lkml.org 
[lkml]   [2014]   [May]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch] mm, slab: suppress out of memory warning unless debug is enabled
On Wed, 7 May 2014 14:36:34 -0700 (PDT) David Rientjes <rientjes@google.com> wrote:

> On Wed, 7 May 2014, Andrew Morton wrote:
>
> > > When the slab or slub allocators cannot allocate additional slab pages, they
> > > emit diagnostic information to the kernel log such as current number of slabs,
> > > number of objects, active objects, etc. This is always coupled with a page
> > > allocation failure warning since it is controlled by !__GFP_NOWARN.
> > >
> > > Suppress this out of memory warning if the allocator is configured without debug
> > > supported. The page allocation failure warning will indicate it is a failed
> > > slab allocation, so this is only useful to diagnose allocator bugs.
> > >
> > > Since CONFIG_SLUB_DEBUG is already enabled by default for the slub allocator,
> > > there is no functional change with this patch. If debug is disabled, however,
> > > the warnings are now suppressed.
> > >
> >
> > I'm not seeing any reason for making this change.
> >
>
> You think the spam in http://marc.info/?l=linux-kernel&m=139927773010514
> is meaningful? It also looks like two different errors when in reality it
> is a single allocation.
>
> Unless you're debugging a slab issue, all the pertinent information is
> already available in the page allocation failure warning emitted by the
> page allocator: we already have the order and gfp mask. We also know it's
> a slab allocation because of the __kmalloc in the call trace.
>
> Does this user care about that there are 207 slabs on node 0 with 207
> objects? Probably only if they are diagnosing a slab problem.

I'd prefer something which can be added to the changelog to address
this omission over a series of rhetorical questions.

> > > @@ -1621,11 +1621,17 @@ __initcall(cpucache_init);
> > > static noinline void
> > > slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid)
> > > {
> > > +#if DEBUG
> > > struct kmem_cache_node *n;
> > > struct page *page;
> > > unsigned long flags;
> > > int node;
> > >
> > > + if (gfpflags & __GFP_NOWARN)
> > > + return;
> > > + if (!printk_ratelimit())
> > > + return;
> >
> > printk_ratelimit() is lame - it uses a single global state. So if
> > random net driver is using printk_ratelimit(), that driver and slab
> > will interfere with each other.
> >
>
> Agreed, but it is a testiment to the uselessness of this information
> already. The page allocation failure warnings are controlled by their own
> ratelimiter, nopage_rs, but that's local to the page allocator. Do you
> prefer that all these ratelimiters be moved to the global namespace for
> generic use?

As these messages are related then it probably makes sense for them to
use a common ratelimit_state, hopefully local to slab.c.



\
 
 \ /
  Last update: 2014-05-08 00:21    [W:0.076 / U:0.876 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site