OK, so when we have memory pressure we can call
percpu_counter_read_positive() with SOFTIRQ enabled, and lockdep
complains...

This bug was probably added in 2008, in commit 1748376b6626a
(net: Use a percpu_counter for sockets_allocated)

I'll have to backport the following patch, can you test it please ?

Thanks !

 net/core/sock.c |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/net/core/sock.c b/net/core/sock.c
index b2e14c0..08fc929 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c

@@ -1818,7 +1818,9 @@ int __sk_mem_schedule(struct sock *sk, int

size, int kind)

 		if (!sk_under_memory_pressure(sk))
 			return 1;
+		local_bh_disable();
 		alloc = sk_sockets_allocated_read_positive(sk);
+		local_bh_enable();
 		if (sk_prot_mem_limits(sk, 2) > alloc *
 		    sk_mem_pages(sk->sk_wmem_queued +
 				 atomic_read(&sk->sk_rmem_alloc) +