malloc mutex and exceptions

malloc mutex and exceptions

[Ignasi Villagrasa]

Hi everyone,

I have been migrating to gcc a server C/C++ based software and I' like to 
control SIGSEGV signal exceptions. So I use sigaction and the corresponding 
handlers to accomplish this target. 

The problem arises when a SIGSEGV happens inside a new operator. I have many 
users connected to the program and the clients suddenly hang.  That's because 
of pthread mutex the malloc and free routines use. The SIGSEGV occurs after 
pthread_mutex_lock call and before pthread_mutex_unlock. 
So, every time a user tries to use this semaphore the program hangs. New and 
delete uses malloc and free with reentrant features. And I'm interested in 
that because of pthread program use, but I have this problem for the 
exception control.

When I have a problem using one of my mutex I can control lock and unlock 
properly, but new and delete mutex relies on library variables.

Is there any way I can control this internal mutex ?

Is there any compiler option to change this behaviour ?

Thanks in advance,

Ignasi Villagrasa.

RE: malloc mutex and exceptions

[Dave Korn]

> -----Original Message-----
> From: gcc-owner On Behalf Of Ignasi Villagrasa
> Sent: 21 March 2004 19:31

> Hi everyone,
> 
> I have been migrating to gcc a server C/C++ based software 
> and I' like to control SIGSEGV signal exceptions. So I use 
> sigaction and the corresponding handlers to accomplish this target. 
> 
> The problem arises when a SIGSEGV happens inside a new 
> operator. 

  Fix the bug in your program that trashes the heap.  SEGV inside the malloc
handler is basically not a recoverable situation.

> I have many users connected to the program and the 
> clients suddenly hang.  That's because of pthread mutex the 
> malloc and free routines use. The SIGSEGV occurs after 
> pthread_mutex_lock call and before pthread_mutex_unlock. 

  Fix the bug in your program that trashes the heap.

> So, every time a user tries to use this semaphore the program 
> hangs. 

  Fix the bug in your program that trashes the heap.

> New and delete uses malloc and free with reentrant 
> features. And I'm interested in that because of pthread 
> program use, but I have this problem for the exception control.
> 
> When I have a problem using one of my mutex I can control 
> lock and unlock properly, but new and delete mutex relies on 
> library variables.
> 
> Is there any way I can control this internal mutex ?

  By not trashing the heap, the new and delete handlers will complete
successfully instead of crashing and get a chance to release the mutex.

> Is there any compiler option to change this behaviour ?

  Unfortunately there's no "-fcorrect-stray-pointer-bugs".  You'll just have
to fix your bug.



    cheers, 
      DaveK
-- 
Can't think of a witty .sigline today....