On 2017/10/11 1:28, Daniel Lezcano wrote:

On 10/10/2017 19:19, Valentin Schneider wrote:

quoted

On 10/10/2017 06:13 PM, Daniel Lezcano wrote:

quoted

On 10/10/2017 19:01, Valentin Schneider wrote:

quoted

On 10/10/2017 05:51 PM, Daniel Lezcano wrote:

quoted

Ok, so apparently there are multiple alarms level in the driver for the
hikey960 [1]. So I prefer to drop this patch for now and take the
hikey960 thermal support first and we can sort out the issue later.

For my information, can you show me the DT snippet you have for the
thermal zones?

Sure thing:

      thermal-zones {

              cls0: cls0 {
                  polling-delay = <1000>;
                  polling-delay-passive = <100>;
                  sustainable-power = <4500>;

                  /* sensor ID */
                  thermal-sensors = <&tsensor 1>;

                  trips {
                      threshold: trip-point@0 {
                          temperature = <65000>;
                          hysteresis = <1000>;
                          type = "passive";
                      };

                      target: trip-point@1 {
                          temperature = <75000>;
                          hysteresis = <1000>;
                          type = "passive";
                      };
                  };

That's strange, regarding your traces:

"
[  118.107357] hisi_thermal fff30000.tsensor: THERMAL ALARM: 70495 >
65000
[  119.182531] hisi_thermal fff30000.tsensor: THERMAL ALARM: 76235 >
65000
[  119.361964] hisi_thermal fff30000.tsensor: THERMAL ALARM: 70495 >
65000
[  119.907865] hisi_thermal fff30000.tsensor: THERMAL ALARM: 75620 >
65000
[  119.959076] hisi_thermal fff30000.tsensor: THERMAL ALARM: 70700 >
65000
"

I was expecting to see more trip points. Did you test the driver with a
70000 trip point?

No, I didn't change any setting other than the tsensor index to make
things work. Mind you, in Kevin's patch series the thermal alarm is
setup with a 4 degrees "lag", i.e. alarms will only be re-triggered if
temperature increases/decreases of at least 4 degrees (which explains
the traces).

This should be a trip point 75000, cross 75000 triger an interrupt and
drop below 71000 triger another interrupt, the multi alarm interrupt is
not suitable for ipa as you discussed before, we should drop the patch
of multi alarm support.

Mmh, the behavior regarding the interrupt is slightly different with the
hi960, perhaps a bit fuzzy regarding how it is handled now. Anyway, we
can live with that now and go further to fix that later, the result is
the same.