Re: [PATCH] Fastwalk: reduce cacheline bouncing of d_count (Changelog@1.1024.1.11)

[Herbert Poetzl]

On Tue, Jul 08, 2003 at 05:53:34PM +0100, Alan Cox wrote:
> On Maw, 2003-07-08 at 17:44, Herbert Poetzl wrote:
> > > Its no big problem to me since I can just back it out of -ac
> > 
> > just curious, because I use this patch since early 2.4.20,
> > are there any reasons to 'back it out of -ac' for you?
> > 
> > anyway I totally agree that the NFS issue pointed out by 
> > Trond should be addressed ...
> 
> Its high risk, its got bugs as Trond already showed and it only
> helps performance on giant SMP boxes. Its all risk and no
> reward. Quota updates get you working 32bit uid quota and
> the interactivity stuff helps all even tho its got some
> risk.

every change (if it's not a bugfix, and even those) bear
a risk, what I like about the fastwalk patch is not the
performance gain on giant SMP boxes, because I do not have
any (unfortunately ;) but the code change from ...

        if (path_init(pathname,  LOOKUP_PARENT, &nd))
                error = path_walk(pathname, &nd);

to
        error = path_lookup(pathname,  LOOKUP_PARENT, &nd);

and

        dentry = cached_lookup(nd->dentry, &this, 0);
        if (!dentry) {
                dentry = real_lookup(nd->dentry, &this, 0);
                err = PTR_ERR(dentry);
                if (IS_ERR(dentry))
                         break;
        }

to
	err = do_lookup(nd, &this, &next, &pinned, 0);

which (at least for me) is more read-/understandable ...

anyway, thanks for you answer,
Herbert