Re: [PATCH] arm64: dts: qcom: shikra: Drop CPU critical trips

[Abel Vesa]

On 26-10-21 17:12:11, Gaurav Kohli wrote:
> 
> 
> On 10/20/2026 6:26 PM, Abel Vesa wrote:
> > On 26-10-19 20:03:47, Gaurav Kohli wrote:
> > > 
> > > 
> > > On 10/15/2026 4:43 PM, Konrad Dybcio wrote:
> > > > On 9/25/26 12:24 PM, Gaurav Kohli wrote:
> > > > > 
> > > > > 
> > > > > On 9/25/2026 3:13 PM, Abel Vesa wrote:
> > > > > > On 26-09-24 19:52:32, Gaurav Kohli wrote:
> > > > > > > Shikra firmware handles CPU thermal mitigation up to 117 C, while
> > > > > > > Linux CPU critical trips trigger at 115 C. This can shut the system
> > > > > > > down before firmware reaches the end of its mitigation range.
> > > > > > > 
> > > > > > > Drop the critical trips from the CPU thermal zones and leave the hot
> > > > > > > trips, non-CPU critical trips and GPUSS passive cooling map unchanged.
> > > > > > 
> > > > > > I thought that was the whole point of having the critical trips set to
> > > > > > 115 C, so that the kernel can gracefully shut down before firmware cuts
> > > > > > it off.
> > > > > > 
> > > > > > What happens if CPU gets to 117 C and the critical trips are removed?
> > > > > > How will firmware thermal mitigation handle it?
> > > > > 
> > > > > Thanks Abel for review.
> > > > > 
> > > > > For the CPU thermal zones on this SoC, hardware-side mitigation is already available at 115 C and 117 C, and hardware shutdown is at 120 C. Because of that, a Linux critical trip at 115 C is not very useful. It overlaps with the hardware mitigation range and can trigger shutdown before the firmware thermal sequence has completed.
> > > > 
> > > > Does the hardware mitigation just pull the plug, or does it signal
> > > > to the OS that it's about to? We absolutely need the OS to be aware,
> > > > so that e.g. the filesystem will be synced.
> > > 
> > > Thanks Konrad for the review.
> > > 
> > > The OS is not notified before the final HW shutdown at 120 C. There is also
> > > no useful margin for a Linux critical trip here, firmware/HW mitigation
> > > starts at 115/117 C and hw shutdown is at 120.
> > > 
> > > So 115 C overlaps with firmware mitigation, while placing it higher is
> > > too close to HW shutdown to be useful.
> > 
> > But having the 115 C one will allow the kernel to take all the necessary
> > steps to bring the system down gracefully, while relying solely on the
> > firmware will just kill the system before kernel can do anything.
> 
> Thanks Abel for the review.
> 
> On this target, 115/117 C firmware mitigation is throttling/prevention, not
> shutdown. The final HW shutdown threshold is 120 C.
> 
> The 115 C Linux critical trip therefore starts shutdown exactly when
> firmware mitigation begins. Moving it higher also leaves too little margin
> before 120 C.
> 
> So this patch drops only the CPU critical trips. Would you prefer keeping a
> Linux critical trip at 118 C ?

Yes. Again, we don't want the firmware to crash the system before kernel
has a chance to bring it down gracefully.

But how does the firmware mitigation (throttling/prevention) is done?
Shouldn't kernel be aware of such thing ?