Re: [RFC] pgflags_t

[Matthew Wilcox <willy@infradead.org>]

On Wed, Oct 06, 2021 at 05:16:51PM +0200, David Hildenbrand wrote:
> On 06.10.21 16:58, Matthew Wilcox wrote:
> > David expressed some unease about the lack of typesafety in patches
> > 1 & 2 of the page->slab conversion [1], and I'll admit to not being
> 
> I yet have to reply to your mail (-EBUSY), I'd even enjoy doing all
> pgflags_t access via the struct page, for example, using simpler "reserved"
> placeholders for such fields in "struct slab" that we really want to move
> out of struct slab later. When accessing flags, we'd cast to "struct page"
> and do the access.

There are several bits that slab/slob/slub use in page->flags currently.
The obvious one is the node ID (whether stored directly or indirected
via section), but also PG_pfmemalloc (aka PG_active), PG_slob_free (aka
PG_private), PG_lock, PG_slab and PG_head,  We might be able to change
that too, but for the moment, how about we not explore that path in the
interest of getting something merged that won't preclude exploring that
path sometime in the future?

> ... but whatever approach we use on that fron, this patch is a step into the
> right direction IMHO.

Thank you!