Skip to contentSkip to conversation

thermal: avoid division by zero in power allocator

PATCH New Andrea Arcangeli 3 participants 4 messages Sep 28, 2015 – Oct 1, 2015 Thermal

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 thermal: avoid division by zero in power allocator

Hello,

This is needed for my workstations or they Oops at boot since v4.3-rc3.

Andrea Arcangeli (1):
thermal: avoid division by zero in power allocator

 drivers/thermal/power_allocator.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

Thanks,
Andrea

#1 lore raw

PATCH 1/1 thermal: avoid division by zero in power allocator

During boot I get a div by zero Oops regression starting in v4.3-rc3.

---
 drivers/thermal/power_allocator.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
drivers/thermal/power_allocator.c +10 -0
index 7ff9627..e570ff0 100644
@@ -144,6 +144,16 @@ static void estimate_pid_constants(struct thermal_zone_device *tz,
 		switch_on_temp = 0;
 
 	temperature_threshold = control_temp - switch_on_temp;
+	/*
+	 * estimate_pid_constants() tries to find appropriate default
+	 * values for thermal zones that don't provide them. If a
+	 * system integrator has configured a thermal zone with two
+	 * passive trip points at the same temperature, that person
+	 * hasn't put any effort to set up the thermal zone properly
+	 * so just give up.
+	 */
+	if (!temperature_threshold)
+		return;
 
 	if (!tz->tzp->k_po || force)
 		tz->tzp->k_po = int_to_frac(sustainable_power) /
#2 lore raw
Quoted 13 lines · On Mon, 28 Sep 2015 23:28:34 +0200 Andrea Arcangeli <[email protected]> wrote:

During boot I get a div by zero Oops regression starting in v4.3-rc3.

...

drivers/thermal/power_allocator.c +3 -0
@@ -144,6 +144,16 @@ static void estimate_pid_constants(struct thermal_zone_device *tz,
 		switch_on_temp = 0;
 
 	temperature_threshold = control_temp - switch_on_temp;
+	/*
+	 * estimate_pid_constants() tries to find appropriate default
+	 * values for thermal zones that don't provide them. If a

drivers/thermal/power_allocator.c static void estimate_pid_constants(struct thermal_zone_device *tz, · line 159

Quoted 10 lines
drivers/thermal/power_allocator.c +7 -0
+	 * system integrator has configured a thermal zone with two
+	 * passive trip points at the same temperature, that person
+	 * hasn't put any effort to set up the thermal zone properly
+	 * so just give up.
+	 */
+	if (!temperature_threshold)
+		return;
 
 	if (!tz->tzp->k_po || force)
 		tz->tzp->k_po = int_to_frac(sustainable_power) /

a) Are we sure this won't leave tz->tzp fields uninitialized?

b) I'm not understanding that code at all. The "proportional" term in a PID controller is supposed to be proportional to the (desired - actual) difference (aka "the error").

But estimate_pid_constants() appears to be setting the "proportional" term to be proportional to 1/error!

Maybe a description of local `temperature_threshold' would help clue me in.

#3 lore raw
Quoted 23 lines · Andrew Morton
Quoted 21 lines · On Mon, 28 Sep 2015 23:28:34 +0200 Andrea Arcangeli <[email protected]> wrote:

During boot I get a div by zero Oops regression starting in v4.3-rc3.

...

drivers/thermal/power_allocator.c +10 -0
@@ -144,6 +144,16 @@ static void estimate_pid_constants(struct thermal_zone_device *tz,
 		switch_on_temp = 0;
 
 	temperature_threshold = control_temp - switch_on_temp;
+	/*
+	 * estimate_pid_constants() tries to find appropriate default
+	 * values for thermal zones that don't provide them. If a
+	 * system integrator has configured a thermal zone with two
+	 * passive trip points at the same temperature, that person
+	 * hasn't put any effort to set up the thermal zone properly
+	 * so just give up.
+	 */
+	if (!temperature_threshold)
+		return;
 
Quoted 4 lines

drivers/thermal/power_allocator.c static void estimate_pid_constants(struct thermal_zone_device *tz, · line 159

Quoted 2 lines
drivers/thermal/power_allocator.c +0 -0
 	if (!tz->tzp->k_po || force)
 		tz->tzp->k_po = int_to_frac(sustainable_power) /

a) Are we sure this won't leave tz->tzp fields uninitialized?

They will be all zeros. That's good enough.

Quoted 6 lines

b) I'm not understanding that code at all. The "proportional" term in a PID controller is supposed to be proportional to the (desired - actual) difference (aka "the error").

But estimate_pid_constants() appears to be setting the "proportional" term to be proportional to 1/error!

estimate_pid_constants() calculate the constants that you use in the PID algorithm. Say:

k_p * error + k_i * integral_of_error + k_d * diff_of_error

This code is calculating a reasonable k_p, k_i and k_d when they are not provided by the platform.

Quoted 2 lines

Maybe a description of local `temperature_threshold' would help clue me in.

The `error' in the above definition is:

target_temperature - current_temperature

whereas `temperature_threshold' is:

target_temperature' - switch_on_temperature'

`switch_on_temperature' is the temperature above which the thermal governor starts operating and throttling cpus (or whatever cooling device is configured).

The switch_on_temperature' and target_temperature' are defined using trip points. A platform that sets two trip points to the same temperature is not properly configured. With Andrea's patch we provide degraded behavior instead of crashing. I agree with that approach (hence my Reviewed-by, maybe it should be an Acked-by?).

Cheers,
Javi