linux-kernel - Re: [PATCH 6/6] sched: Simplify set_affinity

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: <jhjlfbd5vja.mognet@arm.com>
Date:   Wed, 24 Feb 2021 15:34:49 +0000
From:   Valentin Schneider <valentin.schneider@....com>
To:     Peter Zijlstra <peterz@...radead.org>,
        Ingo Molnar <mingo@...nel.org>,
        Thomas Gleixner <tglx@...utronix.de>
Cc:     Vincent Guittot <vincent.guittot@...aro.org>,
        Mel Gorman <mgorman@...e.de>,
        Dietmar Eggemann <dietmar.eggemann@....com>,
        linux-kernel@...r.kernel.org, peterz@...radead.org,
        Andi Kleen <andi@...stfloor.org>
Subject: Re: [PATCH 6/6] sched: Simplify set_affinity_pending refcounts

On 24/02/21 13:24, Peter Zijlstra wrote:
> Now that we have set_affinity_pending::stop_pending to indicate if a
> stopper is in progress, and we have the guarantee that if that stopper
> exists, it will (eventually) complete our @pending we can simplify the
> refcount scheme by no longer counting the stopper thread.
>
> Fixes: 6d337eab041d ("sched: Fix migrate_disable() vs set_cpus_allowed_ptr()")
> Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
> ---
> @@ -2199,12 +2199,16 @@ static int affine_move_task(struct rq *r
>                       push_task = get_task_struct(p);
>               }
>
> +		/*
> +		 * If there are pending waiters, but no pending stop_work,
> +		 * then complete now.
> +		 */
>               pending = p->migration_pending;
> -		if (pending) {
> -			refcount_inc(&pending->refs);
> +		if (pending && !pending->stop_pending) {
>                       p->migration_pending = NULL;
>                       complete = true;
>               }
> +
>               task_rq_unlock(rq, p, rf);
>
>               if (push_task) {
> @@ -2213,7 +2217,7 @@ static int affine_move_task(struct rq *r
>               }
>
>               if (complete)
> -			goto do_complete;
> +			complete_all(&pending->done);

We could've done this in the first place, right? I don't think this path
actually needed to deal with the refcounts (at least not since we started
counting the stoppers).

Musings aside, I believe the above means, for migration_cpu_stop():

  (pending != NULL) => (pending == p->migration_pending)

Since, when ->stop_pending, only the stopper can uninstall
p->migration_pending. This could simplify a few if's.

Also, the fatty comment above affine_move_task() probably needs a bit of
gardening:

---
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 9492f8eb242a..6f649aa2795c 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2165,16 +2165,21 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
  *
  * (1) In the cases covered above. There is one more where the completion is
  * signaled within affine_move_task() itself: when a subsequent affinity request
- * cancels the need for an active migration. Consider:
+ * occurs after the stopper bailed out due to the targeted task still being
+ * Migrate-Disable. Consider:
  *
  *     Initial conditions: P0->cpus_mask = [0, 1]
  *
- *     P0@...0            P1                             P2
- *
- *     migrate_disable();
- *     <preempted>
+ *     CPU0               P1                             P2
+ *     <P0>
+ *       migrate_disable();
+ *       <preempted>
  *                        set_cpus_allowed_ptr(P0, [1]);
  *                          <blocks>
+ *     <migration/0>
+ *       migration_cpu_stop()
+ *         is_migration_disabled()
+ *           <bails>
  *                                                       set_cpus_allowed_ptr(P0, [0, 1]);
  *                                                         <signal completion>
  *                          <awakes>