Re: [PATCH 1/7] dt-bindings: Add DT bindings documentation for Allwinner Thermal Sensor Controller

From: Maxime Ripard
Date: Wed Aug 29 2018 - 09:32:02 EST


On Fri, Aug 24, 2018 at 09:59:21PM +0200, Emmanuel Vadot wrote:
>
> Hi,
>
> On Fri, 24 Aug 2018 16:58:40 +0200
> Maxime Ripard <maxime.ripard@xxxxxxxxxxx> wrote:
>
> > Hi,
> >
> > On Mon, Aug 20, 2018 at 04:27:15PM +0200, Emmanuel Vadot wrote:
> > > On Mon, 20 Aug 2018 16:07:37 +0200
> > > Maxime Ripard <maxime.ripard@xxxxxxxxxxx> wrote:
> > >
> > > > On Mon, Aug 20, 2018 at 07:41:22AM -0600, Rob Herring wrote:
> > > > > On Mon, Aug 20, 2018 at 5:17 AM Maxime Ripard <maxime.ripard@xxxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Sat, Aug 04, 2018 at 09:03:49AM +0200, Emmanuel Vadot wrote:
> > > > > > > This patch adds documentation for Device-Tree bindings for the Allwinner
> > > > > > > Thermal Sensor Controller found on the H3, H5 and A64 SoCs
> > > > > > >
> > > > > > > Signed-off-by: Emmanuel Vadot <manu@xxxxxxxxxxx>
> > > > > >
> > > > > > I'm not going to merge a binding for a device that doesn't have any
> > > > > > driver implemented at the moment in Linux.
> > > > >
> > > > > I'll take it then. Linux is not the only DT client.
> > > >
> > > > Then don't complain if we ever have to break the ABI. We never tested
> > > > that hardware, never had any code running on it, and it *will* cause
> > > > some issues. For example, the calibration data have never been used
> > > > and how they should be represented have never been described, since no
> > > > one ever actually tried to use it.
> > > >
> > > > And sure, Linux is not the only DT client. Just like FreeBSD isn't.
> > >
> > > What kind of data would you need for me to make things better ?
> > > We have the driver in FreeBSD for almost two years now, whne I started
> > > to see how to upstream it I noticed a lot of problems and spent 4 or 5
> > > days to try on all the SoCs mentionned in the serie, the only SoC I
> > > didn't include it the A83T as for some reason I couldn't make the
> > > driver work.
> >
> > Do you have a link to that driver?
>
> https://github.com/freebsd/freebsd/blob/master/sys/arm/allwinner/aw_thermal.c
>
> > How and how much did you test it?
>
> Using a cheap IR thermometer as it's all I can afford.
> Then running multiples cpufreq-a53 process and comparing the result.
> Did the same without using the calibration data, result where closer
> when I used them.

Ok.

> > Are you using the calibration data stored in the SID?
>
> Yes, which is why the serie contain the nvmem cell for the SID.

Ok.

> >
> > > I just hope that you understand that we cannot wait for Linux to
> > > have a driver to have some bindings.
> >
> > Just like I'm sure you can understand that just merging the DT bits
> > without testing anything caused some troubles in the past, and I don't
> > want to discover it in a year from now.
>
> I do, I don't like wrong DT info as much as you do I think. I had
> the unfortunate event to find that the sun4i-a10-timer compatible used
> in every soc is plain wrong for !A10 and !A13, patches comming soon.
> I want those bits merged but I also want that they reflect reality,
> this is not a hard driver to do, NetBSD also have one I think, maybe
> OpenBSD do to.

I guess these one went un-noticed because they are not the default
source of time keeping and events in the kernel.

Thanks!
Maxime

--
Maxime Ripard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

Attachment: signature.asc
Description: PGP signature