x86/tsc: Let high latency PIT fail fast in quick_pit_calibrate()
authorAdrian Hunter <adrian.hunter@intel.com>
Wed, 3 Jun 2015 07:39:46 +0000 (10:39 +0300)
committerThomas Gleixner <tglx@linutronix.de>
Mon, 6 Jul 2015 07:41:00 +0000 (09:41 +0200)
If it takes longer than 12us to read the PIT counter lsb/msb,
then the error margin will never fall below 500ppm within 50ms,
and Fast TSC calibration will always fail.

This patch detects when that will happen and fails fast. Note
the failure message is not printed in that case because:
1. it will always happen on that class of hardware
2. the absence of the message is more informative than its
presence

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Len Brown <lenb@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Link: http://lkml.kernel.org/r/556EB717.9070607@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
arch/x86/kernel/tsc.c

index 5054497..7437b41 100644 (file)
@@ -598,10 +598,19 @@ static unsigned long quick_pit_calibrate(void)
                        if (!pit_expect_msb(0xff-i, &delta, &d2))
                                break;
 
+                       delta -= tsc;
+
+                       /*
+                        * Extrapolate the error and fail fast if the error will
+                        * never be below 500 ppm.
+                        */
+                       if (i == 1 &&
+                           d1 + d2 >= (delta * MAX_QUICK_PIT_ITERATIONS) >> 11)
+                               return 0;
+
                        /*
                         * Iterate until the error is less than 500 ppm
                         */
-                       delta -= tsc;
                        if (d1+d2 >= delta >> 11)
                                continue;