Re: Cache management

[Russell King - ARM Linux <linux@xxxxxxxxxxxxxxxx>]

On Wed, Dec 12, 2007 at 12:33:43PM +0100, Dag-Erling Smørgrav wrote:
> Russell King - ARM Linux <linux@xxxxxxxxxxxxxxxx> writes:
> > Dag-Erling Smørgrav <des@xxxxxxxxx> writes:
> > > It turns out that the name of that function is a lie.
> > Utter rubbish.  They are mapped uncached.
> 
> This isn't really helpful.

Same could be said about your comments - "It turns out that the
name of that function is a lie." _is_ utterly wrong.

Both ioremap_nocache() and ioremap() will _never_ _ever_ create a
mapping with any of the X, C, or B bits in the page table set.  That
means the mapping is neither marked as cacheable or bufferable.

If those functions did create cacheable mappings, the kernel would
fall over, and this list would be flooded with complaints that ARM
kernels were broken.  Things like ethernet drivers just wouldn't work.

The fact that they do work indicates that those functions are working
correctly.

So, the conclusion is that whatever effect you're seeing must either
be down to some peculiar quirk of your hardware, or a bug in your code.

Going around saying that function names are lies is not helpful.

-------------------------------------------------------------------
List admin: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm
FAQ:        http://www.arm.linux.org.uk/mailinglists/faq.php
Etiquette:  http://www.arm.linux.org.uk/mailinglists/etiquette.php