Re: [PATCH v2] mm/zsmalloc: add statistics support

[Minchan Kim <minchan@kernel.org>]

On Fri, Dec 19, 2014 at 04:06:48PM -0800, Andrew Morton wrote:
> On Sat, 20 Dec 2014 08:58:52 +0900 Minchan Kim <minchan@kernel.org> wrote:
> 
> > On Fri, Dec 19, 2014 at 03:45:48PM -0800, Andrew Morton wrote:
> > > On Sat, 20 Dec 2014 08:39:37 +0900 Minchan Kim <minchan@kernel.org> wrote:
> > > 
> > > > Then, we should fix debugfs_create_dir can return errno to propagate the error
> > > > to end user who can know it was failed ENOMEM or EEXIST.
> > > 
> > > Impractical.  Every caller of every debugfs interface will need to be
> > > changed!
> > 
> > If you don't like changing of all of current caller, maybe, we can define
> > debugfs_create_dir_error and use it.
> > 
> > struct dentry *debugfs_create_dir_err(const char *name, struct dentry *parent, int *err)
> > and tweak debugfs_create_dir.
> > struct dentry *debugfs_create_dir(const char *name, struct dentry *parent, int *err)

typo.

I meant this

static struct dentry *__create_file(const char *name, umode_t mode,
                                    struct dentry *parent, void *data,
                                    const struct file_operations *fops,
				    int *err)

> > {
> > 	..
> > 	..
> > 	if (error) {
> > 		*err = error;
> > 		dentry = NULL;
> > 	}
> > }
> > 
> > Why not?
> 
> It involves rehashing a lengthy argument with Greg.

Okay. Then, Ganesh,
please add warn message about duplicaed name possibility althoug
it's unlikely as it is.

Thanks.

> 
> > > 
> > > It's really irritating and dumb.  What we're supposed to do is to
> > > optionally report the failure, then ignore it.  This patch appears to
> > > be OK in that respect.
> > 
> > At least, we should notify to the user why it was failed so he can fix
> > the name if it was duplicated. So if you don't want debugfs, at least
> > I want to warn all of reasons it can fail(at least, duplicated name)
> > to the user.
> 
> Sure.  The debugfs interface design is mistaken.