Re: [PATCH 5/7] arm64: dts: exynos: Add dts files for Samsung Exynos5433 64bit SoC

From: Krzysztof Kozlowski
Date: Tue Aug 16 2016 - 13:52:00 EST


On Tue, Aug 16, 2016 at 09:59:26PM +0900, Chanwoo Choi wrote:
> >> diff --git a/arch/arm64/boot/dts/exynos/exynos5433-tmu.dtsi b/arch/arm64/boot/dts/exynos/exynos5433-tmu.dtsi
> >> new file mode 100644
> >> index 000000000000..175121db367e
> >> --- /dev/null
> >> +++ b/arch/arm64/boot/dts/exynos/exynos5433-tmu.dtsi
> >> @@ -0,0 +1,306 @@
> >> +/*
> >> + * Device tree sources for Exynos5433 thermal zone
> >> + *
> >> + * Copyright (c) 2016 Chanwoo Choi <cw00.choi@xxxxxxxxxxx>
> >> + *
> >> + * This program is free software; you can redistribute it and/or modify
> >> + * it under the terms of the GNU General Public License version 2 as
> >> + * published by the Free Software Foundation.
> >> + */
> >> +
> >> +#include <dt-bindings/thermal/thermal.h>
> >> +
> >> +/ {
> >> +thermal-zones {
> >> + atlas0_thermal: atlas0-thermal {
> >> + thermal-sensors = <&tmu_atlas0>;
> >> + polling-delay-passive = <0>;
> >> + polling-delay = <0>;
> >> + trips {
> >> + atlas0_alert_0: atlas0-alert-0 {
> >> + temperature = <50000>; /* millicelsius */
> >> + hysteresis = <1000>; /* millicelsius */
> >> + type = "active";
> >> + };
> >> + atlas0_alert_1: atlas0-alert-1 {
> >> + temperature = <55000>; /* millicelsius */
> >> + hysteresis = <1000>; /* millicelsius */
> >> + type = "active";
> >> + };
> >> + atlas0_alert_2: atlas0-alert-2 {
> >> + temperature = <60000>; /* millicelsius */
> >> + hysteresis = <1000>; /* millicelsius */
> >> + type = "active";
> >> + };
> >> + atlas0_alert_3: atlas0-alert-3 {
> >> + temperature = <70000>; /* millicelsius */
> >> + hysteresis = <1000>; /* millicelsius */
> >> + type = "active";
> >> + };
> >> + atlas0_alert_4: atlas0-alert-4 {
> >> + temperature = <80000>; /* millicelsius */
> >> + hysteresis = <1000>; /* millicelsius */
> >> + type = "active";
> >> + };
> >> + atlas0_alert_5: atlas0-alert-5 {
> >> + temperature = <90000>; /* millicelsius */
> >> + hysteresis = <1000>; /* millicelsius */
> >> + type = "active";
> >> + };
> >> + atlas0_alert_6: atlas0-alert-6 {
> >> + temperature = <95000>; /* millicelsius */
> >> + hysteresis = <1000>; /* millicelsius */
> >> + type = "active";
> >> + };
> >
> > No critical trip? I think it might be useful to shutdown the system in a
> > user-friendly way.
>
> When I use the critical trip, the following event occur[1].
> But, I guess that this temperature is not correct temperature
> because after completing the kernel booting, the temperature of big.LITTLE/G3D
> are normal when checking the /sys/class/thermal/thermal_zoneX/temp right after booting.
> - Maintain a uniform temperature(38 ~ 45 millicelsius) right after kernel booting.
>
> I guess that the critical interrupt may occur before initializing the exynos tmu.
> But, I don't spend the many time to check the exynos-tmu.c driver.
>
> [1]
> [ 445.122122] thermal thermal_zone0: critical temperature reached(108 C),shutting down
> [ 445.122399] exynos-tmu 10060000.tmu: Temperature sensor ID: 0xa
> [ 445.122588] exynos-tmu 10060000.tmu: Calibration type is 2-point calibration
> [ 445.127942] reboot: Failed to start orderly shutdown: forcing the issue
> [ 445.134586] Emergency Sync complete
> [ 1.097954] reboot: Power down

I understand. Apparently the exynos-tmu driver needs some fixes for
this race. Skipping critical then makes sense.

>
> >
> >> + };
> >> +
> >> + cooling-maps {
> >> + map0 {
> >> + /* Set maximum frequency as 1800MHz */
> >> + trip = <&atlas0_alert_0>;
> >> + cooling-device = <&cpu4 1 1>;
> >
> > Out of curiosity: why choosing specific cooling level (so quite fast
> > the device will slow down) instead of letting cooling framework to
> > decide how much to cool? Any particular reason behind this?
>
> This cooling level is just default value in cooling-maps.
> This value is able to overwrite on dts file.
>
> And the thermal subsystem support the cpu cooling features
> with 'cooling-maps'.
>
> Also, when I tested the performance and stress test
> with GLBenchmark, the temperature of big.LITTLE cores/G3D
> reach easily the critical temperature with 8 online cores.
> So, I add the cooling level aggressively to protect
> the system fault of CPU and GPU and to maintain
> the system state.

I was asking why you do not let cooling framework decide which cooling
level to use but instead you force a specific cooling level. Maybe code
will be a better example. Why not use:
map0 {
/* Set maximum frequency as 1800MHz */
trip = <&atlas0_alert_0>;
cooling-device = <&cpu4 0 1>;
}
map1 {
/* Set maximum frequency as 1700MHz */
trip = <&atlas0_alert_1>;
cooling-device = <&cpu4 1 2>;
};

For higher frequencies it makes even more sense:
map6 {
/* Set maximum frequency as 800MHz */
trip = <&atlas0_alert_6>;
cooling-device = <&cpu4 7 11>;
};

which allows the system to use suitable cooling level to maintain the
balance between performance and temperature dissipance, instead of some
fixed cooling level which might not be accurate to the system load.

Best regards,
Krzysztof