Wait Functions & Sleep(...)

I believe that WaitForSingleObject and WaitForMultipleObjects are what you are looking for. There are other synchronization functions here: Synchronization Functions

I don't get it. Why would you want to drop down into assembly when you can simply use "Critical Sections"?
You can't effectively implement the mechanisms for dealing with thread waiting or scheduling etc. Only the operating system can do that. All the necessary stuff is already done by the OS for you to use. Many cases such as when a thread causes a page fault, already automatically give time to other threads.
If you don't want to reinvent stuff, then why are you trying to duplicate parts of the OS? Don't duplicate these, simply encapsulate them.

Just use Sleep when appropriate, and never mind the drama-queens that say otherwise.

Basically there are two ways to wait for doing mutual exclusion.

1)
First one is busy waiting (using spinlocks), where a flag is polled via test&set/cmpxchg or something similar. This is used for very short critical sections, especially for multiprocessor systems. Here it is very unlikely that two threads want to aquire the thread at the same time. Care must be taken that the thread that aquired the lock is scheduled away while inside the lock, since then the waiting threads will eat up cpu time.
A possible implementation might look like this:

volatile bool locked = false;...// wait while locked, try to perform atomic test and set// testandset() will set the flag to true and return the previous valuewhile (testandset(&locked)){  // busy waiting on smp machine or  // yield to other thread on single processor machine  ...}...// critical section code...// free critical sectionlocked = false;  // assignment should be atomic

To perform an atomic test&set operation, machine instructions like xchg, bts, cmpxchg or similar may be used. On some systems (i.e. most microcontrollers) such an operation may not be available. In this case the operation can be made atomic by locking interrupts. For x86 systems this method works fine.

One thing that will be problematic are multiprocessor systems with shared memory. Here the same memory location may be modified by more than one processor at the same time, resulting in undefined behavior if bts is used for instance. In this case locking the memory bus becomes necessary. The xchg instruction does this automatically. The result is that the code above will block memory accesses by other processors while doing busy waiting for the flag. This can prevented by a slight modification. Instead of doing test&set all the time, the loop waits until the section was free, then tries to atomically aquire the lock:

volatile bool locked = false;...while (1){  // wait until section was free  if (locked)  {    // for single processor system yield to next thread    ...    // was locked, try again    continue;  }  // is free now, try to perform atomical test&set, lock the bus when doing this  if (testandset(&locked) == false)  {    // got the lock, break the loop    break;  }}...// critical section code...// free critical sectionlocked = false;  // assignment should be atomic

As you can see, things are not that easy. Windows provides functions for this that will do the right thing depending on which hardware your application runs on. On the other hand a custom solution may provide a faster implementation, especially when you can inline this assembler opcode. Since busy waiting is used for critical sections where chances are high (i.e. 99%) for the section to be free when trying to lock it.


2)
The second solution involves putting a thread to sleep while waiting for something to happen. This is used when waiting for user input, data read from a disk or similar. WaitForSingleObject() and the other functions are right for this task. Since putting a thread to sleep and waiting it up again needs some time when compared to spinlocks, it may slow things down when done very often or it may result in a high latency. For very short critical sections this would be the wrong choice.

@John Bolton: Yes, I was previously aware of those, however, if you peruse my post again, and perhaps take a moment or two to digest the information given, as well as the question asked, it would be awesome, as your answers, in general, are good ones [smile] I cannot use those wait functions, because they have to wait on a handle, and with a home-rolled local lock, or whatever, a handle is one of the things I do not have.

@iMalc: I am not against the use of critical sections. I don't believe I ever said that [grin] They have a place, and I assure, you they will be used where I feel they should be, but I don't actually see what that has to do with my question? But thanks anyway for the point of view, more of those are always welcome!

@nmi: Thanks man, that's a nice informative post there.. Albeit, not entirely new information to me, seeing as I did some of the foot slogging on this one before posting, but nice nonetheless. I'll have to look for some more info regarding atomic actions and multiprocessor systems...


Seems the general concensus is, don't reinvent, encapsulate [grin] I guess I'm always on the side that isn't the public opinion...