[SRU artful] LP#1747507 starting KVM instances hangs box

Andy Whitcroft apw at canonical.com
Tue Feb 6 16:31:14 UTC 2018


We have early reports of failures when attempting to start KVM VMs on
machines running the retpoline based kernels.  This is triggered by
a flaw in the RSB stuffing code retpoline introduces.  This pull request
pulls in 5 upstream commits (all from the stable 4.14 branch) which
correct this code.

With these applied I am again able to start VMs.

Proposing to add to artful linux where the retpoline patches are
applied.

-apw

The following changes since commit bc3391e235def1ebcded0952bd4418ee2429bedc:

  UBUNTU: [Config] UNMAP_KERNEL_AT_EL0=y && HARDEN_BRANCH_PREDICTOR=y (2018-02-05 18:23:06 +0100)

are available in the Git repository at:

  https://git.launchpad.net/~apw/ubuntu/+source/linux/+git/pti pti/artful-retpoline-intelv1--pull2

for you to fetch changes up to c95f498604801249575313ee5007c378208b2659:

  x86/retpoline: Simplify vmexit_fill_RSB() (2018-02-06 16:23:56 +0000)

----------------------------------------------------------------
  * starting VMs causing retpoline4 to reboot (LP: #1747507) // CVE-2017-5715
    (Spectre v2 retpoline)
    - x86/retpoline: Fill RSB on context switch for affected CPUs
    - x86/retpoline: Add LFENCE to the retpoline/RSB filling RSB macros
    - x86/retpoline: Optimize inline assembler for vmexit_fill_RSB
    - x86/retpoline: Remove the esp/rsp thunk
    - x86/retpoline: Simplify vmexit_fill_RSB()




More information about the kernel-team mailing list