lkml.org 
[lkml]   [2011]   [Nov]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch 1/5]thp: improve the error code path
On Thu, Nov 10, 2011 at 10:33:15AM +0800, Shaohua Li wrote:
> On Thu, 2011-11-10 at 10:18 +0800, Andrea Arcangeli wrote:
> > Hi Shaohua,
> >
> > On Mon, Nov 07, 2011 at 01:17:29PM +0800, Shaohua Li wrote:
> > > On Wed, 2011-10-26 at 09:48 +0800, Shaohua Li wrote:
> > > > On Tue, 2011-10-25 at 19:44 +0800, Andrea Arcangeli wrote:
> > > > > Hello,
> > > > >
> > > > > On Tue, Oct 25, 2011 at 10:58:41AM +0800, Shaohua Li wrote:
> > > > > > +#ifdef CONFIG_SYSFS
> > > > > > + sysfs_remove_group(hugepage_kobj, &khugepaged_attr_group);
> > > > > > +remove_hp_group:
> > > > > > + sysfs_remove_group(hugepage_kobj, &hugepage_attr_group);
> > > > > > +delete_obj:
> > > > > > + kobject_put(hugepage_kobj);
> > > > > > out:
> > > > > > +#endif
> > > > >
> > > > > Adding an ifdef is making the code worse, the whole point of having
> > > > > these functions become noops at build time is to avoid having to add
> > > > > ifdefs in the callers.
> > > > yes, but hugepage_attr_group is defined in CONFIG_SYSFS. And the
> > > > functions are inline functions. They really should be a '#define xxx'.
> >
> > hugepage_attr_group is defined even if CONFIG_SYSFS is not set and I
> > just made a build with CONFIG_SYSFS=n and it builds just fine without
> > any change.
>
> > $ grep CONFIG_SYSFS .config
> > # CONFIG_SYSFS is not set
> >
> > So we can drop 1/5 above.
> this isn't the case in the code. And the code uses hugepage_attr_group
> is already within CONFIG_SYSFS, so your build success.

I thought it was related to a SYSFS=n build failure, so when I
verified it build just fine I didn't really see the point of messing
with sysfs adding more #ifdefs but I see you want to avoid leaking
those two entries if the allocation doesn't succeed.

If the later allocations don't succeed we can panic() and be done with
it. Adding more code will just grow the zImage a bit, but I can
appreciate the more theoretical correctness so I'm ok with it now that
I see the point of it.


\
 
 \ /
  Last update: 2011-11-10 04:01    [from the cache]
©2003-2011 Jasper Spaans