[vfio-users] vfio_pci_read_config failed: Bad address

[Ronnie Swanink]

Hello,

Adding the boot option memmap=64M$0xf8000000 prevents the system from 
booting. There is no output on the screen, and it just hangs right after 
grub.

Is there something else I can try?

Thanks,
Ronnie Swanink



On 17-11-2015 23:41, Alex Williamson wrote:
> 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