[Bug gdb/26819] RISC-V: internal-error: int finish_step_over(execution_control_state*): Assertion

["andrew.burgess at embecosm dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=26819

--- Comment #40 from Andrew Burgess <andrew.burgess at embecosm dot com> ---
Final comment for the night.  I now have this test passing reliably, I changed
the file spike-2-hwthread.cfg (this is one of the files that the steps to
reproduce document has us download), this is the openocd config file.  Within
this file are two lines like this:

  target create $_TARGETNAME_0 riscv -chain-position $_CHIPNAME.cpu -rtos
hwthread
  target create $_TARGETNAME_1 riscv -chain-position $_CHIPNAME.cpu -coreid 1

I changed these lines to be this:

  target create $_TARGETNAME_0 riscv -chain-position $_CHIPNAME.cpu -rtos
hwthread -coreid 0
  target create $_TARGETNAME_1 riscv -chain-position $_CHIPNAME.cpu -rtos
hwthread -coreid 1

And now the test passes just fine.

As I've said in previous comments, I'm no openocd expert, I added the extra
'-rtos hwthread' for the second core as it is the presence of this rtos setting
that is causing openocd to send back the thread-id.  Maybe there's more going
on here and adding this flag is bad for some reason?  I added the '-coreid 0'
to the first core just for symmetry, I suspect 0 is the default.

Unless there's new evidence that comes up, I'm pretty convinced that the issue
here was either openocd has a bug, or openocd was misconfigured, however,  I
think there is an open issue for whether GDB could/should have handled this
situation better?

I think that the packets coming from the remote should be considered user
input, and as such, we should not rely on them being "correct".  As such, when
invalid packets arrive GDB should be throwing an error, not quitting with an
assertion failure.

I'm not really sure where we should be catching this particular error, or what
we should do.  I doubt we can reasonably expect to recover, I think the target
has shown itself to be unreliable in this case (reporting a stop against a
thread that should not be running)...

-- 
You are receiving this mail because:
You are on the CC list for the bug.