2.6.18-rc4 arch/i386/kernel/irq.c::do_IRQ has this code:
if (!irq_desc[irq].handle_irq) {
__do_IRQ(irq, regs);
goto out_exit;
}
#ifdef CONFIG_4KSTACKS
curctx = (union irq_ctx *) current_thread_info();
irqctx = hardirq_ctx[smp_processor_id()];
...
All registered irqs on my system have a NULL handle_irq so do_IRQ
always calls __do_IRQ on the current stack, it never switches to the
hardirq_ctx stack. Removing the test is trivial and it boots and runs
without the test for !irq_desc[irq].handle_irq, but why was the test
put there in the first place?
Note: with that test removed, do_IRQ runs on the hardirq stack, but
struct thread_info is not updated accordingly so backtrace is wrong. I
can fix that as well, but I need to understand the test for handle_irq
first.