Re: [RFC-PATCH 1/2] mm: Add __GFP_NO_LOCKS flag

From: Thomas Gleixner <tglx@linutronix.de>
To: paulmck@kernel.org
Cc: Michal Hocko <mhocko@suse.com>,
	Uladzislau Rezki <urezki@gmail.com>,
	Peter Zijlstra <peterz@infradead.org>,
	LKML <linux-kernel@vger.kernel.org>, RCU <rcu@vger.kernel.org>,
	linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
	Vlastimil Babka <vbabka@suse.cz>,
	Matthew Wilcox <willy@infradead.org>,
	"Theodore Y . Ts'o" <tytso@mit.edu>,
	Joel Fernandes <joel@joelfernandes.org>,
	Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
	Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Subject: Re: [RFC-PATCH 1/2] mm: Add __GFP_NO_LOCKS flag
Date: Wed, 19 Aug 2020 01:26:02 +0200	[thread overview]
Message-ID: <87h7szilit.fsf@nanos.tec.linutronix.de> (raw)
In-Reply-To: <20200818171330.GH27891@paulmck-ThinkPad-P72>

Paul,

On Tue, Aug 18 2020 at 10:13, Paul E. McKenney wrote:
> On Tue, Aug 18, 2020 at 06:55:11PM +0200, Thomas Gleixner wrote:
>> On Tue, Aug 18 2020 at 09:13, Paul E. McKenney wrote:
>> > On Tue, Aug 18, 2020 at 04:43:14PM +0200, Thomas Gleixner wrote:
>> >> Throttling the flooder is incresing robustness far more than reducing
>> >> cache misses.
>> >
>> > True, but it takes time to identify a flooding event that needs to be
>> > throttled (as in milliseconds).  This time cannot be made up.
>> 
>> Not really. A flooding event will deplete your preallocated pages very
>> fast, so you have to go into the allocator and get new ones which
>> naturally throttles the offender.
>
> Should it turn out that we can in fact go into the allocator, completely
> agreed.

You better can for any user space controllable flooding source.

>> So if your open/close thing uses the new single argument free which has
>> to be called from sleepable context then the allocation either gives you
>> a page or that thing has to wait. No fancy extras.
>
> In the single-argument kvfree_rcu() case, completely agreed.
>
>> You still can have a page reserved for your other regular things and
>> once that it gone, you have to fall back to the linked list for
>> those. But when that happens the extra cache misses are not your main
>> problem anymore.
>
> The extra cache misses are a problem in that case because they throttle
> the reclamation, which anti-throttles the producer, especially in the
> case where callback invocation is offloaded.

You still did not explain which contexts can create flooding. I gave you
a complete list a few mails ago, but you still did not tell which of the
contexts can cause flooding.

If it's any context which is not sleepable or controllable in any way,
then any attempt to mitigate it is a lost battle:

  A dependency on allocating memory to free memory is a dead end by
  definition.

Any flooder which is uncontrollable is a bug and no matter what kind of
hacks you provide, it will be able to bring the whole thing down.

So far this looks like you're trying to cure the symptoms, which is
wrong to begin with.

If the flooder is controllable then there is no problem with cache
misses at all unless the RCU free callbacks are not able to catch up
which is yet another problem which you can't cure by allocating more
memory.

Thanks,

        tglx