[Impact]
When running kvm-unit-tests under a guest, it will be paused, requiring a
reset. When running the same test on a host (vmx_nm_test), it will fail.
[Test case]
Grab kvm-unit-tests, build it and run:
TESTNAME=vmx TIMEOUT=90s ACCEL= ./x86/run x86/vmx.flat -smp 1 -cpu host,+vmx -append "vmx_nm_test"
If done inside a guest, when the host runs the bionic 4.15 kernel, the guest
will pause.
[Potential regressions]
Nested KVM could stop working. Floating point could stop working on KVM
guests, though the code that relied on this was already removed from
Bionic.
------------------------------
This issue was first spotted on Mar.16 [1]
The ubuntu_kvm_unit_tests will be interrupted on X-oracle-4.15 on both VM.Standard2.1 and VM.Standard2.16, this is not a regression since it can be reproduced with 4.15.0-1031-oracle #34~16.04.1:
Running '/home/ubuntu/autotest/client/tmp/ubuntu_kvm_unit_tests/src/kvm-unit-tests/tests/vmx_nm_test'
BUILD_HEAD=4671e4ba
timeout -k 1s --foreground 30 /usr/bin/qemu-system-x86_64 -nodefaults -device pc-testdev -device isa-debug-exit,iobase=0xf4,iosize=0x4 -vnc none -serial stdio -device pci-testdev -machine accel=kvm -kernel /tmp/tmp.da3iFrsCzC -smp 1 -cpu host,+vmx -append vmx_nm_test # -initrd /tmp/tmp.h2DFw8L0AF
enabling apic
paging enabled
cr0 = 80010011
cr3 = 477000
cr4 = 20
Test suite: vmx_nm_test
client_loop: send disconnect: Broken pipe
(node disconnected here)
Before the test started, this can be found in syslog:
Apr 13 06:26:25 selfprovisioned-phlin-kvm-unit kernel: [ 1073.529005] L1TF CPU bug present and SMT on, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.
After that, noting was printed and node disconnected.
If you try to run this case manually, it will stop at:
# ./vmx_nm_test
BUILD_HEAD=4671e4ba
ready!!!
timeout -k 1s --foreground 30 /usr/bin/qemu-system-x86_64 -nodefaults -device pc-testdev -device isa-debug-exit,iobase=0xf4,iosize=0x4 -vnc none -serial stdio -device pci-testdev -machine accel=kvm -kernel /tmp/tmp.ZcGrnXu6se -smp 1 -cpu host,+vmx -append vmx_nm_test # -initrd /tmp/tmp.ADjEOAcRKM
enabling ap
(stopped here, even the "enabling apic" string was not printed)
It looks like this is a new test case added since the cycle of 4.15.0-1037.41~16.04.1-oracle
[1] https://bugs.launchpad.net/ubuntu-kernel-tests/+bug/1867623/comments/2
ProblemType: Bug
DistroRelease: Ubuntu 16.04
Package: linux-image-4.15.0-1031-oracle 4.15.0-1031.34~16.04.1
ProcVersionSignature: User Name 4.15.0-1031.34~16.04.1-oracle 4.15.18
Uname: Linux 4.15.0-1031-oracle x86_64
ApportVersion: 2.20.1-0ubuntu2.21
Architecture: amd64
Date: Mon Apr 13 05:18:03 2020
ProcEnviron:
TERM=xterm-256color
PATH=(custom, no user)
XDG_RUNTIME_DIR=<set>
LANG=en_US.UTF-8
SHELL=/bin/bash
SourcePackage: linux-signed-oracle
UpgradeStatus: No upgrade log present (probably fresh install)
On Oracle Bionic 4.15.0-1038.42, it has finished (but failed with bug 1872419) on VM.Standard2.1