2.6.9 PageAnon bug

[Mikael Starvik]

There is at least one architecture supported by 2.6.9 that has no alignment
restrictions what so ever and no struct padding added by the compiler. 

The patch named "rmaplock: PageAnon in mapping" in 2.6.9 doesn't work for
this architecture because it assumes that the address of a member in a
struct can't be odd. 

One possible but ugly patch below. Another possible patch would be to move
i_data above i_bytes and i_sock. I would really like a cleaner patch but I
guess its a bad idea to add a new field to struct page?

Index: fs.h
===================================================================
RCS file: /usr/local/cvs/linux/os/lx25/include/linux/fs.h,v
retrieving revision 1.20
retrieving revision 1.21
diff -r1.20 -r1.21
449c449,453
< 	struct address_space	i_data;
---
> 	/* The LSB in i_data below is used for the PAGE_MAPPING_ANON flag. 
> 	 * This assumes that the address of this member isn't odd which
> 	 * is not true for all architectures. Force the compiler to align
it.
> 	 */
> 	struct address_space	i_data __attribute__ ((aligned(4)));

Anyone who knows about similar usage of bit 0 and/or 1 in pointers
anywhere?

/Mikael 

PS. The architecture I'm referring to is CRIS but there may be more with the
same sloppyness regarding alignment. DS

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/