Re: [PATCH v6 05/14] sched/deadline: Return EBUSY if dl_bw_cpus is zero

[Andrea Righi <arighi@nvidia.com>]

Hi Joel,

On Wed, Jul 02, 2025 at 07:29:30PM -0400, Joel Fernandes wrote:
> Hotplugged CPUs coming online do an enqueue but are not a part of any
> root domain containing cpu_active() CPUs. So in this case, don't mess
> with accounting and we can retry later. Without this patch, we see
> crashes with sched_ext selftest's hotplug test due to divide by zero.
> 
> Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
> ---
>  kernel/sched/deadline.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index 7129b61d548b..0e73577257ad 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -1725,7 +1725,12 @@ int dl_server_apply_params(struct sched_dl_entity *dl_se, u64 runtime, u64 perio
>  	cpus = dl_bw_cpus(cpu);
>  	cap = dl_bw_capacity(cpu);
>  
> -	if (__dl_overflow(dl_b, cap, old_bw, new_bw))
> +	/*
> +	 * Hotplugged CPUs coming online do an enqueue but are not a part of any
> +	 * root domain containing cpu_active() CPUs. So in this case, don't mess
> +	 * with accounting and we can retry later.
> +	 */
> +	if (!cpus || __dl_overflow(dl_b, cap, old_bw, new_bw))
>  		return -EBUSY;

I can trigger the following with the hotplug sched_ext selftest:

[   61.731069] ------------[ cut here ]------------
[   61.731241] WARNING: CPU: 4 PID: 2191 at kernel/sched/deadline.c:1591 dl_server_start+0x9c/0xb0
...
[   61.731552] Sched_ext: hotplug_cbs (enabled+all), task: runnable_at=-1ms
[   61.731555] RIP: 0010:dl_server_start+0x9c/0xb0
...
[   61.732216] Call Trace:
[   61.732239]  <TASK>
[   61.732262]  enqueue_task_scx+0x2bb/0x350
[   61.732298]  enqueue_task+0x2e/0xd0
[   61.732333]  ttwu_do_activate+0xa4/0x2d0
[   61.732368]  try_to_wake_up+0x2a2/0x8e0
[   61.732406]  cpuhp_bringup_ap+0x72/0x250
[   61.732441]  ? __pfx_cpuhp_bringup_ap+0x10/0x10
[   61.732478]  cpuhp_invoke_callback+0x1b0/0x650
[   61.732523]  __cpuhp_invoke_callback_range+0x7e/0xf0
[   61.732566]  _cpu_up+0xea/0x1e0
[   61.732601]  cpu_up+0xc3/0xd0
[   61.732635]  cpu_subsys_online+0x4b/0xd0
[   61.732678]  device_online+0x49/0x80
[   61.732714]  online_store+0x9a/0xd0
[   61.732744]  ? sysfs_kf_write+0x2b/0x70
[   61.732780]  kernfs_fop_write_iter+0x137/0x200
[   61.732822]  vfs_write+0x264/0x5e0
[   61.732867]  ksys_write+0x79/0xf0
[   61.732901]  do_syscall_64+0xbb/0x370
[   61.732939]  entry_SYSCALL_64_after_hwframe+0x77/0x7f

I think we can simply ignore the !cpus case, as it's safe to skip
hotplugged CPUs that perform an enqueue when coming online, without
triggering a warning, and we can always retry applying the server params
later.

Maybe return -EAGAIN when !cpus and handle that gracefully in
dl_server_start()?

Something along these lines (on top of this patch). What do you think?

Thanks,
-Andrea

 kernel/sched/deadline.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index 5d5819d445fb9..78d2a3b0a4cde 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -1587,9 +1587,14 @@ void dl_server_start(struct sched_dl_entity *dl_se)
 	if (!dl_server(dl_se)) {
 		u64 runtime =  50 * NSEC_PER_MSEC;
 		u64 period = 1000 * NSEC_PER_MSEC;
+		int err;
 
-		if (WARN_ON_ONCE(dl_server_apply_params(dl_se, runtime, period, 1)))
+		err = dl_server_apply_params(dl_se, runtime, period, 1);
+		if (err == -EAGAIN)
 			return;
+
+		WARN_ON_ONCE(err);
+
 		dl_se->dl_server = 1;
 		dl_se->dl_defer = 1;
 		setup_new_dl_entity(dl_se);
@@ -1662,7 +1667,10 @@ int dl_server_apply_params(struct sched_dl_entity *dl_se, u64 runtime, u64 perio
 	 * root domain containing cpu_active() CPUs. So in this case, don't mess
 	 * with accounting and we can retry later.
 	 */
-	if (!cpus || __dl_overflow(dl_b, cap, old_bw, new_bw))
+	if (!cpus)
+		return -EAGAIN;
+
+	if (__dl_overflow(dl_b, cap, old_bw, new_bw))
 		return -EBUSY;
 
 	if (init) {