[PATCH v2] ARM: smp: Only expose /sys/.../cpuX/online if hotpluggable

[Magnus Damm]

Hi Stephen,

On Fri, Feb 13, 2015 at 8:59 PM, Stephen Boyd <sboyd at codeaurora.org> wrote:
> On 02/13/15 12:54, Magnus Damm wrote:
>> Hi Stephen,
>>
>> On Fri, Feb 13, 2015 at 8:23 PM, Stephen Boyd <sboyd at codeaurora.org> wrote:
>>> On 02/13/15 12:20, Simon Horman wrote:
>>>> Hi Stephen,
>>>>
>>>> On Fri, Feb 13, 2015 at 10:06:39AM -0800, Stephen Boyd wrote:
>>>>> diff --git a/arch/arm/mach-shmobile/smp-r8a7779.c b/arch/arm/mach-shmobile/smp-r8a7779.c
>>>>> index 3f761f839043..b45206f93ddf 100644
>>>>> --- a/arch/arm/mach-shmobile/smp-r8a7779.c
>>>>> +++ b/arch/arm/mach-shmobile/smp-r8a7779.c
>>>>> @@ -124,19 +124,12 @@ static int r8a7779_cpu_kill(unsigned int cpu)
>>>>>
>>>>>      return 0;
>>>>>  }
>>>>> -
>>>>> -static int r8a7779_cpu_disable(unsigned int cpu)
>>>>> -{
>>>>> -    /* only CPU1->3 have power domains, do not allow hotplug of CPU0 */
>>>>> -    return cpu == 0 ? -EPERM : 0;
>>>>> -}
>>>>>  #endif /* CONFIG_HOTPLUG_CPU */
>>>>>
>>>>>  struct smp_operations r8a7779_smp_ops  __initdata = {
>>>>>      .smp_prepare_cpus       = r8a7779_smp_prepare_cpus,
>>>>>      .smp_boot_secondary     = r8a7779_boot_secondary,
>>>>>  #ifdef CONFIG_HOTPLUG_CPU
>>>>> -    .cpu_disable            = r8a7779_cpu_disable,
>>>>>      .cpu_die                = shmobile_smp_scu_cpu_die,
>>>>>      .cpu_kill               = r8a7779_cpu_kill,
>>>>>  #endif
>>>> Its not clear to me why r8a7779_cpu_disable() has been
>>>> removed rather than replaced by r8a7779_cpu_can_disable()
>>> By default ARM's smp.c assumes that CPU0 can't be hotplugged. The
>>> function is redundant. I guess I should have mentioned that in the
>>> commit text.
>> Thanks for your efforts. Can you please tell me where that limitation
>> is located? I'm quite sure I've brought CPU cores up and down
>> including CPU0, but maybe something is missing?
>>
>>
>
> The assumption is made if there isn't a cpu_disable (now
> cpu_can_disable) op. See platform_cpu_disable() in
> arch/arm/kernel/smp.c. If there isn't such a limitation on a particular
> platform, the platform needs to set the cpu_disable/cpu_can_disable op
> and indicate that there isn't a limitation (by returning 1 for example).

Thanks, I now understand.

The mach-shmobile bits look fine to me, but if I could pick freely
then I would prefer the changes to be broken out a bit more. For
instance, separate the r8a7779_cpu_disable() removal from the callback
renaming and the logic changes.

Also, based on the comment in mcpm_cpu_can_disable() it looks like the
PSCI hook may be executed once only with your change in place?
Hopefully PSCI is OK not being invoked for every CPU shutdown.

Cheers,

/ magnus