Re: [kvm-unit-tests PATCH v1] s390x: Improve stack traces that contain an interrupt frame

[Nina Schoetterl-Glausch <nsg@linux.ibm.com>]

On Mon, 2023-04-17 at 09:57 +0200, Nico Boehr wrote:
> Quoting Nina Schoetterl-Glausch (2023-04-05 14:35:08)
> > When we encounter an unexpected interrupt we print a stack trace.
> > While we can identify the interrupting instruction via the old psw,
> > we don't really have a way to identify callers further up the stack,
> > since we rely on the s390x elf abi calling convention to perform the
> > backtrace. An interrupt is not a call, so there are no guarantees about
> > the contents of the stack and return address registers.
> > If we get lucky their content is as we need it or valid for a previous
> > callee in which case we print one wrong caller and then proceed with the
> > correct ones.
> 
> I did not think too much about it, so it might not work, but how about a
> seperate interrupt stack?
> 
> Then, we could print the interrupt stack trace (which should be correct) and -
> with a warning as you suggest - the maybe incorrect regular stack trace.

Not sure I'm getting the point. Do you want an implementation that doesn't have
the weirdness of using a frame with a special symbol to warn?
We only output a bunch of caller addresses and pretty_print_stacks.py formats that
into a readable stack trace. So by having the special symbol frame there are no
changes needed to that script, but it certainly would be possible do it differently,
e.g. output "STACK: dead beef WARN 0 ffff" and have the script
print the warning if it sees a WARN.