Hello,

I send a dedicated subject message on this topic (second attempt because the first one seems not to be present in ovirt archive..).

Also, the reply to my other related message seems not visible inside list archive page for some reason.

It seems this nasty problem in nested virt using  pc-i440fx-rhel7.X.0 machine type with X >= 3 impacts not only vSphere as main hypervisor for nested KVM, but other hypervisors too (Hyper-V) and other machine types too and could be due to a bug in KVM, so in the kernel, if I understood correctly.

According to this link below 

and its comment by Roman Kagan in June this year:

"

This is a KVM bug. It has been fixed in mainstream Linux in

commit d391f1207067268261add0485f0f34503539c5b0

Author: Vitaly Kuznetsov <email address hidden>

Date: Thu Jan 25 16:37:07 2018 +0100

    x86/kvm/vmx: do not use vm-exit instruction length for fast MMIO when running nested

    I was investigating an issue with seabios >= 1.10 which stopped working

    for nested KVM on Hyper-V. The problem appears to be in

    handle_ept_violation() function: when we do fast mmio we need to skip

    the instruction so we do kvm_skip_emulated_instruction(). This, however,

    depends on VM_EXIT_INSTRUCTION_LEN field being set correctly in VMCS.

    However, this is not the case.

    Intel's manual doesn't mandate VM_EXIT_INSTRUCTION_LEN to be set when

    EPT MISCONFIG occurs. While on real hardware it was observed to be set,

    some hypervisors follow the spec and don't set it; we end up advancing

    IP with some random value.

    I checked with Microsoft and they confirmed they don't fill

    VM_EXIT_INSTRUCTION_LEN on EPT MISCONFIG.

    Fix the issue by doing instruction skip through emulator when running

    nested.

    Fixes: 68c3b4d1676d870f0453c31d5a52e7e65c7448ae

    Suggested-by: Radim Krčmář <email address hidden>

    Suggested-by: Paolo Bonzini <email address hidden>

    Signed-off-by: Vitaly Kuznetsov <email address hidden>

    Acked-by: Michael S. Tsirkin <email address hidden>

    Signed-off-by: Radim Krčmář <email address hidden>

Although the commit mentions Hyper-V as L0 hypervisor, the same problem pertains to ESXi.

The commit is included in v4.16.

"

Is it possible to backport the fix to the kernel provided by plain RHEL/CentOS hosts and/or RHVH/ovirt-ng nodes?

Thanks,

Gianluca