On 01/10/2012 09:15 PM, Linus Torvalds wrote:
> So the wireless on my Macbook Air no longer seems to work with the
> current -git tree.
> 
> The BRCMSMAC driver *used* to work, and no longer does. Also, very
> annoyingly, it's even hard to *compile* the thing, because it used to
> be
> 
>   depends on BCMA=n
> 
> but now it is the exact reverse:
> 
>   depends on BCMA
> 
> so there is no sane way to carry a configuration over from before, and
> things like bisection is a major pain due to having to play idiotic
> configuration games to get it to work across all these modifications.

BCMA support has been added to brcmsmac so now we depend on it. As BCMA
claims the PCI device we had the BCMA=n in place before. I agree that it
makes bisecting a pain in the...

> Ragardless, even once you actually enable BCMA and can get the driver
> to come back, it just doesn't do anything. Maybe there is some
> remaining config problem, but I don't think so. I think the driver is
> just buggered.

Looking at the output I had deja-vu feeling. The issue popped up when we
published part of the patches for BCMA support. I looked into it with
Larry Finger and Rafał Miłecki. We found the reason for the failure, but
it seemed resolved after submitting the remaining patches for BCMA
support. However, it seems to have raised its ugly head.

> Please look into this. I'll obviously be happy test any reasonable suggestions,
> 
>                  Linus
> 

My theory is that the BAR window is not mapped to a valid address. Could
you try to revert the following commit:

commit 439678f8b0fca7aeca06c6581e3679eef618721a
Author: Rafał Miłecki <zajec5@gmail.com>
Date:   Mon Dec 5 19:13:39 2011 +0100

    bcma: pci: use fixed windows when possible

    Some cores are mapped in the fixed way, they registers can be accessed
    all the time.

    Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
    Signed-off-by: John W. Linville <linville@tuxdriver.com>

I have attached the 'revert patch' based on your git master branch.

Gr. AvS