Re: [GIT PULL] PCI fixes for v3.19

[Bjorn Helgaas <bhelgaas@google.com>]

On Mon, Jan 26, 2015 at 01:24:51PM -0800, Tony Luck wrote:
> On Mon, Jan 26, 2015 at 1:02 PM, Bjorn Helgaas <bhelgaas@google.com> wrote:
> > Sorry for the inconvenience.  Can you collect a complete dmesg log and
> > "lspci -vv" output, too (from the kernel with the reverted commit)?
> > That will have more useful information than just /proc/iomem.
> 
> Full dmesg, lspci -vv, and bonus .config (CONFIG_PCI_IOV is indeed
> not set)

The ROM part is something we should fix:

  pci 0000:01:00.1: can't claim BAR 6 [mem 0xfffe0000-0xffffffff pref]: no compatible bridge window

The ROM BAR is probably disabled, and we shouldn't complain about this if
it's disabled.  Yinghai?

The stuff on bus 0000:80 looks like at least partly a firmware problem:

  ACPI: PCI Root Bridge [PCI1] (domain 0000 [bus 80-ff])
  acpi PNP0A08:01: host bridge window [io  0x9000-0xfffe]
  pci 0000:80:01.0: PCI bridge to [bus 81]
  pci 0000:80:01.0:   bridge window [io  0xa000-0xafff]
  pci 0000:80:01.0:   bridge window [mem 0xa0100000-0xa01fffff]
  pci 0000:80:13.0: reg 0x10: [mem 0xa0220000-0xa0220fff]

ACPI told us about an I/O port aperture, but didn't mention any MMIO
apertures through the host bridge.  Therefore, we have no MMIO space to
assign to the devices on [bus 80-ff].

But presumably these devices, e.g., the igb devices at 81:00.0 and 81:00.1
and the mpt device at 83:00.0, actually DO work, so the PCI1 host bridge
must actually forward some MMIO space that ACPI didn't tell us about.  I
think on x86, we would fall back to whatever the firmware left in those
BARs.  We should probably do the same on x86.

Bjorn