Re: [PATCH v6 2/2] thermal: core: Add a back up thermal shutdown mechanism

[Ravikumar]

On Tuesday 18 April 2017 09:59 AM, Keerthy wrote:
> orderly_poweroff is triggered when a graceful shutdown
> of system is desired. This may be used in many critical states of the
> kernel such as when subsystems detects conditions such as critical
> temperature conditions. However, in certain conditions in system
> boot up sequences like those in the middle of driver probes being
> initiated, userspace will be unable to power off the system in a clean
> manner and leaves the system in a critical state. In cases like these,
> the /sbin/poweroff will return success (having forked off to attempt
> powering off the system. However, the system overall will fail to
> completely poweroff (since other modules will be probed) and the system
> is still functional with no userspace (since that would have shut itself
> off).
>
> However, there is no clean way of detecting such failure of userspace
> powering off the system. In such scenarios, it is necessary for a backup
> workqueue to be able to force a shutdown of the system when orderly
> shutdown is not successful after a configurable time period.
Care to add testing information?
> Reported-by: Nishanth Menon <nm@xxxxxx>
> Signed-off-by: Keerthy <j-keerthy@xxxxxx>
> Acked-by: Eduardo Valentin <edubezval@xxxxxxxxx>
> ---
>
> Changes in v6:
>
>    * Rephrased Kconfig description as per Eduardo's feedback.
>    * Added check to verify positive values of delay in milli Seconds.
>
> Changes in v5:
>
>    * Mandated delay for thermal emergency poweroff to be a non-zero value.
>
> Changes in v4:
>
>    * Updated documentation
>    * changed emergency_poweroff_func to thermal_emergency_poweroff_func
>
> Changes in v3:
>
>    * Removed unnecessary mutex init.
>    * Added WARN messages instead of a simple warning message.
>    * Added Documentation.
>
>   Documentation/thermal/sysfs-api.txt | 21 +++++++++++++++
>   drivers/thermal/Kconfig             | 17 ++++++++++++
>   drivers/thermal/thermal_core.c      | 53 +++++++++++++++++++++++++++++++++++++
>   3 files changed, 91 insertions(+)
>
> diff --git a/Documentation/thermal/sysfs-api.txt b/Documentation/thermal/sysfs-api.txt
> index ef473dc..bb9a0a5 100644
> --- a/Documentation/thermal/sysfs-api.txt
> +++ b/Documentation/thermal/sysfs-api.txt
> @@ -582,3 +582,24 @@ platform data is provided, this uses the step_wise throttling policy.
>   This function serves as an arbitrator to set the state of a cooling
>   device. It sets the cooling device to the deepest cooling state if
>   possible.
> +
> +6. thermal_emergency_poweroff:
> +
Should this be in sysfs-api doc?
> +On an event of critical trip temperature crossing. Thermal framework
> +allows the system to shutdown gracefully by calling orderly_poweroff().
> +In the event of a failure of orderly_poweroff() to shut down the system
> +we are in danger of keeping the system alive at undesirably high
> +temperatures. To mitigate this high risk scenario we program a work
> +queue to fire after a pre-determined number of seconds to start
> +an emergency shutdown of the device using the kernel_power_off()
> +function. In case kernel_power_off() fails then finally
> +emergency_restart() is called in the worst case.
> +
> +The delay should be carefully profiled so as to give adequate time for
> +orderly_poweroff(). In case of failure of an orderly_poweroff() the
> +emergency poweroff kicks in after the delay has elapsed and shuts down
> +the system.
> +
In order to come up with an ideal delay, we need to strike a balance between
being paranoid vs being too late.
In a different patch, I tried to justify setting crit temp @120C by quoting
we need to give some time to orderly_poweroff()

So we got T = [3/temp change rate] seconds before the HW issues a reset.

within this T sec we need to give a chance to orderly_poweroff() and when it
fails, bring out the big weapons.

crumb: we might actually be increasing the "temp rate change" by doing a 
lot of IO
access for syncing.
Let us hope someone is trying to cool the system down while we are trying to
save the day..
> +If set to 0 emergency poweroff will not be supported. So a carefully
> +profiled non-zero positive value is a must for emergerncy poweroff to be
> +triggered.
Profiling should be done based on real data than emulation.
That's when we get to know if the memory and IOs listen to the SoC
when the lava is out.
> diff --git a/drivers/thermal/Kconfig b/drivers/thermal/Kconfig
> index 9347401..74bf92b 100644
> --- a/drivers/thermal/Kconfig
> +++ b/drivers/thermal/Kconfig
> @@ -15,6 +15,23 @@ menuconfig THERMAL
>   
>   if THERMAL
>   
> +config THERMAL_EMERGENCY_POWEROFF_DELAY_MS
> +	int "Emergency poweroff delay in milli-seconds"
> +	depends on THERMAL
> +	default 0
> +	help
> +	  Thermal subsystem will issue a graceful shutdown when
> +	  critical temperatures are reached using orderly_poweroff(). In
> +	  case of failure of an orderly_poweroff(), the thermal emergency
> +	  poweroff kicks in after a delay has elapsed and shuts down the system.
> +	  This config is number of milliseconds to delay before emergency
> +	  poweroff kicks in. Similarly to the critical trip point,
> +	  the delay should be carefully profiled so as to give adequate
> +	  time for orderly_poweroff() to finish on regular execution.
> +	  If set to 0 emergency poweroff will not be supported.
> +
> +	  In doubt, leave as 0.
> +
>   config THERMAL_HWMON
>   	bool
>   	prompt "Expose thermal sensors as hwmon device"
> diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
> index 8337c27..b21b9cc 100644
> --- a/drivers/thermal/thermal_core.c
> +++ b/drivers/thermal/thermal_core.c
> @@ -324,6 +324,54 @@ static void handle_non_critical_trips(struct thermal_zone_device *tz,
>   		       def_governor->throttle(tz, trip);
>   }
>   
> +/**
> + * thermal_emergency_poweroff_func - emergency poweroff work after a known delay
may needs to be re-phrased as this func itself can't handle the delay.
> + * @work: work_struct associated with the emergency poweroff function
> + *
> + * This function is called in very critical situations to force
> + * a kernel poweroff after a configurable timeout value.
> + */
> +static void thermal_emergency_poweroff_func(struct work_struct *work)
> +{
> +	/*
> +	 * We have reached here after the emergency thermal shutdown
> +	 * Waiting period has expired. This means orderly_poweroff has
> +	 * not been able to shut off the system for some reason.
> +	 * Try to shut down the system immediately using kernel_power_off
> +	 * if populated
> +	 */
> +	WARN(1, "Attempting kernel_power_off: Temperature too high\n");
> +	kernel_power_off();
> +
> +	/*
> +	 * Worst of the worst case trigger emergency restart
> +	 */
> +	WARN(1, "Attempting emergency_restart: Temperature too high\n");
> +	emergency_restart();
> +}
> +
> +static DECLARE_DELAYED_WORK(thermal_emergency_poweroff_work,
> +			    thermal_emergency_poweroff_func);
> +
> +/**
> + * thermal_emergency_poweroff - Trigger an emergency system poweroff
Here you may say after a pre-set delay.
> + *
> + * This may be called from any critical situation to trigger a system shutdown
> + * after a known period of time. By default this is not scheduled.
This will be called only on a critical temperature event, right?
> + */
> +void thermal_emergency_poweroff(void)
> +{
> +	int poweroff_delay_ms = CONFIG_THERMAL_EMERGENCY_POWEROFF_DELAY_MS;
> +	/*
> +	 * poweroff_delay_ms must be a carefully profiled positive value.
> +	 * Its a must for thermal_emergency_poweroff_work to be scheduled
typo %s/Its/It's/
> +	 */
> +	if (poweroff_delay_ms <= 0)
> +		return;
It may be helpful to provide hint before returning?
"Back up thermal emergency poweroff service is not enabled, set

CONFIG_THERMAL_EMERGENCY_POWEROFF_DELAY_MS to a carefully profiled value to enable this service"

> +	schedule_delayed_work(&thermal_emergency_poweroff_work,
> +			      msecs_to_jiffies(poweroff_delay_ms));
> +}
> +
>   static void handle_critical_trips(struct thermal_zone_device *tz,
>   				  int trip, enum thermal_trip_type trip_type)
>   {
> @@ -346,6 +394,11 @@ static void handle_critical_trips(struct thermal_zone_device *tz,
>   			  tz->temperature / 1000);
>   		mutex_lock(&poweroff_lock);
>   		if (!power_off_triggered) {
> +			/*
> +			 * Queue a backup emergency shutdown in the event of
> +			 * orderly_poweroff failure
> +			 */
> +			thermal_emergency_poweroff();
This comment is misleading because calling the api is not enough to set 
a backup.
>   			orderly_poweroff(true);
>   			power_off_triggered = true;
>   		}
Over all, much needed functionality. Thanks.

Regards,
RK