bus_error during startup guest kernel

[huang-jl]

Describe the bug

During guest kernel startup (after sending InstanceStart to firecracker), the firecracker log will contains the following message:

[anonymous-instance:fc_vcpu 0:WARN:src/vmm/src/arch/x86_64/vcpu.rs:725] vcpu: IO write @ 0xcf8:0x4 failed: bus_error: MissingAddressRange
[anonymous-instance:fc_vcpu 0:WARN:src/vmm/src/arch/x86_64/vcpu.rs:715] vcpu: IO read @ 0xcfc:0x2 failed: bus_error: MissingAddressRange
[anonymous-instance:fc_vcpu 0:WARN:src/vmm/src/arch/x86_64/vcpu.rs:715] vcpu: IO read @ 0x87:0x1 failed: bus_error: MissingAddressRange

To Reproduce

I just follow the get-started, except I run firecracker as follows (i.e., I do not enable pci):

sudo ./firecracker --api-sock "${API_SOCKET}"

Actually, the firecracker just warn!() that error, instead of panic, and the guest was successfully booted and enter the userspace. However, this error message is shown in the firecracker log, I wonder is there something wrong, or it is the expect behaviour when kernel boot up?

Expected behaviour

There should be no such message, which seems to be an error for users.

Environment

• Firecracker version: v.1.13.1
• Host and guest kernel versions: Host is ubuntu 6.8.0-83-generic, guest is 6.1.141 as in get-started
• Rootfs used: As in the get-started
• Architecture: x86_64

Checks

• Have you searched the Firecracker Issues database for similar problems?
• Have you read the existing relevant Firecracker documentation?
• Are you certain the bug being reported is a Firecracker issue?

[huang-jl]

This is the corresponding firecrakcer log generate for ${LOGFILE} as in the get-started.

firecracker.log

[bchalios]

Hi @huang-jl and thanks for reporting this.

This behaviour is expected if you launch a kernel with PCI support and not enable PCI in Firecracker. What happens is that the guest kernel is trying to probe the system's PCI configuration space over the IO ports 0xcf8 and 0xcfc. Since PCI is not enabled Firecracker doesn't handle accesses to these ports and warns these messages.

This should not be a functional problem for your microVM which should boot correctly. It does include though a bit of overhead in the boot process, so you should either explicitly enable PCI when launching Firecracker or use a kernel which isn't PCI-enabled.

Note that the guest also tries to access port 0x87 which is used to probe DMA controllers which Firecracker doesn't support hence you see the same behaviour.

[huang-jl]

@bchalios Thanks for your clear explanation, I understand the reason of these messages.

Actually, I was upgrade from v1.9.1 to v1.13.1. In the docs of v1.9.1, it seems like since the FC start to support ACPI, we should enable CONFIG_PCI for the guest kernel.

Thus, if I choose to use MMIO virtio transport (i.e., do not pass --enable-pci), do I have to bear this warnings and the extra device probe overhead during startup?