x86: Fix sched_clock_cpu for systems with unsynchronized TSC
authorDimitri Sivanich <sivanich@sgi.com>
Mon, 1 Mar 2010 17:48:15 +0000 (11:48 -0600)
committerGreg Kroah-Hartman <gregkh@suse.de>
Thu, 1 Apr 2010 22:58:57 +0000 (15:58 -0700)
commit 14be1f7454ea96ee614467a49cf018a1a383b189 upstream.

On UV systems, the TSC is not synchronized across blades.  The
sched_clock_cpu() function is returning values that can go
backwards  (I've seen as much as 8 seconds) when switching
between cpus.

As each cpu comes up, early_init_intel() will currently set the
sched_clock_stable flag true.  When mark_tsc_unstable() runs, it
clears the flag, but this only occurs once (the first time a cpu
comes up whose TSC is not synchronized with cpu 0).  After this,
early_init_intel() will set the flag again as the next cpu comes
up.

Only set sched_clock_stable if tsc has not been marked unstable.

Signed-off-by: Dimitri Sivanich <sivanich@sgi.com>
Acked-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100301174815.GC8224@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
arch/x86/kernel/cpu/intel.c

index a2a03cf4a489966e793c9b14058f462d60a7e2eb..2a94890149dbddf143e53217f05147e156c2cfb8 100644 (file)
@@ -70,7 +70,8 @@ static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
        if (c->x86_power & (1 << 8)) {
                set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
                set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
-               sched_clock_stable = 1;
+               if (!check_tsc_unstable())
+                       sched_clock_stable = 1;
        }
 
        /*