On Wed, Jun 28, 2023 at 09:41:18AM +0200, David Hildenbrand wrote:

I'm not a friend of these "overlays"; it all only really makes sense to me
once we actually allocate the descriptors dynamically. Maybe some of the
existing/ongoing conversions were different (that's why I was asking for the
difference, as you said the "struct slab" thing was well received).

If they are primarily only unnecessary churn for now (and unclear when/how
it will become useful), I share your opinion.

One of the reasons for doing these conversions "early" is that it helps
people who work on this code know what fields they can actually use in
their memory descriptor.  We have a _lot_ of historical baggage with
people just using random bits in struct page for their own purposes
without necessarily considering the effects on the rest of the system.

By creating specific types for each user of struct page, we can see
what's actually going on.  Before the ptdesc conversion started, I could
not have told you which bits in struct page were used by the s390 code.
I knew they were playing some fun games with the refcount (it's even
documented in the s390 code!) but I didn't know they were using ...
whetever it is; page->private to point to the kvm private data?

So maybe it is harder for MM developers right now to see what fields in
memdesc A overlap with which fields in memdesc B.  That _ought_ not to
be a concern!  We document which fields are available in each memdesc,
and have various assertions to trip when people make things not line up
any more.  There can still be problems, of course; we haven't set the
assertions quite tightly enough in some cases.

People are going to keep adding crap to struct page, and they're going
to keep misusing the crap that's in struct page.  That has to stop.