Le 04/12/2023 à 10:36, aneesh.ku...@kernel.org a écrit :
> From: "Aneesh Kumar K.V (IBM)" <aneesh.ku...@kernel.org>
>
> There used to be a dependency on _PAGE_PRIVILEGED with pte_savedwrite.
> But that got dropped by
> commit 6a56ccbcf6c6 ("mm/autonuma: use can_change_(pte|pmd)_writable() to
> replace savedwrite")
>
> With the change in this patch numa fault pte (pte_protnone()) gets mapped as
> regular user pte
> with RWX cleared (no-access) whereas earlier it used to be mapped
> _PAGE_PRIVILEGED.
>
> Hash fault handling code gets some WARN_ON added in this patch because
> those functions are not expected to get called with _PAGE_READ cleared.
> commit 18061c17c8ec ("powerpc/mm: Update PROTFAULT handling in the page
> fault path") explains the details.
>
> Signed-off-by: Aneesh Kumar K.V (IBM) <aneesh.ku...@kernel.org>
Reviewed-by: Christophe Leroy <christophe.le...@csgroup.eu>
> ---
> arch/powerpc/include/asm/book3s/64/pgtable.h | 10 ++--------
> arch/powerpc/mm/book3s64/hash_utils.c | 7 +++++++
> 2 files changed, 9 insertions(+), 8 deletions(-)
>
> diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h
> b/arch/powerpc/include/asm/book3s/64/pgtable.h
> index cb77eddca54b..927d585652bc 100644
> --- a/arch/powerpc/include/asm/book3s/64/pgtable.h
> +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h
> @@ -17,12 +17,6 @@
> #define _PAGE_EXEC 0x00001 /* execute permission */
> #define _PAGE_WRITE 0x00002 /* write access allowed */
> #define _PAGE_READ 0x00004 /* read access allowed */
> -#define _PAGE_NA _PAGE_PRIVILEGED
> -#define _PAGE_NAX _PAGE_EXEC
> -#define _PAGE_RO _PAGE_READ
> -#define _PAGE_ROX (_PAGE_READ | _PAGE_EXEC)
> -#define _PAGE_RW (_PAGE_READ | _PAGE_WRITE)
> -#define _PAGE_RWX (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC)
> #define _PAGE_PRIVILEGED 0x00008 /* kernel access only */
> #define _PAGE_SAO 0x00010 /* Strong access order */
> #define _PAGE_NON_IDEMPOTENT 0x00020 /* non idempotent memory */
> @@ -532,8 +526,8 @@ static inline bool pte_user(pte_t pte)
> static inline bool pte_access_permitted(pte_t pte, bool write)
> {
> /*
> - * _PAGE_READ is needed for any access and will be
> - * cleared for PROT_NONE
> + * _PAGE_READ is needed for any access and will be cleared for
> + * PROT_NONE. Execute-only mapping via PROT_EXEC also returns false.
> */
> if (!pte_present(pte) || !pte_user(pte) || !pte_read(pte))
> return false;
> diff --git a/arch/powerpc/mm/book3s64/hash_utils.c
> b/arch/powerpc/mm/book3s64/hash_utils.c
> index ad2afa08e62e..0626a25b0d72 100644
> --- a/arch/powerpc/mm/book3s64/hash_utils.c
> +++ b/arch/powerpc/mm/book3s64/hash_utils.c
> @@ -310,9 +310,16 @@ unsigned long htab_convert_pte_flags(unsigned long
> pteflags, unsigned long flags
> else
> rflags |= 0x3;
> }
> + VM_WARN_ONCE(!(pteflags & _PAGE_RWX), "no-access mapping
> request");
> } else {
> if (pteflags & _PAGE_RWX)
> rflags |= 0x2;
> + /*
> + * We should never hit this in normal fault handling because
> + * a permission check (check_pte_access()) will bubble this
> + * to higher level linux handler even for PAGE_NONE.
> + */
> + VM_WARN_ONCE(!(pteflags & _PAGE_RWX), "no-access mapping
> request");
> if (!((pteflags & _PAGE_WRITE) && (pteflags & _PAGE_DIRTY)))
> rflags |= 0x1;
> }
