New eepro100.c v1.09r available for test

[Donald Becker]

Please try the latest version, v1.09r, available from
  http://cesdis.gsfc.nasa.gov/linux/drivers/kern-2.3/index.html
  ftp://cesdis.gsfc.nasa.gov/pub/linux/drivers/kern-2.3/

Note that you will need additional support header files.  Read the web page
or the README file.

This version
   Changes the way the previous descriptor suspend bit is cleared on the x86
   (only) from an atomic bit clear to a byte write.  This might be faster,
   especially on SMP machines, but I really made the change to verify that
   the driver isn't somehow clearing StatusComplete.  This could only happen
   if the clear_bit() was not atomic.  The current theory is that the i82558
   is failing to mark commands as complete with certain PCI bus traffic
   patterns.

   Adds a new error check for skipped commands (see last item).

   Moves the increments of sp->cur_tx so that we should no longer need a
   spinlock in the interrupt handler.  (The spinlock is still there.)

   Shrinks the regions required to be spinlocked.  The spinlocks still have
   the same span as before, but may be moved up to the commented line later.

   Changes the way Tx timeouts are detected in the timer routine.

Please try the new version if you previously experienced transmit timeouts
with v1.09l or later.

I'm especially looking for a new error message 
	printk(KERN_ERR "%s: Command unit failed to mark "
			"command %8.8x as complete at %d.\n",
			dev->name, status, dirty_tx);
which indicates that we recovered from a condition that previously required
a transmit timeout to detect.

Donald Becker					  becker@cesdis.gsfc.nasa.gov
USRA-CESDIS, Center of Excellence in Space Data and Information Sciences.
Code 930.5, Goddard Space Flight Center,  Greenbelt, MD.  20771
301-286-0882	     http://cesdis.gsfc.nasa.gov/people/becker/whoiam.html