the spinlock associated with the appropriate skiplist is acquired
  - timer is inserted into the skiplist
  - the spinlock is released
- Thread B:
  - executes rte_timer_manage()
  - find above timer as expired, add it to run list
  - walk run list, see above timer still in CONFIG state, unlink it from
    run list and continue on
- Thread A:
  - move timer to PENDING state
  - return from rte_timer_reset()
  - timer is now in PENDING state, but not actually linked into skiplist
    and will never get processed further by rte_timer_manage()

This commit fixes this race condition by only releasing the spinlock
after the timer state has been transitioned from CONFIG to PENDING,
which prevents rte_timer_manage() from seeing an incorrect state.

Fixes: 9b15ba895b9f ("timer: use a skip list")
Signed-off-by: Erik Gabriel Carrillo <erik.g.carrillo@intel.com>
---
 lib/librte_timer/rte_timer.c | 28 ++++++++++++++--------------
 1 file changed, 14 insertions(+), 14 deletions(-)

diff --git a/lib/librte_timer/rte_timer.c b/lib/librte_timer/rte_timer.c
index 590488c..30c7b0a 100644
--- a/lib/librte_timer/rte_timer.c
+++ b/lib/librte_timer/rte_timer.c
@@ -241,24 +241,17 @@ timer_get_prev_entries_for_node(struct rte_timer *tim, unsigned tim_lcore,
 	}
 }
 
-/*
- * add in list, lock if needed
+/* call with lock held as necessary
+ * add in list
  * timer must be in config state
  * timer must not be in a list
  */
 static void
-timer_add(struct rte_timer *tim, unsigned tim_lcore, int local_is_locked)
+timer_add(struct rte_timer *tim, unsigned int tim_lcore)
 {
-	unsigned lcore_id = rte_lcore_id();
 	unsigned lvl;
 	struct rte_timer *prev[MAX_SKIPLIST_DEPTH+1];
 
-	/* if timer needs to be scheduled on another core, we need to
-	 * lock the list; if it is on local core, we need to lock if
-	 * we are not called from rte_timer_manage() */
-	if (tim_lcore != lcore_id || !local_is_locked)
-		rte_spinlock_lock(&priv_timer[tim_lcore].list_lock);
-
 	/* find where exactly this element goes in the list of elements
 	 * for each depth. */
 	timer_get_prev_entries(tim->expire, tim_lcore, prev);
@@ -282,9 +275,6 @@ timer_add(struct rte_timer *tim, unsigned tim_lcore, int local_is_locked)
 	 * NOTE: this is not atomic on 32-bit*/
 	priv_timer[tim_lcore].pending_head.expire = priv_timer[tim_lcore].\
 			pending_head.sl_next[0]->expire;
-
-	if (tim_lcore != lcore_id || !local_is_locked)
-		rte_spinlock_unlock(&priv_timer[tim_lcore].list_lock);
 }
 
 /*
@@ -379,8 +369,15 @@ __rte_timer_reset(struct rte_timer *tim, uint64_t expire,
 	tim->f = fct;
 	tim->arg = arg;
 
+	/* if timer needs to be scheduled on another core, we need to
+	 * lock the destination list; if it is on local core, we need to lock if
+	 * we are not called from rte_timer_manage()
+	 */
+	if (tim_lcore != lcore_id || !local_is_locked)
+		rte_spinlock_lock(&priv_timer[tim_lcore].list_lock);
+
 	__TIMER_STAT_ADD(pending, 1);
-	timer_add(tim, tim_lcore, local_is_locked);
+	timer_add(tim, tim_lcore);
 
 	/* update state: as we are in CONFIG state, only us can modify
 	 * the state so we don't need to use cmpset() here */
@@ -389,6 +386,9 @@ __rte_timer_reset(struct rte_timer *tim, uint64_t expire,
 	status.owner = (int16_t)tim_lcore;
 	tim->status.u32 = status.u32;
 
+	if (tim_lcore != lcore_id || !local_is_locked)
+		rte_spinlock_unlock(&priv_timer[tim_lcore].list_lock);
+
 	return 0;
 }
 
-- 
2.6.4

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2018-12-19 18:54 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-11-29 18:53 [dpdk-dev] [PATCH 1/1] timer: fix race condition Erik Gabriel Carrillo
2018-12-19  3:36 ` Thomas Monjalon
2018-12-19  7:57   ` Gavin Hu (Arm Technology China)
2018-12-19 16:11     ` Carrillo, Erik G
2018-12-19 16:09 ` [dpdk-dev] [PATCH v2] " Erik Gabriel Carrillo
2018-12-19 18:54   ` Thomas Monjalon

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).