Messages in this thread |  | | Date | Sat, 28 Jun 2003 20:07:06 -0700 | From | "Martin J. Bligh" <> | Subject | Re: 2.5.73-mm2 |
| |
--William Lee Irwin III <wli@holomorphy.com> wrote (on Saturday, June 28, 2003 19:18:09 -0700):
> On Sat, Jun 28, 2003 at 05:34:05PM -0700, Martin J. Bligh wrote: >> Last time I measured it, it had about a 10% overhead in kernel time. >> Seems like a good thing to keep as an option to me. Bill said he >> had some other code to alleviate the overhead, but I don't think >> it's merged ... I'd rather see UKVA (permanently map the pagetables >> on a per-process basis) merged before it becomes "not an option" - >> that gets rid of all the kmapping. > > There are several orthogonal things going on here. One is dropping the > hooks in the right places to get various concrete tasks done. Another > is general resource scalability vs. raw overhead tradeoffs. The last > one is gathering a wide enough repertoire of core hooks that arches can > use "advanced" techniques like recursive pagetables when they require > various kinds of intervention by the kernel to use. > > This is just another set of hooks we'll need for our end goal, with a > fully functional implementation. It has direct applications and is > completely usable now for resource scalability albeit with some > overhead. Things are all headed in the appropriate directions; the > hooks do not conflict with and do not require any core modifications > whatsoever in order to use in combination with recursive pagetables; > they can simply recover information from already-available places and > transparently replace the highpmd and highpte arch code. > > I can work directly with Dave to arrange a proper demonstration of this > (i.e. fully functional implementation) if need be. I've largely avoided > interceding in recursive pagetable mechanics in order not to duplicate > work.
Right, I'm not against what you're doing - I'm totally for it. My only concern was that whilst it has some overhead, it should stay as a config option (which you did). That lets people make the call of overhead vs resource scaling.
Your patch is fine - just the talk of removing the config option scared me ;-)
M.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
|  |