在 2022/3/22 0:40, Catalin Marinas 写道:
On Mon, Mar 21, 2022 at 02:15:36PM +0800, Tong Tiangen wrote:Considering all your suggestions, The final logic should be: +#define pte_user(pte) (!!(pte_val(pte) & PTE_USER)) +#define pmd_user(pmd) pte_user(pmd_pte(pmd)) +#define pmd_user_exec(pmd) pte_user_exec(pmd_pte(pmd)) +#define pud_user(pud) pte_user(pud_pte(pud)) +static inline bool pte_user_accessible_page(pte_t pte) +{ + return pte_present(pte) && (pte_user(pte)|| pte_user_exec(pte)); +}This is fine.+static inline bool pmd_user_accessible_page(pmd_t pmd) +{ + return pmd_present(pmd) && (pmd_user(pmd)|| pmd_user_exec(pmd)); +}That's fine as well assuming that the function is only called on the set_pmd_at() path where we know that the pmd would be a block mapping (huge page). I think that's the case from a quick look at the current x86 implementation.
Yeah, PTC only check pmd block mapping(huge page) and pud is similar. These code logic will be included in V2. Thanks.
+static inline bool pud_user_accessible_page(pud_t pud) +{ + return pud_present(pud) && pud_user(pud); +}Same here.
