On Wed, Apr 22, 2020 at 7:26 PM Sean Christopherson
<[email protected]> wrote:
>
> Short circuit vmx_check_nested_events() if an exception is pending and
> needs to be injected into L2, priority between coincident events is not
> dependent on exiting behavior. This fixes a bug where a single-step #DB
> that is not intercepted by L1 is incorrectly dropped due to servicing a
> VMX Preemption Timer VM-Exit.
>
> Injected exceptions also need to be blocked if nested VM-Enter is
> pending or an exception was already injected, otherwise injecting the
> exception could overwrite an existing event injection from L1.
> Technically, this scenario should be impossible, i.e. KVM shouldn't
> inject its own exception during nested VM-Enter. This will be addressed
> in a future patch.
>
> Note, event priority between SMI, NMI and INTR is incorrect for L2, e.g.
> SMI should take priority over VM-Exit on NMI/INTR, and NMI that is
> injected into L2 should take priority over VM-Exit INTR. This will also
> be addressed in a future patch.
>
> Fixes: b6b8a1451fc4 ("KVM: nVMX: Rework interception of IRQs and NMIs")
> Reported-by: Jim Mattson <[email protected]>
> Cc: Oliver Upton <[email protected]>
> Cc: Peter Shier <[email protected]>
> Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Jim Mattson <[email protected]>
