watchdog: octeon: missing FROZEN hot plug notifier actions

watchdog: octeon: missing FROZEN hot plug notifier actions

[rcochran]

In drivers/watchdog/octeon-wdt-main.c, the hot plug notifier callback,
octeon_wdt_cpu_callback, does not handle events with CPU_TASKS_FROZEN
set in the action code.

Is this on purpose?  If so, then why is it done that way?

Thanks,
Richard

Re: watchdog: octeon: missing FROZEN hot plug notifier actions

[Guenter Roeck]

On 03/16/2016 01:51 PM, rcochran@linutronix.de wrote:
>
> In drivers/watchdog/octeon-wdt-main.c, the hot plug notifier callback,
> octeon_wdt_cpu_callback, does not handle events with CPU_TASKS_FROZEN
> set in the action code.
>
> Is this on purpose?  If so, then why is it done that way?
>

What do you expect it to do ?

My understanding is that CPU_TASKS_FROZEN is related to suspend/resume operations,
which are not really common for typical applications using Octeon CPUs. If they now
are, it might be worthwhile thinking about adding suspend/resume support to the driver.

Thanks,
Guenter

Re: watchdog: octeon: missing FROZEN hot plug notifier actions

[rcochran]

On Wed, Mar 16, 2016 at 08:18:22PM -0700, Guenter Roeck wrote:
> On 03/16/2016 01:51 PM, rcochran@linutronix.de wrote:
> >
> >In drivers/watchdog/octeon-wdt-main.c, the hot plug notifier callback,
> >octeon_wdt_cpu_callback, does not handle events with CPU_TASKS_FROZEN
> >set in the action code.
> >
> >Is this on purpose?  If so, then why is it done that way?
> >
> 
> What do you expect it to do ?

I would expect that the driver handles FROZEN in the same way,
enabling and disabling the watchdog interrupt on the hot plugged CPU.
It looks like an oversight that FROZEN actions are ignored, but I
wanted to be sure.
 
> My understanding is that CPU_TASKS_FROZEN is related to suspend/resume operations,
> which are not really common for typical applications using Octeon CPUs. If they now
> are, it might be worthwhile thinking about adding suspend/resume support to the driver.

Are they merely not common, or is suspend impossible on this machine?
Even if the use case is uncommon, still the driver should be correct.

Thanks,
Richard

Re: watchdog: octeon: missing FROZEN hot plug notifier actions

[Guenter Roeck]

On 03/17/2016 01:08 AM, rcochran@linutronix.de wrote:
> On Wed, Mar 16, 2016 at 08:18:22PM -0700, Guenter Roeck wrote:
>> On 03/16/2016 01:51 PM, rcochran@linutronix.de wrote:
>>>
>>> In drivers/watchdog/octeon-wdt-main.c, the hot plug notifier callback,
>>> octeon_wdt_cpu_callback, does not handle events with CPU_TASKS_FROZEN
>>> set in the action code.
>>>
>>> Is this on purpose?  If so, then why is it done that way?
>>>
>>
>> What do you expect it to do ?
>
> I would expect that the driver handles FROZEN in the same way,
> enabling and disabling the watchdog interrupt on the hot plugged CPU.
> It looks like an oversight that FROZEN actions are ignored, but I
> wanted to be sure.
>
>> My understanding is that CPU_TASKS_FROZEN is related to suspend/resume operations,
>> which are not really common for typical applications using Octeon CPUs. If they now
>> are, it might be worthwhile thinking about adding suspend/resume support to the driver.
>
> Are they merely not common, or is suspend impossible on this machine?
> Even if the use case is uncommon, still the driver should be correct.
>
The default Octeon configuration at least enables CONFIG_PM and CONFIG_PM_SLEEP,
so one can argue that it should be supported. If you submit a patch, would you
be able to test your changes ?

Thanks,
Guenter

Re: watchdog: octeon: missing FROZEN hot plug notifier actions

[rcochran]

On Thu, Mar 17, 2016 at 06:04:30AM -0700, Guenter Roeck wrote:
> The default Octeon configuration at least enables CONFIG_PM and CONFIG_PM_SLEEP,
> so one can argue that it should be supported. If you submit a patch, would you
> be able to test your changes ?

No, due to -ENODEV.


Sorry,
Richard