[PATCH v5 16/18] thermal/drivers/tegra: Remove unneeded lock when setting a trip point

Daniel Lezcano posted 18 patches 1 year, 8 months ago
[PATCH v5 16/18] thermal/drivers/tegra: Remove unneeded lock when setting a trip point
Posted by Daniel Lezcano 1 year, 8 months ago
The function tegra_tsensor_enable_hw_channel() takes the thermal zone
lock to prevent "a potential" race with a call to set_trips()
callback.

The driver must not play with the thermal framework core code
internals.

The tegra_tsensor_enable_hw_channel() is called by:

 - the suspend / resume callbacks
 - the probe function after the thermal zones are registered

The thermal zone lock taken in this function is supposed to protect
from a call to the set_trips() callback which writes in the same
register.

The potential race is when suspend / resume are called at the same
time as set_trips. This one is called only in
thermal_zone_device_update().

 - At suspend time, the 'in_suspend' is set, thus the
   thermal_zone_device_update() bails out immediately and set_trips is
   not called during this moment.

 - At resume time, the thermal zone is updated at PM_POST_SUSPEND,
   thus the driver has already set the TH2 temperature.

 - At probe time, we register the thermal zone and then we set the
   TH2. The only scenario I can see so far is the interrupt fires, the
   thermal_zone_update() is called exactly at the moment
   tegra_tsensor_enable_hw_channel() a few lines after registering it.

Enable the channels before setting up the interrupt. We close the
potential race window without using the thermal zone's lock.

Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Suggested-by: Thierry Reding <thierry.reding@gmail.com>
---
 drivers/thermal/tegra/tegra30-tsensor.c | 25 ++++++++++++++-----------
 1 file changed, 14 insertions(+), 11 deletions(-)

diff --git a/drivers/thermal/tegra/tegra30-tsensor.c b/drivers/thermal/tegra/tegra30-tsensor.c
index 4b2ea17910cd..cb584a5735ed 100644
--- a/drivers/thermal/tegra/tegra30-tsensor.c
+++ b/drivers/thermal/tegra/tegra30-tsensor.c
@@ -359,9 +359,6 @@ static int tegra_tsensor_enable_hw_channel(const struct tegra_tsensor *ts,
 
 	tegra_tsensor_get_hw_channel_trips(tzd, &hot_trip, &crit_trip);
 
-	/* prevent potential racing with tegra_tsensor_set_trips() */
-	mutex_lock(&tzd->lock);
-
 	dev_info_once(ts->dev, "ch%u: PMC emergency shutdown trip set to %dC\n",
 		      id, DIV_ROUND_CLOSEST(crit_trip, 1000));
 
@@ -404,8 +401,6 @@ static int tegra_tsensor_enable_hw_channel(const struct tegra_tsensor *ts,
 	val |= FIELD_PREP(TSENSOR_SENSOR0_CONFIG0_INTR_THERMAL_RST_EN, 1);
 	writel_relaxed(val, tsc->regs + TSENSOR_SENSOR0_CONFIG0);
 
-	mutex_unlock(&tzd->lock);
-
 	err = thermal_zone_device_enable(tzd);
 	if (err) {
 		dev_err(ts->dev, "ch%u: failed to enable zone: %d\n", id, err);
@@ -585,6 +580,20 @@ static int tegra_tsensor_probe(struct platform_device *pdev)
 			return err;
 	}
 
+	/*
+	 * Enable the channels before setting the interrupt so
+	 * set_trips() can not be called while we are setting up the
+	 * register TSENSOR_SENSOR0_CONFIG1. With this we close a
+	 * potential race window where we are setting up the TH2 and
+	 * the temperature hits TH1 resulting to an update of the
+	 * TSENSOR_SENSOR0_CONFIG1 register in the ISR.
+	 */
+	for (i = 0; i < ARRAY_SIZE(ts->ch); i++) {
+		err = tegra_tsensor_enable_hw_channel(ts, i);
+		if (err)
+			return err;
+	}
+
 	err = devm_request_threaded_irq(&pdev->dev, irq, NULL,
 					tegra_tsensor_isr, IRQF_ONESHOT,
 					"tegra_tsensor", ts);
@@ -592,12 +601,6 @@ static int tegra_tsensor_probe(struct platform_device *pdev)
 		return dev_err_probe(&pdev->dev, err,
 				     "failed to request interrupt\n");
 
-	for (i = 0; i < ARRAY_SIZE(ts->ch); i++) {
-		err = tegra_tsensor_enable_hw_channel(ts, i);
-		if (err)
-			return err;
-	}
-
 	return 0;
 }
 
-- 
2.34.1
Re: [PATCH v5 16/18] thermal/drivers/tegra: Remove unneeded lock when setting a trip point
Posted by Thierry Reding 1 year, 8 months ago
On Wed, Mar 01, 2023 at 09:14:44PM +0100, Daniel Lezcano wrote:
> The function tegra_tsensor_enable_hw_channel() takes the thermal zone
> lock to prevent "a potential" race with a call to set_trips()
> callback.
> 
> The driver must not play with the thermal framework core code
> internals.
> 
> The tegra_tsensor_enable_hw_channel() is called by:
> 
>  - the suspend / resume callbacks
>  - the probe function after the thermal zones are registered
> 
> The thermal zone lock taken in this function is supposed to protect
> from a call to the set_trips() callback which writes in the same
> register.
> 
> The potential race is when suspend / resume are called at the same
> time as set_trips. This one is called only in
> thermal_zone_device_update().
> 
>  - At suspend time, the 'in_suspend' is set, thus the
>    thermal_zone_device_update() bails out immediately and set_trips is
>    not called during this moment.
> 
>  - At resume time, the thermal zone is updated at PM_POST_SUSPEND,
>    thus the driver has already set the TH2 temperature.
> 
>  - At probe time, we register the thermal zone and then we set the
>    TH2. The only scenario I can see so far is the interrupt fires, the
>    thermal_zone_update() is called exactly at the moment
>    tegra_tsensor_enable_hw_channel() a few lines after registering it.
> 
> Enable the channels before setting up the interrupt. We close the
> potential race window without using the thermal zone's lock.
> 
> Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
> Suggested-by: Thierry Reding <thierry.reding@gmail.com>
> ---
>  drivers/thermal/tegra/tegra30-tsensor.c | 25 ++++++++++++++-----------
>  1 file changed, 14 insertions(+), 11 deletions(-)

Acked-by: Thierry Reding <treding@nvidia.com>
Re: [PATCH v5 16/18] thermal/drivers/tegra: Remove unneeded lock when setting a trip point
Posted by Daniel Lezcano 1 year, 8 months ago
On 02/03/2023 10:47, Thierry Reding wrote:
> On Wed, Mar 01, 2023 at 09:14:44PM +0100, Daniel Lezcano wrote:
>> The function tegra_tsensor_enable_hw_channel() takes the thermal zone
>> lock to prevent "a potential" race with a call to set_trips()
>> callback.
>>
>> The driver must not play with the thermal framework core code
>> internals.
>>
>> The tegra_tsensor_enable_hw_channel() is called by:
>>
>>   - the suspend / resume callbacks
>>   - the probe function after the thermal zones are registered
>>
>> The thermal zone lock taken in this function is supposed to protect
>> from a call to the set_trips() callback which writes in the same
>> register.
>>
>> The potential race is when suspend / resume are called at the same
>> time as set_trips. This one is called only in
>> thermal_zone_device_update().
>>
>>   - At suspend time, the 'in_suspend' is set, thus the
>>     thermal_zone_device_update() bails out immediately and set_trips is
>>     not called during this moment.
>>
>>   - At resume time, the thermal zone is updated at PM_POST_SUSPEND,
>>     thus the driver has already set the TH2 temperature.
>>
>>   - At probe time, we register the thermal zone and then we set the
>>     TH2. The only scenario I can see so far is the interrupt fires, the
>>     thermal_zone_update() is called exactly at the moment
>>     tegra_tsensor_enable_hw_channel() a few lines after registering it.
>>
>> Enable the channels before setting up the interrupt. We close the
>> potential race window without using the thermal zone's lock.
>>
>> Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
>> Suggested-by: Thierry Reding <thierry.reding@gmail.com>
>> ---
>>   drivers/thermal/tegra/tegra30-tsensor.c | 25 ++++++++++++++-----------
>>   1 file changed, 14 insertions(+), 11 deletions(-)
> 
> Acked-by: Thierry Reding <treding@nvidia.com>

Thanks for your review

May I ask you to have a look at patch 2 and 8 ?

Thanks

   -- Daniel

-- 
<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