separate per-cpu stacks for executing kprobes handlers arrange to disable interrupts during handler execution
Created attachment 484 [details] Run kprobes handlers on a separate stack - i386 This patch is described in my SystemTap posting of 5/11/05 and in CVS: patches/kprobe_stack (sic).
I've attached a working prototype for i386.
The necessity for this feature is diminished significantly by the SystemTap translator's practice of statically allocating space for the handler's "local" variables.
That practice was factored in at the time this suggestion was initially raised. We still use probably a few hundred bytes of stack, just with kprobes and translator artifacts. The question was whether even those few hundred could pose a problem.
Systemtap probe handlers now check for a minimum of available stack space upon entry, so maybe we don't need to pursue this any more.