[vfio-users] vfio_pci_read_config failed: Bad address

Alex Williamson alex.williamson at redhat.com
Tue Nov 17 22:41:21 UTC 2015


On Tue, 2015-11-17 at 23:15 +0100, Ronnie Swanink wrote:
> Hello,
> 
> The kernel options are:
> [root at localhost ronnie]# cat /proc/cmdline
> BOOT_IMAGE=/vmlinuz-4.3.0-1.vanilla.mainline.knurd.1.fc23.x86_64 
> root=/dev/mapper/fedora-root ro rd.lvm.lv=fedora/root 
> rd.lvm.lv=fedora/swap rhgb quiet intel_iommu=on rd.driver.pre=vfio-pci
> 
> See http://pastebin.com/GwhWbshH for a dump of dmesg
> 
> And to be sure: http://pastebin.com/4kxSuqzA for a dump with the 
> standard 4.2.5-300 kernel.

Looks like a dp55kg Intel motherboard and you seem to have the latest
BIOS.  Unfortunately it's very broken.  I'd have expected better from an
Intel board:

[    0.194621] PCI: MMCONFIG for domain 0000 [bus 00-3f] at [mem 0xf8000000-0xfbffffff] (base 0xf8000000)
[    0.195260] [Firmware Info]: PCI: MMCONFIG at [mem 0xf8000000-0xfbffffff] not reserved in ACPI motherboard resources
[    0.195262] PCI: not using MMCONFIG
[    0.195272] PCI: Using host bridge windows from ACPI; if necessary, use "pci=nocrs" and report a bug
[    0.199537] ACPI: PCI Root Bridge [PCI0] (domain 0000 [bus 00-3e])
[    0.199541] acpi PNP0A08:00: _OSC: OS supports [ASPM ClockPM Segments MSI]
[    0.199599] acpi PNP0A08:00: _OSC: not requesting OS control; OS requires [ExtendedConfig ASPM ClockPM MSI]
[    0.199669] acpi PNP0A08:00: fail to add MMCONFIG information, can't access extended PCI configuration space under this bridge.

You could try manually reserving the MMCONFIG range using a boot option
like memmap=64M$0xf8000000

I've booted my system with pci=nommconf, I don't see nearly as many
-EFAULTs as you got, but it does break my GPU VM.  I think those
accesses are happening through the nvidia config space mirror, that's
the only way the guest in my 440fx VM could get to extended config
space.  That sort of implies you're not getting anywhere with nvidia
assignment until you can get extended config space enabled on your
system.  Thanks,

Alex




More information about the vfio-users mailing list