[Linux-ia64] swiotlb on rx4610, again

[Linux-ia64] swiotlb on rx4610, again

[Matthieu Delahaye]

Hi guys,

I have two rx4610 which ran well until last week under Linux for more than
one year.
They ran with first an unstable and then a stable Debian Woody with the
standard stock kernel (2.4.17) provided by Debian.
I didn't change their hardware configuration since.

Thanks to an ext3 filesystem crash, I had to reinstall Linux on. Using
the same standard process I always used, no way this time to complete
the installation:
Kernel panic: map_single could not allocated software IO TLB (42 bytes)
In interrupt handler - not syncing
Looking on the second one I have, same problem.

I've isolated the problem: setting up a NIC (eepro100) and just perform a ping is enough.
I've found in this mailing list a workaround: swiotlb kernel parameter.

Providing a too high value (32768) made the system reboot before going into
Init. swiotlb@96 works well.

But the fact this kernel worked without this option and that now it is
required, I'm quite "annoyed" hoping this is not an hardware issue. Is there a known problem on rx4610 which may explain that?

Best regards,

Matthieu Delahaye

Re: [Linux-ia64] swiotlb on rx4610, again

[Bjorn Helgaas]

On Thursday 17 April 2003 7:06 am, Matthieu Delahaye wrote:
> Hi guys,
> 
> I have two rx4610 which ran well until last week under Linux for more than
> one year.
> They ran with first an unstable and then a stable Debian Woody with the
> standard stock kernel (2.4.17) provided by Debian.
> I didn't change their hardware configuration since.
> 
> Thanks to an ext3 filesystem crash, I had to reinstall Linux on. Using
> the same standard process I always used, no way this time to complete
> the installation:
> Kernel panic: map_single could not allocated software IO TLB (42 bytes)
> In interrupt handler - not syncing
> Looking on the second one I have, same problem.

Can you give us a little more information about what kernel you're
trying to reinstall?  Often the installer uses a different kernel than
the one you end up running, so perhaps the install kernel is broken.
If you're reinstalling Debian,  you might get more knowledgeable help
on debian-ia64@lists.debian.org.

Bjorn

Re: [Linux-ia64] swiotlb on rx4610, again

[Bdale Garbee]

m.delahaye@esiee.fr (Matthieu Delahaye) writes:

> But the fact this kernel worked without this option and that now it is
> required, I'm quite "annoyed" hoping this is not an hardware issue. 

A couple of general suggestions.

The latest ia64 installer CD for Debian is available online via a link on 
the http://www.debian.org/ports/ia64/ page.  It uses a 2.4.19 kernel build
that received extensive testing on HP's Itanium 1 and 2 systems.

If you're running my binary kernel image packages on a Debian system, then
I have been suggesting the 2.4.19 kernel images from the 'sid' tree regardless
of whether you're running stable, testing, or unstable.  I now recommend the
2.4.20 021210.em18.1 kernels which we have somewhat less experience with, but
which are starting to feel as stable as the 2.4.19 images, and contain many
useful updates.

As has already been indicated, any other questions about running the Debian
distribution on ia64 systems should be addressed to the debian-ia64 list.
Pointers to the list archives and subscription instructions are available on
the ports page mentioned above.

Enjoy!

Bdale

Re: [Linux-ia64] swiotlb on rx4610, again

[Matthieu Delahaye]

On Fri, Apr 18, 2003 at 08:52:43AM -0600, Bdale Garbee wrote:
> m.delahaye@esiee.fr (Matthieu Delahaye) writes:
> 
> > But the fact this kernel worked without this option and that now it is
> > required, I'm quite "annoyed" hoping this is not an hardware issue. 
> 
> A couple of general suggestions.
> 
> [...]

Thanks Bdale for the suggestions. But I think I didn't express correctly the
main objective of my mail:

I can make by servers running under Linux...
and I've justed tell I were using Debian stock kernel just for explaining it
wasn't an home made with a possible error manually inserted, and to tell
I was sure it was the same kernel I was using before my crash.

My problem is not here. I am just trying to know if someone have
an hint on the fact my servers behaviour change over the time with the
same kernel/softwares.

Sure I can add a more recent kernel and this may work correctly without
adding kernel options. But If I have an hardware issue, perhaps it is better to know it now than
latter.

So that's why I've posted here. I just wanted to know if someone else already have noticed that before. No more.
I was quite busy today so that I didn't have perform further tests with
different kernel/configurations and debugging to learn more.

Best regards,

Matthieu