[Linux-ia64] clock stops on test9

[Linux-ia64] clock stops on test9

[Einar Saltnes]

Hi,
I have upgraded a dual Itanium Workstation SDV from A3-stepping/BIOS 50
to B1-stepping/BIOS 60. After getting the upgraded box up and running
stable (had to decrease the FSB speed or move memory as pointed out by
Intel), I had some hard time trying to get linux up and run.
RedHat developer distro 0825 did not install. When taking out one cpu I
could get some further in the installation from the CD, but not finish
it successfully.
To install a newer kernel, I had to swap the disk to my UP Itanium
Workstation SDV A2-stepping/BIOS 50 which has been running stable, more
or less, on the 0829 kernel update from RedHat. Then compile a new
kernel
and swap disks again.
I am now running the test9 kernel on the dual B1-stepping Itanium and it
seems to run stable, more or less. I have also installed
modutils-2.3.17, which was pointed out on this list.

However, there are some problems with the system clock. The BIOS clock
seems to run OK, but in linux the clock seems to stop after some random
time. When using "date", time stands still after some random time! This
gives me some trouble using other utilities e.g. ping and even shutdown.
This shows only on the dual B1-stepping box and not the UP A2-stepping
box. It can run OK for a few minutes or for some hours.
Anyone got a clue?

Which configuration settings should I use regarding the A-stepping /
A1-stepping / B-stepping / B1-stepping settings in menuconfig for my two
Itanium boxes (see below)?
What about the settings if I change back to Dual A3-stepping / BIOS 50 ?
(So far I have been looking at H.J.Lu's settings which was posted on Sep
08)


Workstation SDV #1:
- UP A2-stepping
- Bios 50
- 1GB memory
- QLA1280 scsi-adapter
- SCSI-disk
- LS-120 floppy drive
- IDE CD-ROM
- ATI Rage128 PCI graphics adapter
- linux-2.4.0-test9 UP configuration

Workstation SDV #2:
- Dual B1-stepping
- BIOS 60
- 1GB memory
- QLA1280 scsi-adapter
- SCSI-disk
- LS-120 floppy drive
- IDE CD-ROM
- ATI Rage128 PCI graphics adapter
- linux-2.4.0-test9 SMP configuration

BTW, when talking about BigSurs and Lions how can I tell them apart? It
would be nice to know when something on this list applies to my Itanium
boxes.

Thanks,
Einar
-- 
| Einar Saltnes          |         Scali AS         |
| mailto:eisa@scali.no   |   http://www.scali.com   |
| Phone:+47 22 62 89 50  |  Scalable Linux Systems  |
| Fax  :+47 22 62 89 51  |      and Technology      |

Re: [Linux-ia64] clock stops on test9

[David Mosberger]

>>>>> On Fri, 13 Oct 2000 11:55:10 +0200, Einar Saltnes <eisa@scali.no> said:

  Einar> However, there are some problems with the system clock. The
  Einar> BIOS clock seems to run OK, but in linux the clock seems to
  Einar> stop after some random time. When using "date", time stands
  Einar> still after some random time! This gives me some trouble
  Einar> using other utilities e.g. ping and even shutdown.  This
  Einar> shows only on the dual B1-stepping box and not the UP
  Einar> A2-stepping box. It can run OK for a few minutes or for some
  Einar> hours.  Anyone got a clue?

Doesn't sound familiar.  It's very unlikely to be a hardware problem
because the kernel's timer tick runs off the CPU cycle counter.  Where
did you get the test9 kernel from and what version of the toolchain
was used to compile it?

  Einar> Which configuration settings should I use regarding the
  Einar> A-stepping / A1-stepping / B-stepping / B1-stepping settings
  Einar> in menuconfig for my two Itanium boxes (see below)?  What
  Einar> about the settings if I change back to Dual A3-stepping /
  Einar> BIOS 50 ?  (So far I have been looking at H.J.Lu's settings
  Einar> which was posted on Sep 08)

For A2, turn on the A-step workarounds and nothing else.
For B1, turn on the B-step workarounds and nothing else.

  Einar> BTW, when talking about BigSurs and Lions how can I tell them
  Einar> apart? It would be nice to know when something on this list
  Einar> applies to my Itanium boxes.

That's real easy: if it's rackmounted, it's a Lion, if it's in a
deskside case, it's a Big Sur.

	--david

Re: [Linux-ia64] clock stops on test9

[Einar Saltnes]

David Mosberger wrote:
> 
> >>>>> On Fri, 13 Oct 2000 11:55:10 +0200, Einar Saltnes <eisa@scali.no> said:
> 
>   Einar> However, there are some problems with the system clock. The
>   Einar> BIOS clock seems to run OK, but in linux the clock seems to
>   Einar> stop after some random time. When using "date", time stands
>   Einar> still after some random time! This gives me some trouble
>   Einar> using other utilities e.g. ping and even shutdown.  This
>   Einar> shows only on the dual B1-stepping box and not the UP
>   Einar> A2-stepping box. It can run OK for a few minutes or for some
>   Einar> hours.  Anyone got a clue?
> 
> Doesn't sound familiar.  It's very unlikely to be a hardware problem
> because the kernel's timer tick runs off the CPU cycle counter.  Where
> did you get the test9 kernel from and what version of the toolchain
> was used to compile it?
> 
Source tarball was downloaded from ftp.kernel.org. The ia64 patch was
downloaded from the same site. Kernel was compiled in the RedHat
Developers Distribution of 0825 environment with a test9 kernel.
Toolchain version is

: eisa@scali9-12:eisa; rpm -qi gnupro
Name        : gnupro                       Relocations: /usr 
Version     : 0501                              Vendor: Red Hat, Inc.
Release     : 2                             Build Date: Thu 01 Jun 2000
12:45:11 AM MEST
Install date: Tue 17 Oct 2000 02:10:41 PM MEST      Build Host:
boris.devel.redhat.com
Group       : Development/Tools             Source RPM:
gnupro-0501-2.src.rpm


: eisa@scali9-12:eisa; gcc -v
Reading specs from
/usr/lib/gcc-lib/ia64-redhat-linux/2.9-ia64-000216-final/specs
gcc version 2.9-ia64-000216-final


: eisa@scali9-12:eisa; rpm -qi glibc
Name        : glibc                        Relocations: (not
relocateable)
Version     : 2.1.3                             Vendor: Red Hat, Inc.
Release     : 0.0622.3                      Build Date: Sat 19 Aug 2000
11:55:26 PM MEST
Install date: Tue 03 Oct 2000 05:51:18 PM MEST      Build Host:
boris.devel.redhat.com
Group       : System Environment/Libraries   Source RPM:
glibc-ia64-2.1.3-0.0622.3.src.rpm


I see in my config file that I have enabled both the BSTEP_- and the
B0STEP_SPECIFIC settings. As you point out only BSTEP_ should be
enabled.

----------

OK, I have compiled a new kernel with only BSTEP_ enabled. Same story.
Using "date" I observe that time stops approx 20 minutes after boot of
the new kernel.
However, "cat /proc/efirtc" seems to show correct numbers. BTW, is this
the output of the HWclock? Is this the same source as "date" gets its
numbers from?


Thanks!

Einar
-- 
| Einar Saltnes          |         Scali AS         |
| mailto:eisa@scali.no   |   http://www.scali.com   |
| Phone:+47 22 62 89 50  |  Scalable Linux Systems  |
| Fax  :+47 22 62 89 51  |      and Technology      |

Re: [Linux-ia64] clock stops on test9

[Einar Saltnes]

"Ahna, Christopher J" wrote:
> 
> Einar,
> 
> Are you building your test9 kernel with A1-specific fixes enabled?
> A1-specific fixes deal with interrupt handling and have historically caused
> problems like the one you're seeing on non-A1 stepping Itaniums.  Thanks,
> 
> Chris
> 
Hi,

on my Dual B1-stepping Itanium I have now only enabled the
BSTEP_SPECIFIC setting. But the clock will still stop after some time.

On my UP A2-stepping Itanium, however, I have enabled both the ASTEP_-
and the A1_SPECIFIC setting. On this machine it doesn't seem to be a
problem with the clock. Uptime today is four days.

As David M. pointed out, I will compile a new kernel enabling only the
ASTEP_SPECIFIC setting for the A2-stepping BigSur as this will be the
correct configuration for the actual processor.

I will also make some tests on a Dual A3-stepping BigSur with BIOS 60
which just came in my door...

Thanks!

Einar
-- 
| Einar Saltnes          |         Scali AS         |
| mailto:eisa@scali.no   |   http://www.scali.com   |
| Phone:+47 22 62 89 50  |  Scalable Linux Systems  |
| Fax  :+47 22 62 89 51  |      and Technology      |

Re: [Linux-ia64] clock stops on test9

[Takayoshi Kouchi]

Hi all,

From: Einar Saltnes <eisa@scali.no>
Subject: Re: [Linux-ia64] clock stops on test9

> >   Einar> However, there are some problems with the system clock. The
> >   Einar> BIOS clock seems to run OK, but in linux the clock seems to
> >   Einar> stop after some random time. When using "date", time stands
> >   Einar> still after some random time! This gives me some trouble
> >   Einar> using other utilities e.g. ping and even shutdown.  This
> >   Einar> shows only on the dual B1-stepping box and not the UP
> >   Einar> A2-stepping box. It can run OK for a few minutes or for some
> >   Einar> hours.  Anyone got a clue?
> > 
> > Doesn't sound familiar.  It's very unlikely to be a hardware problem
> > because the kernel's timer tick runs off the CPU cycle counter.  Where
> > did you get the test9 kernel from and what version of the toolchain
> > was used to compile it?

We have seen the same problem as yours.  We have upgraded our BigSur
with an upgrade kit from Intel.  It had originally only one processor,
but we replaced it with two B0 processors.  
While running 2.4.0-test9 kernel, we noticed that timer interrupt of CPU0
would stop at some time after boot.
Below is a /proc/stat of that problem.

luna:~% cat /proc/stat
cpu  932275 0 52697 18446744073709470016
cpu0 372675 0 23555 55456
cpu1 559600 0 29142 18446744073709414560
page 3907 771
swap 1 0
intr 1158056 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1363 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 9299 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6 6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 642 1 2 0 1 0 0 1 1 0 0 0 0 0 0 0 8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1067645 0 0 0 0 0 0 0 0 0 0 0 0 0 0 79080 0
disk_io: (8,0):(1329,601,31038,728,6168) 
ctxt 24320
btime 971772286
processes 541

Once timer interrupt stops, jiffies will not be incremented.
That causes idle time to be negative.

> OK, I have compiled a new kernel with only BSTEP_ enabled. Same story.
> Using "date" I observe that time stops approx 20 minutes after boot of
> the new kernel.
> However, "cat /proc/efirtc" seems to show correct numbers. BTW, is this
> the output of the HWclock? Is this the same source as "date" gets its
> numbers from?

"date" output is calculated from times of timer interrupt.
If CPU0's timer interrupt stops, then counting jiffies (timer interrupt
counter) will also stop.  Then "date" indicates the same time.

We'are investigating why the timer interrupt stops.

---
Takayoshi Kouchi, NEC solutions

Re: [Linux-ia64] clock stops on test9

[Einar Saltnes]

Hi again,

Takayoshi Kouchi wrote:
> 
> Hi all,
> 
> 
> We have seen the same problem as yours.  We have upgraded our BigSur
> with an upgrade kit from Intel.  It had originally only one processor,
> but we replaced it with two B0 processors.
> While running 2.4.0-test9 kernel, we noticed that timer interrupt of CPU0
> would stop at some time after boot.

I can see that cpu0 is the one stopping after some time looking at
/proc/stat. Very interesting!

And to David M.:
I am positive that I have remembered to do "make clean" and "make dep"
before making new kernel. I have observed the same behaviour on three
different kernels, built using only minor differences in the
configuration.

Thank you!

Einar Saltnes
-- 
| Einar Saltnes          |         Scali AS         |
| mailto:eisa@scali.no   |   http://www.scali.com   |
| Phone:+47 22 62 89 50  |  Scalable Linux Systems  |
| Fax  :+47 22 62 89 51  |      and Technology      |