diff mbox series

drivers: thermal: tsens: respect thermal_device_mode in threshold irq reporting

Message ID 20220114031337.24741-1-benl@squareup.com
State Superseded
Headers show
Series drivers: thermal: tsens: respect thermal_device_mode in threshold irq reporting | expand

Commit Message

Benjamin Li Jan. 14, 2022, 3:13 a.m. UTC
'echo disabled > .../thermal_zoneX/mode' will disable the thermal core's
polling mechanism to check for threshold trips. However, tsens supports
an interrupt mechanism to receive notification of trips, implemented in
commit 634e11d5b450 ("drivers: thermal: tsens: Add interrupt support").
This is used sometimes to run performance test cases.

Currently the thermal zone mode that's set by userspace is not checked
before propagating threshold trip events from IRQs. Let's fix this to
restore the abilty to disable thermal throttling at runtime.

====================

Tested on MSM8939 running 5.16.0. This platform has 8 cores; the first
four thermal zones control cpu0-3 and the last zone is for the other four
CPUs together.

  for f in /sys/class/thermal/thermal_zone*; do
    echo "disabled" > $f/mode
    echo $f | paste - $f/type $f/mode
  done

/sys/class/thermal/thermal_zone0	cpu0-thermal	disabled
/sys/class/thermal/thermal_zone1	cpu1-thermal	disabled
/sys/class/thermal/thermal_zone2	cpu2-thermal	disabled
/sys/class/thermal/thermal_zone3	cpu3-thermal	disabled
/sys/class/thermal/thermal_zone4	cpu4567-thermal	disabled

With mitigation thresholds at 75 degC and load running, we can now cruise
past temp=75000 without CPU throttling kicking in.

  watch -n 1 "grep '' /sys/class/thermal/*/temp
      /sys/class/thermal/*/cur_state
      /sys/bus/cpu/devices/cpu*/cpufreq/cpuinfo_cur_freq"

/sys/class/thermal/thermal_zone0/temp:82000
/sys/class/thermal/thermal_zone1/temp:84000
/sys/class/thermal/thermal_zone2/temp:87000
/sys/class/thermal/thermal_zone3/temp:84000
/sys/class/thermal/thermal_zone4/temp:84000
/sys/class/thermal/cooling_device0/cur_state:0
/sys/class/thermal/cooling_device1/cur_state:0
/sys/bus/cpu/devices/cpu0/cpufreq/cpuinfo_cur_freq:1113600
/sys/bus/cpu/devices/cpu1/cpufreq/cpuinfo_cur_freq:1113600
/sys/bus/cpu/devices/cpu2/cpufreq/cpuinfo_cur_freq:1113600
/sys/bus/cpu/devices/cpu3/cpufreq/cpuinfo_cur_freq:1113600
/sys/bus/cpu/devices/cpu4/cpufreq/cpuinfo_cur_freq:800000
/sys/bus/cpu/devices/cpu5/cpufreq/cpuinfo_cur_freq:800000
/sys/bus/cpu/devices/cpu6/cpufreq/cpuinfo_cur_freq:800000
/sys/bus/cpu/devices/cpu7/cpufreq/cpuinfo_cur_freq:800000

Reported-by: Zac Crosby <zac@squareup.com>
Signed-off-by: Benjamin Li <benl@squareup.com>
---
 drivers/thermal/qcom/tsens.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

Comments

Bjorn Andersson Jan. 20, 2022, 12:33 a.m. UTC | #1
On Thu 13 Jan 19:13 PST 2022, Benjamin Li wrote:

> 'echo disabled > .../thermal_zoneX/mode' will disable the thermal core's
> polling mechanism to check for threshold trips. However, tsens supports
> an interrupt mechanism to receive notification of trips, implemented in
> commit 634e11d5b450 ("drivers: thermal: tsens: Add interrupt support").
> This is used sometimes to run performance test cases.
> 
> Currently the thermal zone mode that's set by userspace is not checked
> before propagating threshold trip events from IRQs. Let's fix this to
> restore the abilty to disable thermal throttling at runtime.
> 
> ====================
> 
> Tested on MSM8939 running 5.16.0. This platform has 8 cores; the first
> four thermal zones control cpu0-3 and the last zone is for the other four
> CPUs together.
> 
>   for f in /sys/class/thermal/thermal_zone*; do
>     echo "disabled" > $f/mode
>     echo $f | paste - $f/type $f/mode
>   done
> 
> /sys/class/thermal/thermal_zone0	cpu0-thermal	disabled
> /sys/class/thermal/thermal_zone1	cpu1-thermal	disabled
> /sys/class/thermal/thermal_zone2	cpu2-thermal	disabled
> /sys/class/thermal/thermal_zone3	cpu3-thermal	disabled
> /sys/class/thermal/thermal_zone4	cpu4567-thermal	disabled
> 
> With mitigation thresholds at 75 degC and load running, we can now cruise
> past temp=75000 without CPU throttling kicking in.
> 
>   watch -n 1 "grep '' /sys/class/thermal/*/temp
>       /sys/class/thermal/*/cur_state
>       /sys/bus/cpu/devices/cpu*/cpufreq/cpuinfo_cur_freq"
> 
> /sys/class/thermal/thermal_zone0/temp:82000
> /sys/class/thermal/thermal_zone1/temp:84000
> /sys/class/thermal/thermal_zone2/temp:87000
> /sys/class/thermal/thermal_zone3/temp:84000
> /sys/class/thermal/thermal_zone4/temp:84000
> /sys/class/thermal/cooling_device0/cur_state:0
> /sys/class/thermal/cooling_device1/cur_state:0
> /sys/bus/cpu/devices/cpu0/cpufreq/cpuinfo_cur_freq:1113600
> /sys/bus/cpu/devices/cpu1/cpufreq/cpuinfo_cur_freq:1113600
> /sys/bus/cpu/devices/cpu2/cpufreq/cpuinfo_cur_freq:1113600
> /sys/bus/cpu/devices/cpu3/cpufreq/cpuinfo_cur_freq:1113600
> /sys/bus/cpu/devices/cpu4/cpufreq/cpuinfo_cur_freq:800000
> /sys/bus/cpu/devices/cpu5/cpufreq/cpuinfo_cur_freq:800000
> /sys/bus/cpu/devices/cpu6/cpufreq/cpuinfo_cur_freq:800000
> /sys/bus/cpu/devices/cpu7/cpufreq/cpuinfo_cur_freq:800000
> 
> Reported-by: Zac Crosby <zac@squareup.com>
> Signed-off-by: Benjamin Li <benl@squareup.com>

Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>

Regards,
Bjorn

> ---
>  drivers/thermal/qcom/tsens.c | 15 +++++++++------
>  1 file changed, 9 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c
> index 99a8d9f3e03c..0b6299512e7c 100644
> --- a/drivers/thermal/qcom/tsens.c
> +++ b/drivers/thermal/qcom/tsens.c
> @@ -509,13 +509,16 @@ static irqreturn_t tsens_irq_thread(int irq, void *data)
>  		spin_unlock_irqrestore(&priv->ul_lock, flags);
>  
>  		if (trigger) {
> -			dev_dbg(priv->dev, "[%u] %s: TZ update trigger (%d mC)\n",
> -				hw_id, __func__, temp);
> -			thermal_zone_device_update(s->tzd,
> -						   THERMAL_EVENT_UNSPECIFIED);
> +			if (s->tzd->mode == THERMAL_DEVICE_ENABLED) {
> +				dev_dbg(priv->dev, "[%u] %s: TZ update trigger (%d mC)\n",
> +					hw_id, __func__, temp);
> +				thermal_zone_device_update(s->tzd, THERMAL_EVENT_UNSPECIFIED);
> +			} else {
> +				dev_dbg(priv->dev, "[%u] %s: TZ update trigger (%d mC) skipped as zone disabled\n",
> +					hw_id, __func__, temp);
> +			}
>  		} else {
> -			dev_dbg(priv->dev, "[%u] %s: no violation:  %d\n",
> -				hw_id, __func__, temp);
> +			dev_dbg(priv->dev, "[%u] %s: no violation:  %d\n", hw_id, __func__, temp);
>  		}
>  
>  		if (tsens_version(priv) < VER_0_1) {
> -- 
> 2.17.1
>
diff mbox series

Patch

diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c
index 99a8d9f3e03c..0b6299512e7c 100644
--- a/drivers/thermal/qcom/tsens.c
+++ b/drivers/thermal/qcom/tsens.c
@@ -509,13 +509,16 @@  static irqreturn_t tsens_irq_thread(int irq, void *data)
 		spin_unlock_irqrestore(&priv->ul_lock, flags);
 
 		if (trigger) {
-			dev_dbg(priv->dev, "[%u] %s: TZ update trigger (%d mC)\n",
-				hw_id, __func__, temp);
-			thermal_zone_device_update(s->tzd,
-						   THERMAL_EVENT_UNSPECIFIED);
+			if (s->tzd->mode == THERMAL_DEVICE_ENABLED) {
+				dev_dbg(priv->dev, "[%u] %s: TZ update trigger (%d mC)\n",
+					hw_id, __func__, temp);
+				thermal_zone_device_update(s->tzd, THERMAL_EVENT_UNSPECIFIED);
+			} else {
+				dev_dbg(priv->dev, "[%u] %s: TZ update trigger (%d mC) skipped as zone disabled\n",
+					hw_id, __func__, temp);
+			}
 		} else {
-			dev_dbg(priv->dev, "[%u] %s: no violation:  %d\n",
-				hw_id, __func__, temp);
+			dev_dbg(priv->dev, "[%u] %s: no violation:  %d\n", hw_id, __func__, temp);
 		}
 
 		if (tsens_version(priv) < VER_0_1) {