Re: Recent Power changes and stack_trace_save_tsk_reliable?
From: Joe Lawrence <joe.lawrence@redhat.com>
Date: 2023-08-30 21:48:40
Also in:
live-patching
On 8/30/23 02:37, Michael Ellerman wrote:
Michael Ellerman [off-list ref] writes:quoted
Joe Lawrence [off-list ref] writes:quoted
Hi ppc-dev list, We noticed that our kpatch integration tests started failing on ppc64le when targeting the upstream v6.4 kernel, and then confirmed that the in-tree livepatching kselftests similarly fail, too. From the kselftest results, it appears that livepatch transitions are no longer completing.Hi Joe, Thanks for the report. I thought I was running the livepatch tests, but looks like somewhere along the line my kernel .config lost CONFIG_TEST_LIVEPATCH=m, so I have been running the test but it just skips. :/
That config option is easy to drop if you use `make localmodconfig` to try and expedite the builds :D Been there, done that too many times.
quoted
I can reproduce the failure, and will see if I can bisect it more successfully.It's caused by: eed7c420aac7 ("powerpc: copy_thread differentiate kthreads and user mode threads") Which is obvious in hindsight :) The diff below fixes it for me, can you test that on your setup?
Thanks for the fast triage of this one. The proposed fix works well on our setup. I have yet to try the kpatch integration tests with this, but I can verify that all of the kernel livepatching kselftests now happily run. -- Joe
quoted hunk ↗ jump to hunk
A proper fix will need to be a bit bigger because the comments in there are all slightly wrong now since the above commit. Possibly we can also rework that code more substantially now that copy_thread() is more careful about setting things up, but that would be a follow-up.diff --git a/arch/powerpc/kernel/stacktrace.c b/arch/powerpc/kernel/stacktrace.c index 5de8597eaab8..d0b3509f13ee 100644 --- a/arch/powerpc/kernel/stacktrace.c +++ b/arch/powerpc/kernel/stacktrace.c@@ -73,7 +73,7 @@ int __no_sanitize_address arch_stack_walk_reliable(stack_trace_consume_fn consum bool firstframe; stack_end = stack_page + THREAD_SIZE; - if (!is_idle_task(task)) { + if (!(task->flags & PF_KTHREAD)) { /* * For user tasks, this is the SP value loaded on * kernel entry, see "PACAKSAVE(r13)" in _switch() andcheers