Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/i386/mm/init.c | |
3 | * | |
4 | * Copyright (C) 1995 Linus Torvalds | |
5 | * | |
6 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
7 | */ | |
8 | ||
9 | #include <linux/config.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/signal.h> | |
12 | #include <linux/sched.h> | |
13 | #include <linux/kernel.h> | |
14 | #include <linux/errno.h> | |
15 | #include <linux/string.h> | |
16 | #include <linux/types.h> | |
17 | #include <linux/ptrace.h> | |
18 | #include <linux/mman.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/hugetlb.h> | |
21 | #include <linux/swap.h> | |
22 | #include <linux/smp.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/highmem.h> | |
25 | #include <linux/pagemap.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> | |
30 | #include <linux/efi.h> | |
05039b92 | 31 | #include <linux/memory_hotplug.h> |
27d99f7e | 32 | #include <linux/initrd.h> |
55b2355e | 33 | #include <linux/cpumask.h> |
1da177e4 LT |
34 | |
35 | #include <asm/processor.h> | |
36 | #include <asm/system.h> | |
37 | #include <asm/uaccess.h> | |
38 | #include <asm/pgtable.h> | |
39 | #include <asm/dma.h> | |
40 | #include <asm/fixmap.h> | |
41 | #include <asm/e820.h> | |
42 | #include <asm/apic.h> | |
43 | #include <asm/tlb.h> | |
44 | #include <asm/tlbflush.h> | |
45 | #include <asm/sections.h> | |
46 | ||
47 | unsigned int __VMALLOC_RESERVE = 128 << 20; | |
48 | ||
49 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); | |
50 | unsigned long highstart_pfn, highend_pfn; | |
51 | ||
52 | static int noinline do_test_wp_bit(void); | |
53 | ||
54 | /* | |
55 | * Creates a middle page table and puts a pointer to it in the | |
56 | * given global directory entry. This only returns the gd entry | |
57 | * in non-PAE compilation mode, since the middle layer is folded. | |
58 | */ | |
59 | static pmd_t * __init one_md_table_init(pgd_t *pgd) | |
60 | { | |
61 | pud_t *pud; | |
62 | pmd_t *pmd_table; | |
63 | ||
64 | #ifdef CONFIG_X86_PAE | |
65 | pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE); | |
66 | set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); | |
67 | pud = pud_offset(pgd, 0); | |
68 | if (pmd_table != pmd_offset(pud, 0)) | |
69 | BUG(); | |
70 | #else | |
71 | pud = pud_offset(pgd, 0); | |
72 | pmd_table = pmd_offset(pud, 0); | |
73 | #endif | |
74 | ||
75 | return pmd_table; | |
76 | } | |
77 | ||
78 | /* | |
79 | * Create a page table and place a pointer to it in a middle page | |
80 | * directory entry. | |
81 | */ | |
82 | static pte_t * __init one_page_table_init(pmd_t *pmd) | |
83 | { | |
84 | if (pmd_none(*pmd)) { | |
85 | pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE); | |
86 | set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); | |
87 | if (page_table != pte_offset_kernel(pmd, 0)) | |
88 | BUG(); | |
89 | ||
90 | return page_table; | |
91 | } | |
92 | ||
93 | return pte_offset_kernel(pmd, 0); | |
94 | } | |
95 | ||
96 | /* | |
97 | * This function initializes a certain range of kernel virtual memory | |
98 | * with new bootmem page tables, everywhere page tables are missing in | |
99 | * the given range. | |
100 | */ | |
101 | ||
102 | /* | |
103 | * NOTE: The pagetables are allocated contiguous on the physical space | |
104 | * so we can cache the place of the first one and move around without | |
105 | * checking the pgd every time. | |
106 | */ | |
107 | static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base) | |
108 | { | |
109 | pgd_t *pgd; | |
110 | pud_t *pud; | |
111 | pmd_t *pmd; | |
112 | int pgd_idx, pmd_idx; | |
113 | unsigned long vaddr; | |
114 | ||
115 | vaddr = start; | |
116 | pgd_idx = pgd_index(vaddr); | |
117 | pmd_idx = pmd_index(vaddr); | |
118 | pgd = pgd_base + pgd_idx; | |
119 | ||
120 | for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { | |
121 | if (pgd_none(*pgd)) | |
122 | one_md_table_init(pgd); | |
123 | pud = pud_offset(pgd, vaddr); | |
124 | pmd = pmd_offset(pud, vaddr); | |
125 | for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) { | |
126 | if (pmd_none(*pmd)) | |
127 | one_page_table_init(pmd); | |
128 | ||
129 | vaddr += PMD_SIZE; | |
130 | } | |
131 | pmd_idx = 0; | |
132 | } | |
133 | } | |
134 | ||
135 | static inline int is_kernel_text(unsigned long addr) | |
136 | { | |
137 | if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end) | |
138 | return 1; | |
139 | return 0; | |
140 | } | |
141 | ||
142 | /* | |
143 | * This maps the physical memory to kernel virtual address space, a total | |
144 | * of max_low_pfn pages, by creating page tables starting from address | |
145 | * PAGE_OFFSET. | |
146 | */ | |
147 | static void __init kernel_physical_mapping_init(pgd_t *pgd_base) | |
148 | { | |
149 | unsigned long pfn; | |
150 | pgd_t *pgd; | |
151 | pmd_t *pmd; | |
152 | pte_t *pte; | |
153 | int pgd_idx, pmd_idx, pte_ofs; | |
154 | ||
155 | pgd_idx = pgd_index(PAGE_OFFSET); | |
156 | pgd = pgd_base + pgd_idx; | |
157 | pfn = 0; | |
158 | ||
159 | for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { | |
160 | pmd = one_md_table_init(pgd); | |
161 | if (pfn >= max_low_pfn) | |
162 | continue; | |
163 | for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) { | |
164 | unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET; | |
165 | ||
166 | /* Map with big pages if possible, otherwise create normal page tables. */ | |
167 | if (cpu_has_pse) { | |
168 | unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1; | |
169 | ||
170 | if (is_kernel_text(address) || is_kernel_text(address2)) | |
171 | set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC)); | |
172 | else | |
173 | set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE)); | |
174 | pfn += PTRS_PER_PTE; | |
175 | } else { | |
176 | pte = one_page_table_init(pmd); | |
177 | ||
178 | for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) { | |
179 | if (is_kernel_text(address)) | |
180 | set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC)); | |
181 | else | |
182 | set_pte(pte, pfn_pte(pfn, PAGE_KERNEL)); | |
183 | } | |
184 | } | |
185 | } | |
186 | } | |
187 | } | |
188 | ||
189 | static inline int page_kills_ppro(unsigned long pagenr) | |
190 | { | |
191 | if (pagenr >= 0x70000 && pagenr <= 0x7003F) | |
192 | return 1; | |
193 | return 0; | |
194 | } | |
195 | ||
196 | extern int is_available_memory(efi_memory_desc_t *); | |
197 | ||
5b505b90 | 198 | int page_is_ram(unsigned long pagenr) |
1da177e4 LT |
199 | { |
200 | int i; | |
201 | unsigned long addr, end; | |
202 | ||
203 | if (efi_enabled) { | |
204 | efi_memory_desc_t *md; | |
7ae65fd3 | 205 | void *p; |
1da177e4 | 206 | |
7ae65fd3 MT |
207 | for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { |
208 | md = p; | |
1da177e4 LT |
209 | if (!is_available_memory(md)) |
210 | continue; | |
211 | addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT; | |
212 | end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT; | |
213 | ||
214 | if ((pagenr >= addr) && (pagenr < end)) | |
215 | return 1; | |
216 | } | |
217 | return 0; | |
218 | } | |
219 | ||
220 | for (i = 0; i < e820.nr_map; i++) { | |
221 | ||
222 | if (e820.map[i].type != E820_RAM) /* not usable memory */ | |
223 | continue; | |
224 | /* | |
225 | * !!!FIXME!!! Some BIOSen report areas as RAM that | |
226 | * are not. Notably the 640->1Mb area. We need a sanity | |
227 | * check here. | |
228 | */ | |
229 | addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT; | |
230 | end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT; | |
231 | if ((pagenr >= addr) && (pagenr < end)) | |
232 | return 1; | |
233 | } | |
234 | return 0; | |
235 | } | |
236 | ||
237 | #ifdef CONFIG_HIGHMEM | |
238 | pte_t *kmap_pte; | |
239 | pgprot_t kmap_prot; | |
240 | ||
241 | #define kmap_get_fixmap_pte(vaddr) \ | |
242 | pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr)) | |
243 | ||
244 | static void __init kmap_init(void) | |
245 | { | |
246 | unsigned long kmap_vstart; | |
247 | ||
248 | /* cache the first kmap pte */ | |
249 | kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); | |
250 | kmap_pte = kmap_get_fixmap_pte(kmap_vstart); | |
251 | ||
252 | kmap_prot = PAGE_KERNEL; | |
253 | } | |
254 | ||
255 | static void __init permanent_kmaps_init(pgd_t *pgd_base) | |
256 | { | |
257 | pgd_t *pgd; | |
258 | pud_t *pud; | |
259 | pmd_t *pmd; | |
260 | pte_t *pte; | |
261 | unsigned long vaddr; | |
262 | ||
263 | vaddr = PKMAP_BASE; | |
264 | page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); | |
265 | ||
266 | pgd = swapper_pg_dir + pgd_index(vaddr); | |
267 | pud = pud_offset(pgd, vaddr); | |
268 | pmd = pmd_offset(pud, vaddr); | |
269 | pte = pte_offset_kernel(pmd, vaddr); | |
270 | pkmap_page_table = pte; | |
271 | } | |
272 | ||
c09b4240 | 273 | static void __meminit free_new_highpage(struct page *page) |
05039b92 | 274 | { |
7835e98b | 275 | init_page_count(page); |
05039b92 DH |
276 | __free_page(page); |
277 | totalhigh_pages++; | |
278 | } | |
279 | ||
280 | void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro) | |
1da177e4 LT |
281 | { |
282 | if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) { | |
283 | ClearPageReserved(page); | |
05039b92 | 284 | free_new_highpage(page); |
1da177e4 LT |
285 | } else |
286 | SetPageReserved(page); | |
287 | } | |
288 | ||
05039b92 DH |
289 | static int add_one_highpage_hotplug(struct page *page, unsigned long pfn) |
290 | { | |
291 | free_new_highpage(page); | |
292 | totalram_pages++; | |
293 | #ifdef CONFIG_FLATMEM | |
294 | max_mapnr = max(pfn, max_mapnr); | |
295 | #endif | |
296 | num_physpages++; | |
297 | return 0; | |
298 | } | |
299 | ||
300 | /* | |
301 | * Not currently handling the NUMA case. | |
302 | * Assuming single node and all memory that | |
303 | * has been added dynamically that would be | |
304 | * onlined here is in HIGHMEM | |
305 | */ | |
306 | void online_page(struct page *page) | |
307 | { | |
308 | ClearPageReserved(page); | |
309 | add_one_highpage_hotplug(page, page_to_pfn(page)); | |
310 | } | |
311 | ||
312 | ||
05b79bdc AW |
313 | #ifdef CONFIG_NUMA |
314 | extern void set_highmem_pages_init(int); | |
315 | #else | |
1da177e4 LT |
316 | static void __init set_highmem_pages_init(int bad_ppro) |
317 | { | |
318 | int pfn; | |
319 | for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) | |
05039b92 | 320 | add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro); |
1da177e4 LT |
321 | totalram_pages += totalhigh_pages; |
322 | } | |
05b79bdc | 323 | #endif /* CONFIG_FLATMEM */ |
1da177e4 LT |
324 | |
325 | #else | |
326 | #define kmap_init() do { } while (0) | |
327 | #define permanent_kmaps_init(pgd_base) do { } while (0) | |
328 | #define set_highmem_pages_init(bad_ppro) do { } while (0) | |
329 | #endif /* CONFIG_HIGHMEM */ | |
330 | ||
331 | unsigned long long __PAGE_KERNEL = _PAGE_KERNEL; | |
129f6946 | 332 | EXPORT_SYMBOL(__PAGE_KERNEL); |
1da177e4 LT |
333 | unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC; |
334 | ||
05b79bdc | 335 | #ifdef CONFIG_NUMA |
1da177e4 | 336 | extern void __init remap_numa_kva(void); |
05b79bdc AW |
337 | #else |
338 | #define remap_numa_kva() do {} while (0) | |
1da177e4 LT |
339 | #endif |
340 | ||
341 | static void __init pagetable_init (void) | |
342 | { | |
343 | unsigned long vaddr; | |
344 | pgd_t *pgd_base = swapper_pg_dir; | |
345 | ||
346 | #ifdef CONFIG_X86_PAE | |
347 | int i; | |
348 | /* Init entries of the first-level page table to the zero page */ | |
349 | for (i = 0; i < PTRS_PER_PGD; i++) | |
350 | set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT)); | |
351 | #endif | |
352 | ||
353 | /* Enable PSE if available */ | |
354 | if (cpu_has_pse) { | |
355 | set_in_cr4(X86_CR4_PSE); | |
356 | } | |
357 | ||
358 | /* Enable PGE if available */ | |
359 | if (cpu_has_pge) { | |
360 | set_in_cr4(X86_CR4_PGE); | |
361 | __PAGE_KERNEL |= _PAGE_GLOBAL; | |
362 | __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL; | |
363 | } | |
364 | ||
365 | kernel_physical_mapping_init(pgd_base); | |
366 | remap_numa_kva(); | |
367 | ||
368 | /* | |
369 | * Fixed mappings, only the page table structure has to be | |
370 | * created - mappings will be set by set_fixmap(): | |
371 | */ | |
372 | vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; | |
373 | page_table_range_init(vaddr, 0, pgd_base); | |
374 | ||
375 | permanent_kmaps_init(pgd_base); | |
376 | ||
377 | #ifdef CONFIG_X86_PAE | |
378 | /* | |
379 | * Add low memory identity-mappings - SMP needs it when | |
380 | * starting up on an AP from real-mode. In the non-PAE | |
381 | * case we already have these mappings through head.S. | |
382 | * All user-space mappings are explicitly cleared after | |
383 | * SMP startup. | |
384 | */ | |
c9b02a24 | 385 | set_pgd(&pgd_base[0], pgd_base[USER_PTRS_PER_PGD]); |
1da177e4 LT |
386 | #endif |
387 | } | |
388 | ||
55b2355e | 389 | #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP) |
1da177e4 LT |
390 | /* |
391 | * Swap suspend & friends need this for resume because things like the intel-agp | |
392 | * driver might have split up a kernel 4MB mapping. | |
393 | */ | |
394 | char __nosavedata swsusp_pg_dir[PAGE_SIZE] | |
395 | __attribute__ ((aligned (PAGE_SIZE))); | |
396 | ||
397 | static inline void save_pg_dir(void) | |
398 | { | |
399 | memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE); | |
400 | } | |
401 | #else | |
402 | static inline void save_pg_dir(void) | |
403 | { | |
404 | } | |
405 | #endif | |
406 | ||
407 | void zap_low_mappings (void) | |
408 | { | |
409 | int i; | |
410 | ||
411 | save_pg_dir(); | |
412 | ||
413 | /* | |
414 | * Zap initial low-memory mappings. | |
415 | * | |
416 | * Note that "pgd_clear()" doesn't do it for | |
417 | * us, because pgd_clear() is a no-op on i386. | |
418 | */ | |
419 | for (i = 0; i < USER_PTRS_PER_PGD; i++) | |
420 | #ifdef CONFIG_X86_PAE | |
421 | set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page))); | |
422 | #else | |
423 | set_pgd(swapper_pg_dir+i, __pgd(0)); | |
424 | #endif | |
425 | flush_tlb_all(); | |
426 | } | |
427 | ||
428 | static int disable_nx __initdata = 0; | |
6c231b7b | 429 | u64 __supported_pte_mask __read_mostly = ~_PAGE_NX; |
1da177e4 LT |
430 | |
431 | /* | |
432 | * noexec = on|off | |
433 | * | |
434 | * Control non executable mappings. | |
435 | * | |
436 | * on Enable | |
437 | * off Disable | |
438 | */ | |
439 | void __init noexec_setup(const char *str) | |
440 | { | |
441 | if (!strncmp(str, "on",2) && cpu_has_nx) { | |
442 | __supported_pte_mask |= _PAGE_NX; | |
443 | disable_nx = 0; | |
444 | } else if (!strncmp(str,"off",3)) { | |
445 | disable_nx = 1; | |
446 | __supported_pte_mask &= ~_PAGE_NX; | |
447 | } | |
448 | } | |
449 | ||
450 | int nx_enabled = 0; | |
451 | #ifdef CONFIG_X86_PAE | |
452 | ||
453 | static void __init set_nx(void) | |
454 | { | |
455 | unsigned int v[4], l, h; | |
456 | ||
457 | if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) { | |
458 | cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]); | |
459 | if ((v[3] & (1 << 20)) && !disable_nx) { | |
460 | rdmsr(MSR_EFER, l, h); | |
461 | l |= EFER_NX; | |
462 | wrmsr(MSR_EFER, l, h); | |
463 | nx_enabled = 1; | |
464 | __supported_pte_mask |= _PAGE_NX; | |
465 | } | |
466 | } | |
467 | } | |
468 | ||
469 | /* | |
470 | * Enables/disables executability of a given kernel page and | |
471 | * returns the previous setting. | |
472 | */ | |
473 | int __init set_kernel_exec(unsigned long vaddr, int enable) | |
474 | { | |
475 | pte_t *pte; | |
476 | int ret = 1; | |
477 | ||
478 | if (!nx_enabled) | |
479 | goto out; | |
480 | ||
481 | pte = lookup_address(vaddr); | |
482 | BUG_ON(!pte); | |
483 | ||
484 | if (!pte_exec_kernel(*pte)) | |
485 | ret = 0; | |
486 | ||
487 | if (enable) | |
488 | pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32)); | |
489 | else | |
490 | pte->pte_high |= 1 << (_PAGE_BIT_NX - 32); | |
491 | __flush_tlb_all(); | |
492 | out: | |
493 | return ret; | |
494 | } | |
495 | ||
496 | #endif | |
497 | ||
498 | /* | |
499 | * paging_init() sets up the page tables - note that the first 8MB are | |
500 | * already mapped by head.S. | |
501 | * | |
502 | * This routines also unmaps the page at virtual kernel address 0, so | |
503 | * that we can trap those pesky NULL-reference errors in the kernel. | |
504 | */ | |
505 | void __init paging_init(void) | |
506 | { | |
507 | #ifdef CONFIG_X86_PAE | |
508 | set_nx(); | |
509 | if (nx_enabled) | |
510 | printk("NX (Execute Disable) protection: active\n"); | |
511 | #endif | |
512 | ||
513 | pagetable_init(); | |
514 | ||
515 | load_cr3(swapper_pg_dir); | |
516 | ||
517 | #ifdef CONFIG_X86_PAE | |
518 | /* | |
519 | * We will bail out later - printk doesn't work right now so | |
520 | * the user would just see a hanging kernel. | |
521 | */ | |
522 | if (cpu_has_pae) | |
523 | set_in_cr4(X86_CR4_PAE); | |
524 | #endif | |
525 | __flush_tlb_all(); | |
526 | ||
527 | kmap_init(); | |
528 | } | |
529 | ||
530 | /* | |
531 | * Test if the WP bit works in supervisor mode. It isn't supported on 386's | |
532 | * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This | |
533 | * used to involve black magic jumps to work around some nasty CPU bugs, | |
534 | * but fortunately the switch to using exceptions got rid of all that. | |
535 | */ | |
536 | ||
537 | static void __init test_wp_bit(void) | |
538 | { | |
539 | printk("Checking if this processor honours the WP bit even in supervisor mode... "); | |
540 | ||
541 | /* Any page-aligned address will do, the test is non-destructive */ | |
542 | __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY); | |
543 | boot_cpu_data.wp_works_ok = do_test_wp_bit(); | |
544 | clear_fixmap(FIX_WP_TEST); | |
545 | ||
546 | if (!boot_cpu_data.wp_works_ok) { | |
547 | printk("No.\n"); | |
548 | #ifdef CONFIG_X86_WP_WORKS_OK | |
549 | panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!"); | |
550 | #endif | |
551 | } else { | |
552 | printk("Ok.\n"); | |
553 | } | |
554 | } | |
555 | ||
556 | static void __init set_max_mapnr_init(void) | |
557 | { | |
558 | #ifdef CONFIG_HIGHMEM | |
559 | num_physpages = highend_pfn; | |
560 | #else | |
561 | num_physpages = max_low_pfn; | |
562 | #endif | |
05b79bdc | 563 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
564 | max_mapnr = num_physpages; |
565 | #endif | |
566 | } | |
567 | ||
568 | static struct kcore_list kcore_mem, kcore_vmalloc; | |
569 | ||
570 | void __init mem_init(void) | |
571 | { | |
572 | extern int ppro_with_ram_bug(void); | |
573 | int codesize, reservedpages, datasize, initsize; | |
574 | int tmp; | |
575 | int bad_ppro; | |
576 | ||
05b79bdc | 577 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
578 | if (!mem_map) |
579 | BUG(); | |
580 | #endif | |
581 | ||
582 | bad_ppro = ppro_with_ram_bug(); | |
583 | ||
584 | #ifdef CONFIG_HIGHMEM | |
585 | /* check that fixmap and pkmap do not overlap */ | |
586 | if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) { | |
587 | printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n"); | |
588 | printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n", | |
589 | PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START); | |
590 | BUG(); | |
591 | } | |
592 | #endif | |
593 | ||
594 | set_max_mapnr_init(); | |
595 | ||
596 | #ifdef CONFIG_HIGHMEM | |
597 | high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; | |
598 | #else | |
599 | high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; | |
600 | #endif | |
601 | ||
602 | /* this will put all low memory onto the freelists */ | |
603 | totalram_pages += free_all_bootmem(); | |
604 | ||
605 | reservedpages = 0; | |
606 | for (tmp = 0; tmp < max_low_pfn; tmp++) | |
607 | /* | |
608 | * Only count reserved RAM pages | |
609 | */ | |
610 | if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp))) | |
611 | reservedpages++; | |
612 | ||
613 | set_highmem_pages_init(bad_ppro); | |
614 | ||
615 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
616 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
617 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
618 | ||
619 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); | |
620 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
621 | VMALLOC_END-VMALLOC_START); | |
622 | ||
623 | printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n", | |
624 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), | |
625 | num_physpages << (PAGE_SHIFT-10), | |
626 | codesize >> 10, | |
627 | reservedpages << (PAGE_SHIFT-10), | |
628 | datasize >> 10, | |
629 | initsize >> 10, | |
630 | (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)) | |
631 | ); | |
632 | ||
633 | #ifdef CONFIG_X86_PAE | |
634 | if (!cpu_has_pae) | |
635 | panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!"); | |
636 | #endif | |
637 | if (boot_cpu_data.wp_works_ok < 0) | |
638 | test_wp_bit(); | |
639 | ||
640 | /* | |
641 | * Subtle. SMP is doing it's boot stuff late (because it has to | |
642 | * fork idle threads) - but it also needs low mappings for the | |
643 | * protected-mode entry to work. We zap these entries only after | |
644 | * the WP-bit has been tested. | |
645 | */ | |
646 | #ifndef CONFIG_SMP | |
647 | zap_low_mappings(); | |
648 | #endif | |
649 | } | |
650 | ||
05039b92 DH |
651 | /* |
652 | * this is for the non-NUMA, single node SMP system case. | |
653 | * Specifically, in the case of x86, we will always add | |
654 | * memory to the highmem for now. | |
655 | */ | |
ad8f5797 | 656 | #ifdef CONFIG_MEMORY_HOTPLUG |
05039b92 | 657 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
bc02af93 | 658 | int arch_add_memory(int nid, u64 start, u64 size) |
05039b92 DH |
659 | { |
660 | struct pglist_data *pgdata = &contig_page_data; | |
661 | struct zone *zone = pgdata->node_zones + MAX_NR_ZONES-1; | |
662 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
663 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
664 | ||
665 | return __add_pages(zone, start_pfn, nr_pages); | |
666 | } | |
667 | ||
668 | int remove_memory(u64 start, u64 size) | |
669 | { | |
670 | return -EINVAL; | |
671 | } | |
672 | #endif | |
9d99aaa3 | 673 | #endif |
05039b92 | 674 | |
1da177e4 LT |
675 | kmem_cache_t *pgd_cache; |
676 | kmem_cache_t *pmd_cache; | |
677 | ||
678 | void __init pgtable_cache_init(void) | |
679 | { | |
680 | if (PTRS_PER_PMD > 1) { | |
681 | pmd_cache = kmem_cache_create("pmd", | |
682 | PTRS_PER_PMD*sizeof(pmd_t), | |
683 | PTRS_PER_PMD*sizeof(pmd_t), | |
684 | 0, | |
685 | pmd_ctor, | |
686 | NULL); | |
687 | if (!pmd_cache) | |
688 | panic("pgtable_cache_init(): cannot create pmd cache"); | |
689 | } | |
690 | pgd_cache = kmem_cache_create("pgd", | |
691 | PTRS_PER_PGD*sizeof(pgd_t), | |
692 | PTRS_PER_PGD*sizeof(pgd_t), | |
693 | 0, | |
694 | pgd_ctor, | |
695 | PTRS_PER_PMD == 1 ? pgd_dtor : NULL); | |
696 | if (!pgd_cache) | |
697 | panic("pgtable_cache_init(): Cannot create pgd cache"); | |
698 | } | |
699 | ||
700 | /* | |
701 | * This function cannot be __init, since exceptions don't work in that | |
702 | * section. Put this after the callers, so that it cannot be inlined. | |
703 | */ | |
704 | static int noinline do_test_wp_bit(void) | |
705 | { | |
706 | char tmp_reg; | |
707 | int flag; | |
708 | ||
709 | __asm__ __volatile__( | |
710 | " movb %0,%1 \n" | |
711 | "1: movb %1,%0 \n" | |
712 | " xorl %2,%2 \n" | |
713 | "2: \n" | |
714 | ".section __ex_table,\"a\"\n" | |
715 | " .align 4 \n" | |
716 | " .long 1b,2b \n" | |
717 | ".previous \n" | |
718 | :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)), | |
719 | "=q" (tmp_reg), | |
720 | "=r" (flag) | |
721 | :"2" (1) | |
722 | :"memory"); | |
723 | ||
724 | return flag; | |
725 | } | |
726 | ||
63aaf308 AV |
727 | #ifdef CONFIG_DEBUG_RODATA |
728 | ||
729 | extern char __start_rodata, __end_rodata; | |
730 | void mark_rodata_ro(void) | |
731 | { | |
732 | unsigned long addr = (unsigned long)&__start_rodata; | |
733 | ||
734 | for (; addr < (unsigned long)&__end_rodata; addr += PAGE_SIZE) | |
735 | change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO); | |
736 | ||
737 | printk ("Write protecting the kernel read-only data: %luk\n", | |
738 | (unsigned long)(&__end_rodata - &__start_rodata) >> 10); | |
739 | ||
740 | /* | |
741 | * change_page_attr() requires a global_flush_tlb() call after it. | |
742 | * We do this after the printk so that if something went wrong in the | |
743 | * change, the printk gets out at least to give a better debug hint | |
744 | * of who is the culprit. | |
745 | */ | |
746 | global_flush_tlb(); | |
747 | } | |
748 | #endif | |
749 | ||
9a0b5817 GH |
750 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
751 | { | |
752 | unsigned long addr; | |
753 | ||
754 | for (addr = begin; addr < end; addr += PAGE_SIZE) { | |
755 | ClearPageReserved(virt_to_page(addr)); | |
756 | init_page_count(virt_to_page(addr)); | |
c9cf5528 | 757 | memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE); |
9a0b5817 GH |
758 | free_page(addr); |
759 | totalram_pages++; | |
760 | } | |
761 | printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10); | |
762 | } | |
763 | ||
764 | void free_initmem(void) | |
765 | { | |
766 | free_init_pages("unused kernel memory", | |
767 | (unsigned long)(&__init_begin), | |
768 | (unsigned long)(&__init_end)); | |
769 | } | |
63aaf308 | 770 | |
1da177e4 LT |
771 | #ifdef CONFIG_BLK_DEV_INITRD |
772 | void free_initrd_mem(unsigned long start, unsigned long end) | |
773 | { | |
9a0b5817 | 774 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
775 | } |
776 | #endif | |
9a0b5817 | 777 |