2 #include <asm/pgalloc.h>
3 #include <asm/pgtable.h>
5 #include <asm/fixmap.h>
7 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
9 return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
12 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
17 pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
19 pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
22 pgtable_page_ctor(pte);
26 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
28 pgtable_page_dtor(pte);
29 paravirt_release_pte(page_to_pfn(pte));
30 tlb_remove_page(tlb, pte);
33 #if PAGETABLE_LEVELS > 2
34 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
36 paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
37 tlb_remove_page(tlb, virt_to_page(pmd));
40 #if PAGETABLE_LEVELS > 3
41 void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
43 paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
44 tlb_remove_page(tlb, virt_to_page(pud));
46 #endif /* PAGETABLE_LEVELS > 3 */
47 #endif /* PAGETABLE_LEVELS > 2 */
49 static inline void pgd_list_add(pgd_t *pgd)
51 struct page *page = virt_to_page(pgd);
53 list_add(&page->lru, &pgd_list);
56 static inline void pgd_list_del(pgd_t *pgd)
58 struct page *page = virt_to_page(pgd);
63 #define UNSHARED_PTRS_PER_PGD \
64 (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
66 static void pgd_ctor(void *p)
71 /* Clear usermode parts of PGD */
72 memset(pgd, 0, KERNEL_PGD_BOUNDARY*sizeof(pgd_t));
74 spin_lock_irqsave(&pgd_lock, flags);
76 /* If the pgd points to a shared pagetable level (either the
77 ptes in non-PAE, or shared PMD in PAE), then just copy the
78 references from swapper_pg_dir. */
79 if (PAGETABLE_LEVELS == 2 ||
80 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
81 PAGETABLE_LEVELS == 4) {
82 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
83 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
85 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
86 __pa(swapper_pg_dir) >> PAGE_SHIFT,
91 /* list required to sync kernel mapping updates */
92 if (!SHARED_KERNEL_PMD)
95 spin_unlock_irqrestore(&pgd_lock, flags);
98 static void pgd_dtor(void *pgd)
100 unsigned long flags; /* can be called from interrupt context */
102 if (SHARED_KERNEL_PMD)
105 spin_lock_irqsave(&pgd_lock, flags);
107 spin_unlock_irqrestore(&pgd_lock, flags);
111 * List of all pgd's needed for non-PAE so it can invalidate entries
112 * in both cached and uncached pgd's; not needed for PAE since the
113 * kernel pmd is shared. If PAE were not to share the pmd a similar
114 * tactic would be needed. This is essentially codepath-based locking
115 * against pageattr.c; it is the unique case in which a valid change
116 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
117 * vmalloc faults work because attached pagetables are never freed.
121 #ifdef CONFIG_X86_PAE
123 * Mop up any pmd pages which may still be attached to the pgd.
124 * Normally they will be freed by munmap/exit_mmap, but any pmd we
125 * preallocate which never got a corresponding vma will need to be
128 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
132 for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
135 if (pgd_val(pgd) != 0) {
136 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
138 pgdp[i] = native_make_pgd(0);
140 paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
147 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
148 * updating the top-level pagetable entries to guarantee the
149 * processor notices the update. Since this is expensive, and
150 * all 4 top-level entries are used almost immediately in a
151 * new process's life, we just pre-populate them here.
153 * Also, if we're in a paravirt environment where the kernel pmd is
154 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
155 * and initialize the kernel pmds here.
157 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
163 pud = pud_offset(pgd, 0);
164 for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
165 i++, pud++, addr += PUD_SIZE) {
166 pmd_t *pmd = pmd_alloc_one(mm, addr);
169 pgd_mop_up_pmds(mm, pgd);
173 if (i >= KERNEL_PGD_BOUNDARY)
174 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
175 sizeof(pmd_t) * PTRS_PER_PMD);
177 pud_populate(mm, pud, pmd);
183 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
185 paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
187 /* Note: almost everything apart from _PAGE_PRESENT is
188 reserved at the pmd (PDPT) level. */
189 set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
192 * According to Intel App note "TLBs, Paging-Structure Caches,
193 * and Their Invalidation", April 2007, document 317080-001,
194 * section 8.1: in PAE mode we explicitly have to flush the
195 * TLB via cr3 if the top-level pgd is changed...
197 if (mm == current->active_mm)
198 write_cr3(read_cr3());
200 #else /* !CONFIG_X86_PAE */
201 /* No need to prepopulate any pagetable entries in non-PAE modes. */
202 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
207 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd)
210 #endif /* CONFIG_X86_PAE */
212 pgd_t *pgd_alloc(struct mm_struct *mm)
214 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
216 /* so that alloc_pmd can use it */
221 if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
223 free_page((unsigned long)pgd);
230 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
232 pgd_mop_up_pmds(mm, pgd);
234 free_page((unsigned long)pgd);
237 int ptep_set_access_flags(struct vm_area_struct *vma,
238 unsigned long address, pte_t *ptep,
239 pte_t entry, int dirty)
241 int changed = !pte_same(*ptep, entry);
243 if (changed && dirty) {
245 pte_update_defer(vma->vm_mm, address, ptep);
246 flush_tlb_page(vma, address);
252 int ptep_test_and_clear_young(struct vm_area_struct *vma,
253 unsigned long addr, pte_t *ptep)
257 if (pte_young(*ptep))
258 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
262 pte_update(vma->vm_mm, addr, ptep);
267 int ptep_clear_flush_young(struct vm_area_struct *vma,
268 unsigned long address, pte_t *ptep)
272 young = ptep_test_and_clear_young(vma, address, ptep);
274 flush_tlb_page(vma, address);
281 void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
283 unsigned long address = __fix_to_virt(idx);
285 if (idx >= __end_of_fixed_addresses) {
289 set_pte_vaddr(address, pte);
293 void native_set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
295 __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));