x86: add rdtsc barrier to TSC sync check

Impact: fix incorrectly marked unstable TSC clock

Patch (commit 0d12cdd "sched: improve sched_clock() performance") has
a regression on one of the test systems here.

With the patch, I see:

 checking TSC synchronization [CPU#0 -> CPU#1]:
 Measured 28 cycles TSC warp between CPUs, turning off TSC clock.
 Marking TSC unstable due to check_tsc_sync_source failed

Whereas, without the patch syncs pass fine on all CPUs:

 checking TSC synchronization [CPU#0 -> CPU#1]: passed.

Due to this, TSC is marked unstable, when it is not actually unstable.
This is because syncs in check_tsc_wrap() goes away due to this commit.

As per the discussion on this thread, correct way to fix this is to add
explicit syncs as below?

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

diff --git a/arch/x86/kernel/tsc_sync.c b/arch/x86/kernel/tsc_sync.c
index 9ffb01c31c40a8c9083e9949d065a442a62f46b7..1c0dfbca87c18a05beb42cebaa8e9ffed6d508a1 100644 (file)
--- a/arch/x86/kernel/tsc_sync.c
+++ b/arch/x86/kernel/tsc_sync.c
@@ -46,7 +46,9 @@ static __cpuinit void check_tsc_warp(void)
        cycles_t start, now, prev, end;
        int i;
 
+       rdtsc_barrier();
        start = get_cycles();
+       rdtsc_barrier();
        /*
         * The measurement runs for 20 msecs:
         */
@@ -61,7 +63,9 @@ static __cpuinit void check_tsc_warp(void)
                 */
                __raw_spin_lock(&sync_lock);
                prev = last_tsc;
+               rdtsc_barrier();
                now = get_cycles();
+               rdtsc_barrier();
                last_tsc = now;
                __raw_spin_unlock(&sync_lock);