riscv: Use accessors to page table entries instead of direct dereference
authorAlexandre Ghiti <alexghiti@rivosinc.com>
Wed, 13 Dec 2023 20:30:01 +0000 (21:30 +0100)
committerPalmer Dabbelt <palmer@rivosinc.com>
Wed, 20 Dec 2023 18:48:15 +0000 (10:48 -0800)
As very well explained in commit 20a004e7b017 ("arm64: mm: Use
READ_ONCE/WRITE_ONCE when accessing page tables"), an architecture whose
page table walker can modify the PTE in parallel must use
READ_ONCE()/WRITE_ONCE() macro to avoid any compiler transformation.

So apply that to riscv which is such architecture.

Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Acked-by: Anup Patel <anup@brainfault.org>
Link: https://lore.kernel.org/r/20231213203001.179237-5-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
arch/riscv/include/asm/kfence.h
arch/riscv/include/asm/pgtable-64.h
arch/riscv/include/asm/pgtable.h
arch/riscv/kernel/efi.c
arch/riscv/kvm/mmu.c
arch/riscv/mm/fault.c
arch/riscv/mm/hugetlbpage.c
arch/riscv/mm/kasan_init.c
arch/riscv/mm/pageattr.c
arch/riscv/mm/pgtable.c

index 0bbffd5..7388edd 100644 (file)
@@ -18,9 +18,9 @@ static inline bool kfence_protect_page(unsigned long addr, bool protect)
        pte_t *pte = virt_to_kpte(addr);
 
        if (protect)
-               set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_PRESENT));
+               set_pte(pte, __pte(pte_val(ptep_get(pte)) & ~_PAGE_PRESENT));
        else
-               set_pte(pte, __pte(pte_val(*pte) | _PAGE_PRESENT));
+               set_pte(pte, __pte(pte_val(ptep_get(pte)) | _PAGE_PRESENT));
 
        flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
 
index 5d8431a..b42017d 100644 (file)
@@ -340,13 +340,7 @@ static inline struct page *p4d_page(p4d_t p4d)
 #define pud_index(addr) (((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
 
 #define pud_offset pud_offset
-static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
-{
-       if (pgtable_l4_enabled)
-               return p4d_pgtable(*p4d) + pud_index(address);
-
-       return (pud_t *)p4d;
-}
+pud_t *pud_offset(p4d_t *p4d, unsigned long address);
 
 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
 {
@@ -404,12 +398,6 @@ static inline struct page *pgd_page(pgd_t pgd)
 #define p4d_index(addr) (((addr) >> P4D_SHIFT) & (PTRS_PER_P4D - 1))
 
 #define p4d_offset p4d_offset
-static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
-{
-       if (pgtable_l5_enabled)
-               return pgd_pgtable(*pgd) + p4d_index(address);
-
-       return (p4d_t *)pgd;
-}
+p4d_t *p4d_offset(pgd_t *pgd, unsigned long address);
 
 #endif /* _ASM_RISCV_PGTABLE_64_H */
index c9f4b25..3773f45 100644 (file)
@@ -544,19 +544,12 @@ static inline void pte_clear(struct mm_struct *mm,
        __set_pte_at(ptep, __pte(0));
 }
 
-#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
-static inline int ptep_set_access_flags(struct vm_area_struct *vma,
-                                       unsigned long address, pte_t *ptep,
-                                       pte_t entry, int dirty)
-{
-       if (!pte_same(*ptep, entry))
-               __set_pte_at(ptep, entry);
-       /*
-        * update_mmu_cache will unconditionally execute, handling both
-        * the case that the PTE changed and the spurious fault case.
-        */
-       return true;
-}
+#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS      /* defined in mm/pgtable.c */
+extern int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address,
+                                pte_t *ptep, pte_t entry, int dirty);
+#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG  /* defined in mm/pgtable.c */
+extern int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long address,
+                                    pte_t *ptep);
 
 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
@@ -569,16 +562,6 @@ static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
        return pte;
 }
 
-#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
-static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
-                                           unsigned long address,
-                                           pte_t *ptep)
-{
-       if (!pte_young(*ptep))
-               return 0;
-       return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
-}
-
 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
 static inline void ptep_set_wrprotect(struct mm_struct *mm,
                                      unsigned long address, pte_t *ptep)
index aa6209a..b64bf16 100644 (file)
@@ -60,7 +60,7 @@ int __init efi_create_mapping(struct mm_struct *mm, efi_memory_desc_t *md)
 static int __init set_permissions(pte_t *ptep, unsigned long addr, void *data)
 {
        efi_memory_desc_t *md = data;
-       pte_t pte = READ_ONCE(*ptep);
+       pte_t pte = ptep_get(ptep);
        unsigned long val;
 
        if (md->attribute & EFI_MEMORY_RO) {
index 068c745..a9e2fd7 100644 (file)
@@ -103,7 +103,7 @@ static bool gstage_get_leaf_entry(struct kvm *kvm, gpa_t addr,
        *ptep_level = current_level;
        ptep = (pte_t *)kvm->arch.pgd;
        ptep = &ptep[gstage_pte_index(addr, current_level)];
-       while (ptep && pte_val(*ptep)) {
+       while (ptep && pte_val(ptep_get(ptep))) {
                if (gstage_pte_leaf(ptep)) {
                        *ptep_level = current_level;
                        *ptepp = ptep;
@@ -113,7 +113,7 @@ static bool gstage_get_leaf_entry(struct kvm *kvm, gpa_t addr,
                if (current_level) {
                        current_level--;
                        *ptep_level = current_level;
-                       ptep = (pte_t *)gstage_pte_page_vaddr(*ptep);
+                       ptep = (pte_t *)gstage_pte_page_vaddr(ptep_get(ptep));
                        ptep = &ptep[gstage_pte_index(addr, current_level)];
                } else {
                        ptep = NULL;
@@ -149,25 +149,25 @@ static int gstage_set_pte(struct kvm *kvm, u32 level,
                if (gstage_pte_leaf(ptep))
                        return -EEXIST;
 
-               if (!pte_val(*ptep)) {
+               if (!pte_val(ptep_get(ptep))) {
                        if (!pcache)
                                return -ENOMEM;
                        next_ptep = kvm_mmu_memory_cache_alloc(pcache);
                        if (!next_ptep)
                                return -ENOMEM;
-                       *ptep = pfn_pte(PFN_DOWN(__pa(next_ptep)),
-                                       __pgprot(_PAGE_TABLE));
+                       set_pte(ptep, pfn_pte(PFN_DOWN(__pa(next_ptep)),
+                                             __pgprot(_PAGE_TABLE)));
                } else {
                        if (gstage_pte_leaf(ptep))
                                return -EEXIST;
-                       next_ptep = (pte_t *)gstage_pte_page_vaddr(*ptep);
+                       next_ptep = (pte_t *)gstage_pte_page_vaddr(ptep_get(ptep));
                }
 
                current_level--;
                ptep = &next_ptep[gstage_pte_index(addr, current_level)];
        }
 
-       *ptep = *new_pte;
+       set_pte(ptep, *new_pte);
        if (gstage_pte_leaf(ptep))
                gstage_remote_tlb_flush(kvm, current_level, addr);
 
@@ -239,11 +239,11 @@ static void gstage_op_pte(struct kvm *kvm, gpa_t addr,
 
        BUG_ON(addr & (page_size - 1));
 
-       if (!pte_val(*ptep))
+       if (!pte_val(ptep_get(ptep)))
                return;
 
        if (ptep_level && !gstage_pte_leaf(ptep)) {
-               next_ptep = (pte_t *)gstage_pte_page_vaddr(*ptep);
+               next_ptep = (pte_t *)gstage_pte_page_vaddr(ptep_get(ptep));
                next_ptep_level = ptep_level - 1;
                ret = gstage_level_to_page_size(next_ptep_level,
                                                &next_page_size);
@@ -261,7 +261,7 @@ static void gstage_op_pte(struct kvm *kvm, gpa_t addr,
                if (op == GSTAGE_OP_CLEAR)
                        set_pte(ptep, __pte(0));
                else if (op == GSTAGE_OP_WP)
-                       set_pte(ptep, __pte(pte_val(*ptep) & ~_PAGE_WRITE));
+                       set_pte(ptep, __pte(pte_val(ptep_get(ptep)) & ~_PAGE_WRITE));
                gstage_remote_tlb_flush(kvm, ptep_level, addr);
        }
 }
@@ -603,7 +603,7 @@ bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
                                   &ptep, &ptep_level))
                return false;
 
-       return pte_young(*ptep);
+       return pte_young(ptep_get(ptep));
 }
 
 int kvm_riscv_gstage_map(struct kvm_vcpu *vcpu,
index 90d4ba3..76f1df7 100644 (file)
@@ -136,24 +136,24 @@ static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long a
        pgd = (pgd_t *)pfn_to_virt(pfn) + index;
        pgd_k = init_mm.pgd + index;
 
-       if (!pgd_present(*pgd_k)) {
+       if (!pgd_present(pgdp_get(pgd_k))) {
                no_context(regs, addr);
                return;
        }
-       set_pgd(pgd, *pgd_k);
+       set_pgd(pgd, pgdp_get(pgd_k));
 
        p4d_k = p4d_offset(pgd_k, addr);
-       if (!p4d_present(*p4d_k)) {
+       if (!p4d_present(p4dp_get(p4d_k))) {
                no_context(regs, addr);
                return;
        }
 
        pud_k = pud_offset(p4d_k, addr);
-       if (!pud_present(*pud_k)) {
+       if (!pud_present(pudp_get(pud_k))) {
                no_context(regs, addr);
                return;
        }
-       if (pud_leaf(*pud_k))
+       if (pud_leaf(pudp_get(pud_k)))
                goto flush_tlb;
 
        /*
@@ -161,11 +161,11 @@ static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long a
         * to copy individual PTEs
         */
        pmd_k = pmd_offset(pud_k, addr);
-       if (!pmd_present(*pmd_k)) {
+       if (!pmd_present(pmdp_get(pmd_k))) {
                no_context(regs, addr);
                return;
        }
-       if (pmd_leaf(*pmd_k))
+       if (pmd_leaf(pmdp_get(pmd_k)))
                goto flush_tlb;
 
        /*
@@ -175,7 +175,7 @@ static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long a
         * silently loop forever.
         */
        pte_k = pte_offset_kernel(pmd_k, addr);
-       if (!pte_present(*pte_k)) {
+       if (!pte_present(ptep_get(pte_k))) {
                no_context(regs, addr);
                return;
        }
index b52f021..431596c 100644 (file)
@@ -54,7 +54,7 @@ pte_t *huge_pte_alloc(struct mm_struct *mm,
        }
 
        if (sz == PMD_SIZE) {
-               if (want_pmd_share(vma, addr) && pud_none(*pud))
+               if (want_pmd_share(vma, addr) && pud_none(pudp_get(pud)))
                        pte = huge_pmd_share(mm, vma, addr, pud);
                else
                        pte = (pte_t *)pmd_alloc(mm, pud, addr);
@@ -93,11 +93,11 @@ pte_t *huge_pte_offset(struct mm_struct *mm,
        pmd_t *pmd;
 
        pgd = pgd_offset(mm, addr);
-       if (!pgd_present(*pgd))
+       if (!pgd_present(pgdp_get(pgd)))
                return NULL;
 
        p4d = p4d_offset(pgd, addr);
-       if (!p4d_present(*p4d))
+       if (!p4d_present(p4dp_get(p4d)))
                return NULL;
 
        pud = pud_offset(p4d, addr);
@@ -105,7 +105,7 @@ pte_t *huge_pte_offset(struct mm_struct *mm,
                /* must be pud huge, non-present or none */
                return (pte_t *)pud;
 
-       if (!pud_present(*pud))
+       if (!pud_present(pudp_get(pud)))
                return NULL;
 
        pmd = pmd_offset(pud, addr);
@@ -113,7 +113,7 @@ pte_t *huge_pte_offset(struct mm_struct *mm,
                /* must be pmd huge, non-present or none */
                return (pte_t *)pmd;
 
-       if (!pmd_present(*pmd))
+       if (!pmd_present(pmdp_get(pmd)))
                return NULL;
 
        for_each_napot_order(order) {
@@ -293,7 +293,7 @@ void huge_pte_clear(struct mm_struct *mm,
                    pte_t *ptep,
                    unsigned long sz)
 {
-       pte_t pte = READ_ONCE(*ptep);
+       pte_t pte = ptep_get(ptep);
        int i, pte_num;
 
        if (!pte_napot(pte)) {
index 5e39dcf..e962518 100644 (file)
@@ -31,7 +31,7 @@ static void __init kasan_populate_pte(pmd_t *pmd, unsigned long vaddr, unsigned
        phys_addr_t phys_addr;
        pte_t *ptep, *p;
 
-       if (pmd_none(*pmd)) {
+       if (pmd_none(pmdp_get(pmd))) {
                p = memblock_alloc(PTRS_PER_PTE * sizeof(pte_t), PAGE_SIZE);
                set_pmd(pmd, pfn_pmd(PFN_DOWN(__pa(p)), PAGE_TABLE));
        }
@@ -39,7 +39,7 @@ static void __init kasan_populate_pte(pmd_t *pmd, unsigned long vaddr, unsigned
        ptep = pte_offset_kernel(pmd, vaddr);
 
        do {
-               if (pte_none(*ptep)) {
+               if (pte_none(ptep_get(ptep))) {
                        phys_addr = memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
                        set_pte(ptep, pfn_pte(PFN_DOWN(phys_addr), PAGE_KERNEL));
                        memset(__va(phys_addr), KASAN_SHADOW_INIT, PAGE_SIZE);
@@ -53,7 +53,7 @@ static void __init kasan_populate_pmd(pud_t *pud, unsigned long vaddr, unsigned
        pmd_t *pmdp, *p;
        unsigned long next;
 
-       if (pud_none(*pud)) {
+       if (pud_none(pudp_get(pud))) {
                p = memblock_alloc(PTRS_PER_PMD * sizeof(pmd_t), PAGE_SIZE);
                set_pud(pud, pfn_pud(PFN_DOWN(__pa(p)), PAGE_TABLE));
        }
@@ -63,7 +63,8 @@ static void __init kasan_populate_pmd(pud_t *pud, unsigned long vaddr, unsigned
        do {
                next = pmd_addr_end(vaddr, end);
 
-               if (pmd_none(*pmdp) && IS_ALIGNED(vaddr, PMD_SIZE) && (next - vaddr) >= PMD_SIZE) {
+               if (pmd_none(pmdp_get(pmdp)) && IS_ALIGNED(vaddr, PMD_SIZE) &&
+                   (next - vaddr) >= PMD_SIZE) {
                        phys_addr = memblock_phys_alloc(PMD_SIZE, PMD_SIZE);
                        if (phys_addr) {
                                set_pmd(pmdp, pfn_pmd(PFN_DOWN(phys_addr), PAGE_KERNEL));
@@ -83,7 +84,7 @@ static void __init kasan_populate_pud(p4d_t *p4d,
        pud_t *pudp, *p;
        unsigned long next;
 
-       if (p4d_none(*p4d)) {
+       if (p4d_none(p4dp_get(p4d))) {
                p = memblock_alloc(PTRS_PER_PUD * sizeof(pud_t), PAGE_SIZE);
                set_p4d(p4d, pfn_p4d(PFN_DOWN(__pa(p)), PAGE_TABLE));
        }
@@ -93,7 +94,8 @@ static void __init kasan_populate_pud(p4d_t *p4d,
        do {
                next = pud_addr_end(vaddr, end);
 
-               if (pud_none(*pudp) && IS_ALIGNED(vaddr, PUD_SIZE) && (next - vaddr) >= PUD_SIZE) {
+               if (pud_none(pudp_get(pudp)) && IS_ALIGNED(vaddr, PUD_SIZE) &&
+                   (next - vaddr) >= PUD_SIZE) {
                        phys_addr = memblock_phys_alloc(PUD_SIZE, PUD_SIZE);
                        if (phys_addr) {
                                set_pud(pudp, pfn_pud(PFN_DOWN(phys_addr), PAGE_KERNEL));
@@ -113,7 +115,7 @@ static void __init kasan_populate_p4d(pgd_t *pgd,
        p4d_t *p4dp, *p;
        unsigned long next;
 
-       if (pgd_none(*pgd)) {
+       if (pgd_none(pgdp_get(pgd))) {
                p = memblock_alloc(PTRS_PER_P4D * sizeof(p4d_t), PAGE_SIZE);
                set_pgd(pgd, pfn_pgd(PFN_DOWN(__pa(p)), PAGE_TABLE));
        }
@@ -123,7 +125,8 @@ static void __init kasan_populate_p4d(pgd_t *pgd,
        do {
                next = p4d_addr_end(vaddr, end);
 
-               if (p4d_none(*p4dp) && IS_ALIGNED(vaddr, P4D_SIZE) && (next - vaddr) >= P4D_SIZE) {
+               if (p4d_none(p4dp_get(p4dp)) && IS_ALIGNED(vaddr, P4D_SIZE) &&
+                   (next - vaddr) >= P4D_SIZE) {
                        phys_addr = memblock_phys_alloc(P4D_SIZE, P4D_SIZE);
                        if (phys_addr) {
                                set_p4d(p4dp, pfn_p4d(PFN_DOWN(phys_addr), PAGE_KERNEL));
@@ -145,7 +148,7 @@ static void __init kasan_populate_pgd(pgd_t *pgdp,
        do {
                next = pgd_addr_end(vaddr, end);
 
-               if (pgd_none(*pgdp) && IS_ALIGNED(vaddr, PGDIR_SIZE) &&
+               if (pgd_none(pgdp_get(pgdp)) && IS_ALIGNED(vaddr, PGDIR_SIZE) &&
                    (next - vaddr) >= PGDIR_SIZE) {
                        phys_addr = memblock_phys_alloc(PGDIR_SIZE, PGDIR_SIZE);
                        if (phys_addr) {
@@ -168,7 +171,7 @@ static void __init kasan_early_clear_pud(p4d_t *p4dp,
        if (!pgtable_l4_enabled) {
                pudp = (pud_t *)p4dp;
        } else {
-               base_pud = pt_ops.get_pud_virt(pfn_to_phys(_p4d_pfn(*p4dp)));
+               base_pud = pt_ops.get_pud_virt(pfn_to_phys(_p4d_pfn(p4dp_get(p4dp))));
                pudp = base_pud + pud_index(vaddr);
        }
 
@@ -193,7 +196,7 @@ static void __init kasan_early_clear_p4d(pgd_t *pgdp,
        if (!pgtable_l5_enabled) {
                p4dp = (p4d_t *)pgdp;
        } else {
-               base_p4d = pt_ops.get_p4d_virt(pfn_to_phys(_pgd_pfn(*pgdp)));
+               base_p4d = pt_ops.get_p4d_virt(pfn_to_phys(_pgd_pfn(pgdp_get(pgdp))));
                p4dp = base_p4d + p4d_index(vaddr);
        }
 
@@ -239,14 +242,14 @@ static void __init kasan_early_populate_pud(p4d_t *p4dp,
        if (!pgtable_l4_enabled) {
                pudp = (pud_t *)p4dp;
        } else {
-               base_pud = pt_ops.get_pud_virt(pfn_to_phys(_p4d_pfn(*p4dp)));
+               base_pud = pt_ops.get_pud_virt(pfn_to_phys(_p4d_pfn(p4dp_get(p4dp))));
                pudp = base_pud + pud_index(vaddr);
        }
 
        do {
                next = pud_addr_end(vaddr, end);
 
-               if (pud_none(*pudp) && IS_ALIGNED(vaddr, PUD_SIZE) &&
+               if (pud_none(pudp_get(pudp)) && IS_ALIGNED(vaddr, PUD_SIZE) &&
                    (next - vaddr) >= PUD_SIZE) {
                        phys_addr = __pa((uintptr_t)kasan_early_shadow_pmd);
                        set_pud(pudp, pfn_pud(PFN_DOWN(phys_addr), PAGE_TABLE));
@@ -277,14 +280,14 @@ static void __init kasan_early_populate_p4d(pgd_t *pgdp,
        if (!pgtable_l5_enabled) {
                p4dp = (p4d_t *)pgdp;
        } else {
-               base_p4d = pt_ops.get_p4d_virt(pfn_to_phys(_pgd_pfn(*pgdp)));
+               base_p4d = pt_ops.get_p4d_virt(pfn_to_phys(_pgd_pfn(pgdp_get(pgdp))));
                p4dp = base_p4d + p4d_index(vaddr);
        }
 
        do {
                next = p4d_addr_end(vaddr, end);
 
-               if (p4d_none(*p4dp) && IS_ALIGNED(vaddr, P4D_SIZE) &&
+               if (p4d_none(p4dp_get(p4dp)) && IS_ALIGNED(vaddr, P4D_SIZE) &&
                    (next - vaddr) >= P4D_SIZE) {
                        phys_addr = __pa((uintptr_t)kasan_early_shadow_pud);
                        set_p4d(p4dp, pfn_p4d(PFN_DOWN(phys_addr), PAGE_TABLE));
@@ -305,7 +308,7 @@ static void __init kasan_early_populate_pgd(pgd_t *pgdp,
        do {
                next = pgd_addr_end(vaddr, end);
 
-               if (pgd_none(*pgdp) && IS_ALIGNED(vaddr, PGDIR_SIZE) &&
+               if (pgd_none(pgdp_get(pgdp)) && IS_ALIGNED(vaddr, PGDIR_SIZE) &&
                    (next - vaddr) >= PGDIR_SIZE) {
                        phys_addr = __pa((uintptr_t)kasan_early_shadow_p4d);
                        set_pgd(pgdp, pfn_pgd(PFN_DOWN(phys_addr), PAGE_TABLE));
@@ -381,7 +384,7 @@ static void __init kasan_shallow_populate_pud(p4d_t *p4d,
        do {
                next = pud_addr_end(vaddr, end);
 
-               if (pud_none(*pud_k)) {
+               if (pud_none(pudp_get(pud_k))) {
                        p = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
                        set_pud(pud_k, pfn_pud(PFN_DOWN(__pa(p)), PAGE_TABLE));
                        continue;
@@ -401,7 +404,7 @@ static void __init kasan_shallow_populate_p4d(pgd_t *pgd,
        do {
                next = p4d_addr_end(vaddr, end);
 
-               if (p4d_none(*p4d_k)) {
+               if (p4d_none(p4dp_get(p4d_k))) {
                        p = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
                        set_p4d(p4d_k, pfn_p4d(PFN_DOWN(__pa(p)), PAGE_TABLE));
                        continue;
@@ -420,7 +423,7 @@ static void __init kasan_shallow_populate_pgd(unsigned long vaddr, unsigned long
        do {
                next = pgd_addr_end(vaddr, end);
 
-               if (pgd_none(*pgd_k)) {
+               if (pgd_none(pgdp_get(pgd_k))) {
                        p = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
                        set_pgd(pgd_k, pfn_pgd(PFN_DOWN(__pa(p)), PAGE_TABLE));
                        continue;
@@ -451,7 +454,7 @@ static void __init create_tmp_mapping(void)
 
        /* Copy the last p4d since it is shared with the kernel mapping. */
        if (pgtable_l5_enabled) {
-               ptr = (p4d_t *)pgd_page_vaddr(*pgd_offset_k(KASAN_SHADOW_END));
+               ptr = (p4d_t *)pgd_page_vaddr(pgdp_get(pgd_offset_k(KASAN_SHADOW_END)));
                memcpy(tmp_p4d, ptr, sizeof(p4d_t) * PTRS_PER_P4D);
                set_pgd(&tmp_pg_dir[pgd_index(KASAN_SHADOW_END)],
                        pfn_pgd(PFN_DOWN(__pa(tmp_p4d)), PAGE_TABLE));
@@ -462,7 +465,7 @@ static void __init create_tmp_mapping(void)
 
        /* Copy the last pud since it is shared with the kernel mapping. */
        if (pgtable_l4_enabled) {
-               ptr = (pud_t *)p4d_page_vaddr(*(base_p4d + p4d_index(KASAN_SHADOW_END)));
+               ptr = (pud_t *)p4d_page_vaddr(p4dp_get(base_p4d + p4d_index(KASAN_SHADOW_END)));
                memcpy(tmp_pud, ptr, sizeof(pud_t) * PTRS_PER_PUD);
                set_p4d(&base_p4d[p4d_index(KASAN_SHADOW_END)],
                        pfn_p4d(PFN_DOWN(__pa(tmp_pud)), PAGE_TABLE));
index fc5fc4f..0b5e38e 100644 (file)
@@ -29,7 +29,7 @@ static unsigned long set_pageattr_masks(unsigned long val, struct mm_walk *walk)
 static int pageattr_p4d_entry(p4d_t *p4d, unsigned long addr,
                              unsigned long next, struct mm_walk *walk)
 {
-       p4d_t val = READ_ONCE(*p4d);
+       p4d_t val = p4dp_get(p4d);
 
        if (p4d_leaf(val)) {
                val = __p4d(set_pageattr_masks(p4d_val(val), walk));
@@ -42,7 +42,7 @@ static int pageattr_p4d_entry(p4d_t *p4d, unsigned long addr,
 static int pageattr_pud_entry(pud_t *pud, unsigned long addr,
                              unsigned long next, struct mm_walk *walk)
 {
-       pud_t val = READ_ONCE(*pud);
+       pud_t val = pudp_get(pud);
 
        if (pud_leaf(val)) {
                val = __pud(set_pageattr_masks(pud_val(val), walk));
@@ -55,7 +55,7 @@ static int pageattr_pud_entry(pud_t *pud, unsigned long addr,
 static int pageattr_pmd_entry(pmd_t *pmd, unsigned long addr,
                              unsigned long next, struct mm_walk *walk)
 {
-       pmd_t val = READ_ONCE(*pmd);
+       pmd_t val = pmdp_get(pmd);
 
        if (pmd_leaf(val)) {
                val = __pmd(set_pageattr_masks(pmd_val(val), walk));
@@ -68,7 +68,7 @@ static int pageattr_pmd_entry(pmd_t *pmd, unsigned long addr,
 static int pageattr_pte_entry(pte_t *pte, unsigned long addr,
                              unsigned long next, struct mm_walk *walk)
 {
-       pte_t val = READ_ONCE(*pte);
+       pte_t val = ptep_get(pte);
 
        val = __pte(set_pageattr_masks(pte_val(val), walk));
        set_pte(pte, val);
@@ -108,10 +108,10 @@ static int __split_linear_mapping_pmd(pud_t *pudp,
                    vaddr <= (vaddr & PMD_MASK) && end >= next)
                        continue;
 
-               if (pmd_leaf(*pmdp)) {
+               if (pmd_leaf(pmdp_get(pmdp))) {
                        struct page *pte_page;
-                       unsigned long pfn = _pmd_pfn(*pmdp);
-                       pgprot_t prot = __pgprot(pmd_val(*pmdp) & ~_PAGE_PFN_MASK);
+                       unsigned long pfn = _pmd_pfn(pmdp_get(pmdp));
+                       pgprot_t prot = __pgprot(pmd_val(pmdp_get(pmdp)) & ~_PAGE_PFN_MASK);
                        pte_t *ptep_new;
                        int i;
 
@@ -148,10 +148,10 @@ static int __split_linear_mapping_pud(p4d_t *p4dp,
                    vaddr <= (vaddr & PUD_MASK) && end >= next)
                        continue;
 
-               if (pud_leaf(*pudp)) {
+               if (pud_leaf(pudp_get(pudp))) {
                        struct page *pmd_page;
-                       unsigned long pfn = _pud_pfn(*pudp);
-                       pgprot_t prot = __pgprot(pud_val(*pudp) & ~_PAGE_PFN_MASK);
+                       unsigned long pfn = _pud_pfn(pudp_get(pudp));
+                       pgprot_t prot = __pgprot(pud_val(pudp_get(pudp)) & ~_PAGE_PFN_MASK);
                        pmd_t *pmdp_new;
                        int i;
 
@@ -197,10 +197,10 @@ static int __split_linear_mapping_p4d(pgd_t *pgdp,
                    vaddr <= (vaddr & P4D_MASK) && end >= next)
                        continue;
 
-               if (p4d_leaf(*p4dp)) {
+               if (p4d_leaf(p4dp_get(p4dp))) {
                        struct page *pud_page;
-                       unsigned long pfn = _p4d_pfn(*p4dp);
-                       pgprot_t prot = __pgprot(p4d_val(*p4dp) & ~_PAGE_PFN_MASK);
+                       unsigned long pfn = _p4d_pfn(p4dp_get(p4dp));
+                       pgprot_t prot = __pgprot(p4d_val(p4dp_get(p4dp)) & ~_PAGE_PFN_MASK);
                        pud_t *pudp_new;
                        int i;
 
@@ -406,29 +406,29 @@ bool kernel_page_present(struct page *page)
        pte_t *pte;
 
        pgd = pgd_offset_k(addr);
-       if (!pgd_present(*pgd))
+       if (!pgd_present(pgdp_get(pgd)))
                return false;
-       if (pgd_leaf(*pgd))
+       if (pgd_leaf(pgdp_get(pgd)))
                return true;
 
        p4d = p4d_offset(pgd, addr);
-       if (!p4d_present(*p4d))
+       if (!p4d_present(p4dp_get(p4d)))
                return false;
-       if (p4d_leaf(*p4d))
+       if (p4d_leaf(p4dp_get(p4d)))
                return true;
 
        pud = pud_offset(p4d, addr);
-       if (!pud_present(*pud))
+       if (!pud_present(pudp_get(pud)))
                return false;
-       if (pud_leaf(*pud))
+       if (pud_leaf(pudp_get(pud)))
                return true;
 
        pmd = pmd_offset(pud, addr);
-       if (!pmd_present(*pmd))
+       if (!pmd_present(pmdp_get(pmd)))
                return false;
-       if (pmd_leaf(*pmd))
+       if (pmd_leaf(pmdp_get(pmd)))
                return true;
 
        pte = pte_offset_kernel(pmd, addr);
-       return pte_present(*pte);
+       return pte_present(ptep_get(pte));
 }
index fef4e73..ef887ef 100644 (file)
@@ -5,6 +5,47 @@
 #include <linux/kernel.h>
 #include <linux/pgtable.h>
 
+int ptep_set_access_flags(struct vm_area_struct *vma,
+                         unsigned long address, pte_t *ptep,
+                         pte_t entry, int dirty)
+{
+       if (!pte_same(ptep_get(ptep), entry))
+               __set_pte_at(ptep, entry);
+       /*
+        * update_mmu_cache will unconditionally execute, handling both
+        * the case that the PTE changed and the spurious fault case.
+        */
+       return true;
+}
+
+int ptep_test_and_clear_young(struct vm_area_struct *vma,
+                             unsigned long address,
+                             pte_t *ptep)
+{
+       if (!pte_young(ptep_get(ptep)))
+               return 0;
+       return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
+}
+EXPORT_SYMBOL_GPL(ptep_test_and_clear_young);
+
+#ifdef CONFIG_64BIT
+pud_t *pud_offset(p4d_t *p4d, unsigned long address)
+{
+       if (pgtable_l4_enabled)
+               return p4d_pgtable(p4dp_get(p4d)) + pud_index(address);
+
+       return (pud_t *)p4d;
+}
+
+p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
+{
+       if (pgtable_l5_enabled)
+               return pgd_pgtable(pgdp_get(pgd)) + p4d_index(address);
+
+       return (p4d_t *)pgd;
+}
+#endif
+
 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
 {
@@ -25,7 +66,7 @@ int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
 
 int pud_clear_huge(pud_t *pud)
 {
-       if (!pud_leaf(READ_ONCE(*pud)))
+       if (!pud_leaf(pudp_get(pud)))
                return 0;
        pud_clear(pud);
        return 1;
@@ -33,7 +74,7 @@ int pud_clear_huge(pud_t *pud)
 
 int pud_free_pmd_page(pud_t *pud, unsigned long addr)
 {
-       pmd_t *pmd = pud_pgtable(*pud);
+       pmd_t *pmd = pud_pgtable(pudp_get(pud));
        int i;
 
        pud_clear(pud);
@@ -63,7 +104,7 @@ int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
 
 int pmd_clear_huge(pmd_t *pmd)
 {
-       if (!pmd_leaf(READ_ONCE(*pmd)))
+       if (!pmd_leaf(pmdp_get(pmd)))
                return 0;
        pmd_clear(pmd);
        return 1;
@@ -71,7 +112,7 @@ int pmd_clear_huge(pmd_t *pmd)
 
 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
 {
-       pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
+       pte_t *pte = (pte_t *)pmd_page_vaddr(pmdp_get(pmd));
 
        pmd_clear(pmd);
 
@@ -88,7 +129,7 @@ pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
        pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
 
        VM_BUG_ON(address & ~HPAGE_PMD_MASK);
-       VM_BUG_ON(pmd_trans_huge(*pmdp));
+       VM_BUG_ON(pmd_trans_huge(pmdp_get(pmdp)));
        /*
         * When leaf PTE entries (regular pages) are collapsed into a leaf
         * PMD entry (huge page), a valid non-leaf PTE is converted into a