On Mon, 2016-09-12 at 19:02 -0400, Jamal Hadi Salim wrote:
> On 16-09-12 06:26 PM, Eric Dumazet wrote:
> > On Mon, 2016-09-12 at 18:14 -0400, Jamal Hadi Salim wrote:
> >
> >> I noticed some very weird issues when I took that out.
> >> Running sufficiently large amount of traffic (ping -f is sufficient)
> >> I saw that when i did a dump it took anywhere between 6-15 seconds.
> >> With the read_lock in place response was immediate.
> >> I can go back and run things to verify - but it was very odd.
> >
> > This was on uni processor ?
> >
>
> It was a VM.
>
> > Looks like typical starvation caused by aggressive softirq.
> >
>
> Well, then it is strange that in one case a tc dump of the rule
> was immediate and in the other case it was consistent for 5-15
> seconds.
>
This needs investigation ;)
One possible loop under high stress would be possible in
__gnet_stats_copy_basic(), since we might restart the loop if we are
really really unlucky, but this would have nothing with your patches.
diff --git a/net/core/gen_stats.c b/net/core/gen_stats.c
index
508e051304fb62627e61b5065b2325edd1b84f2e..dc9dd8ae7d5405f76c775278dac7689655b21041
100644
--- a/net/core/gen_stats.c
+++ b/net/core/gen_stats.c
@@ -142,10 +142,14 @@ __gnet_stats_copy_basic(const seqcount_t *running,
return;
}
do {
- if (running)
+ if (running) {
+ local_bh_disable();
seq = read_seqcount_begin(running);
+ }
bstats->bytes = b->bytes;
bstats->packets = b->packets;
+ if (running)
+ local_bh_enable();
} while (running && read_seqcount_retry(running, seq));
}
EXPORT_SYMBOL(__gnet_stats_copy_basic);
