Re: Q: explicit alignment control for the slab allocator

[Mark Hemment]

On Thu, 1 Mar 2001, Manfred Spraul wrote:

> Alan added a CONFIG options for FORCED_DEBUG slab debugging, but there
> is one minor problem with FORCED_DEBUG: FORCED_DEBUG disables
> HW_CACHEALIGN, and several drivers assume that HW_CACHEALIGN implies a
> certain alignment (iirc usb/uhci.c assumes 16-byte alignment)
> 
> I've attached a patch that fixes the explicit alignment control in
> kmem_cache_create().
> 
> The parameter 'offset' [the minimum offset to be used for cache
> coloring] actually is the guaranteed alignment, except that the
> implementation was broken. I've fixed the implementation and renamed
> 'offset' to 'align'.

  As the original author of the slab allocator, I can assure you there is
nothing guaranteed about "offset".  Neither is it to do with any minimum.

  The original idea behind offset was for objects with a "hot" area
greater than a single L1 cache line.  By using offset correctly (and to my
knowledge it has never been used anywhere in the Linux kernel), a SLAB
cache creator (caller of kmem_cache_create()) could ask the SLAB for more
than one colour (space/L1 cache lines) offset between objects.

  As no one uses the feature it could well be broken, but is that a reason
to change its meaning?

Mark