2 #include <asm/pgalloc.h>
5 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
7 return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
10 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
15 pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
17 pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
20 pgtable_page_ctor(pte);
24 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
26 pgtable_page_dtor(pte);
27 paravirt_release_pte(page_to_pfn(pte));
28 tlb_remove_page(tlb, pte);
31 #if PAGETABLE_LEVELS > 2
32 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
34 paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
35 tlb_remove_page(tlb, virt_to_page(pmd));
38 #if PAGETABLE_LEVELS > 3
39 void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
41 tlb_remove_page(tlb, virt_to_page(pud));
43 #endif /* PAGETABLE_LEVELS > 3 */
44 #endif /* PAGETABLE_LEVELS > 2 */
46 static inline void pgd_list_add(pgd_t *pgd)
48 struct page *page = virt_to_page(pgd);
50 list_add(&page->lru, &pgd_list);
53 static inline void pgd_list_del(pgd_t *pgd)
55 struct page *page = virt_to_page(pgd);
61 pgd_t *pgd_alloc(struct mm_struct *mm)
64 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT);
68 spin_lock_irqsave(&pgd_lock, flags);
70 spin_unlock_irqrestore(&pgd_lock, flags);
72 * Copy kernel pointers in from init.
73 * Could keep a freelist or slab cache of those because the kernel
76 boundary = pgd_index(__PAGE_OFFSET);
77 memset(pgd, 0, boundary * sizeof(pgd_t));
78 memcpy(pgd + boundary,
79 init_level4_pgt + boundary,
80 (PTRS_PER_PGD - boundary) * sizeof(pgd_t));
84 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
87 BUG_ON((unsigned long)pgd & (PAGE_SIZE-1));
88 spin_lock_irqsave(&pgd_lock, flags);
90 spin_unlock_irqrestore(&pgd_lock, flags);
91 free_page((unsigned long)pgd);
95 * List of all pgd's needed for non-PAE so it can invalidate entries
96 * in both cached and uncached pgd's; not needed for PAE since the
97 * kernel pmd is shared. If PAE were not to share the pmd a similar
98 * tactic would be needed. This is essentially codepath-based locking
99 * against pageattr.c; it is the unique case in which a valid change
100 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
101 * vmalloc faults work because attached pagetables are never freed.
104 #define UNSHARED_PTRS_PER_PGD \
105 (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD)
107 static void pgd_ctor(void *p)
112 /* Clear usermode parts of PGD */
113 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
115 spin_lock_irqsave(&pgd_lock, flags);
117 /* If the pgd points to a shared pagetable level (either the
118 ptes in non-PAE, or shared PMD in PAE), then just copy the
119 references from swapper_pg_dir. */
120 if (PAGETABLE_LEVELS == 2 ||
121 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD)) {
122 clone_pgd_range(pgd + USER_PTRS_PER_PGD,
123 swapper_pg_dir + USER_PTRS_PER_PGD,
125 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
126 __pa(swapper_pg_dir) >> PAGE_SHIFT,
131 /* list required to sync kernel mapping updates */
132 if (!SHARED_KERNEL_PMD)
135 spin_unlock_irqrestore(&pgd_lock, flags);
138 static void pgd_dtor(void *pgd)
140 unsigned long flags; /* can be called from interrupt context */
142 if (SHARED_KERNEL_PMD)
145 spin_lock_irqsave(&pgd_lock, flags);
147 spin_unlock_irqrestore(&pgd_lock, flags);
150 #ifdef CONFIG_X86_PAE
152 * Mop up any pmd pages which may still be attached to the pgd.
153 * Normally they will be freed by munmap/exit_mmap, but any pmd we
154 * preallocate which never got a corresponding vma will need to be
157 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
161 for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
164 if (pgd_val(pgd) != 0) {
165 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
167 pgdp[i] = native_make_pgd(0);
169 paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
176 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
177 * updating the top-level pagetable entries to guarantee the
178 * processor notices the update. Since this is expensive, and
179 * all 4 top-level entries are used almost immediately in a
180 * new process's life, we just pre-populate them here.
182 * Also, if we're in a paravirt environment where the kernel pmd is
183 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
184 * and initialize the kernel pmds here.
186 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
192 pud = pud_offset(pgd, 0);
193 for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
194 i++, pud++, addr += PUD_SIZE) {
195 pmd_t *pmd = pmd_alloc_one(mm, addr);
198 pgd_mop_up_pmds(mm, pgd);
202 if (i >= USER_PTRS_PER_PGD)
203 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
204 sizeof(pmd_t) * PTRS_PER_PMD);
206 pud_populate(mm, pud, pmd);
212 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
214 paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
216 /* Note: almost everything apart from _PAGE_PRESENT is
217 reserved at the pmd (PDPT) level. */
218 set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
221 * According to Intel App note "TLBs, Paging-Structure Caches,
222 * and Their Invalidation", April 2007, document 317080-001,
223 * section 8.1: in PAE mode we explicitly have to flush the
224 * TLB via cr3 if the top-level pgd is changed...
226 if (mm == current->active_mm)
227 write_cr3(read_cr3());
229 #else /* !CONFIG_X86_PAE */
230 /* No need to prepopulate any pagetable entries in non-PAE modes. */
231 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
236 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd)
239 #endif /* CONFIG_X86_PAE */
241 pgd_t *pgd_alloc(struct mm_struct *mm)
243 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
245 /* so that alloc_pmd can use it */
250 if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
252 free_page((unsigned long)pgd);
259 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
261 pgd_mop_up_pmds(mm, pgd);
263 free_page((unsigned long)pgd);