Re: [PATCH v3 0/2] KVM: nVMX: add tracepoints for nested VM-Enter failures

[Paolo Bonzini <pbonzini@redhat.com>]

On 11/07/19 17:58, Sean Christopherson wrote:
> Debugging VM-Enter failures has been the bane of my existence for years.
> Seeing KVM's VMCS dump format pop up on a console triggers a Pavlovian
> response of swear words and sighs.  As KVM's coverage of VM-Enter checks
> improve, so too do the odds of being able to triage/debug a KVM (or any
> other hypervisor) bug by running the bad KVM build as an L1 guest.
> 
> Improve support for using KVM to debug a buggy VMM by adding tracepoints
> to capture the basic gist of a VM-Enter failure so that extracting said
> information from KVM doesn't require attaching a debugger or modifying
> L0 KVM to manually log failures.
> 
> The captured information is by no means complete or perfect, e.g. I'd
> love to capture *exactly* why a consistency check failed, but logging
> that level of detail would require invasive code changes and might even
> act as a deterrent to adding more checks in KVM.
> 
> v3: Fix a minor snafu in the v2 rebase, and re-rebase to kvm/next
>     (a45ff5994c9c, "Merge tag 'kvm-arm-for-5.3'...")
> 
> v2: Rebase to kvm/queue.
> 
> Sean Christopherson (2):
>   KVM: nVMX: add tracepoint for failed nested VM-Enter
>   KVM: nVMX: trace nested VM-Enter failures detected by H/W
> 
>  arch/x86/include/asm/vmx.h |  14 ++
>  arch/x86/kvm/trace.h       |  22 +++
>  arch/x86/kvm/vmx/nested.c  | 269 ++++++++++++++++++++-----------------
>  arch/x86/kvm/x86.c         |   1 +
>  4 files changed, 179 insertions(+), 127 deletions(-)
> 

Queued, thanks.

Paolo