Patch to kill ioremap_mm

[Benjamin Herrenschmidt]

On Mon, 2005-05-09 at 10:43 -0500, John Rose wrote:
> Hi Ben-
> 
> > Do we need that for normal mmio ioremap mappings or only for PHB IO
> > space ? 
> 
> Just the latter.
> 
> > As I said, the later would stay separate, but I feel we don't
> > need the imalloc infrastructure to handle it. We can probably just
> > directly set/invalidate PTEs for the ranges when needed.
> 
> So you won't use vmalloc to manage the explicit mappings for the PHB
> ranges?  How will you keep up with what's been mapped for a PHB?  Seems
> like this would be necessary to know upon removal.  Or will you simply
> invalidate for the entire PHB range, even for subranges that aren't
> mapped?
> 
> Sweating the details :)

Well, for one, I'm not 100% sure we actually need to remove the linux
PTEs on removal. We only need to remove the hash entries, which can be
done "preventively" easily enough. When a segment is removed, we kick
out hash entries for that segment. If somebody tries to access that
later one, it gets a sigbus since HV will fail inserting an entry.

Ben.