EXA for radeon experimental patch

[Jesse Barnes]

On Thursday, September 1, 2005 11:59 am, Eric Anholt wrote:
> > Why this? The kernel has support for paging, so it could easily free
> > up some continuous pages just by swapping them out if they are used.
>
> It would be nice if it were true.  But getting physically contiguous
> pages any time after boot (of any significant size) is hard.  I'm
> guessing due to wired pages being sprinkled around?

Yep.  Over time, the kernel will allocate non-pageable memory all over 
the system, which makes getting contiguous regions after a few hours of 
uptime pretty difficult or impossible.  There is some work underway to 
mitigate that a bit, but it's not upstream yet.  And even if it was, we 
probably shouldn't count on it since it still won't *guarantee* that we 
get big contiguous chunks when we ask for them (the only way to do that 
would involve unacceptable performance losses on many platforms).

Jesse