[PATCH 1/3] mull'ify multiplication with HZ in __const_udelay() [Was: Re: Too much error in __const_udelay() ?]

[Dominik Brodowski <linux@dominikbrodowski.de>]

John Stultz mentioned on lkml ( http://lkml.org/lkml/2004/6/5/15 ) that
calls to udelay() don't delay long enough, causing trouble e.g. in the USB
subsystem. The following patches address this issue.

Move the multiplication of (loops_per_jiffy * xloops) with HZ into
the "mull" asm operation. This increases the accuracy of the delay functions
largely:

n usec delay on a system with loops_per_jiffy = 1500000 : 

  n 	   before	  after
  1	 1000 ticks	 1499 ticks
 10	14000 ticks	14999 ticks

n usec delay on a system with loops_per_jiffy = 100000 : 

 n 	   before	  after
  1	    0 ticks	   99 ticks
 10	    0 ticks	  999 ticks
100	 9000 ticks	 9999 ticks

As noted by Kurt Garloff, it's necessary to adjust for large loops_per_jiffies,
as the multiplication of it with HZ fails for 4GHz or larger. So, John Stultz
suggested multiplying xloops with 4 first, and multiplying with (HZ/4).

Signed-off-by: Dominik Brodowski <linux@brodo.de>


diff -ruN linux-original/arch/i386/lib/delay.c linux/arch/i386/lib/delay.c
--- linux-original/arch/i386/lib/delay.c	2004-06-14 18:20:27.000000000 +0200
+++ linux/arch/i386/lib/delay.c	2004-06-15 07:48:57.302279400 +0200
@@ -31,10 +31,11 @@
 inline void __const_udelay(unsigned long xloops)
 {
 	int d0;
+	xloops *= 4;
 	__asm__("mull %0"
 		:"=d" (xloops), "=&a" (d0)
-		:"1" (xloops),"0" (current_cpu_data.loops_per_jiffy));
-        __delay(xloops * HZ);
+		:"1" (xloops),"0" (current_cpu_data.loops_per_jiffy * (HZ/4)));
+        __delay(xloops);
 }
 
 void __udelay(unsigned long usecs)