Hi,
I don't think "stepi" should ever execute more than one instruction
(special cases like branch delay slots excepted). However the return
trampoline code skips over such pieces in one go even in the
instruction-stepping mode.
Here's a fix to address this problem. Unfortunately this hook is
currently only used by two platforms -- RS6000/AIX and PA-RISC/HP-UX --
both of which are too exotic for me to test, but the change should be
obvious anyway.
The test suite doesn't really cover instruction-stepping (being too
architecture-specific for generic coverage), but this was noticed in
manual MIPS16 testing with one of the followup changes that uses the
return trampoline hook to handle MIPS16 return thunks.
OK to apply?
2011-12-06 Maciej W. Rozycki <
macro@...>
Maciej W. Rozycki <
macro@...>
gdb/
* infrun.c (handle_inferior_event): Don't proceed through
shared library trampolines if stepping at the machine
instruction level.
Maciej
gdb-return-trampoline-step.diff
Index: gdb-fsf-trunk-quilt/gdb/infrun.c
===================================================================
--- gdb-fsf-trunk-quilt.orig/gdb/infrun.c 2011-11-08 11:25:58.635899589 +0000
+++ gdb-fsf-trunk-quilt/gdb/infrun.c 2011-11-08 11:26:07.545882458 +0000
@@ -5009,7 +5009,8 @@ handle_inferior_event (struct execution_
/* If we're in the return path from a shared library trampoline,
we want to proceed through the trampoline when stepping. */
if (gdbarch_in_solib_return_trampoline (gdbarch,
- stop_pc, ecs->stop_func_name))
+ stop_pc, ecs->stop_func_name)
+ && ecs->event_thread->control.step_over_calls != STEP_OVER_NONE)
{
/* Determine where this trampoline returns. */
CORE_ADDR real_stop_pc;
