[Intel-gfx] [PATCHv3] lib/ratelimit: Lockless ratelimiting

[Steven Rostedt]

On Fri, 20 Jul 2018 16:33:36 +0100
Dmitry Safonov <dima at arista.com> wrote:

> On Fri, 2018-07-20 at 17:09 +0200, Petr Mladek wrote:
> > > > On Tue, 2018-07-03 at 23:56 +0100, Dmitry Safonov wrote:  
> > > > > Currently ratelimit_state is protected with spin_lock. If the
> > > > > .lock
> > > > > is
> > > > > taken at the moment of ___ratelimit() call, the message is
> > > > > suppressed
> > > > > to
> > > > > make ratelimiting robust.
> > > > > 
> > > > > That results in the following issue issue:
> > > > >       CPU0                          CPU1
> > > > > ------------------           ------------------
> > > > > printk_ratelimit()           printk_ratelimit()
> > > > >         |                             |
> > > > >   try_spin_lock()              try_spin_lock()
> > > > >         |                             |
> > > > > time_is_before_jiffies()          return 0; // suppress
> > > > > 
> > > > > So, concurrent call of ___ratelimit() results in silently
> > > > > suppressing
> > > > > one of the messages, regardless if the limit is reached or not.
> > > > > And rc->missed is not increased in such case so the issue is
> > > > > covered
> > > > > from user.
> > > > > 
> > > > > Convert ratelimiting to use atomic counters and drop spin_lock.
> > > > > 
> > > > > Note: That might be unexpected, but with the first interval of
> > > > > messages
> > > > > storm one can print up to burst*2 messages. So, it doesn't
> > > > > guarantee
> > > > > that in *any* interval amount of messages is lesser than burst.
> > > > > But, that differs lightly from previous behavior where one can
> > > > > start
> > > > > burst=5 interval and print 4 messages on the last milliseconds
> > > > > of
> > > > > interval + new 5 messages from new interval (totally 9 messages
> > > > > in
> > > > > lesser than interval value):
> > > > > 
> > > > >    msg0              msg1-msg4 msg0-msg4
> > > > >     |                     |       |
> > > > >     |                     |       |
> > > > >  |--o---------------------o-|-----o--------------------|--> (t)
> > > > >                           <------->
> > > > >                        Lesser than burst
> > > > >   
> > 
> > I am a bit confused. Does this patch fix two problems?
> > 
> >    + Lost rc->missed update when try_spin_lock() fails
> >    + printing up to burst*2 messages in a give interval  
> 
> What I tried to solve here (maybe I could better point it in the commit
> message): is the situation where ratelimit::burst haven't been hit yet
> and there are calls for __ratelimit() from different CPUs.
> So, neither of the messages should be suppressed, but as the spinlock
> is taken already - one of them is dropped and even without updating
> missed counter.
> 
> > It would make the review much easier if you split it into more
> > patches and fix the problems separately.  
> 
> Hmm, not really sure: the patch changes spinlock to atomics and I'm not
> sure it make much sense to add atomics before removing the spinlock..
> 
> > Otherwise, it looks promissing...
> > 
> > Best Regards,
> > Petr
> > 
> > PS: I have vacation the following two weeks. Anyway, please, CC me
> > if you send any new version.  
> 
> Surely, thanks for your attention to the patch (and time).
> 

I'm just catching up from my vacation. What about making rs->missed
into an atomic, and have:

	if (!raw_spin_trylock_irqsave(&rs->lock, flags)) {
		atomic_inc(&rs->missed);
		return 0;
	}

?

You would also need to do:

	if (time_is_before_jiffies(rs->begin + rs->interval)) {
		int missed = atomic_xchg(&rs->missed, 0);
		if (missed) {

So that you don't have a race between checking rs->missed and setting it
to zero.

-- Steve