devel ACK for U6 (conditional: assumes we get a tyan to test with in time)

Hi,
Sorry, I forgot to mention that this is for 32 bit 2.4.21-32

-dc

A little more research on this issue has uncovered a patch for this issue:

ftp://ftp.tyan.com/drivers_linux/s2885/tyan_agp.RHEL3.patch1

This patch was recently included in the RHEL3 kernel (and should included be in
U6), but since we dont have any Tyan hardware on-site I am unable to reproduce
the issue, or verify this fix resolves the problem.

also, booting with "pci=noacpi" may be a workaround (but this is also untested).
This problem seems to appear when using a 32 bit kernel, while AGP bus #4 seems
to be detected properly using a 64 bit kernel.

The tyan patch does not apply properly to redhat 3 update 4 or 5.  I have a
feeling that this patch is for the 2.4.21-15 or earlier, as it appears to have
been released in early 2004.

I did try booting wiht pci=noapci.  dmesg reports that this is not a valid boot
option, so I tried booting with apci=off, and that did not have any effect
except to cause my nvidia card to share an irq with my ethernet adapter.

Yes the 64bit kernel works fine.  I have loaded the x86_64 version and it seed
the agp bridge at pci address 04.  

My belief is that whatever patches were rolled into the 64bit kernel need to be
implemented and tested in the 32 bit version.  I'm not able to write a patch
myself.  But I would be more than happy to test patches.

A previous bug, 108648, reported the same problem.  

I tried to apply several of the patches that were submitted, but they did not
patch properly on 2.4.21-32 or 2.4.21-27.  Indeed you can see that in that
thread that the submitted patches for 32bit were not working right.
  
Then the Redhat resolution seemed to only fix the 64bit kernel.  My guess is
that the fix for 32bit kernel never really happened.

looking for some more debug info here:

cat we get a 'dmesg' output from after the system boots? 

also a 'dmesg' output from when the agpgart module fails to load (log level set
to kern.* being that some usefull debug messages only show up under KERNEL_DEBUG)? 

also worth noting is that there is some AMD-8151 and nVidia code in agpgart_be
that gets executed only when "agp_try_unsupported=1" is passed to the kernel at
boot time. not sure if this will help.

Created attachment 115510 [details]
dmesg right after boot

Created attachment 115511 [details]
dmesg after trying to load agpgart

Attached requested dmesg output.

I have actually tried booting with "agp_try_unsupported=1" based on some
information that I found, but alas, that was ineffective.
Thanks for the tip though

Dan

Any progress on this issue?

Posted a test kernel for Tyan to test on some of their systems.

http://people.redhat.com/bmaly/linux-2.4.21-32.9.EL.tar.gz

Great.  
I've downloaded this as well.  I'm short one tyan box due to hardware failure,
but I will test this kernel when I get it back.

Thanks for your time and efforts.

Tyan did some testing on a S2885 and confirmed the problem exists in our RHEL3
pre-u6 32bit kernel. Tyan now intends to send us a S2885 motherboard so we can
diagnose this issue. 

That's good news.  
Hopefully a fix can be developed in time to make it into u6

just a quick note that we received the hardware from tyan. we should be able to
replicate this issue now.

Created attachment 116768 [details]
Patch to detect AGP bus bus (i386 kernel) for Tyan S2885MB

this patch forces a peer bus scan so that the AMD-8151 AGP bridge is found

A patch has been submitted to rhkernel for inclusion in U6 (pending approval).

Diagnosis:

 The RHEL3 kernel doesnt detect all PCI peer bridges like some AMD-8151
AGP bridges in arch/i386/kernel/pci-pc.c but does in
arch/x86_64/kernel/pci-pc.c. As a  result agpgart.o loads with
a x86_64 kernel but fails to load with an i386 kernel. Additionally
the AGP bridge is absent from lspci output on i386 but shows up on
x86_64.
 This minimalistic patch includes the same x86_64 functionality on i386,
and forces a scan to detect peer host bridges in pcibios_fixup_peer_bridges().
On x86_64 pcibios_last_bus gets set to 0xfe if no mptable is found (this
happens pci_scan_mptable, before pcibios_fixup_peer_bridges() gets called).

Tested this patch on a tyan S2885 board... the AMD-8151 is detected
at boot under i386 (just like x86_64). agpgart.o loads fine as well.

This is great news.  Thanks for your attention and work on this issue.  I really
appreciate it.  I'm assuming that approval of your patch won't be a problem.
When is U6 estimated to be released?
Thanks again

A fix for this problem has just been committed to the RHEL3 U7
patch pool this evening (in kernel version 2.4.21-37.8.EL).

Can you give me a estimated release date for U7?  I really need this fix - was
expecting it in U6...
Thank you

Hi, Daniel.  The U7 public beta period is not scheduled to start until
early January.  We're near the end of the development cycle now.

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHSA-2006-0144.html