Re: [PATCH 04/15] clk: qcom: gcc-sm6375: Add runtime PM

From: Bjorn Andersson <andersson@kernel.org>
To: Stephan Gerhold <stephan@gerhold.net>
Cc: Konrad Dybcio <konrad.dybcio@linaro.org>,
	Andy Gross <agross@kernel.org>,
	Michael Turquette <mturquette@baylibre.com>,
	Stephen Boyd <sboyd@kernel.org>, Rob Herring <robh+dt@kernel.org>,
	Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
	Conor Dooley <conor+dt@kernel.org>,
	Marijn Suijten <marijn.suijten@somainline.org>,
	linux-arm-msm@vger.kernel.org, linux-clk@vger.kernel.org,
	linux-kernel@vger.kernel.org, devicetree@vger.kernel.org
Subject: Re: [PATCH 04/15] clk: qcom: gcc-sm6375: Add runtime PM
Date: Mon, 17 Jul 2023 21:02:29 -0700	[thread overview]
Message-ID: <pv7pcn5xc7xdjigefrwzvbjvp5lmgfziywfpvxn2nl7hjnvyre@kukniawnm5rm> (raw)
In-Reply-To: <ZLVru50N2ukdXnsW@gerhold.net>

On Mon, Jul 17, 2023 at 06:26:35PM +0200, Stephan Gerhold wrote:
> On Mon, Jul 17, 2023 at 05:19:11PM +0200, Konrad Dybcio wrote:
> > The GCC block on SM6375 is powered by the VDD_CX rail. We need to ensure
> > that it's enabled to prevent unwanted power collapse.
> > 
> > Enable runtime PM to keep the power flowing only when necessary.
> > 
> 
> Are you sure this is necessary? If VDD_CX was really possible to fully
> "power collapse" then I would expect that you lose all register
> settings. This is not something we want or can even handle for GCC.
> You would need to restore all frequency settings, branch bits etc etc.
> 

This differ between platforms, some allow us to completely power down CX
while keeping registers state using MX, others require that CX stays in
retention at least.

So, CX isn't the only rail powering GCC. For the most part though, we
have a relationship between frequencies votes for by clients and the
corner of CX, and hence I think the current description is ok...

> Otherwise it's a retention state, but these are covered by the
> corners/levels, not the enable/disable state.
> 

I _think_ we still want to suspend each individual device (and the vote
from Linux), and let the system keep us at retention, instead of off...

> I think most of these power domains are effectively always-on. The
> important part is voting for minimal corners/levels so they can go to
> minimal retention state with vlow/vmin.
> 

When you hit s2idle, you should expect to have the majority of the
rpm(h) PDs be voted off.

Regards,
Bjorn

> Thanks,
> Stephan