Date: Sat, 17 Feb 2024 20:30:03 +0000
From: Conor Dooley <conor@...nel.org>
To: Guenter Roeck <linux@...ck-us.net>
Cc: Naresh Solanki <naresh.solanki@...ements.com>,
	Jean Delvare <jdelvare@...e.com>, Rob Herring <robh+dt@...nel.org>,
	Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
	Conor Dooley <conor+dt@...nel.org>, mazziesaccount@...il.com,
	linux-hwmon@...r.kernel.org, devicetree@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] dt-bindings: hwmon: tda38640: Add interrupt &
 regulator properties

On Sat, Feb 17, 2024 at 07:57:43AM -0800, Guenter Roeck wrote:
> On 2/15/24 03:48, Conor Dooley wrote:
> > On Wed, Feb 14, 2024 at 05:17:04PM -0800, Guenter Roeck wrote:
> > > On 2/14/24 11:55, Conor Dooley wrote:
> > > [ ... ]
> > > > > > Why "vout0" if there's only one output? Is it called that in the
> > > > > > documentation? I had a quick check but only saw it called "vout".
> > > > > > Are there other related devices that would have multiple regulators
> > > > > > that might end up sharing the binding?
> > > > > > 
> > > > > 
> > > > > Primarily because that is what the PMBus core generates for the driver
> > > > > because no one including me was aware that this is unacceptable
> > > > > for single-output drivers.
> > > > 
> > > > Is it unacceptable? If you're implying that I am saying it is, that's
> > > > not what I was doing here - I'm just wondering why it was chosen.
> > > > Numbering when there's only one seems odd, so I was just looking for the
> > > > rationale.
> > > > 
> > > 
> > > Given the tendency of corporate speak (aka "this was a good attempt" for
> > > a complete screwup), and since this did come up before, I did interpret
> > > it along that line. My apologies if that was not the idea.
> > 
> > I'm not gonna go and decree that "vout0" is unacceptable, if it was
> > called that in documentation that I had missed or was convention, I was
> > just gonna say "okay, that sounds reasonable to me".
> > 
> 
> "convention" only if lack of awareness how regulators are supposed to be named
> is a convention.

They're "supposed" to be named whatever the binding says they are named,
but as we've discovered none of these devices actually have bindings
that allow regulators in the first place. I think they should be called
whatever they're called in the documentation for the device, which in
this case was "vout".

> > > Still, I really don't know how to resolve this for existing PMBus drivers
> > > which do register "vout0" even if there is only a single output regulator.
> > 
> > I had a quick look at that series, none of the devices that I checked
> > out there seem to have documented regulators at all. Some of the devices
> > were only documented in trivial-devices.yaml. Relying on the naming of
> > undocumented child nodes is a bug in those drivers & I guess nobody cares
> > about dtbs_check complaints for those platforms. The example that was
> > linked in the other thread doesn't even use a valid compatible :(
> > 	https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/arm/boot/dts/aspeed/aspeed-bmc-delta-ahe50dc.dts?id=8d3dea210042f54b952b481838c1e7dfc4ec751d#n21
> > I guess it uses the i2c device ids to probe on that platform, or have
> > I missed something there?
> > 
> 
> I think that is correct. If I recall correctly, the I2C subsystem no longer
> searches for compatible drivers by only looking at the device id in the
> compatible node, so I guess one has to list "lm25066" instead of "ti,lm25066"
> as compatible to get a match in the i2c subsystem. That is of course
> completely wrong.

If the driver is probing based on i2c_device_id matching, is it correct to
use DT to probe the regulators? (I don't know, that's not some sort of
rhetorical question).

Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)

Powered by blists - more mailing lists