Re: [patch] Real-Time Preemption, -RT-2.6.9-rc4-mm1-U8

[Jens Axboe]

On Thu, Oct 21 2004, Bill Huey wrote:
> On Thu, Oct 21, 2004 at 10:14:43PM +0200, Jens Axboe wrote:
> > On Thu, Oct 21 2004, Bill Huey wrote:
> > > A lot of things are perfectly "valid" in the Linux kernel regarding
> > > stuff like that are a bit irregular. But the preemption work about
> > > to stress these things in ways that was never designed to which is
> > > why these patches are needed. Having a clear use of various locking
> > > conventions is key to getting this system to behave in a predictable
> > > manner. Quite simply, Linux was never targetted to do this and the
> > > sloppiness is showing so it's got to be removed.
> > 
> > I have to disagree, I don't think the above use is either convoluted or
> > sloppy in any way. Now that we have the completion structure, certain
> > things are surely better implemented as such. But the old use is
> > perfectly valid and logical, imho.
> 
> You use a semaphore to protect data, a completion isn't protecting data
> but preserving a certain kind of wait ordering in the code. The
> possibility of overloading the current mutex_t for PI makes for a conceptual
> mismatch when used in this case since having a kind of priority for
> completions is a bit odd. It's better to flat out use a completion
> instead, IMO.

Linux semaphores (being counted) have always been a fine fit for things
like the loop use, where you get to down it 10 times because you have 10
items pending. I know this isn't the traditional mutex and that it
doesn't protect data as such, but is was never abuse. It isn't overload.
Doing it with a traditional mutex (I'm assuming this is what mutex_t is
in Ingos tree) would be overload and a bad idea, indeed.

-- 
Jens Axboe

-
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/