Re: [PATCH 5/6] arm64: use HW PTE accessors in early map_range()
From: Ryan Roberts
Date: Mon Sep 21 2026 - 05:50:51 EST
On 14/09/2026 14:51, Muhammad Usama Anjum wrote:
> map_range() uses HW PTE pointers to build early translation tables at
> each level. Directly testing, decoding, or assigning *tbl treats an
> HW PTE as a SW PTE value, which requires explicit conversion once
> hw_pte_t becomes distinct.
>
> Use __ptep_get() to load a SW PTE value for the empty-entry check and
> child-table address, and __set_pte_nosync() to convert SW PTE values to
> HW PTEs on stores. The no-sync helper performs WRITE_ONCE() without
> adding the completion barriers used by __set_pte().
>
> Signed-off-by: Muhammad Usama Anjum <usama.anjum@xxxxxxx>
Reviewed-by: Ryan Roberts <ryan.roberts@xxxxxxx>
> ---
> arch/arm64/kernel/pi/map_range.c | 13 ++++++++-----
> 1 file changed, 8 insertions(+), 5 deletions(-)
>
> diff --git a/arch/arm64/kernel/pi/map_range.c b/arch/arm64/kernel/pi/map_range.c
> index 94f06da76e333..9c4c8060c5df4 100644
> --- a/arch/arm64/kernel/pi/map_range.c
> +++ b/arch/arm64/kernel/pi/map_range.c
> @@ -51,19 +51,21 @@ void __init map_range(phys_addr_t *pte, u64 start, u64 end, phys_addr_t pa,
>
> while (start < end) {
> u64 next = min((start | lmask) + 1, PAGE_ALIGN(end));
> + pte_t entry = __ptep_get(tbl);
>
> if (level < 2 || (level == 2 && (start | next | pa) & lmask)) {
> /*
> * This chunk needs a finer grained mapping. Create a
> * table mapping if necessary and recurse.
> */
> - if (pte_none(*tbl)) {
> - *tbl = __pte(__phys_to_pte_val(*pte) |
> + if (pte_none(entry)) {
> + entry = __pte(__phys_to_pte_val(*pte) |
> PMD_TYPE_TABLE | PMD_TABLE_UXN);
> - *pte += PTRS_PER_PTE * sizeof(pte_t);
> + __set_pte_nosync(tbl, entry);
> + *pte += PTRS_PER_PTE * sizeof(hw_pte_t);
> }
> map_range(pte, start, next, pa, prot, level + 1,
> - (hw_pte_t *)(__pte_to_phys(*tbl) + va_offset),
> + (hw_pte_t *)(__pte_to_phys(entry) + va_offset),
> may_use_cont, va_offset);
> } else {
> /*
> @@ -81,7 +83,8 @@ void __init map_range(phys_addr_t *pte, u64 start, u64 end, phys_addr_t pa,
> protval &= ~PTE_CONT;
>
> /* Put down a block or page mapping */
> - *tbl = __pte(__phys_to_pte_val(pa) | protval);
> + entry = __pte(__phys_to_pte_val(pa) | protval);
> + __set_pte_nosync(tbl, entry);
> }
> pa += next - start;
> start = next;
>