Re: Overlapping ioremap() calls, set_memory_*() semantics

[Toshi Kani <toshi.kani@hpe.com>]

On Thu, 2016-03-10 at 22:47 -0800, Andy Lutomirski wrote:
> On Mon, Mar 7, 2016 at 9:03 AM, Toshi Kani <toshi.kani@hpe.com> wrote:
> > Let me try to summarize...
> > 
> > The original issue Luis brought up was that drivers written to work
> > with MTRR may create a single ioremap range covering multiple cache
> > attributes since MTRR can overwrite cache attribute of a certain
> > range.  Converting such drivers with PAT-based ioremap interfaces, i.e.
> > ioremap_wc() and ioremap_nocache(), requires a separate ioremap map for
> > each cache attribute, which can be challenging as it may result in
> > overlapping ioremap ranges (in his term) with different cache
> > attributes.
> > 
> > So, Luis asked about 'sematics of overlapping ioremap()' calls.  Hence,
> > I responded that aliasing mapping itself is supported, but alias with
> > different cache attribute is not.  We have checks in place to detect
> > such condition.  Overlapping ioremap calls with a different cache
> > attribute either fails or gets redirected to the existing cache
> > attribute on x86.
> 
> A little off-topic, but someone reminded me recently: most recent CPUs
> have self-snoop.  It's poorly documented, but on self-snooping CPUs, I
> think that a lot of the aliasing issues go away.  We may be able to
> optimize the code quite a bit on these CPUs.

Interesting.  I wonder how much we can rely on this feature.  Yes, by
looking at Intel SDM, it is indeed poorly documented. :-( 

> I also wonder whether we can drop a bunch of the memtype tracking.
> After all, if we have aliases of different types on a self-snooping
> CPU and /dev/mem is locked down hard enough, we could maybe get away
> with letting self-snoop handle all the conflicts.
> 
> (We could also make /dev/mem always do UC if it would help.)

It'd be interesting to know how it performs on an aliased map when it works
correctly... 

Thanks,
-Toshi