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_pt(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_pd(__pa(pmd) >> PAGE_SHIFT);
35 tlb_remove_page(tlb, virt_to_page(pmd));
37 #endif /* PAGETABLE_LEVELS > 2 */
40 static inline void pgd_list_add(pgd_t *pgd)
42 struct page *page = virt_to_page(pgd);
45 spin_lock_irqsave(&pgd_lock, flags);
46 list_add(&page->lru, &pgd_list);
47 spin_unlock_irqrestore(&pgd_lock, flags);
50 static inline void pgd_list_del(pgd_t *pgd)
52 struct page *page = virt_to_page(pgd);
55 spin_lock_irqsave(&pgd_lock, flags);
57 spin_unlock_irqrestore(&pgd_lock, flags);
60 pgd_t *pgd_alloc(struct mm_struct *mm)
63 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT);
68 * Copy kernel pointers in from init.
69 * Could keep a freelist or slab cache of those because the kernel
72 boundary = pgd_index(__PAGE_OFFSET);
73 memset(pgd, 0, boundary * sizeof(pgd_t));
74 memcpy(pgd + boundary,
75 init_level4_pgt + boundary,
76 (PTRS_PER_PGD - boundary) * sizeof(pgd_t));
80 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
82 BUG_ON((unsigned long)pgd & (PAGE_SIZE-1));
84 free_page((unsigned long)pgd);
88 * List of all pgd's needed for non-PAE so it can invalidate entries
89 * in both cached and uncached pgd's; not needed for PAE since the
90 * kernel pmd is shared. If PAE were not to share the pmd a similar
91 * tactic would be needed. This is essentially codepath-based locking
92 * against pageattr.c; it is the unique case in which a valid change
93 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
94 * vmalloc faults work because attached pagetables are never freed.
97 static inline void pgd_list_add(pgd_t *pgd)
99 struct page *page = virt_to_page(pgd);
101 list_add(&page->lru, &pgd_list);
104 static inline void pgd_list_del(pgd_t *pgd)
106 struct page *page = virt_to_page(pgd);
108 list_del(&page->lru);
111 #define UNSHARED_PTRS_PER_PGD \
112 (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD)
114 static void pgd_ctor(void *p)
119 /* Clear usermode parts of PGD */
120 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
122 spin_lock_irqsave(&pgd_lock, flags);
124 /* If the pgd points to a shared pagetable level (either the
125 ptes in non-PAE, or shared PMD in PAE), then just copy the
126 references from swapper_pg_dir. */
127 if (PAGETABLE_LEVELS == 2 ||
128 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD)) {
129 clone_pgd_range(pgd + USER_PTRS_PER_PGD,
130 swapper_pg_dir + USER_PTRS_PER_PGD,
132 paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT,
133 __pa(swapper_pg_dir) >> PAGE_SHIFT,
138 /* list required to sync kernel mapping updates */
139 if (!SHARED_KERNEL_PMD)
142 spin_unlock_irqrestore(&pgd_lock, flags);
145 static void pgd_dtor(void *pgd)
147 unsigned long flags; /* can be called from interrupt context */
149 if (SHARED_KERNEL_PMD)
152 spin_lock_irqsave(&pgd_lock, flags);
154 spin_unlock_irqrestore(&pgd_lock, flags);
157 #ifdef CONFIG_X86_PAE
159 * Mop up any pmd pages which may still be attached to the pgd.
160 * Normally they will be freed by munmap/exit_mmap, but any pmd we
161 * preallocate which never got a corresponding vma will need to be
164 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
168 for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
171 if (pgd_val(pgd) != 0) {
172 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
174 pgdp[i] = native_make_pgd(0);
176 paravirt_release_pd(pgd_val(pgd) >> PAGE_SHIFT);
183 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
184 * updating the top-level pagetable entries to guarantee the
185 * processor notices the update. Since this is expensive, and
186 * all 4 top-level entries are used almost immediately in a
187 * new process's life, we just pre-populate them here.
189 * Also, if we're in a paravirt environment where the kernel pmd is
190 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
191 * and initialize the kernel pmds here.
193 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
199 pud = pud_offset(pgd, 0);
200 for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
201 i++, pud++, addr += PUD_SIZE) {
202 pmd_t *pmd = pmd_alloc_one(mm, addr);
205 pgd_mop_up_pmds(mm, pgd);
209 if (i >= USER_PTRS_PER_PGD)
210 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
211 sizeof(pmd_t) * PTRS_PER_PMD);
213 pud_populate(mm, pud, pmd);
219 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
221 paravirt_alloc_pd(mm, __pa(pmd) >> PAGE_SHIFT);
223 /* Note: almost everything apart from _PAGE_PRESENT is
224 reserved at the pmd (PDPT) level. */
225 set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
228 * According to Intel App note "TLBs, Paging-Structure Caches,
229 * and Their Invalidation", April 2007, document 317080-001,
230 * section 8.1: in PAE mode we explicitly have to flush the
231 * TLB via cr3 if the top-level pgd is changed...
233 if (mm == current->active_mm)
234 write_cr3(read_cr3());
236 #else /* !CONFIG_X86_PAE */
237 /* No need to prepopulate any pagetable entries in non-PAE modes. */
238 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
243 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd)
246 #endif /* CONFIG_X86_PAE */
248 pgd_t *pgd_alloc(struct mm_struct *mm)
250 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
252 /* so that alloc_pd can use it */
257 if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
259 free_page((unsigned long)pgd);
266 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
268 pgd_mop_up_pmds(mm, pgd);
270 free_page((unsigned long)pgd);