| Message ID | 1466571990-12346-1-git-send-email-edubezval@gmail.com (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On Wed, Jun 22, 2016 at 7:06 AM, Eduardo Valentin <edubezval@gmail.com> wrote: > Because several drivers do the following pattern: > .set_mode() > ... > local_data->mode = new_mode; > thermal_zone_device_update(tz); > > makes sense to simply do the thermal_zone_device_update() > in thermal core, after setting the new mode. > > Also, this patch also remove deadlocks on drivers that > call thermal_zone_device_update() on .set_mode(), > as .set_mode() is now called always with tz->lock held. To me, this part of the patch is way more important than the optimization mentioned before. Apparently, the problem is that drivers deadlock, because the thermal_zone_device_update() invoked from ->set_mode() is called under tz->lock. So to address that problem you make the core call thermal_zone_device_update() after ->set_mode() outside of tz->lock and the drivers don't have to do it any more. Is that correct? Thanks, Rafael -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thursday 23 June 2016 05:57 PM, Rafael J. Wysocki wrote: > On Wed, Jun 22, 2016 at 7:06 AM, Eduardo Valentin <edubezval@gmail.com> wrote: >> Because several drivers do the following pattern: >> .set_mode() >> ... >> local_data->mode = new_mode; >> thermal_zone_device_update(tz); >> >> makes sense to simply do the thermal_zone_device_update() >> in thermal core, after setting the new mode. >> >> Also, this patch also remove deadlocks on drivers that >> call thermal_zone_device_update() on .set_mode(), >> as .set_mode() is now called always with tz->lock held. > > To me, this part of the patch is way more important than the > optimization mentioned before. > > Apparently, the problem is that drivers deadlock, because the > thermal_zone_device_update() invoked from ->set_mode() is called under > tz->lock. > > So to address that problem you make the core call > thermal_zone_device_update() after ->set_mode() outside of tz->lock > and the drivers don't have to do it any more. > > Is that correct? Rafael, On my set up, mode_store locks tz->lock and eventually ends up calling of_thermal_set_mode before releasing tz->lock which again tries to lock tz->lock and ends up in a deadlock. http://pastebin.ubuntu.com/17687601/ Regards, Keerthy > > Thanks, > Rafael > -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Jun 23, 2016 at 02:27:12PM +0200, Rafael J. Wysocki wrote: > On Wed, Jun 22, 2016 at 7:06 AM, Eduardo Valentin <edubezval@gmail.com> wrote: > > Because several drivers do the following pattern: > > .set_mode() > > ... > > local_data->mode = new_mode; > > thermal_zone_device_update(tz); > > > > makes sense to simply do the thermal_zone_device_update() > > in thermal core, after setting the new mode. > > > > Also, this patch also remove deadlocks on drivers that > > call thermal_zone_device_update() on .set_mode(), > > as .set_mode() is now called always with tz->lock held. > > To me, this part of the patch is way more important than the > optimization mentioned before. > > Apparently, the problem is that drivers deadlock, because the > thermal_zone_device_update() invoked from ->set_mode() is called under > tz->lock. > > So to address that problem you make the core call > thermal_zone_device_update() after ->set_mode() outside of tz->lock > and the drivers don't have to do it any more. > > Is that correct? Yes this is correct. The optimization is simply a consequence of the bug fix, as reported by Keerthy. > > Thanks, > Rafael -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/acpi/thermal.c b/drivers/acpi/thermal.c index 82707f9..8582b88 100644 --- a/drivers/acpi/thermal.c +++ b/drivers/acpi/thermal.c @@ -519,8 +519,6 @@ static void acpi_thermal_check(void *data) if (!tz->tz_enabled) return; - - thermal_zone_device_update(tz->thermal_zone); } /* sys I/F for generic thermal sysfs support */ diff --git a/drivers/platform/x86/acerhdf.c b/drivers/platform/x86/acerhdf.c index 460fa67..aee33ba 100644 --- a/drivers/platform/x86/acerhdf.c +++ b/drivers/platform/x86/acerhdf.c @@ -405,7 +405,6 @@ static inline void acerhdf_enable_kernelmode(void) kernelmode = 1; thz_dev->polling_delay = interval*1000; - thermal_zone_device_update(thz_dev); pr_notice("kernel mode fan control ON\n"); } diff --git a/drivers/thermal/imx_thermal.c b/drivers/thermal/imx_thermal.c index c5547bd..a413eb6 100644 --- a/drivers/thermal/imx_thermal.c +++ b/drivers/thermal/imx_thermal.c @@ -246,7 +246,6 @@ static int imx_set_mode(struct thermal_zone_device *tz, } data->mode = mode; - thermal_zone_device_update(tz); return 0; } diff --git a/drivers/thermal/of-thermal.c b/drivers/thermal/of-thermal.c index b8e509c..b44c102 100644 --- a/drivers/thermal/of-thermal.c +++ b/drivers/thermal/of-thermal.c @@ -181,9 +181,6 @@ static int of_thermal_set_emul_temp(struct thermal_zone_device *tz, { struct __thermal_zone *data = tz->devdata; - if (!data->ops || !data->ops->set_emul_temp) - return -EINVAL; - return data->ops->set_emul_temp(data->sensor_data, temp); } @@ -292,7 +289,6 @@ static int of_thermal_set_mode(struct thermal_zone_device *tz, mutex_unlock(&tz->lock); data->mode = mode; - thermal_zone_device_update(tz); return 0; } @@ -427,7 +423,9 @@ thermal_zone_of_add_sensor(struct device_node *zone, tzd->ops->get_temp = of_thermal_get_temp; tzd->ops->get_trend = of_thermal_get_trend; - tzd->ops->set_emul_temp = of_thermal_set_emul_temp; + if (ops->set_emul_temp) + tzd->ops->set_emul_temp = of_thermal_set_emul_temp; + mutex_unlock(&tzd->lock); return tzd; diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c index 09da955..bb1ede7 100644 --- a/drivers/thermal/thermal_core.c +++ b/drivers/thermal/thermal_core.c @@ -323,6 +323,21 @@ static void handle_non_critical_trips(struct thermal_zone_device *tz, def_governor->throttle(tz, trip); } +static void thermal_zone_update_stats(struct thermal_zone_device *tz, + int last_trip, int cur_trip) +{ + unsigned long long cur_time = get_jiffies_64(); + struct thermal_zone_stats *cur, *last; + + cur = tz->stats_table[cur_trip]; + last = tz->stats_table[last_trip]; + + cur->counter++; + cur->last_time = cur_time; + + last->time_in += cur_time - last->last_time; +} + static void thermal_tripped_notify(struct thermal_zone_device *tz, int trip, enum thermal_trip_type trip_type, int trip_temp) @@ -333,6 +348,7 @@ static void thermal_tripped_notify(struct thermal_zone_device *tz, NULL }; int upper_trip_hyst, upper_trip_temp, trip_hyst = 0; int ret = 0; + int cur_trip, last_trip; snprintf(tuv_name, sizeof(tuv_name), "THERMAL_ZONE=%s", tz->type); snprintf(tuv_temp, sizeof(tuv_temp), "TEMP=%d", tz->temperature); @@ -344,8 +360,11 @@ static void thermal_tripped_notify(struct thermal_zone_device *tz, mutex_lock(&tz->lock); /* crossing up */ - if (tz->last_temperature < trip_temp && trip_temp < tz->temperature) + if (tz->last_temperature < trip_temp && trip_temp < tz->temperature) { kobject_uevent_env(&tz->device.kobj, KOBJ_CHANGE, msg); + last_trip = trip - 1; + cur_trip = trip; + } if (tz->ops->get_trip_hyst) tz->ops->get_trip_hyst(tz, trip, &trip_hyst); @@ -355,6 +374,8 @@ static void thermal_tripped_notify(struct thermal_zone_device *tz, if (tz->last_temperature > trip_temp && trip_temp > tz->temperature) { snprintf(tuv_trip, sizeof(tuv_trip), "TRIP=%d", trip - 1); kobject_uevent_env(&tz->device.kobj, KOBJ_CHANGE, msg); + last_trip = trip; + cur_trip = trip - 1; } ret = tz->ops->get_trip_temp(tz, trip + 1, &upper_trip_temp); @@ -369,19 +390,15 @@ static void thermal_tripped_notify(struct thermal_zone_device *tz, upper_trip_temp > tz->temperature) kobject_uevent_env(&tz->device.kobj, KOBJ_CHANGE, msg); + thermal_zone_device_update_stats(tz, last_trip, cur_trip); unlock: mutex_unlock(&tz->lock); } static void handle_critical_trips(struct thermal_zone_device *tz, - int trip, enum thermal_trip_type trip_type) + int trip, enum thermal_trip_type trip_type, + int trip_temp) { - int trip_temp; - - tz->ops->get_trip_temp(tz, trip, &trip_temp); - - thermal_tripped_notify(tz, trip, trip_type, trip_temp); - /* If we have not crossed the trip_temp, we do not care. */ if (trip_temp <= 0 || tz->temperature < trip_temp) return; @@ -402,15 +419,21 @@ static void handle_critical_trips(struct thermal_zone_device *tz, static void handle_thermal_trip(struct thermal_zone_device *tz, int trip) { enum thermal_trip_type type; + int trip_temp; + + tz->ops->get_trip_temp(tz, trip, &trip_temp); /* Ignore disabled trip points */ if (test_bit(trip, &tz->trips_disabled)) return; tz->ops->get_trip_type(tz, trip, &type); + tz->ops->get_trip_type(tz, trip, &trip_temp); + + thermal_tripped_notify(tz, trip, trip_type, trip_temp); if (type == THERMAL_TRIP_CRITICAL || type == THERMAL_TRIP_HOT) - handle_critical_trips(tz, trip, type); + handle_critical_trips(tz, trip, type, trip_temp); else handle_non_critical_trips(tz, trip, type); } diff --git a/drivers/thermal/thermal_sysfs.c b/drivers/thermal/thermal_sysfs.c index 743df50..3d0dc30 100644 --- a/drivers/thermal/thermal_sysfs.c +++ b/drivers/thermal/thermal_sysfs.c @@ -100,6 +100,7 @@ mode_store(struct device *dev, struct device_attribute *attr, mutex_lock(&tz->lock); result = tz->ops->set_mode(tz, mode); mutex_unlock(&tz->lock); + thermal_zone_device_update(tz); if (result) return result; diff --git a/drivers/thermal/ti-soc-thermal/ti-thermal-common.c b/drivers/thermal/ti-soc-thermal/ti-thermal-common.c index 15c0a9a..9a5a3a3 100644 --- a/drivers/thermal/ti-soc-thermal/ti-thermal-common.c +++ b/drivers/thermal/ti-soc-thermal/ti-thermal-common.c @@ -205,7 +205,6 @@ static int ti_thermal_set_mode(struct thermal_zone_device *thermal, data->mode = mode; ti_bandgap_write_update_interval(bgp, data->sensor_id, data->ti_thermal->polling_delay); - thermal_zone_device_update(data->ti_thermal); dev_dbg(&thermal->device, "thermal polling set for duration=%d msec\n", data->ti_thermal->polling_delay);
Because several drivers do the following pattern: .set_mode() ... local_data->mode = new_mode; thermal_zone_device_update(tz); makes sense to simply do the thermal_zone_device_update() in thermal core, after setting the new mode. Also, this patch also remove deadlocks on drivers that call thermal_zone_device_update() on .set_mode(), as .set_mode() is now called always with tz->lock held. Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Cc: "Lee, Chun-Yi" <jlee@suse.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Zhang Rui <rui.zhang@intel.com> Cc: Keerthy <j-keerthy@ti.com> Cc: linux-kernel@vger.kernel.org Cc: linux-omap@vger.kernel.org Cc: platform-driver-x86@vger.kernel.org Cc: linux-pm@vger.kernel.org Signed-off-by: Eduardo Valentin <edubezval@gmail.com> --- Rui, Keerthy, I think this patch should take care of the introduced deadlock. Let me know if solves on your end. BR, Eduardo --- drivers/acpi/thermal.c | 2 -- drivers/platform/x86/acerhdf.c | 1 - drivers/thermal/imx_thermal.c | 1 - drivers/thermal/of-thermal.c | 8 ++--- drivers/thermal/thermal_core.c | 41 +++++++++++++++++----- drivers/thermal/thermal_sysfs.c | 1 + drivers/thermal/ti-soc-thermal/ti-thermal-common.c | 1 - 7 files changed, 36 insertions(+), 19 deletions(-)