On Sun, Aug 08, 2021 at 09:29:29PM -0700, Hugh Dickins wrote:

On Sat, 7 Aug 2021, Matthew Wilcox wrote:

quoted

I am, however, of the opinion that 2MB pages give us so much trouble
because they're so very special.  Few people exercise those code paths and
it's easy to break them without noticing.  This is partly why I want to
do arbitrary-order pages.  If everybody is running with compound pages
all the time, we'll see the corner cases often, and people other than
Hugh, Kirill and Mike will be able to work on them.

I don't entirely agree.  I'm all for your use of compound pages in page
cache, but don't think its problems are representative of the problems
in aiming for a PMD (or PUD) bar, with the weird page table transitions
we expect of "THP" there.

I haven't checked: is your use of compound pages in page cache still
limited to the HAVE_ARCH_TRANSPARENT_HUGEPAGE architectures?  When
the others could just as well use compound pages in page cache too.

It is no longer gated by whether TRANSPARENT_HUGEPAGE is enabled or not.
That's a relatively recent change, and I can't say that I've tested it.
I'll give it a try today on a PA-RISC system I've booted recently.

One of the followup pieces of work that I hope somebody other than
myself will undertake is using 64KB PTEs on a 4KB PAGE_SIZE ARM/POWER
machine if the stars align.