| Message ID | 20220120200153.1214-1-benl@squareup.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v3] drivers: thermal: tsens: respect thermal_device_mode in threshold irq reporting | expand |
On 20/01/2022 21:01, Benjamin Li wrote: > 'echo disabled > .../thermal_zoneX/mode' will disable the thermal core's > polling mechanism to check for threshold trips. This is used sometimes to > run performance test cases. > > However, tsens supports an interrupt mechanism to receive notification of > trips, implemented in commit 634e11d5b450 ("drivers: thermal: tsens: Add > interrupt support"). > > 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> > Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> > Signed-off-by: Benjamin Li <benl@squareup.com> > --- > Changes in v3: > - Upgraded logging to dev_info_ratelimited and revised log message. > - Remove unrelated hunk. > > Some drivers that support thermal zone disabling implement a set_mode > operation and simply disable the sensor or the relevant IRQ(s), so they > actually don't log anything when zones are disabled. These drivers are > imx_thermal.c, intel_quark_dts_thermal.c, and int3400_thermal.c. > > For tsens.c, implementing a change_mode would require migrating the driver > from devm_thermal_zone_of_sensor_register to thermal_zone_device_register > (or updating thermal_of.c to add a change_mode operation in thermal_zone_ > of_device_ops). > > stm_thermal.c seems to use this patch's model of not disabling IRQs when > the zone is disabled (they still perform the thermal_zone_device_update > upon IRQ, but return -EAGAIN from their get_temp). What is the concern by changing the core code to have a correct handling of the disabled / enabled state in this driver ? (and by this way give the opportunity to other drivers to fix their code) > Changes in v2: > - Reordered sentences in first part of commit message to make sense. > > drivers/thermal/qcom/tsens.c | 12 ++++++++---- > 1 file changed, 8 insertions(+), 4 deletions(-) > > diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c > index 99a8d9f3e03c..dd0002829536 100644 > --- a/drivers/thermal/qcom/tsens.c > +++ b/drivers/thermal/qcom/tsens.c > @@ -509,10 +509,14 @@ 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_info_ratelimited(priv->dev, "[%u] %s: TZ update trigger (%d mC) skipped - zone disabled, operating outside of safety limits!\n", > + hw_id, __func__, temp); > + } > } else { > dev_dbg(priv->dev, "[%u] %s: no violation: %d\n", > hw_id, __func__, temp);
On Fri, Feb 25, 2022 at 6:02 AM Daniel Lezcano <daniel.lezcano@linaro.org> wrote: > > Some drivers that support thermal zone disabling implement a set_mode > > operation and simply disable the sensor or the relevant IRQ(s), so they > > actually don't log anything when zones are disabled. These drivers are > > imx_thermal.c, intel_quark_dts_thermal.c, and int3400_thermal.c. > > > > For tsens.c, implementing a change_mode would require migrating the driver > > from devm_thermal_zone_of_sensor_register to thermal_zone_device_register > > (or updating thermal_of.c to add a change_mode operation in thermal_zone_ > > of_device_ops). > > > > stm_thermal.c seems to use this patch's model of not disabling IRQs when > > the zone is disabled (they still perform the thermal_zone_device_update > > upon IRQ, but return -EAGAIN from their get_temp). > > What is the concern by changing the core code to have a correct handling > of the disabled / enabled state in this driver ? (and by this way give > the opportunity to other drivers to fix their code)' It seems fine, is that the preference? Updating thermal_of.c to add a change_mode operation in thermal_zone_of_device_ops? Ben > > > -- > <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 25/02/2022 17:46, Benjamin Li wrote: > On Fri, Feb 25, 2022 at 6:02 AM Daniel Lezcano > <daniel.lezcano@linaro.org> wrote: >>> Some drivers that support thermal zone disabling implement a set_mode >>> operation and simply disable the sensor or the relevant IRQ(s), so they >>> actually don't log anything when zones are disabled. These drivers are >>> imx_thermal.c, intel_quark_dts_thermal.c, and int3400_thermal.c. >>> >>> For tsens.c, implementing a change_mode would require migrating the driver >>> from devm_thermal_zone_of_sensor_register to thermal_zone_device_register >>> (or updating thermal_of.c to add a change_mode operation in thermal_zone_ >>> of_device_ops). >>> >>> stm_thermal.c seems to use this patch's model of not disabling IRQs when >>> the zone is disabled (they still perform the thermal_zone_device_update >>> upon IRQ, but return -EAGAIN from their get_temp). >> >> What is the concern by changing the core code to have a correct handling >> of the disabled / enabled state in this driver ? (and by this way give >> the opportunity to other drivers to fix their code)' > > It seems fine, is that the preference? Updating thermal_of.c to add a > change_mode > operation in thermal_zone_of_device_ops? I'm not a big fan of this duplicated ops structure but preferably it would be better to put it there (except if you see a better way to do it)
diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c index 99a8d9f3e03c..dd0002829536 100644 --- a/drivers/thermal/qcom/tsens.c +++ b/drivers/thermal/qcom/tsens.c @@ -509,10 +509,14 @@ 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_info_ratelimited(priv->dev, "[%u] %s: TZ update trigger (%d mC) skipped - zone disabled, operating outside of safety limits!\n", + hw_id, __func__, temp); + } } else { dev_dbg(priv->dev, "[%u] %s: no violation: %d\n", hw_id, __func__, temp);
'echo disabled > .../thermal_zoneX/mode' will disable the thermal core's polling mechanism to check for threshold trips. This is used sometimes to run performance test cases. However, tsens supports an interrupt mechanism to receive notification of trips, implemented in commit 634e11d5b450 ("drivers: thermal: tsens: Add interrupt support"). 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> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> Signed-off-by: Benjamin Li <benl@squareup.com> --- Changes in v3: - Upgraded logging to dev_info_ratelimited and revised log message. - Remove unrelated hunk. Some drivers that support thermal zone disabling implement a set_mode operation and simply disable the sensor or the relevant IRQ(s), so they actually don't log anything when zones are disabled. These drivers are imx_thermal.c, intel_quark_dts_thermal.c, and int3400_thermal.c. For tsens.c, implementing a change_mode would require migrating the driver from devm_thermal_zone_of_sensor_register to thermal_zone_device_register (or updating thermal_of.c to add a change_mode operation in thermal_zone_ of_device_ops). stm_thermal.c seems to use this patch's model of not disabling IRQs when the zone is disabled (they still perform the thermal_zone_device_update upon IRQ, but return -EAGAIN from their get_temp). Changes in v2: - Reordered sentences in first part of commit message to make sense. drivers/thermal/qcom/tsens.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)