Re: [RFC 1/2] PM / suspend: Add platform_suspend_target_state()

[Florian Fainelli <f.fainelli@gmail.com>]

On 06/29/2017 04:00 PM, Rafael J. Wysocki wrote:
> On Thursday, June 22, 2017 06:08:36 PM Florian Fainelli wrote:
>> Add an optional platform_suspend_ops callback: target_state, and a
>> helper function globally visible to get this called:
>> platform_suspend_target_state().
>>
>> This is useful for platform specific drivers that may need to take a
>> slightly different suspend/resume path based on the system's
>> suspend/resume state being entered.
>>
>> Although this callback is optional and documented as such, it requires
>> a platform_suspend_ops::begin callback to be implemented in order to
>> provide an accurate suspend/resume state within the driver that
>> implements this platform_suspend_ops.
>>
>> Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
>> ---
>>  include/linux/suspend.h | 12 ++++++++++++
>>  kernel/power/suspend.c  | 15 +++++++++++++++
>>  2 files changed, 27 insertions(+)
>>
>> diff --git a/include/linux/suspend.h b/include/linux/suspend.h
>> index d9718378a8be..d998a04a90a2 100644
>> --- a/include/linux/suspend.h
>> +++ b/include/linux/suspend.h
>> @@ -172,6 +172,15 @@ static inline void dpm_save_failed_step(enum suspend_stat_step step)
>>   *	Called by the PM core if the suspending of devices fails.
>>   *	This callback is optional and should only be implemented by platforms
>>   *	which require special recovery actions in that situation.
>> + *
>> + * @target_state: Returns the suspend state the suspend_ops will be entering.
>> + * 	Called by device drivers that need to know the platform specific suspend
>> + * 	state the system is about to enter.
>> + * 	This callback is optional and should only be implemented by platforms
>> + * 	which require special handling of power management states within
>> + * 	drivers. It does require @begin to be implemented to provide the suspend
>> + * 	state. Return value is platform_suspend_ops specific, and may be a 1:1
>> + * 	mapping to suspend_state_t when relevant.
>>   */
>>  struct platform_suspend_ops {
>>  	int (*valid)(suspend_state_t state);
>> @@ -184,6 +193,7 @@ struct platform_suspend_ops {
>>  	bool (*suspend_again)(void);
>>  	void (*end)(void);
>>  	void (*recover)(void);
>> +	int (*target_state)(void);
> 
> I would use unsigned int (the sign should not matter).
> 
>>  };
> 
> That's almost what I was thinking about except that the values returned by
> ->target_state should be unique, so it would be good to do something to
> ensure that.
> 
> The concern is as follows.
> 
> Say you have a driver develped for platform X where ->target_state returns
> A for "mem" and B for "standby".  Then, the same IP is re-used on platform Y
> returning B for "mem" and C for "standby" and now the driver cannot
> distinguish between them.
> 
> Moreover, even if they both returned A for "mem" there might be differences
> in how "mem" was defined by each of them and therefore in what the driver was
> expected to do to handle "mem" on X and Y.

That makes sense, would you need the core implementation in
platform_suspend_target_state() to range check what
suspend_ops->target_state() returns against a set of reserved value say,
checking from 0 up to ACPI_S_STATE_COUNT or is there another range you
would like to see being used?

Thanks!
-- 
Florian