Linus Torvalds wrote:

On Tue, 20 Mar 2007, Zachary Amsden wrote:
  

quoted

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.
    

No can do. Think level-triggered. You *need* to disable the interrupt, and 
disabling it at the CPU is the easiest approach. Even so, you need to 
worry about SMP and screaming interrupts at all CPU's, but if you don't 
ack it to the IO-APIC until later, that should be ok (alternatively, you 
need to just mask-and-ack the irq controller).
  

Well, you can keep it masked, but a more important point is that I've 
entirely neglected local interrupts.  This might work for IRQs, but for 
local timer or thermal or IPIs, using the tasklet based replay simply 
will not work.

One of the advantages of doing that is that you only ever have a queue of 
one single entry, which then makes it easier to do the replay.
  

Yes.  Unfortunately now both do_IRQ and all the smp_foo interrupt 
handlers need to detect and queue for replay, but fortunately they all 
have the interrupt number conveniently on the stack.

Zach