Re: 2.3.43 alpha broken with >2g of ram

[Linus Torvalds]

On Sat, 12 Feb 2000, Richard Henderson wrote:
> 
> There aren't many of them, btw, and afaik none of the affected
> systems can even physically _hold_ more than 2G.

Note that I do agree with Andrea in that it is much too draconian to
suddenly require that all drivers be updated to a new interface that is
very different from any old one. 

I would suggest keeping the 2G limit, and just allowing a kernel command
line to override it.

Alternatively, use the same "highmem" support for it that x86 uses for
>4GB pages: anybody using the normal memory allocators would use mem < 2G,
while the high-mem logic comes into play with any allocations in the 2G+
range.

Using highmem will limit the affected drivers a LOT (regular allocations
would never be 2G+), while still allowing a clean way of using high memory
(kmap/kunmap become no-ops, because all of the memory is actually
MAPPABLE, but the bounce buffer logic still gets invoced).

		Linus