Re: priority inversion

Jeff Merkey (jmerkey@timpanogas.com)
Fri, 30 Jul 1999 09:42:30 -0600

Disagree. You can get into some pretty ugly starvation situations on a
system with inversion. The "deadlocks" aren't really deadlocks at all, but
cases where a process gets it's priority knocked down for so long everyone
backs up behind it if it is holding mutexes. I've seen some pretty busted
crap on older versions of Solaris (circa 1991) that used inversion models
before they adopted inheritance. I agree that in most real world situations
these problems don't show up, but there's always someone out there putting
out apps that uses locks excessively. Folks might be splitting hairs here,
but historically, this is the case folks make against inversion. It's an
SMP problem almost exclusively and is tied to how priorities are managed in
the kernel.

Jeff

----- Original Message -----
From: Marc Lehmann <pcg@goof.com>
To: <linux-kernel@vger.rutgers.edu>
Sent: Thursday, July 29, 1999 6:55 PM
Subject: Re: priority inversion

> On Thu, Jul 29, 1999 at 11:50:55AM -0600, Jeff Merkey
<jmerkey@timpanogas.com> wrote:
> > Priority Inversion is **BAD BUSINESS**. Someone whould fix whatever
this
> > person is complaining about. I agree that priority inheritance is slow
and
> > makes for **FAT** sync object code, but it's either this or throw
priorities
> > out of the window in the kernel proper since we will see **LOTS** of
> > deadlocks and busted applications if an inversion model is what we end
up
> > with.
>
> Linux has had this RT support for _quite_ some time. Other unices had as
> well. It would be nice to have priority inheritance, but its by no means
> as crucial as you describe it ;)
>
> --
> -----==- |
> ----==-- _ |
> ---==---(_)__ __ ____ __ Marc Lehmann +--
> --==---/ / _ \/ // /\ \/ / pcg@goof.com |e|
> -=====/_/_//_/\_,_/ /_/\_\ XX11-RIPE --+
> The choice of a GNU generation |
> |
>
> -
> 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/
>

-
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/