Date:   Tue, 27 Jun 2017 16:37:48 -0700
From:   "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     Andrea Parri <parri.andrea@...il.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        priyalee.kushwaha@...el.com, drozdziak1@...il.com,
        Arnd Bergmann <arnd@...db.de>, ldr709@...il.com,
        Thomas Gleixner <tglx@...utronix.de>,
        Peter Zijlstra <peterz@...radead.org>,
        Josh Triplett <josh@...htriplett.org>,
        Nicolas Pitre <nico@...aro.org>,
        Krister Johansen <kjlx@...pleofstupid.com>,
        Vegard Nossum <vegard.nossum@...cle.com>, dcb314@...mail.com,
        Wu Fengguang <fengguang.wu@...el.com>,
        Frederic Weisbecker <fweisbec@...il.com>,
        Rik van Riel <riel@...hat.com>,
        Steven Rostedt <rostedt@...dmis.org>,
        Ingo Molnar <mingo@...nel.org>,
        Alan Stern <stern@...land.harvard.edu>,
        Luc Maranget <luc.maranget@...ia.fr>,
        Jade Alglave <j.alglave@....ac.uk>
Subject: Re: [GIT PULL rcu/next] RCU commits for 4.13

On Tue, Jun 27, 2017 at 02:48:18PM -0700, Linus Torvalds wrote:
> On Tue, Jun 27, 2017 at 1:58 PM, Paul E. McKenney
> <paulmck@...ux.vnet.ibm.com> wrote:
> >
> > So what next?
> >
> > One option would be to weaken the definition of spin_unlock_wait() so
> > that it had acquire semantics but not release semantics.  Alternatively,
> > we could keep the full empty-critical-section semantics and add memory
> > barriers to spin_unlock_wait() implementations, perhaps as shown in the
> > patch below.  I could go either way, though I do have some preference
> > for the stronger semantics.
> >
> > Thoughts?
> 
> I would prefer to just say
> 
>  - document that spin_unlock_wait() has acquire semantics
> 
>  - mindlessly add the smp_mb() to all users
> 
>  - let users then decide if they are ok with just acquire
> 
> That's partly because I think we actually have *fewer* users than we
> have implementations of spin_unlock_wait(). So adding a few smp_mb()'s
> in the users is actually likely the smaller change.

You are right about that!  There are only five invocations of
spin_unlock_wait() in the kernel, with a sixth that has since been
converted to spin_lock() immediately followed by spin_unlock().

> But it's also because then that allows people who *can* say that
> acquire is sufficient to just use it. People who use
> spin_unlock_wait() tend to have some odd performance reason to do so,
> so I think allowing them to use the more light-weight memory ordering
> if it works for them is a good idea.
> 
> But finally, it's partly because I think "acquire" semantics are
> actually the saner ones that we can explain the logic for much more
> clearly.
> 
> Basically, acquire semantics means that you are guaranteed to see any
> changes that were done inside a previously locked region.
> 
> Doesn't that sound like sensible semantics?

It is the semantics that most implementations of spin_unlock_wait()
provide.  Of the six invocations, two of them very clearly rely
only on the acquire semantics and two others already have the needed
memory barriers in place.  I have queued one patch to add smp_mb()
to the remaining spin_unlock_wait() of the surviving five instances,
and another patch to convert the spin_lock/unlock pair to smp_mb()
followed by spin_unlock_wait().

So, yes, it is a sensible set of semantics.  At this point, agreeing
-any- reasonable semantics would be good, as it would allow us to get
locking added to the prototype Linux-kernel memory model. ;-)

> Then, the argument for "smp_mb()" (before the spin_unlock_wait()) becomes:
> 
>  - I did a write that will affect any future lock takes
> 
>  - the smp_mb() now means that that write will be ordered wrt the
> acquire that guarantees we've seen all old actions taken by a lock.
> 
> Does those kinds of semantics make sense to people?

In case the answer is "yes", the (untested) patch below (combining
three commits) shows the changes that I believe would be required.
A preview is also available as individual commits on branch
spin_unlock_wait.2017.06.27a on -rcu here:

	git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git

As usual, thoughts?  ;-)

							Thanx, Paul

------------------------------------------------------------------------

diff --git a/drivers/ata/libata-eh.c b/drivers/ata/libata-eh.c
index ef68232b5222..cc01b77a079a 100644
--- a/drivers/ata/libata-eh.c
+++ b/drivers/ata/libata-eh.c
@@ -704,8 +704,10 @@ void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
 
 		/* initialize eh_tries */
 		ap->eh_tries = ATA_EH_MAX_TRIES;
-	} else
+	} else {
+		smp_mb(); /* Add release semantics for spin_unlock_wait(). */
 		spin_unlock_wait(ap->lock);
+	}
 
 }
 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
diff --git a/include/linux/spinlock.h b/include/linux/spinlock.h
index d9510e8522d4..0c3f54e2a1d1 100644
--- a/include/linux/spinlock.h
+++ b/include/linux/spinlock.h
@@ -373,21 +373,21 @@ static __always_inline int spin_trylock_irq(spinlock_t *lock)
  * spin_unlock_wait - Interpose between successive critical sections
  * @lock: the spinlock whose critical sections are to be interposed.
  *
- * Semantically this is equivalent to a spin_lock() immediately
- * followed by a spin_unlock().  However, most architectures have
- * more efficient implementations in which the spin_unlock_wait()
- * cannot block concurrent lock acquisition, and in some cases
- * where spin_unlock_wait() does not write to the lock variable.
- * Nevertheless, spin_unlock_wait() can have high overhead, so if
- * you feel the need to use it, please check to see if there is
- * a better way to get your job done.
+ * Semantically this is equivalent to a spin_lock() immediately followed
+ * by a mythical spin_unlock() that has no ordering semantics.  However,
+ * most architectures have more efficient implementations in which the
+ * spin_unlock_wait() cannot block concurrent lock acquisition, and in some
+ * cases where spin_unlock_wait() does not write to the lock variable.
+ * Nevertheless, spin_unlock_wait() can have high overhead, so if you
+ * feel the need to use it, please check to see if there is a better way
+ * to get your job done.
  *
- * The ordering guarantees provided by spin_unlock_wait() are:
- *
- * 1.  All accesses preceding the spin_unlock_wait() happen before
- *     any accesses in later critical sections for this same lock.
- * 2.  All accesses following the spin_unlock_wait() happen after
- *     any accesses in earlier critical sections for this same lock.
+ * The spin_unlock_wait() function guarantees that all accesses following
+ * the spin_unlock_wait() happen after any accesses in earlier critical
+ * sections for this same lock.  Please note that it does -not- guarantee
+ * that accesses preceding the spin_unlock_wait() happen before any accesses
+ * in later critical sections for this same lock.  If you need this latter
+ * ordering, precede the spin_unlock_wait() with an smp_mb() or similar.
  */
 static __always_inline void spin_unlock_wait(spinlock_t *lock)
 {
diff --git a/ipc/sem.c b/ipc/sem.c
index 947dc2348271..ef42e55e9dd0 100644
--- a/ipc/sem.c
+++ b/ipc/sem.c
@@ -307,8 +307,8 @@ static void complexmode_enter(struct sem_array *sma)
 
 	for (i = 0; i < sma->sem_nsems; i++) {
 		sem = sma->sem_base + i;
-		spin_lock(&sem->lock);
-		spin_unlock(&sem->lock);
+		smp_mb(); /* Add release semantics for spin_unlock_wait(). */
+		spin_unlock_wait(&sem->lock);
 	}
 }
 

Powered by blists - more mailing lists