[ECOS] select() and signal's

[ECOS] select() and signal's

[Boris V. Guzhov]

Hi,

Are there any way in eCos to interrupt the select() blocking
through a sending of a any signal?

For instance:

// posix thread 1:
void *thr1(void *par)
{
  struct sigaction sa;
  sigset_t mask;

  sigemptyset( &mask );
  sigaddset( &mask, SIGRTMIN );
  pthread_sigmask( SIG_UNBLOCK, &mask, 0 );

  sigemptyset( &sa.sa_mask );
  sa.sa_flags = SA_SIGINFO;
  sa.sa_sigaction = my_sig_handler;
  sigaction( SIGTRTMIN, &sa, NULL);

  fd_set rfds;
  int retval;

  FD_ZERO(&rfds);
  FD_SET(0, &rfds);
  retval = select(1, &rfds, NULL, NULL, NULL);
  if ( retval < 0 )
    perror("select:");
 ...
}

// posix thread 2:
void *thr2(void *par)
{
...
  pthread_kill( th1, SIGRTMIN) ;
...
}


In Linux it works, but in eCos it not works.

How can I  do it in eCos?


Thanks in advance.

--
Boris Guzhov,
St.Petersburg, Russia

Re: [ECOS] select() and signal's

[Jonathan Larmour]

"Boris V. Guzhov" wrote:
> 
> Hi,
> 
> Are there any way in eCos to interrupt the select() blocking
> through a sending of a any signal?
> 
> For instance:
> 
> // posix thread 1:
> void *thr1(void *par)
> {
>   struct sigaction sa;
>   sigset_t mask;
> 
>   sigemptyset( &mask );
>   sigaddset( &mask, SIGRTMIN );
>   pthread_sigmask( SIG_UNBLOCK, &mask, 0 );
> 
>   sigemptyset( &sa.sa_mask );
>   sa.sa_flags = SA_SIGINFO;
>   sa.sa_sigaction = my_sig_handler;
>   sigaction( SIGTRTMIN, &sa, NULL);
> 
>   fd_set rfds;
>   int retval;
> 
>   FD_ZERO(&rfds);
>   FD_SET(0, &rfds);
>   retval = select(1, &rfds, NULL, NULL, NULL);
>   if ( retval < 0 )
>     perror("select:");
>  ...
> }
> 
> // posix thread 2:
> void *thr2(void *par)
> {
> ...
>   pthread_kill( th1, SIGRTMIN) ;
> ...
> }
> 
> In Linux it works, but in eCos it not works.

It is meant to work in eCos. It's meant to return -1 and set errno to
EINTR. Are you sure thr1 isn't being pre-empted for any other reason, e.g.
if thr2 is higher priority.

Try setting a breakpoint on the function cyg_sigqueue and debugging it.

jifl
-- 
Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine

Re: [ECOS] select() and signal's

[Boris V. Guzhov]

But it still not works.
Both threads have an equal priority.
I carefully has looked on cyg_sigqueue() code.
And I don't understand as cyg_sigqueue() function will wake
the blocked  in select() thread.
The cyg_sigqueue() function  calls only  signal_sigwait.broadcast().
But the blocked thread sleeps on other condition variable - selwait,
but not on signal_sigwait.

Thanks in advance.
--
Boris Guzhov,
St.Petersburg, Russia


> > In Linux it works, but in eCos it not works.
>
> It is meant to work in eCos. It's meant to return -1 and set errno to
> EINTR. Are you sure thr1 isn't being pre-empted for any other reason, e.g.
> if thr2 is higher priority.
>
> Try setting a breakpoint on the function cyg_sigqueue and debugging it.
>
> jifl
> --
> Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
> Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine
>

Re: [ECOS] select() and signal's

[Jonathan Larmour]

"Boris V. Guzhov" wrote:
> 
> But it still not works.
> Both threads have an equal priority.
> I carefully has looked on cyg_sigqueue() code.
> And I don't understand as cyg_sigqueue() function will wake
> the blocked  in select() thread.
> The cyg_sigqueue() function  calls only  signal_sigwait.broadcast().
> But the blocked thread sleeps on other condition variable - selwait,
> but not on signal_sigwait.

Try the attached patch and let me know how you get on. I've only done some
simple tests. But I want you to try it with your existing code rather than
me spending more time on it[1]. *Do* let me know the outcome.

Jifl

[1] Since you're not a customer :)
-- 
Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine
? foo
Index: signal.cxx
===================================================================
RCS file: /home/cvs/ecc/ecc/compat/posix/current/src/signal.cxx,v
retrieving revision 1.11
diff -u -5 -p -r1.11 signal.cxx
--- signal.cxx	2001/02/14 01:25:44	1.11
+++ signal.cxx	2001/04/20 20:08:54
@@ -61,12 +61,13 @@
 #include <signal.h>                     // our header
 #include <setjmp.h>
 #include <unistd.h>                     // _exit
 
 #include <cyg/kernel/clock.hxx>
-
+#include <cyg/kernel/thread.hxx>
 #include <cyg/kernel/clock.inl>
+#include <cyg/kernel/thread.inl>
 
 // -------------------------------------------------------------------------
 // Internal definitions
 
 // Handle entry to a signal package function. 
@@ -239,16 +240,33 @@ cyg_bool cyg_sigqueue( const struct sige
         ss->pending = si;
     }
     // else A non-queuable signal, just set it pending
 
     if( thread != NULL )
+    {
         sigaddset( &thread->sigpending, signo );
-    else sigaddset( &sig_pending, signo );
+        // just wake the thread up now if it's blocked somewhere
+        if ((thread->sigpending & ~thread->sigmask) != 0)
+        {
+            thread->thread->set_asr_pending();
+            thread->thread->release();
+        }
+    }
+    else
+    {
+        sigaddset( &sig_pending, signo );
+        // Wake up any threads in sigsuspend() and sigwait().
+        if (!signal_sigwait.get_queue()->empty())
+        {
+            signal_sigwait.broadcast();
+        } 
+        else
+        {
+            cyg_posix_pthread_release_thread( &sig_pending );
+        }
+    }
 
-    // Wake up any threads in sigsuspend() and sigwait().
-    signal_sigwait.broadcast();
-    
     if( !locked ) signal_mutex.unlock();
     
     return true;
 }
 

Re: [ECOS] select() and signal's

[Boris V. Guzhov]

I tried this patch. And it works. But there are some questions.
Now pthread_kill() interrupts not only select() function
but also sem_wait().
As far as I know, semaphores are a general synchronization mechanism and
they can be used with a signal-cathing function (in particular, semaphores
can be posted from a POSIX signal-catching function).  For instance, in
Linux pthread_kill() don't interrupts  sem_wait().

> Try the attached patch and let me know how you get on. I've only done some
> simple tests. But I want you to try it with your existing code rather than
> me spending more time on it[1]. *Do* let me know the outcome.
>
> Jifl
>

Re: [ECOS] select() and signal's

[Boris V. Guzhov]

I tried and it works. But there are some questions.
Now pthread_kill() interrupts not only select(), but
also sem_wait().
As far as I know, the POSIX semaphores are a general and
exclusive synchronization mechanism with a signal-cathing
function. And their functions should not be interrupted by
signals.
In Linux inplementation pthread_kill() not interrupts
the sem_wait().

That's right?

--
Boris Guzhov,
St.Petersburg, Russia



> "Boris V. Guzhov" wrote:
> >
> > But it still not works.
> > Both threads have an equal priority.
> > I carefully has looked on cyg_sigqueue() code.
> > And I don't understand as cyg_sigqueue() function will wake
> > the blocked  in select() thread.
> > The cyg_sigqueue() function  calls only  signal_sigwait.broadcast().
> > But the blocked thread sleeps on other condition variable - selwait,
> > but not on signal_sigwait.
>
> Try the attached patch and let me know how you get on. I've only done some
> simple tests. But I want you to try it with your existing code rather than
> me spending more time on it[1]. *Do* let me know the outcome.
>
> Jifl
>
> [1] Since you're not a customer :)
> --
> Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
> Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine


----------------------------------------------------------------------------
----


> ? foo
> Index: signal.cxx
> ===================================================================
> RCS file: /home/cvs/ecc/ecc/compat/posix/current/src/signal.cxx,v
> retrieving revision 1.11
> diff -u -5 -p -r1.11 signal.cxx
> --- signal.cxx 2001/02/14 01:25:44 1.11
> +++ signal.cxx 2001/04/20 20:08:54
> @@ -61,12 +61,13 @@
>  #include <signal.h>                     // our header
>  #include <setjmp.h>
>  #include <unistd.h>                     // _exit
>
>  #include <cyg/kernel/clock.hxx>
> -
> +#include <cyg/kernel/thread.hxx>
>  #include <cyg/kernel/clock.inl>
> +#include <cyg/kernel/thread.inl>
>
>

// -------------------------------------------------------------------------
>  // Internal definitions
>
>  // Handle entry to a signal package function.
> @@ -239,16 +240,33 @@ cyg_bool cyg_sigqueue( const struct sige
>          ss->pending = si;
>      }
>      // else A non-queuable signal, just set it pending
>
>      if( thread != NULL )
> +    {
>          sigaddset( &thread->sigpending, signo );
> -    else sigaddset( &sig_pending, signo );
> +        // just wake the thread up now if it's blocked somewhere
> +        if ((thread->sigpending & ~thread->sigmask) != 0)
> +        {
> +            thread->thread->set_asr_pending();
> +            thread->thread->release();
> +        }
> +    }
> +    else
> +    {
> +        sigaddset( &sig_pending, signo );
> +        // Wake up any threads in sigsuspend() and sigwait().
> +        if (!signal_sigwait.get_queue()->empty())
> +        {
> +            signal_sigwait.broadcast();
> +        }
> +        else
> +        {
> +            cyg_posix_pthread_release_thread( &sig_pending );
> +        }
> +    }
>
> -    // Wake up any threads in sigsuspend() and sigwait().
> -    signal_sigwait.broadcast();
> -
>      if( !locked ) signal_mutex.unlock();
>
>      return true;
>  }
>
>

Re: [ECOS] select() and signal's

[Jonathan Larmour]

"Boris V. Guzhov" wrote:
> 
> I tried and it works. But there are some questions.
> Now pthread_kill() interrupts not only select(), but
> also sem_wait().
> As far as I know, the POSIX semaphores are a general and
> exclusive synchronization mechanism with a signal-cathing
> function. And their functions should not be interrupted by
> signals.
> In Linux inplementation pthread_kill() not interrupts
> the sem_wait().
> 
> That's right?

My POSIX spec says explicitly "The sem_wait() function shall be
interruptible by the delivery of a signal." So if signals on Linux never
interrupt it, it sounds like Linux is wrong(!)

On doing so, sem_wait should return -1 and set errno to EINTR.

Jifl
-- 
Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine