]>
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 | ||
a3c6018e JB |
140 | static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, |
141 | unsigned long vaddr, pte_t *lastpte) | |
142 | { | |
143 | #ifdef CONFIG_HIGHMEM | |
144 | /* | |
145 | * Something (early fixmap) may already have put a pte | |
146 | * page here, which causes the page table allocation | |
147 | * to become nonlinear. Attempt to fix it, and if it | |
148 | * is still nonlinear then we have to bug. | |
149 | */ | |
150 | int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; | |
151 | int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; | |
152 | ||
153 | if (pmd_idx_kmap_begin != pmd_idx_kmap_end | |
154 | && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin | |
155 | && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end | |
156 | && ((__pa(pte) >> PAGE_SHIFT) < table_start | |
157 | || (__pa(pte) >> PAGE_SHIFT) >= table_end)) { | |
158 | pte_t *newpte; | |
159 | int i; | |
160 | ||
161 | BUG_ON(after_init_bootmem); | |
162 | newpte = alloc_low_page(); | |
163 | for (i = 0; i < PTRS_PER_PTE; i++) | |
164 | set_pte(newpte + i, pte[i]); | |
165 | ||
166 | paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT); | |
167 | set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); | |
168 | BUG_ON(newpte != pte_offset_kernel(pmd, 0)); | |
169 | __flush_tlb_all(); | |
170 | ||
171 | paravirt_release_pte(__pa(pte) >> PAGE_SHIFT); | |
172 | pte = newpte; | |
173 | } | |
174 | BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1) | |
175 | && vaddr > fix_to_virt(FIX_KMAP_END) | |
176 | && lastpte && lastpte + PTRS_PER_PTE != pte); | |
177 | #endif | |
178 | return pte; | |
179 | } | |
180 | ||
1da177e4 | 181 | /* |
8550eb99 | 182 | * This function initializes a certain range of kernel virtual memory |
1da177e4 LT |
183 | * with new bootmem page tables, everywhere page tables are missing in |
184 | * the given range. | |
8550eb99 IM |
185 | * |
186 | * NOTE: The pagetables are allocated contiguous on the physical space | |
187 | * so we can cache the place of the first one and move around without | |
1da177e4 LT |
188 | * checking the pgd every time. |
189 | */ | |
8550eb99 IM |
190 | static void __init |
191 | page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) | |
1da177e4 | 192 | { |
1da177e4 LT |
193 | int pgd_idx, pmd_idx; |
194 | unsigned long vaddr; | |
8550eb99 IM |
195 | pgd_t *pgd; |
196 | pmd_t *pmd; | |
a3c6018e | 197 | pte_t *pte = NULL; |
1da177e4 LT |
198 | |
199 | vaddr = start; | |
200 | pgd_idx = pgd_index(vaddr); | |
201 | pmd_idx = pmd_index(vaddr); | |
202 | pgd = pgd_base + pgd_idx; | |
203 | ||
204 | for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { | |
b239fb25 JF |
205 | pmd = one_md_table_init(pgd); |
206 | pmd = pmd + pmd_index(vaddr); | |
8550eb99 IM |
207 | for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); |
208 | pmd++, pmd_idx++) { | |
a3c6018e JB |
209 | pte = page_table_kmap_check(one_page_table_init(pmd), |
210 | pmd, vaddr, pte); | |
1da177e4 LT |
211 | |
212 | vaddr += PMD_SIZE; | |
213 | } | |
214 | pmd_idx = 0; | |
215 | } | |
216 | } | |
217 | ||
218 | static inline int is_kernel_text(unsigned long addr) | |
219 | { | |
220 | if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end) | |
221 | return 1; | |
222 | return 0; | |
223 | } | |
224 | ||
225 | /* | |
8550eb99 IM |
226 | * This maps the physical memory to kernel virtual address space, a total |
227 | * of max_low_pfn pages, by creating page tables starting from address | |
228 | * PAGE_OFFSET: | |
1da177e4 | 229 | */ |
4e29684c | 230 | static void __init kernel_physical_mapping_init(pgd_t *pgd_base, |
a04ad82d YL |
231 | unsigned long start_pfn, |
232 | unsigned long end_pfn, | |
233 | int use_pse) | |
1da177e4 | 234 | { |
8550eb99 | 235 | int pgd_idx, pmd_idx, pte_ofs; |
1da177e4 LT |
236 | unsigned long pfn; |
237 | pgd_t *pgd; | |
238 | pmd_t *pmd; | |
239 | pte_t *pte; | |
a2699e47 SS |
240 | unsigned pages_2m, pages_4k; |
241 | int mapping_iter; | |
242 | ||
243 | /* | |
244 | * First iteration will setup identity mapping using large/small pages | |
245 | * based on use_pse, with other attributes same as set by | |
246 | * the early code in head_32.S | |
247 | * | |
248 | * Second iteration will setup the appropriate attributes (NX, GLOBAL..) | |
249 | * as desired for the kernel identity mapping. | |
250 | * | |
251 | * This two pass mechanism conforms to the TLB app note which says: | |
252 | * | |
253 | * "Software should not write to a paging-structure entry in a way | |
254 | * that would change, for any linear address, both the page size | |
255 | * and either the page frame or attributes." | |
256 | */ | |
257 | mapping_iter = 1; | |
1da177e4 | 258 | |
a04ad82d YL |
259 | if (!cpu_has_pse) |
260 | use_pse = 0; | |
1da177e4 | 261 | |
a2699e47 SS |
262 | repeat: |
263 | pages_2m = pages_4k = 0; | |
a04ad82d YL |
264 | pfn = start_pfn; |
265 | pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); | |
266 | pgd = pgd_base + pgd_idx; | |
1da177e4 LT |
267 | for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { |
268 | pmd = one_md_table_init(pgd); | |
8550eb99 | 269 | |
a04ad82d YL |
270 | if (pfn >= end_pfn) |
271 | continue; | |
272 | #ifdef CONFIG_X86_PAE | |
273 | pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); | |
274 | pmd += pmd_idx; | |
275 | #else | |
276 | pmd_idx = 0; | |
277 | #endif | |
278 | for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn; | |
f3f20de8 | 279 | pmd++, pmd_idx++) { |
8550eb99 | 280 | unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; |
1da177e4 | 281 | |
8550eb99 IM |
282 | /* |
283 | * Map with big pages if possible, otherwise | |
284 | * create normal page tables: | |
285 | */ | |
a04ad82d | 286 | if (use_pse) { |
8550eb99 | 287 | unsigned int addr2; |
f3f20de8 | 288 | pgprot_t prot = PAGE_KERNEL_LARGE; |
a2699e47 SS |
289 | /* |
290 | * first pass will use the same initial | |
291 | * identity mapping attribute + _PAGE_PSE. | |
292 | */ | |
293 | pgprot_t init_prot = | |
294 | __pgprot(PTE_IDENT_ATTR | | |
295 | _PAGE_PSE); | |
f3f20de8 | 296 | |
8550eb99 | 297 | addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + |
f3f20de8 JF |
298 | PAGE_OFFSET + PAGE_SIZE-1; |
299 | ||
8550eb99 IM |
300 | if (is_kernel_text(addr) || |
301 | is_kernel_text(addr2)) | |
f3f20de8 JF |
302 | prot = PAGE_KERNEL_LARGE_EXEC; |
303 | ||
ce0c0e50 | 304 | pages_2m++; |
a2699e47 SS |
305 | if (mapping_iter == 1) |
306 | set_pmd(pmd, pfn_pmd(pfn, init_prot)); | |
307 | else | |
308 | set_pmd(pmd, pfn_pmd(pfn, prot)); | |
b239fb25 | 309 | |
1da177e4 | 310 | pfn += PTRS_PER_PTE; |
8550eb99 IM |
311 | continue; |
312 | } | |
313 | pte = one_page_table_init(pmd); | |
1da177e4 | 314 | |
a04ad82d YL |
315 | pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); |
316 | pte += pte_ofs; | |
317 | for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; | |
8550eb99 IM |
318 | pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { |
319 | pgprot_t prot = PAGE_KERNEL; | |
a2699e47 SS |
320 | /* |
321 | * first pass will use the same initial | |
322 | * identity mapping attribute. | |
323 | */ | |
324 | pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); | |
f3f20de8 | 325 | |
8550eb99 IM |
326 | if (is_kernel_text(addr)) |
327 | prot = PAGE_KERNEL_EXEC; | |
f3f20de8 | 328 | |
ce0c0e50 | 329 | pages_4k++; |
a2699e47 SS |
330 | if (mapping_iter == 1) |
331 | set_pte(pte, pfn_pte(pfn, init_prot)); | |
332 | else | |
333 | set_pte(pte, pfn_pte(pfn, prot)); | |
1da177e4 LT |
334 | } |
335 | } | |
336 | } | |
a2699e47 SS |
337 | if (mapping_iter == 1) { |
338 | /* | |
339 | * update direct mapping page count only in the first | |
340 | * iteration. | |
341 | */ | |
342 | update_page_count(PG_LEVEL_2M, pages_2m); | |
343 | update_page_count(PG_LEVEL_4K, pages_4k); | |
344 | ||
345 | /* | |
346 | * local global flush tlb, which will flush the previous | |
347 | * mappings present in both small and large page TLB's. | |
348 | */ | |
349 | __flush_tlb_all(); | |
350 | ||
351 | /* | |
352 | * Second iteration will set the actual desired PTE attributes. | |
353 | */ | |
354 | mapping_iter = 2; | |
355 | goto repeat; | |
356 | } | |
1da177e4 LT |
357 | } |
358 | ||
ae531c26 AV |
359 | /* |
360 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
361 | * is valid. The argument is a physical page number. | |
362 | * | |
363 | * | |
364 | * On x86, access has to be given to the first megabyte of ram because that area | |
365 | * contains bios code and data regions used by X and dosemu and similar apps. | |
366 | * Access has to be given to non-kernel-ram areas as well, these contain the PCI | |
367 | * mmio resources as well as potential bios/acpi data regions. | |
368 | */ | |
369 | int devmem_is_allowed(unsigned long pagenr) | |
370 | { | |
371 | if (pagenr <= 256) | |
372 | return 1; | |
e8de1481 AV |
373 | if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) |
374 | return 0; | |
ae531c26 AV |
375 | if (!page_is_ram(pagenr)) |
376 | return 1; | |
377 | return 0; | |
378 | } | |
379 | ||
1da177e4 LT |
380 | pte_t *kmap_pte; |
381 | pgprot_t kmap_prot; | |
382 | ||
8550eb99 IM |
383 | static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) |
384 | { | |
385 | return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), | |
386 | vaddr), vaddr), vaddr); | |
387 | } | |
1da177e4 LT |
388 | |
389 | static void __init kmap_init(void) | |
390 | { | |
391 | unsigned long kmap_vstart; | |
392 | ||
8550eb99 IM |
393 | /* |
394 | * Cache the first kmap pte: | |
395 | */ | |
1da177e4 LT |
396 | kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); |
397 | kmap_pte = kmap_get_fixmap_pte(kmap_vstart); | |
398 | ||
399 | kmap_prot = PAGE_KERNEL; | |
400 | } | |
401 | ||
fd940934 | 402 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
403 | static void __init permanent_kmaps_init(pgd_t *pgd_base) |
404 | { | |
8550eb99 | 405 | unsigned long vaddr; |
1da177e4 LT |
406 | pgd_t *pgd; |
407 | pud_t *pud; | |
408 | pmd_t *pmd; | |
409 | pte_t *pte; | |
1da177e4 LT |
410 | |
411 | vaddr = PKMAP_BASE; | |
412 | page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); | |
413 | ||
414 | pgd = swapper_pg_dir + pgd_index(vaddr); | |
415 | pud = pud_offset(pgd, vaddr); | |
416 | pmd = pmd_offset(pud, vaddr); | |
417 | pte = pte_offset_kernel(pmd, vaddr); | |
8550eb99 | 418 | pkmap_page_table = pte; |
1da177e4 LT |
419 | } |
420 | ||
cc9f7a0c | 421 | static void __init add_one_highpage_init(struct page *page, int pfn) |
1da177e4 | 422 | { |
cc9f7a0c YL |
423 | ClearPageReserved(page); |
424 | init_page_count(page); | |
425 | __free_page(page); | |
426 | totalhigh_pages++; | |
1da177e4 LT |
427 | } |
428 | ||
b5bc6c0e YL |
429 | struct add_highpages_data { |
430 | unsigned long start_pfn; | |
431 | unsigned long end_pfn; | |
b5bc6c0e YL |
432 | }; |
433 | ||
d52d53b8 | 434 | static int __init add_highpages_work_fn(unsigned long start_pfn, |
b5bc6c0e | 435 | unsigned long end_pfn, void *datax) |
1da177e4 | 436 | { |
b5bc6c0e YL |
437 | int node_pfn; |
438 | struct page *page; | |
439 | unsigned long final_start_pfn, final_end_pfn; | |
440 | struct add_highpages_data *data; | |
8550eb99 | 441 | |
b5bc6c0e | 442 | data = (struct add_highpages_data *)datax; |
b5bc6c0e YL |
443 | |
444 | final_start_pfn = max(start_pfn, data->start_pfn); | |
445 | final_end_pfn = min(end_pfn, data->end_pfn); | |
446 | if (final_start_pfn >= final_end_pfn) | |
d52d53b8 | 447 | return 0; |
b5bc6c0e YL |
448 | |
449 | for (node_pfn = final_start_pfn; node_pfn < final_end_pfn; | |
450 | node_pfn++) { | |
451 | if (!pfn_valid(node_pfn)) | |
452 | continue; | |
453 | page = pfn_to_page(node_pfn); | |
cc9f7a0c | 454 | add_one_highpage_init(page, node_pfn); |
23be8c7d | 455 | } |
b5bc6c0e | 456 | |
d52d53b8 YL |
457 | return 0; |
458 | ||
b5bc6c0e YL |
459 | } |
460 | ||
461 | void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn, | |
cc9f7a0c | 462 | unsigned long end_pfn) |
b5bc6c0e YL |
463 | { |
464 | struct add_highpages_data data; | |
465 | ||
466 | data.start_pfn = start_pfn; | |
467 | data.end_pfn = end_pfn; | |
b5bc6c0e YL |
468 | |
469 | work_with_active_regions(nid, add_highpages_work_fn, &data); | |
470 | } | |
471 | ||
8550eb99 | 472 | #ifndef CONFIG_NUMA |
cc9f7a0c | 473 | static void __init set_highmem_pages_init(void) |
1da177e4 | 474 | { |
cc9f7a0c | 475 | add_highpages_with_active_regions(0, highstart_pfn, highend_pfn); |
b5bc6c0e | 476 | |
1da177e4 LT |
477 | totalram_pages += totalhigh_pages; |
478 | } | |
8550eb99 | 479 | #endif /* !CONFIG_NUMA */ |
1da177e4 LT |
480 | |
481 | #else | |
e8e32326 IB |
482 | static inline void permanent_kmaps_init(pgd_t *pgd_base) |
483 | { | |
484 | } | |
485 | static inline void set_highmem_pages_init(void) | |
486 | { | |
487 | } | |
1da177e4 LT |
488 | #endif /* CONFIG_HIGHMEM */ |
489 | ||
b239fb25 | 490 | void __init native_pagetable_setup_start(pgd_t *base) |
1da177e4 | 491 | { |
551889a6 IC |
492 | unsigned long pfn, va; |
493 | pgd_t *pgd; | |
494 | pud_t *pud; | |
495 | pmd_t *pmd; | |
496 | pte_t *pte; | |
b239fb25 JF |
497 | |
498 | /* | |
551889a6 IC |
499 | * Remove any mappings which extend past the end of physical |
500 | * memory from the boot time page table: | |
b239fb25 | 501 | */ |
551889a6 IC |
502 | for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) { |
503 | va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); | |
504 | pgd = base + pgd_index(va); | |
505 | if (!pgd_present(*pgd)) | |
506 | break; | |
507 | ||
508 | pud = pud_offset(pgd, va); | |
509 | pmd = pmd_offset(pud, va); | |
510 | if (!pmd_present(*pmd)) | |
511 | break; | |
512 | ||
513 | pte = pte_offset_kernel(pmd, va); | |
514 | if (!pte_present(*pte)) | |
515 | break; | |
516 | ||
517 | pte_clear(NULL, va, pte); | |
518 | } | |
6944a9c8 | 519 | paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT); |
b239fb25 JF |
520 | } |
521 | ||
522 | void __init native_pagetable_setup_done(pgd_t *base) | |
523 | { | |
b239fb25 JF |
524 | } |
525 | ||
526 | /* | |
527 | * Build a proper pagetable for the kernel mappings. Up until this | |
528 | * point, we've been running on some set of pagetables constructed by | |
529 | * the boot process. | |
530 | * | |
531 | * If we're booting on native hardware, this will be a pagetable | |
551889a6 IC |
532 | * constructed in arch/x86/kernel/head_32.S. The root of the |
533 | * pagetable will be swapper_pg_dir. | |
b239fb25 JF |
534 | * |
535 | * If we're booting paravirtualized under a hypervisor, then there are | |
536 | * more options: we may already be running PAE, and the pagetable may | |
537 | * or may not be based in swapper_pg_dir. In any case, | |
538 | * paravirt_pagetable_setup_start() will set up swapper_pg_dir | |
539 | * appropriately for the rest of the initialization to work. | |
540 | * | |
541 | * In general, pagetable_init() assumes that the pagetable may already | |
542 | * be partially populated, and so it avoids stomping on any existing | |
543 | * mappings. | |
544 | */ | |
e7b37895 | 545 | static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base) |
b239fb25 | 546 | { |
8550eb99 | 547 | unsigned long vaddr, end; |
b239fb25 | 548 | |
1da177e4 LT |
549 | /* |
550 | * Fixed mappings, only the page table structure has to be | |
551 | * created - mappings will be set by set_fixmap(): | |
552 | */ | |
553 | vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; | |
b239fb25 JF |
554 | end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; |
555 | page_table_range_init(vaddr, end, pgd_base); | |
beacfaac | 556 | early_ioremap_reset(); |
e7b37895 YL |
557 | } |
558 | ||
559 | static void __init pagetable_init(void) | |
560 | { | |
561 | pgd_t *pgd_base = swapper_pg_dir; | |
562 | ||
1da177e4 | 563 | permanent_kmaps_init(pgd_base); |
1da177e4 LT |
564 | } |
565 | ||
a6eb84bc | 566 | #ifdef CONFIG_ACPI_SLEEP |
1da177e4 | 567 | /* |
a6eb84bc | 568 | * ACPI suspend needs this for resume, because things like the intel-agp |
1da177e4 LT |
569 | * driver might have split up a kernel 4MB mapping. |
570 | */ | |
a6eb84bc | 571 | char swsusp_pg_dir[PAGE_SIZE] |
8550eb99 | 572 | __attribute__ ((aligned(PAGE_SIZE))); |
1da177e4 LT |
573 | |
574 | static inline void save_pg_dir(void) | |
575 | { | |
576 | memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE); | |
577 | } | |
a6eb84bc | 578 | #else /* !CONFIG_ACPI_SLEEP */ |
1da177e4 LT |
579 | static inline void save_pg_dir(void) |
580 | { | |
581 | } | |
a6eb84bc | 582 | #endif /* !CONFIG_ACPI_SLEEP */ |
1da177e4 | 583 | |
8550eb99 | 584 | void zap_low_mappings(void) |
1da177e4 LT |
585 | { |
586 | int i; | |
587 | ||
1da177e4 LT |
588 | /* |
589 | * Zap initial low-memory mappings. | |
590 | * | |
591 | * Note that "pgd_clear()" doesn't do it for | |
592 | * us, because pgd_clear() is a no-op on i386. | |
593 | */ | |
68db065c | 594 | for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) { |
1da177e4 LT |
595 | #ifdef CONFIG_X86_PAE |
596 | set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page))); | |
597 | #else | |
598 | set_pgd(swapper_pg_dir+i, __pgd(0)); | |
599 | #endif | |
8550eb99 | 600 | } |
1da177e4 LT |
601 | flush_tlb_all(); |
602 | } | |
603 | ||
8550eb99 | 604 | int nx_enabled; |
d5321abe | 605 | |
be43d728 | 606 | pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP); |
6fdc05d4 JF |
607 | EXPORT_SYMBOL_GPL(__supported_pte_mask); |
608 | ||
d5321abe JB |
609 | #ifdef CONFIG_X86_PAE |
610 | ||
8550eb99 | 611 | static int disable_nx __initdata; |
1da177e4 LT |
612 | |
613 | /* | |
614 | * noexec = on|off | |
615 | * | |
616 | * Control non executable mappings. | |
617 | * | |
618 | * on Enable | |
619 | * off Disable | |
620 | */ | |
1a3f239d | 621 | static int __init noexec_setup(char *str) |
1da177e4 | 622 | { |
1a3f239d RR |
623 | if (!str || !strcmp(str, "on")) { |
624 | if (cpu_has_nx) { | |
625 | __supported_pte_mask |= _PAGE_NX; | |
626 | disable_nx = 0; | |
627 | } | |
8550eb99 IM |
628 | } else { |
629 | if (!strcmp(str, "off")) { | |
630 | disable_nx = 1; | |
631 | __supported_pte_mask &= ~_PAGE_NX; | |
632 | } else { | |
633 | return -EINVAL; | |
634 | } | |
635 | } | |
1a3f239d RR |
636 | |
637 | return 0; | |
1da177e4 | 638 | } |
1a3f239d | 639 | early_param("noexec", noexec_setup); |
1da177e4 | 640 | |
1da177e4 LT |
641 | static void __init set_nx(void) |
642 | { | |
643 | unsigned int v[4], l, h; | |
644 | ||
645 | if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) { | |
646 | cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]); | |
8550eb99 | 647 | |
1da177e4 LT |
648 | if ((v[3] & (1 << 20)) && !disable_nx) { |
649 | rdmsr(MSR_EFER, l, h); | |
650 | l |= EFER_NX; | |
651 | wrmsr(MSR_EFER, l, h); | |
652 | nx_enabled = 1; | |
653 | __supported_pte_mask |= _PAGE_NX; | |
654 | } | |
655 | } | |
656 | } | |
1da177e4 LT |
657 | #endif |
658 | ||
90d967e0 YL |
659 | /* user-defined highmem size */ |
660 | static unsigned int highmem_pages = -1; | |
661 | ||
662 | /* | |
663 | * highmem=size forces highmem to be exactly 'size' bytes. | |
664 | * This works even on boxes that have no highmem otherwise. | |
665 | * This also works to reduce highmem size on bigger boxes. | |
666 | */ | |
667 | static int __init parse_highmem(char *arg) | |
668 | { | |
669 | if (!arg) | |
670 | return -EINVAL; | |
671 | ||
672 | highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; | |
673 | return 0; | |
674 | } | |
675 | early_param("highmem", parse_highmem); | |
676 | ||
4769843b IM |
677 | #define MSG_HIGHMEM_TOO_BIG \ |
678 | "highmem size (%luMB) is bigger than pages available (%luMB)!\n" | |
679 | ||
680 | #define MSG_LOWMEM_TOO_SMALL \ | |
681 | "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n" | |
90d967e0 | 682 | /* |
4769843b IM |
683 | * All of RAM fits into lowmem - but if user wants highmem |
684 | * artificially via the highmem=x boot parameter then create | |
685 | * it: | |
90d967e0 | 686 | */ |
4769843b | 687 | void __init lowmem_pfn_init(void) |
90d967e0 | 688 | { |
346cafec | 689 | /* max_low_pfn is 0, we already have early_res support */ |
90d967e0 | 690 | max_low_pfn = max_pfn; |
d88316c2 | 691 | |
4769843b IM |
692 | if (highmem_pages == -1) |
693 | highmem_pages = 0; | |
694 | #ifdef CONFIG_HIGHMEM | |
695 | if (highmem_pages >= max_pfn) { | |
696 | printk(KERN_ERR MSG_HIGHMEM_TOO_BIG, | |
697 | pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); | |
698 | highmem_pages = 0; | |
699 | } | |
700 | if (highmem_pages) { | |
701 | if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) { | |
702 | printk(KERN_ERR MSG_LOWMEM_TOO_SMALL, | |
90d967e0 YL |
703 | pages_to_mb(highmem_pages)); |
704 | highmem_pages = 0; | |
705 | } | |
4769843b IM |
706 | max_low_pfn -= highmem_pages; |
707 | } | |
708 | #else | |
709 | if (highmem_pages) | |
710 | printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); | |
711 | #endif | |
712 | } | |
713 | ||
714 | #define MSG_HIGHMEM_TOO_SMALL \ | |
715 | "only %luMB highmem pages available, ignoring highmem size of %luMB!\n" | |
716 | ||
717 | #define MSG_HIGHMEM_TRIMMED \ | |
718 | "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n" | |
719 | /* | |
720 | * We have more RAM than fits into lowmem - we try to put it into | |
721 | * highmem, also taking the highmem=x boot parameter into account: | |
722 | */ | |
723 | void __init highmem_pfn_init(void) | |
724 | { | |
d88316c2 IM |
725 | max_low_pfn = MAXMEM_PFN; |
726 | ||
4769843b IM |
727 | if (highmem_pages == -1) |
728 | highmem_pages = max_pfn - MAXMEM_PFN; | |
729 | ||
730 | if (highmem_pages + MAXMEM_PFN < max_pfn) | |
731 | max_pfn = MAXMEM_PFN + highmem_pages; | |
732 | ||
733 | if (highmem_pages + MAXMEM_PFN > max_pfn) { | |
734 | printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL, | |
735 | pages_to_mb(max_pfn - MAXMEM_PFN), | |
736 | pages_to_mb(highmem_pages)); | |
737 | highmem_pages = 0; | |
738 | } | |
90d967e0 | 739 | #ifndef CONFIG_HIGHMEM |
4769843b IM |
740 | /* Maximum memory usable is what is directly addressable */ |
741 | printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20); | |
742 | if (max_pfn > MAX_NONPAE_PFN) | |
743 | printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n"); | |
744 | else | |
745 | printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); | |
746 | max_pfn = MAXMEM_PFN; | |
90d967e0 YL |
747 | #else /* !CONFIG_HIGHMEM */ |
748 | #ifndef CONFIG_HIGHMEM64G | |
4769843b IM |
749 | if (max_pfn > MAX_NONPAE_PFN) { |
750 | max_pfn = MAX_NONPAE_PFN; | |
751 | printk(KERN_WARNING MSG_HIGHMEM_TRIMMED); | |
752 | } | |
90d967e0 YL |
753 | #endif /* !CONFIG_HIGHMEM64G */ |
754 | #endif /* !CONFIG_HIGHMEM */ | |
4769843b IM |
755 | } |
756 | ||
757 | /* | |
758 | * Determine low and high memory ranges: | |
759 | */ | |
760 | void __init find_low_pfn_range(void) | |
761 | { | |
762 | /* it could update max_pfn */ | |
763 | ||
d88316c2 | 764 | if (max_pfn <= MAXMEM_PFN) |
4769843b | 765 | lowmem_pfn_init(); |
d88316c2 IM |
766 | else |
767 | highmem_pfn_init(); | |
90d967e0 YL |
768 | } |
769 | ||
b2ac82a0 | 770 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
2ec65f8b | 771 | void __init initmem_init(unsigned long start_pfn, |
b2ac82a0 YL |
772 | unsigned long end_pfn) |
773 | { | |
b2ac82a0 YL |
774 | #ifdef CONFIG_HIGHMEM |
775 | highstart_pfn = highend_pfn = max_pfn; | |
776 | if (max_pfn > max_low_pfn) | |
777 | highstart_pfn = max_low_pfn; | |
778 | memory_present(0, 0, highend_pfn); | |
cb95a13a | 779 | e820_register_active_regions(0, 0, highend_pfn); |
b2ac82a0 YL |
780 | printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", |
781 | pages_to_mb(highend_pfn - highstart_pfn)); | |
782 | num_physpages = highend_pfn; | |
783 | high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; | |
784 | #else | |
785 | memory_present(0, 0, max_low_pfn); | |
cb95a13a | 786 | e820_register_active_regions(0, 0, max_low_pfn); |
b2ac82a0 YL |
787 | num_physpages = max_low_pfn; |
788 | high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; | |
789 | #endif | |
790 | #ifdef CONFIG_FLATMEM | |
791 | max_mapnr = num_physpages; | |
792 | #endif | |
793 | printk(KERN_NOTICE "%ldMB LOWMEM available.\n", | |
794 | pages_to_mb(max_low_pfn)); | |
795 | ||
796 | setup_bootmem_allocator(); | |
b2ac82a0 | 797 | } |
cb95a13a | 798 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
b2ac82a0 | 799 | |
cb95a13a | 800 | static void __init zone_sizes_init(void) |
b2ac82a0 YL |
801 | { |
802 | unsigned long max_zone_pfns[MAX_NR_ZONES]; | |
803 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
804 | max_zone_pfns[ZONE_DMA] = | |
805 | virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
806 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; | |
b2ac82a0 YL |
807 | #ifdef CONFIG_HIGHMEM |
808 | max_zone_pfns[ZONE_HIGHMEM] = highend_pfn; | |
b2ac82a0 YL |
809 | #endif |
810 | ||
811 | free_area_init_nodes(max_zone_pfns); | |
812 | } | |
b2ac82a0 | 813 | |
b2ac82a0 YL |
814 | void __init setup_bootmem_allocator(void) |
815 | { | |
816 | int i; | |
817 | unsigned long bootmap_size, bootmap; | |
818 | /* | |
819 | * Initialize the boot-time allocator (with low memory only): | |
820 | */ | |
821 | bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT; | |
822 | bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT, | |
823 | max_pfn_mapped<<PAGE_SHIFT, bootmap_size, | |
824 | PAGE_SIZE); | |
825 | if (bootmap == -1L) | |
826 | panic("Cannot find bootmem map of size %ld\n", bootmap_size); | |
827 | reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP"); | |
225c37d7 | 828 | |
346cafec YL |
829 | /* don't touch min_low_pfn */ |
830 | bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT, | |
831 | min_low_pfn, max_low_pfn); | |
b2ac82a0 YL |
832 | printk(KERN_INFO " mapped low ram: 0 - %08lx\n", |
833 | max_pfn_mapped<<PAGE_SHIFT); | |
834 | printk(KERN_INFO " low ram: %08lx - %08lx\n", | |
835 | min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT); | |
836 | printk(KERN_INFO " bootmap %08lx - %08lx\n", | |
837 | bootmap, bootmap + bootmap_size); | |
838 | for_each_online_node(i) | |
839 | free_bootmem_with_active_regions(i, max_low_pfn); | |
840 | early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT); | |
841 | ||
4e29684c | 842 | after_init_bootmem = 1; |
b2ac82a0 YL |
843 | } |
844 | ||
0b8fdcbc | 845 | static void __init find_early_table_space(unsigned long end, int use_pse) |
4e29684c | 846 | { |
7482b0e9 | 847 | unsigned long puds, pmds, ptes, tables, start; |
4e29684c YL |
848 | |
849 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; | |
850 | tables = PAGE_ALIGN(puds * sizeof(pud_t)); | |
851 | ||
852 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; | |
853 | tables += PAGE_ALIGN(pmds * sizeof(pmd_t)); | |
854 | ||
0b8fdcbc | 855 | if (use_pse) { |
7482b0e9 | 856 | unsigned long extra; |
a04ad82d YL |
857 | |
858 | extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT); | |
859 | extra += PMD_SIZE; | |
7482b0e9 YL |
860 | ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT; |
861 | } else | |
862 | ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
863 | ||
864 | tables += PAGE_ALIGN(ptes * sizeof(pte_t)); | |
8207c257 | 865 | |
a04ad82d | 866 | /* for fixmap */ |
a3c6018e | 867 | tables += PAGE_ALIGN(__end_of_fixed_addresses * sizeof(pte_t)); |
a04ad82d | 868 | |
4e29684c YL |
869 | /* |
870 | * RED-PEN putting page tables only on node 0 could | |
871 | * cause a hotspot and fill up ZONE_DMA. The page tables | |
872 | * need roughly 0.5KB per GB. | |
873 | */ | |
874 | start = 0x7000; | |
875 | table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT, | |
876 | tables, PAGE_SIZE); | |
877 | if (table_start == -1UL) | |
878 | panic("Cannot find space for the kernel page tables"); | |
879 | ||
880 | table_start >>= PAGE_SHIFT; | |
881 | table_end = table_start; | |
882 | table_top = table_start + (tables>>PAGE_SHIFT); | |
883 | ||
884 | printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n", | |
885 | end, table_start << PAGE_SHIFT, | |
886 | (table_start << PAGE_SHIFT) + tables); | |
887 | } | |
888 | ||
889 | unsigned long __init_refok init_memory_mapping(unsigned long start, | |
890 | unsigned long end) | |
891 | { | |
892 | pgd_t *pgd_base = swapper_pg_dir; | |
a04ad82d YL |
893 | unsigned long start_pfn, end_pfn; |
894 | unsigned long big_page_start; | |
0b8fdcbc SS |
895 | #ifdef CONFIG_DEBUG_PAGEALLOC |
896 | /* | |
897 | * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages. | |
898 | * This will simplify cpa(), which otherwise needs to support splitting | |
899 | * large pages into small in interrupt context, etc. | |
900 | */ | |
901 | int use_pse = 0; | |
902 | #else | |
903 | int use_pse = cpu_has_pse; | |
904 | #endif | |
4e29684c YL |
905 | |
906 | /* | |
907 | * Find space for the kernel direct mapping tables. | |
908 | */ | |
909 | if (!after_init_bootmem) | |
0b8fdcbc | 910 | find_early_table_space(end, use_pse); |
4e29684c YL |
911 | |
912 | #ifdef CONFIG_X86_PAE | |
913 | set_nx(); | |
914 | if (nx_enabled) | |
915 | printk(KERN_INFO "NX (Execute Disable) protection: active\n"); | |
916 | #endif | |
917 | ||
918 | /* Enable PSE if available */ | |
919 | if (cpu_has_pse) | |
920 | set_in_cr4(X86_CR4_PSE); | |
921 | ||
922 | /* Enable PGE if available */ | |
923 | if (cpu_has_pge) { | |
924 | set_in_cr4(X86_CR4_PGE); | |
ef5e94af | 925 | __supported_pte_mask |= _PAGE_GLOBAL; |
4e29684c YL |
926 | } |
927 | ||
a04ad82d YL |
928 | /* |
929 | * Don't use a large page for the first 2/4MB of memory | |
930 | * because there are often fixed size MTRRs in there | |
931 | * and overlapping MTRRs into large pages can cause | |
932 | * slowdowns. | |
933 | */ | |
934 | big_page_start = PMD_SIZE; | |
935 | ||
936 | if (start < big_page_start) { | |
937 | start_pfn = start >> PAGE_SHIFT; | |
938 | end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT); | |
939 | } else { | |
940 | /* head is not big page alignment ? */ | |
941 | start_pfn = start >> PAGE_SHIFT; | |
942 | end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT) | |
943 | << (PMD_SHIFT - PAGE_SHIFT); | |
944 | } | |
945 | if (start_pfn < end_pfn) | |
946 | kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0); | |
947 | ||
948 | /* big page range */ | |
949 | start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT) | |
950 | << (PMD_SHIFT - PAGE_SHIFT); | |
951 | if (start_pfn < (big_page_start >> PAGE_SHIFT)) | |
952 | start_pfn = big_page_start >> PAGE_SHIFT; | |
953 | end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT); | |
954 | if (start_pfn < end_pfn) | |
955 | kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, | |
0b8fdcbc | 956 | use_pse); |
a04ad82d YL |
957 | |
958 | /* tail is not big page alignment ? */ | |
959 | start_pfn = end_pfn; | |
960 | if (start_pfn > (big_page_start>>PAGE_SHIFT)) { | |
961 | end_pfn = end >> PAGE_SHIFT; | |
962 | if (start_pfn < end_pfn) | |
963 | kernel_physical_mapping_init(pgd_base, start_pfn, | |
964 | end_pfn, 0); | |
965 | } | |
4e29684c | 966 | |
e7b37895 YL |
967 | early_ioremap_page_table_range_init(pgd_base); |
968 | ||
4e29684c YL |
969 | load_cr3(swapper_pg_dir); |
970 | ||
971 | __flush_tlb_all(); | |
972 | ||
973 | if (!after_init_bootmem) | |
974 | reserve_early(table_start << PAGE_SHIFT, | |
975 | table_end << PAGE_SHIFT, "PGTABLE"); | |
976 | ||
caadbdce YL |
977 | if (!after_init_bootmem) |
978 | early_memtest(start, end); | |
979 | ||
4e29684c YL |
980 | return end >> PAGE_SHIFT; |
981 | } | |
982 | ||
e7b37895 | 983 | |
1da177e4 LT |
984 | /* |
985 | * paging_init() sets up the page tables - note that the first 8MB are | |
986 | * already mapped by head.S. | |
987 | * | |
988 | * This routines also unmaps the page at virtual kernel address 0, so | |
989 | * that we can trap those pesky NULL-reference errors in the kernel. | |
990 | */ | |
991 | void __init paging_init(void) | |
992 | { | |
1da177e4 LT |
993 | pagetable_init(); |
994 | ||
1da177e4 LT |
995 | __flush_tlb_all(); |
996 | ||
997 | kmap_init(); | |
11cd0bc1 YL |
998 | |
999 | /* | |
1000 | * NOTE: at this point the bootmem allocator is fully available. | |
1001 | */ | |
11cd0bc1 YL |
1002 | sparse_init(); |
1003 | zone_sizes_init(); | |
1da177e4 LT |
1004 | } |
1005 | ||
1006 | /* | |
1007 | * Test if the WP bit works in supervisor mode. It isn't supported on 386's | |
f7f17a67 DV |
1008 | * and also on some strange 486's. All 586+'s are OK. This used to involve |
1009 | * black magic jumps to work around some nasty CPU bugs, but fortunately the | |
1010 | * switch to using exceptions got rid of all that. | |
1da177e4 | 1011 | */ |
1da177e4 LT |
1012 | static void __init test_wp_bit(void) |
1013 | { | |
d7d119d7 IM |
1014 | printk(KERN_INFO |
1015 | "Checking if this processor honours the WP bit even in supervisor mode..."); | |
1da177e4 LT |
1016 | |
1017 | /* Any page-aligned address will do, the test is non-destructive */ | |
1018 | __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY); | |
1019 | boot_cpu_data.wp_works_ok = do_test_wp_bit(); | |
1020 | clear_fixmap(FIX_WP_TEST); | |
1021 | ||
1022 | if (!boot_cpu_data.wp_works_ok) { | |
d7d119d7 | 1023 | printk(KERN_CONT "No.\n"); |
1da177e4 | 1024 | #ifdef CONFIG_X86_WP_WORKS_OK |
d7d119d7 IM |
1025 | panic( |
1026 | "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!"); | |
1da177e4 LT |
1027 | #endif |
1028 | } else { | |
d7d119d7 | 1029 | printk(KERN_CONT "Ok.\n"); |
1da177e4 LT |
1030 | } |
1031 | } | |
1032 | ||
8550eb99 | 1033 | static struct kcore_list kcore_mem, kcore_vmalloc; |
1da177e4 LT |
1034 | |
1035 | void __init mem_init(void) | |
1036 | { | |
1da177e4 | 1037 | int codesize, reservedpages, datasize, initsize; |
cc9f7a0c | 1038 | int tmp; |
1da177e4 | 1039 | |
cfb80c9e JF |
1040 | pci_iommu_alloc(); |
1041 | ||
05b79bdc | 1042 | #ifdef CONFIG_FLATMEM |
8d8f3cbe | 1043 | BUG_ON(!mem_map); |
1da177e4 | 1044 | #endif |
1da177e4 LT |
1045 | /* this will put all low memory onto the freelists */ |
1046 | totalram_pages += free_all_bootmem(); | |
1047 | ||
1048 | reservedpages = 0; | |
1049 | for (tmp = 0; tmp < max_low_pfn; tmp++) | |
1050 | /* | |
8550eb99 | 1051 | * Only count reserved RAM pages: |
1da177e4 LT |
1052 | */ |
1053 | if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp))) | |
1054 | reservedpages++; | |
1055 | ||
cc9f7a0c | 1056 | set_highmem_pages_init(); |
1da177e4 LT |
1057 | |
1058 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
1059 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
1060 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
1061 | ||
8550eb99 IM |
1062 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
1063 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
1da177e4 LT |
1064 | VMALLOC_END-VMALLOC_START); |
1065 | ||
8550eb99 IM |
1066 | printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, " |
1067 | "%dk reserved, %dk data, %dk init, %ldk highmem)\n", | |
1da177e4 LT |
1068 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
1069 | num_physpages << (PAGE_SHIFT-10), | |
1070 | codesize >> 10, | |
1071 | reservedpages << (PAGE_SHIFT-10), | |
1072 | datasize >> 10, | |
1073 | initsize >> 10, | |
1074 | (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)) | |
1075 | ); | |
1076 | ||
d7d119d7 | 1077 | printk(KERN_INFO "virtual kernel memory layout:\n" |
8550eb99 | 1078 | " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" |
052e7994 | 1079 | #ifdef CONFIG_HIGHMEM |
8550eb99 | 1080 | " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n" |
052e7994 | 1081 | #endif |
8550eb99 IM |
1082 | " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" |
1083 | " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" | |
1084 | " .init : 0x%08lx - 0x%08lx (%4ld kB)\n" | |
1085 | " .data : 0x%08lx - 0x%08lx (%4ld kB)\n" | |
1086 | " .text : 0x%08lx - 0x%08lx (%4ld kB)\n", | |
1087 | FIXADDR_START, FIXADDR_TOP, | |
1088 | (FIXADDR_TOP - FIXADDR_START) >> 10, | |
052e7994 JF |
1089 | |
1090 | #ifdef CONFIG_HIGHMEM | |
8550eb99 IM |
1091 | PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, |
1092 | (LAST_PKMAP*PAGE_SIZE) >> 10, | |
052e7994 JF |
1093 | #endif |
1094 | ||
8550eb99 IM |
1095 | VMALLOC_START, VMALLOC_END, |
1096 | (VMALLOC_END - VMALLOC_START) >> 20, | |
052e7994 | 1097 | |
8550eb99 IM |
1098 | (unsigned long)__va(0), (unsigned long)high_memory, |
1099 | ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, | |
052e7994 | 1100 | |
8550eb99 IM |
1101 | (unsigned long)&__init_begin, (unsigned long)&__init_end, |
1102 | ((unsigned long)&__init_end - | |
1103 | (unsigned long)&__init_begin) >> 10, | |
052e7994 | 1104 | |
8550eb99 IM |
1105 | (unsigned long)&_etext, (unsigned long)&_edata, |
1106 | ((unsigned long)&_edata - (unsigned long)&_etext) >> 10, | |
052e7994 | 1107 | |
8550eb99 IM |
1108 | (unsigned long)&_text, (unsigned long)&_etext, |
1109 | ((unsigned long)&_etext - (unsigned long)&_text) >> 10); | |
052e7994 | 1110 | |
beeb4195 JB |
1111 | /* |
1112 | * Check boundaries twice: Some fundamental inconsistencies can | |
1113 | * be detected at build time already. | |
1114 | */ | |
1115 | #define __FIXADDR_TOP (-PAGE_SIZE) | |
1116 | #ifdef CONFIG_HIGHMEM | |
1117 | BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); | |
1118 | BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE); | |
1119 | #endif | |
1120 | #define high_memory (-128UL << 20) | |
1121 | BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END); | |
1122 | #undef high_memory | |
1123 | #undef __FIXADDR_TOP | |
1124 | ||
052e7994 | 1125 | #ifdef CONFIG_HIGHMEM |
8550eb99 IM |
1126 | BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); |
1127 | BUG_ON(VMALLOC_END > PKMAP_BASE); | |
052e7994 | 1128 | #endif |
beeb4195 | 1129 | BUG_ON(VMALLOC_START >= VMALLOC_END); |
8550eb99 | 1130 | BUG_ON((unsigned long)high_memory > VMALLOC_START); |
052e7994 | 1131 | |
1da177e4 LT |
1132 | if (boot_cpu_data.wp_works_ok < 0) |
1133 | test_wp_bit(); | |
1134 | ||
61165d7a | 1135 | save_pg_dir(); |
1da177e4 | 1136 | zap_low_mappings(); |
1da177e4 LT |
1137 | } |
1138 | ||
ad8f5797 | 1139 | #ifdef CONFIG_MEMORY_HOTPLUG |
bc02af93 | 1140 | int arch_add_memory(int nid, u64 start, u64 size) |
05039b92 | 1141 | { |
7c7e9425 | 1142 | struct pglist_data *pgdata = NODE_DATA(nid); |
776ed98b | 1143 | struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM; |
05039b92 DH |
1144 | unsigned long start_pfn = start >> PAGE_SHIFT; |
1145 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
1146 | ||
c04fc586 | 1147 | return __add_pages(nid, zone, start_pfn, nr_pages); |
05039b92 | 1148 | } |
9d99aaa3 | 1149 | #endif |
05039b92 | 1150 | |
1da177e4 LT |
1151 | /* |
1152 | * This function cannot be __init, since exceptions don't work in that | |
1153 | * section. Put this after the callers, so that it cannot be inlined. | |
1154 | */ | |
8550eb99 | 1155 | static noinline int do_test_wp_bit(void) |
1da177e4 LT |
1156 | { |
1157 | char tmp_reg; | |
1158 | int flag; | |
1159 | ||
1160 | __asm__ __volatile__( | |
8550eb99 IM |
1161 | " movb %0, %1 \n" |
1162 | "1: movb %1, %0 \n" | |
1163 | " xorl %2, %2 \n" | |
1da177e4 | 1164 | "2: \n" |
f832ff18 | 1165 | _ASM_EXTABLE(1b,2b) |
1da177e4 LT |
1166 | :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)), |
1167 | "=q" (tmp_reg), | |
1168 | "=r" (flag) | |
1169 | :"2" (1) | |
1170 | :"memory"); | |
8550eb99 | 1171 | |
1da177e4 LT |
1172 | return flag; |
1173 | } | |
1174 | ||
63aaf308 | 1175 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
1176 | const int rodata_test_data = 0xC3; |
1177 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
63aaf308 | 1178 | |
63aaf308 AV |
1179 | void mark_rodata_ro(void) |
1180 | { | |
6fb14755 JB |
1181 | unsigned long start = PFN_ALIGN(_text); |
1182 | unsigned long size = PFN_ALIGN(_etext) - start; | |
63aaf308 | 1183 | |
8f0f996e SR |
1184 | #ifndef CONFIG_DYNAMIC_FTRACE |
1185 | /* Dynamic tracing modifies the kernel text section */ | |
4e4eee0e MD |
1186 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
1187 | printk(KERN_INFO "Write protecting the kernel text: %luk\n", | |
1188 | size >> 10); | |
0c42f392 AK |
1189 | |
1190 | #ifdef CONFIG_CPA_DEBUG | |
4e4eee0e MD |
1191 | printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n", |
1192 | start, start+size); | |
1193 | set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT); | |
0c42f392 | 1194 | |
4e4eee0e MD |
1195 | printk(KERN_INFO "Testing CPA: write protecting again\n"); |
1196 | set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT); | |
602033ed | 1197 | #endif |
8f0f996e SR |
1198 | #endif /* CONFIG_DYNAMIC_FTRACE */ |
1199 | ||
6fb14755 JB |
1200 | start += size; |
1201 | size = (unsigned long)__end_rodata - start; | |
6d238cc4 | 1202 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
d7d119d7 IM |
1203 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
1204 | size >> 10); | |
edeed305 | 1205 | rodata_test(); |
63aaf308 | 1206 | |
0c42f392 | 1207 | #ifdef CONFIG_CPA_DEBUG |
d7d119d7 | 1208 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size); |
6d238cc4 | 1209 | set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); |
0c42f392 | 1210 | |
d7d119d7 | 1211 | printk(KERN_INFO "Testing CPA: write protecting again\n"); |
6d238cc4 | 1212 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
0c42f392 | 1213 | #endif |
63aaf308 AV |
1214 | } |
1215 | #endif | |
1216 | ||
9a0b5817 GH |
1217 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
1218 | { | |
ee01f112 IM |
1219 | #ifdef CONFIG_DEBUG_PAGEALLOC |
1220 | /* | |
1221 | * If debugging page accesses then do not free this memory but | |
1222 | * mark them not present - any buggy init-section access will | |
1223 | * create a kernel page fault: | |
1224 | */ | |
1225 | printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", | |
1226 | begin, PAGE_ALIGN(end)); | |
1227 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
1228 | #else | |
86f03989 IM |
1229 | unsigned long addr; |
1230 | ||
3c1df68b AV |
1231 | /* |
1232 | * We just marked the kernel text read only above, now that | |
1233 | * we are going to free part of that, we need to make that | |
1234 | * writeable first. | |
1235 | */ | |
1236 | set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); | |
1237 | ||
9a0b5817 | 1238 | for (addr = begin; addr < end; addr += PAGE_SIZE) { |
e3ebadd9 LT |
1239 | ClearPageReserved(virt_to_page(addr)); |
1240 | init_page_count(virt_to_page(addr)); | |
1241 | memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE); | |
1242 | free_page(addr); | |
9a0b5817 GH |
1243 | totalram_pages++; |
1244 | } | |
6fb14755 | 1245 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
ee01f112 | 1246 | #endif |
9a0b5817 GH |
1247 | } |
1248 | ||
1249 | void free_initmem(void) | |
1250 | { | |
1251 | free_init_pages("unused kernel memory", | |
e3ebadd9 LT |
1252 | (unsigned long)(&__init_begin), |
1253 | (unsigned long)(&__init_end)); | |
9a0b5817 | 1254 | } |
63aaf308 | 1255 | |
1da177e4 LT |
1256 | #ifdef CONFIG_BLK_DEV_INITRD |
1257 | void free_initrd_mem(unsigned long start, unsigned long end) | |
1258 | { | |
e3ebadd9 | 1259 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
1260 | } |
1261 | #endif | |
d2dbf343 YL |
1262 | |
1263 | int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, | |
1264 | int flags) | |
1265 | { | |
1266 | return reserve_bootmem(phys, len, flags); | |
1267 | } |