Re: [PATCH, RFC] Earlier I2C initialization

[David Brownell <david-b@pacbell.net>]

On Wednesday 11 June 2008, Maciej W. Rozycki wrote:
> > That being said, I'm not sure if the comparison with the PCI subsystem
> > holds... I am under the impression that PCI bus handling doesn't
> > require dedicated drivers? At least I can't see any under drivers/pci.
> 
>  Of course it does require them.  It is just due to their very nature they
> tend to be placed under arch/,

PCI root hubs, yes.  The drivers/pci/hotplug bridges are slightly
more generic, ditto drivers/pci/pcie and the CardBus bridges in
drivers/pcmcia.  At one point, lack of a generic (non-hotplug) PCI
bridge driver was viewed as a weakness of that driver stack.

Also, drivers/acpi/pci_root.c binds the root at subsys_initcall.
That's done *after* some earlier PCI magic; I never bothered
to sort through that little maze.


> although there are some cases where the 
> same system controller can be used for a range of processors (e.g. some
> Marvell chips can be used either with MIPS or PowerPC CPUs) and they might
> be arguably put in a place more suitable for sharing between
> architectures.  See arch/mips/pci/ for an example of a generous bunch of
> PCI host drivers.

Which, for the record, get very early initialization using two
different mechanisms:

  - many use arch_initcall()

  - the "arch" subtree is linked before the "drivers" subtree

I don't think I2C needs to worry about arch_initcall just now,
but if necessary it could initialize earlier than subsys_initcall.

- Dave