lkml.org 
[lkml]   [2011]   [Oct]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH RFC V2 3/5] jump_label: if a key has already been initialized, don't nop it out
On Mon, Oct 10, 2011 at 12:58:19PM -0700, Jeremy Fitzhardinge wrote:
> > Hi,
> >
> > I just realized that the early call to jump_label_inc(), isn't being
> > honored with this patch until later when we invoke jump_label_init().
> > That strikes me as being inconsistent. When jump_label_inc() returns we
> > should expect the branch to be updated.
>
> Why is that? It looks to me like it will unconditionally update the
> instruction, irrespective of whether _init() has been called?
>

No. jump_label_init(), sets up key->entries, to point into the jump
table...before that jump_label_update(), doesn't know where the table is
located, and will just return, without doing the update.


> > Thus, I think what probably want is to add a new 'int jump_label_init'
> > flag. If its not set we can call 'jump_label_init()' from
> > jump_label_inc()/dec().
>
> Hm. I worry that it may end up calling jump_label_init() in an
> unexpected context, especially since it may well be config-dependent, or
> adding a jump_label_inc() later on starts mysteriously failing.

good point.

>
> > And jump_label_init() can avoid initialization
> > if its already set.
>
> That doesn't seem worthwhile in itself. I suspect the number of "early"
> jump_label_incs will be very small (or we should look at doing the init
> earlier).
>
> J

I have it as 'early_initcall()', but perhaps it should be moved into
init/main.c. I don't think there's any reason it can't be done super
early. So I think this might be the best answer. It will also simplify
your series.

Thanks,

-Jason


\
 
 \ /
  Last update: 2011-10-10 22:13    [from the cache]
©2003-2011 Jasper Spaans