2 #include <asm/pgalloc.h>
3 #include <asm/pgtable.h>
6 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
8 return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
11 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
16 pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
18 pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
21 pgtable_page_ctor(pte);
25 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
27 pgtable_page_dtor(pte);
28 paravirt_release_pte(page_to_pfn(pte));
29 tlb_remove_page(tlb, pte);
32 #if PAGETABLE_LEVELS > 2
33 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
35 paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
36 tlb_remove_page(tlb, virt_to_page(pmd));
39 #if PAGETABLE_LEVELS > 3
40 void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
42 paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
43 tlb_remove_page(tlb, virt_to_page(pud));
45 #endif /* PAGETABLE_LEVELS > 3 */
46 #endif /* PAGETABLE_LEVELS > 2 */
48 static inline void pgd_list_add(pgd_t *pgd)
50 struct page *page = virt_to_page(pgd);
52 list_add(&page->lru, &pgd_list);
55 static inline void pgd_list_del(pgd_t *pgd)
57 struct page *page = virt_to_page(pgd);
62 #define UNSHARED_PTRS_PER_PGD \
63 (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
65 static void pgd_ctor(void *p)
70 /* Clear usermode parts of PGD */
71 memset(pgd, 0, KERNEL_PGD_BOUNDARY*sizeof(pgd_t));
73 spin_lock_irqsave(&pgd_lock, flags);
75 /* If the pgd points to a shared pagetable level (either the
76 ptes in non-PAE, or shared PMD in PAE), then just copy the
77 references from swapper_pg_dir. */
78 if (PAGETABLE_LEVELS == 2 ||
79 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
80 PAGETABLE_LEVELS == 4) {
81 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
82 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
84 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
85 __pa(swapper_pg_dir) >> PAGE_SHIFT,
90 /* list required to sync kernel mapping updates */
91 if (!SHARED_KERNEL_PMD)
94 spin_unlock_irqrestore(&pgd_lock, flags);
97 static void pgd_dtor(void *pgd)
99 unsigned long flags; /* can be called from interrupt context */
101 if (SHARED_KERNEL_PMD)
104 spin_lock_irqsave(&pgd_lock, flags);
106 spin_unlock_irqrestore(&pgd_lock, flags);
110 * List of all pgd's needed for non-PAE so it can invalidate entries
111 * in both cached and uncached pgd's; not needed for PAE since the
112 * kernel pmd is shared. If PAE were not to share the pmd a similar
113 * tactic would be needed. This is essentially codepath-based locking
114 * against pageattr.c; it is the unique case in which a valid change
115 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
116 * vmalloc faults work because attached pagetables are never freed.
120 #ifdef CONFIG_X86_PAE
122 * Mop up any pmd pages which may still be attached to the pgd.
123 * Normally they will be freed by munmap/exit_mmap, but any pmd we
124 * preallocate which never got a corresponding vma will need to be
127 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
131 for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
134 if (pgd_val(pgd) != 0) {
135 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
137 pgdp[i] = native_make_pgd(0);
139 paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
146 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
147 * updating the top-level pagetable entries to guarantee the
148 * processor notices the update. Since this is expensive, and
149 * all 4 top-level entries are used almost immediately in a
150 * new process's life, we just pre-populate them here.
152 * Also, if we're in a paravirt environment where the kernel pmd is
153 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
154 * and initialize the kernel pmds here.
156 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
162 pud = pud_offset(pgd, 0);
163 for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
164 i++, pud++, addr += PUD_SIZE) {
165 pmd_t *pmd = pmd_alloc_one(mm, addr);
168 pgd_mop_up_pmds(mm, pgd);
172 if (i >= KERNEL_PGD_BOUNDARY)
173 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
174 sizeof(pmd_t) * PTRS_PER_PMD);
176 pud_populate(mm, pud, pmd);
182 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
184 paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
186 /* Note: almost everything apart from _PAGE_PRESENT is
187 reserved at the pmd (PDPT) level. */
188 set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
191 * According to Intel App note "TLBs, Paging-Structure Caches,
192 * and Their Invalidation", April 2007, document 317080-001,
193 * section 8.1: in PAE mode we explicitly have to flush the
194 * TLB via cr3 if the top-level pgd is changed...
196 if (mm == current->active_mm)
197 write_cr3(read_cr3());
199 #else /* !CONFIG_X86_PAE */
200 /* No need to prepopulate any pagetable entries in non-PAE modes. */
201 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
206 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd)
209 #endif /* CONFIG_X86_PAE */
211 pgd_t *pgd_alloc(struct mm_struct *mm)
213 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
215 /* so that alloc_pmd can use it */
220 if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
222 free_page((unsigned long)pgd);
229 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
231 pgd_mop_up_pmds(mm, pgd);
233 free_page((unsigned long)pgd);
236 int ptep_set_access_flags(struct vm_area_struct *vma,
237 unsigned long address, pte_t *ptep,
238 pte_t entry, int dirty)
240 int changed = !pte_same(*ptep, entry);
242 if (changed && dirty) {
244 pte_update_defer(vma->vm_mm, address, ptep);
245 flush_tlb_page(vma, address);
251 int ptep_test_and_clear_young(struct vm_area_struct *vma,
252 unsigned long addr, pte_t *ptep)
256 if (pte_young(*ptep))
257 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
261 pte_update(vma->vm_mm, addr, ptep);
266 int ptep_clear_flush_young(struct vm_area_struct *vma,
267 unsigned long address, pte_t *ptep)
271 young = ptep_test_and_clear_young(vma, address, ptep);
273 flush_tlb_page(vma, address);