[ILUG] kernel programming curiosity

[Fergal Daly]

Thanks, I have it now, but now my question is: would they not be better off
replacing

#define get_user(x,ptr)             \
({  int __ret_gu,__val_gu;            \
  switch(sizeof (*(ptr))) {         \  
  case 1:  __get_user_x(1,__ret_gu,__val_gu,ptr); break;    \
  case 2:  __get_user_x(2,__ret_gu,__val_gu,ptr); break;    \
  case 4:  __get_user_x(4,__ret_gu,__val_gu,ptr); break;    \
  default: __get_user_x(X,__ret_gu,__val_gu,ptr); break;    \
  }               \
  (x) = (__typeof__(*(ptr)))__val_gu;       \
  __ret_gu;             \
})

with

#define get_user(x,ptr)             \
({  int __ret_gu,__val_gu;            \
  __get_user_x(sizeof (*(ptr)),__ret_gu,__val_gu,ptr); \
  (x) = (__typeof__(*(ptr)))__val_gu;       \
  __ret_gu;             \
})

Benefits:
- gcc doesn't have to optimise away the other __put_users because
  they are never there in the first place.
- calling put_user with a pointer to something of the wrong size still
  breaks just as badly as it used to, except you now have a hint to what
  broke it because insmod will complain about __put_user_8 not existing
- if other sizes become legal you don't have make any changes to this
  header file
- less code

Or am I missing something?

Fergal


On Sun, Apr 22, 2001 at 03:59:22PM +0100, Mark McLoughlin wrote:
> On Sat, 21 Apr 2001, Fergal Daly wrote:
> 
> > The bit that's still really confusing me is where is gcc getting the capital
> > X from, I have grepped all the .h and .c files of 2.2.17 and 2.2.19 and
> > there is no mention of put_user_X, only put_user_x. Any idea what that's
> > about?
> 
> I think the whole point is there is no put_user_X. So if you try and
> do a put_user with something that's more the 4 bytes, it won't link.
> If you look at the cpp output, it's the default branch
> 
> default: __asm__ __volatile__("call __put_user_" "X"
> 
> This doesn't work without optimising (-fthread-jumps, I think) turned
> on because with this turned on all defunct branches of the switch get
> optimised away so there are references to put_user_{1,2,4,X}. If you
> do have it turned on, you're left with only a reference to the
> put_user you need, and if that's put_user_X you're screwed....
> 
> Good Luck,
> Mark.
> 
> 
> -- 
> Irish Linux Users' Group: ilug at linux.ie
> http://www.linux.ie/mailman/listinfo/ilug for (un)subscription information.
> List maintainer: listmaster at linux.ie

--