linux-kernel - Re: [PATCH 4/5] i2c: qcom-cci: Add OPP table support and enforce FAST

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <aL6rYq5gmX8CdeE-@linaro.org>
Date: Mon, 8 Sep 2025 12:09:38 +0200
From: Stephan Gerhold <stephan.gerhold@...aro.org>
To: Konrad Dybcio <konrad.dybcio@....qualcomm.com>
Cc: Konrad Dybcio <konradybcio@...nel.org>,
	Bjorn Andersson <andersson@...nel.org>,
	Rob Herring <robh@...nel.org>,
	Krzysztof Kozlowski <krzk+dt@...nel.org>,
	Conor Dooley <conor+dt@...nel.org>,
	Bryan O'Donoghue <bryan.odonoghue@...aro.org>,
	Loic Poulain <loic.poulain@....qualcomm.com>,
	Robert Foss <rfoss@...nel.org>, Andi Shyti <andi.shyti@...nel.org>,
	Marijn Suijten <marijn.suijten@...ainline.org>,
	linux-arm-msm@...r.kernel.org, devicetree@...r.kernel.org,
	linux-kernel@...r.kernel.org, linux-i2c@...r.kernel.org
Subject: Re: [PATCH 4/5] i2c: qcom-cci: Add OPP table support and enforce
 FAST_PLUS requirements

On Mon, Sep 08, 2025 at 12:00:13PM +0200, Konrad Dybcio wrote:
> On 9/8/25 11:57 AM, Stephan Gerhold wrote:
> > On Mon, Sep 08, 2025 at 11:49:52AM +0200, Konrad Dybcio wrote:
> >> On 9/8/25 10:46 AM, Stephan Gerhold wrote:
> >>> On Mon, Sep 08, 2025 at 10:43:50AM +0200, Konrad Dybcio wrote:
> >>>> On 9/8/25 10:36 AM, Stephan Gerhold wrote:
> >>>>> On Thu, Sep 04, 2025 at 04:31:23PM +0200, Konrad Dybcio wrote:
> >>>>>> From: Konrad Dybcio <konrad.dybcio@....qualcomm.com>
> >>>>>>
> >>>>>> The CCI clock has voltage requirements, which need to be described
> >>>>>> through an OPP table.
> >>>>>>
> >>>>>> The 1 MHz FAST_PLUS mode requires the CCI core clock runs at 37,5 MHz
> >>>>>> (which is a value common across all SoCs), since it's not possible to
> >>>>>> reach the required timings with the default 19.2 MHz rate.
> >>>>>>
> >>>>>> Address both issues by introducing an OPP table and using it to vote
> >>>>>> for the faster rate.
> >>>>>>
> >>>>>> Signed-off-by: Konrad Dybcio <konrad.dybcio@....qualcomm.com>
> >>>>>
> >>>>> Using an OPP table for a single static rate that remains the same over
> >>>>> the whole lifetime of the driver feels like overkill to me. Couldn't you
> >>>>> just put the "required-opps" directly into the device node so that it is
> >>>>> automatically applied when the device goes in/out of runtime suspend?
> >>>>>
> >>>>> And since you need to make DT additions anyway, couldn't you just use
> >>>>> "assigned-clock-rates" to avoid the need for a driver patch entirely? We
> >>>>> use that for e.g. USB clocks as well.
> >>>>
> >>>> This is futureproofing, in case someone invents FastMode++ with a higher
> >>>> dvfs requirement or for when the driver adds presets for a 19.2 MHz CCI
> >>>> clock which would (marginally) decrease power consumption
> >>>>
> >>>
> >>> If 19.2 MHz CCI clock is feasible and has lower voltage requirements,
> >>> then I would expect a separate entry for 19.2 MHz in the OPP table of
> >>> PATCH 5/5? The DT is unrelated to what functionality you implement in
> >>> the driver, and that would make the OPP table look less useless. :-)
> >>
> >> The frequency plan for 8280 does not recommend any rate != 37.5 MHz
> >>
> >> For x1e80100 however, the lovsvs_d1 corner is recommended to be 30
> >> (yes, thirty) MHz, sourced from CAM_PLL8 for $reasons
> >>
> > 
> > The 37.5 MHz rate still exists on X1E I presume, or are you saying we
> > need more changes to support those odd 30 MHz?
> 
> Yes, any corner over lowsvs_d1 is 37.5, sourced from cam_pll0
> 
> > Personally, I'm not fully convinced there is ever going to be a use case
> > of someone using a "non-standard" frequency. Even if "FastMode++" is
> > invented most devices will probably want to use it.
> 
> Not really, there's no reason to make your i2c bus go fastfastfast if
> the devices on the other end can't cope with it
> 
> > And the voltage
> > requirements we're currently talking about here like "low svs" during
> > camera use cases are kind of negligible compared to others too.
> 
> Again, this is an I2C controller that seems to be associated with
> cameras.. No image data has to actually be processed for the
> communications to take place and you can attach any odd device
> 

My point is: In the unlikely case that support for faster I2C speeds is
added in newer SoCs, I think you'd just get a new "standard" base clock
frequency, add a new cci_data struct with adjusted timings and everyone
will use that (even for the lower I2C speeds). I doubt anyone will
bother adjusting and validating this for just one "corner"/voltage level
less. There are much more effective targets for power optimization than
the few bytes of I2C communication. :-)

Thanks,
Stephan