Re: [PATCH] x86/mm/pat: Don't leak reservations at the physical address limit
From: Xu Yilun
Date: Fri Nov 20 2026 - 12:46:09 EST
On Sat, Nov 07, 2026 at 09:25:57PM -0800, Chris Down wrote:
> memtype_reserve() and memtype_free() pass their physical addresses
> through sanitize_phys(). Both take an exclusive end address, but
> sanitize_phys() expects an inclusive one.
This reads like both memtype_reserve/free() need fix, but in fact
memtype_reserve() is already fixed, you are simply following up the fix
for memtype_free().
Make it explicit by referencing the previous fix, that makes review
easier.
>
> So this has the effect that for a range that ends exactly at the top of
> the physical address space, the mask turns the end into 0.
>
> As a result the ISA range check succeeds and memtype_free() returns 0
> without removing the reservation. A new reservation that overlaps
> existing entries takes on their memory type, so the stale entry goes on
> choosing the cache mode for that range.
>
> Let's fix this by doing some annoying off by one twiddling.
>
> Fixes: aeef8b5089b7 ("x86/pat: Pass valid address to sanitize_phys()")
Maybe you want to reference the previous fix, but that's not what Fixes
is for. Obviously, aeef8b5089b7 is not the one that cause the issue.
Please make the back porting work well.
> Cc: <stable@xxxxxxxxxxxxxxx>
> Signed-off-by: Chris Down <chris@xxxxxxxxxxxxxx>
> ---
> arch/x86/mm/pat/memtype.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
> index cf94fb561310..007d833b4c00 100644
> --- a/arch/x86/mm/pat/memtype.c
> +++ b/arch/x86/mm/pat/memtype.c
> @@ -613,7 +613,7 @@ int memtype_free(u64 start, u64 end)
> return 0;
>
> start = sanitize_phys(start);
> - end = sanitize_phys(end);
> + end = sanitize_phys(end - 1) + 1;
>
> /* Low ISA region is always mapped WB. No need to track */
> if (x86_platform.is_untracked_pat_range(start, end))
>
> base-commit: 5cdc3746fa163fb4e8d05fa3cabf46bc6aa7df75