[Linux-ia64] PCI_BUSES_TO_SCAN off by 1?

[Linux-ia64] PCI_BUSES_TO_SCAN off by 1?

[Van Maren, Kevin]

Hi all,

Is there any reason we don't scan the last PCI bus on IA64?
arch/ia64/kernel/pci.c has PCI_BUSES_TO_SCAN set to 255 instead
of 256, which prevents the kernel from scanning bus 255.  This
causes is some problems since some of our machines (ie, the
Itanium2 "tiger") place chipset registers in PCI config space on
the last PCI bus.  I've run with it set to 256 on the Lion and
some other machines with no apparent problems.

Thanks,
Kevin

Re: [Linux-ia64] PCI_BUSES_TO_SCAN off by 1?

[Bjorn Helgaas]

> Is there any reason we don't scan the last PCI bus on IA64?
> arch/ia64/kernel/pci.c has PCI_BUSES_TO_SCAN set to 255 instead
> of 256, which prevents the kernel from scanning bus 255.  This
> causes is some problems since some of our machines (ie, the
> Itanium2 "tiger") place chipset registers in PCI config space on
> the last PCI bus.

I agree, it sounds like an off-by-one error.

Along this line, I had been hoping that if we find any PCI root
bridges via ACPI, we could skip the blind scanning of buses 0-255.
It sounds like skipping the blind scan might break tiger, though.
I assume it has root bridges in ACPI like BigSur does, but evidently
these chipset registers on bus 255 aren't found that way?  Having to
do both ACPI and blind scans feels wrong somehow, but I can't point
to a spec to back that up.

Bjorn

Re: [Linux-ia64] PCI_BUSES_TO_SCAN off by 1?

[Jack Steiner]

> 
> Hi all,
> 
> Is there any reason we don't scan the last PCI bus on IA64?
> arch/ia64/kernel/pci.c has PCI_BUSES_TO_SCAN set to 255 instead
> of 256, which prevents the kernel from scanning bus 255.  This
> causes is some problems since some of our machines (ie, the
> Itanium2 "tiger") place chipset registers in PCI config space on
> the last PCI bus.  I've run with it set to 256 on the Lion and
> some other machines with no apparent problems.
> 
> Thanks,
> Kevin

I opened PV 866403 for this....



> 
> 
> _______________________________________________
> Linux-IA64 mailing list
> Linux-IA64@linuxia64.org
> http://lists.linuxia64.org/lists/listinfo/linux-ia64
> 


-- 
Thanks

Jack Steiner    (651-683-5302)   (vnet 233-5302)      steiner@sgi.com

Re: [Linux-ia64] PCI_BUSES_TO_SCAN off by 1?

[KOCHI, Takayoshi]

On Fri, 30 Aug 2002 15:11:20 -0600
Bjorn Helgaas <bjorn_helgaas@hp.com> wrote:

> > Is there any reason we don't scan the last PCI bus on IA64?
> > arch/ia64/kernel/pci.c has PCI_BUSES_TO_SCAN set to 255 instead
> > of 256, which prevents the kernel from scanning bus 255.  This
> > causes is some problems since some of our machines (ie, the
> > Itanium2 "tiger") place chipset registers in PCI config space on
> > the last PCI bus.
> 
> I agree, it sounds like an off-by-one error.
> 
> Along this line, I had been hoping that if we find any PCI root
> bridges via ACPI, we could skip the blind scanning of buses 0-255.
> It sounds like skipping the blind scan might break tiger, though.

AFAIK, tiger defines PCI bus 255 as a host-bus bridge in ACPI.

Thanks,
-- 
KOCHI, Takayoshi <t-kouchi@cq.jp.nec.com/t-kouchi@mvf.biglobe.ne.jp>

Re: [Linux-ia64] PCI_BUSES_TO_SCAN off by 1?

[Bjorn Helgaas]

On Friday 30 August 2002 3:11 pm, Bjorn Helgaas wrote:
> Along this line, I had been hoping that if we find any PCI root
> bridges via ACPI, we could skip the blind scanning of buses 0-255.

To complete this thread, it looks like we can't do this.  My
i2000 reports PCI root bridges for buses 0, 1, 2, and 3 via
ACPI.  The 460GX chipset, including the CSRs for the AGP
bridge, appears in config space on PCI bus 4.  ACPI doesn't
tell us about bus 4, so we have to do the blind scan to find
that stuff.

Bjorn

Re: [Linux-ia64] PCI_BUSES_TO_SCAN off by 1?

[David Mosberger]

>>>>> On Tue, 22 Oct 2002 16:20:52 -0600, Bjorn Helgaas <bjorn_helgaas@hp.com> said:

  Bjorn> To complete this thread, it looks like we can't do this.  My
  Bjorn> i2000 reports PCI root bridges for buses 0, 1, 2, and 3 via
  Bjorn> ACPI.  The 460GX chipset, including the CSRs for the AGP
  Bjorn> bridge, appears in config space on PCI bus 4.  ACPI doesn't
  Bjorn> tell us about bus 4, so we have to do the blind scan to find
  Bjorn> that stuff.

We no longer do the blind scan in 2.5.39 and it seems to work fine on
the big sur I tested it on:

	$ uname -v
	#33 SMP Sat Sep 28 10:58:07 PDT 2002
	$ dmesg|fgrep 'Probing PCI'
	PCI: Probing PCI hardware on bus (00)
	PCI: Probing PCI hardware on bus (01)
	PCI: Probing PCI hardware on bus (02)
	PCI: Probing PCI hardware on bus (03)
	PCI: Probing PCI hardware on bus (04)

Could it be that your i2000 has old firmware?  My big sur reports the
following firmware version (via Stephane's smbiosview utility):

	version      : W460GXBS2.86E.0117C.P09.200108091154

  --david