Issue Type: | Bug |
---|---|
Priority: | 3 - Elevated |
Status: | Resolved |
Created at: | 2018-01-04T14:18:14.235Z |
Updated at: | 2018-06-09T14:18:59.166Z |
Created by: | Former user |
---|---|
Reported by: | Former user |
Assigned to: | Former user |
Fixed: A fix for this issue is checked into the tree and tested.
(Resolution Date: 2018-03-28T19:55:23.400Z)
2018-03-15 Nibelheim (Release Date: 2018-03-15)
The fundamental approach behind the fix is to:
I can't speak for @accountid:62561a9f9770e60071696c8f and @accountid:62431b8f258562006fa2866a who aren't around right now, but this has been running (with the follow on PCID fixes) on staging-1 and staging-2 for quite a while now.
In terms of explicit testing, a lot of stepping through traps and syscalls, testing with kpti=0, hitting the unusual trap paths (such as explicit ints from userspace), etc., LDT usage, was done. The changes were loaded in various ways - as well as staging, a bunch of KVM load (starts/stops/stress-ng, etc.) as well as Illumos builds.
Aside from the staging machines this was also tested in various ways under VMWare, and a few differently sized machines.
Additionally this was ported across to the bhyve branch just to check we're not going to clash there.
illumos-joyent commit d85fbfe15cf9925f83722b6d62da49d549af615c (branch master, by Alex Wilson)
OS-6547 Implement KPTI
Reviewed by: John Levon <john.levon@joyent.com>
Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Approved by: Jerry Jelinek <jerry.jelinek@joyent.com>