Re: [REPOST PATCH 3/4] slab: introduce byte sized index for thefreelist of a slab

From: Christoph Lameter
Date: Tue Sep 10 2013 - 17:25:14 EST


On Tue, 10 Sep 2013, Joonsoo Kim wrote:

> On Mon, Sep 09, 2013 at 02:44:03PM +0000, Christoph Lameter wrote:
> > On Mon, 9 Sep 2013, Joonsoo Kim wrote:
> >
> > > 32 byte is not minimum object size, minimum *kmalloc* object size
> > > in default configuration. There are some slabs that their object size is
> > > less than 32 byte. If we have a 8 byte sized kmem_cache, it has 512 objects
> > > in 4K page.
> >
> > As far as I can recall only SLUB supports 8 byte objects. SLABs mininum
> > has always been 32 bytes.
>
> No.
> There are many slabs that their object size are less than 32 byte.
> And I can also create a 8 byte sized slab in my kernel with SLAB.

Well the minimum size for the kmalloc array is 32 bytes. These are custom
slabs. KMALLOC_SHIFT_LOW is set to 5 in include/linux/slab.h.

Ok so there are some slabs like that. Hmmm.. We have sizes 16 and 24 in
your list. 16*256 is still 4096. So this would still work fine if we would
forbid a size of 8 or increase that by default to 16.

> > On x86 f.e. it would add useless branching. The branches are never taken.
> > You only need these if you do bad things to the system like requiring
> > large contiguous allocs.
>
> As I said before, since there is a possibility that some runtime loaded modules
> use a 8 byte sized slab, we can't determine index size in compile time. Otherwise
> we should always use short int sized index and I think that it is worse than
> adding a branch.

We can enforce a mininum slab size and an order limit so that it fits. And
then there would be no additional branching.


--
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/