2011/10/3 KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

> On Sun, 2 Oct 2011 21:57:07 +0800
> Wei Yang <weiyang.kernel@gmail.com> wrote:
>
> > Dear experts,
> >
> > I am viewing the source code of __release_region() in kernel/resource.c.
> > And I have one comment for the performance issue.
> >
> > For example, we have a resource tree like this.
> > 10-89
> >    20-79
> >        30-49
> >        55-59
> >        60-64
> >        65-69
> >    80-89
> > 100-279
> >
> > If the caller wants to release a region of [50,59], the original code
> will
> > execute four times in the for loop in the subtree of 20-79.
> >
> > After changing the code below, it will execute two times instead.
> >
> > By using the "git annotate", I see this code is committed by Linus as the
> > initial version. So don't get more information about why this code is
> > written
> > in this way.
> >
> > Maybe the case I thought will not happen in the real world?
> >
> > Your comment is warmly welcome. :)
> >
> > diff --git a/kernel/resource.c b/kernel/resource.c
> > index 8461aea..81525b4 100644
> > --- a/kernel/resource.c
> > +++ b/kernel/resource.c
> > @@ -931,7 +931,7 @@ void __release_region(struct resource *parent,
> > resource_size_t start,
> >        for (;;) {
> >                struct resource *res = *p;
> >
> > -               if (!res)
> > +               if (!res || res->start > start)
>
> Hmm ?
>        res->start > end ?
>
>  I think res->start > start is fine.
__release_region will release the exact the region, no overlap.
So if res->start > start, this means there is no exact region to release.
The required to release region doesn't exist.


> Thanks,
> -Kame
>
>


-- 
Wei Yang
Help You, Help Me