From: Hanna Linder <hannal@us.ibm.com>
To: Paul Menage <pmenage@ensim.com>
Cc: Hanna Linder <hannal@us.ibm.com>,
viro@math.psu.edu, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] 2.5.6-pre3 Fast Walk Dcache
Date: Fri, 08 Mar 2002 17:30:55 -0800 [thread overview]
Message-ID: <30770000.1015637455@w-hlinder.des> (raw)
In-Reply-To: <E16jRuZ-00076W-00@pmenage-dt.ensim.com>
In-Reply-To: <E16jRuZ-00076W-00@pmenage-dt.ensim.com>
--On Friday, March 08, 2002 13:28:23 -0800 Paul Menage <pmenage@ensim.com> wrote:
>
> 1) You're missing parentheses in cached_lookup_nd() and path_lookup():
>
Oops. Operator precedence strikes again. They are in
cached_lookup_nd() and link_path_walk(). I have made
the changes.
> 2) Since cached_lookup_nd() calls __d_lookup() and hence
> __dget_locked(), it's not clear how you actually avoid incrementing the
> d_count values of the dentries, other than the root/cwd dentries. Can
> you explain the logic in a little more detail?
The frequency with which root/cwd dentries are in any given path
probably cause the majority of the cacheline bouncing of d_count.
So the logic there is clear. However, you make a good point and
I'm looking at this.
> 3) If you replace walk_init_root() and path_lookup() with something
> like the following, you can pull the ugliness of walk_init_root() out
> of path_lookup(). Basically, make walk_init_root() recognise
> LOOKUP_LOCKED and take the dcache_lock rather than grabbing refcounts.
> walk_init_root() drops the LOOKUP_LOCKED flag if necessary while
> calling __emul_lookup_dentry() to avoid additional complexity. If
> walk_init_root() returns 0, then the dcache lock wasn't taken,
> regardless of whether the nd.flags had LOOKUP_LOCKED set.
>
> static inline int
> walk_init_root(const char *name, struct nameidata *nd)
> {
> unsigned int flags = nd->flags;
> read_lock(¤t->fs->lock);
> if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
>
> if(flags & LOOKUP_LOCKED)
> nd->flags &= ~LOOKUP_LOCKED;
>
> nd->mnt = mntget(current->fs->altrootmnt);
> nd->dentry = dget(current->fs->altroot);
The first reaction I have is that it breaks the consistancy
between a flag and what the flag represents. Having the dcache_lock
held without the LOOKUP_LOCKED flag in this part of the code might
cause deadlocks or lead to hard-to-maintain code.
I appreciate you taking the time to provide such thoughtful and
deatailed comments and I will look at the whole patch again with
these comments in mind. Next week expect a new and improved
version!
Thanks.
Hanna
next prev parent reply other threads:[~2002-03-09 1:29 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <0C01A29FBAE24448A792F5C68F5EA47D2260CB@nasdaq.ms.ensim.com>
2002-03-08 21:28 ` Paul Menage
2002-03-09 1:30 ` Hanna Linder [this message]
2002-03-09 2:03 ` [PATCH] 2.5.6 Fast Walk Dcache (improved) Paul Menage
2002-03-09 2:59 ` Paul Menage
2002-03-10 19:51 ` Hanna V Linder
2002-03-11 7:13 ` Paul Menage
2002-03-12 7:16 ` Paul Menage
2002-03-08 0:25 [PATCH] 2.5.6-pre3 Fast Walk Dcache Hanna Linder
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=30770000.1015637455@w-hlinder.des \
--to=hannal@us.ibm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=pmenage@ensim.com \
--cc=viro@math.psu.edu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®