Re: Linux 2.5.63

[george anzinger <george@mvista.com>]

Christoph Hellwig wrote:
>>George Anzinger <george@mvista.com>:
>>  o POSIX clocks & timers
> 
> 
> Care to explain what  FOLD_NANO_SLEEP_INTO_CLOCK_NANO_SLEEP
> is supposed to do?  It's always defined in signal.h, so we can
> aswell get rid of it..

It is there in case some one might want nanosleep to NOT be folded 
into clock_nanosleep.  For a while this was a moving target and I got 
a bit paranoid  :)  I see no real reason to keep it...
> 
> And what's this:
> 
> #ifndef div_long_long_rem
> +#include <asm/div64.h>
> +
> +#define div_long_long_rem(dividend,divisor,remainder) ({ \
> +                      u64 result = dividend;           \
> +                      *remainder = do_div(result,divisor); \
> +                      result; })
> +
> +#endif                         /* ifndef div_long_long_rem */
> 
> Any reason you can't just use do_div directly like everyone else? :)

Actually I have coded a better function as part of the expanded 
high-res-timers which does a 64-bit/32-bit div in a much cleaner way. 
  Again, it is part of the full high-res-timers patch.  I have been 
thinking of submitting the complete set of these math routines outside 
of the high-res patch.  They are designed to make scaled math easy.  I 
  have both a generic and a i386 header file, but they still need a 
bit of testing.

The issue is getting around the C limitation of not being able to use 
the div and mpy instructions that take 64-bits/32-bits and return 2 
32-bit results and the mpy which takes 2 32-bit operands and returns a 
64-bit result.

For scaled operations, they also roll a shift into the mix in an 
efficient way (i.e. a small inline asm function).


-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml