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> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/bootmem.h> | |
23 | #include <linux/proc_fs.h> | |
59170891 | 24 | #include <linux/pci.h> |
6fb14755 | 25 | #include <linux/pfn.h> |
c9cf5528 | 26 | #include <linux/poison.h> |
17a941d8 | 27 | #include <linux/dma-mapping.h> |
44df75e6 MT |
28 | #include <linux/module.h> |
29 | #include <linux/memory_hotplug.h> | |
ae32b129 | 30 | #include <linux/nmi.h> |
1da177e4 LT |
31 | |
32 | #include <asm/processor.h> | |
33 | #include <asm/system.h> | |
34 | #include <asm/uaccess.h> | |
35 | #include <asm/pgtable.h> | |
36 | #include <asm/pgalloc.h> | |
37 | #include <asm/dma.h> | |
38 | #include <asm/fixmap.h> | |
39 | #include <asm/e820.h> | |
40 | #include <asm/apic.h> | |
41 | #include <asm/tlb.h> | |
42 | #include <asm/mmu_context.h> | |
43 | #include <asm/proto.h> | |
44 | #include <asm/smp.h> | |
2bc0414e | 45 | #include <asm/sections.h> |
1da177e4 LT |
46 | |
47 | #ifndef Dprintk | |
48 | #define Dprintk(x...) | |
49 | #endif | |
50 | ||
e6584504 | 51 | const struct dma_mapping_ops* dma_ops; |
17a941d8 MBY |
52 | EXPORT_SYMBOL(dma_ops); |
53 | ||
e18c6874 AK |
54 | static unsigned long dma_reserve __initdata; |
55 | ||
1da177e4 LT |
56 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
57 | ||
58 | /* | |
59 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the | |
60 | * physical space so we can cache the place of the first one and move | |
61 | * around without checking the pgd every time. | |
62 | */ | |
63 | ||
64 | void show_mem(void) | |
65 | { | |
e92343cc AK |
66 | long i, total = 0, reserved = 0; |
67 | long shared = 0, cached = 0; | |
1da177e4 LT |
68 | pg_data_t *pgdat; |
69 | struct page *page; | |
70 | ||
e92343cc | 71 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 72 | show_free_areas(); |
e92343cc | 73 | printk(KERN_INFO "Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); |
1da177e4 | 74 | |
ec936fc5 | 75 | for_each_online_pgdat(pgdat) { |
1da177e4 | 76 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
ae32b129 KR |
77 | /* this loop can take a while with 256 GB and 4k pages |
78 | so update the NMI watchdog */ | |
79 | if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) { | |
80 | touch_nmi_watchdog(); | |
81 | } | |
12710a56 BP |
82 | if (!pfn_valid(pgdat->node_start_pfn + i)) |
83 | continue; | |
1da177e4 LT |
84 | page = pfn_to_page(pgdat->node_start_pfn + i); |
85 | total++; | |
e92343cc AK |
86 | if (PageReserved(page)) |
87 | reserved++; | |
88 | else if (PageSwapCache(page)) | |
89 | cached++; | |
90 | else if (page_count(page)) | |
91 | shared += page_count(page) - 1; | |
1da177e4 LT |
92 | } |
93 | } | |
e92343cc AK |
94 | printk(KERN_INFO "%lu pages of RAM\n", total); |
95 | printk(KERN_INFO "%lu reserved pages\n",reserved); | |
96 | printk(KERN_INFO "%lu pages shared\n",shared); | |
97 | printk(KERN_INFO "%lu pages swap cached\n",cached); | |
1da177e4 LT |
98 | } |
99 | ||
1da177e4 LT |
100 | int after_bootmem; |
101 | ||
5f44a669 | 102 | static __init void *spp_getpage(void) |
1da177e4 LT |
103 | { |
104 | void *ptr; | |
105 | if (after_bootmem) | |
106 | ptr = (void *) get_zeroed_page(GFP_ATOMIC); | |
107 | else | |
108 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
109 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) | |
110 | panic("set_pte_phys: cannot allocate page data %s\n", after_bootmem?"after bootmem":""); | |
111 | ||
112 | Dprintk("spp_getpage %p\n", ptr); | |
113 | return ptr; | |
114 | } | |
115 | ||
5f44a669 | 116 | static __init void set_pte_phys(unsigned long vaddr, |
1da177e4 LT |
117 | unsigned long phys, pgprot_t prot) |
118 | { | |
119 | pgd_t *pgd; | |
120 | pud_t *pud; | |
121 | pmd_t *pmd; | |
122 | pte_t *pte, new_pte; | |
123 | ||
124 | Dprintk("set_pte_phys %lx to %lx\n", vaddr, phys); | |
125 | ||
126 | pgd = pgd_offset_k(vaddr); | |
127 | if (pgd_none(*pgd)) { | |
128 | printk("PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
129 | return; | |
130 | } | |
131 | pud = pud_offset(pgd, vaddr); | |
132 | if (pud_none(*pud)) { | |
133 | pmd = (pmd_t *) spp_getpage(); | |
134 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER)); | |
135 | if (pmd != pmd_offset(pud, 0)) { | |
136 | printk("PAGETABLE BUG #01! %p <-> %p\n", pmd, pmd_offset(pud,0)); | |
137 | return; | |
138 | } | |
139 | } | |
140 | pmd = pmd_offset(pud, vaddr); | |
141 | if (pmd_none(*pmd)) { | |
142 | pte = (pte_t *) spp_getpage(); | |
143 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER)); | |
144 | if (pte != pte_offset_kernel(pmd, 0)) { | |
145 | printk("PAGETABLE BUG #02!\n"); | |
146 | return; | |
147 | } | |
148 | } | |
149 | new_pte = pfn_pte(phys >> PAGE_SHIFT, prot); | |
150 | ||
151 | pte = pte_offset_kernel(pmd, vaddr); | |
152 | if (!pte_none(*pte) && | |
153 | pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) | |
154 | pte_ERROR(*pte); | |
155 | set_pte(pte, new_pte); | |
156 | ||
157 | /* | |
158 | * It's enough to flush this one mapping. | |
159 | * (PGE mappings get flushed as well) | |
160 | */ | |
161 | __flush_tlb_one(vaddr); | |
162 | } | |
163 | ||
164 | /* NOTE: this is meant to be run only at boot */ | |
5f44a669 AK |
165 | void __init |
166 | __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t prot) | |
1da177e4 LT |
167 | { |
168 | unsigned long address = __fix_to_virt(idx); | |
169 | ||
170 | if (idx >= __end_of_fixed_addresses) { | |
171 | printk("Invalid __set_fixmap\n"); | |
172 | return; | |
173 | } | |
174 | set_pte_phys(address, phys, prot); | |
175 | } | |
176 | ||
a3142c8e | 177 | unsigned long __meminitdata table_start, table_end; |
1da177e4 | 178 | |
dafe41ee | 179 | static __meminit void *alloc_low_page(unsigned long *phys) |
1da177e4 | 180 | { |
dafe41ee | 181 | unsigned long pfn = table_end++; |
1da177e4 LT |
182 | void *adr; |
183 | ||
44df75e6 MT |
184 | if (after_bootmem) { |
185 | adr = (void *)get_zeroed_page(GFP_ATOMIC); | |
186 | *phys = __pa(adr); | |
187 | return adr; | |
188 | } | |
189 | ||
1da177e4 LT |
190 | if (pfn >= end_pfn) |
191 | panic("alloc_low_page: ran out of memory"); | |
dafe41ee VG |
192 | |
193 | adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); | |
44df75e6 | 194 | memset(adr, 0, PAGE_SIZE); |
dafe41ee VG |
195 | *phys = pfn * PAGE_SIZE; |
196 | return adr; | |
197 | } | |
1da177e4 | 198 | |
dafe41ee | 199 | static __meminit void unmap_low_page(void *adr) |
1da177e4 | 200 | { |
44df75e6 MT |
201 | |
202 | if (after_bootmem) | |
203 | return; | |
204 | ||
dafe41ee | 205 | early_iounmap(adr, PAGE_SIZE); |
1da177e4 LT |
206 | } |
207 | ||
f2d3efed | 208 | /* Must run before zap_low_mappings */ |
a3142c8e | 209 | __meminit void *early_ioremap(unsigned long addr, unsigned long size) |
f2d3efed | 210 | { |
dafe41ee VG |
211 | unsigned long vaddr; |
212 | pmd_t *pmd, *last_pmd; | |
213 | int i, pmds; | |
214 | ||
215 | pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
216 | vaddr = __START_KERNEL_map; | |
217 | pmd = level2_kernel_pgt; | |
218 | last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1; | |
219 | for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) { | |
220 | for (i = 0; i < pmds; i++) { | |
221 | if (pmd_present(pmd[i])) | |
222 | goto next; | |
223 | } | |
224 | vaddr += addr & ~PMD_MASK; | |
225 | addr &= PMD_MASK; | |
226 | for (i = 0; i < pmds; i++, addr += PMD_SIZE) | |
227 | set_pmd(pmd + i,__pmd(addr | _KERNPG_TABLE | _PAGE_PSE)); | |
228 | __flush_tlb(); | |
229 | return (void *)vaddr; | |
230 | next: | |
231 | ; | |
f2d3efed | 232 | } |
dafe41ee VG |
233 | printk("early_ioremap(0x%lx, %lu) failed\n", addr, size); |
234 | return NULL; | |
f2d3efed AK |
235 | } |
236 | ||
237 | /* To avoid virtual aliases later */ | |
a3142c8e | 238 | __meminit void early_iounmap(void *addr, unsigned long size) |
f2d3efed | 239 | { |
dafe41ee VG |
240 | unsigned long vaddr; |
241 | pmd_t *pmd; | |
242 | int i, pmds; | |
243 | ||
244 | vaddr = (unsigned long)addr; | |
245 | pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
246 | pmd = level2_kernel_pgt + pmd_index(vaddr); | |
247 | for (i = 0; i < pmds; i++) | |
248 | pmd_clear(pmd + i); | |
f2d3efed AK |
249 | __flush_tlb(); |
250 | } | |
251 | ||
44df75e6 | 252 | static void __meminit |
6ad91658 | 253 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end) |
44df75e6 | 254 | { |
6ad91658 | 255 | int i = pmd_index(address); |
44df75e6 | 256 | |
6ad91658 | 257 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
44df75e6 | 258 | unsigned long entry; |
6ad91658 | 259 | pmd_t *pmd = pmd_page + pmd_index(address); |
44df75e6 | 260 | |
5f51e139 JB |
261 | if (address >= end) { |
262 | if (!after_bootmem) | |
263 | for (; i < PTRS_PER_PMD; i++, pmd++) | |
264 | set_pmd(pmd, __pmd(0)); | |
44df75e6 MT |
265 | break; |
266 | } | |
6ad91658 KM |
267 | |
268 | if (pmd_val(*pmd)) | |
269 | continue; | |
270 | ||
44df75e6 MT |
271 | entry = _PAGE_NX|_PAGE_PSE|_KERNPG_TABLE|_PAGE_GLOBAL|address; |
272 | entry &= __supported_pte_mask; | |
273 | set_pmd(pmd, __pmd(entry)); | |
274 | } | |
275 | } | |
276 | ||
277 | static void __meminit | |
278 | phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end) | |
279 | { | |
6ad91658 KM |
280 | pmd_t *pmd = pmd_offset(pud,0); |
281 | spin_lock(&init_mm.page_table_lock); | |
282 | phys_pmd_init(pmd, address, end); | |
283 | spin_unlock(&init_mm.page_table_lock); | |
284 | __flush_tlb_all(); | |
44df75e6 MT |
285 | } |
286 | ||
6ad91658 | 287 | static void __meminit phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end) |
1da177e4 | 288 | { |
6ad91658 | 289 | int i = pud_index(addr); |
44df75e6 | 290 | |
44df75e6 | 291 | |
6ad91658 | 292 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE ) { |
6ad91658 KM |
293 | unsigned long pmd_phys; |
294 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 LT |
295 | pmd_t *pmd; |
296 | ||
6ad91658 | 297 | if (addr >= end) |
1da177e4 | 298 | break; |
1da177e4 | 299 | |
6ad91658 | 300 | if (!after_bootmem && !e820_any_mapped(addr,addr+PUD_SIZE,0)) { |
1da177e4 LT |
301 | set_pud(pud, __pud(0)); |
302 | continue; | |
303 | } | |
304 | ||
6ad91658 KM |
305 | if (pud_val(*pud)) { |
306 | phys_pmd_update(pud, addr, end); | |
307 | continue; | |
308 | } | |
309 | ||
dafe41ee | 310 | pmd = alloc_low_page(&pmd_phys); |
44df75e6 | 311 | spin_lock(&init_mm.page_table_lock); |
1da177e4 | 312 | set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE)); |
6ad91658 | 313 | phys_pmd_init(pmd, addr, end); |
44df75e6 | 314 | spin_unlock(&init_mm.page_table_lock); |
dafe41ee | 315 | unmap_low_page(pmd); |
1da177e4 LT |
316 | } |
317 | __flush_tlb(); | |
318 | } | |
319 | ||
320 | static void __init find_early_table_space(unsigned long end) | |
321 | { | |
6c5acd16 | 322 | unsigned long puds, pmds, tables, start; |
1da177e4 LT |
323 | |
324 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; | |
325 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; | |
326 | tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) + | |
327 | round_up(pmds * sizeof(pmd_t), PAGE_SIZE); | |
328 | ||
ee408c79 AK |
329 | /* RED-PEN putting page tables only on node 0 could |
330 | cause a hotspot and fill up ZONE_DMA. The page tables | |
331 | need roughly 0.5KB per GB. */ | |
332 | start = 0x8000; | |
333 | table_start = find_e820_area(start, end, tables); | |
1da177e4 LT |
334 | if (table_start == -1UL) |
335 | panic("Cannot find space for the kernel page tables"); | |
336 | ||
337 | table_start >>= PAGE_SHIFT; | |
338 | table_end = table_start; | |
44df75e6 MT |
339 | |
340 | early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n", | |
5f51e139 JB |
341 | end, table_start << PAGE_SHIFT, |
342 | (table_start << PAGE_SHIFT) + tables); | |
1da177e4 LT |
343 | } |
344 | ||
345 | /* Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
346 | This runs before bootmem is initialized and gets pages directly from the | |
347 | physical memory. To access them they are temporarily mapped. */ | |
44df75e6 | 348 | void __meminit init_memory_mapping(unsigned long start, unsigned long end) |
1da177e4 LT |
349 | { |
350 | unsigned long next; | |
351 | ||
352 | Dprintk("init_memory_mapping\n"); | |
353 | ||
354 | /* | |
355 | * Find space for the kernel direct mapping tables. | |
356 | * Later we should allocate these tables in the local node of the memory | |
357 | * mapped. Unfortunately this is done currently before the nodes are | |
358 | * discovered. | |
359 | */ | |
44df75e6 MT |
360 | if (!after_bootmem) |
361 | find_early_table_space(end); | |
1da177e4 LT |
362 | |
363 | start = (unsigned long)__va(start); | |
364 | end = (unsigned long)__va(end); | |
365 | ||
366 | for (; start < end; start = next) { | |
1da177e4 | 367 | unsigned long pud_phys; |
44df75e6 MT |
368 | pgd_t *pgd = pgd_offset_k(start); |
369 | pud_t *pud; | |
370 | ||
371 | if (after_bootmem) | |
d2ae5b5f | 372 | pud = pud_offset(pgd, start & PGDIR_MASK); |
44df75e6 | 373 | else |
dafe41ee | 374 | pud = alloc_low_page(&pud_phys); |
44df75e6 | 375 | |
1da177e4 LT |
376 | next = start + PGDIR_SIZE; |
377 | if (next > end) | |
378 | next = end; | |
379 | phys_pud_init(pud, __pa(start), __pa(next)); | |
44df75e6 MT |
380 | if (!after_bootmem) |
381 | set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys)); | |
dafe41ee | 382 | unmap_low_page(pud); |
1da177e4 LT |
383 | } |
384 | ||
44df75e6 MT |
385 | if (!after_bootmem) |
386 | asm volatile("movq %%cr4,%0" : "=r" (mmu_cr4_features)); | |
1da177e4 | 387 | __flush_tlb_all(); |
1da177e4 LT |
388 | } |
389 | ||
2b97690f | 390 | #ifndef CONFIG_NUMA |
1da177e4 LT |
391 | void __init paging_init(void) |
392 | { | |
6391af17 MG |
393 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
394 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
395 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
396 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
397 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
398 | ||
44df75e6 MT |
399 | memory_present(0, 0, end_pfn); |
400 | sparse_init(); | |
5cb248ab | 401 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
402 | } |
403 | #endif | |
404 | ||
405 | /* Unmap a kernel mapping if it exists. This is useful to avoid prefetches | |
406 | from the CPU leading to inconsistent cache lines. address and size | |
407 | must be aligned to 2MB boundaries. | |
408 | Does nothing when the mapping doesn't exist. */ | |
409 | void __init clear_kernel_mapping(unsigned long address, unsigned long size) | |
410 | { | |
411 | unsigned long end = address + size; | |
412 | ||
413 | BUG_ON(address & ~LARGE_PAGE_MASK); | |
414 | BUG_ON(size & ~LARGE_PAGE_MASK); | |
415 | ||
416 | for (; address < end; address += LARGE_PAGE_SIZE) { | |
417 | pgd_t *pgd = pgd_offset_k(address); | |
418 | pud_t *pud; | |
419 | pmd_t *pmd; | |
420 | if (pgd_none(*pgd)) | |
421 | continue; | |
422 | pud = pud_offset(pgd, address); | |
423 | if (pud_none(*pud)) | |
424 | continue; | |
425 | pmd = pmd_offset(pud, address); | |
426 | if (!pmd || pmd_none(*pmd)) | |
427 | continue; | |
428 | if (0 == (pmd_val(*pmd) & _PAGE_PSE)) { | |
429 | /* Could handle this, but it should not happen currently. */ | |
430 | printk(KERN_ERR | |
431 | "clear_kernel_mapping: mapping has been split. will leak memory\n"); | |
432 | pmd_ERROR(*pmd); | |
433 | } | |
434 | set_pmd(pmd, __pmd(0)); | |
435 | } | |
436 | __flush_tlb_all(); | |
437 | } | |
438 | ||
44df75e6 MT |
439 | /* |
440 | * Memory hotplug specific functions | |
44df75e6 | 441 | */ |
44df75e6 MT |
442 | void online_page(struct page *page) |
443 | { | |
444 | ClearPageReserved(page); | |
7835e98b | 445 | init_page_count(page); |
44df75e6 MT |
446 | __free_page(page); |
447 | totalram_pages++; | |
448 | num_physpages++; | |
449 | } | |
450 | ||
bc02af93 | 451 | #ifdef CONFIG_MEMORY_HOTPLUG |
9d99aaa3 AK |
452 | /* |
453 | * Memory is added always to NORMAL zone. This means you will never get | |
454 | * additional DMA/DMA32 memory. | |
455 | */ | |
bc02af93 | 456 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 457 | { |
bc02af93 | 458 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 459 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
44df75e6 MT |
460 | unsigned long start_pfn = start >> PAGE_SHIFT; |
461 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
462 | int ret; | |
463 | ||
45e0b78b KM |
464 | init_memory_mapping(start, (start + size -1)); |
465 | ||
44df75e6 MT |
466 | ret = __add_pages(zone, start_pfn, nr_pages); |
467 | if (ret) | |
468 | goto error; | |
469 | ||
44df75e6 MT |
470 | return ret; |
471 | error: | |
472 | printk("%s: Problem encountered in __add_pages!\n", __func__); | |
473 | return ret; | |
474 | } | |
bc02af93 | 475 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 MT |
476 | |
477 | int remove_memory(u64 start, u64 size) | |
478 | { | |
479 | return -EINVAL; | |
480 | } | |
481 | EXPORT_SYMBOL_GPL(remove_memory); | |
482 | ||
8243229f | 483 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
4942e998 KM |
484 | int memory_add_physaddr_to_nid(u64 start) |
485 | { | |
486 | return 0; | |
487 | } | |
8c2676a5 | 488 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
4942e998 KM |
489 | #endif |
490 | ||
45e0b78b KM |
491 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
492 | ||
493 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
bc02af93 YG |
494 | /* |
495 | * Memory Hotadd without sparsemem. The mem_maps have been allocated in advance, | |
496 | * just online the pages. | |
497 | */ | |
498 | int __add_pages(struct zone *z, unsigned long start_pfn, unsigned long nr_pages) | |
499 | { | |
500 | int err = -EIO; | |
501 | unsigned long pfn; | |
502 | unsigned long total = 0, mem = 0; | |
503 | for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++) { | |
504 | if (pfn_valid(pfn)) { | |
505 | online_page(pfn_to_page(pfn)); | |
506 | err = 0; | |
507 | mem++; | |
508 | } | |
509 | total++; | |
510 | } | |
511 | if (!err) { | |
512 | z->spanned_pages += total; | |
513 | z->present_pages += mem; | |
514 | z->zone_pgdat->node_spanned_pages += total; | |
515 | z->zone_pgdat->node_present_pages += mem; | |
516 | } | |
517 | return err; | |
518 | } | |
45e0b78b | 519 | #endif |
44df75e6 | 520 | |
1da177e4 LT |
521 | static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, kcore_modules, |
522 | kcore_vsyscall; | |
523 | ||
524 | void __init mem_init(void) | |
525 | { | |
0a43e4bf | 526 | long codesize, reservedpages, datasize, initsize; |
1da177e4 | 527 | |
0dc243ae | 528 | pci_iommu_alloc(); |
1da177e4 | 529 | |
1da177e4 LT |
530 | /* clear the zero-page */ |
531 | memset(empty_zero_page, 0, PAGE_SIZE); | |
532 | ||
533 | reservedpages = 0; | |
534 | ||
535 | /* this will put all low memory onto the freelists */ | |
2b97690f | 536 | #ifdef CONFIG_NUMA |
0a43e4bf | 537 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 538 | #else |
0a43e4bf | 539 | totalram_pages = free_all_bootmem(); |
1da177e4 | 540 | #endif |
5cb248ab MG |
541 | reservedpages = end_pfn - totalram_pages - |
542 | absent_pages_in_range(0, end_pfn); | |
1da177e4 LT |
543 | |
544 | after_bootmem = 1; | |
545 | ||
546 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
547 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
548 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
549 | ||
550 | /* Register memory areas for /proc/kcore */ | |
551 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); | |
552 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
553 | VMALLOC_END-VMALLOC_START); | |
554 | kclist_add(&kcore_kernel, &_stext, _end - _stext); | |
555 | kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); | |
556 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, | |
557 | VSYSCALL_END - VSYSCALL_START); | |
558 | ||
0a43e4bf | 559 | printk("Memory: %luk/%luk available (%ldk kernel code, %ldk reserved, %ldk data, %ldk init)\n", |
1da177e4 LT |
560 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
561 | end_pfn << (PAGE_SHIFT-10), | |
562 | codesize >> 10, | |
563 | reservedpages << (PAGE_SHIFT-10), | |
564 | datasize >> 10, | |
565 | initsize >> 10); | |
1da177e4 LT |
566 | } |
567 | ||
d167a518 | 568 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
1da177e4 LT |
569 | { |
570 | unsigned long addr; | |
571 | ||
d167a518 GH |
572 | if (begin >= end) |
573 | return; | |
574 | ||
6fb14755 | 575 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
d167a518 | 576 | for (addr = begin; addr < end; addr += PAGE_SIZE) { |
e3ebadd9 LT |
577 | ClearPageReserved(virt_to_page(addr)); |
578 | init_page_count(virt_to_page(addr)); | |
579 | memset((void *)(addr & ~(PAGE_SIZE-1)), | |
580 | POISON_FREE_INITMEM, PAGE_SIZE); | |
6fb14755 JB |
581 | if (addr >= __START_KERNEL_map) |
582 | change_page_attr_addr(addr, 1, __pgprot(0)); | |
e3ebadd9 | 583 | free_page(addr); |
1da177e4 LT |
584 | totalram_pages++; |
585 | } | |
6fb14755 JB |
586 | if (addr > __START_KERNEL_map) |
587 | global_flush_tlb(); | |
d167a518 GH |
588 | } |
589 | ||
590 | void free_initmem(void) | |
591 | { | |
d167a518 | 592 | free_init_pages("unused kernel memory", |
e3ebadd9 LT |
593 | (unsigned long)(&__init_begin), |
594 | (unsigned long)(&__init_end)); | |
1da177e4 LT |
595 | } |
596 | ||
67df197b AV |
597 | #ifdef CONFIG_DEBUG_RODATA |
598 | ||
67df197b AV |
599 | void mark_rodata_ro(void) |
600 | { | |
e3ebadd9 | 601 | unsigned long start = (unsigned long)_stext, end; |
67df197b | 602 | |
6fb14755 JB |
603 | #ifdef CONFIG_HOTPLUG_CPU |
604 | /* It must still be possible to apply SMP alternatives. */ | |
605 | if (num_possible_cpus() > 1) | |
e3ebadd9 | 606 | start = (unsigned long)_etext; |
6fb14755 | 607 | #endif |
0864a4e2 AV |
608 | |
609 | #ifdef CONFIG_KPROBES | |
610 | start = (unsigned long)__start_rodata; | |
611 | #endif | |
612 | ||
e3ebadd9 LT |
613 | end = (unsigned long)__end_rodata; |
614 | start = (start + PAGE_SIZE - 1) & PAGE_MASK; | |
615 | end &= PAGE_MASK; | |
616 | if (end <= start) | |
617 | return; | |
618 | ||
619 | change_page_attr_addr(start, (end - start) >> PAGE_SHIFT, PAGE_KERNEL_RO); | |
67df197b | 620 | |
6fb14755 | 621 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 622 | (end - start) >> 10); |
67df197b AV |
623 | |
624 | /* | |
625 | * change_page_attr_addr() requires a global_flush_tlb() call after it. | |
626 | * We do this after the printk so that if something went wrong in the | |
627 | * change, the printk gets out at least to give a better debug hint | |
628 | * of who is the culprit. | |
629 | */ | |
630 | global_flush_tlb(); | |
631 | } | |
632 | #endif | |
633 | ||
1da177e4 LT |
634 | #ifdef CONFIG_BLK_DEV_INITRD |
635 | void free_initrd_mem(unsigned long start, unsigned long end) | |
636 | { | |
e3ebadd9 | 637 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
638 | } |
639 | #endif | |
640 | ||
641 | void __init reserve_bootmem_generic(unsigned long phys, unsigned len) | |
642 | { | |
2b97690f | 643 | #ifdef CONFIG_NUMA |
1da177e4 | 644 | int nid = phys_to_nid(phys); |
5e58a02a AK |
645 | #endif |
646 | unsigned long pfn = phys >> PAGE_SHIFT; | |
647 | if (pfn >= end_pfn) { | |
648 | /* This can happen with kdump kernels when accessing firmware | |
649 | tables. */ | |
650 | if (pfn < end_pfn_map) | |
651 | return; | |
652 | printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n", | |
653 | phys, len); | |
654 | return; | |
655 | } | |
656 | ||
657 | /* Should check here against the e820 map to avoid double free */ | |
658 | #ifdef CONFIG_NUMA | |
1da177e4 LT |
659 | reserve_bootmem_node(NODE_DATA(nid), phys, len); |
660 | #else | |
661 | reserve_bootmem(phys, len); | |
662 | #endif | |
0e0b864e | 663 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
e18c6874 | 664 | dma_reserve += len / PAGE_SIZE; |
0e0b864e MG |
665 | set_dma_reserve(dma_reserve); |
666 | } | |
1da177e4 LT |
667 | } |
668 | ||
669 | int kern_addr_valid(unsigned long addr) | |
670 | { | |
671 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; | |
672 | pgd_t *pgd; | |
673 | pud_t *pud; | |
674 | pmd_t *pmd; | |
675 | pte_t *pte; | |
676 | ||
677 | if (above != 0 && above != -1UL) | |
678 | return 0; | |
679 | ||
680 | pgd = pgd_offset_k(addr); | |
681 | if (pgd_none(*pgd)) | |
682 | return 0; | |
683 | ||
684 | pud = pud_offset(pgd, addr); | |
685 | if (pud_none(*pud)) | |
686 | return 0; | |
687 | ||
688 | pmd = pmd_offset(pud, addr); | |
689 | if (pmd_none(*pmd)) | |
690 | return 0; | |
691 | if (pmd_large(*pmd)) | |
692 | return pfn_valid(pmd_pfn(*pmd)); | |
693 | ||
694 | pte = pte_offset_kernel(pmd, addr); | |
695 | if (pte_none(*pte)) | |
696 | return 0; | |
697 | return pfn_valid(pte_pfn(*pte)); | |
698 | } | |
699 | ||
700 | #ifdef CONFIG_SYSCTL | |
701 | #include <linux/sysctl.h> | |
702 | ||
703 | extern int exception_trace, page_fault_trace; | |
704 | ||
705 | static ctl_table debug_table2[] = { | |
c37ce032 EB |
706 | { |
707 | .ctl_name = 99, | |
708 | .procname = "exception-trace", | |
709 | .data = &exception_trace, | |
710 | .maxlen = sizeof(int), | |
711 | .mode = 0644, | |
712 | .proc_handler = proc_dointvec | |
713 | }, | |
714 | {} | |
1da177e4 LT |
715 | }; |
716 | ||
717 | static ctl_table debug_root_table2[] = { | |
c37ce032 EB |
718 | { |
719 | .ctl_name = CTL_DEBUG, | |
720 | .procname = "debug", | |
721 | .mode = 0555, | |
722 | .child = debug_table2 | |
723 | }, | |
724 | {} | |
1da177e4 LT |
725 | }; |
726 | ||
727 | static __init int x8664_sysctl_init(void) | |
728 | { | |
0b4d4147 | 729 | register_sysctl_table(debug_root_table2); |
1da177e4 LT |
730 | return 0; |
731 | } | |
732 | __initcall(x8664_sysctl_init); | |
733 | #endif | |
734 | ||
103efcd9 | 735 | /* A pseudo VMA to allow ptrace access for the vsyscall page. This only |
1e014410 AK |
736 | covers the 64bit vsyscall page now. 32bit has a real VMA now and does |
737 | not need special handling anymore. */ | |
1da177e4 LT |
738 | |
739 | static struct vm_area_struct gate_vma = { | |
740 | .vm_start = VSYSCALL_START, | |
103efcd9 EP |
741 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES << PAGE_SHIFT), |
742 | .vm_page_prot = PAGE_READONLY_EXEC, | |
743 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
744 | }; |
745 | ||
1da177e4 LT |
746 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
747 | { | |
748 | #ifdef CONFIG_IA32_EMULATION | |
1e014410 AK |
749 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
750 | return NULL; | |
1da177e4 LT |
751 | #endif |
752 | return &gate_vma; | |
753 | } | |
754 | ||
755 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
756 | { | |
757 | struct vm_area_struct *vma = get_gate_vma(task); | |
1e014410 AK |
758 | if (!vma) |
759 | return 0; | |
1da177e4 LT |
760 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
761 | } | |
762 | ||
763 | /* Use this when you have no reliable task/vma, typically from interrupt | |
764 | * context. It is less reliable than using the task's vma and may give | |
765 | * false positives. | |
766 | */ | |
767 | int in_gate_area_no_task(unsigned long addr) | |
768 | { | |
1e014410 | 769 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 770 | } |
2e1c49db ZN |
771 | |
772 | void *alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size) | |
773 | { | |
774 | return __alloc_bootmem_core(pgdat->bdata, size, | |
775 | SMP_CACHE_BYTES, (4UL*1024*1024*1024), 0); | |
776 | } |