[anton@samba.org: [PATCH]: Bug in ppc32 ld.so]

[Geoff Keating]

> Date: Tue, 16 Jul 2002 11:02:05 -0700
> From: Anton Blanchard <anton@samba.org>

> > No, I don't agree, you haven't given any reason why ld.so should do this.
> 
> If the page is not marked guarded the cpu can instruction prefetch into the
> page. Can you guarantee that current and future CPUs will never prefetch
> into the icache clean area?

Can you come up with an example of how the CPU might do this?
Remember, this code is not used for ld.so's PLT itself.  At the time
this routine is running, there should be no references to the
newly-created PLT in the instruction sequence from that point, nor
would sequential fetching reach the code.

> When I mentioned the current behaviour one of the ppc architects warned me
> we were asking for trouble.

Perhaps you didn't fully explain the situation?

-- 
- Geoffrey Keating <geoffk@geoffk.org> <geoffk@redhat.com>