Re: Fw: Questions about VDSO

[Benjamin Herrenschmidt <benh at au1 dot ibm dot com>]

On Thu, 2004-11-04 at 20:02 -0800, Roland McGrath wrote:

> You are the one who raised the issue, so I leave it to you to say what
> issue you are actually talking about.
> 
> > Ok, that's just ignorance on my side
> 
> Yes.  If you want to hack the dynamic linker, you should start by
> understand its most basic implementation plan.

At this point, I don't want to hack it, I want to understand what is the
best way to go, of course, if it end up requiring hacking the dynamic
linker, I'll start digging or working with somebody like Steve or you
who is familiar enough with it to hack it ;) Let's be constructive a bit
here please.

Note also that it's most basic implementation plan is rather difficult
to understand even by reading the source unless you have been around for
a long time, unless there is some nice piece of documentation that I've
missed ? I'm sorry to say that the source code isn't self explanatory.

So rather than talking about issues, let's talk about what I want to do,
and I would appreciate if you could tell me what you think is the best
way to go... Is anybody here interested in making things happening or
turning people down is really the local national sport as it seem to be
so far ?

In the meantime, I'm preparing an updated ppc64 vDSO patch that adds the
signal trampoline part so we can get that part done at least. 

I'll add one function too (gettimeofday userland) so we can use it as a
playground to see the best way of "hooking" those functions to userland.
There are various debates so far: should we have ld.so be able to link
the aplication directly to the vdso for things like that, or should
glibc always be in the middle, that sort of thing.

Ben.