Very Computer

I've posted this elsewhere but got no responses, and wonder if any of
you kernel experts could help...

I installed Mandrake 8.1 on my system that uses an Athlon 1000 in an
Epox 8KTA3 mobo, which uses the VIA KT133A / 686B chipset.  It also
has an Acer 4432 ATAPI CD-RW on /hdc and an old Toshiba 4x CDROM on
/hdd.  Kernel was 2.4.8-mdk.

The CD-RW was set up with SCSI emulation, so there was no /dev/hdc
(couldn't even ls it); there was instead /dev/scd0.  When I tried to
mount the CD-RW, it sat about 20 seconds and then kernel panic
occured.  Something about kapm-idled. But I don't really understand
how to read these messages.

Others have told me that VIA chips are buggy.  I haven't seen any
VIA-specific changes in the kernels up to 2.4.13 though.  I went to
Viahardware.com and saw a lot of talk about problems with sound cards,
but I'm using only motherboard built-in sound (SB emulation) which
works fine. And the system is very stable in Windows 98.  I didn't
notice problems in  Red Hat 7.1, and only used Mandrake 8.0 for a
short time before switching to Red Hat, in part because it couldn't
play music CDs.  (Well, Red Hat turned out not to be great at it
either, so that IDE chain is a bit suspicious.) Finally, I upgraded
the BIOS, which viahardware suggested helps.  No difference.  BTW the
Acer CD-RW's firmware is also uprevved to the latest level, not
original.

Wondering if this was a Mandrake-specific problem, I reformatted and
installed Red Hat 7.2.  This uses kernel 2.4.7.  Alas same problem.
So it really seems to be in the 2.4 series kernel.  Maybe the Acer
CD-RW is a bit odd, or not compatible with the way the kernel's VIA
drivers work.

Nothing's visible in KDE but when I try to mount it from console mode,
I can at least see the last 24 lines of the kernel panic.  Here's what
it looked like under Mandrake, mostly, from hand-transcription, with
some ellipses of my own:

Unable to handle kernel paging request at virtual address a6b40767
 printing eip:
c884d323
*pde=00000000
Oops: 0002
CPU: 0
EIP: 0010:[<c884d323>]
E flags: 00010002
... a bunch of registers I didn't get...
process kapm-idled (pid:4, startpage = c1253000)
stack: ...
call trace: ...
code: c7 80 7c 01 00 00 00 00 07 00 8b 3e 24 85 ff 0f 84 68 01 00
<0>Kernel panic:  Aiee, killing interrupt handler!
In interrupt handler - not syncing

Now I'm not a kernel guru or anything, just trying to be a user, but
it does seem to involve interrupts, and I don't know what kapm-idled
is. Does all of this mean anything to anyone reading it?  Any ideas
that might let me run Linux on this system?  Thanks!

Quote:>(couldn't even ls it); there was instead /dev/scd0.  When I tried to
>mount the CD-RW, it sat about 20 seconds and then kernel panic
>occured.  Something about kapm-idled. But I don't really understand
>how to read these messages.

I have a Netfinity 7000M10.  Mounting /dev/cdrom under straight non-updated
RedHat 7.1 (2.4.2 kernel) gives me a message about turning off UDMA for
the cdrom - and everything is ok.  After I updated the kernel to 2.4.13,
it does NOT give this message.  Instead it hangs forever and sometimes
complains about a DMA timeout.

the hdc is just a device node. The driver (ide) could still hit it, note that
you ide-scsi to deal with the disk not ide-cdrom, this may mean supplying a
hdc=ignore parameter.

The VIA bugs that have been mentioned are either IDE UDMA related, drop to
DMA/PIO to check or Athlon/VIA problems, the latter only becomes apparent if
you use athlon specific optimisations.

karl.

Since the problem occurs in an interrupt handler the current
process might not be related to the problem.

It might always be happening while kapm-idled is the current
process because there is no other process trying to use any
CPU time. I would try to reproduce the error message while a
process is using as much CPU time as it can get.

In other words try runing this program, and at the same time
from the console do what produced the panic:

int main(){while(1);}

--
Kasper Dupont

Quote:>The VIA bugs that have been mentioned are either IDE UDMA related, drop
>to DMA/PIO to check or Athlon/VIA problems, the latter only becomes
>apparent if you use athlon specific optimisations.

I set the BIOS to disable UDMA, though it had been reading the disk as PIO4
anyway, not UDMA.  No difference.  I also notice this time (RedHat, not
Mandrake) that the IDE CD-ROM at /hdd also causes a kernel panic when I try
to mount it.

Alas, I set up RedHat to boot graphically, and I can't figure out how to
turn off X Windows and sit at a plain old shell to watch the kernel panic
happen.  I hate not knowing dumb stuff like this.

I also acknowledge that the kapmd-idle process was probably not the real
problem, but the interrupt handler was.  Could this perhaps have anything
to do with the particular pattern of shared interrupts on the PCI? I
woudln't think so since IDE2 is always on its own IRQ.

I don't have any Athlon optimizations turned on, just the kernel out of the
box.  When I tried building a kernel in Mandrake, nothing stood out as
obviously relevant.

More ideas anyone?  I've thought of swapping hdd and hdc (master/slave
jumpers), though that might confuse the Windows system, which alas is the
only OS that's now working right.  And RedHat's KDE looks pretty nice now.
(I can't say the same for their GNOME, but hey deciding between them is the
Linux game.)

And (showing myself for the Linux near-ignoramus that I am), what's the
name of the file that stores the stuff that scrolls past you during boot?
--
Fred R. Goldstein   k1io   fgoldstein"at" wn.net
These are my own opinions. You expect anyone else to agree?

Tried CTRL+ALT+F2? (And probably CTRL+ALT+F7 to get back)

--
Kasper Dupont

Is this an all drives, you can use hdparm on each drive. which kernel are you
running?

Quote:

> Alas, I set up RedHat to boot graphically, and I can't figure out how to
> turn off X Windows and sit at a plain old shell to watch the kernel panic
> happen. I hate not knowing dumb stuff like this.

change the initdefault line to have 3 instead of 5. On RH the boot messages
get put put into a file /var/log/dmesg but there is also /var/log/messages as
well.

That's both CD-ROM drives, one being an -RW.  Kernel is 2.4.7 (Red Hat 7.2)
but it also happens on 2.4.8-mdk (Mandrake 8.1).

Quote:>change the initdefault line to have 3 instead of 5. On RH the boot
>messages get put put into a file /var/log/dmesg but there is also
>/var/log/messages as well.

From the /var/log/messages, I notice the line
Applying VIA southbridge workaround
near the top of the sequence, so the kernel does recognize the VIA chip,
but I'm not sure what that "workaround" is.  Or whether the 686A and 686B
are supposed to be handled the same way.  It does use the VP_IDE driver.

Mounting the CD in text mode, the kernel panic message this time was during
process "swapper", which lends credence to the theory that it's an
interrupt problem unrelated to the task at hand.  It still ends with the
message:
In interrupt handler - not syncing

So far nothing in the BIOS stands out as being able to help, but then there
are a lot of BIOS options.  I'm pretty near the fail-safe defaults.  I do
have the CD-RW as master, not slave, against its maker's recommendations,
but this arrangement seems to be more stable in Windows.

And, of course, it installs from CD-ROM nicely.  THAT kernel is happy
enough.  But that isn't "mount"ed from inside Linux, but booted by the
BIOS.
--
Fred R. Goldstein   k1io   fgoldstein"at" wn.net
These are my own opinions. You expect anyone else to agree?

I too have mobo with a VIA chipset - Apollo + - One chip is a VT 82C596, the
other one I cant read.
Motherboard is Slot one  made by 1st Motherboard/Novus with a 450MHZ Intel P3
CPU. I have had
many varied problems running 2.4 kernels on this mobo. 2.2 kernels run fine. I
would like to be able to
run 2.4 kernels so that I can use iptables. I tried the 2.4 kernel on an older,
oh so much slower P1 233MHZ
MMX mobo with an Intel PIIX chipset and it ran fine. This is what I am
currently using. If the VIA chipset
is buggy, making a fix may be difficult. I hope the kernel gurus are working on
a work-around. I am currently
running 2.4.12 on this machine, haven't tried it on the VIA machine. Also the
distro I am using is Slackware 8.0.
Sorry, I'm not a fan of RedHat/SysV clones.

Just thought I'd put in my two cents worth.

                                                       Dave

Can you check the ide-scsi and scsi cdrom driver route as well as the ide-cd
route.  the device will be something like /dev/sr0.
What about later releases of the kernels?

Quote:

>>change the initdefault line to have 3 instead of 5. On RH the boot messages
>>get put put into a file /var/log/dmesg but there is also /var/log/messages
>>as well.

> From the /var/log/messages, I notice the line Applying VIA southbridge
> workaround
> near the top of the sequence, so the kernel does recognize the VIA chip, but
> I'm not sure what that "workaround" is.  Or whether the 686A and 686B are
> supposed to be handled the same way.  It does use the VP_IDE driver.

Nor do I, but it sounds like the UDMA fix not the Athlon/VIA fix.

Quote:> Mounting the CD in text mode, the kernel panic message this time was during
> process "swapper", which lends credence to the theory that it's an interrupt
> problem unrelated to the task at hand.  It still ends with the message:
> In interrupt handler - not syncing

yep

Quote:> So far nothing in the BIOS stands out as being able to help, but then there
> are a lot of BIOS options.  I'm pretty near the fail-safe defaults.  I do
> have the CD-RW as master, not slave, against its maker's recommendations,
> but this arrangement seems to be more stable in Windows.

BIOS manufacturers can be very sloppy, don't be surprised that the fail-safe
isn't. Thats wht there are utilities like powertweak.

Quote:> And, of course, it installs from CD-ROM nicely.  THAT kernel is happy
> enough.  But that isn't "mount"ed from inside Linux, but booted by the BIOS.

When it installs packages the CD has to be mounted. It might very well create
a RAM disk for a basic filesystem but the packages are read from a mounted CD
filesystem.

karl.

Quote:>Can you check the ide-scsi and scsi cdrom driver route as well as the
>ide-cd route.  the device will be something like /dev/sr0.
>What about later releases of the kernels?

IDE-CD is /dev/scd0.  If I boot without the hdc=ide-scsi suffix, then it's
/hdc, and works normally, though it does mount more slowly than the
old Toshiba 4x CD-ROM.

I've updated to 2.4.9, thanks to Red Hat Network's up2date command (slick!
even though it crashed at the end).  Not a fix, but better diagnostics, see
below.

Quote:>BIOS manufacturers can be very sloppy, don't be surprised that the
>fail-safe isn't. Thats wht there are utilities like powertweak.

You think it'll help?  Where do I find it?

With 2.4.9, I get much more useful info in the kernel panic.  Here is the
New Improved Kernel Panic message tail.  I've only hand-transcribed the
first of the big hex numbers, but I think I have all the rest of those
lines.  Does this give any useful clues?  It happens around 20 seconds
after issuing a mount command.

Call Trace [<C0187bd8>] ide_error [kernel] 0x128
[<c886b49c>] __insmod_ide-scsi_S.text_L5196 [ide-scsi] 0x43
  ide_timer_expiry [kernel] 0x172
  .rodata.str1.1 [kernel] 0x4a86
  ide_timer_expiry [kernel] 0x0
  __run_timers [kernel] 0xb0
  run_all_timers [kernel] 0x17
  bh_action [kernel] 0x16
  tasklet_hi_action [kernel] 0x52
  do_softirq [kernel] 0x47
  do_IRQ [kernel] 0x90
  default_idle [kernel] 0x0
  call_do_IRQ [0x5]
  default_idle [kernel] 0x0
  default_idle [kernel] 0x23
  cpu_idle [kernel] 0x42
  stext [kernel] 0x0
Kernel panic:  Aiee, killing interrupt handler!
in interrupt handler - not syncing

Thanks again!
--
Fred R. Goldstein   k1io   fgoldstein"at" wn.net
These are my own opinions. You expect anyone else to agree?

with /dev/scd0 ide-scsi and sr modules are used. With /dev/hdX (wrt cdrom)
ide-cd is used.  The two are separate but whether it produces different
results I don't know.  The stack trace you gave indicated the ide-scsi route.
Unload the ide-scsi modules and access the ide-cd (/dev/hdX) instead.

...

Quote:

>>BIOS manufacturers can be very sloppy, don't be surprised that the fail-safe
>>isn't. Thats wht there are utilities like powertweak.

> You think it'll help?  Where do I find it?

try

powertweak.sourceforge.net

karl.

If I boot *without* the "hdc=ide-scsi" line in grub.conf, then /hdc is
a normal CD-ROM and works.  But then I can't use the CD-RW as a CD-RW,
just as a CD-ROM. A workaround, but hardly a solution.

Quote:> powertweak.sourceforge.net

Thanks.  Will try.  I'm sure tweaking the various registers will be
great fun, whether or not it helps. ;-)

Well that indicates a bug in ide-scsi or scsi-modules, probably ide-scsi. The
only other issues that have cropped up are ide cables but your ok with the
ide-cd driver (which gets used at installation).  What does hdparm -i
/dev/hdc say.

karl.

fg>> If I boot *without* the "hdc=ide-scsi" line in grub.conf, then /hdc is

Quote:>> a normal CD-ROM and works.  But then I can't use the CD-RW as a CD-RW,
>> just as a CD-ROM. A workaround, but hardly a solution.

>Well that indicates a bug in ide-scsi or scsi-modules, probably
>ide-scsi. The only other issues that have cropped up are ide cables but
>your ok with the ide-cd driver (which gets used at installation).  What
>does hdparm -i /dev/hdc say.

Here's what I get, with hdc feeding ide-scsi:

/dev/hdc:

 Model=4X4X32, FwRev=3.LS, SerialNo=

 Config={ Removeable DTR<=5Mbs DTR>10Mbs nonMagnetic }

 RawCHS=0/0/0, TrkSize=0, SectSize=0, ECCbytes=0

 BuffType=DualPortCache, BuffSize=1024kB, MaxMultSect=0

 (maybe): CurCHS=0/0/0, CurSects=0, LBA=yes, LBAsects=0

 IORDY=on/off, tPIO={min:120,w/IORDY:120}, tDMA={min:120,rec:120}

 PIO modes: pio0 pio1 pio2 pio3 pio4

 DMA modes: sdma0 sdma1 sdma2 mdma0 mdma1 *mdma2

 AdvancedPM=no

Here's the old CD-ROM drive, for comparison:

/dev/hdd:

 Model=HITACHI CDR-7730, FwRev=0008a, SerialNo=

 Config={ Removeable DTR<=5Mbs DTR>10Mbs nonMagnetic }

 RawCHS=0/0/0, TrkSize=0, SectSize=0, ECCbytes=0

 BuffType=unknown, BuffSize=64kB, MaxMultSect=0

 (maybe): CurCHS=0/0/0, CurSects=0, LBA=yes, LBAsects=0

 IORDY=yes

 PIO modes: pio0 pio1 pio2

 DMA modes:

 AdvancedPM=no

Thanks for the help!

[In case this matters, when I boot, the computer's own BIOS shows hdc as
PIO4, vs. PIO2 for the old drive, but Linux' boot sequence (in /var...)
shows the kernel thinking that the BIOS has hdc set for DMA, like hda.
Probably of no import.]
--
Fred R. Goldstein   k1io   fgoldstein"at" wn.net
These are my own opinions. You expect anyone else to agree?