Re: [OpenOCD-user] [openocd-user@lists.sourceforge.net: v0.11.0-rc2. Found bug when stepping or continue with debugger.]

[Antonio B. <bor...@gm...>]

On Thu, Feb 25, 2021 at 12:12 PM Alexander Gabitov
<ga...@pl...> wrote:
>
> Hi, Antonio,
>
> you are right, I looked at line
> src/server/gdb_server.c:2753   if (retval != ERROR_OK)
> and saw it like:     if (retval == ERROR_OK)
>
> I was wrong about the place where it happened but still the bug have a
> place.
>

Hi Alexander,
please keep the mailing list in copy so everyone can benefit by this
conversation.

Could you please try again after removing this event handler
mb86h61.cpu configure -event resumed { ... }
from your config file?

I replicated the issue on a cortex_m target by simply adding a dummy
<target_name> configure -event resumed {echo "in event resumed"}

I want to understand if there is some other condition in your script
that causes the issue or removing the event "resumed" is enough.

Regards,
Antonio


> In attachment you can see the place in the log where bug happens and my
> comments. I guess arm11_poll() is called in callback:
>           target_resume() -> arm11_resume() ->
> target_call_event_callbacks(target, TARGET_EVENT_RESUMED):
>
> Listing of "target.c":
> ......
>      target_handle_event(target, event); /* Gabitov: May be arm11_poll()
> was called here */
>
>      while (callback) {
>          next_callback = callback->next;
>          callback->callback(target, event, callback->priv); /* Gabitov:
> or which is more probably arm11_poll() was called here */
>          callback = next_callback;
>      }
> ......
>
> 24.02.2021 21:47, Antonio Borneo пишет:
> > Hi Alexander,
> >
> > you point to the line
> > src/server/gdb_server.c:2753  retval = target_poll(target);
> > but that line is executed only when
> > src/server/gdb_server.c:2746  retval = target_resume(target, 1, 0, 0, 0);
> > returns error.
> >
> > In the logs you have provided I do not see such error returned by
> > target_resume().
> > Look like your code calls target_poll() even when target_resume()
> > returns ERROR_OK!
> > Can you please clarify this?
> >
> > So far I cannot replicate this issue, but I do not have an ARM11.
> > Maybe the problem is in arm11_resume()
> >
> > Regards
> > Antonio
> >
> >
> > On Wed, Feb 24, 2021 at 5:28 PM Paul Fertser <fer...@gm...> wrote:
> >>
> >> --
> >> Be free, use free (http://www.gnu.org/philosophy/free-sw.html) software!
> >> mailto:fer...@gm...
> >>
> >>
> >>
> >> ---------- Forwarded message ----------
> >> From: "Габитов Александр Фаритович via OpenOCD-user" <ope...@li...>
> >> To: ope...@li...
> >> Cc:
> >> Bcc:
> >> Date: Wed, 24 Feb 2021 16:24:20 +0500
> >> Subject: [OpenOCD-user] v0.11.0-rc2. Found bug when stepping or continue with debugger.
> >> Hello!
> >>
> >> Recently I had to update the program from version 8-rc1 to version 11-rc2(from archive) to support the J-link v9 debugger. While debugging with a new version of the program when stepping, or when starting the program after a breakpoint - the control did not return to the console, while in the console I saw the following lines:
> >>
> >>      target halted in ARM state due to breakpoint, current mode: System
> >>      cpsr: 0x6000011f pc: 0x41703480
> >>
> >> I found that the problem is in the function 'gdb_handle_vcont_packet' (gdb_server.c):
> >>
> >>      /* simple case, a continue packet */
> >>      if (parse[0] == 'c') {
> >>          gdb_running_type = 'c';
> >>          LOG_DEBUG("target %s continue", target_name(target));
> >>          log_add_callback(gdb_log_callback, connection);
> >>          retval = target_resume(target, 1, 0, 0, 0);
> >>          if (retval == ERROR_TARGET_NOT_HALTED)
> >>              LOG_INFO("target %s was not halted when resume was requested", target_name(target));
> >>
> >>          /* poll target in an attempt to make its internal state consistent */
> >>          if (retval != ERROR_OK) {
> >>              retval = target_poll(target);
> >>              if (retval != ERROR_OK)
> >>                  LOG_DEBUG("error polling target %s after failed resume", target_name(target));
> >>          }
> >>
> >>          /*
> >>           * We don't report errors to gdb here, move frontend_state to
> >>           * TARGET_RUNNING to stay in sync with gdb's expectation of the
> >>           * target state
> >>           */
> >>          gdb_connection->frontend_state = TARGET_RUNNING;
> >>          target_call_event_callbacks(target, TARGET_EVENT_GDB_START);
> >>
> >>          return true;
> >>      }
> >>
> >> I will explain how I did it. I added 3 breakpoints and tried to navigate between them using the continue command. The problem is in string [ retval = target_poll(target); ]. Polling the target here leads to updating it state (if target hits breakpoint before polling), and than when we running frontend it never sees changing of target state. As a result we got target halted but running frontend. The previous version (v0.10.0) works fine, cause it doesnot poll target during process of continue command.
> >>
> >> The same problem when i do stepping (over or into).
> >>
> >> I should mention that sometimes everything works fine because target hits breakpoint after polling in continue command.
> >>
> >> I'll attach console log, debug log and CFG file.
> >>
> >> Note: I added some more debug meassages to source codes and most of them i have marked with label - Gabitov: or gabitov:, but some are unmarked (for example - 'enter DSCR_CORE_HALTED'). So if you find unrecognized message in debug log probably it's mine. The source codes version is 0.11.0-rc2 (from archive - https://sourceforge.net/projects/openocd/files/openocd/0.11.0-rc2/openocd-0.11.0-rc2.zip).
> >>
> >> I will provide more detailed information if needed.
> >>
> >> ------------------
> >>
> >> Alexander Gabitov, Engineer, LLC PLANAR
> >>
> >>
> >> _______________________________________________
> >> OpenOCD-user mailing list
> >> Ope...@li...
> >> https://lists.sourceforge.net/lists/listinfo/openocd-user
>
> --
> Best regards,
> Alexander Gabitov, engineer, "PLANAR" LLC
>