]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/x86_64/mm/init.c | |
3 | * | |
4 | * Copyright (C) 1995 Linus Torvalds | |
5 | * Copyright (C) 2000 Pavel Machek <pavel@suse.cz> | |
6 | * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
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/swap.h> | |
19 | #include <linux/smp.h> | |
20 | #include <linux/init.h> | |
11034d55 | 21 | #include <linux/initrd.h> |
1da177e4 LT |
22 | #include <linux/pagemap.h> |
23 | #include <linux/bootmem.h> | |
24 | #include <linux/proc_fs.h> | |
59170891 | 25 | #include <linux/pci.h> |
6fb14755 | 26 | #include <linux/pfn.h> |
c9cf5528 | 27 | #include <linux/poison.h> |
17a941d8 | 28 | #include <linux/dma-mapping.h> |
44df75e6 MT |
29 | #include <linux/module.h> |
30 | #include <linux/memory_hotplug.h> | |
ae32b129 | 31 | #include <linux/nmi.h> |
1da177e4 LT |
32 | |
33 | #include <asm/processor.h> | |
34 | #include <asm/system.h> | |
35 | #include <asm/uaccess.h> | |
36 | #include <asm/pgtable.h> | |
37 | #include <asm/pgalloc.h> | |
38 | #include <asm/dma.h> | |
39 | #include <asm/fixmap.h> | |
40 | #include <asm/e820.h> | |
41 | #include <asm/apic.h> | |
42 | #include <asm/tlb.h> | |
43 | #include <asm/mmu_context.h> | |
44 | #include <asm/proto.h> | |
45 | #include <asm/smp.h> | |
2bc0414e | 46 | #include <asm/sections.h> |
718fc13b | 47 | #include <asm/kdebug.h> |
aaa64e04 | 48 | #include <asm/numa.h> |
7bfeab9a | 49 | #include <asm/cacheflush.h> |
1da177e4 | 50 | |
064d25f1 YL |
51 | /* |
52 | * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries. | |
53 | * The direct mapping extends to max_pfn_mapped, so that we can directly access | |
54 | * apertures, ACPI and other tables without having to play with fixmaps. | |
55 | */ | |
56 | unsigned long max_pfn_mapped; | |
57 | ||
e18c6874 AK |
58 | static unsigned long dma_reserve __initdata; |
59 | ||
1da177e4 LT |
60 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
61 | ||
00d1c5e0 IM |
62 | int direct_gbpages __meminitdata |
63 | #ifdef CONFIG_DIRECT_GBPAGES | |
64 | = 1 | |
65 | #endif | |
66 | ; | |
67 | ||
68 | static int __init parse_direct_gbpages_off(char *arg) | |
69 | { | |
70 | direct_gbpages = 0; | |
71 | return 0; | |
72 | } | |
73 | early_param("nogbpages", parse_direct_gbpages_off); | |
74 | ||
75 | static int __init parse_direct_gbpages_on(char *arg) | |
76 | { | |
77 | direct_gbpages = 1; | |
78 | return 0; | |
79 | } | |
80 | early_param("gbpages", parse_direct_gbpages_on); | |
81 | ||
1da177e4 LT |
82 | /* |
83 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the | |
84 | * physical space so we can cache the place of the first one and move | |
85 | * around without checking the pgd every time. | |
86 | */ | |
87 | ||
88 | void show_mem(void) | |
89 | { | |
e92343cc AK |
90 | long i, total = 0, reserved = 0; |
91 | long shared = 0, cached = 0; | |
1da177e4 | 92 | struct page *page; |
14a62c34 | 93 | pg_data_t *pgdat; |
1da177e4 | 94 | |
e92343cc | 95 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 96 | show_free_areas(); |
ec936fc5 | 97 | for_each_online_pgdat(pgdat) { |
14a62c34 TG |
98 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
99 | /* | |
100 | * This loop can take a while with 256 GB and | |
101 | * 4k pages so defer the NMI watchdog: | |
102 | */ | |
103 | if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) | |
ae32b129 | 104 | touch_nmi_watchdog(); |
14a62c34 | 105 | |
12710a56 BP |
106 | if (!pfn_valid(pgdat->node_start_pfn + i)) |
107 | continue; | |
14a62c34 | 108 | |
1da177e4 LT |
109 | page = pfn_to_page(pgdat->node_start_pfn + i); |
110 | total++; | |
e92343cc AK |
111 | if (PageReserved(page)) |
112 | reserved++; | |
113 | else if (PageSwapCache(page)) | |
114 | cached++; | |
115 | else if (page_count(page)) | |
116 | shared += page_count(page) - 1; | |
14a62c34 | 117 | } |
1da177e4 | 118 | } |
14a62c34 TG |
119 | printk(KERN_INFO "%lu pages of RAM\n", total); |
120 | printk(KERN_INFO "%lu reserved pages\n", reserved); | |
121 | printk(KERN_INFO "%lu pages shared\n", shared); | |
122 | printk(KERN_INFO "%lu pages swap cached\n", cached); | |
1da177e4 LT |
123 | } |
124 | ||
1da177e4 LT |
125 | int after_bootmem; |
126 | ||
5f44a669 | 127 | static __init void *spp_getpage(void) |
14a62c34 | 128 | { |
1da177e4 | 129 | void *ptr; |
14a62c34 | 130 | |
1da177e4 | 131 | if (after_bootmem) |
14a62c34 | 132 | ptr = (void *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
133 | else |
134 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
14a62c34 TG |
135 | |
136 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { | |
137 | panic("set_pte_phys: cannot allocate page data %s\n", | |
138 | after_bootmem ? "after bootmem" : ""); | |
139 | } | |
1da177e4 | 140 | |
10f22dde | 141 | pr_debug("spp_getpage %p\n", ptr); |
14a62c34 | 142 | |
1da177e4 | 143 | return ptr; |
14a62c34 | 144 | } |
1da177e4 | 145 | |
d494a961 | 146 | void |
0814e0ba | 147 | set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte) |
1da177e4 | 148 | { |
1da177e4 LT |
149 | pud_t *pud; |
150 | pmd_t *pmd; | |
d494a961 | 151 | pte_t *pte; |
1da177e4 | 152 | |
0814e0ba | 153 | pud = pud_page + pud_index(vaddr); |
1da177e4 | 154 | if (pud_none(*pud)) { |
14a62c34 | 155 | pmd = (pmd_t *) spp_getpage(); |
bb23e403 | 156 | pud_populate(&init_mm, pud, pmd); |
1da177e4 | 157 | if (pmd != pmd_offset(pud, 0)) { |
10f22dde | 158 | printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", |
14a62c34 | 159 | pmd, pmd_offset(pud, 0)); |
1da177e4 LT |
160 | return; |
161 | } | |
162 | } | |
163 | pmd = pmd_offset(pud, vaddr); | |
164 | if (pmd_none(*pmd)) { | |
165 | pte = (pte_t *) spp_getpage(); | |
bb23e403 | 166 | pmd_populate_kernel(&init_mm, pmd, pte); |
1da177e4 | 167 | if (pte != pte_offset_kernel(pmd, 0)) { |
10f22dde | 168 | printk(KERN_ERR "PAGETABLE BUG #02!\n"); |
1da177e4 LT |
169 | return; |
170 | } | |
171 | } | |
1da177e4 LT |
172 | |
173 | pte = pte_offset_kernel(pmd, vaddr); | |
70c9f590 | 174 | if (!pte_none(*pte) && pte_val(new_pte) && |
1da177e4 LT |
175 | pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) |
176 | pte_ERROR(*pte); | |
177 | set_pte(pte, new_pte); | |
178 | ||
179 | /* | |
180 | * It's enough to flush this one mapping. | |
181 | * (PGE mappings get flushed as well) | |
182 | */ | |
183 | __flush_tlb_one(vaddr); | |
184 | } | |
185 | ||
0814e0ba EH |
186 | void |
187 | set_pte_vaddr(unsigned long vaddr, pte_t pteval) | |
188 | { | |
189 | pgd_t *pgd; | |
190 | pud_t *pud_page; | |
191 | ||
192 | pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval)); | |
193 | ||
194 | pgd = pgd_offset_k(vaddr); | |
195 | if (pgd_none(*pgd)) { | |
196 | printk(KERN_ERR | |
197 | "PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
198 | return; | |
199 | } | |
200 | pud_page = (pud_t*)pgd_page_vaddr(*pgd); | |
201 | set_pte_vaddr_pud(pud_page, vaddr, pteval); | |
202 | } | |
203 | ||
3a9e189d JS |
204 | /* |
205 | * Create large page table mappings for a range of physical addresses. | |
206 | */ | |
207 | static void __init __init_extra_mapping(unsigned long phys, unsigned long size, | |
208 | pgprot_t prot) | |
209 | { | |
210 | pgd_t *pgd; | |
211 | pud_t *pud; | |
212 | pmd_t *pmd; | |
213 | ||
214 | BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK)); | |
215 | for (; size; phys += PMD_SIZE, size -= PMD_SIZE) { | |
216 | pgd = pgd_offset_k((unsigned long)__va(phys)); | |
217 | if (pgd_none(*pgd)) { | |
218 | pud = (pud_t *) spp_getpage(); | |
219 | set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE | | |
220 | _PAGE_USER)); | |
221 | } | |
222 | pud = pud_offset(pgd, (unsigned long)__va(phys)); | |
223 | if (pud_none(*pud)) { | |
224 | pmd = (pmd_t *) spp_getpage(); | |
225 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | | |
226 | _PAGE_USER)); | |
227 | } | |
228 | pmd = pmd_offset(pud, phys); | |
229 | BUG_ON(!pmd_none(*pmd)); | |
230 | set_pmd(pmd, __pmd(phys | pgprot_val(prot))); | |
231 | } | |
232 | } | |
233 | ||
234 | void __init init_extra_mapping_wb(unsigned long phys, unsigned long size) | |
235 | { | |
236 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE); | |
237 | } | |
238 | ||
239 | void __init init_extra_mapping_uc(unsigned long phys, unsigned long size) | |
240 | { | |
241 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE); | |
242 | } | |
243 | ||
31eedd82 | 244 | /* |
88f3aec7 IM |
245 | * The head.S code sets up the kernel high mapping: |
246 | * | |
247 | * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) | |
31eedd82 TG |
248 | * |
249 | * phys_addr holds the negative offset to the kernel, which is added | |
250 | * to the compile time generated pmds. This results in invalid pmds up | |
251 | * to the point where we hit the physaddr 0 mapping. | |
252 | * | |
253 | * We limit the mappings to the region from _text to _end. _end is | |
254 | * rounded up to the 2MB boundary. This catches the invalid pmds as | |
255 | * well, as they are located before _text: | |
256 | */ | |
257 | void __init cleanup_highmap(void) | |
258 | { | |
259 | unsigned long vaddr = __START_KERNEL_map; | |
260 | unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1; | |
261 | pmd_t *pmd = level2_kernel_pgt; | |
262 | pmd_t *last_pmd = pmd + PTRS_PER_PMD; | |
263 | ||
264 | for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) { | |
2884f110 | 265 | if (pmd_none(*pmd)) |
31eedd82 TG |
266 | continue; |
267 | if (vaddr < (unsigned long) _text || vaddr > end) | |
268 | set_pmd(pmd, __pmd(0)); | |
269 | } | |
270 | } | |
271 | ||
75175278 AK |
272 | static unsigned long __initdata table_start; |
273 | static unsigned long __meminitdata table_end; | |
d86623a0 | 274 | static unsigned long __meminitdata table_top; |
1da177e4 | 275 | |
dafe41ee | 276 | static __meminit void *alloc_low_page(unsigned long *phys) |
14a62c34 | 277 | { |
dafe41ee | 278 | unsigned long pfn = table_end++; |
1da177e4 LT |
279 | void *adr; |
280 | ||
44df75e6 MT |
281 | if (after_bootmem) { |
282 | adr = (void *)get_zeroed_page(GFP_ATOMIC); | |
283 | *phys = __pa(adr); | |
14a62c34 | 284 | |
44df75e6 MT |
285 | return adr; |
286 | } | |
287 | ||
d86623a0 | 288 | if (pfn >= table_top) |
14a62c34 | 289 | panic("alloc_low_page: ran out of memory"); |
dafe41ee VG |
290 | |
291 | adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); | |
44df75e6 | 292 | memset(adr, 0, PAGE_SIZE); |
dafe41ee VG |
293 | *phys = pfn * PAGE_SIZE; |
294 | return adr; | |
295 | } | |
1da177e4 | 296 | |
dafe41ee | 297 | static __meminit void unmap_low_page(void *adr) |
14a62c34 | 298 | { |
44df75e6 MT |
299 | if (after_bootmem) |
300 | return; | |
301 | ||
dafe41ee | 302 | early_iounmap(adr, PAGE_SIZE); |
14a62c34 | 303 | } |
1da177e4 | 304 | |
7b16eb89 | 305 | static unsigned long __meminit |
4f9c11dd JF |
306 | phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end) |
307 | { | |
308 | unsigned pages = 0; | |
7b16eb89 | 309 | unsigned long last_map_addr = end; |
4f9c11dd | 310 | int i; |
7b16eb89 | 311 | |
4f9c11dd JF |
312 | pte_t *pte = pte_page + pte_index(addr); |
313 | ||
314 | for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) { | |
315 | ||
316 | if (addr >= end) { | |
317 | if (!after_bootmem) { | |
318 | for(; i < PTRS_PER_PTE; i++, pte++) | |
319 | set_pte(pte, __pte(0)); | |
320 | } | |
321 | break; | |
322 | } | |
323 | ||
324 | if (pte_val(*pte)) | |
325 | continue; | |
326 | ||
327 | if (0) | |
328 | printk(" pte=%p addr=%lx pte=%016lx\n", | |
329 | pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte); | |
330 | set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL)); | |
7b16eb89 | 331 | last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE; |
4f9c11dd JF |
332 | pages++; |
333 | } | |
334 | update_page_count(PG_LEVEL_4K, pages); | |
7b16eb89 YL |
335 | |
336 | return last_map_addr; | |
4f9c11dd JF |
337 | } |
338 | ||
7b16eb89 | 339 | static unsigned long __meminit |
4f9c11dd JF |
340 | phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end) |
341 | { | |
342 | pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd); | |
343 | ||
7b16eb89 | 344 | return phys_pte_init(pte, address, end); |
4f9c11dd JF |
345 | } |
346 | ||
cc615032 | 347 | static unsigned long __meminit |
b50efd2a YL |
348 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end, |
349 | unsigned long page_size_mask) | |
44df75e6 | 350 | { |
ce0c0e50 | 351 | unsigned long pages = 0; |
7b16eb89 | 352 | unsigned long last_map_addr = end; |
ce0c0e50 | 353 | |
6ad91658 | 354 | int i = pmd_index(address); |
44df75e6 | 355 | |
6ad91658 | 356 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
4f9c11dd | 357 | unsigned long pte_phys; |
6ad91658 | 358 | pmd_t *pmd = pmd_page + pmd_index(address); |
4f9c11dd | 359 | pte_t *pte; |
44df75e6 | 360 | |
5f51e139 | 361 | if (address >= end) { |
14a62c34 | 362 | if (!after_bootmem) { |
5f51e139 JB |
363 | for (; i < PTRS_PER_PMD; i++, pmd++) |
364 | set_pmd(pmd, __pmd(0)); | |
14a62c34 | 365 | } |
44df75e6 MT |
366 | break; |
367 | } | |
6ad91658 | 368 | |
4f9c11dd | 369 | if (pmd_val(*pmd)) { |
22b45144 | 370 | if (!pmd_large(*pmd)) |
7b16eb89 YL |
371 | last_map_addr = phys_pte_update(pmd, address, |
372 | end); | |
4f9c11dd JF |
373 | continue; |
374 | } | |
375 | ||
b50efd2a | 376 | if (page_size_mask & (1<<PG_LEVEL_2M)) { |
4f9c11dd JF |
377 | pages++; |
378 | set_pte((pte_t *)pmd, | |
379 | pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
7b16eb89 | 380 | last_map_addr = (address & PMD_MASK) + PMD_SIZE; |
6ad91658 | 381 | continue; |
4f9c11dd | 382 | } |
6ad91658 | 383 | |
4f9c11dd | 384 | pte = alloc_low_page(&pte_phys); |
7b16eb89 | 385 | last_map_addr = phys_pte_init(pte, address, end); |
4f9c11dd JF |
386 | unmap_low_page(pte); |
387 | ||
388 | pmd_populate_kernel(&init_mm, pmd, __va(pte_phys)); | |
44df75e6 | 389 | } |
ce0c0e50 | 390 | update_page_count(PG_LEVEL_2M, pages); |
7b16eb89 | 391 | return last_map_addr; |
44df75e6 MT |
392 | } |
393 | ||
cc615032 | 394 | static unsigned long __meminit |
b50efd2a YL |
395 | phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end, |
396 | unsigned long page_size_mask) | |
44df75e6 | 397 | { |
14a62c34 | 398 | pmd_t *pmd = pmd_offset(pud, 0); |
cc615032 AK |
399 | unsigned long last_map_addr; |
400 | ||
6ad91658 | 401 | spin_lock(&init_mm.page_table_lock); |
b50efd2a | 402 | last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask); |
6ad91658 KM |
403 | spin_unlock(&init_mm.page_table_lock); |
404 | __flush_tlb_all(); | |
cc615032 | 405 | return last_map_addr; |
44df75e6 MT |
406 | } |
407 | ||
cc615032 | 408 | static unsigned long __meminit |
b50efd2a YL |
409 | phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end, |
410 | unsigned long page_size_mask) | |
14a62c34 | 411 | { |
ce0c0e50 | 412 | unsigned long pages = 0; |
cc615032 | 413 | unsigned long last_map_addr = end; |
6ad91658 | 414 | int i = pud_index(addr); |
44df75e6 | 415 | |
14a62c34 | 416 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { |
6ad91658 KM |
417 | unsigned long pmd_phys; |
418 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 LT |
419 | pmd_t *pmd; |
420 | ||
6ad91658 | 421 | if (addr >= end) |
1da177e4 | 422 | break; |
1da177e4 | 423 | |
14a62c34 TG |
424 | if (!after_bootmem && |
425 | !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { | |
426 | set_pud(pud, __pud(0)); | |
1da177e4 | 427 | continue; |
14a62c34 | 428 | } |
1da177e4 | 429 | |
6ad91658 | 430 | if (pud_val(*pud)) { |
ef925766 | 431 | if (!pud_large(*pud)) |
b50efd2a YL |
432 | last_map_addr = phys_pmd_update(pud, addr, end, |
433 | page_size_mask); | |
ef925766 AK |
434 | continue; |
435 | } | |
436 | ||
b50efd2a | 437 | if (page_size_mask & (1<<PG_LEVEL_1G)) { |
ce0c0e50 | 438 | pages++; |
ef925766 AK |
439 | set_pte((pte_t *)pud, |
440 | pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
cc615032 | 441 | last_map_addr = (addr & PUD_MASK) + PUD_SIZE; |
6ad91658 KM |
442 | continue; |
443 | } | |
444 | ||
dafe41ee | 445 | pmd = alloc_low_page(&pmd_phys); |
14a62c34 | 446 | |
44df75e6 | 447 | spin_lock(&init_mm.page_table_lock); |
b50efd2a | 448 | last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask); |
4f9c11dd JF |
449 | unmap_low_page(pmd); |
450 | pud_populate(&init_mm, pud, __va(pmd_phys)); | |
44df75e6 | 451 | spin_unlock(&init_mm.page_table_lock); |
14a62c34 | 452 | |
1da177e4 | 453 | } |
1a2b4412 | 454 | __flush_tlb_all(); |
ce0c0e50 | 455 | update_page_count(PG_LEVEL_1G, pages); |
cc615032 | 456 | |
1a0db38e | 457 | return last_map_addr; |
14a62c34 | 458 | } |
1da177e4 | 459 | |
4f9c11dd | 460 | static unsigned long __meminit |
b50efd2a YL |
461 | phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end, |
462 | unsigned long page_size_mask) | |
4f9c11dd JF |
463 | { |
464 | pud_t *pud; | |
465 | ||
466 | pud = (pud_t *)pgd_page_vaddr(*pgd); | |
467 | ||
b50efd2a | 468 | return phys_pud_init(pud, addr, end, page_size_mask); |
4f9c11dd JF |
469 | } |
470 | ||
1da177e4 LT |
471 | static void __init find_early_table_space(unsigned long end) |
472 | { | |
c2e6d65b | 473 | unsigned long puds, pmds, ptes, tables, start; |
1da177e4 LT |
474 | |
475 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; | |
ef925766 | 476 | tables = round_up(puds * sizeof(pud_t), PAGE_SIZE); |
c2e6d65b YL |
477 | if (direct_gbpages) { |
478 | unsigned long extra; | |
479 | extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT); | |
480 | pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT; | |
481 | } else | |
482 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; | |
483 | tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE); | |
484 | ||
485 | if (cpu_has_pse) { | |
486 | unsigned long extra; | |
487 | extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT); | |
488 | ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
489 | } else | |
490 | ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
491 | tables += round_up(ptes * sizeof(pte_t), PAGE_SIZE); | |
1da177e4 | 492 | |
14a62c34 TG |
493 | /* |
494 | * RED-PEN putting page tables only on node 0 could | |
495 | * cause a hotspot and fill up ZONE_DMA. The page tables | |
496 | * need roughly 0.5KB per GB. | |
497 | */ | |
498 | start = 0x8000; | |
24a5da73 | 499 | table_start = find_e820_area(start, end, tables, PAGE_SIZE); |
1da177e4 LT |
500 | if (table_start == -1UL) |
501 | panic("Cannot find space for the kernel page tables"); | |
502 | ||
503 | table_start >>= PAGE_SHIFT; | |
504 | table_end = table_start; | |
d86623a0 | 505 | table_top = table_start + (tables >> PAGE_SHIFT); |
44df75e6 | 506 | |
d86623a0 YL |
507 | printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n", |
508 | end, table_start << PAGE_SHIFT, table_top << PAGE_SHIFT); | |
1da177e4 LT |
509 | } |
510 | ||
ef925766 AK |
511 | static void __init init_gbpages(void) |
512 | { | |
513 | if (direct_gbpages && cpu_has_gbpages) | |
514 | printk(KERN_INFO "Using GB pages for direct mapping\n"); | |
515 | else | |
516 | direct_gbpages = 0; | |
517 | } | |
518 | ||
03273184 | 519 | #ifdef CONFIG_MEMTEST |
c64df707 YL |
520 | |
521 | static void __init memtest(unsigned long start_phys, unsigned long size, | |
522 | unsigned pattern) | |
272b9cad YL |
523 | { |
524 | unsigned long i; | |
525 | unsigned long *start; | |
526 | unsigned long start_bad; | |
527 | unsigned long last_bad; | |
528 | unsigned long val; | |
529 | unsigned long start_phys_aligned; | |
530 | unsigned long count; | |
531 | unsigned long incr; | |
532 | ||
533 | switch (pattern) { | |
534 | case 0: | |
535 | val = 0UL; | |
536 | break; | |
537 | case 1: | |
538 | val = -1UL; | |
539 | break; | |
540 | case 2: | |
541 | val = 0x5555555555555555UL; | |
542 | break; | |
543 | case 3: | |
544 | val = 0xaaaaaaaaaaaaaaaaUL; | |
545 | break; | |
546 | default: | |
547 | return; | |
548 | } | |
549 | ||
550 | incr = sizeof(unsigned long); | |
551 | start_phys_aligned = ALIGN(start_phys, incr); | |
552 | count = (size - (start_phys_aligned - start_phys))/incr; | |
553 | start = __va(start_phys_aligned); | |
554 | start_bad = 0; | |
555 | last_bad = 0; | |
556 | ||
557 | for (i = 0; i < count; i++) | |
558 | start[i] = val; | |
559 | for (i = 0; i < count; i++, start++, start_phys_aligned += incr) { | |
560 | if (*start != val) { | |
561 | if (start_phys_aligned == last_bad + incr) { | |
562 | last_bad += incr; | |
563 | } else { | |
564 | if (start_bad) { | |
dcfe9465 | 565 | printk(KERN_CONT "\n %016lx bad mem addr %016lx - %016lx reserved", |
272b9cad YL |
566 | val, start_bad, last_bad + incr); |
567 | reserve_early(start_bad, last_bad - start_bad, "BAD RAM"); | |
568 | } | |
569 | start_bad = last_bad = start_phys_aligned; | |
570 | } | |
571 | } | |
572 | } | |
573 | if (start_bad) { | |
dcfe9465 | 574 | printk(KERN_CONT "\n %016lx bad mem addr %016lx - %016lx reserved", |
272b9cad YL |
575 | val, start_bad, last_bad + incr); |
576 | reserve_early(start_bad, last_bad - start_bad, "BAD RAM"); | |
577 | } | |
578 | ||
579 | } | |
580 | ||
03273184 YL |
581 | /* default is disabled */ |
582 | static int memtest_pattern __initdata; | |
c64df707 | 583 | |
272b9cad YL |
584 | static int __init parse_memtest(char *arg) |
585 | { | |
586 | if (arg) | |
c64df707 | 587 | memtest_pattern = simple_strtoul(arg, NULL, 0); |
272b9cad YL |
588 | return 0; |
589 | } | |
590 | ||
591 | early_param("memtest", parse_memtest); | |
592 | ||
593 | static void __init early_memtest(unsigned long start, unsigned long end) | |
594 | { | |
27df66a4 | 595 | u64 t_start, t_size; |
272b9cad YL |
596 | unsigned pattern; |
597 | ||
c64df707 YL |
598 | if (!memtest_pattern) |
599 | return; | |
600 | ||
601 | printk(KERN_INFO "early_memtest: pattern num %d", memtest_pattern); | |
272b9cad YL |
602 | for (pattern = 0; pattern < memtest_pattern; pattern++) { |
603 | t_start = start; | |
604 | t_size = 0; | |
605 | while (t_start < end) { | |
606 | t_start = find_e820_area_size(t_start, &t_size, 1); | |
607 | ||
608 | /* done ? */ | |
609 | if (t_start >= end) | |
610 | break; | |
611 | if (t_start + t_size > end) | |
612 | t_size = end - t_start; | |
613 | ||
27df66a4 AM |
614 | printk(KERN_CONT "\n %016llx - %016llx pattern %d", |
615 | (unsigned long long)t_start, | |
616 | (unsigned long long)t_start + t_size, pattern); | |
272b9cad YL |
617 | |
618 | memtest(t_start, t_size, pattern); | |
619 | ||
620 | t_start += t_size; | |
621 | } | |
622 | } | |
c64df707 | 623 | printk(KERN_CONT "\n"); |
272b9cad | 624 | } |
c64df707 YL |
625 | #else |
626 | static void __init early_memtest(unsigned long start, unsigned long end) | |
627 | { | |
628 | } | |
629 | #endif | |
272b9cad | 630 | |
b50efd2a YL |
631 | static unsigned long __init kernel_physical_mapping_init(unsigned long start, |
632 | unsigned long end, | |
633 | unsigned long page_size_mask) | |
14a62c34 | 634 | { |
1da177e4 | 635 | |
b50efd2a | 636 | unsigned long next, last_map_addr = end; |
1da177e4 LT |
637 | |
638 | start = (unsigned long)__va(start); | |
639 | end = (unsigned long)__va(end); | |
640 | ||
641 | for (; start < end; start = next) { | |
44df75e6 | 642 | pgd_t *pgd = pgd_offset_k(start); |
14a62c34 | 643 | unsigned long pud_phys; |
44df75e6 MT |
644 | pud_t *pud; |
645 | ||
4f9c11dd JF |
646 | next = start + PGDIR_SIZE; |
647 | if (next > end) | |
648 | next = end; | |
649 | ||
650 | if (pgd_val(*pgd)) { | |
b50efd2a YL |
651 | last_map_addr = phys_pud_update(pgd, __pa(start), |
652 | __pa(end), page_size_mask); | |
4f9c11dd JF |
653 | continue; |
654 | } | |
655 | ||
44df75e6 | 656 | if (after_bootmem) |
d2ae5b5f | 657 | pud = pud_offset(pgd, start & PGDIR_MASK); |
44df75e6 | 658 | else |
dafe41ee | 659 | pud = alloc_low_page(&pud_phys); |
44df75e6 | 660 | |
b50efd2a YL |
661 | last_map_addr = phys_pud_init(pud, __pa(start), __pa(next), |
662 | page_size_mask); | |
4f9c11dd | 663 | unmap_low_page(pud); |
574977a2 JF |
664 | pgd_populate(&init_mm, pgd_offset_k(start), |
665 | __va(pud_phys)); | |
14a62c34 | 666 | } |
1da177e4 | 667 | |
b50efd2a YL |
668 | return last_map_addr; |
669 | } | |
7b16eb89 YL |
670 | |
671 | struct map_range { | |
672 | unsigned long start; | |
673 | unsigned long end; | |
674 | unsigned page_size_mask; | |
675 | }; | |
676 | ||
677 | #define NR_RANGE_MR 5 | |
678 | ||
679 | static int save_mr(struct map_range *mr, int nr_range, | |
680 | unsigned long start_pfn, unsigned long end_pfn, | |
681 | unsigned long page_size_mask) | |
682 | { | |
683 | ||
684 | if (start_pfn < end_pfn) { | |
685 | if (nr_range >= NR_RANGE_MR) | |
686 | panic("run out of range for init_memory_mapping\n"); | |
687 | mr[nr_range].start = start_pfn<<PAGE_SHIFT; | |
688 | mr[nr_range].end = end_pfn<<PAGE_SHIFT; | |
689 | mr[nr_range].page_size_mask = page_size_mask; | |
690 | nr_range++; | |
691 | } | |
692 | ||
693 | return nr_range; | |
694 | } | |
695 | ||
b50efd2a YL |
696 | /* |
697 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
698 | * This runs before bootmem is initialized and gets pages directly from | |
699 | * the physical memory. To access them they are temporarily mapped. | |
700 | */ | |
701 | unsigned long __init_refok init_memory_mapping(unsigned long start, | |
702 | unsigned long end) | |
703 | { | |
7b16eb89 | 704 | unsigned long last_map_addr = 0; |
b50efd2a | 705 | unsigned long page_size_mask = 0; |
c2e6d65b | 706 | unsigned long start_pfn, end_pfn; |
b50efd2a | 707 | |
7b16eb89 YL |
708 | struct map_range mr[NR_RANGE_MR]; |
709 | int nr_range, i; | |
710 | ||
b50efd2a YL |
711 | printk(KERN_INFO "init_memory_mapping\n"); |
712 | ||
713 | /* | |
714 | * Find space for the kernel direct mapping tables. | |
715 | * | |
716 | * Later we should allocate these tables in the local node of the | |
717 | * memory mapped. Unfortunately this is done currently before the | |
718 | * nodes are discovered. | |
719 | */ | |
7b16eb89 | 720 | if (!after_bootmem) |
b50efd2a | 721 | init_gbpages(); |
b50efd2a YL |
722 | |
723 | if (direct_gbpages) | |
724 | page_size_mask |= 1 << PG_LEVEL_1G; | |
725 | if (cpu_has_pse) | |
726 | page_size_mask |= 1 << PG_LEVEL_2M; | |
727 | ||
7b16eb89 YL |
728 | memset(mr, 0, sizeof(mr)); |
729 | nr_range = 0; | |
730 | ||
731 | /* head if not big page alignment ?*/ | |
c2e6d65b YL |
732 | start_pfn = start >> PAGE_SHIFT; |
733 | end_pfn = ((start + (PMD_SIZE - 1)) >> PMD_SHIFT) | |
734 | << (PMD_SHIFT - PAGE_SHIFT); | |
7b16eb89 | 735 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); |
c2e6d65b YL |
736 | |
737 | /* big page (2M) range*/ | |
738 | start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT) | |
739 | << (PMD_SHIFT - PAGE_SHIFT); | |
740 | end_pfn = ((start + (PUD_SIZE - 1))>>PUD_SHIFT) | |
741 | << (PUD_SHIFT - PAGE_SHIFT); | |
742 | if (end_pfn > ((end>>PUD_SHIFT)<<(PUD_SHIFT - PAGE_SHIFT))) | |
743 | end_pfn = ((end>>PUD_SHIFT)<<(PUD_SHIFT - PAGE_SHIFT)); | |
7b16eb89 YL |
744 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, |
745 | page_size_mask & (1<<PG_LEVEL_2M)); | |
c2e6d65b YL |
746 | |
747 | /* big page (1G) range */ | |
748 | start_pfn = end_pfn; | |
749 | end_pfn = (end>>PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT); | |
7b16eb89 YL |
750 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, |
751 | page_size_mask & | |
752 | ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); | |
c2e6d65b YL |
753 | |
754 | /* tail is not big page (1G) alignment */ | |
755 | start_pfn = end_pfn; | |
756 | end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT); | |
7b16eb89 YL |
757 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, |
758 | page_size_mask & (1<<PG_LEVEL_2M)); | |
759 | ||
c2e6d65b YL |
760 | /* tail is not big page (2M) alignment */ |
761 | start_pfn = end_pfn; | |
762 | end_pfn = end>>PAGE_SHIFT; | |
7b16eb89 YL |
763 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); |
764 | ||
765 | for (i = 0; i < nr_range; i++) | |
766 | printk(KERN_DEBUG " %010lx - %010lx page %s\n", | |
767 | mr[i].start, mr[i].end, | |
768 | (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":( | |
769 | (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k")); | |
770 | ||
771 | if (!after_bootmem) | |
772 | find_early_table_space(end); | |
773 | ||
774 | for (i = 0; i < nr_range; i++) | |
c2e6d65b | 775 | last_map_addr = kernel_physical_mapping_init( |
7b16eb89 YL |
776 | mr[i].start, mr[i].end, |
777 | mr[i].page_size_mask); | |
b50efd2a | 778 | |
44df75e6 | 779 | if (!after_bootmem) |
f51c9452 | 780 | mmu_cr4_features = read_cr4(); |
1da177e4 | 781 | __flush_tlb_all(); |
75175278 | 782 | |
b50efd2a | 783 | if (!after_bootmem && table_end > table_start) |
24a5da73 YL |
784 | reserve_early(table_start << PAGE_SHIFT, |
785 | table_end << PAGE_SHIFT, "PGTABLE"); | |
272b9cad | 786 | |
b50efd2a YL |
787 | printk(KERN_INFO "last_map_addr: %lx end: %lx\n", |
788 | last_map_addr, end); | |
789 | ||
272b9cad | 790 | if (!after_bootmem) |
b50efd2a | 791 | early_memtest(start, end); |
cc615032 | 792 | |
1a0db38e | 793 | return last_map_addr >> PAGE_SHIFT; |
1da177e4 LT |
794 | } |
795 | ||
2b97690f | 796 | #ifndef CONFIG_NUMA |
1f75d7e3 YL |
797 | void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn) |
798 | { | |
799 | unsigned long bootmap_size, bootmap; | |
800 | ||
801 | bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT; | |
802 | bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size, | |
803 | PAGE_SIZE); | |
804 | if (bootmap == -1L) | |
805 | panic("Cannot find bootmem map of size %ld\n", bootmap_size); | |
346cafec YL |
806 | /* don't touch min_low_pfn */ |
807 | bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT, | |
808 | 0, end_pfn); | |
1f75d7e3 YL |
809 | e820_register_active_regions(0, start_pfn, end_pfn); |
810 | free_bootmem_with_active_regions(0, end_pfn); | |
811 | early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT); | |
812 | reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT); | |
813 | } | |
814 | ||
1da177e4 LT |
815 | void __init paging_init(void) |
816 | { | |
6391af17 | 817 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
14a62c34 | 818 | |
6391af17 MG |
819 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
820 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
821 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
c987d12f | 822 | max_zone_pfns[ZONE_NORMAL] = max_pfn; |
6391af17 | 823 | |
c987d12f | 824 | memory_present(0, 0, max_pfn); |
44df75e6 | 825 | sparse_init(); |
5cb248ab | 826 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
827 | } |
828 | #endif | |
829 | ||
44df75e6 MT |
830 | /* |
831 | * Memory hotplug specific functions | |
44df75e6 | 832 | */ |
bc02af93 | 833 | #ifdef CONFIG_MEMORY_HOTPLUG |
9d99aaa3 AK |
834 | /* |
835 | * Memory is added always to NORMAL zone. This means you will never get | |
836 | * additional DMA/DMA32 memory. | |
837 | */ | |
bc02af93 | 838 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 839 | { |
bc02af93 | 840 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 841 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
cc615032 | 842 | unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT; |
44df75e6 MT |
843 | unsigned long nr_pages = size >> PAGE_SHIFT; |
844 | int ret; | |
845 | ||
cc615032 AK |
846 | last_mapped_pfn = init_memory_mapping(start, start + size-1); |
847 | if (last_mapped_pfn > max_pfn_mapped) | |
848 | max_pfn_mapped = last_mapped_pfn; | |
45e0b78b | 849 | |
44df75e6 | 850 | ret = __add_pages(zone, start_pfn, nr_pages); |
10f22dde | 851 | WARN_ON(1); |
44df75e6 | 852 | |
44df75e6 | 853 | return ret; |
44df75e6 | 854 | } |
bc02af93 | 855 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 | 856 | |
8243229f | 857 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
4942e998 KM |
858 | int memory_add_physaddr_to_nid(u64 start) |
859 | { | |
860 | return 0; | |
861 | } | |
8c2676a5 | 862 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
4942e998 KM |
863 | #endif |
864 | ||
45e0b78b KM |
865 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
866 | ||
ae531c26 AV |
867 | /* |
868 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
869 | * is valid. The argument is a physical page number. | |
870 | * | |
871 | * | |
872 | * On x86, access has to be given to the first megabyte of ram because that area | |
873 | * contains bios code and data regions used by X and dosemu and similar apps. | |
874 | * Access has to be given to non-kernel-ram areas as well, these contain the PCI | |
875 | * mmio resources as well as potential bios/acpi data regions. | |
876 | */ | |
877 | int devmem_is_allowed(unsigned long pagenr) | |
878 | { | |
879 | if (pagenr <= 256) | |
880 | return 1; | |
881 | if (!page_is_ram(pagenr)) | |
882 | return 1; | |
883 | return 0; | |
884 | } | |
885 | ||
886 | ||
14a62c34 TG |
887 | static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, |
888 | kcore_modules, kcore_vsyscall; | |
1da177e4 LT |
889 | |
890 | void __init mem_init(void) | |
891 | { | |
0a43e4bf | 892 | long codesize, reservedpages, datasize, initsize; |
1da177e4 | 893 | |
0dc243ae | 894 | pci_iommu_alloc(); |
1da177e4 | 895 | |
48ddb154 | 896 | /* clear_bss() already clear the empty_zero_page */ |
1da177e4 LT |
897 | |
898 | reservedpages = 0; | |
899 | ||
900 | /* this will put all low memory onto the freelists */ | |
2b97690f | 901 | #ifdef CONFIG_NUMA |
0a43e4bf | 902 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 903 | #else |
0a43e4bf | 904 | totalram_pages = free_all_bootmem(); |
1da177e4 | 905 | #endif |
c987d12f YL |
906 | reservedpages = max_pfn - totalram_pages - |
907 | absent_pages_in_range(0, max_pfn); | |
1da177e4 LT |
908 | after_bootmem = 1; |
909 | ||
910 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
911 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
912 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
913 | ||
914 | /* Register memory areas for /proc/kcore */ | |
14a62c34 TG |
915 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
916 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
1da177e4 LT |
917 | VMALLOC_END-VMALLOC_START); |
918 | kclist_add(&kcore_kernel, &_stext, _end - _stext); | |
919 | kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); | |
14a62c34 | 920 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
1da177e4 LT |
921 | VSYSCALL_END - VSYSCALL_START); |
922 | ||
10f22dde | 923 | printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " |
14a62c34 | 924 | "%ldk reserved, %ldk data, %ldk init)\n", |
1da177e4 | 925 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
c987d12f | 926 | max_pfn << (PAGE_SHIFT-10), |
1da177e4 LT |
927 | codesize >> 10, |
928 | reservedpages << (PAGE_SHIFT-10), | |
929 | datasize >> 10, | |
930 | initsize >> 10); | |
76ebd054 TG |
931 | |
932 | cpa_init(); | |
1da177e4 LT |
933 | } |
934 | ||
d167a518 | 935 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
1da177e4 | 936 | { |
bfc734b2 | 937 | unsigned long addr = begin; |
1da177e4 | 938 | |
bfc734b2 | 939 | if (addr >= end) |
d167a518 GH |
940 | return; |
941 | ||
ee01f112 IM |
942 | /* |
943 | * If debugging page accesses then do not free this memory but | |
944 | * mark them not present - any buggy init-section access will | |
945 | * create a kernel page fault: | |
946 | */ | |
947 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
948 | printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", | |
949 | begin, PAGE_ALIGN(end)); | |
950 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
951 | #else | |
6fb14755 | 952 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
14a62c34 | 953 | |
bfc734b2 | 954 | for (; addr < end; addr += PAGE_SIZE) { |
e3ebadd9 LT |
955 | ClearPageReserved(virt_to_page(addr)); |
956 | init_page_count(virt_to_page(addr)); | |
957 | memset((void *)(addr & ~(PAGE_SIZE-1)), | |
958 | POISON_FREE_INITMEM, PAGE_SIZE); | |
e3ebadd9 | 959 | free_page(addr); |
1da177e4 LT |
960 | totalram_pages++; |
961 | } | |
ee01f112 | 962 | #endif |
d167a518 GH |
963 | } |
964 | ||
965 | void free_initmem(void) | |
966 | { | |
d167a518 | 967 | free_init_pages("unused kernel memory", |
e3ebadd9 LT |
968 | (unsigned long)(&__init_begin), |
969 | (unsigned long)(&__init_end)); | |
1da177e4 LT |
970 | } |
971 | ||
67df197b | 972 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
973 | const int rodata_test_data = 0xC3; |
974 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
67df197b | 975 | |
67df197b AV |
976 | void mark_rodata_ro(void) |
977 | { | |
4e4eee0e | 978 | unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata); |
67df197b | 979 | |
6fb14755 | 980 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 981 | (end - start) >> 10); |
984bb80d AV |
982 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
983 | ||
984 | /* | |
985 | * The rodata section (but not the kernel text!) should also be | |
986 | * not-executable. | |
987 | */ | |
988 | start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; | |
989 | set_memory_nx(start, (end - start) >> PAGE_SHIFT); | |
67df197b | 990 | |
1a487252 AV |
991 | rodata_test(); |
992 | ||
0c42f392 | 993 | #ifdef CONFIG_CPA_DEBUG |
10f22dde | 994 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); |
6d238cc4 | 995 | set_memory_rw(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 996 | |
10f22dde | 997 | printk(KERN_INFO "Testing CPA: again\n"); |
6d238cc4 | 998 | set_memory_ro(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 999 | #endif |
67df197b | 1000 | } |
4e4eee0e | 1001 | |
67df197b AV |
1002 | #endif |
1003 | ||
1da177e4 LT |
1004 | #ifdef CONFIG_BLK_DEV_INITRD |
1005 | void free_initrd_mem(unsigned long start, unsigned long end) | |
1006 | { | |
e3ebadd9 | 1007 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
1008 | } |
1009 | #endif | |
1010 | ||
d2dbf343 YL |
1011 | int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, |
1012 | int flags) | |
14a62c34 | 1013 | { |
2b97690f | 1014 | #ifdef CONFIG_NUMA |
8b3cd09e | 1015 | int nid, next_nid; |
6a07a0ed | 1016 | int ret; |
5e58a02a AK |
1017 | #endif |
1018 | unsigned long pfn = phys >> PAGE_SHIFT; | |
14a62c34 | 1019 | |
c987d12f | 1020 | if (pfn >= max_pfn) { |
14a62c34 TG |
1021 | /* |
1022 | * This can happen with kdump kernels when accessing | |
1023 | * firmware tables: | |
1024 | */ | |
67794292 | 1025 | if (pfn < max_pfn_mapped) |
8b2ef1d7 | 1026 | return -EFAULT; |
14a62c34 | 1027 | |
6a07a0ed | 1028 | printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n", |
5e58a02a | 1029 | phys, len); |
8b2ef1d7 | 1030 | return -EFAULT; |
5e58a02a AK |
1031 | } |
1032 | ||
1033 | /* Should check here against the e820 map to avoid double free */ | |
1034 | #ifdef CONFIG_NUMA | |
8b3cd09e YL |
1035 | nid = phys_to_nid(phys); |
1036 | next_nid = phys_to_nid(phys + len - 1); | |
1037 | if (nid == next_nid) | |
8b2ef1d7 | 1038 | ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags); |
8b3cd09e | 1039 | else |
8b2ef1d7 BW |
1040 | ret = reserve_bootmem(phys, len, flags); |
1041 | ||
1042 | if (ret != 0) | |
1043 | return ret; | |
1044 | ||
14a62c34 | 1045 | #else |
72a7fe39 | 1046 | reserve_bootmem(phys, len, BOOTMEM_DEFAULT); |
1da177e4 | 1047 | #endif |
8b3cd09e | 1048 | |
0e0b864e | 1049 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
e18c6874 | 1050 | dma_reserve += len / PAGE_SIZE; |
0e0b864e MG |
1051 | set_dma_reserve(dma_reserve); |
1052 | } | |
8b2ef1d7 BW |
1053 | |
1054 | return 0; | |
1da177e4 LT |
1055 | } |
1056 | ||
14a62c34 TG |
1057 | int kern_addr_valid(unsigned long addr) |
1058 | { | |
1da177e4 | 1059 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
14a62c34 TG |
1060 | pgd_t *pgd; |
1061 | pud_t *pud; | |
1062 | pmd_t *pmd; | |
1063 | pte_t *pte; | |
1da177e4 LT |
1064 | |
1065 | if (above != 0 && above != -1UL) | |
14a62c34 TG |
1066 | return 0; |
1067 | ||
1da177e4 LT |
1068 | pgd = pgd_offset_k(addr); |
1069 | if (pgd_none(*pgd)) | |
1070 | return 0; | |
1071 | ||
1072 | pud = pud_offset(pgd, addr); | |
1073 | if (pud_none(*pud)) | |
14a62c34 | 1074 | return 0; |
1da177e4 LT |
1075 | |
1076 | pmd = pmd_offset(pud, addr); | |
1077 | if (pmd_none(*pmd)) | |
1078 | return 0; | |
14a62c34 | 1079 | |
1da177e4 LT |
1080 | if (pmd_large(*pmd)) |
1081 | return pfn_valid(pmd_pfn(*pmd)); | |
1082 | ||
1083 | pte = pte_offset_kernel(pmd, addr); | |
1084 | if (pte_none(*pte)) | |
1085 | return 0; | |
14a62c34 | 1086 | |
1da177e4 LT |
1087 | return pfn_valid(pte_pfn(*pte)); |
1088 | } | |
1089 | ||
14a62c34 TG |
1090 | /* |
1091 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
1092 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
1093 | * not need special handling anymore: | |
1094 | */ | |
1da177e4 | 1095 | static struct vm_area_struct gate_vma = { |
14a62c34 TG |
1096 | .vm_start = VSYSCALL_START, |
1097 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), | |
1098 | .vm_page_prot = PAGE_READONLY_EXEC, | |
1099 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
1100 | }; |
1101 | ||
1da177e4 LT |
1102 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
1103 | { | |
1104 | #ifdef CONFIG_IA32_EMULATION | |
1e014410 AK |
1105 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
1106 | return NULL; | |
1da177e4 LT |
1107 | #endif |
1108 | return &gate_vma; | |
1109 | } | |
1110 | ||
1111 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
1112 | { | |
1113 | struct vm_area_struct *vma = get_gate_vma(task); | |
14a62c34 | 1114 | |
1e014410 AK |
1115 | if (!vma) |
1116 | return 0; | |
14a62c34 | 1117 | |
1da177e4 LT |
1118 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
1119 | } | |
1120 | ||
14a62c34 TG |
1121 | /* |
1122 | * Use this when you have no reliable task/vma, typically from interrupt | |
1123 | * context. It is less reliable than using the task's vma and may give | |
1124 | * false positives: | |
1da177e4 LT |
1125 | */ |
1126 | int in_gate_area_no_task(unsigned long addr) | |
1127 | { | |
1e014410 | 1128 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 1129 | } |
2e1c49db | 1130 | |
2aae950b AK |
1131 | const char *arch_vma_name(struct vm_area_struct *vma) |
1132 | { | |
1133 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
1134 | return "[vdso]"; | |
1135 | if (vma == &gate_vma) | |
1136 | return "[vsyscall]"; | |
1137 | return NULL; | |
1138 | } | |
0889eba5 CL |
1139 | |
1140 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
1141 | /* | |
1142 | * Initialise the sparsemem vmemmap using huge-pages at the PMD level. | |
1143 | */ | |
c2b91e2e YL |
1144 | static long __meminitdata addr_start, addr_end; |
1145 | static void __meminitdata *p_start, *p_end; | |
1146 | static int __meminitdata node_start; | |
1147 | ||
14a62c34 TG |
1148 | int __meminit |
1149 | vmemmap_populate(struct page *start_page, unsigned long size, int node) | |
0889eba5 CL |
1150 | { |
1151 | unsigned long addr = (unsigned long)start_page; | |
1152 | unsigned long end = (unsigned long)(start_page + size); | |
1153 | unsigned long next; | |
1154 | pgd_t *pgd; | |
1155 | pud_t *pud; | |
1156 | pmd_t *pmd; | |
1157 | ||
1158 | for (; addr < end; addr = next) { | |
7c934d39 | 1159 | void *p = NULL; |
0889eba5 CL |
1160 | |
1161 | pgd = vmemmap_pgd_populate(addr, node); | |
1162 | if (!pgd) | |
1163 | return -ENOMEM; | |
14a62c34 | 1164 | |
0889eba5 CL |
1165 | pud = vmemmap_pud_populate(pgd, addr, node); |
1166 | if (!pud) | |
1167 | return -ENOMEM; | |
1168 | ||
7c934d39 JF |
1169 | if (!cpu_has_pse) { |
1170 | next = (addr + PAGE_SIZE) & PAGE_MASK; | |
1171 | pmd = vmemmap_pmd_populate(pud, addr, node); | |
1172 | ||
1173 | if (!pmd) | |
1174 | return -ENOMEM; | |
1175 | ||
1176 | p = vmemmap_pte_populate(pmd, addr, node); | |
14a62c34 | 1177 | |
0889eba5 CL |
1178 | if (!p) |
1179 | return -ENOMEM; | |
1180 | ||
7c934d39 JF |
1181 | addr_end = addr + PAGE_SIZE; |
1182 | p_end = p + PAGE_SIZE; | |
14a62c34 | 1183 | } else { |
7c934d39 JF |
1184 | next = pmd_addr_end(addr, end); |
1185 | ||
1186 | pmd = pmd_offset(pud, addr); | |
1187 | if (pmd_none(*pmd)) { | |
1188 | pte_t entry; | |
1189 | ||
1190 | p = vmemmap_alloc_block(PMD_SIZE, node); | |
1191 | if (!p) | |
1192 | return -ENOMEM; | |
1193 | ||
1194 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, | |
1195 | PAGE_KERNEL_LARGE); | |
1196 | set_pmd(pmd, __pmd(pte_val(entry))); | |
1197 | ||
7c934d39 JF |
1198 | /* check to see if we have contiguous blocks */ |
1199 | if (p_end != p || node_start != node) { | |
1200 | if (p_start) | |
1201 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
1202 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
1203 | addr_start = addr; | |
1204 | node_start = node; | |
1205 | p_start = p; | |
1206 | } | |
49c980df YL |
1207 | |
1208 | addr_end = addr + PMD_SIZE; | |
1209 | p_end = p + PMD_SIZE; | |
7c934d39 JF |
1210 | } else |
1211 | vmemmap_verify((pte_t *)pmd, node, addr, next); | |
14a62c34 | 1212 | } |
7c934d39 | 1213 | |
0889eba5 | 1214 | } |
0889eba5 CL |
1215 | return 0; |
1216 | } | |
c2b91e2e YL |
1217 | |
1218 | void __meminit vmemmap_populate_print_last(void) | |
1219 | { | |
1220 | if (p_start) { | |
1221 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
1222 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
1223 | p_start = NULL; | |
1224 | p_end = NULL; | |
1225 | node_start = 0; | |
1226 | } | |
1227 | } | |
0889eba5 | 1228 | #endif |