Question about threads and signals

Question about threads and signals

[Jamie Lokier]

Dear POSIX thread experts (i.e. Ingo and Ulrich),

I have a teensy question.  Using the new Linux 2.5 threads + glibc/NPTL:

1. If a signal is delivered to a thread, is it masked for the duration of
   the handler in (a) just that thread or (b) all threads?

   In other words, if I have 3 threads and SIGIO is not blocked in any
   of them, is it possible for my SIGIO handler to be called up to 3
   times concurrently?  Or is the blocked mask somehow shared?

   Is the same thing true of SIGCHLD?  SIGSEGV?

2. Is this true of POSIX threads in general, or just Linux?

Thanks,
-- Jamie

Re: Question about threads and signals

[Ulrich Drepper]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Jamie Lokier wrote:

> 1. If a signal is delivered to a thread, is it masked for the duration of
>    the handler in (a) just that thread or (b) all threads?

(a)

>    In other words, if I have 3 threads and SIGIO is not blocked in any
>    of them, is it possible for my SIGIO handler to be called up to 3
>    times concurrently?  Or is the blocked mask somehow shared?

Masks are never shared.


>    Is the same thing true of SIGCHLD?  SIGSEGV?

Yes.  Up to the point where a fatal signal isn't caught and the process
is killed.  At that point all threads except the one responsible for the
termination is stopped and then terminated.


> 2. Is this true of POSIX threads in general, or just Linux?

Well, the above is what POSIX requires and what I think we've
implemented.  These requirements are essential for programs which do
much of their work in signal handlers.  Creating more threads which
mainly just sit around but can react to signals is a valid programming
model.

- -- 
- --------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)

iD8DBQE+KM8m2ijCOnn/RHQRAlcEAJ0W27Ju6gq4xhT7A0PGr2IJCGfp0ACfa2NY
2mmvlgXLC2Xm8UYnU+rD6cE=
=7eDw
-----END PGP SIGNATURE-----

Re: Question about threads and signals

[Jamie Lokier]

Ulrich Drepper wrote:
> > 2. Is this true of POSIX threads in general, or just Linux?
> 
> Well, the above is what POSIX requires and what I think we've
> implemented.  These requirements are essential for programs which do
> much of their work in signal handlers.

So, in a program which uses pthread_create(), a SIGCHLD handler must
either (a) be reentrant, or (b) be unblocked (or installed) in one
thread only.

> Creating more threads which mainly just sit around but can react to
> signals is a valid programming model.

Indeed, as Ingo pointed out, some signals are like poor man's threads,
and that programming model makes that explicit.  I like that, because
it means the signal handler can validly do everything normal code can
do, including use mutexes and other locking primitives which may sleep.

(Some other signals are intrinsically thread-local though, such as
SIGSEGV and SIGFPE - they are synchronous exception handlers rather
than asynchronous event handlers).

Thanks for your quick answer,
-- Jamie