Bug 53861

Summary:	nVMX: inaccuracy in emulation of entry failure
Product:	Virtualization	Reporter:	Nadav Har'El (nyh)
Component:	kvm	Assignee:	virtualization_kvm
Status:	NEW ---
Severity:	enhancement
Priority:	P1
Hardware:	All
OS:	Linux
Kernel Version:		Subsystem:
Regression:	No	Bisected commit-id:
Bug Depends on:
Bug Blocks:	94971, 53601

Description Nadav Har'El 2013-02-14 15:13:28 UTC

Emulation of nested entry (L1->L2) failure is rather involved, and there are two kinds of entry failures - some recognized before the vmcs02 was touched (and nested_vmx_failValid/Invalid() is used), and some after we started to touch vmcs02 (and nested_vmx_entry_failure() is used). This whole business is explained in the Intel SDM, section "VM-entry failures during or after loading guest state".

But where's a corner cases related to *buggy L0* that we probably do not emulate sensibly:

Imagine that L0 runs L2 for L1, and succeeds, but then exits to L0 for some reason and L0 handles this event (without L1's involvement) and wants to resume L2. What if this entry fails, e.g., because we (L0) filled some vmcs02 field incorrectly? Neither nested_vmx_failValid() or nested_vmx_entry_failure() are appropriate because L2 did run for a while and most likely changed vmcs02 (so we need to update vmcs12 with prepare_vmcs12()).

This can only happen due to L0 bug (which sets something wrong in the vmcs) so perhaps the best solution is just to kill L1 in this case? Is there a better solution?