Messages in this thread | | | Date | Mon, 28 Apr 2008 22:25:52 +0200 | From | Ingo Molnar <> | Subject | Re: [patch 0/2] Immediate Values - jump patching update |
| |
* H. Peter Anvin <hpa@zytor.com> wrote:
> Mathieu Desnoyers wrote: >> >> Thanks to Peter for the review. > > Just in case someone gets the wrong idea... > > I still think this is the completely wrong approach.
hm, can it result in a broken kernel? If yes, how? Or are your objections more higher level?
i actually like that we end up with something rather NOP-ish, with some mild 'ambient' impact due to the extra constraints that state value visibility brings with it. The in-source impact of the markers is minimal, especially with Peter Zijstra's wrappers. The scheduler markers at least are also expected to stay pretty stable as well.
the syscall markers should be done less intrusively - and i think they can be done less intrusively.
but we need to get past this current impasse, the optimizations that Mathieu has done to markers are pretty impressive so far. The overhead is not zero, but it gets quite close to it and the SystemTap and kprobes people are happy with it as well.
> I'd use stronger terms, but Al Viro would sue me for copyright > infringement.
it would clearly fall under the Fair Use Doctrine and perhaps also under the Doctrine of Severe Necessities, so dont be shy!
Ingo
| |