tree cb028e6c87dd694717e23597db9de0ce83972327
parent 9cacfafb6850a144ae33523993e3870fc8b6220d
author Barret Rhoden <brho@cs.berkeley.edu> 1484345623 -0500
committer Barret Rhoden <brho@cs.berkeley.edu> 1484751602 -0500

x86: vmm: Rework VMRESUME logic

The old code assumed that if we found a loaded VMCS (non-partial TF, VMCS
still sitting around), then it was ready to be resumed.  This would only be
true if we previously actually ran the VMCS.

It's possible for us to have the VMCS loaded, but never launched.  This was
pretty easy to do: launch it as an SCP.  Then the first time we tried to
run the VM, it was on Core 0, where it had been created and was still
sitting around.

When dealing with a VMCS, we must do the following:
- load it
- launch it
- resume it.

The old code thought that 'loaded' (lingering on the core) meant 'ready to
resume.'

The bug was pretty easy to find and sort out, once I had VMEXITs reflected
to userspace (instead of locking up the machine), and I could see the error
number.  (exit_qual == 5 -> VMRESUME with non-launched VMCS).

Signed-off-by: Barret Rhoden <brho@cs.berkeley.edu>