On Sun, Mar 12, 2000 at 09:53:29AM -0800, Linus Torvalds wrote:
> Doing an __sti in interrupts will let higher-priority interrupts in. As it
> always has.
Previously, _any_ interrupt was let to progress. That is, hardware
priorities were ignored by Linux.
> I think it's probably stupid most of the time, and shows that something is
> wrong (an interrupt should not take long enough for it to matter that we
> do an __sti), but sometimes that kind of stupidity is a result of horrible
> hardware latencies (eg the time it takes to read a sector from the IDE
> disk using programmed IO).
So if we have a hardware determined higher priority level IDE than level
network, then it previously was the case that if the IDE driver was polite
and did an __sti, the network was assured of a relatively low latency. However,
the network now may need to wait until the programmed i/o completes.
Probably programmed i/o should rather happen in soft_interrupt space, but
with the current drivers it does not.
> Quite regardless of how you do interrupts: it doesn't matter where youput
> the ACK's, you always need to make sure that irq masking etc is correct,
> and you must NOT allow a context switch while an interrupt handler is
> still running.
Ok. Ingo: does your "low latency" patch violate this rule?
> You're making these problems up.
Yes. I thought that the hardware worked like 8259s, but it is somewhat
more sensible.
> > I really dislike hardware interrupt priorities since they have nothing
> > to do with the OS' ideas of importance and are complex to work with.
>
> I agree. I think interrupt controllers should be simple masks, nothing
> more, nothing less.
>
> However, I'm not designing the hardware.
It's amazing how stubbornly hardware folks refuse to consider OS in their
designs.
---------------------------------------------------------
Victor Yodaiken
FSMLabs: www.fsmlabs.com www.rtlinux.com
FSMLabs is a servicemark and a service of
VJY Associates L.L.C, New Mexico.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Mar 15 2000 - 21:00:22 EST