Messages in this thread | | | Date | Fri, 30 May 2008 17:15:46 +0200 | From | "Jiri Hladky" <> | Subject | Bug in arch/i386/lib/delay.c file, delay_loop function |
| |
Hi all,
when trying to understand how Bogomips are implemented I have found bug in arch/i386/lib/delay.c file, delay_loop function
/* simple loop based delay: */ static void delay_loop(unsigned long loops) { int d0;
__asm__ __volatile__( "\tjmp 1f\n" ".align 16\n" "1:\tjmp 2f\n" ".align 16\n" "2:\tdecl %0\n\tjns 2b" :"=&a" (d0) :"0" (loops)); }
The function fails for loops > 2^31+1. It because SF is set when dec returns numbers > 2^31
The fix is to use jnz instruction instead of jns (and add one decl instruction to the end to have exactly the same number of loops as in original version):
__asm__ __volatile__( "\tjmp 1f\n" ".align 16\n" "1:\tjmp 2f\n" ".align 16\n" "2:\tdecl %0\n\tjnz 2b\n" "decl %0" :"=&a" (d0) :"0" (loops));
IMHO, d0 is not needed at all so that we can further simplify the code: static void delay_loop(unsigned long loops) __asm__ __volatile__( "\tjmp 1f\n" ".align 16\n" "1:\tjmp 2f\n" ".align 16\n" "2:\tdecl %0\n\tjnz 2b\n" "decl %0" :/*we don't need output */ :"a" (loops)); }
I will attach three small C-program to test it delay-orig.c - original loop from kernel source code delay-fixed.c - fixed loop delay-fixed1.c - fixed loop without d0 variable
Outputs: ============== delay-orig.c ================== time delay-orig 2147483649 loops 2147483649 loops 2147483649 do -2147483648
real 0m0.002s user 0m0.000s sys 0m0.000s
================== delay-fixed.c ============= time delay-fixed 2147483649 loops 2147483649 loops 2147483649 do -1
real 0m1.025s user 0m1.024s sys 0m0.000s
========== delay-fixed1.c ===================== time delay-fixed1 2147483649 loops 2147483649 loops 2147483649
real 0m1.073s user 0m1.060s sys 0m0.004s
and update kernel source file arch/i386/lib/delay.c. Please let me know if these modifications make sense.
Thanks a lot Jiri [unhandled content-type:application/x-bzip2] | |