On Thu, Oct 03, 2024 at 09:04:30AM -0700, Paul E. McKenney wrote:
> On Thu, Oct 03, 2024 at 04:22:40PM +0200, Peter Zijlstra wrote:
> > On Thu, Oct 03, 2024 at 05:45:47AM -0700, Paul E. McKenney wrote:
> >
> > > I ran 100*TREE03 for 18 hours each, and got 23 instances of *something*
> > > happening (and I need to suppress stalls on the repeat). One of the
> > > earlier bugs happened early, but sadly not this one.
> >
> > Damn, I don't have the amount of CPU hours available you mention in your
> > later email. I'll just go up the rounds to 20 minutes and see if
> > something wants to go bang before I have to shut down the noise
> > pollution for the day...
>
> Indeed, this was one reason I was soliciting debug patches. ;-)
Sooo... I was contemplating if something like the below might perhaps
help some. It's a bit of a mess (I'll try and clean up if/when it
actually proves to work), but it compiles and survives a hand full of 1m
runs.
I'll try and give it more runs tomorrow when I can power up the big
machines again -- unless you've already told me it's crap by then :-)
---
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 43e453ab7e20..1fe850788195 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7010,20 +7010,20 @@ int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flag
}
EXPORT_SYMBOL(default_wake_function);
-void __setscheduler_prio(struct task_struct *p, int prio)
+const struct sched_class *__setscheduler_class(struct task_struct *p, int prio)
{
if (dl_prio(prio))
- p->sched_class = &dl_sched_class;
- else if (rt_prio(prio))
- p->sched_class = &rt_sched_class;
+ return &dl_sched_class;
+
+ if (rt_prio(prio))
+ return &rt_sched_class;
+
#ifdef CONFIG_SCHED_CLASS_EXT
- else if (task_should_scx(p))
- p->sched_class = &ext_sched_class;
+ if (task_should_scx(p))
+ return &ext_sched_class;
#endif
- else
- p->sched_class = &fair_sched_class;
- p->prio = prio;
+ return &fair_sched_class;
}
#ifdef CONFIG_RT_MUTEXES
@@ -7069,7 +7069,7 @@ void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
{
int prio, oldprio, queued, running, queue_flag =
DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
- const struct sched_class *prev_class;
+ const struct sched_class *prev_class, *next_class;
struct rq_flags rf;
struct rq *rq;
@@ -7127,6 +7127,11 @@ void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
queue_flag &= ~DEQUEUE_MOVE;
prev_class = p->sched_class;
+ next_class = __setscheduler_class(p, prio);
+
+ if (prev_class != next_class && p->se.sched_delayed)
+ dequeue_task(rq, p, DEQUEUE_SLEEP | DEQUEUE_DELAYED | DEQUEUE_NOCLOCK);
+
queued = task_on_rq_queued(p);
running = task_current(rq, p);
if (queued)
@@ -7164,7 +7169,9 @@ void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
p->rt.timeout = 0;
}
- __setscheduler_prio(p, prio);
+ p->sched_class = next_class;
+ p->prio = prio;
+
check_class_changing(rq, p, prev_class);
if (queued)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index ab497fafa7be..c157d4860a3b 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -13177,22 +13177,6 @@ static void attach_task_cfs_rq(struct task_struct *p)
static void switched_from_fair(struct rq *rq, struct task_struct *p)
{
detach_task_cfs_rq(p);
- /*
- * Since this is called after changing class, this is a little weird
- * and we cannot use DEQUEUE_DELAYED.
- */
- if (p->se.sched_delayed) {
- /* First, dequeue it from its new class' structures */
- dequeue_task(rq, p, DEQUEUE_NOCLOCK | DEQUEUE_SLEEP);
- /*
- * Now, clean up the fair_sched_class side of things
- * related to sched_delayed being true and that wasn't done
- * due to the generic dequeue not using DEQUEUE_DELAYED.
- */
- finish_delayed_dequeue_entity(&p->se);
- p->se.rel_deadline = 0;
- __block_task(rq, p);
- }
}
static void switched_to_fair(struct rq *rq, struct task_struct *p)
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index b1c3588a8f00..fba524c81c63 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -3797,7 +3797,7 @@ static inline int rt_effective_prio(struct task_struct *p, int prio)
extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
-extern void __setscheduler_prio(struct task_struct *p, int prio);
+extern const struct sched_class *__setscheduler_class(struct task_struct *p, int prio);
extern void set_load_weight(struct task_struct *p, bool update_load);
extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
diff --git a/kernel/sched/syscalls.c b/kernel/sched/syscalls.c
index aa70beee9895..0470bcc3d204 100644
--- a/kernel/sched/syscalls.c
+++ b/kernel/sched/syscalls.c
@@ -529,7 +529,7 @@ int __sched_setscheduler(struct task_struct *p,
{
int oldpolicy = -1, policy = attr->sched_policy;
int retval, oldprio, newprio, queued, running;
- const struct sched_class *prev_class;
+ const struct sched_class *prev_class, *next_class;
struct balance_callback *head;
struct rq_flags rf;
int reset_on_fork;
@@ -706,6 +706,12 @@ int __sched_setscheduler(struct task_struct *p,
queue_flags &= ~DEQUEUE_MOVE;
}
+ prev_class = p->sched_class;
+ next_class = __setscheduler_class(p, newprio);
+
+ if (prev_class != next_class && p->se.sched_delayed)
+ dequeue_task(rq, p, DEQUEUE_SLEEP | DEQUEUE_DELAYED | DEQUEUE_NOCLOCK);
+
queued = task_on_rq_queued(p);
running = task_current(rq, p);
if (queued)
@@ -713,11 +719,10 @@ int __sched_setscheduler(struct task_struct *p,
if (running)
put_prev_task(rq, p);
- prev_class = p->sched_class;
-
if (!(attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)) {
__setscheduler_params(p, attr);
- __setscheduler_prio(p, newprio);
+ p->sched_class = next_class;
+ p->prio = newprio;
}
__setscheduler_uclamp(p, attr);
check_class_changing(rq, p, prev_class);
