Fixes for bz-2121368
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2121368
Add a few more backports to the retbleed backport. These were skipped and manual conflict resolution lead a missing ret
instruction at the end of the ftrace trampoline for vulnerable AMD CPUs. This finally manifested in CKI and bugged as bz-2121368. From Josh P:
I found that it's a bug in the retbleed ftrace backport. Confusingly, the upstream text_gen_insn() actually writes the instruction to memory, whereas the RHEL9 version does not. So on retbleed-affected AMD, the ftrace trampoline is missing a return (specifically, jmp __x86_return_thunk).
Interesting to note in this patchset:
- ("x86/ibt,ftrace: Make function-graph play nice") : A previously avoided backport, undoes the mentioned conflict resolution from d172ce9b ("x86/ftrace: Use alternative RET encoding") so that ...
- ("x86/ftrace: Use alternative RET encoding") : Backported a second time, now with no conflicts
Signed-off-by: Joe Lawrence joe.lawrence@redhat.com
Edited by Joe Lawrence