Re: [PATCH 00/14] sched: Support shared runqueue locking

[Andrea Righi <arighi@nvidia.com>]

On Thu, Sep 11, 2025 at 11:58:45AM +0200, Peter Zijlstra wrote:
> On Wed, Sep 10, 2025 at 08:35:55PM +0200, Peter Zijlstra wrote:
> 
> > I'll go untangle it, but probably something for tomorrow, I'm bound to
> > make a mess of it now :-)
> 
> Best I could come up with is something like this. I tried a few other
> approaches, but they all turned into a bigger mess.
> 
> Let me go try and run this.

With this one it's complaining about lockdep_assert_held(p->srq_lock):

[   19.055730] WARNING: CPU: 2 PID: 368 at kernel/sched/core.c:10840 sched_change_begin+0x2ac/0x3e0
...
[   19.056468] RIP: 0010:sched_change_begin+0x2ac/0x3e0
...
[   19.057217] RSP: 0018:ffffa9f7805bbde8 EFLAGS: 00010046
[   19.057359] RAX: 0000000000000000 RBX: ffff97ae04880000 RCX: 0000000000000001
[   19.057464] RDX: 0000000000000046 RSI: ffff97ae01715518 RDI: ffff97ae027f0b68
[   19.057568] RBP: 0000000000000082 R08: 0000000000000001 R09: 0000000000000001
[   19.057706] R10: 0000000000000001 R11: 0000000000000001 R12: ffff97ae3bdbcc80
[   19.057833] R13: ffff97ae93c48000 R14: ffff97ae3b717f20 R15: 0000000000000000
[   19.057973] FS:  00007f18999edb00(0000) GS:ffff97ae93c48000(0000) knlGS:0000000000000000
[   19.058112] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   19.058223] CR2: 000055e1e6b0246c CR3: 0000000102ce8000 CR4: 0000000000750ef0
[   19.058460] PKRU: 55555554
[   19.058561] Call Trace:
[   19.058604]  <TASK>
[   19.058675]  __set_cpus_allowed_ptr_locked+0x17c/0x230
[   19.058769]  __set_cpus_allowed_ptr+0x64/0xa0
[   19.058853]  __sched_setaffinity+0x72/0x100
[   19.058920]  sched_setaffinity+0x261/0x2f0
[   19.058985]  __x64_sys_sched_setaffinity+0x50/0x80
[   19.059084]  do_syscall_64+0xbb/0x370
[   19.059158]  entry_SYSCALL_64_after_hwframe+0x77/0x7f
[   19.059236] RIP: 0033:0x7f189a3bd25b

Thanks,
-Andrea

> 
> ---
>  kernel/sched/core.c |   10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
> 
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -2481,11 +2481,11 @@ static inline bool is_cpu_allowed(struct
>   * Returns (locked) new rq. Old rq's lock is released.
>   */
>  static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf,
> -				   struct task_struct *p, int new_cpu)
> +				   struct task_struct *p, int new_cpu, int flags)
>  {
>  	lockdep_assert_rq_held(rq);
>  
> -	deactivate_task(rq, p, DEQUEUE_NOCLOCK);
> +	deactivate_task(rq, p, flags | DEQUEUE_NOCLOCK);
>  	set_task_cpu(p, new_cpu);
>  	rq_unlock(rq, rf);
>  
> @@ -2493,7 +2493,7 @@ static struct rq *move_queued_task(struc
>  
>  	rq_lock(rq, rf);
>  	WARN_ON_ONCE(task_cpu(p) != new_cpu);
> -	activate_task(rq, p, 0);
> +	activate_task(rq, p, flags);
>  	wakeup_preempt(rq, p, 0);
>  
>  	return rq;
> @@ -2533,7 +2533,7 @@ static struct rq *__migrate_task(struct
>  	if (!is_cpu_allowed(p, dest_cpu))
>  		return rq;
>  
> -	rq = move_queued_task(rq, rf, p, dest_cpu);
> +	rq = move_queued_task(rq, rf, p, dest_cpu, 0);
>  
>  	return rq;
>  }
> @@ -3007,7 +3007,7 @@ static int affine_move_task(struct rq *r
>  
>  		if (!is_migration_disabled(p)) {
>  			if (task_on_rq_queued(p))
> -				rq = move_queued_task(rq, rf, p, dest_cpu);
> +				rq = move_queued_task(rq, rf, p, dest_cpu, DEQUEUE_LOCKED);
>  
>  			if (!pending->stop_pending) {
>  				p->migration_pending = NULL;