[RFC,v3,25/35] arm64: mte: Reserve tag block for the zero page

Commit Message

Alexandru Elisei Jan. 25, 2024, 4:42 p.m. UTC

On arm64, when a page is mapped as tagged, its tags are zeroed for two
reasons:

* To prevent leakage of tags to userspace.

* To allow userspace to access the contents of the page with having to set
  the tags explicitely (bits 59:56 of an userspace pointer are zero, which
  correspond to tag 0b0000).

The zero page receives special treatment, as the tags for the zero page are
zeroed when the MTE feature is being enabled. This is done for performance
reasons - the tags are zeroed once, instead of every time the page is
mapped.

When the tags for the zero page are zeroed, tag storage is not yet enabled.
Reserve tag storage for the page immediately after tag storage management
becomes enabled.

Note that zeroing tags before tag storage management is enabled is safe to
do because the tag storage pages are reserved at that point.

Signed-off-by: Alexandru Elisei <[e-mail]>
---

Changes since rfc v2:

* Expanded commit message (David Hildenbrand)

 arch/arm64/kernel/mte_tag_storage.c | 2 ++
 1 file changed, 2 insertions(+)

Patch

diff --git a/arch/arm64/kernel/mte_tag_storage.c b/arch/arm64/kernel/mte_tag_storage.c
index 8c347f4855e4..1c8469781870 100644
--- a/arch/arm64/kernel/mte_tag_storage.c
+++ b/arch/arm64/kernel/mte_tag_storage.c
@@ -363,6 +363,8 @@  static int __init mte_enable_tag_storage(void)
 			goto out_disabled;
 	}
 
+	reserve_tag_storage(ZERO_PAGE(0), 0, GFP_HIGHUSER);
+
 	static_branch_enable(&tag_storage_enabled_key);
 	pr_info("MTE tag storage region management enabled");