Re: [PATCH] pmac_zilog: insert correct failure path for device numbers being taken

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Fri, 2004-07-16 at 21:05, Albert Cahalan wrote:
> > I'll talk to him at KS/OLS and see if we can come up with
> > some solution, this is actually a regression since 2.4
> > could "offset" macserial, so we could accomodate, for
> > example, a driver for a pcmcia modem _and_ the zilog ports.
> 
> That's the wrong way around. The zilog ports are always
> there, and thus could have stable numbers. The PCMCIA
> ports can not have stable numbers; they might be gone even.

Nice ideal vision ...

> In general, the platform-specific (motherboard, generally)
> ports should get to grab device numbers first. Anthing
> connected by a normally hot-plug bus goes last. Plain PCI
> is in the middle, because PCI cards are occasionally moved.

I'm not talking about the best solution that will make everybody
happy, but at this point in 2.6, whatever will fix the problem, I
doubt russel would accept a patch changing the 8250 driver in any
significant way

> For a PC, serial ports hanging off the motherboard's LPC bus
> (what amounts to built-in ISA) should go before serial ports
> that might be on normal PCI cards, which in turn go before
> those on PCMCIA.
-- 
Benjamin Herrenschmidt <benh@kernel.crashing.org>