Commit | Line | Data |
---|---|---|
4f76cd38 | 1 | #include <linux/mm.h> |
5a0e3ad6 | 2 | #include <linux/gfp.h> |
4f76cd38 | 3 | #include <asm/pgalloc.h> |
ee5aa8d3 | 4 | #include <asm/pgtable.h> |
4f76cd38 | 5 | #include <asm/tlb.h> |
a1d5a869 | 6 | #include <asm/fixmap.h> |
4f76cd38 | 7 | |
9e730237 VN |
8 | #define PGALLOC_GFP GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO |
9 | ||
14315592 IC |
10 | #ifdef CONFIG_HIGHPTE |
11 | #define PGALLOC_USER_GFP __GFP_HIGHMEM | |
12 | #else | |
13 | #define PGALLOC_USER_GFP 0 | |
14 | #endif | |
15 | ||
16 | gfp_t __userpte_alloc_gfp = PGALLOC_GFP | PGALLOC_USER_GFP; | |
17 | ||
4f76cd38 JF |
18 | pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) |
19 | { | |
9e730237 | 20 | return (pte_t *)__get_free_page(PGALLOC_GFP); |
4f76cd38 JF |
21 | } |
22 | ||
23 | pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address) | |
24 | { | |
25 | struct page *pte; | |
26 | ||
14315592 | 27 | pte = alloc_pages(__userpte_alloc_gfp, 0); |
4f76cd38 JF |
28 | if (pte) |
29 | pgtable_page_ctor(pte); | |
30 | return pte; | |
31 | } | |
32 | ||
14315592 IC |
33 | static int __init setup_userpte(char *arg) |
34 | { | |
35 | if (!arg) | |
36 | return -EINVAL; | |
37 | ||
38 | /* | |
39 | * "userpte=nohigh" disables allocation of user pagetables in | |
40 | * high memory. | |
41 | */ | |
42 | if (strcmp(arg, "nohigh") == 0) | |
43 | __userpte_alloc_gfp &= ~__GFP_HIGHMEM; | |
44 | else | |
45 | return -EINVAL; | |
46 | return 0; | |
47 | } | |
48 | early_param("userpte", setup_userpte); | |
49 | ||
9e1b32ca | 50 | void ___pte_free_tlb(struct mmu_gather *tlb, struct page *pte) |
397f687a JF |
51 | { |
52 | pgtable_page_dtor(pte); | |
6944a9c8 | 53 | paravirt_release_pte(page_to_pfn(pte)); |
397f687a JF |
54 | tlb_remove_page(tlb, pte); |
55 | } | |
56 | ||
170fdff7 | 57 | #if PAGETABLE_LEVELS > 2 |
9e1b32ca | 58 | void ___pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd) |
170fdff7 | 59 | { |
6944a9c8 | 60 | paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT); |
1de14c3c DH |
61 | /* |
62 | * NOTE! For PAE, any changes to the top page-directory-pointer-table | |
63 | * entries need a full cr3 reload to flush. | |
64 | */ | |
65 | #ifdef CONFIG_X86_PAE | |
66 | tlb->need_flush_all = 1; | |
67 | #endif | |
170fdff7 JF |
68 | tlb_remove_page(tlb, virt_to_page(pmd)); |
69 | } | |
5a5f8f42 JF |
70 | |
71 | #if PAGETABLE_LEVELS > 3 | |
9e1b32ca | 72 | void ___pud_free_tlb(struct mmu_gather *tlb, pud_t *pud) |
5a5f8f42 | 73 | { |
2761fa09 | 74 | paravirt_release_pud(__pa(pud) >> PAGE_SHIFT); |
5a5f8f42 JF |
75 | tlb_remove_page(tlb, virt_to_page(pud)); |
76 | } | |
77 | #endif /* PAGETABLE_LEVELS > 3 */ | |
170fdff7 JF |
78 | #endif /* PAGETABLE_LEVELS > 2 */ |
79 | ||
4f76cd38 JF |
80 | static inline void pgd_list_add(pgd_t *pgd) |
81 | { | |
82 | struct page *page = virt_to_page(pgd); | |
4f76cd38 | 83 | |
4f76cd38 | 84 | list_add(&page->lru, &pgd_list); |
4f76cd38 JF |
85 | } |
86 | ||
87 | static inline void pgd_list_del(pgd_t *pgd) | |
88 | { | |
89 | struct page *page = virt_to_page(pgd); | |
4f76cd38 | 90 | |
4f76cd38 | 91 | list_del(&page->lru); |
4f76cd38 JF |
92 | } |
93 | ||
4f76cd38 | 94 | #define UNSHARED_PTRS_PER_PGD \ |
68db065c | 95 | (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD) |
4f76cd38 | 96 | |
617d34d9 JF |
97 | |
98 | static void pgd_set_mm(pgd_t *pgd, struct mm_struct *mm) | |
99 | { | |
100 | BUILD_BUG_ON(sizeof(virt_to_page(pgd)->index) < sizeof(mm)); | |
101 | virt_to_page(pgd)->index = (pgoff_t)mm; | |
102 | } | |
103 | ||
104 | struct mm_struct *pgd_page_get_mm(struct page *page) | |
105 | { | |
106 | return (struct mm_struct *)page->index; | |
107 | } | |
108 | ||
109 | static void pgd_ctor(struct mm_struct *mm, pgd_t *pgd) | |
4f76cd38 | 110 | { |
4f76cd38 JF |
111 | /* If the pgd points to a shared pagetable level (either the |
112 | ptes in non-PAE, or shared PMD in PAE), then just copy the | |
113 | references from swapper_pg_dir. */ | |
114 | if (PAGETABLE_LEVELS == 2 || | |
85958b46 JF |
115 | (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) || |
116 | PAGETABLE_LEVELS == 4) { | |
68db065c JF |
117 | clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY, |
118 | swapper_pg_dir + KERNEL_PGD_BOUNDARY, | |
4f76cd38 | 119 | KERNEL_PGD_PTRS); |
4f76cd38 JF |
120 | } |
121 | ||
122 | /* list required to sync kernel mapping updates */ | |
617d34d9 JF |
123 | if (!SHARED_KERNEL_PMD) { |
124 | pgd_set_mm(pgd, mm); | |
4f76cd38 | 125 | pgd_list_add(pgd); |
617d34d9 | 126 | } |
4f76cd38 JF |
127 | } |
128 | ||
17b74627 | 129 | static void pgd_dtor(pgd_t *pgd) |
4f76cd38 | 130 | { |
4f76cd38 JF |
131 | if (SHARED_KERNEL_PMD) |
132 | return; | |
133 | ||
a79e53d8 | 134 | spin_lock(&pgd_lock); |
4f76cd38 | 135 | pgd_list_del(pgd); |
a79e53d8 | 136 | spin_unlock(&pgd_lock); |
4f76cd38 JF |
137 | } |
138 | ||
85958b46 JF |
139 | /* |
140 | * List of all pgd's needed for non-PAE so it can invalidate entries | |
141 | * in both cached and uncached pgd's; not needed for PAE since the | |
142 | * kernel pmd is shared. If PAE were not to share the pmd a similar | |
143 | * tactic would be needed. This is essentially codepath-based locking | |
144 | * against pageattr.c; it is the unique case in which a valid change | |
145 | * of kernel pagetables can't be lazily synchronized by vmalloc faults. | |
146 | * vmalloc faults work because attached pagetables are never freed. | |
6d49e352 | 147 | * -- nyc |
85958b46 JF |
148 | */ |
149 | ||
4f76cd38 | 150 | #ifdef CONFIG_X86_PAE |
d8d5900e JF |
151 | /* |
152 | * In PAE mode, we need to do a cr3 reload (=tlb flush) when | |
153 | * updating the top-level pagetable entries to guarantee the | |
154 | * processor notices the update. Since this is expensive, and | |
155 | * all 4 top-level entries are used almost immediately in a | |
156 | * new process's life, we just pre-populate them here. | |
157 | * | |
158 | * Also, if we're in a paravirt environment where the kernel pmd is | |
159 | * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate | |
160 | * and initialize the kernel pmds here. | |
161 | */ | |
162 | #define PREALLOCATED_PMDS UNSHARED_PTRS_PER_PGD | |
163 | ||
164 | void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd) | |
165 | { | |
166 | paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT); | |
167 | ||
168 | /* Note: almost everything apart from _PAGE_PRESENT is | |
169 | reserved at the pmd (PDPT) level. */ | |
170 | set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT)); | |
171 | ||
172 | /* | |
173 | * According to Intel App note "TLBs, Paging-Structure Caches, | |
174 | * and Their Invalidation", April 2007, document 317080-001, | |
175 | * section 8.1: in PAE mode we explicitly have to flush the | |
176 | * TLB via cr3 if the top-level pgd is changed... | |
177 | */ | |
4981d01e | 178 | flush_tlb_mm(mm); |
d8d5900e JF |
179 | } |
180 | #else /* !CONFIG_X86_PAE */ | |
181 | ||
182 | /* No need to prepopulate any pagetable entries in non-PAE modes. */ | |
183 | #define PREALLOCATED_PMDS 0 | |
184 | ||
185 | #endif /* CONFIG_X86_PAE */ | |
186 | ||
187 | static void free_pmds(pmd_t *pmds[]) | |
188 | { | |
189 | int i; | |
190 | ||
191 | for(i = 0; i < PREALLOCATED_PMDS; i++) | |
192 | if (pmds[i]) | |
193 | free_page((unsigned long)pmds[i]); | |
194 | } | |
195 | ||
196 | static int preallocate_pmds(pmd_t *pmds[]) | |
197 | { | |
198 | int i; | |
199 | bool failed = false; | |
200 | ||
201 | for(i = 0; i < PREALLOCATED_PMDS; i++) { | |
9e730237 | 202 | pmd_t *pmd = (pmd_t *)__get_free_page(PGALLOC_GFP); |
d8d5900e JF |
203 | if (pmd == NULL) |
204 | failed = true; | |
205 | pmds[i] = pmd; | |
206 | } | |
207 | ||
208 | if (failed) { | |
209 | free_pmds(pmds); | |
210 | return -ENOMEM; | |
211 | } | |
212 | ||
213 | return 0; | |
214 | } | |
215 | ||
4f76cd38 JF |
216 | /* |
217 | * Mop up any pmd pages which may still be attached to the pgd. | |
218 | * Normally they will be freed by munmap/exit_mmap, but any pmd we | |
219 | * preallocate which never got a corresponding vma will need to be | |
220 | * freed manually. | |
221 | */ | |
222 | static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp) | |
223 | { | |
224 | int i; | |
225 | ||
d8d5900e | 226 | for(i = 0; i < PREALLOCATED_PMDS; i++) { |
4f76cd38 JF |
227 | pgd_t pgd = pgdp[i]; |
228 | ||
229 | if (pgd_val(pgd) != 0) { | |
230 | pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd); | |
231 | ||
232 | pgdp[i] = native_make_pgd(0); | |
233 | ||
6944a9c8 | 234 | paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT); |
4f76cd38 JF |
235 | pmd_free(mm, pmd); |
236 | } | |
237 | } | |
238 | } | |
239 | ||
d8d5900e | 240 | static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[]) |
4f76cd38 JF |
241 | { |
242 | pud_t *pud; | |
243 | unsigned long addr; | |
244 | int i; | |
245 | ||
cf3e5050 JF |
246 | if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */ |
247 | return; | |
248 | ||
4f76cd38 | 249 | pud = pud_offset(pgd, 0); |
4f76cd38 | 250 | |
d8d5900e JF |
251 | for (addr = i = 0; i < PREALLOCATED_PMDS; |
252 | i++, pud++, addr += PUD_SIZE) { | |
253 | pmd_t *pmd = pmds[i]; | |
4f76cd38 | 254 | |
68db065c | 255 | if (i >= KERNEL_PGD_BOUNDARY) |
4f76cd38 JF |
256 | memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]), |
257 | sizeof(pmd_t) * PTRS_PER_PMD); | |
258 | ||
259 | pud_populate(mm, pud, pmd); | |
260 | } | |
4f76cd38 | 261 | } |
1ec1fe73 | 262 | |
d8d5900e | 263 | pgd_t *pgd_alloc(struct mm_struct *mm) |
1ec1fe73 | 264 | { |
d8d5900e JF |
265 | pgd_t *pgd; |
266 | pmd_t *pmds[PREALLOCATED_PMDS]; | |
1ec1fe73 | 267 | |
9e730237 | 268 | pgd = (pgd_t *)__get_free_page(PGALLOC_GFP); |
d8d5900e JF |
269 | |
270 | if (pgd == NULL) | |
271 | goto out; | |
272 | ||
273 | mm->pgd = pgd; | |
274 | ||
275 | if (preallocate_pmds(pmds) != 0) | |
276 | goto out_free_pgd; | |
277 | ||
278 | if (paravirt_pgd_alloc(mm) != 0) | |
279 | goto out_free_pmds; | |
1ec1fe73 IM |
280 | |
281 | /* | |
d8d5900e JF |
282 | * Make sure that pre-populating the pmds is atomic with |
283 | * respect to anything walking the pgd_list, so that they | |
284 | * never see a partially populated pgd. | |
1ec1fe73 | 285 | */ |
a79e53d8 | 286 | spin_lock(&pgd_lock); |
4f76cd38 | 287 | |
617d34d9 | 288 | pgd_ctor(mm, pgd); |
d8d5900e | 289 | pgd_prepopulate_pmd(mm, pgd, pmds); |
4f76cd38 | 290 | |
a79e53d8 | 291 | spin_unlock(&pgd_lock); |
4f76cd38 JF |
292 | |
293 | return pgd; | |
d8d5900e JF |
294 | |
295 | out_free_pmds: | |
296 | free_pmds(pmds); | |
297 | out_free_pgd: | |
298 | free_page((unsigned long)pgd); | |
299 | out: | |
300 | return NULL; | |
4f76cd38 JF |
301 | } |
302 | ||
303 | void pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
304 | { | |
305 | pgd_mop_up_pmds(mm, pgd); | |
306 | pgd_dtor(pgd); | |
eba0045f | 307 | paravirt_pgd_free(mm, pgd); |
4f76cd38 JF |
308 | free_page((unsigned long)pgd); |
309 | } | |
ee5aa8d3 | 310 | |
0f9a921c RR |
311 | /* |
312 | * Used to set accessed or dirty bits in the page table entries | |
313 | * on other architectures. On x86, the accessed and dirty bits | |
314 | * are tracked by hardware. However, do_wp_page calls this function | |
315 | * to also make the pte writeable at the same time the dirty bit is | |
316 | * set. In that case we do actually need to write the PTE. | |
317 | */ | |
ee5aa8d3 JF |
318 | int ptep_set_access_flags(struct vm_area_struct *vma, |
319 | unsigned long address, pte_t *ptep, | |
320 | pte_t entry, int dirty) | |
321 | { | |
322 | int changed = !pte_same(*ptep, entry); | |
323 | ||
324 | if (changed && dirty) { | |
325 | *ptep = entry; | |
326 | pte_update_defer(vma->vm_mm, address, ptep); | |
ee5aa8d3 JF |
327 | } |
328 | ||
329 | return changed; | |
330 | } | |
f9fbf1a3 | 331 | |
db3eb96f AA |
332 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
333 | int pmdp_set_access_flags(struct vm_area_struct *vma, | |
334 | unsigned long address, pmd_t *pmdp, | |
335 | pmd_t entry, int dirty) | |
336 | { | |
337 | int changed = !pmd_same(*pmdp, entry); | |
338 | ||
339 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); | |
340 | ||
341 | if (changed && dirty) { | |
342 | *pmdp = entry; | |
343 | pmd_update_defer(vma->vm_mm, address, pmdp); | |
5e4bf1a5 IM |
344 | /* |
345 | * We had a write-protection fault here and changed the pmd | |
346 | * to to more permissive. No need to flush the TLB for that, | |
347 | * #PF is architecturally guaranteed to do that and in the | |
348 | * worst-case we'll generate a spurious fault. | |
349 | */ | |
db3eb96f AA |
350 | } |
351 | ||
352 | return changed; | |
353 | } | |
354 | #endif | |
355 | ||
f9fbf1a3 JF |
356 | int ptep_test_and_clear_young(struct vm_area_struct *vma, |
357 | unsigned long addr, pte_t *ptep) | |
358 | { | |
359 | int ret = 0; | |
360 | ||
361 | if (pte_young(*ptep)) | |
362 | ret = test_and_clear_bit(_PAGE_BIT_ACCESSED, | |
48e23957 | 363 | (unsigned long *) &ptep->pte); |
f9fbf1a3 JF |
364 | |
365 | if (ret) | |
366 | pte_update(vma->vm_mm, addr, ptep); | |
367 | ||
368 | return ret; | |
369 | } | |
c20311e1 | 370 | |
db3eb96f AA |
371 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
372 | int pmdp_test_and_clear_young(struct vm_area_struct *vma, | |
373 | unsigned long addr, pmd_t *pmdp) | |
374 | { | |
375 | int ret = 0; | |
376 | ||
377 | if (pmd_young(*pmdp)) | |
378 | ret = test_and_clear_bit(_PAGE_BIT_ACCESSED, | |
f2d6bfe9 | 379 | (unsigned long *)pmdp); |
db3eb96f AA |
380 | |
381 | if (ret) | |
382 | pmd_update(vma->vm_mm, addr, pmdp); | |
383 | ||
384 | return ret; | |
385 | } | |
386 | #endif | |
387 | ||
c20311e1 JF |
388 | int ptep_clear_flush_young(struct vm_area_struct *vma, |
389 | unsigned long address, pte_t *ptep) | |
390 | { | |
391 | int young; | |
392 | ||
393 | young = ptep_test_and_clear_young(vma, address, ptep); | |
394 | if (young) | |
395 | flush_tlb_page(vma, address); | |
396 | ||
397 | return young; | |
398 | } | |
7c7e6e07 | 399 | |
db3eb96f AA |
400 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
401 | int pmdp_clear_flush_young(struct vm_area_struct *vma, | |
402 | unsigned long address, pmd_t *pmdp) | |
403 | { | |
404 | int young; | |
405 | ||
406 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); | |
407 | ||
408 | young = pmdp_test_and_clear_young(vma, address, pmdp); | |
409 | if (young) | |
410 | flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE); | |
411 | ||
412 | return young; | |
413 | } | |
414 | ||
415 | void pmdp_splitting_flush(struct vm_area_struct *vma, | |
416 | unsigned long address, pmd_t *pmdp) | |
417 | { | |
418 | int set; | |
419 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); | |
420 | set = !test_and_set_bit(_PAGE_BIT_SPLITTING, | |
f2d6bfe9 | 421 | (unsigned long *)pmdp); |
db3eb96f AA |
422 | if (set) { |
423 | pmd_update(vma->vm_mm, address, pmdp); | |
424 | /* need tlb flush only to serialize against gup-fast */ | |
425 | flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE); | |
426 | } | |
427 | } | |
428 | #endif | |
429 | ||
fd862dde GP |
430 | /** |
431 | * reserve_top_address - reserves a hole in the top of kernel address space | |
432 | * @reserve - size of hole to reserve | |
433 | * | |
434 | * Can be used to relocate the fixmap area and poke a hole in the top | |
435 | * of kernel address space to make room for a hypervisor. | |
436 | */ | |
437 | void __init reserve_top_address(unsigned long reserve) | |
438 | { | |
439 | #ifdef CONFIG_X86_32 | |
440 | BUG_ON(fixmaps_set > 0); | |
441 | printk(KERN_INFO "Reserving virtual address space above 0x%08x\n", | |
442 | (int)-reserve); | |
443 | __FIXADDR_TOP = -reserve - PAGE_SIZE; | |
fd862dde GP |
444 | #endif |
445 | } | |
446 | ||
7c7e6e07 JF |
447 | int fixmaps_set; |
448 | ||
aeaaa59c | 449 | void __native_set_fixmap(enum fixed_addresses idx, pte_t pte) |
7c7e6e07 JF |
450 | { |
451 | unsigned long address = __fix_to_virt(idx); | |
452 | ||
453 | if (idx >= __end_of_fixed_addresses) { | |
454 | BUG(); | |
455 | return; | |
456 | } | |
aeaaa59c | 457 | set_pte_vaddr(address, pte); |
7c7e6e07 JF |
458 | fixmaps_set++; |
459 | } | |
aeaaa59c | 460 | |
3b3809ac MH |
461 | void native_set_fixmap(enum fixed_addresses idx, phys_addr_t phys, |
462 | pgprot_t flags) | |
aeaaa59c JF |
463 | { |
464 | __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags)); | |
465 | } |