On Tue, Jan 18, 2011 at 01:00:20PM -0800, Daniel Walker wrote:

Just greping the driver I found at least one writel() .. I'm not really
concerned with that tho. I use this driver, I need this driver to work.
You broke the driver, and your NAK'ing the fix ..

If drivers use documented internal APIs, they _will_ break, and that's
tough luck.  The advertised API (DMA-mapping) is what we guarantee to
support and fixup drivers for.

If you don't know that a driver is using an internal API then you can't
preempt it breaking.  The answer is drivers must not use internal APIs.

Had the driver been properly reviewed before being merged, it should
have been caught before it was merged.

I've given you the outline of a proper fix, and you've probably already
spent longer discussing it than it would take you to cut'n'paste the
code into kernel and test it...