thermal: core: Two fixes for 6.12
cover thermal: core: Two fixes for 6.12
Hi Everyone,
These patches address two thermal core issues that should better be taken care of in 6.12.
The first patch deals with the handling of polling delays. It could be 6.11-rc material even, but it may change the behavior somewhat and it's better to avoid regressing the kernel late in the cycle.
The other one adds some sanity checks for the temperature and hysteresis of writable trips, to prevent trip low temperature from falling below THERMAL_TEMP_INVALID due to an invalid temperature/hysteresis combination.
Thanks!
PATCH 1/2 thermal: core: Fix rounding of delay jiffies
From: Rafael J. Wysocki <[email protected]>
Using round_jiffies() in thermal_set_delay_jiffies() is invalid because its argument should be time in the future in absolute jiffies and it computes the result with respect to the current jiffies value at the invocation time. Fortunately, in the majority of cases it does not make any difference due to the time_is_after_jiffies() check in round_jiffies_common().
While using round_jiffies_relative() instead of round_jiffies() might reflect the intent a bit better, it still would not be defensible because that function should be called when the timer is about to be set and it is not suitable for pre-computation of delay values.
Accordingly, drop thermal_set_delay_jiffies() altogether, simply convert polling_delay and passive_delay to jiffies during thermal zone initialization and make thermal_zone_device_set_polling() call round_jiffies_relative() on the delay if it is greather than 1 second.
--- drivers/thermal/thermal_core.c | 23 ++++++++++------------- 1 file changed, 10 insertions(+), 13 deletions(-)
drivers/thermal/thermal_core.c +10 -13
@@ -323,11 +323,15 @@ static void thermal_zone_broken_disable(static void thermal_zone_device_set_polling (struct thermal_zone_device *tz,unsigned long delay) { -if (delay) -mod_delayed_work (system_freezable_power_efficient_wq, - &tz->poll_queue, delay); -else +if (!delay) {cancel_delayed_work (&tz->poll_queue); +return ; + } + +if (delay >HZ ) + delay =round_jiffies_relative (delay); + +mod_delayed_work (system_freezable_power_efficient_wq, &tz->poll_queue, delay); }static void thermal_zone_recheck (struct thermal_zone_device *tz,int error)
@@ -1312,13 +1316,6 @@ void thermal_cooling_device_unregister(s }EXPORT_SYMBOL_GPL (thermal_cooling_device_unregister); -static void thermal_set_delay_jiffies (unsigned long *delay_jiffies,int delay_ms) -{ - *delay_jiffies =msecs_to_jiffies (delay_ms); -if (delay_ms >1000 ) - *delay_jiffies =round_jiffies (*delay_jiffies); -} -int thermal_zone_get_crit_temp (struct thermal_zone_device *tz,int *temp) {const struct thermal_trip_desc *td;
@@ -1465,8 +1462,8 @@ thermal_zone_device_register_with_trips( td->threshold =INT_MAX ; } -thermal_set_delay_jiffies (&tz->passive_delay_jiffies, passive_delay); -thermal_set_delay_jiffies (&tz->polling_delay_jiffies, polling_delay); + tz->polling_delay_jiffies =msecs_to_jiffies (polling_delay); + tz->passive_delay_jiffies =msecs_to_jiffies (passive_delay); tz->recheck_delay_jiffies =THERMAL_RECHECK_DELAY ;/* sys I/F */
Re: [PATCH v1 1/2] thermal: core: Fix rounding of delay jiffies
Quoted 14 lines · Rafael J. Wysocki
From: Rafael J. Wysocki <[email protected]>
Using
round_jiffies()inthermal_set_delay_jiffies()is invalid because its argument should be time in the future in absolute jiffies and it computes the result with respect to the current jiffies value at the invocation time. Fortunately, in the majority of cases it does not make any difference due to thetime_is_after_jiffies()check inround_jiffies_common().While using
round_jiffies_relative()instead ofround_jiffies()might reflect the intent a bit better, it still would not be defensible because that function should be called when the timer is about to be set and it is not suitable for pre-computation of delay values.
Quoted 4 lines
Accordingly, drop
thermal_set_delay_jiffies()altogether, simply convertpolling_delayandpassive_delayto jiffies during thermal zone initialization and makethermal_zone_device_set_polling()callround_jiffies_relative()on the delay if it is greather than 1 second.
For the record:
In the history, the code was:
+ if (delay > 1000)
+ schedule_delayed_work(&(tz->poll_queue),
+ round_jiffies(msecs_to_jiffies(delay)));
+ else
+ schedule_delayed_work(&(tz->poll_queue),
+ msecs_to_jiffies(delay));
And the initial commit 21bc42ab852549f4a547d18d77e0e4d1b24ffd96:
"ACPI: thermal: use round_jiffies when thermal zone polling is enabled"
Good catch !
Quoted 59 lines
Fixes 17d399cd9c89 ("thermal/core: Precompute the delays from msecs to jiffies")Fixes e5f2cda61d06 ("thermal/core: Move thermal_set_delay_jiffies to static")Signed-off-by Rafael J. Wysocki <[email protected]>--- drivers/thermal/thermal_core.c | 23 ++++++++++------------- 1 file changed, 10 insertions(+), 13 deletions(-)Index:
linux-pm/drivers/thermal/thermal_core.c
===================================================================drivers/thermal/thermal_core.c +10
-13@@ -323,11 +323,15 @@ static void thermal_zone_broken_disable(static void thermal_zone_device_set_polling (struct thermal_zone_device *tz,unsigned long delay) { -if (delay) -mod_delayed_work (system_freezable_power_efficient_wq, - &tz->poll_queue, delay); -else +if (!delay) {cancel_delayed_work (&tz->poll_queue); +return ; + } + +if (delay >HZ ) + delay =round_jiffies_relative (delay); + +mod_delayed_work (system_freezable_power_efficient_wq, &tz->poll_queue, delay); }static void thermal_zone_recheck (struct thermal_zone_device *tz,int error)@@ -1312,13 +1316,6 @@ void thermal_cooling_device_unregister(s }EXPORT_SYMBOL_GPL (thermal_cooling_device_unregister); -static void thermal_set_delay_jiffies (unsigned long *delay_jiffies,int delay_ms) -{ - *delay_jiffies =msecs_to_jiffies (delay_ms); -if (delay_ms >1000 ) - *delay_jiffies =round_jiffies (*delay_jiffies); -} -int thermal_zone_get_crit_temp (struct thermal_zone_device *tz,int *temp) {const struct thermal_trip_desc *td;@@ -1465,8 +1462,8 @@ thermal_zone_device_register_with_trips( td->threshold =INT_MAX ; } -thermal_set_delay_jiffies (&tz->passive_delay_jiffies, passive_delay); -thermal_set_delay_jiffies (&tz->polling_delay_jiffies, polling_delay); + tz->polling_delay_jiffies =msecs_to_jiffies (polling_delay); + tz->passive_delay_jiffies =msecs_to_jiffies (passive_delay); tz->recheck_delay_jiffies =THERMAL_RECHECK_DELAY ;/* sys I/F */
Signature · 5 lines
<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
PATCH 2/2 thermal: sysfs: Add sanity checks for trip temperature and hysteresis
In reply to Rafael J. WysockiFrom: Rafael J. Wysocki <[email protected]>
Add sanity checks for new trip temperature and hysteresis values to trip_point_temp_store() and trip_point_hyst_store() to prevent trip point thresholds from falling below THERMAL_TEMP_INVALID.
However, still allow user space to pass THERMAL_TEMP_INVALID as the new trip temperature value to invalidate the trip if necessary.
--- drivers/thermal/thermal_sysfs.c | 38 ++++++++++++++++++++++++++------------ 1 file changed, 26 insertions(+), 12 deletions(-)
drivers/thermal/thermal_sysfs.c +26 -12
@@ -111,18 +111,25 @@ trip_point_temp_store(struct device *devmutex_lock (&tz->lock); -if (temp != trip->temperature) { -if (tz->ops.set_trip_temp) { - ret = tz->ops.set_trip_temp (tz, trip, temp); -if (ret) - goto unlock; - } +if (temp == trip->temperature) + goto unlock; -thermal_zone_set_trip_temp (tz, trip, temp); +if (temp !=THERMAL_TEMP_INVALID && + temp <= trip->hysteresis +THERMAL_TEMP_INVALID ) { + ret = -EINVAL ; + goto unlock; + } -__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); +if (tz->ops.set_trip_temp) { + ret = tz->ops.set_trip_temp (tz, trip, temp); +if (ret) + goto unlock; } +thermal_zone_set_trip_temp (tz, trip, temp); + +__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); + unlock:mutex_unlock (&tz->lock);
@@ -152,15 +159,22 @@ trip_point_hyst_store(struct device *devmutex_lock (&tz->lock); -if (hyst != trip->hysteresis) { -thermal_zone_set_trip_hyst (tz, trip, hyst); +if (hyst == trip->hysteresis) + goto unlock; -__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); +if (hyst +THERMAL_TEMP_INVALID >= trip->temperature) { + ret = -EINVAL ; + goto unlock; } +thermal_zone_set_trip_hyst (tz, trip, hyst); + +__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); + +unlock:mutex_unlock (&tz->lock); -return count; +return ret ? ret : count; } static ssize_t
Re: [PATCH v1 2/2] thermal: sysfs: Add sanity checks for trip temperature and hysteresis
Quoted 26 lines · Rafael J. Wysocki
From: Rafael J. Wysocki <[email protected]>
Add sanity checks for new trip temperature and hysteresis values to
trip_point_temp_store()andtrip_point_hyst_store()to prevent trip point thresholds from falling belowTHERMAL_TEMP_INVALID.However, still allow user space to pass
THERMAL_TEMP_INVALIDas the new trip temperature value to invalidate the trip if necessary.Fixes be0a3600aa1e ("thermal: sysfs: Rework the handling of trip point updates")Cc 6.8+ <[email protected]> # 6.8+Signed-off-by Rafael J. Wysocki <[email protected]>--- drivers/thermal/thermal_sysfs.c | 38 ++++++++++++++++++++++++++------------ 1 file changed, 26 insertions(+), 12 deletions(-)Index:
linux-pm/drivers/thermal/thermal_sysfs.c
===================================================================drivers/thermal/thermal_sysfs.c +0
-2@@ -111,18 +111,25 @@ trip_point_temp_store(struct device *devmutex_lock (&tz->lock); -if (temp != trip->temperature) { -if (tz->ops.set_trip_temp) {
drivers/thermal/thermal_sysfs.c trip_point_temp_store(struct device *dev · line 118
Quoted 10 lines
drivers/thermal/thermal_sysfs.c +4
-5- ret = tz->ops.set_trip_temp (tz, trip, temp); -if (ret) - goto unlock; - } +if (temp == trip->temperature) + goto unlock; -thermal_zone_set_trip_temp (tz, trip, temp); +if (temp !=THERMAL_TEMP_INVALID && + temp <= trip->hysteresis +THERMAL_TEMP_INVALID ) {
It seems to me the condition is hard to understand.
temp <= trip->hysteresis + THERMAL_TEMP_INVALID
==>
temp - trip->hysteresis <= THERMAL_TEMP_INVALID
Could be the test below simpler to understand ?
if (trip->hysteresis && temp - trip->hysteresis < THERMAL_TEMP_INVALID))
I think more sanity check may be needed also.
if (temp < THERMAL_TEMP_INVALID)
Quoted 48 lines
drivers/thermal/thermal_sysfs.c +22
-5+ ret = -EINVAL ; + goto unlock; + } -__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); +if (tz->ops.set_trip_temp) { + ret = tz->ops.set_trip_temp (tz, trip, temp); +if (ret) + goto unlock; } +thermal_zone_set_trip_temp (tz, trip, temp); + +__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); + unlock:mutex_unlock (&tz->lock);@@ -152,15 +159,22 @@ trip_point_hyst_store(struct device *devmutex_lock (&tz->lock); -if (hyst != trip->hysteresis) { -thermal_zone_set_trip_hyst (tz, trip, hyst); +if (hyst == trip->hysteresis) + goto unlock; -__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); +if (hyst +THERMAL_TEMP_INVALID >= trip->temperature) { + ret = -EINVAL ; + goto unlock; } +thermal_zone_set_trip_hyst (tz, trip, hyst); + +__thermal_zone_device_update (tz,THERMAL_TRIP_CHANGED ); + +unlock:mutex_unlock (&tz->lock); -return count; +return ret ? ret : count; } static ssize_t
Signature · 5 lines
<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
Rafael J. Wysocki
maintainerRe: [PATCH v1 2/2] thermal: sysfs: Add sanity checks for trip temperature and hysteresis
Quoted 35 lines · Daniel Lezcano
Quoted 33 lines · Rafael J. Wysocki
From: Rafael J. Wysocki <[email protected]>
Add sanity checks for new trip temperature and hysteresis values to
trip_point_temp_store()andtrip_point_hyst_store()to prevent trip point thresholds from falling belowTHERMAL_TEMP_INVALID.However, still allow user space to pass
THERMAL_TEMP_INVALIDas the new trip temperature value to invalidate the trip if necessary.Fixes be0a3600aa1e ("thermal: sysfs: Rework the handling of trip point updates")Cc 6.8+ <[email protected]> # 6.8+Signed-off-by Rafael J. Wysocki <[email protected]>--- drivers/thermal/thermal_sysfs.c | 38 ++++++++++++++++++++++++++------------ 1 file changed, 26 insertions(+), 12 deletions(-)Index:
linux-pm/drivers/thermal/thermal_sysfs.c
===================================================================drivers/thermal/thermal_sysfs.c +2
-6@@ -111,18 +111,25 @@ trip_point_temp_store(struct device *devmutex_lock (&tz->lock); -if (temp != trip->temperature) { -if (tz->ops.set_trip_temp) { - ret = tz->ops.set_trip_temp (tz, trip, temp); -if (ret) - goto unlock; - } +if (temp == trip->temperature) + goto unlock;
Quoted 5 lines
drivers/thermal/thermal_sysfs.c trip_point_temp_store(struct device *dev · line 118
Quoted 3 lines
drivers/thermal/thermal_sysfs.c +2
-1-thermal_zone_set_trip_temp (tz, trip, temp); +if (temp !=THERMAL_TEMP_INVALID && + temp <= trip->hysteresis +THERMAL_TEMP_INVALID ) {It seems to me the condition is hard to understand.
That's not the key consideration here though.
Quoted 2 lines
temp <= trip->hysteresis + THERMAL_TEMP_INVALID
This cannot overflow because trip->hysteresis is non-negative.
Quoted 4 lines
==>temp - trip->hysteresis <= THERMAL_TEMP_INVALID
But this can.
Quoted 10 lines
Could be the test below simpler to understand ?
if (trip->hysteresis && temp - trip->hysteresis < THERMAL_TEMP_INVALID))I think more sanity check may be needed also.
if (temp <
THERMAL_TEMP_INVALID)
With my version of the check above this is not necessary (unless I'm missing something}.
Quoted 3 lines
Quoted 3 lines
drivers/thermal/thermal_sysfs.c +3
-0+ ret = -EINVAL ; + goto unlock; + }
Rafael J. Wysocki
maintainerRe: [PATCH v1 2/2] thermal: sysfs: Add sanity checks for trip temperature and hysteresis
Quoted 49 lines · Rafael J. Wysocki
Quoted 35 lines · Daniel Lezcano
Quoted 33 lines · Rafael J. Wysocki
From: Rafael J. Wysocki <[email protected]>
Add sanity checks for new trip temperature and hysteresis values to
trip_point_temp_store()andtrip_point_hyst_store()to prevent trip point thresholds from falling belowTHERMAL_TEMP_INVALID.However, still allow user space to pass
THERMAL_TEMP_INVALIDas the new trip temperature value to invalidate the trip if necessary.Fixes be0a3600aa1e ("thermal: sysfs: Rework the handling of trip point updates")Cc 6.8+ <[email protected]> # 6.8+Signed-off-by Rafael J. Wysocki <[email protected]>--- drivers/thermal/thermal_sysfs.c | 38 ++++++++++++++++++++++++++------------ 1 file changed, 26 insertions(+), 12 deletions(-)Index:
linux-pm/drivers/thermal/thermal_sysfs.c
===================================================================drivers/thermal/thermal_sysfs.c +2
-6@@ -111,18 +111,25 @@ trip_point_temp_store(struct device *devmutex_lock (&tz->lock); -if (temp != trip->temperature) { -if (tz->ops.set_trip_temp) { - ret = tz->ops.set_trip_temp (tz, trip, temp); -if (ret) - goto unlock; - } +if (temp == trip->temperature) + goto unlock;Quoted 5 lines
drivers/thermal/thermal_sysfs.c trip_point_temp_store(struct device *dev · line 118
Quoted 3 lines
drivers/thermal/thermal_sysfs.c +2
-1-thermal_zone_set_trip_temp (tz, trip, temp); +if (temp !=THERMAL_TEMP_INVALID && + temp <= trip->hysteresis +THERMAL_TEMP_INVALID ) {It seems to me the condition is hard to understand.
That's not the key consideration here though.
Quoted 2 lines
temp <= trip->hysteresis + THERMAL_TEMP_INVALID
Quoted 8 lines
This cannot overflow because
trip->hysteresisis non-negative.Quoted 4 lines
==>temp - trip->hysteresis <= THERMAL_TEMP_INVALIDBut this can.
Well, I think I should add a comment there to point that out or people will try to "clean it up".
Also note that in the hysteresis case the condition can be
if (trip->temperature - hyst <= THERMAL_TEMP_INVALID) {
because trip->temperature is never below THERMAL_TEMP_INVALID there.
Moreover, setting the hysteresis when the temperature is THERMAL_TRIP_INVALID does not make much sense.
I'll send a v2.
Quoted 17 lines
Quoted 10 lines
Could be the test below simpler to understand ?
if (trip->hysteresis && temp - trip->hysteresis < THERMAL_TEMP_INVALID))I think more sanity check may be needed also.
if (temp <
THERMAL_TEMP_INVALID)With my version of the check above this is not necessary (unless I'm missing something}.
Quoted 3 lines
Quoted 3 lines
drivers/thermal/thermal_sysfs.c +3
-0+ ret = -EINVAL ; + goto unlock; + }