lkml.org 
[lkml]   [2008]   [Dec]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH -v3 6/8] fsnotify: add group priorities
From
Date
On Thu, 2008-11-27 at 17:25 +0100, Peter Zijlstra wrote:
> On Tue, 2008-11-25 at 12:21 -0500, Eric Paris wrote:
> > In preperation for blocking fsnotify calls group priorities must be added.
> > When multiple groups request the same event type the lowest priority group
> > will receive the notification first.
>
> > @@ -114,9 +117,26 @@ struct fsnotify_group *fsnotify_find_group(unsigned int group_num, unsigned long
> >
> > group->ops = ops;
> >
> > - /* add it */
> > - list_add_rcu(&group->group_list, &fsnotify_groups);
> > + /* Do we need to be the first entry? */
> > + if (list_empty(&fsnotify_groups)) {
> > + list_add_rcu(&group->group_list, &fsnotify_groups);
> > + goto out;
> > + }
> > +
> > + list_for_each_entry(group_iter, &fsnotify_groups, group_list) {
> > + /* insert in front of this one? */
> > + if (priority < group_iter->priority) {
> > + /* I used list_add_tail() to insert in front of group_iter... */
> > + list_add_tail_rcu(&group->group_list, &group_iter->group_list);
> > + break;
> > + }
> >
> > + /* are we at the end? if so insert at end */
> > + if (list_is_last(&group_iter->group_list, &fsnotify_groups)) {
> > + list_add_tail_rcu(&group->group_list, &fsnotify_groups);
> > + break;
> > + }
> > + }
> > out:
> > mutex_unlock(&fsnotify_grp_mutex);
> > fsnotify_recalc_global_mask();
>
> What priority range do you need to cater for, and how many groups?

On a typical system I'd expect to see one group for dnotify (rpmidmapd
uses dnotify so most systems will end up having 1 group I would expect)

inotify I wouldn't expect more than 3-4 inotify_init() calls

fsnotify I wouldn't imagine more than 3 groups.

So total we are talking about maybe 10 groups on a system really making
use of fs notification?

> I can
> imagine for many groups and limit range a priority list might be better
> suited.

talking about plist.h? Since I don't allow 2 groups with the same
priority I'd say a lot of the plist code would just be overhead (the
prio list and the node list would be the same)

That's not a big deal since I don't really care about the add/remove
code paths since they are all notification overhead/setup/teardown. I
would think that cleaner simpler code would probably be a better idea
rather than performance for these areas especially since it looks like
the speed critical parts of plists (list_for_each_entry) would be the
exact same.

what I don't see is plists being protected by RCU and looking at
plist_del it doesn't seem like it would be rcu safe. RCU safe plists
might be a good idea, but for now I think I should just do my own
priority listing so I don't have to hold a lock while I walk the group
list (that path is VERY hot)

-Eric



\
 
 \ /
  Last update: 2008-12-01 16:25    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site