blob: 83765328e2659dbcd2ed2307f10d5ce4181a2c8c [file] [log] [blame]
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -07001#include <linux/mm.h>
2#include <asm/pgalloc.h>
3#include <asm/tlb.h>
4
5pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
6{
7 return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
8}
9
10pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
11{
12 struct page *pte;
13
14#ifdef CONFIG_HIGHPTE
15 pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
16#else
17 pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
18#endif
19 if (pte)
20 pgtable_page_ctor(pte);
21 return pte;
22}
23
Jeremy Fitzhardinge397f6872008-03-17 16:36:57 -070024void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
25{
26 pgtable_page_dtor(pte);
27 paravirt_release_pt(page_to_pfn(pte));
28 tlb_remove_page(tlb, pte);
29}
30
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070031#ifdef CONFIG_X86_64
32static inline void pgd_list_add(pgd_t *pgd)
33{
34 struct page *page = virt_to_page(pgd);
35 unsigned long flags;
36
37 spin_lock_irqsave(&pgd_lock, flags);
38 list_add(&page->lru, &pgd_list);
39 spin_unlock_irqrestore(&pgd_lock, flags);
40}
41
42static inline void pgd_list_del(pgd_t *pgd)
43{
44 struct page *page = virt_to_page(pgd);
45 unsigned long flags;
46
47 spin_lock_irqsave(&pgd_lock, flags);
48 list_del(&page->lru);
49 spin_unlock_irqrestore(&pgd_lock, flags);
50}
51
52pgd_t *pgd_alloc(struct mm_struct *mm)
53{
54 unsigned boundary;
55 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT);
56 if (!pgd)
57 return NULL;
58 pgd_list_add(pgd);
59 /*
60 * Copy kernel pointers in from init.
61 * Could keep a freelist or slab cache of those because the kernel
62 * part never changes.
63 */
64 boundary = pgd_index(__PAGE_OFFSET);
65 memset(pgd, 0, boundary * sizeof(pgd_t));
66 memcpy(pgd + boundary,
67 init_level4_pgt + boundary,
68 (PTRS_PER_PGD - boundary) * sizeof(pgd_t));
69 return pgd;
70}
71
72void pgd_free(struct mm_struct *mm, pgd_t *pgd)
73{
74 BUG_ON((unsigned long)pgd & (PAGE_SIZE-1));
75 pgd_list_del(pgd);
76 free_page((unsigned long)pgd);
77}
78#else
79/*
80 * List of all pgd's needed for non-PAE so it can invalidate entries
81 * in both cached and uncached pgd's; not needed for PAE since the
82 * kernel pmd is shared. If PAE were not to share the pmd a similar
83 * tactic would be needed. This is essentially codepath-based locking
84 * against pageattr.c; it is the unique case in which a valid change
85 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
86 * vmalloc faults work because attached pagetables are never freed.
87 * -- wli
88 */
89static inline void pgd_list_add(pgd_t *pgd)
90{
91 struct page *page = virt_to_page(pgd);
92
93 list_add(&page->lru, &pgd_list);
94}
95
96static inline void pgd_list_del(pgd_t *pgd)
97{
98 struct page *page = virt_to_page(pgd);
99
100 list_del(&page->lru);
101}
102
103#define UNSHARED_PTRS_PER_PGD \
104 (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD)
105
106static void pgd_ctor(void *p)
107{
108 pgd_t *pgd = p;
109 unsigned long flags;
110
111 /* Clear usermode parts of PGD */
112 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
113
114 spin_lock_irqsave(&pgd_lock, flags);
115
116 /* If the pgd points to a shared pagetable level (either the
117 ptes in non-PAE, or shared PMD in PAE), then just copy the
118 references from swapper_pg_dir. */
119 if (PAGETABLE_LEVELS == 2 ||
120 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD)) {
121 clone_pgd_range(pgd + USER_PTRS_PER_PGD,
122 swapper_pg_dir + USER_PTRS_PER_PGD,
123 KERNEL_PGD_PTRS);
124 paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT,
125 __pa(swapper_pg_dir) >> PAGE_SHIFT,
126 USER_PTRS_PER_PGD,
127 KERNEL_PGD_PTRS);
128 }
129
130 /* list required to sync kernel mapping updates */
131 if (!SHARED_KERNEL_PMD)
132 pgd_list_add(pgd);
133
134 spin_unlock_irqrestore(&pgd_lock, flags);
135}
136
137static void pgd_dtor(void *pgd)
138{
139 unsigned long flags; /* can be called from interrupt context */
140
141 if (SHARED_KERNEL_PMD)
142 return;
143
144 spin_lock_irqsave(&pgd_lock, flags);
145 pgd_list_del(pgd);
146 spin_unlock_irqrestore(&pgd_lock, flags);
147}
148
149#ifdef CONFIG_X86_PAE
150/*
151 * Mop up any pmd pages which may still be attached to the pgd.
152 * Normally they will be freed by munmap/exit_mmap, but any pmd we
153 * preallocate which never got a corresponding vma will need to be
154 * freed manually.
155 */
156static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
157{
158 int i;
159
160 for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
161 pgd_t pgd = pgdp[i];
162
163 if (pgd_val(pgd) != 0) {
164 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
165
166 pgdp[i] = native_make_pgd(0);
167
168 paravirt_release_pd(pgd_val(pgd) >> PAGE_SHIFT);
169 pmd_free(mm, pmd);
170 }
171 }
172}
173
174/*
175 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
176 * updating the top-level pagetable entries to guarantee the
177 * processor notices the update. Since this is expensive, and
178 * all 4 top-level entries are used almost immediately in a
179 * new process's life, we just pre-populate them here.
180 *
181 * Also, if we're in a paravirt environment where the kernel pmd is
182 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
183 * and initialize the kernel pmds here.
184 */
185static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
186{
187 pud_t *pud;
188 unsigned long addr;
189 int i;
190
191 pud = pud_offset(pgd, 0);
192 for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
193 i++, pud++, addr += PUD_SIZE) {
194 pmd_t *pmd = pmd_alloc_one(mm, addr);
195
196 if (!pmd) {
197 pgd_mop_up_pmds(mm, pgd);
198 return 0;
199 }
200
201 if (i >= USER_PTRS_PER_PGD)
202 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
203 sizeof(pmd_t) * PTRS_PER_PMD);
204
205 pud_populate(mm, pud, pmd);
206 }
207
208 return 1;
209}
Ingo Molnar1ec1fe72008-03-19 20:30:40 +0100210
211void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
212{
213 paravirt_alloc_pd(mm, __pa(pmd) >> PAGE_SHIFT);
214
215 /* Note: almost everything apart from _PAGE_PRESENT is
216 reserved at the pmd (PDPT) level. */
217 set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
218
219 /*
220 * According to Intel App note "TLBs, Paging-Structure Caches,
221 * and Their Invalidation", April 2007, document 317080-001,
222 * section 8.1: in PAE mode we explicitly have to flush the
223 * TLB via cr3 if the top-level pgd is changed...
224 */
225 if (mm == current->active_mm)
226 write_cr3(read_cr3());
227}
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700228#else /* !CONFIG_X86_PAE */
229/* No need to prepopulate any pagetable entries in non-PAE modes. */
230static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
231{
232 return 1;
233}
234
235static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd)
236{
237}
238#endif /* CONFIG_X86_PAE */
239
240pgd_t *pgd_alloc(struct mm_struct *mm)
241{
242 pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
243
244 /* so that alloc_pd can use it */
245 mm->pgd = pgd;
246 if (pgd)
247 pgd_ctor(pgd);
248
249 if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
250 pgd_dtor(pgd);
251 free_page((unsigned long)pgd);
252 pgd = NULL;
253 }
254
255 return pgd;
256}
257
258void pgd_free(struct mm_struct *mm, pgd_t *pgd)
259{
260 pgd_mop_up_pmds(mm, pgd);
261 pgd_dtor(pgd);
262 free_page((unsigned long)pgd);
263}
264#endif