| Message ID | 1452181973.8255.213.camel@edumazet-glaptop2.roam.corp.google.com |
|---|---|
| State | RFC, archived |
| Delegated to: | David Miller |
| Headers | show |
On Thu, Jan 07, 2016 at 07:52:53AM -0800, Eric Dumazet wrote: > On Thu, 2016-01-07 at 10:32 -0500, Dave Jones wrote: > > =============================== > > [ INFO: suspicious RCU usage. ] > > 4.4.0-rc8-firewall+ #1 Not tainted > > ------------------------------- > > net/ipv6/tcp_ipv6.c:465 suspicious rcu_dereference_check() usage! > > > > other info that might help us debug this: > > > > > > rcu_scheduler_active = 1, debug_locks = 1 > > 1 lock held by swapper/1/0: > > #0: (((&req->rsk_timer))){+.-...}, at: [<ffffffff99113285>] call_timer_fn+0x5/0x3f0 > > > > stack backtrace: > > CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.4.0-rc8-firewall+ #1 > > 0000000000000000 ffff8801d7a07b28 ffffffff9948b3b5 ffff8801d6046500 > > ffff8801d7a07b58 ffffffff990e9b7a ffff8801cd356240 0000000000000000 > > ffff8801d23b1698 ffff8801d23b0c40 ffff8801d7a07ba8 ffffffff99b635d2 > > Call Trace: > > <IRQ> [<ffffffff9948b3b5>] dump_stack+0x4e/0x79 > > [<ffffffff990e9b7a>] lockdep_rcu_suspicious+0xea/0x110 > > [<ffffffff99b635d2>] tcp_v6_send_synack+0x2c2/0x350 > > [<ffffffff99a8550d>] tcp_rtx_synack+0xdd/0x180 > > [<ffffffff99a85430>] ? tcp_send_probe0+0x1a0/0x1a0 > > [<ffffffff99a5f0a4>] reqsk_timer_handler+0x4c4/0x530 > > [<ffffffff99a5ebe0>] ? inet_csk_reqsk_queue_drop+0x3a0/0x3a0 > > [<ffffffff990e0f2b>] ? __lock_is_held+0x9b/0xd0 > > [<ffffffff991133b2>] call_timer_fn+0x132/0x3f0 > > [<ffffffff99113285>] ? call_timer_fn+0x5/0x3f0 > > [<ffffffff99a5ebe0>] ? inet_csk_reqsk_queue_drop+0x3a0/0x3a0 > > [<ffffffff99113280>] ? process_timeout+0x10/0x10 > > [<ffffffff990e44d2>] ? trace_hardirqs_on_caller+0x192/0x2a0 > > [<ffffffff990b52ea>] ? preempt_count_sub+0x1a/0x130 > > [<ffffffff99113cab>] run_timer_softirq+0x47b/0x590 > > [<ffffffff99a5ebe0>] ? inet_csk_reqsk_queue_drop+0x3a0/0x3a0 > > [<ffffffff99113830>] ? internal_add_timer+0x110/0x110 > > [<ffffffff990e0eb8>] ? __lock_is_held+0x28/0xd0 > > [<ffffffff99076fd2>] __do_softirq+0x1b2/0x5c0 > > [<ffffffff9907762c>] irq_exit+0xfc/0x110 > > [<ffffffff99c12abf>] smp_apic_timer_interrupt+0x5f/0x70 > > [<ffffffff99c112bb>] apic_timer_interrupt+0x8b/0x90 > > <EOI> [<ffffffff998dd547>] ? cpuidle_enter_state+0x1c7/0x460 > > [<ffffffff998dd542>] ? cpuidle_enter_state+0x1c2/0x460 > > [<ffffffff99107289>] ? rcu_eqs_enter_common+0x139/0x280 > > [<ffffffff998dd847>] cpuidle_enter+0x17/0x20 > > [<ffffffff990dbd92>] cpu_startup_entry+0x4d2/0x5b0 > > [<ffffffff990db8c0>] ? default_idle_call+0x60/0x60 > > [<ffffffff9912d714>] ? clockevents_config_and_register+0x64/0x70 > > [<ffffffff9904d785>] ? setup_APIC_timer+0x115/0x120 > > [<ffffffff9904be1a>] start_secondary+0x23a/0x2a0 > > [<ffffffff9904bbe0>] ? set_cpu_sibling_map+0x9c0/0x9c0 > > > > Apparently RCU lockdep thinks a softirq handler (timer soft irq) needs > rcu_read_lock() before rcu_dereference() > > This is not clear if this is a lockdep false positive. > > Paul, can you remind me why it is needed, as a softirq handler is not > allowed to schedule or be preempted ? Hello, Eric! If this were rcu_dereference_bh(), then you would be OK as is, given that you are in a softirq handler. But for rcu_dereference(), lockdep does indeed insist on an rcu_read_lock(). Yes, you would in fact be OK with the current implementation (I think, anyway), even with preemptible RCU, but that is an accident of implementation. Is the required rcu_read_lock() and rcu_read_unlock() resulting in a performance problem? Thanx, Paul > diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c > index 6b8a8a9091fa..bd100b47c717 100644 > --- a/net/ipv6/tcp_ipv6.c > +++ b/net/ipv6/tcp_ipv6.c > @@ -462,8 +462,10 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, > if (np->repflow && ireq->pktopts) > fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts)); > > + rcu_read_lock(); > err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt), > np->tclass); > + rcu_read_unlock(); > err = net_xmit_eval(err); > } > > > -- 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
On Thu, 2016-01-07 at 08:20 -0800, Paul E. McKenney wrote: > This is not clear if this is a lockdep false positive. > > > > Paul, can you remind me why it is needed, as a softirq handler is not > > allowed to schedule or be preempted ? > > Hello, Eric! > > If this were rcu_dereference_bh(), then you would be OK as is, given that > you are in a softirq handler. But for rcu_dereference(), lockdep does > indeed insist on an rcu_read_lock(). Yes, you would in fact be OK with > the current implementation (I think, anyway), even with preemptible RCU, > but that is an accident of implementation. > > Is the required rcu_read_lock() and rcu_read_unlock() resulting in a > performance problem? No performance problem. This comes from my 45f6fad84cc305103b28d73482b344d7f5b76f39 commit ("ipv6: add complete rcu protection around np->opt") I added the rcu_read_lock()/unlock() sections where I thought they were needed, and when I considered tcp_v6_send_synack() case, my reasoning was that we were holding rcu_read_lock() in the normal non retransmit case, since the SYN packet is processed under rcu_read_lock() protection, and wrongly assumed the timer irq was also holding rcu_read_lock() Also my RCU lockdep enabled tests did not trigger the warning seen by Dave Jones.... Strange... I will submit a formal patch. Thanks ! -- 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/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c index 6b8a8a9091fa..bd100b47c717 100644 --- a/net/ipv6/tcp_ipv6.c +++ b/net/ipv6/tcp_ipv6.c @@ -462,8 +462,10 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, if (np->repflow && ireq->pktopts) fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts)); + rcu_read_lock(); err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt), np->tclass); + rcu_read_unlock(); err = net_xmit_eval(err); }