[POWERPC] Simplify smp_space_timers
authorwill schmidt <will_schmidt@vnet.ibm.com>
Fri, 11 May 2007 13:34:16 +0000 (23:34 +1000)
committerPaul Mackerras <paulus@samba.org>
Sat, 12 May 2007 01:32:47 +0000 (11:32 +1000)
Greatly simplify the function smp_space_timers.

The stolen time calculation (per comment within the code) doesn't need the
half-jiffy stagger any more.  There isn't an issue with bouncing off global
locks, so we really shouldn't need any sort of staggering at all.

However, the last_jiffy value still needs to be set.   This removes the
extra stagger logic, and just sets the values.

This change should benefit applications that rely on barrier
synchronization, and will help cut down OS jitter.

Boot tested across the board (G5,power3,power4,power5,970mp blade).

Signed-off-by: Will Schmidt <will_schmidt@vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
arch/powerpc/kernel/time.c

index 7cedef8f5f70c2ae397c057746200a02da31f5cf..2c8564d54e4d009a67527463d1aa2a8c7f3f9d6f 100644 (file)
@@ -711,30 +711,15 @@ void wakeup_decrementer(void)
 void __init smp_space_timers(unsigned int max_cpus)
 {
        int i;
-       unsigned long half = tb_ticks_per_jiffy / 2;
-       unsigned long offset = tb_ticks_per_jiffy / max_cpus;
        u64 previous_tb = per_cpu(last_jiffy, boot_cpuid);
 
        /* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
        previous_tb -= tb_ticks_per_jiffy;
-       /*
-        * The stolen time calculation for POWER5 shared-processor LPAR
-        * systems works better if the two threads' timebase interrupts
-        * are staggered by half a jiffy with respect to each other.
-        */
+
        for_each_possible_cpu(i) {
                if (i == boot_cpuid)
                        continue;
-               if (i == (boot_cpuid ^ 1))
-                       per_cpu(last_jiffy, i) =
-                               per_cpu(last_jiffy, boot_cpuid) - half;
-               else if (i & 1)
-                       per_cpu(last_jiffy, i) =
-                               per_cpu(last_jiffy, i ^ 1) + half;
-               else {
-                       previous_tb += offset;
-                       per_cpu(last_jiffy, i) = previous_tb;
-               }
+               per_cpu(last_jiffy, i) = previous_tb;
        }
 }
 #endif