linux-kernel - Re: RCU tests for Maple Tree

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Mon, 17 May 2021 09:53:10 -0700
From:   "Paul E. McKenney" <paulmck@...nel.org>
To:     Liam Howlett <liam.howlett@...cle.com>
Cc:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>
Subject: Re: RCU tests for Maple Tree

On Mon, May 17, 2021 at 09:50:04AM -0700, Paul E. McKenney wrote:
> On Mon, May 17, 2021 at 04:30:53PM +0000, Liam Howlett wrote:
> > * Paul E. McKenney <paulmck@...nel.org> [210517 11:40]:
> > > Hello, Liam!
> > > 
> > > Apologies for my being so slow here, but just wanted to double-check my
> > > understanding of this code.
> > > 
> > > There appear to be two tests that execute from run_check_rcu():
> > > 
> > > o	rcu_loop().  This appears to have RCU readers scanning the tree
> > > 	while an updater is adding a single range.  (Or replacing it,
> > > 	as the case might be.)
> > > 
> > > o	rcu_val().  This appears to have RCU readers repeatedly reading a
> > > 	given value while an updater is adding/replacing a single range.
> > > 	The test complains if no one sees the new value.
> > > 
> > > These tests appear to be the only use of threads, though perhaps the
> > > test harness has some way of creating threads that I missed.
> > > 
> > > Are there other tests that I should be looking for?
> > 
> > No, those are the only ones I'm running with threads right now.  I think
> > all RCU tests are run from check_rcu() iirc.  This did yield results of
> > failures that had to be addressed so I'm somewhat confident that it's
> > actually working.
> 
> OK, I guess I can feel relieved that I can still read code.  ;-)

Oh, and I should hasten to add that for a data structure fully protected
by a reader-writer lock, sequential tests can be reasonably effective.
At least assuming readers really truly only read...

							Thanx, Paul

> > >From your wording I'm gathering I need to increase this by a lot more
> > test cases?
> 
> I would feel better with the addition of something that looked more
> like a stress test.  For but one example, is there some combination
> of several successive update operations that can mess up slow readers
> (that is, readers that are interrupted or preempted, and thus have
> multiple updates happen while they are traversing the tree)?  If so,
> the current tests will not find it.
> 
> 							Thanx, Paul