Re: [PATCH] Abstracted Priority Inheritance for RT

[Esben Nielsen <simlo@phys.au.dk>]

On Thu, 2 Jun 2005, Daniel Walker wrote:

> On Thu, 2005-06-02 at 17:18 +0200, Esben Nielsen wrote:
> 
> > Good :-)
> > I asked the question because I considered (and started but didn't
> > have time) doing what you have done. I wanted to generalise the rt_mutex
> > to have real rw_lock as well - which was dropped due to the
> > non-deterministc behavioir even with PI. To do that I needed to have the
> > recursion and the callback..
> 
> I'm not planning to do a real rw-lock, but I hope this generic PI will
> help with that. 

I came to the conclusion that an rw-lock would make all writers
non-deterministic. I had a small discussion on this list with Ingo about
it. He had also come to the same conclusion.

> I'm still not completely satisfied with this callback
> structure , but I don't see a better way to do it. Do you have an
> suggestions for replacing it?
>
No, I got stock there too.

But right now the following ideas spring to my mind:
If it is to solve the problem of having a callback wrap every use
in macroes and use the TYPE_EQUAL() to expclicit call the right function.
Only if the explicit type is unknown in the macro use the callback. That
should optimize stuff a little bit.. Just a wild idea.

If it is explicitly for PI you can do a thing like
 waiter->get_next_waiter();
to resolve the chain of waiters. Then you can have the PI algotithm work
iteratively without knowing the explicit kind of lock involved.


> Daniel
> 

Esben

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