On Tue, Oct 31, 2023 at 04:20:33PM +0100, Peter Zijlstra wrote:
> On Tue, Oct 31, 2023 at 07:24:13AM -0700, Paul E. McKenney wrote:
>
> > So, at least until GCC catches up to clang's code generation, I take it
> > that you don't want WRITE_ONCE() for that ->nvcsw increment. Thoughts on
> > ->on_rq?
>
> I've not done the patch yet, but I suspect those would be fine, those
> are straight up stores, hard to get wrong (famous last words).
Assuming that the reads are already either marked with READ_ONCE() or
are under appropriate locks, my immediate thought would be something
like the all-too-lightly tested patch shown below.
The ASSERT_EXCLUSIVE_WRITER() causes KCSAN to complain if there is a
concurrent store of any kind to the location.
Of course, please feel free to ignore. Thoughts?
Thanx, Paul
------------------------------------------------------------------------
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 81885748871d..aeace19ad7f5 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2124,12 +2124,14 @@ void activate_task(struct rq *rq, struct task_struct *p, int flags)
enqueue_task(rq, p, flags);
- p->on_rq = TASK_ON_RQ_QUEUED;
+ WRITE_ONCE(p->on_rq, TASK_ON_RQ_QUEUED);
+ ASSERT_EXCLUSIVE_WRITER(p->on_rq);
}
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
{
- p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING;
+ WRITE_ONCE(p->on_rq, (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING);
+ ASSERT_EXCLUSIVE_WRITER(p->on_rq);
dequeue_task(rq, p, flags);
}
