On Fri, Mar 20, 2026 at 03:59:42PM +0100, Ard Biesheuvel wrote:
> From: Ard Biesheuvel <[email protected]>
> 
> Nothing in the fixmap or in its vicinity requires executable
> permissions, and given that it is placed at exactly 1 GiB from the end
> of the virtual address space, we can safely set the hierarchical XN
> attributes on the level 2 table entries covering the fixmap, without
> running the risk of inadvertently taking away the executable permissions
> on an adjacent mappings.
> 
> This is a hardening measure that reduces the risk of the fixmap being
> abused to create executable mappings in the kernel address space.
> 
> Signed-off-by: Ard Biesheuvel <[email protected]>
> ---
>  arch/arm64/mm/fixmap.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/mm/fixmap.c b/arch/arm64/mm/fixmap.c
> index c5c5425791da..c3dd3c868cf5 100644
> --- a/arch/arm64/mm/fixmap.c
> +++ b/arch/arm64/mm/fixmap.c
> @@ -48,7 +48,8 @@ static void __init early_fixmap_init_pte(pmd_t *pmdp, 
> unsigned long addr)
>       if (pmd_none(pmd)) {
>               ptep = bm_pte[BM_PTE_TABLE_IDX(addr)];
>               __pmd_populate(pmdp, __pa_symbol(ptep),
> -                            PMD_TYPE_TABLE | PMD_TABLE_AF);
> +                            PMD_TYPE_TABLE | PMD_TABLE_AF |
> +                            PMD_TABLE_PXN | PMD_TABLE_UXN);
>       }
>  }

Sashiko reckons this breaks kpti. I think that's valid but I couldn't
reproduce it on qemu (maybe it doesn't implement hierarchical
permissions). Then I tried FVP and the whole series panics (unrelated to
kpti). With kvm-arm.mode=protected, I think kvm_ksym_ref() is lm_alias()
and we have kvm_hyp_init_symbols() trying to flush the bss.

I'll drop it for now.

Unable to handle kernel paging request at virtual address fff00000748f7000
Mem abort info:
  ESR = 0x0000000096000147
** replaying previous printk message **
  EC = 0x25: DABT (current EL), IL = 32 bits
  SET = 0, FnV = 0
  EA = 0, S1PTW = 0
  FSC = 0x07: level 3 translation fault
Data abort info:
  ISV = 0, ISS = 0x00000147, ISS2 = 0x00000000
  CM = 1, WnR = 1, TnD = 0, TagAccess = 0
  GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
swapper pgtable: 4k pages, 52-bit VAs, pgdp=00000000f40ad000
[fff00000748f7000] pgd=18000008fffff403, p4d=18000008ffffe403, 
pud=18000008ffffd403, pmd=18000008fffe9403, pte=00e80000f48f7406
Internal error: Oops: 0000000096000147 [#1]  SMP
Modules linked in:
CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.0.0-rc3-00013-g6bb20b972b8c 
#2 PREEMPT
Hardware name: FVP Base RevC (DT)
pstate: 81400005 (Nzcv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
pc : dcache_clean_inval_poc+0x24/0x48
lr : kvm_arm_init+0xb48/0x1338
sp : ffff80008005bd20
x29: ffff80008005bd60 x28: ffff95c63230b000 x27: 0000000000000004
x26: 0000000000000001 x25: ffff95c631a160c0 x24: 0000000000000008
x23: 0000000000000000 x22: 0000000000000000 x21: fff00000748f7000
x20: 0000000000002000 x19: 0001101131111112 x18: 00000000ffffffff
x17: ffff95c6321fdd88 x16: 0000000031427afd x15: 0000000000000100
x14: 0000000000000000 x13: 0000000000077a9a x12: fff000087f805650
x11: fff000087f805630 x10: ffffc1ffe005db08 x9 : 0000000000000000
x8 : 0000000080000000 x7 : ffff80008005bc50 x6 : 000f63580145a000
x5 : ffff95c6322f8000 x4 : 0000000000000000 x3 : 000000000000003f
x2 : 0000000000000040 x1 : fff00000748f9000 x0 : fff00000748f7000
Call trace:
 dcache_clean_inval_poc+0x24/0x48 (P)
 do_one_initcall+0x60/0x1d4
 kernel_init_freeable+0x24c/0x2d4
 kernel_init+0x24/0x140
 ret_from_fork+0x10/0x20
Code: 9ac32042 d1000443 8a230000 d503201f (d50b7e20)
---[ end trace 0000000000000000 ]---

-- 
Catalin

Reply via email to