2017-04-12 09:13:43

by Keerthy

[permalink] [raw]
Subject: [PATCH] thermal: core: Allow orderly_poweroff to be called only once

thermal_zone_device_check --> thermal_zone_device_update -->
handle_thermal_trip --> handle_critical_trips --> orderly_poweroff

The above sequence happens every 250/500 mS based on the configuration.
The orderly_poweroff function is getting called every 250/500 mS.
With a full fledged file system it takes at least 5-10 Seconds to
power off gracefully.

In that period due to the thermal_zone_device_check triggering
periodically the thermal work queues bombard with
orderly_poweroff calls multiple times eventually leading to
failures in gracefully powering off the system.

Make sure that orderly_poweroff is called only once.

Fixes: 0c01ebbfd3caf1d ("Thermal: Remove throttling logic out of thermal_sys.c")
Reported-by: Nishanth Menon <[email protected]>
Signed-off-by: Keerthy <[email protected]>
---
drivers/thermal/thermal_core.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
index 11f0675..4e68ce0 100644
--- a/drivers/thermal/thermal_core.c
+++ b/drivers/thermal/thermal_core.c
@@ -326,6 +326,7 @@ static void handle_critical_trips(struct thermal_zone_device *tz,
int trip, enum thermal_trip_type trip_type)
{
int trip_temp;
+ static bool power_off_triggered;

tz->ops->get_trip_temp(tz, trip, &trip_temp);

@@ -338,11 +339,12 @@ static void handle_critical_trips(struct thermal_zone_device *tz,
if (tz->ops->notify)
tz->ops->notify(tz, trip, trip_type);

- if (trip_type == THERMAL_TRIP_CRITICAL) {
+ if (trip_type == THERMAL_TRIP_CRITICAL && !power_off_triggered) {
dev_emerg(&tz->device,
"critical temperature reached(%d C),shutting down\n",
tz->temperature / 1000);
orderly_poweroff(true);
+ power_off_triggered = true;
}
}

--
1.9.1


2017-04-13 05:06:48

by Eduardo Valentin

[permalink] [raw]
Subject: Re: [PATCH] thermal: core: Allow orderly_poweroff to be called only once

Hey,

On Wed, Apr 12, 2017 at 02:42:12PM +0530, Keerthy wrote:
> thermal_zone_device_check --> thermal_zone_device_update -->
> handle_thermal_trip --> handle_critical_trips --> orderly_poweroff
>
> The above sequence happens every 250/500 mS based on the configuration.
> The orderly_poweroff function is getting called every 250/500 mS.
> With a full fledged file system it takes at least 5-10 Seconds to
> power off gracefully.
>
> In that period due to the thermal_zone_device_check triggering
> periodically the thermal work queues bombard with
> orderly_poweroff calls multiple times eventually leading to
> failures in gracefully powering off the system.
>
> Make sure that orderly_poweroff is called only once.
>
> Fixes: 0c01ebbfd3caf1d ("Thermal: Remove throttling logic out of thermal_sys.c")

Why? how the above commit introduced this issue?

> Reported-by: Nishanth Menon <[email protected]>
> Signed-off-by: Keerthy <[email protected]>
> ---
> drivers/thermal/thermal_core.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
> index 11f0675..4e68ce0 100644
> --- a/drivers/thermal/thermal_core.c
> +++ b/drivers/thermal/thermal_core.c
> @@ -326,6 +326,7 @@ static void handle_critical_trips(struct thermal_zone_device *tz,
> int trip, enum thermal_trip_type trip_type)
> {
> int trip_temp;
> + static bool power_off_triggered;
>

I am not convinced this change does what you say it does..

> tz->ops->get_trip_temp(tz, trip, &trip_temp);
>
> @@ -338,11 +339,12 @@ static void handle_critical_trips(struct thermal_zone_device *tz,
> if (tz->ops->notify)
> tz->ops->notify(tz, trip, trip_type);
>
> - if (trip_type == THERMAL_TRIP_CRITICAL) {
> + if (trip_type == THERMAL_TRIP_CRITICAL && !power_off_triggered) {
> dev_emerg(&tz->device,
> "critical temperature reached(%d C),shutting down\n",
> tz->temperature / 1000);
> orderly_poweroff(true);

Imagine you have two zones about to enter critical path. thermal zone 0
first gets into critical, and at this line (btween orderly_poweroff()
and you set your flag), it gets preempted. Then thermal zone 1 also
enters the critical path, then happens to be scheduled, you essentially
end up still calling orderly_poweroff() twice, which is what this patch
is supposed to avoid.

You better simply use a lock to serialize code execution.

BR,

> + power_off_triggered = true;
> }
> }
>
> --
> 1.9.1
>


Attachments:
(No filename) (2.44 kB)
signature.asc (819.00 B)
Digital signature
Download all attachments

2017-04-13 05:17:24

by Keerthy

[permalink] [raw]
Subject: Re: [PATCH] thermal: core: Allow orderly_poweroff to be called only once



On Thursday 13 April 2017 10:36 AM, Eduardo Valentin wrote:
> Hey,
>
> On Wed, Apr 12, 2017 at 02:42:12PM +0530, Keerthy wrote:
>> thermal_zone_device_check --> thermal_zone_device_update -->
>> handle_thermal_trip --> handle_critical_trips --> orderly_poweroff
>>
>> The above sequence happens every 250/500 mS based on the configuration.
>> The orderly_poweroff function is getting called every 250/500 mS.
>> With a full fledged file system it takes at least 5-10 Seconds to
>> power off gracefully.
>>
>> In that period due to the thermal_zone_device_check triggering
>> periodically the thermal work queues bombard with
>> orderly_poweroff calls multiple times eventually leading to
>> failures in gracefully powering off the system.
>>
>> Make sure that orderly_poweroff is called only once.
>>
>> Fixes: 0c01ebbfd3caf1d ("Thermal: Remove throttling logic out of thermal_sys.c")
>
> Why? how the above commit introduced this issue?

git blame showed me that. I can remove this. Let me know if you have the
right commit that introduced this.

>
>> Reported-by: Nishanth Menon <[email protected]>
>> Signed-off-by: Keerthy <[email protected]>
>> ---
>> drivers/thermal/thermal_core.c | 4 +++-
>> 1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
>> index 11f0675..4e68ce0 100644
>> --- a/drivers/thermal/thermal_core.c
>> +++ b/drivers/thermal/thermal_core.c
>> @@ -326,6 +326,7 @@ static void handle_critical_trips(struct thermal_zone_device *tz,
>> int trip, enum thermal_trip_type trip_type)
>> {
>> int trip_temp;
>> + static bool power_off_triggered;
>>
>
> I am not convinced this change does what you say it does..
>
>> tz->ops->get_trip_temp(tz, trip, &trip_temp);
>>
>> @@ -338,11 +339,12 @@ static void handle_critical_trips(struct thermal_zone_device *tz,
>> if (tz->ops->notify)
>> tz->ops->notify(tz, trip, trip_type);
>>
>> - if (trip_type == THERMAL_TRIP_CRITICAL) {
>> + if (trip_type == THERMAL_TRIP_CRITICAL && !power_off_triggered) {
>> dev_emerg(&tz->device,
>> "critical temperature reached(%d C),shutting down\n",
>> tz->temperature / 1000);
>> orderly_poweroff(true);
>
> Imagine you have two zones about to enter critical path. thermal zone 0
> first gets into critical, and at this line (btween orderly_poweroff()
> and you set your flag), it gets preempted. Then thermal zone 1 also
> enters the critical path, then happens to be scheduled, you essentially
> end up still calling orderly_poweroff() twice, which is what this patch
> is supposed to avoid.
>
> You better simply use a lock to serialize code execution.

Okay understood. Will do it in the next version.

>
> BR,
>
>> + power_off_triggered = true;
>> }
>> }
>>
>> --
>> 1.9.1
>>