hCO;mbI`gYhHQV5^CT1A^BL@[>6iTbKebAoF9`lV:   
   From: "Lars Erdmann"    
      
      
   "James J. Weinkam"  schrieb im Newsbeitrag    
   news:jd6ftn$oni$1@speranza.aioe.org...   
   > Steven Levine wrote:   
   >>   
   >> What is true is that within the sorted statements, certain sets of   
   >> statements must be in a specific order.  Typically this is to satisfy   
   >> hardware requirements such as Lar's describes for the EHCI driver.   
   >> Sometimes, it is to satisfy timing requirements or work around   
   >> implementation errors.  For example, it is recommonend that helperw be   
   >> the first RUN statement.  This gives it a better chance to finishing   
   >> its initialization before the network applications are started.   
   >>   
   > Unfortunately, we mortals who are not privy to the information and    
   > resources needed to fix the problem are forced to resort to workarounds    
   > such as you describe.   
   >   
   > What astonishes me is that in an operating system developed largely by    
   > IBM, with all their corporate experience, starting in the late eighties,    
   > such matters were left to chance. In a properly designed operating system    
   > nothing is left to chance. Process synchronization primitives have been    
   > well known since the sixties. I think the word that best describes such a    
   > design is incompetence. Sad.   
      
   Hm, the "synchronization issues" between the HC drivers is nothing that IBM    
   could have foreseen years ago.   
   The reason why you don't see this problem with for example Linux is because    
   those guys kind of built this sync stuff right into   
   their USB stack (or into the kernel, I didn't fully read their source code).   
   We have to live with the fact that there will be no new updates to the    
   kernel. And reimplementing the complete USB stack is much too much work.   
   I am currently thinking of adding synchronization semaphores so that a    
   "companion host controller" waits until its associated   
   main host controller is done with initialization before it finishes with its    
   initialization.   
   But that's not that easy to implement and as always with new code, might add    
   new bugs which I will have to find and fix ...   
      
      
   Lars   
      
      
   --- Internet Rex 2.31   
    * Origin: Arcor (1:261/20.999)