On Sat, Oct 28, 2006 at 02:10:09PM -0700, David Brownell wrote:

On Saturday 28 October 2006 4:21 am, Christoph Hellwig wrote:

quoted

This is really awkward and against what we do in any other driver.

Awkward, yes -- which is why I posted the non-awkward version,
which is repeated below.  (No thanks to "diff" for making the
patch ugly though; the resulting code is clean and non-awkward,
moving that function helped.)

Against what other drivers do?  Since "usbnet.c" is infrastructure
code, not a driver, your comment can't apply.  Infrastructure uses
conditional compilation routinely in such cases.

But remember that the actual drivers follow the standard convention
("select MII") given Randy's patch #1 of 2.

Ah sorry - I missed that.

I still don't quite like the approach.  What about simply putting
the mii using functions into usbnet-mii.c and let makefile doing
all the work?  This would require a second set of ethtool ops,
but I'd actually consider that a cleanup, as it makes clear which
one we're using and allows to kill all the checks for non-mii
hardware in the methods.