On Wed, Jul 17, 2019 at 01:01:27PM +0200, Miroslav Benes wrote:

quoted

On a related note, do you think it would be feasible to extend (in
another patchset) the reliable stack unwinding code a bit so that we
could feed it pre-baked stacks ... then we could verify that the code
was finding interesting scenarios.  That was a passing thought I had
back when Nicolai and I were debugging the ppc64le exception frame
marker bug, but didn't think it worth the time/effort at the time.

That is an interesting thought. It would help the testing a lot. I will 
make a note in my todo list.

Another idea I had for reliable unwinder testing: add a
CONFIG_RELIABLE_STACKTRACE_DEBUG option which does a periodic stack
trace and warns if it doesn't reach the end.  It could triggered from a
periodic NMI, or from schedule().

-- 
Josh