Re: [patch 02/11] x86 architecture implementation of Hardware Breakpoint interfaces

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Wed, 2009-03-11 at 13:12 +0100, Ingo Molnar wrote:
> 
> #3 is probably the most informative (and hence probably the
>    best) variant. It also leaves policy of how to resolve the 
>    conflict to the admin.

Agreed.
> 
> Would be nice to have it simple. Reluctance regarding this 
> patchset is mostly rooted in that diffstat above.
> 
> The changes it does in the x86 architecture code are nice 
> generalizations and cleanups. Both the scheduler, task 
> startup/exit and ptrace bits look pretty sane in terms of 
> factoring out debug register details. But the breakpoint 
> management looks very complex

I agree there is some interest in generalization and cleanup, especially
as far as userspace APIs go, though it's a hard nut to crack as every
CPU family out there has some subtle differences in the way breakpoints
or watchpoints work (for example, alignment constraints, ability to do
ranges, the way they handle kernel vs. user, etc...)

I'm not yet sold.

Cheers,
Ben.