Messages in this thread | | | Date | Tue, 31 Jan 2012 15:30:23 +0100 | From | Ingo Molnar <> | Subject | Re: x86: clean up smpboot.c's use of udelay+schedule |
| |
* Arjan van de Ven <arjan@infradead.org> wrote:
> On Tue, 31 Jan 2012 13:43:41 +0100 > Ingo Molnar <mingo@elte.hu> wrote: > > > + usleep_range(100, 200); > > > > I'm wondering whether we could shorten this delay to say 10 > > usecs and thus save 0.1 msecs (or more) from a typical SMP > > bootup? > > doesn't matter really; [...]
It matters somewhat, especially if PeterZ's suggestion is used, which is far more clean as well. The magic delays are not really justified anymore.
> [...] bringing up a cpu is several orders more expensive (> > 100msec in 3.2, in 3.3 this got optimized to maybe 30 msec) > 0.1 msec is the least of anyone's worries at this point ;-)
It's 3% of the 30 msecs overhead.
> ( would be nice if this was a completion, but this is rather > fragile code in general... at least not making it spin is an > incremental improvement )
Completions arent hard to use and the scheduler should be up and running at this stage already.
Thanks,
Ingo
| |