Re: AW: [ros-kernel]Fw: Boot problem with CVS changes from friday/saturday.

[<we...@cw...>]

[A bit more explanation.]

Quoting we...@cw...:
> Yes, I agree. The problem is that the irq is unmasked then the
> interrupt flag is set then we go on to do a lot more processing. 
>
The same interrupt can then occur again before we have executed the handler
eventually overflowing the stack for a frequently occuring event.
 
> 1. Set per-interrupt flag.
> 2. If per-interrupt flag was already set, increment per-interrupt
> counter and
> return.
> 3. Enable interrupts.
> 4. Call device handler.
> 5. Check the per-interrupt counter, if greater than zero then goto to
> step 4.
> 6. Clear the per-interrupt flag.
> 7. Execute DPCs, APCs and thread dispatching.
> 
This is safe since: a) no more one (and a bit) instances of each interrupt
handler can be active at a time; b) DPCs and thread dispatches is done at
DISPATCH_LEVEL, APCs at APC_LEVEL so these can never be executed. 

The per-interrupt flag can be reduced to just a check that the irql level
is currently less than the priority of the interrupt and the interrupt
handler epilog can also check for queued instances of lower priority
interrupts.