[PATCH v2] clocksource: arm_global_timer: fix suspend resume

[John Stultz]

On Fri, Nov 20, 2015 at 11:28 AM, Grygorii Strashko
<grygorii.strashko at ti.com> wrote:
> On 11/20/2015 09:09 PM, John Stultz wrote:
>> On Fri, Nov 20, 2015 at 10:35 AM, Grygorii Strashko
>> <grygorii.strashko at ti.com> wrote:
>>> Hi Santosh,
>>>
>>> On 11/20/2015 07:23 PM, santosh shilimkar wrote:
>>>> + Thomas, Marc
>>>>
>>>> On 11/20/2015 5:57 AM, Grygorii Strashko wrote:
>>>>> Now the System stall is observed on TI AM437x based board
>>>>> (am437x-gp-evm) during resuming from System suspend when ARM Global
>>>>> timer is selected as clocksource device - SysRq are working, but
>>>>> nothing else. The reason of stall is that ARM Global timer loses its
>>>>> contexts.
>>>>>
>>>>> The reason of stall is that ARM Global timer loses its contexts during
>>>>> System suspend:
>>>>>      GT_CONTROL.TIMER_ENABLE = 0 (unbanked)
>>>>>      GT_COUNTERx = 0
>>>>>
>>>>> Hence, update ARM Global timer driver to reflect above behaviour
>>>>> - re-enable ARM Global timer on resume GT_CONTROL.TIMER_ENABLE = 1
>>>>> - ensure clocksource and clockevent devices have coresponding flags
>>>>>     (CLOCK_SOURCE_SUSPEND_NONSTOP and CLOCK_EVT_FEAT_C3STOP) set
>>>>>     depending on presence of "always-on" DT property.
>>>>>
>>>> Something which loses context in low power states can't be
>>>> called "always-on"
>>>
>>> Sry, it's kinda new area for me and I could make mistakes.
>>>
>>> While working on this patch I've:
>>>   - re-used implementation from ARM arch timer
>>> commit 82a5619410d4c4df65c04272db198eca5a867c18
>>> Author: Lorenzo Pieralisi <lorenzo.pieralisi at arm.com>
>>> Date:   Tue Apr 8 10:04:32 2014 +0100
>>>
>>>      clocksource: arch_arm_timer: Fix age-old arch timer C3STOP detection issue
>>>
>>>
>>> - and followed timekeeping.txt:
>>> "Typically the clock source is a monotonic, atomic counter which will provide
>>>   n bits which count from 0 to 2^(n-1) and then wraps around to 0 and start over.
>>> It will ideally NEVER stop ticking as long as the system is running. It
>>> may stop during system suspend."
>>> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>>
>>> And that exactly my use-case: I'd like to use ARM GT as clocksource
>>> and with CPUIdle = n. But System suspend has to be allowed.
>>>
>>>
>>>>
>>>> Also if the clock-soucre can't tick in the low power states
>>>> then that device shouldn't be used as a clock-source.
>>>
>>> Agree. clocksource can't (except with suspend). Have I missed something?
>>
>> I think the point Stantosh is making is that if the clocksource stops
>> in suspend (which is allowed) you should not be setting
>> CLOCK_SOURCE_SUSPEND_NONSTOP (which promises the clocksource doesn't
>> stop in suspend, so it can be used for suspend timing as well).
>>
>
> Ok. Thanks. I got it now. Adding CLOCK_SOURCE_SUSPEND_NONSTOP is mistake.
>
>> The contradictory part in your patch is that you're also setting the
>> clockevent logic as  CLOCK_EVT_FEAT_C3STOP, which flags that the
>> clockevent hardware might stop in low-power idle states (ie: not
>> suspend, but while the system is running).  Usually hardware that
>> halts in low-power mode (like the apic on some x86 machines) is not
>> also used for timekeeping (instead we use the hpet/acpi there).
>
> Sry, I've set CLOCK_EVT_FEAT_C3STOP if "always-on" = *false*

You might also consider renaming that value from always_on to
something more descriptive, given the subtlety of the different states
here. Maybe instead use a flag called halts_in_idle or something?

thanks
-john