Re: [PATCH v24 08/11] sched: Avoid donor->sched_class->yield_task() null traversal

[K Prateek Nayak <kprateek.nayak@amd.com>]

Hello John,

On 12/30/2025 11:31 AM, K Prateek Nayak wrote:
> On 11/25/2025 4:01 AM, John Stultz wrote:
>> With proxy-exec once we do return migration from ttwu(), if a
>> task is proxying for a waiting donor, and the donor is woken up,
>> we switch the rq->donor to point to idle briefly until we can
>> re-enter __schedule().
>>
>> However, if a task that was acting as a proxy calls into
>> yield() right after the donor is switched to idle, it may
>> trip a null  pointer traversal, because the idle task doesn't
>> have a yield_task() pointer.
> 
> I thought that was a transient state that should not be observed by the
> running task.
> 
> Since NEED_RESCHED is retained, we'll just go though another pass of
> __schedule() loop and the task should not observe rq->donor as idle in
> do_sched_yield() as only the current task can yield on the local CPU.
> 
> Do we have a splat that suggests this happens?

So I think I found my answer (at least one of them):

  find_proxy_task()
    /* case DEACTIVATE_DONOR */
    proxy_deactivate(rq, donor)
      proxy_resched_idle(rq); /* Switched donor to rq->idle. */
      try_to_block_task(rq, donor, &state, true)
        if (signal_pending_state(task_state, donor))
          WRITE_ONCE(p->__state, TASK_RUNNING)
          return false; /* Blocking fails. */

    /* If deactivate fails, force return */
    p = donor;
    return p

  next = p; /* Donor is rq->idle. */


This should be illegal and I think we should either force a "pick_again"
if proxy_deactivate() fails (can it get stuck on an infinite loop?) or
we should fix the donor relation before running "p".

We can also push the proxy_resched_idle() into try_to_block_task() and
only do it once we are past all the early returns and if
task_current_donor(rq, p).

... and as I write this I realize we can have this via
proxy_needs_return() so I guess we need this patch after all and
proxy_needs_return() should do resched_curr() for that case too so we
can re-evaluate the donor context on the CPU where we are stealing away
the donor from.

Quick question: Can we avoid that proxy_resched_idle() in
proxy_needs_return() by retaining PROXY_WAKING and letting
proxy_force_return() handle donor's migration too?

  (Seems to survive the same set of challenges I've been putting my
   suggestions through)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 700fd08b2392..e01a61f1955e 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -3720,6 +3720,12 @@ static inline bool proxy_needs_return(struct rq *rq, struct task_struct *p)
 	if (!sched_proxy_exec())
 		return false;
 
+	/* For current, and donor, let proxy_force_return() handle return. */
+	if (task_current(rq, p) || task_current_donor(rq, p)) {
+		resched_curr(rq);
+		return false;
+	}
+
 	guard(raw_spinlock)(&p->blocked_lock);
 
 	/* If task isn't PROXY_WAKING, we don't need to do return migration */
@@ -3728,20 +3734,12 @@ static inline bool proxy_needs_return(struct rq *rq, struct task_struct *p)
 
 	__clear_task_blocked_on(p, PROXY_WAKING);
 
-	/* If already current, don't need to return migrate */
-	if (task_current(rq, p))
-		return false;
-
 	/* If wake_cpu is targeting this cpu, don't bother return migrating */
 	if (p->wake_cpu == cpu_of(rq)) {
 		resched_curr(rq);
 		return false;
 	}
 
-	/* If we're return migrating the rq->donor, switch it out for idle */
-	if (task_current_donor(rq, p))
-		proxy_resched_idle(rq);
-
 	/* (ab)Use DEQUEUE_SPECIAL to ensure task is always blocked here. */
 	block_task(rq, p, DEQUEUE_NOCLOCK | DEQUEUE_SPECIAL);
 	return true;
---

I think you'll have a note somewhere that says why this is an absolutely
terrible idea :-)

-- 
Thanks and Regards,
Prateek

  “Program testing can be used to show the presence of bugs,
   but never to show their absence!”  ― Edsger W. Dijkstra