Re: [PATCH] mm/slub: Make __ksize() faster

[Matthew Wilcox <willy@infradead.org>]

On Fri, Mar 08, 2024 at 12:16:09AM -0500, Kent Overstreet wrote:
> On Fri, Mar 08, 2024 at 04:48:11AM +0000, Matthew Wilcox wrote:
> > On Thu, Mar 07, 2024 at 10:13:22PM -0500, Kent Overstreet wrote:
> > > This steals it to make __ksize() faster; it's now a single dependent
> > > load, instead of two. This is going to be important for tracking the
> > > amount of memory stranded by RCU, which we want to be able to do if
> > > we're going to be freeing all pagecache folios (and perhaps all folios)
> > > via RCU.
> > 
> > Do you have any measurements?  I don't necessarily object, it's just
> > that I would have thought we were already bringing slab->slab_cache
> > into cache as part of the freeing operation.
> 
> for kfree() yes, but for kfree_rcu() no - that'll happen significantly
> later.
> 
> I don't have measurements that it /matters/, but since we have the u32
> there now this seems like a good use for it.

There are potentiually better uses for those bits.  We could turn
folio_test_slab() into a PageType test, freeing up a page flag.