Re: [PATCH 3/3] soc: samsung: exynos-pmu: fix error paths in cpuhotplug/idle states setup

[Alexey Klimov]

Hi Peter,

On Thu Jun 11, 2026 at 8:07 AM BST, Peter Griffin wrote:
> Hi Alexey,
>
> On Wed, 10 Jun 2026 at 16:07, Alexey Klimov <alexey.klimov@xxxxxxxxxx> wrote:
>>
>> On Wed Jun 10, 2026 at 2:34 PM BST, Peter Griffin wrote:
>> > Hi Alexey,
>>
>> Hi Peter,
>>
>> > Thanks for your patch!
>> >
>> > On Fri, 5 Jun 2026 at 21:19, Alexey Klimov <alexey.klimov@xxxxxxxxxx> wrote:
>> >>
>> >> The setup_cpuhp_and_cpuidle() initialisation sequence currently ignores
>> >> the return values of cpuhp_setup_state(), cpu_pm_register_notifier(), and
>> >> register_reboot_notifier(). If any of these registrations fail during
>> >> probe() routine, the driver returns 0, leaving the driver partially
>> >> configured.
>> >
>> > I originally made the failure non-fatal because the system still boots
>> > without the notifiers registered (and all other Arm64 Exynos SoCs
>> > upstream don't register notifiers and AFAICT have broken cpu hotplug
>> > and cpu idle).
>> >
>> > In hindsight, that seems like a mistake. I think your patch to fully
>> > unwind everything in case of failure makes more sense.  See small
>> > comment below about destroy_cpuhp_and_cpuidle()
>>
>> Wait, setup_cpuhp_and_cpuidle() should be non-fatal and shouldn't
>> return any errors?
>
> I suggest you re-read my above comment above ^^

Could you please clarify what specifically addresses my question about
notifiers?

Looking further into this, it seems that, for instance, if one of the
hotplug states fails to register then tracking of pmu_context->in_cpuhp
becomes broken.
If reboot notifier silently fails to be registered, then it is unclear how
this from gs101_cpu_pmu_offline() supposed to work:

/* Ignore CPU_PM_ENTER event in reboot or suspend sequence. */
if (pmu_context->sys_insuspend || pmu_context->sys_inreboot) {
	raw_spin_unlock(&pmu_context->cpupm_lock);
	return NOTIFY_OK;
}

If c2 idles are used during reboot/shutdown then they fail or what?

I am not saying that patch is correct and some rework is needed but I don't
get why we should completely ignore errors from hotplug states registration
and should not check registration of notifiers. At least warning should be
shown to user that pm functionality might be unreliable.

Best regards,
Alexey.