Subj : Re: Recursive mutexes
To   : comp.programming.threads
From : doug
Date : Sun Mar 27 2005 12:22 pm


"Chris Thomasson" <_no_damn_spam_cristom@_no_damn_comcast.net_spam> wrote in 
message news:HMadndFiKdmz0t7fRVn-2g@comcast.com...
>> Not in my system. Why you assume I'm using windows? :-)
>
> DOH!
>
> I instantly thought windows because you used EnterCriticalSection. I 
> failed to see the call to ExitCriticalSection. Windows has 
> EnterCriticalSection/LeaveCriticalSection API's. Thats what confuused me.
>
>
>
>
>>(In fact, AFAIK, windows does support recursive mutexes).
>
> Yes it does.
>
>
>> No, in my case they're implemented with a semaphore, so calling it twice 
>> in a row
>> in the same thread would deadlock. This is an embedded system, and the 
>> synchronization support
>> isn't very good. It's quite minimal. This is why I need to do it by 
>> myself.
>
> Ok, you can do it like this:
>
>
> void my_mutex::acquire() {
> if (m_owner == thread_id) {
>    ++m_count;
>    return;
> }
This may not be relevant (depending on your OS/hardware) - but the above 
will not work on a multi-CPU system.  There is no memory barrier op before 
the read of m_owner, so a cpu could read an out of date value from its 
cache.
Bug would manifest itself as a thread recursively acquiring mutex (i.e. has 
not called ExitCriticalSection() - the barrier which 'writes' the value of 
m_owner), but on different cpu, seeing an old value of m_owner, skipping the 
increment of m_count and (according to your description of the behaviour) 
hanging on EnterCriticalSection.

>
> EnterCriticalSection();
> m_owner = thread_id;
> }
>
>
>
>
> void my_mutex::release() {
> if (m_count) {
>    --m_count;
>    if ( m_count ) { return; }
> }
>
> m_owner = 0;
> ExitCriticalSection();
> }
>
>
>
>

.