On Fri, 2021-09-24 at 07:22 +0200, Mike Galbraith wrote:
> Greetings,
>
> I met this lockdep_assert_preemption_disabled() gripe when starting
> ltp zram tests.
Before actually, I just wasn't paying attention.
<snip gripe>
net: bridge: mcast: Disable preemption in br_multicast_update_querier() for RT
write_seqcount_begin() in br_multicast_update_querier() complains about
preemption not being disabled during RT softirq processesing. Since the
read side spins, disable preemption across the tiny write side for RT.
Signed-off-by: Mike Galbraith <efault@xxxxxx>
---
net/bridge/br_multicast.c | 4 ++++
1 file changed, 4 insertions(+)
--- a/net/bridge/br_multicast.c
+++ b/net/bridge/br_multicast.c
@@ -1679,10 +1679,14 @@ static void br_multicast_update_querier(
{
lockdep_assert_held_once(&brmctx->br->multicast_lock);
+ if (IS_ENABLED(CONFIG_PREEMPT_RT))
+ preempt_disable();
write_seqcount_begin(&querier->seq);
querier->port_ifidx = ifindex;
memcpy(&querier->addr, saddr, sizeof(*saddr));
write_seqcount_end(&querier->seq);
+ if (IS_ENABLED(CONFIG_PREEMPT_RT))
+ preempt_enable();
}
static void br_multicast_send_query(struct net_bridge_mcast *brmctx,
