]>
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> |
718fc13b | 46 | #include <asm/kdebug.h> |
aaa64e04 | 47 | #include <asm/numa.h> |
1da177e4 | 48 | |
14a62c34 | 49 | const struct dma_mapping_ops *dma_ops; |
17a941d8 MBY |
50 | EXPORT_SYMBOL(dma_ops); |
51 | ||
e18c6874 AK |
52 | static unsigned long dma_reserve __initdata; |
53 | ||
1da177e4 LT |
54 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
55 | ||
56 | /* | |
57 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the | |
58 | * physical space so we can cache the place of the first one and move | |
59 | * around without checking the pgd every time. | |
60 | */ | |
61 | ||
62 | void show_mem(void) | |
63 | { | |
e92343cc AK |
64 | long i, total = 0, reserved = 0; |
65 | long shared = 0, cached = 0; | |
1da177e4 | 66 | struct page *page; |
14a62c34 | 67 | pg_data_t *pgdat; |
1da177e4 | 68 | |
e92343cc | 69 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 70 | show_free_areas(); |
14a62c34 TG |
71 | printk(KERN_INFO "Free swap: %6ldkB\n", |
72 | nr_swap_pages << (PAGE_SHIFT-10)); | |
1da177e4 | 73 | |
ec936fc5 | 74 | for_each_online_pgdat(pgdat) { |
14a62c34 TG |
75 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
76 | /* | |
77 | * This loop can take a while with 256 GB and | |
78 | * 4k pages so defer the NMI watchdog: | |
79 | */ | |
80 | if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) | |
ae32b129 | 81 | touch_nmi_watchdog(); |
14a62c34 | 82 | |
12710a56 BP |
83 | if (!pfn_valid(pgdat->node_start_pfn + i)) |
84 | continue; | |
14a62c34 | 85 | |
1da177e4 LT |
86 | page = pfn_to_page(pgdat->node_start_pfn + i); |
87 | total++; | |
e92343cc AK |
88 | if (PageReserved(page)) |
89 | reserved++; | |
90 | else if (PageSwapCache(page)) | |
91 | cached++; | |
92 | else if (page_count(page)) | |
93 | shared += page_count(page) - 1; | |
14a62c34 | 94 | } |
1da177e4 | 95 | } |
14a62c34 TG |
96 | printk(KERN_INFO "%lu pages of RAM\n", total); |
97 | printk(KERN_INFO "%lu reserved pages\n", reserved); | |
98 | printk(KERN_INFO "%lu pages shared\n", shared); | |
99 | printk(KERN_INFO "%lu pages swap cached\n", cached); | |
1da177e4 LT |
100 | } |
101 | ||
1da177e4 LT |
102 | int after_bootmem; |
103 | ||
5f44a669 | 104 | static __init void *spp_getpage(void) |
14a62c34 | 105 | { |
1da177e4 | 106 | void *ptr; |
14a62c34 | 107 | |
1da177e4 | 108 | if (after_bootmem) |
14a62c34 | 109 | ptr = (void *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
110 | else |
111 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
14a62c34 TG |
112 | |
113 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { | |
114 | panic("set_pte_phys: cannot allocate page data %s\n", | |
115 | after_bootmem ? "after bootmem" : ""); | |
116 | } | |
1da177e4 | 117 | |
10f22dde | 118 | pr_debug("spp_getpage %p\n", ptr); |
14a62c34 | 119 | |
1da177e4 | 120 | return ptr; |
14a62c34 | 121 | } |
1da177e4 | 122 | |
14a62c34 TG |
123 | static __init void |
124 | set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot) | |
1da177e4 LT |
125 | { |
126 | pgd_t *pgd; | |
127 | pud_t *pud; | |
128 | pmd_t *pmd; | |
129 | pte_t *pte, new_pte; | |
130 | ||
10f22dde | 131 | pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys); |
1da177e4 LT |
132 | |
133 | pgd = pgd_offset_k(vaddr); | |
134 | if (pgd_none(*pgd)) { | |
10f22dde IM |
135 | printk(KERN_ERR |
136 | "PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
1da177e4 LT |
137 | return; |
138 | } | |
139 | pud = pud_offset(pgd, vaddr); | |
140 | if (pud_none(*pud)) { | |
14a62c34 | 141 | pmd = (pmd_t *) spp_getpage(); |
1da177e4 LT |
142 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER)); |
143 | if (pmd != pmd_offset(pud, 0)) { | |
10f22dde | 144 | printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", |
14a62c34 | 145 | pmd, pmd_offset(pud, 0)); |
1da177e4 LT |
146 | return; |
147 | } | |
148 | } | |
149 | pmd = pmd_offset(pud, vaddr); | |
150 | if (pmd_none(*pmd)) { | |
151 | pte = (pte_t *) spp_getpage(); | |
152 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER)); | |
153 | if (pte != pte_offset_kernel(pmd, 0)) { | |
10f22dde | 154 | printk(KERN_ERR "PAGETABLE BUG #02!\n"); |
1da177e4 LT |
155 | return; |
156 | } | |
157 | } | |
158 | new_pte = pfn_pte(phys >> PAGE_SHIFT, prot); | |
159 | ||
160 | pte = pte_offset_kernel(pmd, vaddr); | |
161 | if (!pte_none(*pte) && | |
162 | pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) | |
163 | pte_ERROR(*pte); | |
164 | set_pte(pte, new_pte); | |
165 | ||
166 | /* | |
167 | * It's enough to flush this one mapping. | |
168 | * (PGE mappings get flushed as well) | |
169 | */ | |
170 | __flush_tlb_one(vaddr); | |
171 | } | |
172 | ||
173 | /* NOTE: this is meant to be run only at boot */ | |
14a62c34 TG |
174 | void __init |
175 | __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot) | |
1da177e4 LT |
176 | { |
177 | unsigned long address = __fix_to_virt(idx); | |
178 | ||
179 | if (idx >= __end_of_fixed_addresses) { | |
10f22dde | 180 | printk(KERN_ERR "Invalid __set_fixmap\n"); |
1da177e4 LT |
181 | return; |
182 | } | |
183 | set_pte_phys(address, phys, prot); | |
184 | } | |
185 | ||
75175278 AK |
186 | static unsigned long __initdata table_start; |
187 | static unsigned long __meminitdata table_end; | |
1da177e4 | 188 | |
dafe41ee | 189 | static __meminit void *alloc_low_page(unsigned long *phys) |
14a62c34 | 190 | { |
dafe41ee | 191 | unsigned long pfn = table_end++; |
1da177e4 LT |
192 | void *adr; |
193 | ||
44df75e6 MT |
194 | if (after_bootmem) { |
195 | adr = (void *)get_zeroed_page(GFP_ATOMIC); | |
196 | *phys = __pa(adr); | |
14a62c34 | 197 | |
44df75e6 MT |
198 | return adr; |
199 | } | |
200 | ||
14a62c34 TG |
201 | if (pfn >= end_pfn) |
202 | panic("alloc_low_page: ran out of memory"); | |
dafe41ee VG |
203 | |
204 | adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); | |
44df75e6 | 205 | memset(adr, 0, PAGE_SIZE); |
dafe41ee VG |
206 | *phys = pfn * PAGE_SIZE; |
207 | return adr; | |
208 | } | |
1da177e4 | 209 | |
dafe41ee | 210 | static __meminit void unmap_low_page(void *adr) |
14a62c34 | 211 | { |
44df75e6 MT |
212 | if (after_bootmem) |
213 | return; | |
214 | ||
dafe41ee | 215 | early_iounmap(adr, PAGE_SIZE); |
14a62c34 | 216 | } |
1da177e4 | 217 | |
f2d3efed | 218 | /* Must run before zap_low_mappings */ |
a3142c8e | 219 | __meminit void *early_ioremap(unsigned long addr, unsigned long size) |
f2d3efed | 220 | { |
dafe41ee | 221 | pmd_t *pmd, *last_pmd; |
14a62c34 | 222 | unsigned long vaddr; |
dafe41ee VG |
223 | int i, pmds; |
224 | ||
225 | pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
226 | vaddr = __START_KERNEL_map; | |
227 | pmd = level2_kernel_pgt; | |
228 | last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1; | |
14a62c34 | 229 | |
dafe41ee VG |
230 | for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) { |
231 | for (i = 0; i < pmds; i++) { | |
232 | if (pmd_present(pmd[i])) | |
14a62c34 | 233 | goto continue_outer_loop; |
dafe41ee VG |
234 | } |
235 | vaddr += addr & ~PMD_MASK; | |
236 | addr &= PMD_MASK; | |
14a62c34 | 237 | |
dafe41ee | 238 | for (i = 0; i < pmds; i++, addr += PMD_SIZE) |
929fd589 | 239 | set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC)); |
1a2b4412 | 240 | __flush_tlb_all(); |
14a62c34 | 241 | |
dafe41ee | 242 | return (void *)vaddr; |
14a62c34 | 243 | continue_outer_loop: |
dafe41ee | 244 | ; |
f2d3efed | 245 | } |
10f22dde | 246 | printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size); |
14a62c34 | 247 | |
dafe41ee | 248 | return NULL; |
f2d3efed AK |
249 | } |
250 | ||
14a62c34 TG |
251 | /* |
252 | * To avoid virtual aliases later: | |
253 | */ | |
a3142c8e | 254 | __meminit void early_iounmap(void *addr, unsigned long size) |
f2d3efed | 255 | { |
dafe41ee VG |
256 | unsigned long vaddr; |
257 | pmd_t *pmd; | |
258 | int i, pmds; | |
259 | ||
260 | vaddr = (unsigned long)addr; | |
261 | pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
262 | pmd = level2_kernel_pgt + pmd_index(vaddr); | |
14a62c34 | 263 | |
dafe41ee VG |
264 | for (i = 0; i < pmds; i++) |
265 | pmd_clear(pmd + i); | |
14a62c34 | 266 | |
1a2b4412 | 267 | __flush_tlb_all(); |
f2d3efed AK |
268 | } |
269 | ||
44df75e6 | 270 | static void __meminit |
6ad91658 | 271 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end) |
44df75e6 | 272 | { |
6ad91658 | 273 | int i = pmd_index(address); |
44df75e6 | 274 | |
6ad91658 | 275 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
6ad91658 | 276 | pmd_t *pmd = pmd_page + pmd_index(address); |
44df75e6 | 277 | |
5f51e139 | 278 | if (address >= end) { |
14a62c34 | 279 | if (!after_bootmem) { |
5f51e139 JB |
280 | for (; i < PTRS_PER_PMD; i++, pmd++) |
281 | set_pmd(pmd, __pmd(0)); | |
14a62c34 | 282 | } |
44df75e6 MT |
283 | break; |
284 | } | |
6ad91658 KM |
285 | |
286 | if (pmd_val(*pmd)) | |
287 | continue; | |
288 | ||
d4f71f79 AK |
289 | set_pte((pte_t *)pmd, |
290 | pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
44df75e6 MT |
291 | } |
292 | } | |
293 | ||
294 | static void __meminit | |
295 | phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end) | |
296 | { | |
14a62c34 | 297 | pmd_t *pmd = pmd_offset(pud, 0); |
6ad91658 KM |
298 | spin_lock(&init_mm.page_table_lock); |
299 | phys_pmd_init(pmd, address, end); | |
300 | spin_unlock(&init_mm.page_table_lock); | |
301 | __flush_tlb_all(); | |
44df75e6 MT |
302 | } |
303 | ||
14a62c34 TG |
304 | static void __meminit |
305 | phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end) | |
306 | { | |
6ad91658 | 307 | int i = pud_index(addr); |
44df75e6 | 308 | |
14a62c34 | 309 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { |
6ad91658 KM |
310 | unsigned long pmd_phys; |
311 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 LT |
312 | pmd_t *pmd; |
313 | ||
6ad91658 | 314 | if (addr >= end) |
1da177e4 | 315 | break; |
1da177e4 | 316 | |
14a62c34 TG |
317 | if (!after_bootmem && |
318 | !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { | |
319 | set_pud(pud, __pud(0)); | |
1da177e4 | 320 | continue; |
14a62c34 | 321 | } |
1da177e4 | 322 | |
6ad91658 KM |
323 | if (pud_val(*pud)) { |
324 | phys_pmd_update(pud, addr, end); | |
325 | continue; | |
326 | } | |
327 | ||
dafe41ee | 328 | pmd = alloc_low_page(&pmd_phys); |
14a62c34 | 329 | |
44df75e6 | 330 | spin_lock(&init_mm.page_table_lock); |
1da177e4 | 331 | set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE)); |
6ad91658 | 332 | phys_pmd_init(pmd, addr, end); |
44df75e6 | 333 | spin_unlock(&init_mm.page_table_lock); |
14a62c34 | 334 | |
dafe41ee | 335 | unmap_low_page(pmd); |
1da177e4 | 336 | } |
1a2b4412 | 337 | __flush_tlb_all(); |
14a62c34 | 338 | } |
1da177e4 LT |
339 | |
340 | static void __init find_early_table_space(unsigned long end) | |
341 | { | |
6c5acd16 | 342 | unsigned long puds, pmds, tables, start; |
1da177e4 LT |
343 | |
344 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; | |
345 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; | |
346 | tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) + | |
347 | round_up(pmds * sizeof(pmd_t), PAGE_SIZE); | |
348 | ||
14a62c34 TG |
349 | /* |
350 | * RED-PEN putting page tables only on node 0 could | |
351 | * cause a hotspot and fill up ZONE_DMA. The page tables | |
352 | * need roughly 0.5KB per GB. | |
353 | */ | |
354 | start = 0x8000; | |
24a5da73 | 355 | table_start = find_e820_area(start, end, tables, PAGE_SIZE); |
1da177e4 LT |
356 | if (table_start == -1UL) |
357 | panic("Cannot find space for the kernel page tables"); | |
358 | ||
359 | table_start >>= PAGE_SHIFT; | |
360 | table_end = table_start; | |
44df75e6 MT |
361 | |
362 | early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n", | |
5f51e139 JB |
363 | end, table_start << PAGE_SHIFT, |
364 | (table_start << PAGE_SHIFT) + tables); | |
1da177e4 LT |
365 | } |
366 | ||
14a62c34 TG |
367 | /* |
368 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
369 | * This runs before bootmem is initialized and gets pages directly from | |
370 | * the physical memory. To access them they are temporarily mapped. | |
371 | */ | |
b6fd6ecb | 372 | void __init_refok init_memory_mapping(unsigned long start, unsigned long end) |
14a62c34 TG |
373 | { |
374 | unsigned long next; | |
1da177e4 | 375 | |
10f22dde | 376 | pr_debug("init_memory_mapping\n"); |
1da177e4 | 377 | |
14a62c34 | 378 | /* |
1da177e4 | 379 | * Find space for the kernel direct mapping tables. |
14a62c34 TG |
380 | * |
381 | * Later we should allocate these tables in the local node of the | |
382 | * memory mapped. Unfortunately this is done currently before the | |
383 | * nodes are discovered. | |
1da177e4 | 384 | */ |
44df75e6 MT |
385 | if (!after_bootmem) |
386 | find_early_table_space(end); | |
1da177e4 LT |
387 | |
388 | start = (unsigned long)__va(start); | |
389 | end = (unsigned long)__va(end); | |
390 | ||
391 | for (; start < end; start = next) { | |
44df75e6 | 392 | pgd_t *pgd = pgd_offset_k(start); |
14a62c34 | 393 | unsigned long pud_phys; |
44df75e6 MT |
394 | pud_t *pud; |
395 | ||
396 | if (after_bootmem) | |
d2ae5b5f | 397 | pud = pud_offset(pgd, start & PGDIR_MASK); |
44df75e6 | 398 | else |
dafe41ee | 399 | pud = alloc_low_page(&pud_phys); |
44df75e6 | 400 | |
1da177e4 | 401 | next = start + PGDIR_SIZE; |
14a62c34 TG |
402 | if (next > end) |
403 | next = end; | |
1da177e4 | 404 | phys_pud_init(pud, __pa(start), __pa(next)); |
44df75e6 MT |
405 | if (!after_bootmem) |
406 | set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys)); | |
dafe41ee | 407 | unmap_low_page(pud); |
14a62c34 | 408 | } |
1da177e4 | 409 | |
44df75e6 | 410 | if (!after_bootmem) |
f51c9452 | 411 | mmu_cr4_features = read_cr4(); |
1da177e4 | 412 | __flush_tlb_all(); |
75175278 | 413 | |
24a5da73 YL |
414 | if (!after_bootmem) |
415 | reserve_early(table_start << PAGE_SHIFT, | |
416 | table_end << PAGE_SHIFT, "PGTABLE"); | |
1da177e4 LT |
417 | } |
418 | ||
2b97690f | 419 | #ifndef CONFIG_NUMA |
1da177e4 LT |
420 | void __init paging_init(void) |
421 | { | |
6391af17 | 422 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
14a62c34 | 423 | |
6391af17 MG |
424 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
425 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
426 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
427 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
428 | ||
44df75e6 MT |
429 | memory_present(0, 0, end_pfn); |
430 | sparse_init(); | |
5cb248ab | 431 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
432 | } |
433 | #endif | |
434 | ||
44df75e6 MT |
435 | /* |
436 | * Memory hotplug specific functions | |
44df75e6 | 437 | */ |
44df75e6 MT |
438 | void online_page(struct page *page) |
439 | { | |
440 | ClearPageReserved(page); | |
7835e98b | 441 | init_page_count(page); |
44df75e6 MT |
442 | __free_page(page); |
443 | totalram_pages++; | |
444 | num_physpages++; | |
445 | } | |
446 | ||
bc02af93 | 447 | #ifdef CONFIG_MEMORY_HOTPLUG |
9d99aaa3 AK |
448 | /* |
449 | * Memory is added always to NORMAL zone. This means you will never get | |
450 | * additional DMA/DMA32 memory. | |
451 | */ | |
bc02af93 | 452 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 453 | { |
bc02af93 | 454 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 455 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
44df75e6 MT |
456 | unsigned long start_pfn = start >> PAGE_SHIFT; |
457 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
458 | int ret; | |
459 | ||
14a62c34 | 460 | init_memory_mapping(start, start + size-1); |
45e0b78b | 461 | |
44df75e6 | 462 | ret = __add_pages(zone, start_pfn, nr_pages); |
10f22dde | 463 | WARN_ON(1); |
44df75e6 | 464 | |
44df75e6 | 465 | return ret; |
44df75e6 | 466 | } |
bc02af93 | 467 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 | 468 | |
8243229f | 469 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
4942e998 KM |
470 | int memory_add_physaddr_to_nid(u64 start) |
471 | { | |
472 | return 0; | |
473 | } | |
8c2676a5 | 474 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
4942e998 KM |
475 | #endif |
476 | ||
45e0b78b KM |
477 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
478 | ||
14a62c34 TG |
479 | static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, |
480 | kcore_modules, kcore_vsyscall; | |
1da177e4 LT |
481 | |
482 | void __init mem_init(void) | |
483 | { | |
0a43e4bf | 484 | long codesize, reservedpages, datasize, initsize; |
1da177e4 | 485 | |
0dc243ae | 486 | pci_iommu_alloc(); |
1da177e4 | 487 | |
48ddb154 | 488 | /* clear_bss() already clear the empty_zero_page */ |
1da177e4 | 489 | |
f2633105 IM |
490 | /* temporary debugging - double check it's true: */ |
491 | { | |
492 | int i; | |
493 | ||
494 | for (i = 0; i < 1024; i++) | |
495 | WARN_ON_ONCE(empty_zero_page[i]); | |
496 | } | |
497 | ||
1da177e4 LT |
498 | reservedpages = 0; |
499 | ||
500 | /* this will put all low memory onto the freelists */ | |
2b97690f | 501 | #ifdef CONFIG_NUMA |
0a43e4bf | 502 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 503 | #else |
0a43e4bf | 504 | totalram_pages = free_all_bootmem(); |
1da177e4 | 505 | #endif |
5cb248ab MG |
506 | reservedpages = end_pfn - totalram_pages - |
507 | absent_pages_in_range(0, end_pfn); | |
1da177e4 LT |
508 | after_bootmem = 1; |
509 | ||
510 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
511 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
512 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
513 | ||
514 | /* Register memory areas for /proc/kcore */ | |
14a62c34 TG |
515 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
516 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
1da177e4 LT |
517 | VMALLOC_END-VMALLOC_START); |
518 | kclist_add(&kcore_kernel, &_stext, _end - _stext); | |
519 | kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); | |
14a62c34 | 520 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
1da177e4 LT |
521 | VSYSCALL_END - VSYSCALL_START); |
522 | ||
10f22dde | 523 | printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " |
14a62c34 | 524 | "%ldk reserved, %ldk data, %ldk init)\n", |
1da177e4 LT |
525 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
526 | end_pfn << (PAGE_SHIFT-10), | |
527 | codesize >> 10, | |
528 | reservedpages << (PAGE_SHIFT-10), | |
529 | datasize >> 10, | |
530 | initsize >> 10); | |
1da177e4 LT |
531 | } |
532 | ||
d167a518 | 533 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
1da177e4 LT |
534 | { |
535 | unsigned long addr; | |
536 | ||
d167a518 GH |
537 | if (begin >= end) |
538 | return; | |
539 | ||
ee01f112 IM |
540 | /* |
541 | * If debugging page accesses then do not free this memory but | |
542 | * mark them not present - any buggy init-section access will | |
543 | * create a kernel page fault: | |
544 | */ | |
545 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
546 | printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", | |
547 | begin, PAGE_ALIGN(end)); | |
548 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
549 | #else | |
6fb14755 | 550 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
14a62c34 | 551 | |
d167a518 | 552 | for (addr = begin; addr < end; addr += PAGE_SIZE) { |
e3ebadd9 LT |
553 | ClearPageReserved(virt_to_page(addr)); |
554 | init_page_count(virt_to_page(addr)); | |
555 | memset((void *)(addr & ~(PAGE_SIZE-1)), | |
556 | POISON_FREE_INITMEM, PAGE_SIZE); | |
e3ebadd9 | 557 | free_page(addr); |
1da177e4 LT |
558 | totalram_pages++; |
559 | } | |
ee01f112 | 560 | #endif |
d167a518 GH |
561 | } |
562 | ||
563 | void free_initmem(void) | |
564 | { | |
d167a518 | 565 | free_init_pages("unused kernel memory", |
e3ebadd9 LT |
566 | (unsigned long)(&__init_begin), |
567 | (unsigned long)(&__init_end)); | |
1da177e4 LT |
568 | } |
569 | ||
67df197b | 570 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
571 | const int rodata_test_data = 0xC3; |
572 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
67df197b | 573 | |
67df197b AV |
574 | void mark_rodata_ro(void) |
575 | { | |
e3ebadd9 | 576 | unsigned long start = (unsigned long)_stext, end; |
67df197b | 577 | |
602033ed LT |
578 | #ifdef CONFIG_HOTPLUG_CPU |
579 | /* It must still be possible to apply SMP alternatives. */ | |
580 | if (num_possible_cpus() > 1) | |
581 | start = (unsigned long)_etext; | |
582 | #endif | |
583 | ||
584 | #ifdef CONFIG_KPROBES | |
585 | start = (unsigned long)__start_rodata; | |
586 | #endif | |
14a62c34 | 587 | |
e3ebadd9 LT |
588 | end = (unsigned long)__end_rodata; |
589 | start = (start + PAGE_SIZE - 1) & PAGE_MASK; | |
590 | end &= PAGE_MASK; | |
591 | if (end <= start) | |
592 | return; | |
593 | ||
67df197b | 594 | |
6fb14755 | 595 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 596 | (end - start) >> 10); |
984bb80d AV |
597 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
598 | ||
599 | /* | |
600 | * The rodata section (but not the kernel text!) should also be | |
601 | * not-executable. | |
602 | */ | |
603 | start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; | |
604 | set_memory_nx(start, (end - start) >> PAGE_SHIFT); | |
67df197b | 605 | |
1a487252 AV |
606 | rodata_test(); |
607 | ||
0c42f392 | 608 | #ifdef CONFIG_CPA_DEBUG |
10f22dde | 609 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); |
6d238cc4 | 610 | set_memory_rw(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 611 | |
10f22dde | 612 | printk(KERN_INFO "Testing CPA: again\n"); |
6d238cc4 | 613 | set_memory_ro(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 614 | #endif |
67df197b AV |
615 | } |
616 | #endif | |
617 | ||
1da177e4 LT |
618 | #ifdef CONFIG_BLK_DEV_INITRD |
619 | void free_initrd_mem(unsigned long start, unsigned long end) | |
620 | { | |
e3ebadd9 | 621 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
622 | } |
623 | #endif | |
624 | ||
14a62c34 TG |
625 | void __init reserve_bootmem_generic(unsigned long phys, unsigned len) |
626 | { | |
2b97690f | 627 | #ifdef CONFIG_NUMA |
1da177e4 | 628 | int nid = phys_to_nid(phys); |
5e58a02a AK |
629 | #endif |
630 | unsigned long pfn = phys >> PAGE_SHIFT; | |
14a62c34 | 631 | |
5e58a02a | 632 | if (pfn >= end_pfn) { |
14a62c34 TG |
633 | /* |
634 | * This can happen with kdump kernels when accessing | |
635 | * firmware tables: | |
636 | */ | |
5e58a02a AK |
637 | if (pfn < end_pfn_map) |
638 | return; | |
14a62c34 | 639 | |
5e58a02a AK |
640 | printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n", |
641 | phys, len); | |
642 | return; | |
643 | } | |
644 | ||
645 | /* Should check here against the e820 map to avoid double free */ | |
646 | #ifdef CONFIG_NUMA | |
72a7fe39 | 647 | reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT); |
14a62c34 | 648 | #else |
72a7fe39 | 649 | reserve_bootmem(phys, len, BOOTMEM_DEFAULT); |
1da177e4 | 650 | #endif |
0e0b864e | 651 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
e18c6874 | 652 | dma_reserve += len / PAGE_SIZE; |
0e0b864e MG |
653 | set_dma_reserve(dma_reserve); |
654 | } | |
1da177e4 LT |
655 | } |
656 | ||
14a62c34 TG |
657 | int kern_addr_valid(unsigned long addr) |
658 | { | |
1da177e4 | 659 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
14a62c34 TG |
660 | pgd_t *pgd; |
661 | pud_t *pud; | |
662 | pmd_t *pmd; | |
663 | pte_t *pte; | |
1da177e4 LT |
664 | |
665 | if (above != 0 && above != -1UL) | |
14a62c34 TG |
666 | return 0; |
667 | ||
1da177e4 LT |
668 | pgd = pgd_offset_k(addr); |
669 | if (pgd_none(*pgd)) | |
670 | return 0; | |
671 | ||
672 | pud = pud_offset(pgd, addr); | |
673 | if (pud_none(*pud)) | |
14a62c34 | 674 | return 0; |
1da177e4 LT |
675 | |
676 | pmd = pmd_offset(pud, addr); | |
677 | if (pmd_none(*pmd)) | |
678 | return 0; | |
14a62c34 | 679 | |
1da177e4 LT |
680 | if (pmd_large(*pmd)) |
681 | return pfn_valid(pmd_pfn(*pmd)); | |
682 | ||
683 | pte = pte_offset_kernel(pmd, addr); | |
684 | if (pte_none(*pte)) | |
685 | return 0; | |
14a62c34 | 686 | |
1da177e4 LT |
687 | return pfn_valid(pte_pfn(*pte)); |
688 | } | |
689 | ||
14a62c34 TG |
690 | /* |
691 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
692 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
693 | * not need special handling anymore: | |
694 | */ | |
1da177e4 | 695 | static struct vm_area_struct gate_vma = { |
14a62c34 TG |
696 | .vm_start = VSYSCALL_START, |
697 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), | |
698 | .vm_page_prot = PAGE_READONLY_EXEC, | |
699 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
700 | }; |
701 | ||
1da177e4 LT |
702 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
703 | { | |
704 | #ifdef CONFIG_IA32_EMULATION | |
1e014410 AK |
705 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
706 | return NULL; | |
1da177e4 LT |
707 | #endif |
708 | return &gate_vma; | |
709 | } | |
710 | ||
711 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
712 | { | |
713 | struct vm_area_struct *vma = get_gate_vma(task); | |
14a62c34 | 714 | |
1e014410 AK |
715 | if (!vma) |
716 | return 0; | |
14a62c34 | 717 | |
1da177e4 LT |
718 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
719 | } | |
720 | ||
14a62c34 TG |
721 | /* |
722 | * Use this when you have no reliable task/vma, typically from interrupt | |
723 | * context. It is less reliable than using the task's vma and may give | |
724 | * false positives: | |
1da177e4 LT |
725 | */ |
726 | int in_gate_area_no_task(unsigned long addr) | |
727 | { | |
1e014410 | 728 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 729 | } |
2e1c49db | 730 | |
2aae950b AK |
731 | const char *arch_vma_name(struct vm_area_struct *vma) |
732 | { | |
733 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
734 | return "[vdso]"; | |
735 | if (vma == &gate_vma) | |
736 | return "[vsyscall]"; | |
737 | return NULL; | |
738 | } | |
0889eba5 CL |
739 | |
740 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
741 | /* | |
742 | * Initialise the sparsemem vmemmap using huge-pages at the PMD level. | |
743 | */ | |
14a62c34 TG |
744 | int __meminit |
745 | vmemmap_populate(struct page *start_page, unsigned long size, int node) | |
0889eba5 CL |
746 | { |
747 | unsigned long addr = (unsigned long)start_page; | |
748 | unsigned long end = (unsigned long)(start_page + size); | |
749 | unsigned long next; | |
750 | pgd_t *pgd; | |
751 | pud_t *pud; | |
752 | pmd_t *pmd; | |
753 | ||
754 | for (; addr < end; addr = next) { | |
755 | next = pmd_addr_end(addr, end); | |
756 | ||
757 | pgd = vmemmap_pgd_populate(addr, node); | |
758 | if (!pgd) | |
759 | return -ENOMEM; | |
14a62c34 | 760 | |
0889eba5 CL |
761 | pud = vmemmap_pud_populate(pgd, addr, node); |
762 | if (!pud) | |
763 | return -ENOMEM; | |
764 | ||
765 | pmd = pmd_offset(pud, addr); | |
766 | if (pmd_none(*pmd)) { | |
767 | pte_t entry; | |
14a62c34 TG |
768 | void *p; |
769 | ||
770 | p = vmemmap_alloc_block(PMD_SIZE, node); | |
0889eba5 CL |
771 | if (!p) |
772 | return -ENOMEM; | |
773 | ||
14a62c34 TG |
774 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, |
775 | PAGE_KERNEL_LARGE); | |
0889eba5 CL |
776 | set_pmd(pmd, __pmd(pte_val(entry))); |
777 | ||
778 | printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n", | |
779 | addr, addr + PMD_SIZE - 1, p, node); | |
14a62c34 | 780 | } else { |
0889eba5 | 781 | vmemmap_verify((pte_t *)pmd, node, addr, next); |
14a62c34 | 782 | } |
0889eba5 | 783 | } |
0889eba5 CL |
784 | return 0; |
785 | } | |
786 | #endif |