Re: [RFC] disable_irq()/enable_irq() semantics and ide-probe.c

[Linus Torvalds]

On Thu, 9 Oct 2003 viro@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx wrote:
> 
> The underlined bit is absent on alpha version of the same function.
> 
> Note that this piece is bogus - if .action is NULL, we are already caught
> by IRQ_INPROGRESS check.  So it's not exactly a bug, but considering
> your arguments about exact same check slightly earlier in handle_irq()...

Yes. I'm definitely not claiming the code is beautiful. 

I think it happens to be working ;)

> It's from cset1.437.22.19 by mingo; the same changeset had done unconditional
> removal of IRQ_INPROGRESS, so there it made sense.  After the irq.c part
> had been reverted (1.497.61.30 from you), i8259.c one should be killed
> too, AFAICS...

Yeah, the IRQ_INPROGRESS removal in handle_irq() was buggy: it caused the
bit to be spuriously cleared if an interrupt happened while the previous
interrupt was active (which will _not_ happen in the i8259 case, but does
happen in the edge-triggered case).

The problem, I think, is that all this code grew fairly organically, and 
nobody ever sat down and wrote down the rules. 

Which is why I think it _works_, but it's clearly nonoptimal and sometimes 
confusing.

I suspect the 2.4.x situation is even worse. 

		Linus

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/