linux-kernel - Re: [PATCH memory-model 2/4] tools/memory-model: Add example for heuristic lockless reads

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

Message-ID: <20210723181138.GA48833@rowland.harvard.edu>
Date:   Fri, 23 Jul 2021 14:11:38 -0400
From:   Alan Stern <stern@...land.harvard.edu>
To:     "Paul E. McKenney" <paulmck@...nel.org>
Cc:     linux-kernel@...r.kernel.org, linux-arch@...r.kernel.org,
        kernel-team@...com, mingo@...nel.org, parri.andrea@...il.com,
        will@...nel.org, peterz@...radead.org, boqun.feng@...il.com,
        npiggin@...il.com, dhowells@...hat.com, j.alglave@....ac.uk,
        luc.maranget@...ia.fr, akiyks@...il.com,
        Manfred Spraul <manfred@...orfullife.com>
Subject: Re: [PATCH memory-model 2/4] tools/memory-model: Add example for
 heuristic lockless reads

On Fri, Jul 23, 2021 at 10:30:10AM -0700, Paul E. McKenney wrote:
> On Fri, Jul 23, 2021 at 12:59:47PM -0400, Alan Stern wrote:
> > On Fri, Jul 23, 2021 at 09:24:31AM -0700, Paul E. McKenney wrote:
> > > On Thu, Jul 22, 2021 at 10:08:46PM -0400, Alan Stern wrote:
> > 
> > > > > +	void do_something_locked(struct foo *fp)
> > > > > +	{
> > > > > +		bool gf = true;
> > > > > +
> > > > > +		/* IMPORTANT: Heuristic plus spin_lock()! */
> > > > > +		if (!data_race(global_flag)) {
> > > > > +			spin_lock(&fp->f_lock);
> > > > > +			if (!smp_load_acquire(&global_flag)) {
> > 
> > > > > +	void begin_global(void)
> > > > > +	{
> > > > > +		int i;
> > > > > +
> > > > > +		spin_lock(&global_lock);
> > > > > +		WRITE_ONCE(global_flag, true);
> > > > 
> > > > Why does this need to be WRITE_ONCE?  It still races with the first read 
> > > > of global_flag above.
> > > 
> > > But also with the smp_load_acquire() of global_flag, right?
> > 
> > What I'm curious about is why, given these two races, you notate one of 
> > them by changing a normal write to WRITE_ONCE and you notate the other 
> > by changing a normal read to a data_race() read.  Why not handle them 
> > both the same way?
> 
> Because the code can tolerate the first read returning complete nonsense,
> but needs the value from the second read to be exact at that point in
> time.

In other words, if the second read races with the WRITE_ONCE, it needs to 
get either the value before the write or the value after the write; 
nothing else will do because it isn't a heuristic here.  Fair point.

>  (If the value changes immediately after being read, the fact that
> ->f_lock is held prevents begin_global() from completing.)

This seems like something worth explaining in the document.  That 
"IMPORTANT" comment doesn't really get the full point across.

Alan