Linus Torvalds wrote:

On Tue, 20 Mar 2007, Zachary Amsden wrote:
  

quoted

void local_irq_restore(int enabled)
{
   pda.intr_mask = enabled;
   /*
    * note there is a window here where softirqs are not processed by
    * the interrupt handler, but that is not a problem, since it will
    * get done here in the outer enable of any nested pair.
    */
   if (enabled)
       local_bh_enable();
}
    

Actually, this one is more complicated. You also need to actually enable 
hardware interrupts again if they got disabled by an interrupt actually 
occurring while the "soft-interrupt" was disabled.
  

Actually, I was thinking the irq handlers would just not mess around 
with eflags on the stack, just call the chip to ack the interrupt and 
re-enable hardware interrupts when they left, since that is free anyway 
with the iret.  Maybe leaving irqs disabled is better.

Anyway, it really *should* be pretty damn simple. No need to disable 
preemption, there should be no events that can *cause* it, since all 
interrupts get headed off at the pass.. (the return-from-interrupt thng 
should already notice that it's returning to an interrupts-disabled 
section and not try to do any preemption).
  
What did I miss?
  

I wasn't disabling preemption to actually disable preemption.  I was 
just using bh_disable as a global hammer to stop softirqs (thus the irq 
replay tasklet) from running during the normal irq_exit path.  Then, we 
can just use the existing software IRQ replay code, and I think barely 
any new code (queue_irq(), etc) has to be written.

Zach