x86/mm: Ensure forced page table splitting
authorDave Hansen <dave.hansen@linux.intel.com>
Fri, 18 Nov 2022 15:16:16 +0000 (07:16 -0800)
committerDave Hansen <dave.hansen@linux.intel.com>
Thu, 15 Dec 2022 18:37:28 +0000 (10:37 -0800)
commit3e844d842d49cdbe61a4b338bdd512654179488a
treefea8b53ff4d583986b98be3bbf60f1ee65cee0b7
parent1cfaac2400c73378e78182a706be0f3ac8b93cd7
x86/mm: Ensure forced page table splitting

There are a few kernel users like kfence that require 4k pages to work
correctly and do not support large mappings.  They use set_memory_4k()
to break down those large mappings.

That, in turn relies on cpa_data->force_split option to indicate to
set_memory code that it should split page tables regardless of whether
the need to be.

But, a recent change added an optimization which would return early
if a set_memory request came in that did not change permissions.  It
did not consult ->force_split and would mistakenly optimize away the
splitting that set_memory_4k() needs.  This broke kfence.

Skip the same-permission optimization when ->force_split is set.

Fixes: 127960a05548 ("x86/mm: Inhibit _PAGE_NX changes from cpa_process_alias()")
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Tested-by: Marco Elver <elver@google.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/CA+G9fYuFxZTxkeS35VTZMXwQvohu73W3xbZ5NtjebsVvH6hCuA@mail.gmail.com/
arch/x86/mm/pat/set_memory.c