Messages in this thread | | | Date | Mon, 25 May 2009 07:22:02 +0200 | From | Eric Dumazet <> | Subject | Re: regression: unregister_netdev() unusably slow |
| |
Benjamin LaHaise a écrit : > On Mon, May 25, 2009 at 12:47:39AM +0200, Eric Dumazet wrote: >> There is a strong dependancy against HZ >> BTW, I am using TREE_RCU > > I'm using CLASSIC_RCU. The bisect just completed, and it points to RCU. > It makes some degree of sense since I'm testing on an otherwise idle > machine. That said, where is fixing it going to make sense? I'm not > opposed to having device unregister take a few timer ticks, but there > has to be some way of exposing parallelism to the system, and since the > synchronize_net() calls are done under rntl_lock(), none is possible at > present. Hrm.
Thanks Ben, this bisection indeed confirms how nasty synchronize_rcu() is :)
Time to include Paul and lkml in the discussion, and find a better solution than one provided in February.
> > -ben > > bf51935f3e988e0ed6f34b55593e5912f990750a is first bad commit > commit bf51935f3e988e0ed6f34b55593e5912f990750a > Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com> > Date: Tue Feb 17 06:01:30 2009 -0800 > > x86, rcu: fix strange load average and ksoftirqd behavior > > Damien Wyart reported high ksoftirqd CPU usage (20%) on an > otherwise idle system. > > The function-graph trace Damien provided: > ... > diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c > > index a546f55..bd4da2a 100644 > --- a/arch/x86/kernel/process_32.c > +++ b/arch/x86/kernel/process_32.c > @@ -104,9 +104,6 @@ void cpu_idle(void) > check_pgt_cache(); > rmb(); > > - if (rcu_pending(cpu)) > - rcu_check_callbacks(cpu, 0); > - > if (cpu_is_offline(cpu)) > play_dead(); > > > --
Paul, this commit makes net device unregister very slow (more than 100 ms if CONFIG_NO_HZ is set), while it used to be pretty fast in previous kernels.
Quoting Ben : " I just ran a few L2TP tests against 2.6.30-rc7, and it looks like network device deletion has become unusably slow. At least in 2.6.27.10, deleting 1000 network interfaces takes less than 2 seconds of real time. The same test run under 2.6.30-rc7 is taking hundreds of seconds to delete 1000 interfaces at a rate of about 5 per second. The interfaces all share the same local ip address, but each have a single route to a unique client ip address."
Device unregister is a synchronize_rcu() abuser (three calls to dismantle a vlan...) so delaying rcu callbacks can be pretty expensive for it.
I wonder if the real root of the problem was not discovered in the meantime, by commit 64ca5ab913f1594ef316556e65f5eae63ff50cee rcu: increment quiescent state counter in ksoftirqd()
Maybe this commit solved Damien Wyart problem as well, and we can revert commit bf51935f3e988e0ed6f34b55593e5912f990750a ?
Thank you
-- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |