Re: [PATCH 5/5] mm: Do not map the shadow stack as THP

On Wed, Feb 25, 2026 at 01:02:36PM +0000, Mark Brown wrote:

On Tue, Feb 24, 2026 at 05:57:57PM +0000, Catalin Marinas wrote:

quoted

The default shadow stack size allocated on first prctl() for the main
thread or subsequently on clone() is either half of RLIMIT_STACK or half
of a thread's stack size (for arm64). Both of these are likely to be
suitable for a THP allocation and the kernel is more aggressive in
creating such mappings. However, it does not make much sense to use a
huge page. It didn't make sense for the normal stacks either, see commit
c4608d1bf7c6 ("mm: mmap: map MAP_STACK to VM_NOHUGEPAGE").

Reviewed-by: Mark Brown <broonie@kernel.org>

Thanks.

The create THP and immediately splitting it pattern is very clear when
checking the mm behaviour on new GCSs, this should help performance.

If the first access is a write, the kernel allocates a THP from the
start without subsequent splitting. Also since 6.13 (commit 1ced09e0331f
"mm: allocate THP on hugezeropage wp-fault"), we go for another THP on
write. It's still wasting memory and time to zero the full 2MB when
it's highly unlikely we'd ever use that much for a shadow stack.

-- 
Catalin