[ath9k-devel] [PATCH 1/3] ath9k: Fix build error on ARM

[linux@arm.linux.org.uk (Russell King - ARM Linux)]

On Wed, Feb 05, 2014 at 04:32:46AM -0800, Joe Perches wrote:
> On Wed, 2014-02-05 at 11:50 +0000, Russell King - ARM Linux wrote:
> > On Tue, Feb 04, 2014 at 08:36:36AM -0800, Joe Perches wrote:
> > > On Tue, 2014-02-04 at 08:03 +0100, Holger Schurig wrote:
> > > > Joe, look in linux/arch/arm/include/asm/delay.h. The macro udelay
> > > > cannot handle large values because of lost-of-precision.
> > > > 
> > > > IMHO udelay on ARM is broken, because it also cannot work with fast
> > > > ARM processors (where bogomips >= 3355, which is in sight now). It's
> > > > just not broken enought that someone did something against it ...   so
> > > > the current kludge is good enought.
> > > 
> > > Maybe something like this would be better?
> > 
> > No, the point of __bad_udelay() is that people doing stupidly large
> > udelay()s result in build errors,
> 
> Apparently, people just convert stupidly large udelay()s
> to mdelay and not be bothered.

And that's the correct answer.  Having udelay(10000) rather than mdelay(10)
is a sign that they weren't paying that much attention when writing the
code.

> Perhaps there should be some runtime udelay > maximum supported check.

Having both a runtime check _and_ a compile time check would actually
be a good thing, but any runtime check needs to be suitably rate-
limited.

The compile time check is very important because it catches a lot of
cases which wouldn't otherwise be found (eg, in drivers which hardly
anyone uses on ARM.)

Maybe the compile time check should be something which is implemented
in a cross-architecture way in linux/delay.h with the maximum set to
the lowest that any architecture can do?

-- 
FTTC broadband for 0.8mile line: 5.8Mbps down 500kbps up.  Estimation
in database were 13.1 to 19Mbit for a good line, about 7.5+ for a bad.
Estimate before purchase was "up to 13.2Mbit".