Critics on L4 (NOT Re: shared memory Beispiel?)

[Adam Lackorzynski]

Hi,

first, please use a proper quoting style. Your style is very unusual and
irritating.


On Wed Aug 31, 2005 at 21:00:16 +0800, yuan Darwin-r62832 wrote:
> >When those features are going to be implemented?
> >
> Fiasco implements IO Flexpages. Neither Redirectors nor Clans and Chiefs 
> are currently implemented in Fiasco and probably never will. I cannot 
> speak for plans regarding Pistachio.
> 
> [Darwin Yuan] I didn't read the source code of Fiasco. However, from
> the documents I got from its website, I found that Clans & Chiefs was
> implemented in its kernel ( It might be removed now). But I don't

C&C were never implemented, and will never be, in any kernel.

> think it's a good design that this policy was involved in L4 kernel,
> now that L4 claims it's a real microkernel. However, I think those
> kind of policies could be implemented outside kernel. Therefore, I
> think Redirector is a good mechanism to achieve those kind of
> policies. That's why I think redirector is very important for
> implementing a flexible system on L4.

So what we have now are inflexible systems?

> >Now that those features are very important for implementing a flexible 
> >OS on L4 u-kernel.
> >  
> >
> Both redirectors and IO flexpages are not essential to get a system up 
> and running. Their purpose is to control and restrict communication 
> between threads respectively to prevent a thread from reading / writing 
> IO-ports. Feel free to implement those features if you need them.
> 
> [Darwin Yuan] Yes, That's true. However, a real system should makes
> peripheral equipment work. IO Fpage is very essential for implementing
> hardware driver. 

Not having IO protection doesn't mean you cannot use drivers! Not having
IO protection means that everyone can whack your system because everyone
runs with IO privileges, that's it. And you need to do this because you
cannot have drivers otherwise. We lived with that for some time but now
we have proper IO protection so everything in fine in this regard.

> >I think L4Ka::Pistachio is more clean & flexible, why Fiasco put so 
> >many complicated & inflexible policies inside L4 kernel?
> >  
> >
> Which policies are you referring to? Can you elaborate on these points?
> 
> [Darwin Yuan] For example, Clans & Chiefs.   Also, I think "Task" is a

That's not the fault of Fiasco, C&C is in the V2 spec. These are
different things. And as already mentioned C&C where never implemented
as some clever guys noticed before that there are some problems.

> redundant abstract/concept, "Address Space" is a much better one. By

What's the difference between Task and AS?

> "Thread" & "Address Space", L4Ka::Pistachio makes the abstract very
> refined, and allows to construct very flexible system. All together,
> the APIs of X.2 is much better than X.0.

X.2 is the successor of V2/X0 so this is no real surprise. But in which
way does the API decide how one's building flexible or inflexible
systems?

> Why don't you like the design of L4Linux or do you mean Fiasco?
> 
> [Darwin Yuan] Actually, I don't like the APIs & mechanism of Fiasco.
> Now that L4Linux is based on those APIs & mechanism, I don't like this
> version of L4Linux.

Cool. So what would I need to do to make it please you?




Adam
-- 
Adam                 adam at os.inf.tu-dresden.de
  Lackorzynski         http://os.inf.tu-dresden.de/~adam/