SMP 440GX+ hang on boot (2.4.16)

SMP 440GX+ hang on boot (2.4.16)

[Jeff Gustafson]

	Apparently this is a known problem.  On this mother board we have a
AIC7xxx controller.  We also have a DAC960.  When the system tries to mount
the drives on the DAC960 (nothing is connected to the AIC7xxx), the system
hangs.  Supposedly the problem is only with UP kernels, but we get hangs 
with a SMP compiled kernel!
	The funny thing is that the kernels from RH 7.2 (2.4.7-10) and Alan
Cox's latest (2.4.13-ac8) work just fine.  The 2.4.16 kernel does not 
work.  Is there a possiblity that the fix could posted to linux-kernel?  
Maybe it could be put into the upcoming 2.4.17 release.  Please??
	Note that if you search the RedHat Bugilla site for "440GX+" you'll 
find references to this problem.

				...Jeff

Re: SMP 440GX+ hang on boot (2.4.16)

[Alan Cox]

> AIC7xxx controller.  We also have a DAC960.  When the system tries to mount
> the drives on the DAC960 (nothing is connected to the AIC7xxx), the system
> hangs.  Supposedly the problem is only with UP kernels, but we get hangs 
> with a SMP compiled kernel!
> work.  Is there a possiblity that the fix could posted to linux-kernel?  

I have no idea why the -ac kernel happens to work in your case. The
following is the standard RH message on the subject. Since -ac works and
2.4.16/17 doesn't I'd be suspicious that you may be seeing a different
problem. 

---------------------

Most implementations using the 440GX chipset require the "apic"
option to function correctly.  When this is the case, providing
DMI information to Red Hat as documented below[1] may allow us to
automate that setting in the future.

On some systems, the "apic" boot flag may not work.  Unfortunately,
because of Intel Proprietary information, these platforms are not
supported at this time.

[1] wget http://people.redhat.com/arjanv/dmidecode.c
    gcc dmidecode.c -o dmidecode
    ./dmidecode | \
       mail -s 'requires apic option to install' hardwarebugs-list@redhat.com

---------------------


Intel to my knowledge are still failing to provide either BIOSes with
suitable $PIR IRQ routing table data or explanations of how to work around
the problems with the 440GX. You may wish to consult your supplier for
advice, especially if the machine was sold to you for the purpose of
running Linux.

Alan

Re: SMP 440GX+ hang on boot (2.4.16)

[arjan]

In article <20011208122147.C62182F055@alpha.zimage.com> you wrote:
>        Apparently this is a known problem.  On this mother board we have a
> AIC7xxx controller.  We also have a DAC960.  When the system tries to mount
> the drives on the DAC960 (nothing is connected to the AIC7xxx), the system
> hangs.  Supposedly the problem is only with UP kernels, but we get hangs 
> with a SMP compiled kernel!
>        The funny thing is that the kernels from RH 7.2 (2.4.7-10) and Alan
> Cox's latest (2.4.13-ac8) work just fine.  The 2.4.16 kernel does not 
> work.  Is there a possiblity that the fix could posted to linux-kernel?  
> Maybe it could be put into the upcoming 2.4.17 release.  Please??

1) Did you enable ACPI in your own kernel ?
2) Does "noapic" make a difference ?

and if neither is true, well, even Intel officially says the 440GX is 
unsupported in that case.

Re: SMP 440GX+ hang on boot (2.4.16)

[Jeff Gustafson]

> > AIC7xxx controller.  We also have a DAC960.  When the system tries to mount
> > the drives on the DAC960 (nothing is connected to the AIC7xxx), the system
> > hangs.  Supposedly the problem is only with UP kernels, but we get hangs 
> > with a SMP compiled kernel!
> > work.  Is there a possiblity that the fix could posted to linux-kernel?  
> 
> I have no idea why the -ac kernel happens to work in your case. The
> following is the standard RH message on the subject. Since -ac works and
> 2.4.16/17 doesn't I'd be suspicious that you may be seeing a different
> problem. 

	It could be.  I'm going to try to boot off the Adaptec chip and 
see if that works.  It seems that the DAC960 doesn't allow Linux to mount 
the filesystem (no VFS... 'scream' message).  So it could be a DAC960 
driver issue?  

> ---------------------
> 
> Most implementations using the 440GX chipset require the "apic"
> option to function correctly.  When this is the case, providing
> 
> ---------------------

	The thing is that this machine installed RH 7.2 perfectly.  We 
didn't use the 'apic' option to install.  After installation, the machine 
booted without any issues.  It wasn't until we tried 2.4.16 (and 2.4.14) 
that we ran into issues booting. 

> Intel to my knowledge are still failing to provide either BIOSes with
> suitable $PIR IRQ routing table data or explanations of how to work around
> the problems with the 440GX. 

	ARGE!

> You may wish to consult your supplier for
> advice, especially if the machine was sold to you for the purpose of
> running Linux.

	Wow!  Was that a subtle slam?  I wonder which vendor you had in
mind?  ;)  I expect it's the same name that's on this 440GX+ box.

				...Jeff

Re: SMP 440GX+ hang on boot (2.4.16)

[Jeff Gustafson]

> >        The funny thing is that the kernels from RH 7.2 (2.4.7-10) and Alan
> > Cox's latest (2.4.13-ac8) work just fine.  The 2.4.16 kernel does not 
> > work.  Is there a possiblity that the fix could posted to linux-kernel?  
> > Maybe it could be put into the upcoming 2.4.17 release.  Please??
> 
> 1) Did you enable ACPI in your own kernel ?

	No ACPI.  We used RH's 7.2 i686 SMP config file as a starting 
point to build 2.4.16.

> 2) Does "noapic" make a difference ?

	That made it worse.  Just gets stuck right in the middle of
initalizing the driver.  Before it would initialize, but couldn't mount
the root partition.

				...Jeff