Re: [PATCH RFA] Linux threads, SIGINT handling

Re: [PATCH RFA] Linux threads, SIGINT handling

[Mark Kettenis]

   Date: Fri, 25 May 2001 12:18:37 -0700
   From: Michael Snyder <msnyder@cygnus.com>

   Mark, 

   When a user types ^C(break) at a tty on Linux, SIGINT is sent to every 
   thread in the process group (whether by the kernel or by the manager 
   thread, I do not know).  The effect is that, if you type ^C once, you
   have to say "continue" N times (once for every user thread).

It's the kernel that sends all those SIGINTs, since to the kernel our
program is just a bunch of processes sharing their VM.  And indeed,
the effect is rather annoying.

   This patch handles that problem by discarding the redundant SIGINT
   events (ie. all of them except the one that was actually detected in
   lin_lwp_wait).

I don't think it's OK to unconditionally discard the SIGINT events.
The process might want to do something special with those, and may
depend on getting them in every thread.  How about changing the

   else if (WSTOPSIG (status) == SIGINT)

into

   else if (WSTOPSIG (status) == SIGINT
            && signal_pass_state (SIGINT) == 0)

?  Not passing SIGINT to the inferior is GDB's default, so this will
work fine for most people.  And this way people can say

   "handle SIGINT pass"

ignore the warning, and have SIGINT delivered to all threads.

If you agree, it's OK to check this in with that change.

Mark

Re: [PATCH RFA] Linux threads, SIGINT handling

[Michael Snyder]

Mark Kettenis wrote:
> 
>    Date: Fri, 25 May 2001 12:18:37 -0700
>    From: Michael Snyder <msnyder@cygnus.com>
> 
>    Mark,
> 
>    When a user types ^C(break) at a tty on Linux, SIGINT is sent to every
>    thread in the process group (whether by the kernel or by the manager
>    thread, I do not know).  The effect is that, if you type ^C once, you
>    have to say "continue" N times (once for every user thread).
> 
> It's the kernel that sends all those SIGINTs, since to the kernel our
> program is just a bunch of processes sharing their VM.  And indeed,
> the effect is rather annoying.
> 
>    This patch handles that problem by discarding the redundant SIGINT
>    events (ie. all of them except the one that was actually detected in
>    lin_lwp_wait).
> 
> I don't think it's OK to unconditionally discard the SIGINT events.
> The process might want to do something special with those, and may
> depend on getting them in every thread.  How about changing the
> 
>    else if (WSTOPSIG (status) == SIGINT)
> 
> into
> 
>    else if (WSTOPSIG (status) == SIGINT
>             && signal_pass_state (SIGINT) == 0)
> 
> ?  Not passing SIGINT to the inferior is GDB's default, so this will
> work fine for most people.  And this way people can say
> 
>    "handle SIGINT pass"
> 
> ignore the warning, and have SIGINT delivered to all threads.
> 
> If you agree, it's OK to check this in with that change.

Agreed and committed.

Michael