Sutter’s Mill

Posted in Concurrency | 2 Comments

2 Responses

1. on 2008-01-14 at 2:59 am (no name)

   Hi Mr Herb,

       I have read the article in dr dobbs. I would like to give one another problem if we call a method with lock from a critical section. Suppose we have the following code.

   Processor::ProcessAcquiredData()

   {

     while(m_bisdatareadprocessingstoped)

     {

       Lock();

       int* p = buf.read();

       //do processing with p

       Unlock();

     }

   }

   int* Buff::read()

   {

       WaitforSingleObject(DataEvent);

       return GetdataPointer();

   }

   Processor::StopProcessing()

   {

       m_bisdatareadprocessingstoped = false;

       WaitforSingleObject(m_handletoprocessacquiredthread);

       // Do necessary state change

   }

   Now suppose if i want to stop the thread one becase user want an immediate process stoping.

   If we write this the program will get locked up. The locking up will occur if the data prcessing is stopped and if the waiting is inside Buff::read(). In such a case it becomes impossible to give a grace exit to thread. I think this is also an addition not to bring out any kind of wait inside another wait.

   
   
2. on 2009-07-02 at 12:52 pm Daniel Roberts

   Hello,
   In the section “Avoidance: Noncritical Calls”, option (a) says “move the call out of the critical section if you didn’t need the exclusion anyway.” Is this a reliable option? Afterall, the compiler can reorder code INTO a critical section (just not out of a critical section) as outlined in your previous article, “Use Critical Sections (Preferably Locks) to Eliminate Races.”

   Thanks for your time.
   Cheers

   
   

Comments are closed.