2.4.7 softirq incorrectness.

[Rusty Russell <rusty@rustcorp.com.au>]

This current code is bogus.  Consider:
	spin_lock_irqsave(flags);	
	cpu_raise_softirq(this_cpu, NET_RX_SOFTIRQ);
	spin_unlock_irqrestore(flags);

Oops... softirq not run until the next interrupt.  So, EITHER:

1) make local_irq_restore check for pending softirqs in as we do for
   local_bh_enable(), and get rid of the wakeup_softirqd() in
   cpu_raise_softirq().  ie. all "exits" from in_interrupt == true are
   symmetrical.

*OR*

2) Change the check in cpu_raise_softirq to:
	if (!in_hw_irq_handler(cpu))
		wakeup_softirqd(cpu);

   and implement in_hw_irq_handler() on all platforms.  Then get rid of
   the test in local_bh_enable().

Please pick one approach or the other, and stick with it!  The current
code does half of each, badly. 8(

Thanks,
Rusty.
--
Premature optmztion is rt of all evl. --DK