On Thu, Oct 14, 2021 at 07:07:38PM +0100, Mark Rutland wrote:

On Fri, Oct 08, 2021 at 09:17:07AM -0700, Josh Poimboeuf wrote:

quoted

On Fri, Oct 08, 2021 at 03:45:59PM +0200, Peter Zijlstra wrote:

quoted

quoted

stack_trace_save_tsk() *shouldn't* skip anything unless we've explicitly
told it to via skipnr, because I'd expect that

It's what most archs happen to do today and is what
stack_trace_save_tsk() as implemented using arch_stack_walk() does.
Which is I think the closest to canonical we have.

Ah; and arch_stack_walk() itself shouldn't skip anything, which gives
the consistent low-level semantic I wanted.

quoted

It *is* confusing though.  Even if 'nosched' may be the normally
desired behavior, stack_trace_save_tsk() should probably be named
stack_trace_save_tsk_nosched().

I agree that'd be less confusing!

Josh, am I right in my understanding that the reliable stacktrace
functions *shouldn't* skip sched functions, or should those similarly
gain a _nosched suffix?

Correct, the reliable variants need to see the entire call stack and
therefore they shouldn't skip sched functions.

-- 
Josh