Re: Endless calls to xas_split_alloc() due to corrupted xarray entry

From: Matthew Wilcox
Date: Wed Jun 19 2024 - 16:51:09 EST


On Wed, Jun 19, 2024 at 08:48:28AM -0700, Linus Torvalds wrote:
> On Wed, 19 Jun 2024 at 07:31, Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> >
> > Actually, it's 11. We can't split an order-12 folio because we'd have
> > to allocate two levels of radix tree, and I decided that was too much
> > work. Also, I didn't know that ARM used order-13 PMD size at the time.
> >
> > I think this is the best fix (modulo s/12/11/).
>
> Can we use some more descriptive thing than the magic constant 11 that
> is clearly very subtle.
>
> Is it "XA_CHUNK_SHIFT * 2 - 1"
>
> IOW, something like
>
> #define MAX_XAS_ORDER (XA_CHUNK_SHIFT * 2 - 1)
> #define MAX_PAGECACHE_ORDER min(HPAGE_PMD_ORDER,12)
>
> except for the non-TRANSPARENT_HUGEPAGE case where it currently does
>
> #define MAX_PAGECACHE_ORDER 8
>
> and I assume that "8" is just "random round value, smaller than 11"?

It's actually documented:

/*
* There are some parts of the kernel which assume that PMD entries
* are exactly HPAGE_PMD_ORDER. Those should be fixed, but until then,
* limit the maximum allocation order to PMD size. I'm not aware of any
* assumptions about maximum order if THP are disabled, but 8 seems like
* a good order (that's 1MB if you're using 4kB pages)
*/
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
#define MAX_PAGECACHE_ORDER HPAGE_PMD_ORDER
#else
#define MAX_PAGECACHE_ORDER 8
#endif

although I'm not even sure if we use it if CONFIG_TRANSPARENT_HUGEPAGE
is disabled. All the machinery to split pages is gated by
CONFIG_TRANSPARENT_HUGEPAGE, so I think we end up completely ignoring
it. I used to say "somebody should do the work to split out
CONFIG_LARGE_FOLIOS from CONFIG_TRANSPARENT_HUGEPAGE", but now I think
that nobody cares about the architectures that don't support it,
and it's not worth anybody's time pretending that we do.