]>
Commit | Line | Data |
---|---|---|
4f76cd38 JF |
1 | #include <linux/mm.h> |
2 | #include <asm/pgalloc.h> | |
3 | #include <asm/tlb.h> | |
4 | ||
5 | pte_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 | ||
10 | pgtable_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 | ||
397f687a JF |
24 | void __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 | ||
170fdff7 JF |
31 | #if PAGETABLE_LEVELS > 2 |
32 | void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd) | |
33 | { | |
34 | paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT); | |
35 | tlb_remove_page(tlb, virt_to_page(pmd)); | |
36 | } | |
5a5f8f42 JF |
37 | |
38 | #if PAGETABLE_LEVELS > 3 | |
39 | void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud) | |
40 | { | |
41 | tlb_remove_page(tlb, virt_to_page(pud)); | |
42 | } | |
43 | #endif /* PAGETABLE_LEVELS > 3 */ | |
170fdff7 JF |
44 | #endif /* PAGETABLE_LEVELS > 2 */ |
45 | ||
4f76cd38 JF |
46 | static inline void pgd_list_add(pgd_t *pgd) |
47 | { | |
48 | struct page *page = virt_to_page(pgd); | |
4f76cd38 | 49 | |
4f76cd38 | 50 | list_add(&page->lru, &pgd_list); |
4f76cd38 JF |
51 | } |
52 | ||
53 | static inline void pgd_list_del(pgd_t *pgd) | |
54 | { | |
55 | struct page *page = virt_to_page(pgd); | |
4f76cd38 | 56 | |
4f76cd38 | 57 | list_del(&page->lru); |
4f76cd38 JF |
58 | } |
59 | ||
39415855 | 60 | #ifdef CONFIG_X86_64 |
4f76cd38 JF |
61 | pgd_t *pgd_alloc(struct mm_struct *mm) |
62 | { | |
63 | unsigned boundary; | |
64 | pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT); | |
39415855 | 65 | unsigned long flags; |
4f76cd38 JF |
66 | if (!pgd) |
67 | return NULL; | |
39415855 | 68 | spin_lock_irqsave(&pgd_lock, flags); |
4f76cd38 | 69 | pgd_list_add(pgd); |
39415855 | 70 | spin_unlock_irqrestore(&pgd_lock, flags); |
4f76cd38 JF |
71 | /* |
72 | * Copy kernel pointers in from init. | |
73 | * Could keep a freelist or slab cache of those because the kernel | |
74 | * part never changes. | |
75 | */ | |
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)); | |
81 | return pgd; | |
82 | } | |
83 | ||
84 | void pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
85 | { | |
39415855 | 86 | unsigned long flags; |
4f76cd38 | 87 | BUG_ON((unsigned long)pgd & (PAGE_SIZE-1)); |
39415855 | 88 | spin_lock_irqsave(&pgd_lock, flags); |
4f76cd38 | 89 | pgd_list_del(pgd); |
39415855 | 90 | spin_unlock_irqrestore(&pgd_lock, flags); |
4f76cd38 JF |
91 | free_page((unsigned long)pgd); |
92 | } | |
93 | #else | |
94 | /* | |
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. | |
102 | * -- wli | |
103 | */ | |
4f76cd38 JF |
104 | #define UNSHARED_PTRS_PER_PGD \ |
105 | (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD) | |
106 | ||
107 | static void pgd_ctor(void *p) | |
108 | { | |
109 | pgd_t *pgd = p; | |
110 | unsigned long flags; | |
111 | ||
112 | /* Clear usermode parts of PGD */ | |
113 | memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t)); | |
114 | ||
115 | spin_lock_irqsave(&pgd_lock, flags); | |
116 | ||
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, | |
124 | KERNEL_PGD_PTRS); | |
125 | paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT, | |
126 | __pa(swapper_pg_dir) >> PAGE_SHIFT, | |
127 | USER_PTRS_PER_PGD, | |
128 | KERNEL_PGD_PTRS); | |
129 | } | |
130 | ||
131 | /* list required to sync kernel mapping updates */ | |
132 | if (!SHARED_KERNEL_PMD) | |
133 | pgd_list_add(pgd); | |
134 | ||
135 | spin_unlock_irqrestore(&pgd_lock, flags); | |
136 | } | |
137 | ||
138 | static void pgd_dtor(void *pgd) | |
139 | { | |
140 | unsigned long flags; /* can be called from interrupt context */ | |
141 | ||
142 | if (SHARED_KERNEL_PMD) | |
143 | return; | |
144 | ||
145 | spin_lock_irqsave(&pgd_lock, flags); | |
146 | pgd_list_del(pgd); | |
147 | spin_unlock_irqrestore(&pgd_lock, flags); | |
148 | } | |
149 | ||
150 | #ifdef CONFIG_X86_PAE | |
151 | /* | |
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 | |
155 | * freed manually. | |
156 | */ | |
157 | static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp) | |
158 | { | |
159 | int i; | |
160 | ||
161 | for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) { | |
162 | pgd_t pgd = pgdp[i]; | |
163 | ||
164 | if (pgd_val(pgd) != 0) { | |
165 | pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd); | |
166 | ||
167 | pgdp[i] = native_make_pgd(0); | |
168 | ||
169 | paravirt_release_pd(pgd_val(pgd) >> PAGE_SHIFT); | |
170 | pmd_free(mm, pmd); | |
171 | } | |
172 | } | |
173 | } | |
174 | ||
175 | /* | |
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. | |
181 | * | |
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. | |
185 | */ | |
186 | static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd) | |
187 | { | |
188 | pud_t *pud; | |
189 | unsigned long addr; | |
190 | int i; | |
191 | ||
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); | |
196 | ||
197 | if (!pmd) { | |
198 | pgd_mop_up_pmds(mm, pgd); | |
199 | return 0; | |
200 | } | |
201 | ||
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); | |
205 | ||
206 | pud_populate(mm, pud, pmd); | |
207 | } | |
208 | ||
209 | return 1; | |
210 | } | |
1ec1fe73 IM |
211 | |
212 | void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd) | |
213 | { | |
214 | paravirt_alloc_pd(mm, __pa(pmd) >> PAGE_SHIFT); | |
215 | ||
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)); | |
219 | ||
220 | /* | |
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... | |
225 | */ | |
226 | if (mm == current->active_mm) | |
227 | write_cr3(read_cr3()); | |
228 | } | |
4f76cd38 JF |
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) | |
232 | { | |
233 | return 1; | |
234 | } | |
235 | ||
236 | static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd) | |
237 | { | |
238 | } | |
239 | #endif /* CONFIG_X86_PAE */ | |
240 | ||
241 | pgd_t *pgd_alloc(struct mm_struct *mm) | |
242 | { | |
243 | pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
244 | ||
245 | /* so that alloc_pd can use it */ | |
246 | mm->pgd = pgd; | |
247 | if (pgd) | |
248 | pgd_ctor(pgd); | |
249 | ||
250 | if (pgd && !pgd_prepopulate_pmd(mm, pgd)) { | |
251 | pgd_dtor(pgd); | |
252 | free_page((unsigned long)pgd); | |
253 | pgd = NULL; | |
254 | } | |
255 | ||
256 | return pgd; | |
257 | } | |
258 | ||
259 | void pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
260 | { | |
261 | pgd_mop_up_pmds(mm, pgd); | |
262 | pgd_dtor(pgd); | |
263 | free_page((unsigned long)pgd); | |
264 | } | |
265 | #endif |