Re: [PATCH 1/2] kmemleak: Delete an error message for a failed memory allocation in two functions

[Dan Carpenter]

On Mon, Aug 14, 2017 at 03:38:04PM +0100, Catalin Marinas wrote:
> On Mon, Aug 14, 2017 at 04:02:21PM +0300, Dan Carpenter wrote:
> > On Mon, Aug 14, 2017 at 12:14:32PM +0100, Catalin Marinas wrote:
> > > On Mon, Aug 14, 2017 at 11:35:02AM +0200, SF Markus Elfring wrote:
> > > > From: Markus Elfring <elfring@xxxxxxxxxxxxxxxxxxxxx>
> > > > Date: Mon, 14 Aug 2017 10:50:22 +0200
> > > > 
> > > > Omit an extra message for a memory allocation failure in these functions.
> > > > 
> > > > This issue was detected by using the Coccinelle software.
> > > > 
> > > > Signed-off-by: Markus Elfring <elfring@xxxxxxxxxxxxxxxxxxxxx>
> > > > ---
> > > >  mm/kmemleak.c | 5 +----
> > > >  1 file changed, 1 insertion(+), 4 deletions(-)
> > > > 
> > > > diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> > > > index 7780cd83a495..c6c798d90b2e 100644
> > > > --- a/mm/kmemleak.c
> > > > +++ b/mm/kmemleak.c
> > > > @@ -555,7 +555,6 @@ static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
> > > >  
> > > >  	object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
> > > >  	if (!object) {
> > > > -		pr_warn("Cannot allocate a kmemleak_object structure\n");
> > > >  		kmemleak_disable();
> > > 
> > > I don't really get what this patch is trying to achieve. Given that
> > > kmemleak will be disabled after this, I'd rather know why it happened.
> > 
> > kmem_cache_alloc() will generate a stack trace and a bunch of more
> > useful information if it fails.  The allocation isn't likely to fail,
> > but if it does you will know.  The extra message is just wasting RAM.
> 
> Currently kmemleak uses __GFP_NOWARN for its own metadata allocation, so
> we wouldn't see the sl*b warnings. I don't fully remember why I went for
> this gfp flag, probably not to interfere with other messages printed by
> the allocator (kmemleak_alloc is called from within sl*b).
> 
> I'm fine to drop __GFP_NOWARN and remove those extra messages.
> 

Ah.  I considered that, but didn't see a NOWARN in the diff and was too
lazy to check the code.  Probably I would just leave it as-is.


regards,
dan carpenter