Re: [PATCH] i2c-mux-pca954x: allow downstream bus numbers to be specified in the dts

[Laurent Pinchart <laurent.pinchart-ryLnwIuWjnjg/C1BVhZhaw@public.gmane.org>]

Hi Frank,

On Thursday 17 April 2014 16:26:04 Frank Bormann wrote:
> A user-space application, which is being configured with the first mux bus
> number, executes i2cget/i2cset commands, specifying the i2c bus number and
> expecting them to be incremental from the first number.

I won't venture to comment on whether bus numbering has ever been considered 
to be a kernel ABI, other people should be able to comment on that. Any chance 
to fix the application to find the bus numbers dynamically ?

> On Thu, Apr 17, 2014 at 4:15 PM, Laurent Pinchart wrote:
> > On Thursday 17 April 2014 13:54:44 Frank Bormann wrote:
> > > Hi Laurent,
> > > 
> > > I have a pca9546 on one of my i2c buses. This will create four mux buses
> > > in Linux. One of those mux buses then has a pca9542 connected to it,
> > > creating another two mux buses on top of the first ones.
> > > 
> > > The 3.8 kernel, I was initially using enumerated, this as:
> > > 
> > > pca9546: 3, 4, 5, 6
> > > pca9542: 7, 8
> > > 
> > > However, the new 3.12 kernel, I am using now, has changed that
> > > enumeration to:
> > > 
> > > pca9546: 3, 4, 7, 8
> > > pca9542: 5, 6
> > > 
> > > As you may imagine, the pca9542 is connected to bus 4. Apparently,
> > > previously, it would finish the initalization of pca9546 before dealing
> > > with the pca9542. Now it seems to do the pca9542 initialization as soon
> > > as it sees it on mux bus 4. My application however expects the pca9546
> > > buses to be on incremental bus numbers.
> > 
> > I understand that the bus numbers changed, but you still haven't explain
> > *why* you need to have fixed bus numbers. Why do you need to know the bus
> > number at all ?

-- 
Regards,

Laurent Pinchart