< Previous by Date Date Index Next by Date >
< Previous in Thread Thread Index Next in Thread >

Re: [reSIProcate] unsafe static Mutex object initialization


I was referring to the ThreadIf stuff as safe since its indirectly only initialized during global static initialization. It would unsafe during regular runtime as you mentioned.

Aron Rosenberg
Sr. Director, Engineering,
LifeSize, a division of Logitech




On Sat, Jul 30, 2011 at 11:15 AM, Byron Campen <bcampen@xxxxxxxxxxxx> wrote:
Actually, this too can be problematic, because you can get multiple initializations of the Mutex if multiple threads try to construct the same type of object at once. And, if you have any static functions that try to use the mutex, you're sunk because it might not be initialized. A better way to do this is the following:

In the header:

class FoobaJooba
{
static Mutex& getMutex()
{
static Mutex mMutex;
return mMutex;
}

// Nobody should actually touch this, or use it to try to determine
// whether the mutex has been initialized; this bool could be anything
// during static initialization. It exists solely to cause getMutex()
// to be called sometime during static init.
static bool ensureInitialized;
};

In the implementation file:

bool FoobaJooba::ensureInitialized=(&initMutex()!=0);

This ensures that getMutex() is called sometime during static initialization (it might be called multiple times, but this is ok, since there is only one thread of execution), so multiple init problems don't crop up, and it also ensures that if somebody tries to access the mutex during static init _before_ ensureInitialized has been initialized, the mutex is just initialized a little bit early.

Now, this is still not perfect; it doesn't prevent nastiness during static _de_initilization.

Best regards,
Byron Campen


Best regards,
Byron Campen

The best way to handle this is actually to make the mutex be a static pointer which gets allocated during a class constructor. See ThreadIf class and mTlsDestructorsMutex and friends for the best way to implement this.

-Aron

Aron Rosenberg
LifeSize, a division of Logitech


On Thu, Jul 21, 2011 at 11:25 AM, Daniel Pocock <daniel@xxxxxxxxxxxxxxxxxxxxx> wrote:



I notice a few places where a Mutex objected is constructed, typically
as a static member of a class:

$ grep Mutex rutil/*.hxx | grep static
Log.hxx:      static Mutex _mutex;
Random.hxx:      static Mutex mMutex;
Time.hxx:                  static Mutex mWrapCounterMutex;
Time.hxx:                  static Mutex mMutex;

The order in which the constructors of these objects are called is
non-deterministic

It is actually possible that an code section depending on the Mutex
could be attempting to lock on the Mutex before the Mutex itself is
initialized



For UNIX, it may be possible to initialize Mutex::mId =
PTHREAD_MUTEX_INITIALIZER, or even define Mutex like so:

#define STATIC_MUTEX_INIT PTHREAD_MUTEX_INITIALIZER
typedef pthread_mutex_t Mutex;

and then in some file.cxx:

Mutex mMutex = STATIC_MUTEX_INIT;



However, the situation for Windows is not quite the same:

http://locklessinc.com/articles/pthreads_on_windows/

suggests a hack:

#define PTHREAD_MUTEX_INITIALIZER {(void*)-1,-1,0,0,0,0}

that depends on the internal structure of the Windows type CRITICAL_SECTION

Has anyone else got any thoughts on this, how it should be done on
Windows, and any other platform considerations that I haven't raised?



_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel

_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel