From: Tobias Klauser <[email protected]>
wanpipe.txt and wan-router.txt in Documentation/networking
contain the exact same information (diff between the two shows no
difference). This patch removes the reference to this document in
drivers/net/wan/Kconfig.
Signed-off-by: Tobias Klauser <[email protected]>
Signed-off-by: Domen Puncer <[email protected]>
---
Kconfig | 2 +-
1 files changed, 1 insertion(+), 1 deletion(-)
Index: quilt/drivers/net/wan/Kconfig
===================================================================
--- quilt.orig/drivers/net/wan/Kconfig
+++ quilt/drivers/net/wan/Kconfig
@@ -435,7 +435,7 @@ config VENDOR_SANGOMA
the driver to support.
If you have one or more of these cards, say M to this option;
- and read <file:Documentation/networking/wanpipe.txt>.
+ and read <file:Documentation/networking/wan-router.txt>.
To compile this driver as a module, choose M here: the
module will be called wanpipe.
--
Hello,
We have several SMP machines (Tyan Tiger MPX motherboard, 2
AthlonMP 1900+ CPU, linux-2.4.21-20.EL). When running some time
sensitive programs, I observed that the values of gettimeofday () jumped
sometimes on a couple of machines (other machines are fine), from
several hundreds milliseconds to a couple of seconds.
I searched online and tried to figure out why this happened. It
seems there are a lot of people who complained about the clock drift
issue of gettimeofday () on SMP machines. But I still could not get the
answers about what is the exact cause of this issue and how to fix it.
Could somebody tell me some clues? Can this be solved by
upgrading the kernel to 2.6? Thanks a lot.
By the way, I have read Linux SMP HOWTO and the following documents.
http://seclists.org/lists/linux-kernel/2002/Jun/1524.html
http://seclists.org/lists/linux-kernel/2002/Jun/1550.html
http://alphalinux.org/archives/linux-alpha/April2000/0015.html
I could not get enough clues from them yet, maybe because my limited
knowledge about the kernel.
Regards,
[Liangchen Zheng]
On Sat, 2005-05-28 at 18:37 -0700, Liangchen Zheng wrote:
> Hello,
> We have several SMP machines (Tyan Tiger MPX motherboard, 2
> AthlonMP 1900+ CPU, linux-2.4.21-20.EL). When running some time
> sensitive programs, I observed that the values of gettimeofday () jumped
> sometimes on a couple of machines (other machines are fine), from
> several hundreds milliseconds to a couple of seconds.
Are you running NTP on those machines by any chance?
Lee
On Sat, 2005-05-28 at 18:37 -0700, Liangchen Zheng wrote:
> Hello,
> We have several SMP machines (Tyan Tiger MPX motherboard, 2
> AthlonMP 1900+ CPU, linux-2.4.21-20.EL). When running some time
> sensitive programs, I observed that the values of gettimeofday () jumped
> sometimes on a couple of machines (other machines are fine), from
> several hundreds milliseconds to a couple of seconds.
try "notsc" as boot option
Liangchen Zheng wrote:
> Hello,
> We have several SMP machines (Tyan Tiger MPX motherboard, 2
> AthlonMP 1900+ CPU, linux-2.4.21-20.EL). When running some time
> sensitive programs, I observed that the values of gettimeofday() jumped
> sometimes on a couple of machines (other machines are fine), from
> several hundreds milliseconds to a couple of seconds.
That sounds like what I described here:
http://lkml.org/lkml/2005/4/4/57
2 options as far as I can see.
1. Use irq affinity to bind the timer irq to a particular CPU,
while using CPU affinity to bind your process to a particular CPU.
2. Change the code to maintain a last_tsc_low for each CPU.
--
P?draig Brady - http://www.pixelbeat.org
--
It seems the problem is not caused by ntpd.
The issue is still there after I stopped the ntpd daemon. I tested it
for a long time and found the clock drifted around every 1 second. The
distances of the clock drift are interleaved with both positive and
negative values. For example, if it goes 300ms forward, it will go
backward around 300ms at the next time. Because the occurrences of the
clock drift are so regular, I doubt it was caused only by one source. I
am still wondering how I can discover it.
There were also 2 other suggestions I got from here the last
week. I will try them to see if it can solve my problem.
Thanks.
Regards,
Liangchen
-----Original Message-----
From: Lee Revell [mailto:[email protected]]
Sent: Saturday, May 28, 2005 7:00 PM
To: Liangchen Zheng
Subject: RE: The values of gettimeofday() jumps.
On Sat, 2005-05-28 at 19:00 -0700, Liangchen Zheng wrote:
> Yes. There is an ntp server in the LAN. Is the problem caused by this?
I
> saw some document said the clock drift issue can be solved by using
> ntpd.
> So if the problem is caused by ntp, how can I solve it?
> Thanks a lot.
Well, if you are running ntpd on all of the machines (and not just
periodically running ntpdate), it should slew the clock gradually, so
you should not see any big jumps.
If all machines are running ntpd locally, and all are configured to use
the same NTP server, all clocks should be in sync all the time.
Lee