We are seeing thermal shutdown on Intel based mobile workstations, the
shutdown happens during the first trip handle in
thermal_zone_device_register():
kernel: thermal thermal_zone15: critical temperature reached (101 C), shutting down
However, we shouldn't do a thermal shutdown here, since
1) We may want to use a dedicated daemon, Intel's thermald in this case,
to handle thermal shutdown.
2) For ACPI based system, _CRT doesn't mean shutdown unless it's inside
ThermalZone. ACPI Spec, 11.4.4 _CRT (Critical Temperature):
"... If this object it present under a device, the device’s driver
evaluates this object to determine the device’s critical cooling
temperature trip point. This value may then be used by the device’s
driver to program an internal device temperature sensor trip point."
So a "critical trip" here merely means we should take a more aggressive
cooling method.
So add an indication to let thermal core know it should leave thermal
device to userspace to handle.
Signed-off-by: Kai-Heng Feng <[email protected]>
---
drivers/thermal/thermal_core.c | 3 +++
include/linux/thermal.h | 2 ++
2 files changed, 5 insertions(+)
diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
index c6d74bc1c90b..6561e3767529 100644
--- a/drivers/thermal/thermal_core.c
+++ b/drivers/thermal/thermal_core.c
@@ -1477,6 +1477,9 @@ thermal_zone_device_register(const char *type, int trips, int mask,
goto unregister;
}
+ if (tz->tzp && tz->tzp->userspace)
+ thermal_zone_device_disable(tz);
+
mutex_lock(&thermal_list_lock);
list_add_tail(&tz->node, &thermal_tz_list);
mutex_unlock(&thermal_list_lock);
diff --git a/include/linux/thermal.h b/include/linux/thermal.h
index d07ea27e72a9..e8e8fac78fc8 100644
--- a/include/linux/thermal.h
+++ b/include/linux/thermal.h
@@ -247,6 +247,8 @@ struct thermal_zone_params {
*/
bool no_hwmon;
+ bool userspace;
+
int num_tbps; /* Number of tbp entries */
struct thermal_bind_params *tbp;
--
2.29.2
The device isn't present under ACPI ThermalZone, and there's a dedicated
userspace daemon for this thermal device.
Let thermal core know it shouldn't handle trips to avoid surprising
thermal shutdown.
Signed-off-by: Kai-Heng Feng <[email protected]>
---
drivers/thermal/intel/intel_pch_thermal.c | 11 +++++------
1 file changed, 5 insertions(+), 6 deletions(-)
diff --git a/drivers/thermal/intel/intel_pch_thermal.c b/drivers/thermal/intel/intel_pch_thermal.c
index 3b813ebb6ca1..e55e6318d733 100644
--- a/drivers/thermal/intel/intel_pch_thermal.c
+++ b/drivers/thermal/intel/intel_pch_thermal.c
@@ -270,6 +270,10 @@ static struct thermal_zone_device_ops tzd_ops = {
.get_trip_temp = pch_get_trip_temp,
};
+static struct thermal_zone_params tzd_params = {
+ .userspace = true,
+};
+
enum board_ids {
board_hsw,
board_wpt,
@@ -346,21 +350,16 @@ static int intel_pch_thermal_probe(struct pci_dev *pdev,
goto error_cleanup;
ptd->tzd = thermal_zone_device_register(bi->name, nr_trips, 0, ptd,
- &tzd_ops, NULL, 0, 0);
+ &tzd_ops, &tzd_params, 0, 0);
if (IS_ERR(ptd->tzd)) {
dev_err(&pdev->dev, "Failed to register thermal zone %s\n",
bi->name);
err = PTR_ERR(ptd->tzd);
goto error_cleanup;
}
- err = thermal_zone_device_enable(ptd->tzd);
- if (err)
- goto err_unregister;
return 0;
-err_unregister:
- thermal_zone_device_unregister(ptd->tzd);
error_cleanup:
iounmap(ptd->hw_base);
error_release:
--
2.29.2
The device isn't present under ACPI ThermalZone, and there's a dedicated
userspace daemon for this thermal device.
Let thermal core know it shouldn't handle trips to avoid surprising
thermal shutdown.
Signed-off-by: Kai-Heng Feng <[email protected]>
---
drivers/thermal/intel/int340x_thermal/int3400_thermal.c | 1 +
.../thermal/intel/int340x_thermal/int340x_thermal_zone.c | 6 +-----
2 files changed, 2 insertions(+), 5 deletions(-)
diff --git a/drivers/thermal/intel/int340x_thermal/int3400_thermal.c b/drivers/thermal/intel/int340x_thermal/int3400_thermal.c
index 0966551cbaaa..2002bc96eb3c 100644
--- a/drivers/thermal/intel/int340x_thermal/int3400_thermal.c
+++ b/drivers/thermal/intel/int340x_thermal/int3400_thermal.c
@@ -439,6 +439,7 @@ static struct thermal_zone_device_ops int3400_thermal_ops = {
static struct thermal_zone_params int3400_thermal_params = {
.governor_name = "user_space",
.no_hwmon = true,
+ .userspace = true,
};
static void int3400_setup_gddv(struct int3400_thermal_priv *priv)
diff --git a/drivers/thermal/intel/int340x_thermal/int340x_thermal_zone.c b/drivers/thermal/intel/int340x_thermal/int340x_thermal_zone.c
index 6e479deff76b..a103eb42ef2d 100644
--- a/drivers/thermal/intel/int340x_thermal/int340x_thermal_zone.c
+++ b/drivers/thermal/intel/int340x_thermal/int340x_thermal_zone.c
@@ -208,6 +208,7 @@ EXPORT_SYMBOL_GPL(int340x_thermal_read_trips);
static struct thermal_zone_params int340x_thermal_params = {
.governor_name = "user_space",
.no_hwmon = true,
+ .userspace = true,
};
struct int34x_thermal_zone *int340x_thermal_zone_add(struct acpi_device *adev,
@@ -259,14 +260,9 @@ struct int34x_thermal_zone *int340x_thermal_zone_add(struct acpi_device *adev,
ret = PTR_ERR(int34x_thermal_zone->zone);
goto err_thermal_zone;
}
- ret = thermal_zone_device_enable(int34x_thermal_zone->zone);
- if (ret)
- goto err_enable;
return int34x_thermal_zone;
-err_enable:
- thermal_zone_device_unregister(int34x_thermal_zone->zone);
err_thermal_zone:
acpi_lpat_free_conversion_table(int34x_thermal_zone->lpat_table);
kfree(int34x_thermal_zone->aux_trips);
--
2.29.2
On Tue, 2020-12-01 at 02:22 +0800, Kai-Heng Feng wrote:
> > On Dec 1, 2020, at 02:13, Srinivas Pandruvada <
> > [email protected]> wrote:
>
> [snipped]
>
> > > > What about creating an new callback
> > > >
> > > > enum thermal_trip_status {
> > > > THERMAL_TRIP_DISABLED = 0,
> > > > THERMAL_TRIP_ENABLED,
> > > > };
> > > >
> > > > int get_trip_status(struct thermal_zone_device *, int trip,
> > > > enum
> > > > thermal_trip_status *state);
> > > >
> > > > Then in
> > > > static void handle_thermal_trip(struct thermal_zone_device *tz,
> > > > int
> > > > trip)
> > > > {
> > > >
> > > > /* before tz->ops->get_trip_temp(tz, trip, &trip_temp); */
> > > > if (tz->ops->get_trip_status) {
> > > > enum thermal_trip_status *status;
> > > >
> > > > if (!tz->ops->get_trip_status(tz, trip, &status)) {
> > > > if (status == THERMAL_TRIP_DISABLED)
> > > > return;
> > > > }
> > > > }
> > > > ...
> > > > ...
> > > >
> > > > }
> > > >
> > > >
> > > > This callback will help the cases:
> > > > - Allows drivers to selectively disable certain trips during
> > > > init
> > > > state
> > > > or system resume where there can be spikes or always. int340x
> > > > drivers
> > > > can disable always.
> > >
> > > This sounds really great. This is indeed can happen on system
> > > resume,
> > > before userspace process thaw.
> > >
> > > > - Still give options for drivers to handle critical trip even
> > > > if
> > > > they
> > > > are bound to user space governors. User space process may be
> > > > dead,
> > > > so
> > > > still allow kernel to process graceful shutdown
> > >
> > > To make the scenario happen, do we need a new sysfs to let
> > > usespace
> > > enable it with THERMAL_TRIP_ENABLED?
> > This should be drivers call not user space.
>
> Understood. So after thermal_zone_device_register(), the driver can
> decide to what to return on get_trip_temp().
get_trip_status()
> Let me work on a new patch if there's no other concern.
Better to wait for confirmation from Daniel and others.
Thanks,
Srinivas
>
> Kai-Heng
>
> > Thanks,
> > Srinivas
> On Dec 1, 2020, at 02:39, Srinivas Pandruvada <[email protected]> wrote:
>
> On Tue, 2020-12-01 at 02:22 +0800, Kai-Heng Feng wrote:
>>> On Dec 1, 2020, at 02:13, Srinivas Pandruvada <
>>> [email protected]> wrote:
>>
>> [snipped]
>>
>>>>> What about creating an new callback
>>>>>
>>>>> enum thermal_trip_status {
>>>>> THERMAL_TRIP_DISABLED = 0,
>>>>> THERMAL_TRIP_ENABLED,
>>>>> };
>>>>>
>>>>> int get_trip_status(struct thermal_zone_device *, int trip,
>>>>> enum
>>>>> thermal_trip_status *state);
>>>>>
>>>>> Then in
>>>>> static void handle_thermal_trip(struct thermal_zone_device *tz,
>>>>> int
>>>>> trip)
>>>>> {
>>>>>
>>>>> /* before tz->ops->get_trip_temp(tz, trip, &trip_temp); */
>>>>> if (tz->ops->get_trip_status) {
>>>>> enum thermal_trip_status *status;
>>>>>
>>>>> if (!tz->ops->get_trip_status(tz, trip, &status)) {
>>>>> if (status == THERMAL_TRIP_DISABLED)
>>>>> return;
>>>>> }
>>>>> }
>>>>> ...
>>>>> ...
>>>>>
>>>>> }
>>>>>
>>>>>
>>>>> This callback will help the cases:
>>>>> - Allows drivers to selectively disable certain trips during
>>>>> init
>>>>> state
>>>>> or system resume where there can be spikes or always. int340x
>>>>> drivers
>>>>> can disable always.
>>>>
>>>> This sounds really great. This is indeed can happen on system
>>>> resume,
>>>> before userspace process thaw.
>>>>
>>>>> - Still give options for drivers to handle critical trip even
>>>>> if
>>>>> they
>>>>> are bound to user space governors. User space process may be
>>>>> dead,
>>>>> so
>>>>> still allow kernel to process graceful shutdown
>>>>
>>>> To make the scenario happen, do we need a new sysfs to let
>>>> usespace
>>>> enable it with THERMAL_TRIP_ENABLED?
>>> This should be drivers call not user space.
>>
>> Understood. So after thermal_zone_device_register(), the driver can
>> decide to what to return on get_trip_temp().
> get_trip_status()
>
>> Let me work on a new patch if there's no other concern.
> Better to wait for confirmation from Daniel and others.
Daniel,
Do you like Srinivas' proposed solution?
I hope we can find a solution in upstream kernel soon.
Kai-Heng
>
> Thanks,
> Srinivas
>
>>
>> Kai-Heng
>>
>>> Thanks,
>>> Srinivas
On 07/12/2020 06:36, Kai-Heng Feng wrote:
>
>
>> On Dec 1, 2020, at 02:39, Srinivas Pandruvada <[email protected]> wrote:
>>
>> On Tue, 2020-12-01 at 02:22 +0800, Kai-Heng Feng wrote:
>>>> On Dec 1, 2020, at 02:13, Srinivas Pandruvada <
>>>> [email protected]> wrote:
>>>
>>> [snipped]
>>>
>>>>>> What about creating an new callback
>>>>>>
>>>>>> enum thermal_trip_status {
>>>>>> THERMAL_TRIP_DISABLED = 0,
>>>>>> THERMAL_TRIP_ENABLED,
>>>>>> };
>>>>>>
>>>>>> int get_trip_status(struct thermal_zone_device *, int trip,
>>>>>> enum
>>>>>> thermal_trip_status *state);
>>>>>>
>>>>>> Then in
>>>>>> static void handle_thermal_trip(struct thermal_zone_device *tz,
>>>>>> int
>>>>>> trip)
>>>>>> {
>>>>>>
>>>>>> /* before tz->ops->get_trip_temp(tz, trip, &trip_temp); */
>>>>>> if (tz->ops->get_trip_status) {
>>>>>> enum thermal_trip_status *status;
>>>>>>
>>>>>> if (!tz->ops->get_trip_status(tz, trip, &status)) {
>>>>>> if (status == THERMAL_TRIP_DISABLED)
>>>>>> return;
>>>>>> }
>>>>>> }
>>>>>> ...
>>>>>> ...
>>>>>>
>>>>>> }
>>>>>>
>>>>>>
>>>>>> This callback will help the cases:
>>>>>> - Allows drivers to selectively disable certain trips during
>>>>>> init
>>>>>> state
>>>>>> or system resume where there can be spikes or always. int340x
>>>>>> drivers
>>>>>> can disable always.
>>>>>
>>>>> This sounds really great. This is indeed can happen on system
>>>>> resume,
>>>>> before userspace process thaw.
>>>>>
>>>>>> - Still give options for drivers to handle critical trip even
>>>>>> if
>>>>>> they
>>>>>> are bound to user space governors. User space process may be
>>>>>> dead,
>>>>>> so
>>>>>> still allow kernel to process graceful shutdown
>>>>>
>>>>> To make the scenario happen, do we need a new sysfs to let
>>>>> usespace
>>>>> enable it with THERMAL_TRIP_ENABLED?
>>>> This should be drivers call not user space.
>>>
>>> Understood. So after thermal_zone_device_register(), the driver can
>>> decide to what to return on get_trip_temp().
>> get_trip_status()
>>
>>> Let me work on a new patch if there's no other concern.
>> Better to wait for confirmation from Daniel and others.
>
> Daniel,
>
> Do you like Srinivas' proposed solution?
>
> I hope we can find a solution in upstream kernel soon.
(just trying to figure out the full context)
If the device is enumerated outside of a thermal zone, the sensor should
not register in the thermal zone no ?
--
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs
Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
On Wed, 2020-12-09 at 10:30 +0100, Daniel Lezcano wrote:
> On 07/12/2020 06:36, Kai-Heng Feng wrote:
> >
> > > On Dec 1, 2020, at 02:39, Srinivas Pandruvada <
> > > [email protected]> wrote:
> > >
> > > On Tue, 2020-12-01 at 02:22 +0800, Kai-Heng Feng wrote:
> > > > > On Dec 1, 2020, at 02:13, Srinivas Pandruvada <
> > > > > [email protected]> wrote:
> > > >
> > > > [snipped]
> > > >
> > > > > > > What about creating an new callback
> > > > > > >
> > > > > > > enum thermal_trip_status {
> > > > > > > THERMAL_TRIP_DISABLED = 0,
> > > > > > > THERMAL_TRIP_ENABLED,
> > > > > > > };
> > > > > > >
> > > > > > > int get_trip_status(struct thermal_zone_device *, int
> > > > > > > trip,
> > > > > > > enum
> > > > > > > thermal_trip_status *state);
> > > > > > >
> > > > > > > Then in
> > > > > > > static void handle_thermal_trip(struct
> > > > > > > thermal_zone_device *tz,
> > > > > > > int
> > > > > > > trip)
> > > > > > > {
> > > > > > >
> > > > > > > /* before tz->ops->get_trip_temp(tz, trip, &trip_temp);
> > > > > > > */
> > > > > > > if (tz->ops->get_trip_status) {
> > > > > > > enum thermal_trip_status *status;
> > > > > > >
> > > > > > > if (!tz->ops->get_trip_status(tz, trip, &status)) {
> > > > > > > if (status == THERMAL_TRIP_DISABLED)
> > > > > > > return;
> > > > > > > }
> > > > > > > }
> > > > > > > ...
> > > > > > > ...
> > > > > > >
> > > > > > > }
> > > > > > >
> > > > > > >
> > > > > > > This callback will help the cases:
> > > > > > > - Allows drivers to selectively disable certain trips
> > > > > > > during
> > > > > > > init
> > > > > > > state
> > > > > > > or system resume where there can be spikes or always.
> > > > > > > int340x
> > > > > > > drivers
> > > > > > > can disable always.
> > > > > >
> > > > > > This sounds really great. This is indeed can happen on
> > > > > > system
> > > > > > resume,
> > > > > > before userspace process thaw.
> > > > > >
> > > > > > > - Still give options for drivers to handle critical trip
> > > > > > > even
> > > > > > > if
> > > > > > > they
> > > > > > > are bound to user space governors. User space process may
> > > > > > > be
> > > > > > > dead,
> > > > > > > so
> > > > > > > still allow kernel to process graceful shutdown
> > > > > >
> > > > > > To make the scenario happen, do we need a new sysfs to let
> > > > > > usespace
> > > > > > enable it with THERMAL_TRIP_ENABLED?
> > > > > This should be drivers call not user space.
> > > >
> > > > Understood. So after thermal_zone_device_register(), the driver
> > > > can
> > > > decide to what to return on get_trip_temp().
> > > get_trip_status()
> > >
> > > > Let me work on a new patch if there's no other concern.
> > > Better to wait for confirmation from Daniel and others.
> >
> > Daniel,
> >
> > Do you like Srinivas' proposed solution?
> >
> > I hope we can find a solution in upstream kernel soon.
>
> (just trying to figure out the full context)
>
> If the device is enumerated outside of a thermal zone, the sensor
> should
> not register in the thermal zone no ?
Other trips are fine, so sensor registry is still required for passive
and active control. We just need to ignore critical trip. These table
are tested by OEM on Windows, where critical trip doesn't result in
immediate shutdown.
Thanks,
Srinivas
>
>
>
On Sun, Nov 29, 2020 at 9:36 PM Kai-Heng Feng
<[email protected]> wrote:
>
> We are seeing thermal shutdown on Intel based mobile workstations, the
> shutdown happens during the first trip handle in
> thermal_zone_device_register():
> kernel: thermal thermal_zone15: critical temperature reached (101 C), shutting down
Is the temperature reported by the thermal zone actually correct here?
101 C seems extremely excessive.
On Tue, Dec 15, 2020 at 2:22 AM Matthew Garrett <[email protected]> wrote:
>
> On Sun, Nov 29, 2020 at 9:36 PM Kai-Heng Feng
> <[email protected]> wrote:
> >
> > We are seeing thermal shutdown on Intel based mobile workstations, the
> > shutdown happens during the first trip handle in
> > thermal_zone_device_register():
> > kernel: thermal thermal_zone15: critical temperature reached (101 C), shutting down
>
> Is the temperature reported by the thermal zone actually correct here?
> 101 C seems extremely excessive.
According to ODM/OEM, it's correct.
It's a short period when Intel Turbo Boost kicks in.
Kai-Heng