RE: why mprotect(PROT_SEM) ?

> Well, my explanation is already outdated :(
> ...
> ...
> PROT_SEM was planned for doing this - but you can just pass 
> the value into
> mprotect(), that's all. It's not doing anything.
> 
> The futex code itself takes care for pinning the page - this goal is
> reached with get_user_pages().

> From David Howell
> Cc: Dr. Uwe Girlich; pthreads-devel
> Subject: RE: [pthreads-devel] why mprotect(PROT_SEM) ?

> That's good news then, we can remove the mprotect syscall 
> from the NGPT 
> runtime library if the kernel futex code is doing it. Still 
> might be an
> advantage in allocating the user-mode futex structs in the same page, 
> should yield less pinned pages.

Well, I am not that sure, we'd need Rusty to confirm that. AFAIK, this could
be right, as the vcache mechanism should take care of it, but I am not still
that sure we can just wipe it out [for 2.4 kernels]. 

This is what I got from Rusty - to me it clarifies it enough [I am ">",
Rusty is not indexed]:

------------------------------
> The PROT_SEM flag needs to be applied to any shared area - how we define
> shared here? [this is my last hope for simpliciation of the problem,
though
> I am pesimistic]:
> 
> - shared between _different_ processes with DIFFERENT VMAs

This one.  If it's in the same address space, you *don't* need it.

> If it were the first one, it would mean, for me, that between clones of
the
> same process I don't need to set PROT_SEM and that would simplify the
> internal spinlocks a _huge_ lot.

Yes.

> However, I don't see a reason why that
> should happen, normal processes sharing a page are the same that two
clones
> sharing all ... basically, I loose again.

Same address space: you're fine.  It's executing at a different
address, and maybe has slightly different kernel datastructures, but
it is (to futexes) the same program.
---------------------------

So, PROT_SEM needed for 2.4, no PROT_SEM needed for 2.5? that's the
question. 

Inaky Perez-Gonzalez -- Not speaking for Intel - opinions are my own [or my
fault]