Re: __get_user usage in mm/slab.c

[Andi Kleen]

On Tue, Mar 12, 2002 at 12:58:53PM +0100, Roman Zippel wrote:
> Hi,
> 
> The way __get_user is currently used in mm/slab.c is not portable. It
> breaks on arch which have seperate user/kernel memory space. It still
> works during boot or from kernel threads, but /proc/slabinfo shows only 
> broken entries or if a module creates a slab cache, I got lots of
> warnings.
> We have to at least insert a "set_fs(get_fs())", but IMO a separate
> interface would be better. Any opinions?

I agree that a separate interface would be better, one that guarantees to
handle exceptions on the m68k and other archs with separate address spaces too.
I use that facility quite regularly in architecture specific code, sorry
for letting it slip into portable code. 
I guess set_fs(KERNEL_DS); __*_user() will not catch exceptions on m68k
currently, right? 

-Andi