Re: WARNING: at mm/page_alloc.c:1159 get_page_from_freelist+0x325/0x655()

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> On Wed, 17 Jun 2009, Ingo Molnar wrote:
> > 
> > a new warning started popping up today, in the new page allocator 
> > code. The allocation came from kmemleak:
> 
> We should probably print out the order.
> 
> Right now it warns about any order but 0, and I think that's 
> likely bogus. It's fine to allow small orders (I'd suggest 0-2), 
> since we should always be able to get those, and small kmalloc's 
> generally do want more than one page just to avoid crazy 
> fragmentation issues.
> 
> See, for example, the whole 'slab_break_gfp_order' logic in 
> mm/slab.c: it very much expects to be able to use order-1 
> allocations for kmalloc() if there is enough memory (where 
> "enough" is actually just 32MB). And slub seems to put some limit 
> at PAGE_ALLOC_COSTLY_ORDER (3).
> 
> So apart from anything else (ie this particular case is possibly 
> fixable in kmemleak), I do think that we should likely allow at 
> least order-1 and possible order-2 allocations with __GFP_NOFAIL 
> too.

I saw about half a dozen of different warning patterns during the 
day, so the warning definitely feels a bit over-eager.

	Ingo