| Message ID | ffd56913c060fc717a74b87ba1a154125c9db2c4.1407796232.git.mleitner@redhat.com |
|---|---|
| State | Not Applicable, archived |
| Delegated to: | David Miller |
| Headers | show |
On Mo, 2014-08-11 at 19:41 -0300, Marcelo Ricardo Leitner wrote: > When rt_intern_hash() has to deal with neighbour cache overflowing, > it triggers the route cache garbage collector in an attempt to free > some references on neighbour entries. > > Such call cannot be done async but should also not run in parallel with > an already-running one, so that they don't collapse fighting over the > hash lock entries. > > This patch thus blocks parallel executions by ignoring the call if > garbage collector is already running on another CPU. > > We don't use a spinlock for it because now it runs on a work queue and > we want it to be schedulable in-between the hash indexes. > > Signed-off-by: Marcelo Ricardo Leitner <mleitner@redhat.com> > Cc: Hannes Frederic Sowa <hannes@redhat.com> Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org> -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 2ad1cdb2c9e45f5ec5a6c1fa9b781657b1c292ec..68153bb86b62b500febd2a134f71348fa78295d9 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -986,6 +986,7 @@ static void __do_rt_garbage_collect(int elasticity, int min_interval) static unsigned long last_gc; static int rover; static int equilibrium; + static int rt_gc_barrier; struct rtable *rth; struct rtable __rcu **rthp; unsigned long now = jiffies; @@ -999,6 +1000,12 @@ static void __do_rt_garbage_collect(int elasticity, int min_interval) RT_CACHE_STAT_INC(gc_total); + if (cmpxchg(&rt_gc_barrier, 0, 1) != 0) { + /* busy */ + RT_CACHE_STAT_INC(gc_ignored); + return; + } + if (now - last_gc < min_interval && entries < ip_rt_max_size) { RT_CACHE_STAT_INC(gc_ignored); @@ -1089,6 +1096,7 @@ static void __do_rt_garbage_collect(int elasticity, int min_interval) if (net_ratelimit()) pr_warn("dst cache overflow\n"); RT_CACHE_STAT_INC(gc_dst_overflow); + rt_gc_barrier = 0; return; work_done: @@ -1097,7 +1105,9 @@ work_done: dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh || dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh) expire = ip_rt_gc_timeout; -out: return; +out: + rt_gc_barrier = 0; + return; } static void __rt_garbage_collect(struct work_struct *w)
When rt_intern_hash() has to deal with neighbour cache overflowing, it triggers the route cache garbage collector in an attempt to free some references on neighbour entries. Such call cannot be done async but should also not run in parallel with an already-running one, so that they don't collapse fighting over the hash lock entries. This patch thus blocks parallel executions by ignoring the call if garbage collector is already running on another CPU. We don't use a spinlock for it because now it runs on a work queue and we want it to be schedulable in-between the hash indexes. Signed-off-by: Marcelo Ricardo Leitner <mleitner@redhat.com> Cc: Hannes Frederic Sowa <hannes@redhat.com> --- net/ipv4/route.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)