Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/i386/mm/pgtable.c | |
3 | */ | |
4 | ||
1da177e4 LT |
5 | #include <linux/sched.h> |
6 | #include <linux/kernel.h> | |
7 | #include <linux/errno.h> | |
8 | #include <linux/mm.h> | |
27eb0b28 | 9 | #include <linux/nmi.h> |
1da177e4 LT |
10 | #include <linux/swap.h> |
11 | #include <linux/smp.h> | |
12 | #include <linux/highmem.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/spinlock.h> | |
052e7994 | 16 | #include <linux/module.h> |
f1d1a842 | 17 | #include <linux/quicklist.h> |
1da177e4 LT |
18 | |
19 | #include <asm/system.h> | |
20 | #include <asm/pgtable.h> | |
21 | #include <asm/pgalloc.h> | |
22 | #include <asm/fixmap.h> | |
23 | #include <asm/e820.h> | |
24 | #include <asm/tlb.h> | |
25 | #include <asm/tlbflush.h> | |
26 | ||
27 | void show_mem(void) | |
28 | { | |
29 | int total = 0, reserved = 0; | |
30 | int shared = 0, cached = 0; | |
31 | int highmem = 0; | |
32 | struct page *page; | |
33 | pg_data_t *pgdat; | |
34 | unsigned long i; | |
208d54e5 | 35 | unsigned long flags; |
1da177e4 | 36 | |
f90e7185 | 37 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 38 | show_free_areas(); |
f90e7185 | 39 | printk(KERN_INFO "Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); |
ec936fc5 | 40 | for_each_online_pgdat(pgdat) { |
208d54e5 | 41 | pgdat_resize_lock(pgdat, &flags); |
1da177e4 | 42 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
27eb0b28 PB |
43 | if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) |
44 | touch_nmi_watchdog(); | |
408fde81 | 45 | page = pgdat_page_nr(pgdat, i); |
1da177e4 LT |
46 | total++; |
47 | if (PageHighMem(page)) | |
48 | highmem++; | |
49 | if (PageReserved(page)) | |
50 | reserved++; | |
51 | else if (PageSwapCache(page)) | |
52 | cached++; | |
53 | else if (page_count(page)) | |
54 | shared += page_count(page) - 1; | |
55 | } | |
208d54e5 | 56 | pgdat_resize_unlock(pgdat, &flags); |
1da177e4 | 57 | } |
f90e7185 CL |
58 | printk(KERN_INFO "%d pages of RAM\n", total); |
59 | printk(KERN_INFO "%d pages of HIGHMEM\n", highmem); | |
60 | printk(KERN_INFO "%d reserved pages\n", reserved); | |
61 | printk(KERN_INFO "%d pages shared\n", shared); | |
62 | printk(KERN_INFO "%d pages swap cached\n", cached); | |
6f4e1e50 | 63 | |
b1e7a8fd | 64 | printk(KERN_INFO "%lu pages dirty\n", global_page_state(NR_FILE_DIRTY)); |
ce866b34 CL |
65 | printk(KERN_INFO "%lu pages writeback\n", |
66 | global_page_state(NR_WRITEBACK)); | |
65ba55f5 | 67 | printk(KERN_INFO "%lu pages mapped\n", global_page_state(NR_FILE_MAPPED)); |
972d1a7b CL |
68 | printk(KERN_INFO "%lu pages slab\n", |
69 | global_page_state(NR_SLAB_RECLAIMABLE) + | |
70 | global_page_state(NR_SLAB_UNRECLAIMABLE)); | |
df849a15 CL |
71 | printk(KERN_INFO "%lu pages pagetables\n", |
72 | global_page_state(NR_PAGETABLE)); | |
1da177e4 LT |
73 | } |
74 | ||
75 | /* | |
76 | * Associate a virtual page frame with a given physical page frame | |
77 | * and protection flags for that frame. | |
78 | */ | |
79 | static void set_pte_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags) | |
80 | { | |
81 | pgd_t *pgd; | |
82 | pud_t *pud; | |
83 | pmd_t *pmd; | |
84 | pte_t *pte; | |
85 | ||
86 | pgd = swapper_pg_dir + pgd_index(vaddr); | |
87 | if (pgd_none(*pgd)) { | |
88 | BUG(); | |
89 | return; | |
90 | } | |
91 | pud = pud_offset(pgd, vaddr); | |
92 | if (pud_none(*pud)) { | |
93 | BUG(); | |
94 | return; | |
95 | } | |
96 | pmd = pmd_offset(pud, vaddr); | |
97 | if (pmd_none(*pmd)) { | |
98 | BUG(); | |
99 | return; | |
100 | } | |
101 | pte = pte_offset_kernel(pmd, vaddr); | |
b0bfece4 | 102 | if (pgprot_val(flags)) |
aa506dc7 | 103 | set_pte_present(&init_mm, vaddr, pte, pfn_pte(pfn, flags)); |
b0bfece4 JB |
104 | else |
105 | pte_clear(&init_mm, vaddr, pte); | |
1da177e4 LT |
106 | |
107 | /* | |
108 | * It's enough to flush this one mapping. | |
109 | * (PGE mappings get flushed as well) | |
110 | */ | |
111 | __flush_tlb_one(vaddr); | |
112 | } | |
113 | ||
114 | /* | |
115 | * Associate a large virtual page frame with a given physical page frame | |
116 | * and protection flags for that frame. pfn is for the base of the page, | |
117 | * vaddr is what the page gets mapped to - both must be properly aligned. | |
118 | * The pmd must already be instantiated. Assumes PAE mode. | |
119 | */ | |
120 | void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags) | |
121 | { | |
122 | pgd_t *pgd; | |
123 | pud_t *pud; | |
124 | pmd_t *pmd; | |
125 | ||
126 | if (vaddr & (PMD_SIZE-1)) { /* vaddr is misaligned */ | |
f90e7185 | 127 | printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n"); |
1da177e4 LT |
128 | return; /* BUG(); */ |
129 | } | |
130 | if (pfn & (PTRS_PER_PTE-1)) { /* pfn is misaligned */ | |
f90e7185 | 131 | printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n"); |
1da177e4 LT |
132 | return; /* BUG(); */ |
133 | } | |
134 | pgd = swapper_pg_dir + pgd_index(vaddr); | |
135 | if (pgd_none(*pgd)) { | |
f90e7185 | 136 | printk(KERN_WARNING "set_pmd_pfn: pgd_none\n"); |
1da177e4 LT |
137 | return; /* BUG(); */ |
138 | } | |
139 | pud = pud_offset(pgd, vaddr); | |
140 | pmd = pmd_offset(pud, vaddr); | |
141 | set_pmd(pmd, pfn_pmd(pfn, flags)); | |
142 | /* | |
143 | * It's enough to flush this one mapping. | |
144 | * (PGE mappings get flushed as well) | |
145 | */ | |
146 | __flush_tlb_one(vaddr); | |
147 | } | |
148 | ||
052e7994 | 149 | static int fixmaps; |
052e7994 JF |
150 | unsigned long __FIXADDR_TOP = 0xfffff000; |
151 | EXPORT_SYMBOL(__FIXADDR_TOP); | |
052e7994 | 152 | |
1da177e4 LT |
153 | void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags) |
154 | { | |
155 | unsigned long address = __fix_to_virt(idx); | |
156 | ||
157 | if (idx >= __end_of_fixed_addresses) { | |
158 | BUG(); | |
159 | return; | |
160 | } | |
161 | set_pte_pfn(address, phys >> PAGE_SHIFT, flags); | |
052e7994 JF |
162 | fixmaps++; |
163 | } | |
164 | ||
165 | /** | |
166 | * reserve_top_address - reserves a hole in the top of kernel address space | |
167 | * @reserve - size of hole to reserve | |
168 | * | |
169 | * Can be used to relocate the fixmap area and poke a hole in the top | |
170 | * of kernel address space to make room for a hypervisor. | |
171 | */ | |
172 | void reserve_top_address(unsigned long reserve) | |
173 | { | |
174 | BUG_ON(fixmaps > 0); | |
7ce0bcfd ZA |
175 | printk(KERN_INFO "Reserving virtual address space above 0x%08x\n", |
176 | (int)-reserve); | |
052e7994 JF |
177 | __FIXADDR_TOP = -reserve - PAGE_SIZE; |
178 | __VMALLOC_RESERVE += reserve; | |
1da177e4 LT |
179 | } |
180 | ||
181 | pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) | |
182 | { | |
183 | return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO); | |
184 | } | |
185 | ||
2f569afd | 186 | pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address) |
1da177e4 LT |
187 | { |
188 | struct page *pte; | |
189 | ||
190 | #ifdef CONFIG_HIGHPTE | |
191 | pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0); | |
192 | #else | |
193 | pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0); | |
194 | #endif | |
2f569afd MS |
195 | if (pte) |
196 | pgtable_page_ctor(pte); | |
1da177e4 LT |
197 | return pte; |
198 | } | |
199 | ||
1da177e4 LT |
200 | /* |
201 | * List of all pgd's needed for non-PAE so it can invalidate entries | |
202 | * in both cached and uncached pgd's; not needed for PAE since the | |
203 | * kernel pmd is shared. If PAE were not to share the pmd a similar | |
204 | * tactic would be needed. This is essentially codepath-based locking | |
205 | * against pageattr.c; it is the unique case in which a valid change | |
206 | * of kernel pagetables can't be lazily synchronized by vmalloc faults. | |
207 | * vmalloc faults work because attached pagetables are never freed. | |
1da177e4 LT |
208 | * -- wli |
209 | */ | |
1da177e4 LT |
210 | static inline void pgd_list_add(pgd_t *pgd) |
211 | { | |
212 | struct page *page = virt_to_page(pgd); | |
e3ed910d JF |
213 | |
214 | list_add(&page->lru, &pgd_list); | |
1da177e4 LT |
215 | } |
216 | ||
217 | static inline void pgd_list_del(pgd_t *pgd) | |
218 | { | |
e3ed910d JF |
219 | struct page *page = virt_to_page(pgd); |
220 | ||
221 | list_del(&page->lru); | |
1da177e4 LT |
222 | } |
223 | ||
e618c957 JF |
224 | #define UNSHARED_PTRS_PER_PGD \ |
225 | (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD) | |
f1d1a842 | 226 | |
e618c957 | 227 | static void pgd_ctor(void *p) |
1da177e4 | 228 | { |
e618c957 | 229 | pgd_t *pgd = p; |
1da177e4 LT |
230 | unsigned long flags; |
231 | ||
e618c957 | 232 | /* Clear usermode parts of PGD */ |
5311ab62 JF |
233 | memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t)); |
234 | ||
235 | spin_lock_irqsave(&pgd_lock, flags); | |
1da177e4 | 236 | |
e618c957 JF |
237 | /* If the pgd points to a shared pagetable level (either the |
238 | ptes in non-PAE, or shared PMD in PAE), then just copy the | |
239 | references from swapper_pg_dir. */ | |
240 | if (PAGETABLE_LEVELS == 2 || | |
241 | (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD)) { | |
242 | clone_pgd_range(pgd + USER_PTRS_PER_PGD, | |
5311ab62 JF |
243 | swapper_pg_dir + USER_PTRS_PER_PGD, |
244 | KERNEL_PGD_PTRS); | |
e618c957 JF |
245 | paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT, |
246 | __pa(swapper_pg_dir) >> PAGE_SHIFT, | |
247 | USER_PTRS_PER_PGD, | |
248 | KERNEL_PGD_PTRS); | |
249 | } | |
5311ab62 | 250 | |
e618c957 JF |
251 | /* list required to sync kernel mapping updates */ |
252 | if (!SHARED_KERNEL_PMD) | |
5311ab62 | 253 | pgd_list_add(pgd); |
e618c957 JF |
254 | |
255 | spin_unlock_irqrestore(&pgd_lock, flags); | |
5311ab62 | 256 | } |
1da177e4 | 257 | |
2378569d | 258 | static void pgd_dtor(void *pgd) |
1da177e4 LT |
259 | { |
260 | unsigned long flags; /* can be called from interrupt context */ | |
261 | ||
f1d1a842 CL |
262 | if (SHARED_KERNEL_PMD) |
263 | return; | |
5311ab62 | 264 | |
1da177e4 LT |
265 | spin_lock_irqsave(&pgd_lock, flags); |
266 | pgd_list_del(pgd); | |
267 | spin_unlock_irqrestore(&pgd_lock, flags); | |
268 | } | |
269 | ||
8fe3deef JF |
270 | #ifdef CONFIG_X86_PAE |
271 | /* | |
272 | * Mop up any pmd pages which may still be attached to the pgd. | |
273 | * Normally they will be freed by munmap/exit_mmap, but any pmd we | |
274 | * preallocate which never got a corresponding vma will need to be | |
275 | * freed manually. | |
276 | */ | |
5e541973 | 277 | static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp) |
8fe3deef JF |
278 | { |
279 | int i; | |
280 | ||
508bebbb | 281 | for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) { |
8fe3deef JF |
282 | pgd_t pgd = pgdp[i]; |
283 | ||
284 | if (pgd_val(pgd) != 0) { | |
285 | pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd); | |
286 | ||
287 | pgdp[i] = native_make_pgd(0); | |
288 | ||
289 | paravirt_release_pd(pgd_val(pgd) >> PAGE_SHIFT); | |
5e541973 | 290 | pmd_free(mm, pmd); |
8fe3deef JF |
291 | } |
292 | } | |
293 | } | |
294 | ||
295 | /* | |
296 | * In PAE mode, we need to do a cr3 reload (=tlb flush) when | |
297 | * updating the top-level pagetable entries to guarantee the | |
298 | * processor notices the update. Since this is expensive, and | |
299 | * all 4 top-level entries are used almost immediately in a | |
300 | * new process's life, we just pre-populate them here. | |
508bebbb JF |
301 | * |
302 | * Also, if we're in a paravirt environment where the kernel pmd is | |
303 | * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate | |
304 | * and initialize the kernel pmds here. | |
8fe3deef JF |
305 | */ |
306 | static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd) | |
307 | { | |
308 | pud_t *pud; | |
309 | unsigned long addr; | |
310 | int i; | |
311 | ||
312 | pud = pud_offset(pgd, 0); | |
508bebbb JF |
313 | for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD; |
314 | i++, pud++, addr += PUD_SIZE) { | |
8fe3deef JF |
315 | pmd_t *pmd = pmd_alloc_one(mm, addr); |
316 | ||
317 | if (!pmd) { | |
5e541973 | 318 | pgd_mop_up_pmds(mm, pgd); |
8fe3deef JF |
319 | return 0; |
320 | } | |
321 | ||
508bebbb JF |
322 | if (i >= USER_PTRS_PER_PGD) |
323 | memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]), | |
324 | sizeof(pmd_t) * PTRS_PER_PMD); | |
325 | ||
8fe3deef JF |
326 | pud_populate(mm, pud, pmd); |
327 | } | |
328 | ||
329 | return 1; | |
330 | } | |
331 | #else /* !CONFIG_X86_PAE */ | |
332 | /* No need to prepopulate any pagetable entries in non-PAE modes. */ | |
333 | static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd) | |
334 | { | |
335 | return 1; | |
336 | } | |
337 | ||
5e541973 | 338 | static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp) |
8fe3deef JF |
339 | { |
340 | } | |
341 | #endif /* CONFIG_X86_PAE */ | |
342 | ||
1da177e4 LT |
343 | pgd_t *pgd_alloc(struct mm_struct *mm) |
344 | { | |
f1d1a842 | 345 | pgd_t *pgd = quicklist_alloc(0, GFP_KERNEL, pgd_ctor); |
1da177e4 | 346 | |
6c435456 JF |
347 | mm->pgd = pgd; /* so that alloc_pd can use it */ |
348 | ||
8fe3deef JF |
349 | if (pgd && !pgd_prepopulate_pmd(mm, pgd)) { |
350 | quicklist_free(0, pgd_dtor, pgd); | |
351 | pgd = NULL; | |
352 | } | |
353 | ||
1da177e4 | 354 | return pgd; |
1da177e4 LT |
355 | } |
356 | ||
5e541973 | 357 | void pgd_free(struct mm_struct *mm, pgd_t *pgd) |
1da177e4 | 358 | { |
5e541973 | 359 | pgd_mop_up_pmds(mm, pgd); |
f1d1a842 | 360 | quicklist_free(0, pgd_dtor, pgd); |
1da177e4 | 361 | } |
f1d1a842 CL |
362 | |
363 | void check_pgt_cache(void) | |
364 | { | |
365 | quicklist_trim(0, pgd_dtor, 25, 16); | |
366 | } | |
5aa05085 IM |
367 | |
368 | void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte) | |
369 | { | |
2f569afd | 370 | pgtable_page_dtor(pte); |
5aa05085 IM |
371 | paravirt_release_pt(page_to_pfn(pte)); |
372 | tlb_remove_page(tlb, pte); | |
373 | } | |
374 | ||
375 | #ifdef CONFIG_X86_PAE | |
376 | ||
377 | void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd) | |
378 | { | |
5aa05085 IM |
379 | paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT); |
380 | tlb_remove_page(tlb, virt_to_page(pmd)); | |
381 | } | |
382 | ||
383 | #endif |