[vfio-users] vfio_pci_read_config failed: Bad address
Ronnie Swanink
ronnie at ronnieswanink.nl
Wed Nov 18 00:29:35 UTC 2015
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
More information about the vfio-users
mailing list