On 11/22/21 12:40, Christoph Lameter wrote:

On Mon, 22 Nov 2021, Vlastimil Babka wrote:

quoted

It's no problem to have the real beginning of the object aligned, and the
prepended header not. The code already does that before this patch for the
kmalloc power-of-two alignments, where e.g. the object can be aligned to 256
bytes, but the prepended header to a smaller ARCH_KMALLOC_MINALIGN /
ARCH_SLAB_MINALIGN.

Ok but then the first object in a page may still have those issues.

Hmm, that's right. I guess we should also distinguish ARCH_KMALLOC_MINALIGN
for kmalloc paths, and ARCH_SLAB_MINALIGN for kmem_cache_alloc paths. Seems
the latter is generally smaller, thus some holes left by kmalloc allocations
could be filled later by kmem_cache_alloc allocations.

quoted

quoted

I dont know but it seems that making slob that sophisticated is counter
productive. Remove SLOB?

I wouldn't mind, but somebody might :)

Well run a space efficiency analysis after this patch. If the memory used
is larger than SLUB (with the configuration for minimal data footprint)
then there is no reason for SLOB to continue.

Makes sense.