TSC and real-time clock slippage with 2.6.2

TSC and real-time clock slippage with 2.6.2

[Ian Chard]

Hi,

Ever since I upgraded from 2.4.20 to the 2.6 tree, I've had a problem
with real-time clock slippage and hard hangs on my Athlon XP 2500+
(1830MHz according to /proc/cpuinfo).  I've kept an eye on the list and
have applied new patches as the problem seems to be known about, but as
the problem's still there with 2.6.2 I thought it was about time I
reared my ugly head.

At or shortly after boot time, I get the "Losing too many ticks!"
message (this seems to be related to how hard the system is working --
if it runs an fsck, the message appears immediately).  Then, while the
system is running, the real-time clock will lose time: the more jobs use
the CPU, the more time I lose.  Occasionally, the system will oops or
hard-hang altogether (which could be an unrelated driver issue; it is
pretty unusual).

I'm willing to test any patches you clever folk want to throw at me, or
alternatively if there's an easy solution I'll try anything.

Please cc: me on any replies, as my phone line is made of wet string and
would never cope with being subscribed!

Thanks for any help
- Ian

-- 
Ian Chard <ian@chard.org>
"Resting" Unix sysadmin and RHCE
near Linlithgow, central Scotland  |  sedimentation fault - beach dumped

Re: TSC and real-time clock slippage with 2.6.2

[john stultz]

On Wed, 2004-02-04 at 05:34, Ian Chard wrote:
> Ever since I upgraded from 2.4.20 to the 2.6 tree, I've had a problem
> with real-time clock slippage and hard hangs on my Athlon XP 2500+
> (1830MHz according to /proc/cpuinfo).  I've kept an eye on the list and
> have applied new patches as the problem seems to be known about, but as
> the problem's still there with 2.6.2 I thought it was about time I
> reared my ugly head.
> 
> At or shortly after boot time, I get the "Losing too many ticks!"
> message (this seems to be related to how hard the system is working --
> if it runs an fsck, the message appears immediately).  Then, while the
> system is running, the real-time clock will lose time: the more jobs use
> the CPU, the more time I lose.  Occasionally, the system will oops or
> hard-hang altogether (which could be an unrelated driver issue; it is
> pretty unusual).
> 
> I'm willing to test any patches you clever folk want to throw at me, or
> alternatively if there's an easy solution I'll try anything.

Are you using the amd76x_pm module?

Also, does time still drift backward with the following patch?

Could you also send me your dmesg and ntp drift file for both with and
without this patch?

thanks
-john

===== arch/i386/kernel/timers/timer_tsc.c 1.35 vs edited =====
--- 1.35/arch/i386/kernel/timers/timer_tsc.c	Wed Jan  7 00:31:11 2004
+++ edited/arch/i386/kernel/timers/timer_tsc.c	Tue Jan 20 13:22:54 2004
@@ -226,7 +226,7 @@
 	delta += delay_at_last_interrupt;
 	lost = delta/(1000000/HZ);
 	delay = delta%(1000000/HZ);
-	if (lost >= 2) {
+	if (0 && (lost >= 2)) {
 		jiffies_64 += lost-1;
 
 		/* sanity check to ensure we're not always losing ticks */