2013-05-02 23:13:22

by Thomas Pedersen

Subject: [PATCH] mac80211: clean up mesh RCU warning

When running a kernel with CONFIG_PROVE_RCU, the minstrel
rate update code would complain about being called without
the RCU read lock held. This is because
mesh_sta_info_init() may or may not have been called under
RCU read lock. Perform some bureaucratic critical section
toggling to always call mesh_sta_info_init() under RCU,
and silence this warning.

Signed-off-by: Thomas Pedersen <[email protected]>
---
net/mac80211/mesh_plink.c | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/net/mac80211/mesh_plink.c b/net/mac80211/mesh_plink.c
index 09bebed..4058bd7 100644
--- a/net/mac80211/mesh_plink.c
+++ b/net/mac80211/mesh_plink.c
@@ -473,13 +473,15 @@ mesh_sta_info_get(struct ieee80211_sub_if_data *sdata,
rcu_read_unlock();
/* can't run atomic */
sta = mesh_sta_info_alloc(sdata, addr, elems);
- if (!sta) {
- rcu_read_lock();
+
+ rcu_read_lock();
+ if (!sta)
return NULL;
- }

mesh_sta_info_init(sdata, sta, elems, true);

+ rcu_read_unlock();
+ /* also might sleep */
if (sta_info_insert_rcu(sta))
return NULL;
}
--
1.7.10.4

2013-05-03 06:41:07

by Johannes Berg

Subject: Re: [PATCH] mac80211: clean up mesh RCU warning

On Thu, 2013-05-02 at 16:11 -0700, Thomas Pedersen wrote:
> When running a kernel with CONFIG_PROVE_RCU, the minstrel
> rate update code would complain about being called without
> the RCU read lock held. This is because
> mesh_sta_info_init() may or may not have been called under
> RCU read lock. Perform some bureaucratic critical section
> toggling to always call mesh_sta_info_init() under RCU,
> and silence this warning.

This was discussed in another thread a while ago, but unfortunately
nothing came out of it. I really don't think this is the right solution,
since the code that this calls doesn't even use rcu_dereference()
properly.

Felix, please come up with a proper fix soon, otherwise I'm going to
revert your changes for 3.10.

johannes