Re: [RFC PATCH 2/3] kmemleak: Add callbacks to the bootmem allocator

[Catalin Marinas <catalin.marinas@arm.com>]

On Tue, 2009-07-07 at 18:53 +0200, Johannes Weiner wrote:
> On Tue, Jul 07, 2009 at 10:08:50AM +0300, Pekka Enberg wrote:
> > On Mon, 2009-07-06 at 11:51 +0100, Catalin Marinas wrote:
> > > @@ -597,7 +601,9 @@ restart:
> > >  void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
> > >  					unsigned long goal)
> > >  {
> > > -	return ___alloc_bootmem_nopanic(size, align, goal, 0);
> > > +	void *ptr =  ___alloc_bootmem_nopanic(size, align, goal, 0);
> > > +	kmemleak_alloc(ptr, size, 1, GFP_KERNEL);
> > > +	return ptr;
> 
> You may get an object from kzalloc() here, I don't think you want to
> track that (again), right?

You are write, I missed the alloc_arch_preferred_bootmem() function
which may call kzalloc().

> Pekka already worked out all the central places to catch 'slab already
> available' allocations, they can probably help you place the hooks.

It seems that alloc_bootmem_core() is central to all the bootmem
allocations. Is it OK to place the kmemleak_alloc hook only in this
function?

diff --git a/mm/bootmem.c b/mm/bootmem.c
index 5a649a0..74cbb34 100644
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -520,6 +520,7 @@ find_block:
 		region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
 				start_off);
 		memset(region, 0, size);
+		kmemleak_alloc(region, size, 1, 0);
 		return region;
 	}
 

> > > +	kmemleak_alloc(ptr, size, 1, GFP_KERNEL);
> 
> These GFP_KERNEL startled me.  We know for sure that this code runs in
> earlylog mode only and gfp is unused, right?  Can you perhaps just
> pass 0 for gfp instead?

Yes, indeed.

Thanks for your comments.

-- 
Catalin

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>