Re: CVS commit: src/share/man/man4

[Simon Burge <simonb%NetBSD.org@localhost>]

Paul Goyette wrote:

> On Fri, 12 Sep 2008, Simon Burge wrote:
> 
> >>>> Document new I2C_SCAN option.
> >>>
> >> Added the following warning:
> >>
> >>    Scan each i2c bus to determine which addresses respond.  Note
> >>    that using this option can access some devices in such a manner
> >>    as to leave them in an unstable or unuseable state, and might
> >>    prevent those devices from being properly matched and/or
> >>    attached.
> >
> > Can you expand this a bit more?  Leaving some devices in an "unstable or
> > unuseable state" is rather an understatement - enabling this option will
> > stop my Dell 640m laptop from rebooting.  I suspect there's some sort of
> > power controller on the I2C bus that we blindly try to probe and then
> > leave in some state that renders it useless.
> >
> > From what I understand, there's no generic way to probe an I2C bus or
> > SMbus, and therefore any unknown device(s) that are probed with the
> > current code might be accessed in a way that could do anything that we
> > can't predict.
> >
> > I'd prefer to see some large warning along the lines of "If you enable
> > this, you may experience funny problems, up to and including not being
> > able to reboot your machine (or even worse).  Don't enable this unless
> > you really know what you're doing.".
> 
> How about if I reword it like this:
> 
>       Scan each i2c bus to determine which addresses respond.
> 
>       WARNING! Using this option can access some devices in such a
>       manner as to leave them in an unstable or unuseable state, and
>       can prevent those devices from being properly matched and/or
>       attached.  It can also lock up the entire i2c bus and even
>       prevent a machine from completing the boot process.  Don't
>       use this option unless you know what you're doing and can
>       accept all sorts of unforseen consequences.

Looks good.

> It would really be nice if system/motherboard vendors could tell us
> what they've put on the bus so we can actually behave intelligently.
> Even better, it would've been nice if the i2c folks had thought things
> through a bit more and insisted on some minimal device identification 
> protocol.  (Of course, that would've made things more expensive, I 
> guess.)

Exactly (the "more expensive" comment)!

Cheers,
Simon.