Re: kernel-2.2.1-undefined references.

[Paul Derbyshire <pderbysh at usa dot net>]

At 10:12 PM 2/14/99 -0500, you wrote:
>I don't think so. How the hell else can you write C code that works with
>data that changes asynchronously, e.g. because of multiple threads or
>because of hardware or interrupts or signal handlers or...

By the way, don't say use mutexes and semaphores.
1. Hardware doesn't understand these things, and you do need software
   to interface to hardware.
2. A modal signal/exception handler would be in deep trouble if it
   encountered a mutex that was locked and decided to wait for it to
   unlock...and wait, and wait, and wait...
3. The semaphores and mutexes need to be coded as well. They need
   state variables, and those state variables need to be modified
   asynchronously, *without* using a mutex or semaphore as this would
   be a chicken and egg problem! So the mutex flag or semaphore
   counter needs to be 'volatile', so code like this:

   void mutex::get_lock (void) throw () {
     suspend_task_switching();
     while (locked) {
       resume_task_switching();
       yield();
       suspend_task_switching();
     }
     locked=true;
     resume_task_switching();
   }

   doesn't hang in an infinite loop because 'locked' was optimized
   into a register and the while loop therefore never exits since its
   register copy of 'locked' never changes! So we need

 private:
   volatile bool locked;

   or the mutex class won't work worth a damn.

     
-- 
   .*.  "Clouds are not spheres, mountains are not cones, coastlines are not
-()  <  circles, and bark is not smooth, nor does lightning travel in a
   `*'  straight line."    -------------------------------------------------
        -- B. Mandelbrot  |http://surf.to/pgd.net
_____________________ ____|________     Paul Derbyshire     pderbysh@usa.net
Programmer & Humanist|ICQ: 10423848|