tree 0d8c906b0f511014b815cb246a84100274c7702a
parent baa3fb322ac8623cdc78b6d65ea884209494af1b
author Alex Legg <alexlegg@google.com> 1544243830 +1100
committer CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> 1544655886 +0000

[x86][hypervisor] Clear tracked exceptions when injecting an interrupt

Also, prioritise NMIs above other interrupts and exceptions.

This was causing flakiness in the guest integration tests on release
builds through the following order of events:
1. Linux guest begins a safe rdmsr
2. Hypervisor correctly responds by tracking a GP fault
3. On resume, a timer interrupt was prioritised before the exception
4. The guest exits again, then resumes injecting the GP fault
5. Guest panics because it receives an exception during the timer
interrupt handler.

This is likely also the cause for similar debian_guest failures in the
past.

There are two additional changes to follow: 1) Tidy up the use of
X86_INT_PLATFORM_BASE to use X86_INT_MAX_INTEL_DEFINED instead, and
handling undefined vectors. 2) Improved testing (See MAC-225).

MAC-223 #done

TEST=Run guest_integration_tests in a loop with a release build

Change-Id: Ide64857f5442b023ac49c6369a2de8aaa305e2fe