[PATCH] i386: run BIOS PCI detection before direct

[PATCH] i386: run BIOS PCI detection before direct

[Dave Hansen]

from 2.6.16-rc3-mm1 through at least 2.6.16-rc6-mm1 a patch from
Andi Kleen, titled

        x86_64-i386-pci-ordering.patch

which is now called:

	gregkh-pci-pci-give-pci-config-access-initialization-a-defined-ordering.patch

has caused a 4-way PIII Xeon (non-NUMA) to stop detecting its SCSI
card.  I believe this is also the issue keeping -mm from booting
on "elm3b67" from http://test.kernel.org/. 

The following patch reverts the ordering of the PCI detection code
to always run the BIOS initialization, first.  As far as I can
tell, this was the original behavior, and it makes my machine boot
again.

Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
---

 work-dave/arch/i386/pci/init.c |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)

diff -puN arch/i386/pci/init.c~run-bios-first-take2 arch/i386/pci/init.c
--- work/arch/i386/pci/init.c~run-bios-first-take2	2006-03-16 15:55:00.000000000 -0800
+++ work-dave/arch/i386/pci/init.c	2006-03-16 15:55:47.000000000 -0800
@@ -12,13 +12,13 @@ static __init int pci_access_init(void)
 #endif
 	if (raw_pci_ops)
 		return 0;
-#ifdef CONFIG_PCI_DIRECT
-	pci_direct_init();
+#ifdef CONFIG_PCI_BIOS
+	pci_pcbios_init();
 #endif
 	if (raw_pci_ops)
 		return 0;
-#ifdef CONFIG_PCI_BIOS
-	pci_pcbios_init();
+#ifdef CONFIG_PCI_DIRECT
+	pci_direct_init();
 #endif
 	return 0;
 }
_

Re: [PATCH] i386: run BIOS PCI detection before direct

[Andy Whitcroft]

Dave Hansen wrote:
> from 2.6.16-rc3-mm1 through at least 2.6.16-rc6-mm1 a patch from
> Andi Kleen, titled
> 
>         x86_64-i386-pci-ordering.patch
> 
> which is now called:
> 
> 	gregkh-pci-pci-give-pci-config-access-initialization-a-defined-ordering.patch
> 
> has caused a 4-way PIII Xeon (non-NUMA) to stop detecting its SCSI
> card.  I believe this is also the issue keeping -mm from booting
> on "elm3b67" from http://test.kernel.org/. 
> 
> The following patch reverts the ordering of the PCI detection code
> to always run the BIOS initialization, first.  As far as I can
> tell, this was the original behavior, and it makes my machine boot
> again.
> 
> Signed-off-by: Dave Hansen <haveblue@us.ibm.com>

Ran this through the nightly regression suite on the affected machine
and it boots fine with this patch applied.

-apw

Re: [PATCH] i386: run BIOS PCI detection before direct

[Andi Kleen]

On Friday 17 March 2006 11:38, Andy Whitcroft wrote:
> Dave Hansen wrote:
> > from 2.6.16-rc3-mm1 through at least 2.6.16-rc6-mm1 a patch from
> > Andi Kleen, titled
> > 
> >         x86_64-i386-pci-ordering.patch
> > 
> > which is now called:
> > 
> > 	gregkh-pci-pci-give-pci-config-access-initialization-a-defined-ordering.patch
> > 
> > has caused a 4-way PIII Xeon (non-NUMA) to stop detecting its SCSI
> > card.  I believe this is also the issue keeping -mm from booting
> > on "elm3b67" from http://test.kernel.org/. 
> > 
> > The following patch reverts the ordering of the PCI detection code
> > to always run the BIOS initialization, first.  As far as I can
> > tell, this was the original behavior, and it makes my machine boot
> > again.
> > 
> > Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
> 
> Ran this through the nightly regression suite on the affected machine
> and it boots fine with this patch applied.

I fixed this up my copy of the patch.

Also fixed the warning with CONFIG_ACPI=n

Thanks,

-Andi

Re: [PATCH] i386: run BIOS PCI detection before direct

[Andi Kleen]

On Wednesday 22 March 2006 23:08, Greg KH wrote:

> 
> Care to send me that copy of the patch so I can forward it on?

ftp://ftp.firstfloor.org/pub/ak/x86_64/quilt/patches/i386-pci-ordering

-Andi

Re: [PATCH] i386: run BIOS PCI detection before direct

[Greg KH]

On Wed, Mar 22, 2006 at 02:10:47PM +0100, Andi Kleen wrote:
> On Friday 17 March 2006 11:38, Andy Whitcroft wrote:
> > Dave Hansen wrote:
> > > from 2.6.16-rc3-mm1 through at least 2.6.16-rc6-mm1 a patch from
> > > Andi Kleen, titled
> > > 
> > >         x86_64-i386-pci-ordering.patch
> > > 
> > > which is now called:
> > > 
> > > 	gregkh-pci-pci-give-pci-config-access-initialization-a-defined-ordering.patch
> > > 
> > > has caused a 4-way PIII Xeon (non-NUMA) to stop detecting its SCSI
> > > card.  I believe this is also the issue keeping -mm from booting
> > > on "elm3b67" from http://test.kernel.org/. 
> > > 
> > > The following patch reverts the ordering of the PCI detection code
> > > to always run the BIOS initialization, first.  As far as I can
> > > tell, this was the original behavior, and it makes my machine boot
> > > again.
> > > 
> > > Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
> > 
> > Ran this through the nightly regression suite on the affected machine
> > and it boots fine with this patch applied.
> 
> I fixed this up my copy of the patch.
> 
> Also fixed the warning with CONFIG_ACPI=n

Care to send me that copy of the patch so I can forward it on?

thanks,

greg k-h