The sys_mmap2 syscall on m68k, cris, and mips

["H. Peter Anvin" <hpa@zytor.com>]

Hello, sorry for the wide distribution;

I am trying to track down the story on the sys_mmap2 system call. 
Currently, to the best of my knowledge, glibc, µClibc, and klibc 
(haven't looked at newlib recently) all assume that sys_mmap2() is the 
way to map large files on 32-bit architectures, and that the offset 
argument is always shifted over by 12.

As indicated by this post, this is the case on most architectures:

http://marc.theaimsgroup.com/?l=linux-kernel&m=114066586932670&w=2

The odd ones out are m68k, cris, and MIPS.  Ralf Baechle has indicated 
that non-4K-pages are currently experimental on MIPS, and that he would 
be open to changing MIPS to fit the rest of the pattern.  That leaves 
m68k and cris.

Thus, I would like to hear your opinion.  One opinion is to define that 
the shift is fixed, but architecture-specific, and introduce an 
MMAP2_PAGE_SHIFT #define that libraries can pick up.  (That would 
require m68k to change at least on the Sun3 platform, though.  I don't 
think there is a huge installed base there, though.)

	-hpa