Re: [PATCH 2/2] arm64: dts: mediatek: cherry: Add CPU supply dependency to cpufreq-hw

From: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
To: "Viresh Kumar" <viresh.kumar@linaro.org>,
	"Nícolas F. R. A. Prado" <nfraprado@collabora.com>
Cc: "Rafael J . Wysocki" <rafael@kernel.org>,
	kernel@collabora.com, Conor Dooley <conor+dt@kernel.org>,
	Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
	Matthias Brugger <matthias.bgg@gmail.com>,
	Rob Herring <robh+dt@kernel.org>,
	devicetree@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, linux-mediatek@lists.infradead.org
Subject: Re: [PATCH 2/2] arm64: dts: mediatek: cherry: Add CPU supply dependency to cpufreq-hw
Date: Tue, 2 Jan 2024 10:09:30 +0100	[thread overview]
Message-ID: <a288d5c8-6eb9-49f8-bcbc-f2b41e51c797@collabora.com> (raw)
In-Reply-To: <20240102061123.frvwltggqqh6m7pm@vireshk-i7>

Il 02/01/24 07:11, Viresh Kumar ha scritto:
> On 29-12-23, 18:28, Nícolas F. R. A. Prado wrote:
>> When the mediatek-cpufreq-hw driver enables the hardware (by
>> writing to REG_FREQ_ENABLE), if the regulator supplying the voltage to
>> the big CPUs hasn't probed yet, the platform hangs shortly after and
>> "rcu: INFO: rcu_preempt detected stalls on CPUs/tasks" are printed in
>> the log.
>>
>> To prevent this from happening, describe the big CPUs regulator in the
>> performance-controller DT node, so that devlink ensures the regulator
>> has been probed and configured before the frequency scaling hardware is
>> probed and enabled.
>>
>> Signed-off-by: Nícolas F. R. A. Prado <nfraprado@collabora.com>
>>
>> ---
>>
>>   arch/arm64/boot/dts/mediatek/mt8195-cherry.dtsi | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/arch/arm64/boot/dts/mediatek/mt8195-cherry.dtsi b/arch/arm64/boot/dts/mediatek/mt8195-cherry.dtsi
>> index dd5b89b73190..505da60eee90 100644
>> --- a/arch/arm64/boot/dts/mediatek/mt8195-cherry.dtsi
>> +++ b/arch/arm64/boot/dts/mediatek/mt8195-cherry.dtsi
>> @@ -502,6 +502,10 @@ &pcie1 {
>>   	pinctrl-0 = <&pcie1_pins_default>;
>>   };
>>   
>> +&performance {
>> +	big-cpus-supply = <&mt6315_6_vbuck1>;
>> +};
>> +
>>   &pio {
>>   	mediatek,rsel-resistance-in-si-unit;
>>   	pinctrl-names = "default";
> 
> I think the regulator needs to be mentioned in the CPU's node and not
> here ?
> 

Even if the regulator voltage is being changed by firmware with cpufreq-hw, the
actual regulators should go to each CPU node and not in the cpufreq driver node,
I agree with Viresh.

Besides, that's the same thing that we're doing with mediatek-cpufreq as well...
and since we're talking about that, we should also do something about this such
that we stop declaring `regulator-always-on` for CPU cores in devicetree, but
this is probably slightly out of context for what you're trying to do here, so,
read that as an "extra consideration" :-)

Cheers,
Angelo