Re: Slab corruption in 2.6.16-rc5-mm2

[Linus Torvalds]

On Thu, 9 Mar 2006, Martin J. Bligh wrote:
> > 
> > DEBUG_PAGEALLOC in particular is *fantastic* at making bugs hide.
> > I've lost many an hour trying to pin bugs down due to that.
> 
> Is this backwards? We're saying DEBUG_PAGEALLOC is bad?

DEBUG_PAGEALLOC is great for finding the really stupid kinds of bugs, and 
it's definitely worth doing every once in a while.

However, DEBUG_PAGEALLOC makes many things orders of magnitude slower, and 
it eats memory like mad (because it turns some slabs into whole pages - 
but it still doesn't help small allocation debugging that much). So unlike 
DEBUG_SLAB, it's not reasonable to have it on all the time.

IOW, DEBUG_SLAB is something that a distro kernel can reasonably enable 
for users by default (I think fedora-devel does, for example). In 
contrast, DEBUG_PAGEALLOC is more of a "useful for special cases" thing, 
where you want to validate that there's nothing _obviously_ bad going on.

> Do we NOT want to have DEBUG_SLAB and DEBUG_PAGEALLOC both enabled?

I suspect that once DEBUG_PAGEALLOC is on, whether you do DEBUG_SLAB or 
not is a toss-up. The interesting cases tend to be

 - neither: usable for benchmarking
 - DEBUG_SLAB: perfectly usable for normal work
 - DEBUG_PAGEALLOC (with or without DEBUG_SLAB): debugging tool only

At least that's my opinion, maybe others have other experiences.

			Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/