> From: alan@lxorguk.ukuu.org.uk (Alan Cox)
> Date: Wed, 4 Jun 1997 20:30:21 +0100 (BST)
>
> > 1) The memory mapped by 'vremap' is marked cacheable. This
> > is (almost certainly) incorrect as it is memory that is
> > actually device memory. Note that the mmap call used by
> > X servers marks the memory as uncached.
>
> The PC architecture is supposed to handle this for you. It could well
> be an issue on MIPS and friends. The X one is hack and shouldnt be needed
>
> Indeed, I recall reading somewhere in the 440FX Natoma (or 430HX) chipset
> documentation that there was no caching on any PCI memory accesses.
last i recall the mechanizm is two-sided, once there are possible MTTR
settings for a memory area, second there is the caching bit in the page
table entry. The CPU uses the worst setting.
so if you have a BIOS that say sets some more clever MTTR value for
certain memory ranges, it might make a difference wether you set the page
table entry caching or nocaching. [there was one report that
ioremap_nocache() on 2.1 made difference on a high end device driver.]
-- mingo