[PATCH v3 07/14] slub: Remove page->counters

[Vlastimil Babka <vbabka@suse.cz>]

On 04/18/2018 08:49 PM, Matthew Wilcox wrote:
> From: Matthew Wilcox <mawilcox@microsoft.com>
> 
> Use page->private instead, now that these two fields are in the same
> location.  Include a compile-time assert that the fields don't get out
> of sync.
> 
> Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>

Why not retain a small union of "counters" and inuse/objects/frozens
within the SLUB's sub-structure? IMHO it would be more obvious and
reduce churn?

...

> @@ -358,17 +359,10 @@ static __always_inline void slab_unlock(struct page *page)
>  
>  static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
>  {
> -	struct page tmp;
> -	tmp.counters = counters_new;
> -	/*
> -	 * page->counters can cover frozen/inuse/objects as well
> -	 * as page->_refcount.  If we assign to ->counters directly
> -	 * we run the risk of losing updates to page->_refcount, so
> -	 * be careful and only assign to the fields we need.
> -	 */
> -	page->frozen  = tmp.frozen;
> -	page->inuse   = tmp.inuse;
> -	page->objects = tmp.objects;

BTW was this ever safe to begin with? IIRC bitfields are frowned upon as
a potential RMW. Or is there still at least guarantee the RMW happens
only within the 32bit struct and not the whole 64bit word, which used to
include also _refcount?