Re: [PATCH v1] i2c: npcm7xx: Support changing bus speed using debugfs.

From: Andy Shevchenko <andy.shevchenko@gmail.com>
To: Avi Fishman <avifishman70@gmail.com>
Cc: Alex Qiu <xqiu@google.com>,
	Benjamin Fair <benjaminfair@google.com>,
	OpenBMC Maillist <openbmc@lists.ozlabs.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Tali Perry <tali.perry1@gmail.com>, Wolfram Sang <wsa@kernel.org>,
	Linux I2C <linux-i2c@vger.kernel.org>,
	Tomer Maimon <tmaimon77@gmail.com>
Subject: Re: [PATCH v1] i2c: npcm7xx: Support changing bus speed using debugfs.
Date: Thu, 1 Oct 2020 21:51:17 +0300	[thread overview]
Message-ID: <CAHp75Ver-O8=3eKHsHoKgpdd0sBQhLiC+UAqM=4C-gUfMBteuA@mail.gmail.com> (raw)
In-Reply-To: <CAKKbWA62G+B7p-bc7TRoa22iJNGdTUaopQVj3S3_FHh43ntasA@mail.gmail.com>

On Thu, Oct 1, 2020 at 9:42 PM Avi Fishman <avifishman70@gmail.com> wrote:
>
> Tali indeed pointed our major customers (Alex represent one of them :)
> that this feature must be handled carefully since it may break the
> communication and they are aware of that. Nevertheless they still want
> this feature, they already reviewed this and accepted it (in internal
> mails)
>
> So we will appreciate if this will be accepted.
>
> On Thu, Oct 1, 2020 at 9:27 PM Alex Qiu <xqiu@google.com> wrote:
> >
> > On Thu, Oct 1, 2020 at 10:41 AM Andy Shevchenko
> > <andy.shevchenko@gmail.com> wrote:
> > >
> > > On Thu, Oct 01, 2020 at 08:13:49PM +0300, Avi Fishman wrote:
> > > > Hi Andy,
> > > >
> > > > Customers using BMC with complex i2c topology asked us to support
> > > > changing bus frequency at run time, for example same device will
> > > > communicate with one slave at 100Kbp/s and another with 400kbp/s and
> > > > maybe also with smae device at different speed (for example an i2c
> > > > mux).
> > > > This is not only for debug.
> > >
> > > The above design is fragile to start with. If you have connected peripheral
> > > devices with different speed limitations and you try to access faster one the
> > > slower ones may block and break the bus which will need recovery.
> > >
> >
> > Hi Andy,
> >
> > To clarify, we are using a single read-only image to support multiple
> > configurations, so the supported bus rate of the devices are not known
> > at compile time, but at runtime. We start with 100 kHz, and go 400 kHz
> > if applicable. FYI, we are using 5.1 kernel, however I don't know much
> > about DT overlay.

I see. So, there are following statements:
 - the elaboration is good but I guess needs to be added somewhere in
form of the documentation
 - the internal schedules or so are not crucial for the upstream (it
rather sounds like a bribing the judge)
 - the current approach, if I'm not mistaken, is using debugfs, which
is not ABI and it's good
 - I'm not a maintainer here, but I don't like the approach

Let the maintainer decide.

-- 
With Best Regards,
Andy Shevchenko