linux-kernel - Re: [RFC PATCH 1/1] rcu: use atomic_read(v) instead of atomic_add

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]

Date:	Wed, 9 Jul 2014 00:00:10 -0400
From:	Pranith Kumar <bobby.prani@...il.com>
To:	Paul McKenney <paulmck@...ux.vnet.ibm.com>
Cc:	Josh Triplett <josh@...htriplett.org>,
	"open list:READ-COPY UPDATE..." <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH 1/1] rcu: use atomic_read(v) instead of
 atomic_add_return(0, v)

On Tue, Jul 8, 2014 at 8:07 PM, Paul E. McKenney
<paulmck@...ux.vnet.ibm.com> wrote:
> On Tue, Jul 08, 2014 at 06:55:45PM -0400, Pranith Kumar wrote:
>> atomic_add_return() invalidates the cache line in other processors where-as
>> atomic_read does not. I don't see why we would need invalidation in this case.
>> If indeed it was need a comment would be helpful for readers. Otherwise doesn't
>> using atomic_read() make more sense here? RFC!
>>
>> replace atomic_add_return(0, v) with atomic_read(v) as the latter is better.
>>
>> Signed-off-by: Pranith Kumar <bobby.prani@...il.com>
>
> This will break RCU -- the full memory barriers implied both before
> and after atomic_add_return() are needed in order for RCU to be able to
> avoid death due to memory reordering.
>
> That said, I have considered replacing the atomic_add_return() with:
>
>         smp_mb();
>         ... = atomic_read(...);
>         smp_mb();
>
> However, this is a very ticklish change, and would need serious thought
> and even more serious testing.
>

Thank you for looking at the RFC. I tried understanding the code
deeper and found that the ordering which is needed here is actually on
dynticks_snap.
The ordering currently (by way of atomic_add_return) is on
rdp->dynticks->dynticks which I think is not right.

Looking at the history of the code led me to rev. 23b5c8fa01b723 which
makes me think that the above statement is true. I think providing
ordering guarantees on dynticks_snap should be enough.

I have added an updated patch below. However, it is really hard(and
error prone) to convince oneself that this is right. So I will not
pursue this further if you think this is wrong. You definitely know
better than me :)

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 1b70cb6..bbccd0a 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -891,7 +891,7 @@ static int rcu_is_cpu_rrupt_from_idle(void)
 static int dyntick_save_progress_counter(struct rcu_data *rdp,
                                         bool *isidle, unsigned long *maxj)
 {
-       rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
+       smp_store_release(&rdp->dynticks_snap,
atomic_read(&rdp->dynticks->dynticks));
        rcu_sysidle_check_cpu(rdp, isidle, maxj);
        if ((rdp->dynticks_snap & 0x1) == 0) {
                trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
@@ -920,8 +920,8 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
        int *rcrmp;
        unsigned int snap;

-       curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks);
-       snap = (unsigned int)rdp->dynticks_snap;
+       curr = (unsigned int)atomic_read(&rdp->dynticks->dynticks);
+       snap = (unsigned int)smp_load_acquire(&rdp->dynticks_snap);

        /*
         * If the CPU passed through or entered a dynticks idle phase with


-- 
Pranith
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/