This is targeted at 2.4.20, but should apply cleanly to any of the 2.5
kernels too.
The clock frequency found in the Alpha HWRPB may be significantly off from
the actual frequency. The existing test is not stringent enough to weed out
errors that will swamp ntp; this patch reduces the allowable error from 1%
(10,000 PPM) to 250 PPM.
Kudos to Hal Murray <[email protected]> for insightful guidance;
tested on
my 3-way Alphaserver 4100 5/466 with very happy results. :-)
--- kernel-source-2.4.20/arch/alpha/kernel/time.c 2002-02-27
05:12:19.000000000 -0500
+++ rsb-patch/arch/alpha/kernel/time.c 2003-06-05 12:51:10.000000000 -0400
@@ -24,6 +24,8 @@
* 2000-08-13 Jan-Benedict Glaw <[email protected]>
* Fixed time_init to be aware of epoches != 1900. This prevents
* booting up in 2048 for me;) Code is stolen from rtc.c.
+ * 2003-06-03 R. Scott Bailey <[email protected]>
+ * Tighten sanity in time_init from 1% (10,000 PPM) to 250 PPM
*/
#include <linux/config.h>
#include <linux/errno.h>
@@ -288,7 +290,7 @@
time_init(void)
{
unsigned int year, mon, day, hour, min, sec, cc1, cc2, epoch;
- unsigned long cycle_freq, one_percent;
+ unsigned long cycle_freq, tolerance;
long diff;
/* Calibrate CPU clock -- attempt #1. */
@@ -306,13 +308,13 @@
cycle_freq = hwrpb->cycle_freq;
if (est_cycle_freq) {
- /* If the given value is within 1% of what we calculated,
+ /* If the given value is within 250 PPM of what we
calculated,
accept it. Otherwise, use what we found. */
- one_percent = cycle_freq / 100;
+ tolerance = cycle_freq / 4000;
diff = cycle_freq - est_cycle_freq;
if (diff < 0)
diff = -diff;
- if (diff > one_percent) {
+ if (diff > tolerance) {
cycle_freq = est_cycle_freq;
printk("HWRPB cycle frequency bogus. "
"Estimated %lu Hz\n", cycle_freq);