Similarly to bug #2185, because we don't have a separate stack for kprobe handling (bug #931), we may have stack exhaustion problems in ordinary usage without reentrancy. We go out of our way to minimize stack usage, but we need to - estimate or measure stack usage of a worst-case scenario (kprobes handler, deeply nested systemtap expression, calling into the most complex runtime routine) - add code to the probe entry functions to check for available stack space, and punt (skip probe) if it's not sufficient
patch committed. The default MINSTACKSPACE may need to be tuned in the future.