]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * |
3 | * Copyright (C) 1995 Linus Torvalds | |
4 | * | |
5 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
6 | */ | |
7 | ||
1da177e4 LT |
8 | #include <linux/module.h> |
9 | #include <linux/signal.h> | |
10 | #include <linux/sched.h> | |
11 | #include <linux/kernel.h> | |
12 | #include <linux/errno.h> | |
13 | #include <linux/string.h> | |
14 | #include <linux/types.h> | |
15 | #include <linux/ptrace.h> | |
16 | #include <linux/mman.h> | |
17 | #include <linux/mm.h> | |
18 | #include <linux/hugetlb.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/smp.h> | |
21 | #include <linux/init.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/pagemap.h> | |
cfb80c9e | 24 | #include <linux/pci.h> |
6fb14755 | 25 | #include <linux/pfn.h> |
c9cf5528 | 26 | #include <linux/poison.h> |
1da177e4 LT |
27 | #include <linux/bootmem.h> |
28 | #include <linux/slab.h> | |
29 | #include <linux/proc_fs.h> | |
05039b92 | 30 | #include <linux/memory_hotplug.h> |
27d99f7e | 31 | #include <linux/initrd.h> |
55b2355e | 32 | #include <linux/cpumask.h> |
1da177e4 | 33 | |
f832ff18 | 34 | #include <asm/asm.h> |
46eaa670 | 35 | #include <asm/bios_ebda.h> |
1da177e4 LT |
36 | #include <asm/processor.h> |
37 | #include <asm/system.h> | |
38 | #include <asm/uaccess.h> | |
39 | #include <asm/pgtable.h> | |
40 | #include <asm/dma.h> | |
41 | #include <asm/fixmap.h> | |
42 | #include <asm/e820.h> | |
43 | #include <asm/apic.h> | |
8550eb99 | 44 | #include <asm/bugs.h> |
1da177e4 LT |
45 | #include <asm/tlb.h> |
46 | #include <asm/tlbflush.h> | |
a5a19c63 | 47 | #include <asm/pgalloc.h> |
1da177e4 | 48 | #include <asm/sections.h> |
b239fb25 | 49 | #include <asm/paravirt.h> |
551889a6 | 50 | #include <asm/setup.h> |
7bfeab9a | 51 | #include <asm/cacheflush.h> |
1da177e4 LT |
52 | |
53 | unsigned int __VMALLOC_RESERVE = 128 << 20; | |
54 | ||
f361a450 | 55 | unsigned long max_low_pfn_mapped; |
67794292 | 56 | unsigned long max_pfn_mapped; |
7d1116a9 | 57 | |
1da177e4 LT |
58 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
59 | unsigned long highstart_pfn, highend_pfn; | |
60 | ||
8550eb99 | 61 | static noinline int do_test_wp_bit(void); |
1da177e4 | 62 | |
4e29684c YL |
63 | |
64 | static unsigned long __initdata table_start; | |
65 | static unsigned long __meminitdata table_end; | |
66 | static unsigned long __meminitdata table_top; | |
67 | ||
68 | static int __initdata after_init_bootmem; | |
69 | ||
d6be89ad | 70 | static __init void *alloc_low_page(void) |
4e29684c YL |
71 | { |
72 | unsigned long pfn = table_end++; | |
73 | void *adr; | |
74 | ||
75 | if (pfn >= table_top) | |
76 | panic("alloc_low_page: ran out of memory"); | |
77 | ||
78 | adr = __va(pfn * PAGE_SIZE); | |
79 | memset(adr, 0, PAGE_SIZE); | |
4e29684c YL |
80 | return adr; |
81 | } | |
82 | ||
1da177e4 LT |
83 | /* |
84 | * Creates a middle page table and puts a pointer to it in the | |
85 | * given global directory entry. This only returns the gd entry | |
86 | * in non-PAE compilation mode, since the middle layer is folded. | |
87 | */ | |
88 | static pmd_t * __init one_md_table_init(pgd_t *pgd) | |
89 | { | |
90 | pud_t *pud; | |
91 | pmd_t *pmd_table; | |
8550eb99 | 92 | |
1da177e4 | 93 | #ifdef CONFIG_X86_PAE |
b239fb25 | 94 | if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { |
4e29684c YL |
95 | if (after_init_bootmem) |
96 | pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE); | |
97 | else | |
d6be89ad | 98 | pmd_table = (pmd_t *)alloc_low_page(); |
6944a9c8 | 99 | paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT); |
b239fb25 JF |
100 | set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); |
101 | pud = pud_offset(pgd, 0); | |
8550eb99 | 102 | BUG_ON(pmd_table != pmd_offset(pud, 0)); |
a376f30a Z |
103 | |
104 | return pmd_table; | |
b239fb25 JF |
105 | } |
106 | #endif | |
1da177e4 LT |
107 | pud = pud_offset(pgd, 0); |
108 | pmd_table = pmd_offset(pud, 0); | |
8550eb99 | 109 | |
1da177e4 LT |
110 | return pmd_table; |
111 | } | |
112 | ||
113 | /* | |
114 | * Create a page table and place a pointer to it in a middle page | |
8550eb99 | 115 | * directory entry: |
1da177e4 LT |
116 | */ |
117 | static pte_t * __init one_page_table_init(pmd_t *pmd) | |
118 | { | |
b239fb25 | 119 | if (!(pmd_val(*pmd) & _PAGE_PRESENT)) { |
509a80c4 IM |
120 | pte_t *page_table = NULL; |
121 | ||
4e29684c | 122 | if (after_init_bootmem) { |
509a80c4 | 123 | #ifdef CONFIG_DEBUG_PAGEALLOC |
4e29684c | 124 | page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE); |
509a80c4 | 125 | #endif |
4e29684c YL |
126 | if (!page_table) |
127 | page_table = | |
509a80c4 | 128 | (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE); |
d6be89ad JB |
129 | } else |
130 | page_table = (pte_t *)alloc_low_page(); | |
b239fb25 | 131 | |
6944a9c8 | 132 | paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT); |
1da177e4 | 133 | set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); |
b239fb25 | 134 | BUG_ON(page_table != pte_offset_kernel(pmd, 0)); |
1da177e4 | 135 | } |
509a80c4 | 136 | |
1da177e4 LT |
137 | return pte_offset_kernel(pmd, 0); |
138 | } | |
139 | ||
11124411 TH |
140 | void __init populate_extra_pte(unsigned long vaddr) |
141 | { | |
142 | int pgd_idx = pgd_index(vaddr); | |
143 | int pmd_idx = pmd_index(vaddr); | |
144 | pmd_t *pmd; | |
145 | ||
146 | pmd = one_md_table_init(swapper_pg_dir + pgd_idx); | |
147 | one_page_table_init(pmd + pmd_idx); | |
148 | } | |
149 | ||
a3c6018e JB |
150 | static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, |
151 | unsigned long vaddr, pte_t *lastpte) | |
152 | { | |
153 | #ifdef CONFIG_HIGHMEM | |
154 | /* | |
155 | * Something (early fixmap) may already have put a pte | |
156 | * page here, which causes the page table allocation | |
157 | * to become nonlinear. Attempt to fix it, and if it | |
158 | * is still nonlinear then we have to bug. | |
159 | */ | |
160 | int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; | |
161 | int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; | |
162 | ||
163 | if (pmd_idx_kmap_begin != pmd_idx_kmap_end | |
164 | && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin | |
165 | && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end | |
166 | && ((__pa(pte) >> PAGE_SHIFT) < table_start | |
167 | || (__pa(pte) >> PAGE_SHIFT) >= table_end)) { | |
168 | pte_t *newpte; | |
169 | int i; | |
170 | ||
171 | BUG_ON(after_init_bootmem); | |
172 | newpte = alloc_low_page(); | |
173 | for (i = 0; i < PTRS_PER_PTE; i++) | |
174 | set_pte(newpte + i, pte[i]); | |
175 | ||
176 | paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT); | |
177 | set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); | |
178 | BUG_ON(newpte != pte_offset_kernel(pmd, 0)); | |
179 | __flush_tlb_all(); | |
180 | ||
181 | paravirt_release_pte(__pa(pte) >> PAGE_SHIFT); | |
182 | pte = newpte; | |
183 | } | |
184 | BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1) | |
185 | && vaddr > fix_to_virt(FIX_KMAP_END) | |
186 | && lastpte && lastpte + PTRS_PER_PTE != pte); | |
187 | #endif | |
188 | return pte; | |
189 | } | |
190 | ||
1da177e4 | 191 | /* |
8550eb99 | 192 | * This function initializes a certain range of kernel virtual memory |
1da177e4 LT |
193 | * with new bootmem page tables, everywhere page tables are missing in |
194 | * the given range. | |
8550eb99 IM |
195 | * |
196 | * NOTE: The pagetables are allocated contiguous on the physical space | |
197 | * so we can cache the place of the first one and move around without | |
1da177e4 LT |
198 | * checking the pgd every time. |
199 | */ | |
8550eb99 IM |
200 | static void __init |
201 | page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) | |
1da177e4 | 202 | { |
1da177e4 LT |
203 | int pgd_idx, pmd_idx; |
204 | unsigned long vaddr; | |
8550eb99 IM |
205 | pgd_t *pgd; |
206 | pmd_t *pmd; | |
a3c6018e | 207 | pte_t *pte = NULL; |
1da177e4 LT |
208 | |
209 | vaddr = start; | |
210 | pgd_idx = pgd_index(vaddr); | |
211 | pmd_idx = pmd_index(vaddr); | |
212 | pgd = pgd_base + pgd_idx; | |
213 | ||
214 | for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { | |
b239fb25 JF |
215 | pmd = one_md_table_init(pgd); |
216 | pmd = pmd + pmd_index(vaddr); | |
8550eb99 IM |
217 | for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); |
218 | pmd++, pmd_idx++) { | |
a3c6018e JB |
219 | pte = page_table_kmap_check(one_page_table_init(pmd), |
220 | pmd, vaddr, pte); | |
1da177e4 LT |
221 | |
222 | vaddr += PMD_SIZE; | |
223 | } | |
224 | pmd_idx = 0; | |
225 | } | |
226 | } | |
227 | ||
228 | static inline int is_kernel_text(unsigned long addr) | |
229 | { | |
230 | if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end) | |
231 | return 1; | |
232 | return 0; | |
233 | } | |
234 | ||
235 | /* | |
8550eb99 IM |
236 | * This maps the physical memory to kernel virtual address space, a total |
237 | * of max_low_pfn pages, by creating page tables starting from address | |
238 | * PAGE_OFFSET: | |
1da177e4 | 239 | */ |
4e29684c | 240 | static void __init kernel_physical_mapping_init(pgd_t *pgd_base, |
a04ad82d YL |
241 | unsigned long start_pfn, |
242 | unsigned long end_pfn, | |
243 | int use_pse) | |
1da177e4 | 244 | { |
8550eb99 | 245 | int pgd_idx, pmd_idx, pte_ofs; |
1da177e4 LT |
246 | unsigned long pfn; |
247 | pgd_t *pgd; | |
248 | pmd_t *pmd; | |
249 | pte_t *pte; | |
a2699e47 SS |
250 | unsigned pages_2m, pages_4k; |
251 | int mapping_iter; | |
252 | ||
253 | /* | |
254 | * First iteration will setup identity mapping using large/small pages | |
255 | * based on use_pse, with other attributes same as set by | |
256 | * the early code in head_32.S | |
257 | * | |
258 | * Second iteration will setup the appropriate attributes (NX, GLOBAL..) | |
259 | * as desired for the kernel identity mapping. | |
260 | * | |
261 | * This two pass mechanism conforms to the TLB app note which says: | |
262 | * | |
263 | * "Software should not write to a paging-structure entry in a way | |
264 | * that would change, for any linear address, both the page size | |
265 | * and either the page frame or attributes." | |
266 | */ | |
267 | mapping_iter = 1; | |
1da177e4 | 268 | |
a04ad82d YL |
269 | if (!cpu_has_pse) |
270 | use_pse = 0; | |
1da177e4 | 271 | |
a2699e47 SS |
272 | repeat: |
273 | pages_2m = pages_4k = 0; | |
a04ad82d YL |
274 | pfn = start_pfn; |
275 | pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); | |
276 | pgd = pgd_base + pgd_idx; | |
1da177e4 LT |
277 | for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { |
278 | pmd = one_md_table_init(pgd); | |
8550eb99 | 279 | |
a04ad82d YL |
280 | if (pfn >= end_pfn) |
281 | continue; | |
282 | #ifdef CONFIG_X86_PAE | |
283 | pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); | |
284 | pmd += pmd_idx; | |
285 | #else | |
286 | pmd_idx = 0; | |
287 | #endif | |
288 | for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn; | |
f3f20de8 | 289 | pmd++, pmd_idx++) { |
8550eb99 | 290 | unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; |
1da177e4 | 291 | |
8550eb99 IM |
292 | /* |
293 | * Map with big pages if possible, otherwise | |
294 | * create normal page tables: | |
295 | */ | |
a04ad82d | 296 | if (use_pse) { |
8550eb99 | 297 | unsigned int addr2; |
f3f20de8 | 298 | pgprot_t prot = PAGE_KERNEL_LARGE; |
a2699e47 SS |
299 | /* |
300 | * first pass will use the same initial | |
301 | * identity mapping attribute + _PAGE_PSE. | |
302 | */ | |
303 | pgprot_t init_prot = | |
304 | __pgprot(PTE_IDENT_ATTR | | |
305 | _PAGE_PSE); | |
f3f20de8 | 306 | |
8550eb99 | 307 | addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + |
f3f20de8 JF |
308 | PAGE_OFFSET + PAGE_SIZE-1; |
309 | ||
8550eb99 IM |
310 | if (is_kernel_text(addr) || |
311 | is_kernel_text(addr2)) | |
f3f20de8 JF |
312 | prot = PAGE_KERNEL_LARGE_EXEC; |
313 | ||
ce0c0e50 | 314 | pages_2m++; |
a2699e47 SS |
315 | if (mapping_iter == 1) |
316 | set_pmd(pmd, pfn_pmd(pfn, init_prot)); | |
317 | else | |
318 | set_pmd(pmd, pfn_pmd(pfn, prot)); | |
b239fb25 | 319 | |
1da177e4 | 320 | pfn += PTRS_PER_PTE; |
8550eb99 IM |
321 | continue; |
322 | } | |
323 | pte = one_page_table_init(pmd); | |
1da177e4 | 324 | |
a04ad82d YL |
325 | pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); |
326 | pte += pte_ofs; | |
327 | for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; | |
8550eb99 IM |
328 | pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { |
329 | pgprot_t prot = PAGE_KERNEL; | |
a2699e47 SS |
330 | /* |
331 | * first pass will use the same initial | |
332 | * identity mapping attribute. | |
333 | */ | |
334 | pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); | |
f3f20de8 | 335 | |
8550eb99 IM |
336 | if (is_kernel_text(addr)) |
337 | prot = PAGE_KERNEL_EXEC; | |
f3f20de8 | 338 | |
ce0c0e50 | 339 | pages_4k++; |
a2699e47 SS |
340 | if (mapping_iter == 1) |
341 | set_pte(pte, pfn_pte(pfn, init_prot)); | |
342 | else | |
343 | set_pte(pte, pfn_pte(pfn, prot)); | |
1da177e4 LT |
344 | } |
345 | } | |
346 | } | |
a2699e47 SS |
347 | if (mapping_iter == 1) { |
348 | /* | |
349 | * update direct mapping page count only in the first | |
350 | * iteration. | |
351 | */ | |
352 | update_page_count(PG_LEVEL_2M, pages_2m); | |
353 | update_page_count(PG_LEVEL_4K, pages_4k); | |
354 | ||
355 | /* | |
356 | * local global flush tlb, which will flush the previous | |
357 | * mappings present in both small and large page TLB's. | |
358 | */ | |
359 | __flush_tlb_all(); | |
360 | ||
361 | /* | |
362 | * Second iteration will set the actual desired PTE attributes. | |
363 | */ | |
364 | mapping_iter = 2; | |
365 | goto repeat; | |
366 | } | |
1da177e4 LT |
367 | } |
368 | ||
ae531c26 AV |
369 | /* |
370 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
371 | * is valid. The argument is a physical page number. | |
372 | * | |
373 | * | |
374 | * On x86, access has to be given to the first megabyte of ram because that area | |
375 | * contains bios code and data regions used by X and dosemu and similar apps. | |
376 | * Access has to be given to non-kernel-ram areas as well, these contain the PCI | |
377 | * mmio resources as well as potential bios/acpi data regions. | |
378 | */ | |
379 | int devmem_is_allowed(unsigned long pagenr) | |
380 | { | |
381 | if (pagenr <= 256) | |
382 | return 1; | |
e8de1481 AV |
383 | if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) |
384 | return 0; | |
ae531c26 AV |
385 | if (!page_is_ram(pagenr)) |
386 | return 1; | |
387 | return 0; | |
388 | } | |
389 | ||
1da177e4 LT |
390 | pte_t *kmap_pte; |
391 | pgprot_t kmap_prot; | |
392 | ||
8550eb99 IM |
393 | static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) |
394 | { | |
395 | return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), | |
396 | vaddr), vaddr), vaddr); | |
397 | } | |
1da177e4 LT |
398 | |
399 | static void __init kmap_init(void) | |
400 | { | |
401 | unsigned long kmap_vstart; | |
402 | ||
8550eb99 IM |
403 | /* |
404 | * Cache the first kmap pte: | |
405 | */ | |
1da177e4 LT |
406 | kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); |
407 | kmap_pte = kmap_get_fixmap_pte(kmap_vstart); | |
408 | ||
409 | kmap_prot = PAGE_KERNEL; | |
410 | } | |
411 | ||
fd940934 | 412 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
413 | static void __init permanent_kmaps_init(pgd_t *pgd_base) |
414 | { | |
8550eb99 | 415 | unsigned long vaddr; |
1da177e4 LT |
416 | pgd_t *pgd; |
417 | pud_t *pud; | |
418 | pmd_t *pmd; | |
419 | pte_t *pte; | |
1da177e4 LT |
420 | |
421 | vaddr = PKMAP_BASE; | |
422 | page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); | |
423 | ||
424 | pgd = swapper_pg_dir + pgd_index(vaddr); | |
425 | pud = pud_offset(pgd, vaddr); | |
426 | pmd = pmd_offset(pud, vaddr); | |
427 | pte = pte_offset_kernel(pmd, vaddr); | |
8550eb99 | 428 | pkmap_page_table = pte; |
1da177e4 LT |
429 | } |
430 | ||
cc9f7a0c | 431 | static void __init add_one_highpage_init(struct page *page, int pfn) |
1da177e4 | 432 | { |
cc9f7a0c YL |
433 | ClearPageReserved(page); |
434 | init_page_count(page); | |
435 | __free_page(page); | |
436 | totalhigh_pages++; | |
1da177e4 LT |
437 | } |
438 | ||
b5bc6c0e YL |
439 | struct add_highpages_data { |
440 | unsigned long start_pfn; | |
441 | unsigned long end_pfn; | |
b5bc6c0e YL |
442 | }; |
443 | ||
d52d53b8 | 444 | static int __init add_highpages_work_fn(unsigned long start_pfn, |
b5bc6c0e | 445 | unsigned long end_pfn, void *datax) |
1da177e4 | 446 | { |
b5bc6c0e YL |
447 | int node_pfn; |
448 | struct page *page; | |
449 | unsigned long final_start_pfn, final_end_pfn; | |
450 | struct add_highpages_data *data; | |
8550eb99 | 451 | |
b5bc6c0e | 452 | data = (struct add_highpages_data *)datax; |
b5bc6c0e YL |
453 | |
454 | final_start_pfn = max(start_pfn, data->start_pfn); | |
455 | final_end_pfn = min(end_pfn, data->end_pfn); | |
456 | if (final_start_pfn >= final_end_pfn) | |
d52d53b8 | 457 | return 0; |
b5bc6c0e YL |
458 | |
459 | for (node_pfn = final_start_pfn; node_pfn < final_end_pfn; | |
460 | node_pfn++) { | |
461 | if (!pfn_valid(node_pfn)) | |
462 | continue; | |
463 | page = pfn_to_page(node_pfn); | |
cc9f7a0c | 464 | add_one_highpage_init(page, node_pfn); |
23be8c7d | 465 | } |
b5bc6c0e | 466 | |
d52d53b8 YL |
467 | return 0; |
468 | ||
b5bc6c0e YL |
469 | } |
470 | ||
471 | void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn, | |
cc9f7a0c | 472 | unsigned long end_pfn) |
b5bc6c0e YL |
473 | { |
474 | struct add_highpages_data data; | |
475 | ||
476 | data.start_pfn = start_pfn; | |
477 | data.end_pfn = end_pfn; | |
b5bc6c0e YL |
478 | |
479 | work_with_active_regions(nid, add_highpages_work_fn, &data); | |
480 | } | |
481 | ||
8550eb99 | 482 | #ifndef CONFIG_NUMA |
cc9f7a0c | 483 | static void __init set_highmem_pages_init(void) |
1da177e4 | 484 | { |
cc9f7a0c | 485 | add_highpages_with_active_regions(0, highstart_pfn, highend_pfn); |
b5bc6c0e | 486 | |
1da177e4 LT |
487 | totalram_pages += totalhigh_pages; |
488 | } | |
8550eb99 | 489 | #endif /* !CONFIG_NUMA */ |
1da177e4 LT |
490 | |
491 | #else | |
e8e32326 IB |
492 | static inline void permanent_kmaps_init(pgd_t *pgd_base) |
493 | { | |
494 | } | |
495 | static inline void set_highmem_pages_init(void) | |
496 | { | |
497 | } | |
1da177e4 LT |
498 | #endif /* CONFIG_HIGHMEM */ |
499 | ||
b239fb25 | 500 | void __init native_pagetable_setup_start(pgd_t *base) |
1da177e4 | 501 | { |
551889a6 IC |
502 | unsigned long pfn, va; |
503 | pgd_t *pgd; | |
504 | pud_t *pud; | |
505 | pmd_t *pmd; | |
506 | pte_t *pte; | |
b239fb25 JF |
507 | |
508 | /* | |
551889a6 IC |
509 | * Remove any mappings which extend past the end of physical |
510 | * memory from the boot time page table: | |
b239fb25 | 511 | */ |
551889a6 IC |
512 | for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) { |
513 | va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); | |
514 | pgd = base + pgd_index(va); | |
515 | if (!pgd_present(*pgd)) | |
516 | break; | |
517 | ||
518 | pud = pud_offset(pgd, va); | |
519 | pmd = pmd_offset(pud, va); | |
520 | if (!pmd_present(*pmd)) | |
521 | break; | |
522 | ||
523 | pte = pte_offset_kernel(pmd, va); | |
524 | if (!pte_present(*pte)) | |
525 | break; | |
526 | ||
527 | pte_clear(NULL, va, pte); | |
528 | } | |
6944a9c8 | 529 | paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT); |
b239fb25 JF |
530 | } |
531 | ||
532 | void __init native_pagetable_setup_done(pgd_t *base) | |
533 | { | |
b239fb25 JF |
534 | } |
535 | ||
536 | /* | |
537 | * Build a proper pagetable for the kernel mappings. Up until this | |
538 | * point, we've been running on some set of pagetables constructed by | |
539 | * the boot process. | |
540 | * | |
541 | * If we're booting on native hardware, this will be a pagetable | |
551889a6 IC |
542 | * constructed in arch/x86/kernel/head_32.S. The root of the |
543 | * pagetable will be swapper_pg_dir. | |
b239fb25 JF |
544 | * |
545 | * If we're booting paravirtualized under a hypervisor, then there are | |
546 | * more options: we may already be running PAE, and the pagetable may | |
547 | * or may not be based in swapper_pg_dir. In any case, | |
548 | * paravirt_pagetable_setup_start() will set up swapper_pg_dir | |
549 | * appropriately for the rest of the initialization to work. | |
550 | * | |
551 | * In general, pagetable_init() assumes that the pagetable may already | |
552 | * be partially populated, and so it avoids stomping on any existing | |
553 | * mappings. | |
554 | */ | |
e7b37895 | 555 | static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base) |
b239fb25 | 556 | { |
8550eb99 | 557 | unsigned long vaddr, end; |
b239fb25 | 558 | |
1da177e4 LT |
559 | /* |
560 | * Fixed mappings, only the page table structure has to be | |
561 | * created - mappings will be set by set_fixmap(): | |
562 | */ | |
563 | vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; | |
b239fb25 JF |
564 | end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; |
565 | page_table_range_init(vaddr, end, pgd_base); | |
beacfaac | 566 | early_ioremap_reset(); |
e7b37895 YL |
567 | } |
568 | ||
569 | static void __init pagetable_init(void) | |
570 | { | |
571 | pgd_t *pgd_base = swapper_pg_dir; | |
572 | ||
1da177e4 | 573 | permanent_kmaps_init(pgd_base); |
1da177e4 LT |
574 | } |
575 | ||
a6eb84bc | 576 | #ifdef CONFIG_ACPI_SLEEP |
1da177e4 | 577 | /* |
a6eb84bc | 578 | * ACPI suspend needs this for resume, because things like the intel-agp |
1da177e4 LT |
579 | * driver might have split up a kernel 4MB mapping. |
580 | */ | |
a6eb84bc | 581 | char swsusp_pg_dir[PAGE_SIZE] |
8550eb99 | 582 | __attribute__ ((aligned(PAGE_SIZE))); |
1da177e4 LT |
583 | |
584 | static inline void save_pg_dir(void) | |
585 | { | |
586 | memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE); | |
587 | } | |
a6eb84bc | 588 | #else /* !CONFIG_ACPI_SLEEP */ |
1da177e4 LT |
589 | static inline void save_pg_dir(void) |
590 | { | |
591 | } | |
a6eb84bc | 592 | #endif /* !CONFIG_ACPI_SLEEP */ |
1da177e4 | 593 | |
8550eb99 | 594 | void zap_low_mappings(void) |
1da177e4 LT |
595 | { |
596 | int i; | |
597 | ||
1da177e4 LT |
598 | /* |
599 | * Zap initial low-memory mappings. | |
600 | * | |
601 | * Note that "pgd_clear()" doesn't do it for | |
602 | * us, because pgd_clear() is a no-op on i386. | |
603 | */ | |
68db065c | 604 | for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) { |
1da177e4 LT |
605 | #ifdef CONFIG_X86_PAE |
606 | set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page))); | |
607 | #else | |
608 | set_pgd(swapper_pg_dir+i, __pgd(0)); | |
609 | #endif | |
8550eb99 | 610 | } |
1da177e4 LT |
611 | flush_tlb_all(); |
612 | } | |
613 | ||
8550eb99 | 614 | int nx_enabled; |
d5321abe | 615 | |
be43d728 | 616 | pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP); |
6fdc05d4 JF |
617 | EXPORT_SYMBOL_GPL(__supported_pte_mask); |
618 | ||
d5321abe JB |
619 | #ifdef CONFIG_X86_PAE |
620 | ||
8550eb99 | 621 | static int disable_nx __initdata; |
1da177e4 LT |
622 | |
623 | /* | |
624 | * noexec = on|off | |
625 | * | |
626 | * Control non executable mappings. | |
627 | * | |
628 | * on Enable | |
629 | * off Disable | |
630 | */ | |
1a3f239d | 631 | static int __init noexec_setup(char *str) |
1da177e4 | 632 | { |
1a3f239d RR |
633 | if (!str || !strcmp(str, "on")) { |
634 | if (cpu_has_nx) { | |
635 | __supported_pte_mask |= _PAGE_NX; | |
636 | disable_nx = 0; | |
637 | } | |
8550eb99 IM |
638 | } else { |
639 | if (!strcmp(str, "off")) { | |
640 | disable_nx = 1; | |
641 | __supported_pte_mask &= ~_PAGE_NX; | |
642 | } else { | |
643 | return -EINVAL; | |
644 | } | |
645 | } | |
1a3f239d RR |
646 | |
647 | return 0; | |
1da177e4 | 648 | } |
1a3f239d | 649 | early_param("noexec", noexec_setup); |
1da177e4 | 650 | |
1da177e4 LT |
651 | static void __init set_nx(void) |
652 | { | |
653 | unsigned int v[4], l, h; | |
654 | ||
655 | if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) { | |
656 | cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]); | |
8550eb99 | 657 | |
1da177e4 LT |
658 | if ((v[3] & (1 << 20)) && !disable_nx) { |
659 | rdmsr(MSR_EFER, l, h); | |
660 | l |= EFER_NX; | |
661 | wrmsr(MSR_EFER, l, h); | |
662 | nx_enabled = 1; | |
663 | __supported_pte_mask |= _PAGE_NX; | |
664 | } | |
665 | } | |
666 | } | |
1da177e4 LT |
667 | #endif |
668 | ||
90d967e0 YL |
669 | /* user-defined highmem size */ |
670 | static unsigned int highmem_pages = -1; | |
671 | ||
672 | /* | |
673 | * highmem=size forces highmem to be exactly 'size' bytes. | |
674 | * This works even on boxes that have no highmem otherwise. | |
675 | * This also works to reduce highmem size on bigger boxes. | |
676 | */ | |
677 | static int __init parse_highmem(char *arg) | |
678 | { | |
679 | if (!arg) | |
680 | return -EINVAL; | |
681 | ||
682 | highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; | |
683 | return 0; | |
684 | } | |
685 | early_param("highmem", parse_highmem); | |
686 | ||
687 | /* | |
688 | * Determine low and high memory ranges: | |
689 | */ | |
2ec65f8b | 690 | void __init find_low_pfn_range(void) |
90d967e0 | 691 | { |
2ec65f8b YL |
692 | /* it could update max_pfn */ |
693 | ||
346cafec | 694 | /* max_low_pfn is 0, we already have early_res support */ |
90d967e0 YL |
695 | |
696 | max_low_pfn = max_pfn; | |
697 | if (max_low_pfn > MAXMEM_PFN) { | |
698 | if (highmem_pages == -1) | |
699 | highmem_pages = max_pfn - MAXMEM_PFN; | |
700 | if (highmem_pages + MAXMEM_PFN < max_pfn) | |
701 | max_pfn = MAXMEM_PFN + highmem_pages; | |
702 | if (highmem_pages + MAXMEM_PFN > max_pfn) { | |
703 | printk(KERN_WARNING "only %luMB highmem pages " | |
704 | "available, ignoring highmem size of %uMB.\n", | |
705 | pages_to_mb(max_pfn - MAXMEM_PFN), | |
706 | pages_to_mb(highmem_pages)); | |
707 | highmem_pages = 0; | |
708 | } | |
709 | max_low_pfn = MAXMEM_PFN; | |
710 | #ifndef CONFIG_HIGHMEM | |
711 | /* Maximum memory usable is what is directly addressable */ | |
712 | printk(KERN_WARNING "Warning only %ldMB will be used.\n", | |
713 | MAXMEM>>20); | |
714 | if (max_pfn > MAX_NONPAE_PFN) | |
715 | printk(KERN_WARNING | |
716 | "Use a HIGHMEM64G enabled kernel.\n"); | |
717 | else | |
718 | printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); | |
719 | max_pfn = MAXMEM_PFN; | |
720 | #else /* !CONFIG_HIGHMEM */ | |
721 | #ifndef CONFIG_HIGHMEM64G | |
722 | if (max_pfn > MAX_NONPAE_PFN) { | |
723 | max_pfn = MAX_NONPAE_PFN; | |
724 | printk(KERN_WARNING "Warning only 4GB will be used." | |
725 | "Use a HIGHMEM64G enabled kernel.\n"); | |
726 | } | |
727 | #endif /* !CONFIG_HIGHMEM64G */ | |
728 | #endif /* !CONFIG_HIGHMEM */ | |
729 | } else { | |
730 | if (highmem_pages == -1) | |
731 | highmem_pages = 0; | |
732 | #ifdef CONFIG_HIGHMEM | |
733 | if (highmem_pages >= max_pfn) { | |
734 | printk(KERN_ERR "highmem size specified (%uMB) is " | |
735 | "bigger than pages available (%luMB)!.\n", | |
736 | pages_to_mb(highmem_pages), | |
737 | pages_to_mb(max_pfn)); | |
738 | highmem_pages = 0; | |
739 | } | |
740 | if (highmem_pages) { | |
741 | if (max_low_pfn - highmem_pages < | |
742 | 64*1024*1024/PAGE_SIZE){ | |
743 | printk(KERN_ERR "highmem size %uMB results in " | |
744 | "smaller than 64MB lowmem, ignoring it.\n" | |
745 | , pages_to_mb(highmem_pages)); | |
746 | highmem_pages = 0; | |
747 | } | |
748 | max_low_pfn -= highmem_pages; | |
749 | } | |
750 | #else | |
751 | if (highmem_pages) | |
752 | printk(KERN_ERR "ignoring highmem size on non-highmem" | |
753 | " kernel!\n"); | |
754 | #endif | |
755 | } | |
90d967e0 YL |
756 | } |
757 | ||
b2ac82a0 | 758 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
2ec65f8b | 759 | void __init initmem_init(unsigned long start_pfn, |
b2ac82a0 YL |
760 | unsigned long end_pfn) |
761 | { | |
b2ac82a0 YL |
762 | #ifdef CONFIG_HIGHMEM |
763 | highstart_pfn = highend_pfn = max_pfn; | |
764 | if (max_pfn > max_low_pfn) | |
765 | highstart_pfn = max_low_pfn; | |
766 | memory_present(0, 0, highend_pfn); | |
cb95a13a | 767 | e820_register_active_regions(0, 0, highend_pfn); |
b2ac82a0 YL |
768 | printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", |
769 | pages_to_mb(highend_pfn - highstart_pfn)); | |
770 | num_physpages = highend_pfn; | |
771 | high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; | |
772 | #else | |
773 | memory_present(0, 0, max_low_pfn); | |
cb95a13a | 774 | e820_register_active_regions(0, 0, max_low_pfn); |
b2ac82a0 YL |
775 | num_physpages = max_low_pfn; |
776 | high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; | |
777 | #endif | |
778 | #ifdef CONFIG_FLATMEM | |
779 | max_mapnr = num_physpages; | |
780 | #endif | |
781 | printk(KERN_NOTICE "%ldMB LOWMEM available.\n", | |
782 | pages_to_mb(max_low_pfn)); | |
783 | ||
784 | setup_bootmem_allocator(); | |
b2ac82a0 | 785 | } |
cb95a13a | 786 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
b2ac82a0 | 787 | |
cb95a13a | 788 | static void __init zone_sizes_init(void) |
b2ac82a0 YL |
789 | { |
790 | unsigned long max_zone_pfns[MAX_NR_ZONES]; | |
791 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
792 | max_zone_pfns[ZONE_DMA] = | |
793 | virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
794 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; | |
b2ac82a0 YL |
795 | #ifdef CONFIG_HIGHMEM |
796 | max_zone_pfns[ZONE_HIGHMEM] = highend_pfn; | |
b2ac82a0 YL |
797 | #endif |
798 | ||
799 | free_area_init_nodes(max_zone_pfns); | |
800 | } | |
b2ac82a0 | 801 | |
b2ac82a0 YL |
802 | void __init setup_bootmem_allocator(void) |
803 | { | |
804 | int i; | |
805 | unsigned long bootmap_size, bootmap; | |
806 | /* | |
807 | * Initialize the boot-time allocator (with low memory only): | |
808 | */ | |
809 | bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT; | |
810 | bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT, | |
811 | max_pfn_mapped<<PAGE_SHIFT, bootmap_size, | |
812 | PAGE_SIZE); | |
813 | if (bootmap == -1L) | |
814 | panic("Cannot find bootmem map of size %ld\n", bootmap_size); | |
815 | reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP"); | |
225c37d7 | 816 | |
346cafec YL |
817 | /* don't touch min_low_pfn */ |
818 | bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT, | |
819 | min_low_pfn, max_low_pfn); | |
b2ac82a0 YL |
820 | printk(KERN_INFO " mapped low ram: 0 - %08lx\n", |
821 | max_pfn_mapped<<PAGE_SHIFT); | |
822 | printk(KERN_INFO " low ram: %08lx - %08lx\n", | |
823 | min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT); | |
824 | printk(KERN_INFO " bootmap %08lx - %08lx\n", | |
825 | bootmap, bootmap + bootmap_size); | |
826 | for_each_online_node(i) | |
827 | free_bootmem_with_active_regions(i, max_low_pfn); | |
828 | early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT); | |
829 | ||
4e29684c | 830 | after_init_bootmem = 1; |
b2ac82a0 YL |
831 | } |
832 | ||
0b8fdcbc | 833 | static void __init find_early_table_space(unsigned long end, int use_pse) |
4e29684c | 834 | { |
7482b0e9 | 835 | unsigned long puds, pmds, ptes, tables, start; |
4e29684c YL |
836 | |
837 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; | |
838 | tables = PAGE_ALIGN(puds * sizeof(pud_t)); | |
839 | ||
840 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; | |
841 | tables += PAGE_ALIGN(pmds * sizeof(pmd_t)); | |
842 | ||
0b8fdcbc | 843 | if (use_pse) { |
7482b0e9 | 844 | unsigned long extra; |
a04ad82d YL |
845 | |
846 | extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT); | |
847 | extra += PMD_SIZE; | |
7482b0e9 YL |
848 | ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT; |
849 | } else | |
850 | ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
851 | ||
852 | tables += PAGE_ALIGN(ptes * sizeof(pte_t)); | |
8207c257 | 853 | |
a04ad82d | 854 | /* for fixmap */ |
a3c6018e | 855 | tables += PAGE_ALIGN(__end_of_fixed_addresses * sizeof(pte_t)); |
a04ad82d | 856 | |
4e29684c YL |
857 | /* |
858 | * RED-PEN putting page tables only on node 0 could | |
859 | * cause a hotspot and fill up ZONE_DMA. The page tables | |
860 | * need roughly 0.5KB per GB. | |
861 | */ | |
862 | start = 0x7000; | |
863 | table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT, | |
864 | tables, PAGE_SIZE); | |
865 | if (table_start == -1UL) | |
866 | panic("Cannot find space for the kernel page tables"); | |
867 | ||
868 | table_start >>= PAGE_SHIFT; | |
869 | table_end = table_start; | |
870 | table_top = table_start + (tables>>PAGE_SHIFT); | |
871 | ||
872 | printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n", | |
873 | end, table_start << PAGE_SHIFT, | |
874 | (table_start << PAGE_SHIFT) + tables); | |
875 | } | |
876 | ||
877 | unsigned long __init_refok init_memory_mapping(unsigned long start, | |
878 | unsigned long end) | |
879 | { | |
880 | pgd_t *pgd_base = swapper_pg_dir; | |
a04ad82d YL |
881 | unsigned long start_pfn, end_pfn; |
882 | unsigned long big_page_start; | |
0b8fdcbc SS |
883 | #ifdef CONFIG_DEBUG_PAGEALLOC |
884 | /* | |
885 | * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages. | |
886 | * This will simplify cpa(), which otherwise needs to support splitting | |
887 | * large pages into small in interrupt context, etc. | |
888 | */ | |
889 | int use_pse = 0; | |
890 | #else | |
891 | int use_pse = cpu_has_pse; | |
892 | #endif | |
4e29684c YL |
893 | |
894 | /* | |
895 | * Find space for the kernel direct mapping tables. | |
896 | */ | |
897 | if (!after_init_bootmem) | |
0b8fdcbc | 898 | find_early_table_space(end, use_pse); |
4e29684c YL |
899 | |
900 | #ifdef CONFIG_X86_PAE | |
901 | set_nx(); | |
902 | if (nx_enabled) | |
903 | printk(KERN_INFO "NX (Execute Disable) protection: active\n"); | |
904 | #endif | |
905 | ||
906 | /* Enable PSE if available */ | |
907 | if (cpu_has_pse) | |
908 | set_in_cr4(X86_CR4_PSE); | |
909 | ||
910 | /* Enable PGE if available */ | |
911 | if (cpu_has_pge) { | |
912 | set_in_cr4(X86_CR4_PGE); | |
ef5e94af | 913 | __supported_pte_mask |= _PAGE_GLOBAL; |
4e29684c YL |
914 | } |
915 | ||
a04ad82d YL |
916 | /* |
917 | * Don't use a large page for the first 2/4MB of memory | |
918 | * because there are often fixed size MTRRs in there | |
919 | * and overlapping MTRRs into large pages can cause | |
920 | * slowdowns. | |
921 | */ | |
922 | big_page_start = PMD_SIZE; | |
923 | ||
924 | if (start < big_page_start) { | |
925 | start_pfn = start >> PAGE_SHIFT; | |
926 | end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT); | |
927 | } else { | |
928 | /* head is not big page alignment ? */ | |
929 | start_pfn = start >> PAGE_SHIFT; | |
930 | end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT) | |
931 | << (PMD_SHIFT - PAGE_SHIFT); | |
932 | } | |
933 | if (start_pfn < end_pfn) | |
934 | kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0); | |
935 | ||
936 | /* big page range */ | |
937 | start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT) | |
938 | << (PMD_SHIFT - PAGE_SHIFT); | |
939 | if (start_pfn < (big_page_start >> PAGE_SHIFT)) | |
940 | start_pfn = big_page_start >> PAGE_SHIFT; | |
941 | end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT); | |
942 | if (start_pfn < end_pfn) | |
943 | kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, | |
0b8fdcbc | 944 | use_pse); |
a04ad82d YL |
945 | |
946 | /* tail is not big page alignment ? */ | |
947 | start_pfn = end_pfn; | |
948 | if (start_pfn > (big_page_start>>PAGE_SHIFT)) { | |
949 | end_pfn = end >> PAGE_SHIFT; | |
950 | if (start_pfn < end_pfn) | |
951 | kernel_physical_mapping_init(pgd_base, start_pfn, | |
952 | end_pfn, 0); | |
953 | } | |
4e29684c | 954 | |
e7b37895 YL |
955 | early_ioremap_page_table_range_init(pgd_base); |
956 | ||
4e29684c YL |
957 | load_cr3(swapper_pg_dir); |
958 | ||
959 | __flush_tlb_all(); | |
960 | ||
961 | if (!after_init_bootmem) | |
962 | reserve_early(table_start << PAGE_SHIFT, | |
963 | table_end << PAGE_SHIFT, "PGTABLE"); | |
964 | ||
caadbdce YL |
965 | if (!after_init_bootmem) |
966 | early_memtest(start, end); | |
967 | ||
4e29684c YL |
968 | return end >> PAGE_SHIFT; |
969 | } | |
970 | ||
e7b37895 | 971 | |
1da177e4 LT |
972 | /* |
973 | * paging_init() sets up the page tables - note that the first 8MB are | |
974 | * already mapped by head.S. | |
975 | * | |
976 | * This routines also unmaps the page at virtual kernel address 0, so | |
977 | * that we can trap those pesky NULL-reference errors in the kernel. | |
978 | */ | |
979 | void __init paging_init(void) | |
980 | { | |
1da177e4 LT |
981 | pagetable_init(); |
982 | ||
1da177e4 LT |
983 | __flush_tlb_all(); |
984 | ||
985 | kmap_init(); | |
11cd0bc1 YL |
986 | |
987 | /* | |
988 | * NOTE: at this point the bootmem allocator is fully available. | |
989 | */ | |
11cd0bc1 YL |
990 | sparse_init(); |
991 | zone_sizes_init(); | |
1da177e4 LT |
992 | } |
993 | ||
994 | /* | |
995 | * Test if the WP bit works in supervisor mode. It isn't supported on 386's | |
f7f17a67 DV |
996 | * and also on some strange 486's. All 586+'s are OK. This used to involve |
997 | * black magic jumps to work around some nasty CPU bugs, but fortunately the | |
998 | * switch to using exceptions got rid of all that. | |
1da177e4 | 999 | */ |
1da177e4 LT |
1000 | static void __init test_wp_bit(void) |
1001 | { | |
d7d119d7 IM |
1002 | printk(KERN_INFO |
1003 | "Checking if this processor honours the WP bit even in supervisor mode..."); | |
1da177e4 LT |
1004 | |
1005 | /* Any page-aligned address will do, the test is non-destructive */ | |
1006 | __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY); | |
1007 | boot_cpu_data.wp_works_ok = do_test_wp_bit(); | |
1008 | clear_fixmap(FIX_WP_TEST); | |
1009 | ||
1010 | if (!boot_cpu_data.wp_works_ok) { | |
d7d119d7 | 1011 | printk(KERN_CONT "No.\n"); |
1da177e4 | 1012 | #ifdef CONFIG_X86_WP_WORKS_OK |
d7d119d7 IM |
1013 | panic( |
1014 | "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!"); | |
1da177e4 LT |
1015 | #endif |
1016 | } else { | |
d7d119d7 | 1017 | printk(KERN_CONT "Ok.\n"); |
1da177e4 LT |
1018 | } |
1019 | } | |
1020 | ||
8550eb99 | 1021 | static struct kcore_list kcore_mem, kcore_vmalloc; |
1da177e4 LT |
1022 | |
1023 | void __init mem_init(void) | |
1024 | { | |
1da177e4 | 1025 | int codesize, reservedpages, datasize, initsize; |
cc9f7a0c | 1026 | int tmp; |
1da177e4 | 1027 | |
cfb80c9e JF |
1028 | pci_iommu_alloc(); |
1029 | ||
05b79bdc | 1030 | #ifdef CONFIG_FLATMEM |
8d8f3cbe | 1031 | BUG_ON(!mem_map); |
1da177e4 | 1032 | #endif |
1da177e4 LT |
1033 | /* this will put all low memory onto the freelists */ |
1034 | totalram_pages += free_all_bootmem(); | |
1035 | ||
1036 | reservedpages = 0; | |
1037 | for (tmp = 0; tmp < max_low_pfn; tmp++) | |
1038 | /* | |
8550eb99 | 1039 | * Only count reserved RAM pages: |
1da177e4 LT |
1040 | */ |
1041 | if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp))) | |
1042 | reservedpages++; | |
1043 | ||
cc9f7a0c | 1044 | set_highmem_pages_init(); |
1da177e4 LT |
1045 | |
1046 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
1047 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
1048 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
1049 | ||
8550eb99 IM |
1050 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
1051 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
1da177e4 LT |
1052 | VMALLOC_END-VMALLOC_START); |
1053 | ||
8550eb99 IM |
1054 | printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, " |
1055 | "%dk reserved, %dk data, %dk init, %ldk highmem)\n", | |
1da177e4 LT |
1056 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
1057 | num_physpages << (PAGE_SHIFT-10), | |
1058 | codesize >> 10, | |
1059 | reservedpages << (PAGE_SHIFT-10), | |
1060 | datasize >> 10, | |
1061 | initsize >> 10, | |
1062 | (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)) | |
1063 | ); | |
1064 | ||
d7d119d7 | 1065 | printk(KERN_INFO "virtual kernel memory layout:\n" |
8550eb99 | 1066 | " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" |
052e7994 | 1067 | #ifdef CONFIG_HIGHMEM |
8550eb99 | 1068 | " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n" |
052e7994 | 1069 | #endif |
8550eb99 IM |
1070 | " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" |
1071 | " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" | |
1072 | " .init : 0x%08lx - 0x%08lx (%4ld kB)\n" | |
1073 | " .data : 0x%08lx - 0x%08lx (%4ld kB)\n" | |
1074 | " .text : 0x%08lx - 0x%08lx (%4ld kB)\n", | |
1075 | FIXADDR_START, FIXADDR_TOP, | |
1076 | (FIXADDR_TOP - FIXADDR_START) >> 10, | |
052e7994 JF |
1077 | |
1078 | #ifdef CONFIG_HIGHMEM | |
8550eb99 IM |
1079 | PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, |
1080 | (LAST_PKMAP*PAGE_SIZE) >> 10, | |
052e7994 JF |
1081 | #endif |
1082 | ||
8550eb99 IM |
1083 | VMALLOC_START, VMALLOC_END, |
1084 | (VMALLOC_END - VMALLOC_START) >> 20, | |
052e7994 | 1085 | |
8550eb99 IM |
1086 | (unsigned long)__va(0), (unsigned long)high_memory, |
1087 | ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, | |
052e7994 | 1088 | |
8550eb99 IM |
1089 | (unsigned long)&__init_begin, (unsigned long)&__init_end, |
1090 | ((unsigned long)&__init_end - | |
1091 | (unsigned long)&__init_begin) >> 10, | |
052e7994 | 1092 | |
8550eb99 IM |
1093 | (unsigned long)&_etext, (unsigned long)&_edata, |
1094 | ((unsigned long)&_edata - (unsigned long)&_etext) >> 10, | |
052e7994 | 1095 | |
8550eb99 IM |
1096 | (unsigned long)&_text, (unsigned long)&_etext, |
1097 | ((unsigned long)&_etext - (unsigned long)&_text) >> 10); | |
052e7994 | 1098 | |
beeb4195 JB |
1099 | /* |
1100 | * Check boundaries twice: Some fundamental inconsistencies can | |
1101 | * be detected at build time already. | |
1102 | */ | |
1103 | #define __FIXADDR_TOP (-PAGE_SIZE) | |
1104 | #ifdef CONFIG_HIGHMEM | |
1105 | BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); | |
1106 | BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE); | |
1107 | #endif | |
1108 | #define high_memory (-128UL << 20) | |
1109 | BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END); | |
1110 | #undef high_memory | |
1111 | #undef __FIXADDR_TOP | |
1112 | ||
052e7994 | 1113 | #ifdef CONFIG_HIGHMEM |
8550eb99 IM |
1114 | BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); |
1115 | BUG_ON(VMALLOC_END > PKMAP_BASE); | |
052e7994 | 1116 | #endif |
beeb4195 | 1117 | BUG_ON(VMALLOC_START >= VMALLOC_END); |
8550eb99 | 1118 | BUG_ON((unsigned long)high_memory > VMALLOC_START); |
052e7994 | 1119 | |
1da177e4 LT |
1120 | if (boot_cpu_data.wp_works_ok < 0) |
1121 | test_wp_bit(); | |
1122 | ||
61165d7a | 1123 | save_pg_dir(); |
1da177e4 | 1124 | zap_low_mappings(); |
1da177e4 LT |
1125 | } |
1126 | ||
ad8f5797 | 1127 | #ifdef CONFIG_MEMORY_HOTPLUG |
bc02af93 | 1128 | int arch_add_memory(int nid, u64 start, u64 size) |
05039b92 | 1129 | { |
7c7e9425 | 1130 | struct pglist_data *pgdata = NODE_DATA(nid); |
776ed98b | 1131 | struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM; |
05039b92 DH |
1132 | unsigned long start_pfn = start >> PAGE_SHIFT; |
1133 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
1134 | ||
c04fc586 | 1135 | return __add_pages(nid, zone, start_pfn, nr_pages); |
05039b92 | 1136 | } |
9d99aaa3 | 1137 | #endif |
05039b92 | 1138 | |
1da177e4 LT |
1139 | /* |
1140 | * This function cannot be __init, since exceptions don't work in that | |
1141 | * section. Put this after the callers, so that it cannot be inlined. | |
1142 | */ | |
8550eb99 | 1143 | static noinline int do_test_wp_bit(void) |
1da177e4 LT |
1144 | { |
1145 | char tmp_reg; | |
1146 | int flag; | |
1147 | ||
1148 | __asm__ __volatile__( | |
8550eb99 IM |
1149 | " movb %0, %1 \n" |
1150 | "1: movb %1, %0 \n" | |
1151 | " xorl %2, %2 \n" | |
1da177e4 | 1152 | "2: \n" |
f832ff18 | 1153 | _ASM_EXTABLE(1b,2b) |
1da177e4 LT |
1154 | :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)), |
1155 | "=q" (tmp_reg), | |
1156 | "=r" (flag) | |
1157 | :"2" (1) | |
1158 | :"memory"); | |
8550eb99 | 1159 | |
1da177e4 LT |
1160 | return flag; |
1161 | } | |
1162 | ||
63aaf308 | 1163 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
1164 | const int rodata_test_data = 0xC3; |
1165 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
63aaf308 | 1166 | |
63aaf308 AV |
1167 | void mark_rodata_ro(void) |
1168 | { | |
6fb14755 JB |
1169 | unsigned long start = PFN_ALIGN(_text); |
1170 | unsigned long size = PFN_ALIGN(_etext) - start; | |
63aaf308 | 1171 | |
8f0f996e SR |
1172 | #ifndef CONFIG_DYNAMIC_FTRACE |
1173 | /* Dynamic tracing modifies the kernel text section */ | |
4e4eee0e MD |
1174 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
1175 | printk(KERN_INFO "Write protecting the kernel text: %luk\n", | |
1176 | size >> 10); | |
0c42f392 AK |
1177 | |
1178 | #ifdef CONFIG_CPA_DEBUG | |
4e4eee0e MD |
1179 | printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n", |
1180 | start, start+size); | |
1181 | set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT); | |
0c42f392 | 1182 | |
4e4eee0e MD |
1183 | printk(KERN_INFO "Testing CPA: write protecting again\n"); |
1184 | set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT); | |
602033ed | 1185 | #endif |
8f0f996e SR |
1186 | #endif /* CONFIG_DYNAMIC_FTRACE */ |
1187 | ||
6fb14755 JB |
1188 | start += size; |
1189 | size = (unsigned long)__end_rodata - start; | |
6d238cc4 | 1190 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
d7d119d7 IM |
1191 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
1192 | size >> 10); | |
edeed305 | 1193 | rodata_test(); |
63aaf308 | 1194 | |
0c42f392 | 1195 | #ifdef CONFIG_CPA_DEBUG |
d7d119d7 | 1196 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size); |
6d238cc4 | 1197 | set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); |
0c42f392 | 1198 | |
d7d119d7 | 1199 | printk(KERN_INFO "Testing CPA: write protecting again\n"); |
6d238cc4 | 1200 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
0c42f392 | 1201 | #endif |
63aaf308 AV |
1202 | } |
1203 | #endif | |
1204 | ||
9a0b5817 GH |
1205 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
1206 | { | |
ee01f112 IM |
1207 | #ifdef CONFIG_DEBUG_PAGEALLOC |
1208 | /* | |
1209 | * If debugging page accesses then do not free this memory but | |
1210 | * mark them not present - any buggy init-section access will | |
1211 | * create a kernel page fault: | |
1212 | */ | |
1213 | printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", | |
1214 | begin, PAGE_ALIGN(end)); | |
1215 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
1216 | #else | |
86f03989 IM |
1217 | unsigned long addr; |
1218 | ||
3c1df68b AV |
1219 | /* |
1220 | * We just marked the kernel text read only above, now that | |
1221 | * we are going to free part of that, we need to make that | |
1222 | * writeable first. | |
1223 | */ | |
1224 | set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); | |
1225 | ||
9a0b5817 | 1226 | for (addr = begin; addr < end; addr += PAGE_SIZE) { |
e3ebadd9 LT |
1227 | ClearPageReserved(virt_to_page(addr)); |
1228 | init_page_count(virt_to_page(addr)); | |
1229 | memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE); | |
1230 | free_page(addr); | |
9a0b5817 GH |
1231 | totalram_pages++; |
1232 | } | |
6fb14755 | 1233 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
ee01f112 | 1234 | #endif |
9a0b5817 GH |
1235 | } |
1236 | ||
1237 | void free_initmem(void) | |
1238 | { | |
1239 | free_init_pages("unused kernel memory", | |
e3ebadd9 LT |
1240 | (unsigned long)(&__init_begin), |
1241 | (unsigned long)(&__init_end)); | |
9a0b5817 | 1242 | } |
63aaf308 | 1243 | |
1da177e4 LT |
1244 | #ifdef CONFIG_BLK_DEV_INITRD |
1245 | void free_initrd_mem(unsigned long start, unsigned long end) | |
1246 | { | |
e3ebadd9 | 1247 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
1248 | } |
1249 | #endif | |
d2dbf343 YL |
1250 | |
1251 | int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, | |
1252 | int flags) | |
1253 | { | |
1254 | return reserve_bootmem(phys, len, flags); | |
1255 | } |