[17/31] Introduce GDB_THREAD_OPTION_EXIT thread option, fix step-over-thread-exit

  When stepping over a breakpoint with displaced stepping, GDB needs to
be informed if the stepped thread exits, otherwise the displaced
stepping buffer that was allocated to that thread leaks, and this can
result in deadlock, with other threads waiting for their turn to
displaced step, but their turn never comes.

Similarly, when stepping over a breakpoint in line, GDB also needs to
be informed if the stepped thread exits, so that is can clear the step
over state and re-resume threads.

This commit makes it possible for GDB to ask the target to report
thread exit events for a given thread, using the new "thread options"
mechanism introduced by a previous patch.

This only adds the core bits.  Following patches in the series will
teach the Linux backends (native & gdbserver) to handle the
GDB_THREAD_OPTION_EXIT option, and then a later patch will make use of
these thread exit events to clean up displaced stepping and inline
stepping state properly.

Change-Id: I96b719fdf7fee94709e98bb3a90751d8134f3a38
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27338
---
 gdb/infrun.c        | 15 ++++++++++-----
 gdb/remote.c        |  9 +++++++++
 gdb/target/target.c |  1 +
 gdb/target/target.h |  4 ++++
 4 files changed, 24 insertions(+), 5 deletions(-)
  

Pedro Alves <pedro@palves.net> writes:

> When stepping over a breakpoint with displaced stepping, GDB needs to
> be informed if the stepped thread exits, otherwise the displaced
> stepping buffer that was allocated to that thread leaks, and this can
> result in deadlock, with other threads waiting for their turn to
> displaced step, but their turn never comes.
>
> Similarly, when stepping over a breakpoint in line, GDB also needs to
> be informed if the stepped thread exits, so that is can clear the step
> over state and re-resume threads.
>
> This commit makes it possible for GDB to ask the target to report
> thread exit events for a given thread, using the new "thread options"
> mechanism introduced by a previous patch.
>
> This only adds the core bits.  Following patches in the series will
> teach the Linux backends (native & gdbserver) to handle the
> GDB_THREAD_OPTION_EXIT option, and then a later patch will make use of
> these thread exit events to clean up displaced stepping and inline
> stepping state properly.

LGTM.

Reviewed-By: Andrew Burgess <aburgess@redhat.com>

Thanks,
Andrew

>
> Change-Id: I96b719fdf7fee94709e98bb3a90751d8134f3a38
> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27338
> ---
>  gdb/infrun.c        | 15 ++++++++++-----
>  gdb/remote.c        |  9 +++++++++
>  gdb/target/target.c |  1 +
>  gdb/target/target.h |  4 ++++
>  4 files changed, 24 insertions(+), 5 deletions(-)
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 6fdffb31884..e47e3c688e7 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -2445,24 +2445,29 @@ do_target_resume (ptid_t resume_ptid, bool step, enum gdb_signal sig)
>    else
>      target_pass_signals (signal_pass);
>  
> -  /* Request that the target report thread-{created,cloned} events in
> -     the following situations:
> +  /* Request that the target report thread-{created,cloned,exited}
> +     events in the following situations:
>  
>       - If we are performing an in-line step-over-breakpoint, then we
>         will remove a breakpoint from the target and only run the
>         current thread.  We don't want any new thread (spawned by the
> -       step) to start running, as it might miss the breakpoint.
> +       step) to start running, as it might miss the breakpoint.  We
> +       need to clear the step-over state if the stepped thread exits,
> +       so we also enable thread-exit events.
>  
>       - If we are stepping over a breakpoint out of line (displaced
>         stepping) then we won't remove a breakpoint from the target,
>         but, if the step spawns a new clone thread, then we will need
>         to fixup the $pc address in the clone child too, so we need it
> -       to start stopped.
> +       to start stopped.  We need to release the displaced stepping
> +       buffer if the stepped thread exits, so we also enable
> +       thread-exit events.
>    */
>    if (step_over_info_valid_p ()
>        || displaced_step_in_progress_thread (tp))
>      {
> -      gdb_thread_options options = GDB_THREAD_OPTION_CLONE;
> +      gdb_thread_options options
> +	= GDB_THREAD_OPTION_CLONE | GDB_THREAD_OPTION_EXIT;
>        if (target_supports_set_thread_options (options))
>  	tp->set_thread_options (options);
>        else
> diff --git a/gdb/remote.c b/gdb/remote.c
> index f7ab8523fd5..0d2b7c09a07 100644
> --- a/gdb/remote.c
> +++ b/gdb/remote.c
> @@ -3998,6 +3998,15 @@ remote_target::update_thread_list ()
>  	      if (has_single_non_exited_thread (tp->inf))
>  		continue;
>  
> +	      /* Do not remove the thread if we've requested to be
> +		 notified of its exit.  For example, the thread may be
> +		 displaced stepping, infrun will need to handle the
> +		 exit event, and displaced stepping info is recorded
> +		 in the thread object.  If we deleted the thread now,
> +		 we'd lose that info.  */
> +	      if ((tp->thread_options () & GDB_THREAD_OPTION_EXIT) != 0)
> +		continue;
> +
>  	      /* Not found.  */
>  	      delete_thread (tp);
>  	    }
> diff --git a/gdb/target/target.c b/gdb/target/target.c
> index 453167a2ad4..2dc033e63b7 100644
> --- a/gdb/target/target.c
> +++ b/gdb/target/target.c
> @@ -196,6 +196,7 @@ to_string (gdb_thread_options options)
>  {
>    static constexpr gdb_thread_options::string_mapping mapping[] = {
>      MAP_ENUM_FLAG (GDB_THREAD_OPTION_CLONE),
> +    MAP_ENUM_FLAG (GDB_THREAD_OPTION_EXIT),
>    };
>    return options.to_string (mapping);
>  }
> diff --git a/gdb/target/target.h b/gdb/target/target.h
> index 139e371e8d8..4e8839c5667 100644
> --- a/gdb/target/target.h
> +++ b/gdb/target/target.h
> @@ -34,6 +34,10 @@ enum gdb_thread_option : unsigned
>    /* Tell the target to report TARGET_WAITKIND_THREAD_CLONED events
>       for the thread.  */
>    GDB_THREAD_OPTION_CLONE = 1 << 0,
> +
> +  /* Tell the target to report TARGET_WAITKIND_THREAD_EXIT events for
> +     the thread.  */
> +  GDB_THREAD_OPTION_EXIT = 1 << 1,
>  };
>  
>  DEF_ENUM_FLAGS_TYPE (enum gdb_thread_option, gdb_thread_options);
> -- 
> 2.36.0