]>
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> |
7bfeab9a | 48 | #include <asm/cacheflush.h> |
1da177e4 | 49 | |
14a62c34 | 50 | const struct dma_mapping_ops *dma_ops; |
17a941d8 MBY |
51 | EXPORT_SYMBOL(dma_ops); |
52 | ||
e18c6874 AK |
53 | static unsigned long dma_reserve __initdata; |
54 | ||
1da177e4 LT |
55 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
56 | ||
00d1c5e0 IM |
57 | int direct_gbpages __meminitdata |
58 | #ifdef CONFIG_DIRECT_GBPAGES | |
59 | = 1 | |
60 | #endif | |
61 | ; | |
62 | ||
63 | static int __init parse_direct_gbpages_off(char *arg) | |
64 | { | |
65 | direct_gbpages = 0; | |
66 | return 0; | |
67 | } | |
68 | early_param("nogbpages", parse_direct_gbpages_off); | |
69 | ||
70 | static int __init parse_direct_gbpages_on(char *arg) | |
71 | { | |
72 | direct_gbpages = 1; | |
73 | return 0; | |
74 | } | |
75 | early_param("gbpages", parse_direct_gbpages_on); | |
76 | ||
1da177e4 LT |
77 | /* |
78 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the | |
79 | * physical space so we can cache the place of the first one and move | |
80 | * around without checking the pgd every time. | |
81 | */ | |
82 | ||
83 | void show_mem(void) | |
84 | { | |
e92343cc AK |
85 | long i, total = 0, reserved = 0; |
86 | long shared = 0, cached = 0; | |
1da177e4 | 87 | struct page *page; |
14a62c34 | 88 | pg_data_t *pgdat; |
1da177e4 | 89 | |
e92343cc | 90 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 91 | show_free_areas(); |
ec936fc5 | 92 | for_each_online_pgdat(pgdat) { |
14a62c34 TG |
93 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
94 | /* | |
95 | * This loop can take a while with 256 GB and | |
96 | * 4k pages so defer the NMI watchdog: | |
97 | */ | |
98 | if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) | |
ae32b129 | 99 | touch_nmi_watchdog(); |
14a62c34 | 100 | |
12710a56 BP |
101 | if (!pfn_valid(pgdat->node_start_pfn + i)) |
102 | continue; | |
14a62c34 | 103 | |
1da177e4 LT |
104 | page = pfn_to_page(pgdat->node_start_pfn + i); |
105 | total++; | |
e92343cc AK |
106 | if (PageReserved(page)) |
107 | reserved++; | |
108 | else if (PageSwapCache(page)) | |
109 | cached++; | |
110 | else if (page_count(page)) | |
111 | shared += page_count(page) - 1; | |
14a62c34 | 112 | } |
1da177e4 | 113 | } |
14a62c34 TG |
114 | printk(KERN_INFO "%lu pages of RAM\n", total); |
115 | printk(KERN_INFO "%lu reserved pages\n", reserved); | |
116 | printk(KERN_INFO "%lu pages shared\n", shared); | |
117 | printk(KERN_INFO "%lu pages swap cached\n", cached); | |
1da177e4 LT |
118 | } |
119 | ||
1da177e4 LT |
120 | int after_bootmem; |
121 | ||
5f44a669 | 122 | static __init void *spp_getpage(void) |
14a62c34 | 123 | { |
1da177e4 | 124 | void *ptr; |
14a62c34 | 125 | |
1da177e4 | 126 | if (after_bootmem) |
14a62c34 | 127 | ptr = (void *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
128 | else |
129 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
14a62c34 TG |
130 | |
131 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { | |
132 | panic("set_pte_phys: cannot allocate page data %s\n", | |
133 | after_bootmem ? "after bootmem" : ""); | |
134 | } | |
1da177e4 | 135 | |
10f22dde | 136 | pr_debug("spp_getpage %p\n", ptr); |
14a62c34 | 137 | |
1da177e4 | 138 | return ptr; |
14a62c34 | 139 | } |
1da177e4 | 140 | |
14a62c34 TG |
141 | static __init void |
142 | set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot) | |
1da177e4 LT |
143 | { |
144 | pgd_t *pgd; | |
145 | pud_t *pud; | |
146 | pmd_t *pmd; | |
147 | pte_t *pte, new_pte; | |
148 | ||
10f22dde | 149 | pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys); |
1da177e4 LT |
150 | |
151 | pgd = pgd_offset_k(vaddr); | |
152 | if (pgd_none(*pgd)) { | |
10f22dde IM |
153 | printk(KERN_ERR |
154 | "PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
1da177e4 LT |
155 | return; |
156 | } | |
157 | pud = pud_offset(pgd, vaddr); | |
158 | if (pud_none(*pud)) { | |
14a62c34 | 159 | pmd = (pmd_t *) spp_getpage(); |
1da177e4 LT |
160 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER)); |
161 | if (pmd != pmd_offset(pud, 0)) { | |
10f22dde | 162 | printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", |
14a62c34 | 163 | pmd, pmd_offset(pud, 0)); |
1da177e4 LT |
164 | return; |
165 | } | |
166 | } | |
167 | pmd = pmd_offset(pud, vaddr); | |
168 | if (pmd_none(*pmd)) { | |
169 | pte = (pte_t *) spp_getpage(); | |
170 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER)); | |
171 | if (pte != pte_offset_kernel(pmd, 0)) { | |
10f22dde | 172 | printk(KERN_ERR "PAGETABLE BUG #02!\n"); |
1da177e4 LT |
173 | return; |
174 | } | |
175 | } | |
176 | new_pte = pfn_pte(phys >> PAGE_SHIFT, prot); | |
177 | ||
178 | pte = pte_offset_kernel(pmd, vaddr); | |
179 | if (!pte_none(*pte) && | |
180 | pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) | |
181 | pte_ERROR(*pte); | |
182 | set_pte(pte, new_pte); | |
183 | ||
184 | /* | |
185 | * It's enough to flush this one mapping. | |
186 | * (PGE mappings get flushed as well) | |
187 | */ | |
188 | __flush_tlb_one(vaddr); | |
189 | } | |
190 | ||
31eedd82 | 191 | /* |
88f3aec7 IM |
192 | * The head.S code sets up the kernel high mapping: |
193 | * | |
194 | * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) | |
31eedd82 TG |
195 | * |
196 | * phys_addr holds the negative offset to the kernel, which is added | |
197 | * to the compile time generated pmds. This results in invalid pmds up | |
198 | * to the point where we hit the physaddr 0 mapping. | |
199 | * | |
200 | * We limit the mappings to the region from _text to _end. _end is | |
201 | * rounded up to the 2MB boundary. This catches the invalid pmds as | |
202 | * well, as they are located before _text: | |
203 | */ | |
204 | void __init cleanup_highmap(void) | |
205 | { | |
206 | unsigned long vaddr = __START_KERNEL_map; | |
207 | unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1; | |
208 | pmd_t *pmd = level2_kernel_pgt; | |
209 | pmd_t *last_pmd = pmd + PTRS_PER_PMD; | |
210 | ||
211 | for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) { | |
212 | if (!pmd_present(*pmd)) | |
213 | continue; | |
214 | if (vaddr < (unsigned long) _text || vaddr > end) | |
215 | set_pmd(pmd, __pmd(0)); | |
216 | } | |
217 | } | |
218 | ||
1da177e4 | 219 | /* NOTE: this is meant to be run only at boot */ |
14a62c34 TG |
220 | void __init |
221 | __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot) | |
1da177e4 LT |
222 | { |
223 | unsigned long address = __fix_to_virt(idx); | |
224 | ||
225 | if (idx >= __end_of_fixed_addresses) { | |
10f22dde | 226 | printk(KERN_ERR "Invalid __set_fixmap\n"); |
1da177e4 LT |
227 | return; |
228 | } | |
229 | set_pte_phys(address, phys, prot); | |
230 | } | |
231 | ||
75175278 AK |
232 | static unsigned long __initdata table_start; |
233 | static unsigned long __meminitdata table_end; | |
1da177e4 | 234 | |
dafe41ee | 235 | static __meminit void *alloc_low_page(unsigned long *phys) |
14a62c34 | 236 | { |
dafe41ee | 237 | unsigned long pfn = table_end++; |
1da177e4 LT |
238 | void *adr; |
239 | ||
44df75e6 MT |
240 | if (after_bootmem) { |
241 | adr = (void *)get_zeroed_page(GFP_ATOMIC); | |
242 | *phys = __pa(adr); | |
14a62c34 | 243 | |
44df75e6 MT |
244 | return adr; |
245 | } | |
246 | ||
14a62c34 TG |
247 | if (pfn >= end_pfn) |
248 | panic("alloc_low_page: ran out of memory"); | |
dafe41ee VG |
249 | |
250 | adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); | |
44df75e6 | 251 | memset(adr, 0, PAGE_SIZE); |
dafe41ee VG |
252 | *phys = pfn * PAGE_SIZE; |
253 | return adr; | |
254 | } | |
1da177e4 | 255 | |
dafe41ee | 256 | static __meminit void unmap_low_page(void *adr) |
14a62c34 | 257 | { |
44df75e6 MT |
258 | if (after_bootmem) |
259 | return; | |
260 | ||
dafe41ee | 261 | early_iounmap(adr, PAGE_SIZE); |
14a62c34 | 262 | } |
1da177e4 | 263 | |
f2d3efed | 264 | /* Must run before zap_low_mappings */ |
a3142c8e | 265 | __meminit void *early_ioremap(unsigned long addr, unsigned long size) |
f2d3efed | 266 | { |
dafe41ee | 267 | pmd_t *pmd, *last_pmd; |
14a62c34 | 268 | unsigned long vaddr; |
dafe41ee VG |
269 | int i, pmds; |
270 | ||
271 | pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
272 | vaddr = __START_KERNEL_map; | |
273 | pmd = level2_kernel_pgt; | |
274 | last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1; | |
14a62c34 | 275 | |
dafe41ee VG |
276 | for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) { |
277 | for (i = 0; i < pmds; i++) { | |
278 | if (pmd_present(pmd[i])) | |
14a62c34 | 279 | goto continue_outer_loop; |
dafe41ee VG |
280 | } |
281 | vaddr += addr & ~PMD_MASK; | |
282 | addr &= PMD_MASK; | |
14a62c34 | 283 | |
dafe41ee | 284 | for (i = 0; i < pmds; i++, addr += PMD_SIZE) |
929fd589 | 285 | set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC)); |
1a2b4412 | 286 | __flush_tlb_all(); |
14a62c34 | 287 | |
dafe41ee | 288 | return (void *)vaddr; |
14a62c34 | 289 | continue_outer_loop: |
dafe41ee | 290 | ; |
f2d3efed | 291 | } |
10f22dde | 292 | printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size); |
14a62c34 | 293 | |
dafe41ee | 294 | return NULL; |
f2d3efed AK |
295 | } |
296 | ||
14a62c34 TG |
297 | /* |
298 | * To avoid virtual aliases later: | |
299 | */ | |
a3142c8e | 300 | __meminit void early_iounmap(void *addr, unsigned long size) |
f2d3efed | 301 | { |
dafe41ee VG |
302 | unsigned long vaddr; |
303 | pmd_t *pmd; | |
304 | int i, pmds; | |
305 | ||
306 | vaddr = (unsigned long)addr; | |
307 | pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
308 | pmd = level2_kernel_pgt + pmd_index(vaddr); | |
14a62c34 | 309 | |
dafe41ee VG |
310 | for (i = 0; i < pmds; i++) |
311 | pmd_clear(pmd + i); | |
14a62c34 | 312 | |
1a2b4412 | 313 | __flush_tlb_all(); |
f2d3efed AK |
314 | } |
315 | ||
44df75e6 | 316 | static void __meminit |
6ad91658 | 317 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end) |
44df75e6 | 318 | { |
6ad91658 | 319 | int i = pmd_index(address); |
44df75e6 | 320 | |
6ad91658 | 321 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
6ad91658 | 322 | pmd_t *pmd = pmd_page + pmd_index(address); |
44df75e6 | 323 | |
5f51e139 | 324 | if (address >= end) { |
14a62c34 | 325 | if (!after_bootmem) { |
5f51e139 JB |
326 | for (; i < PTRS_PER_PMD; i++, pmd++) |
327 | set_pmd(pmd, __pmd(0)); | |
14a62c34 | 328 | } |
44df75e6 MT |
329 | break; |
330 | } | |
6ad91658 KM |
331 | |
332 | if (pmd_val(*pmd)) | |
333 | continue; | |
334 | ||
d4f71f79 AK |
335 | set_pte((pte_t *)pmd, |
336 | pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
44df75e6 MT |
337 | } |
338 | } | |
339 | ||
340 | static void __meminit | |
341 | phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end) | |
342 | { | |
14a62c34 | 343 | pmd_t *pmd = pmd_offset(pud, 0); |
6ad91658 KM |
344 | spin_lock(&init_mm.page_table_lock); |
345 | phys_pmd_init(pmd, address, end); | |
346 | spin_unlock(&init_mm.page_table_lock); | |
347 | __flush_tlb_all(); | |
44df75e6 MT |
348 | } |
349 | ||
14a62c34 TG |
350 | static void __meminit |
351 | phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end) | |
352 | { | |
6ad91658 | 353 | int i = pud_index(addr); |
44df75e6 | 354 | |
14a62c34 | 355 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { |
6ad91658 KM |
356 | unsigned long pmd_phys; |
357 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 LT |
358 | pmd_t *pmd; |
359 | ||
6ad91658 | 360 | if (addr >= end) |
1da177e4 | 361 | break; |
1da177e4 | 362 | |
14a62c34 TG |
363 | if (!after_bootmem && |
364 | !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { | |
365 | set_pud(pud, __pud(0)); | |
1da177e4 | 366 | continue; |
14a62c34 | 367 | } |
1da177e4 | 368 | |
6ad91658 | 369 | if (pud_val(*pud)) { |
ef925766 AK |
370 | if (!pud_large(*pud)) |
371 | phys_pmd_update(pud, addr, end); | |
372 | continue; | |
373 | } | |
374 | ||
375 | if (direct_gbpages) { | |
376 | set_pte((pte_t *)pud, | |
377 | pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
6ad91658 KM |
378 | continue; |
379 | } | |
380 | ||
dafe41ee | 381 | pmd = alloc_low_page(&pmd_phys); |
14a62c34 | 382 | |
44df75e6 | 383 | spin_lock(&init_mm.page_table_lock); |
1da177e4 | 384 | set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE)); |
6ad91658 | 385 | phys_pmd_init(pmd, addr, end); |
44df75e6 | 386 | spin_unlock(&init_mm.page_table_lock); |
14a62c34 | 387 | |
dafe41ee | 388 | unmap_low_page(pmd); |
1da177e4 | 389 | } |
1a2b4412 | 390 | __flush_tlb_all(); |
14a62c34 | 391 | } |
1da177e4 LT |
392 | |
393 | static void __init find_early_table_space(unsigned long end) | |
394 | { | |
6c5acd16 | 395 | unsigned long puds, pmds, tables, start; |
1da177e4 LT |
396 | |
397 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; | |
ef925766 AK |
398 | tables = round_up(puds * sizeof(pud_t), PAGE_SIZE); |
399 | if (!direct_gbpages) { | |
400 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; | |
401 | tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE); | |
402 | } | |
1da177e4 | 403 | |
14a62c34 TG |
404 | /* |
405 | * RED-PEN putting page tables only on node 0 could | |
406 | * cause a hotspot and fill up ZONE_DMA. The page tables | |
407 | * need roughly 0.5KB per GB. | |
408 | */ | |
409 | start = 0x8000; | |
24a5da73 | 410 | table_start = find_e820_area(start, end, tables, PAGE_SIZE); |
1da177e4 LT |
411 | if (table_start == -1UL) |
412 | panic("Cannot find space for the kernel page tables"); | |
413 | ||
414 | table_start >>= PAGE_SHIFT; | |
415 | table_end = table_start; | |
44df75e6 MT |
416 | |
417 | early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n", | |
5f51e139 JB |
418 | end, table_start << PAGE_SHIFT, |
419 | (table_start << PAGE_SHIFT) + tables); | |
1da177e4 LT |
420 | } |
421 | ||
ef925766 AK |
422 | static void __init init_gbpages(void) |
423 | { | |
424 | if (direct_gbpages && cpu_has_gbpages) | |
425 | printk(KERN_INFO "Using GB pages for direct mapping\n"); | |
426 | else | |
427 | direct_gbpages = 0; | |
428 | } | |
429 | ||
c64df707 YL |
430 | #ifdef CONFIG_MEMTEST_BOOTPARAM |
431 | ||
432 | static void __init memtest(unsigned long start_phys, unsigned long size, | |
433 | unsigned pattern) | |
272b9cad YL |
434 | { |
435 | unsigned long i; | |
436 | unsigned long *start; | |
437 | unsigned long start_bad; | |
438 | unsigned long last_bad; | |
439 | unsigned long val; | |
440 | unsigned long start_phys_aligned; | |
441 | unsigned long count; | |
442 | unsigned long incr; | |
443 | ||
444 | switch (pattern) { | |
445 | case 0: | |
446 | val = 0UL; | |
447 | break; | |
448 | case 1: | |
449 | val = -1UL; | |
450 | break; | |
451 | case 2: | |
452 | val = 0x5555555555555555UL; | |
453 | break; | |
454 | case 3: | |
455 | val = 0xaaaaaaaaaaaaaaaaUL; | |
456 | break; | |
457 | default: | |
458 | return; | |
459 | } | |
460 | ||
461 | incr = sizeof(unsigned long); | |
462 | start_phys_aligned = ALIGN(start_phys, incr); | |
463 | count = (size - (start_phys_aligned - start_phys))/incr; | |
464 | start = __va(start_phys_aligned); | |
465 | start_bad = 0; | |
466 | last_bad = 0; | |
467 | ||
468 | for (i = 0; i < count; i++) | |
469 | start[i] = val; | |
470 | for (i = 0; i < count; i++, start++, start_phys_aligned += incr) { | |
471 | if (*start != val) { | |
472 | if (start_phys_aligned == last_bad + incr) { | |
473 | last_bad += incr; | |
474 | } else { | |
475 | if (start_bad) { | |
dcfe9465 | 476 | printk(KERN_CONT "\n %016lx bad mem addr %016lx - %016lx reserved", |
272b9cad YL |
477 | val, start_bad, last_bad + incr); |
478 | reserve_early(start_bad, last_bad - start_bad, "BAD RAM"); | |
479 | } | |
480 | start_bad = last_bad = start_phys_aligned; | |
481 | } | |
482 | } | |
483 | } | |
484 | if (start_bad) { | |
dcfe9465 | 485 | printk(KERN_CONT "\n %016lx bad mem addr %016lx - %016lx reserved", |
272b9cad YL |
486 | val, start_bad, last_bad + incr); |
487 | reserve_early(start_bad, last_bad - start_bad, "BAD RAM"); | |
488 | } | |
489 | ||
490 | } | |
491 | ||
c64df707 YL |
492 | static int memtest_pattern __initdata = CONFIG_MEMTEST_BOOTPARAM_VALUE; |
493 | ||
272b9cad YL |
494 | static int __init parse_memtest(char *arg) |
495 | { | |
496 | if (arg) | |
c64df707 | 497 | memtest_pattern = simple_strtoul(arg, NULL, 0); |
272b9cad YL |
498 | return 0; |
499 | } | |
500 | ||
501 | early_param("memtest", parse_memtest); | |
502 | ||
503 | static void __init early_memtest(unsigned long start, unsigned long end) | |
504 | { | |
505 | unsigned long t_start, t_size; | |
506 | unsigned pattern; | |
507 | ||
c64df707 YL |
508 | if (!memtest_pattern) |
509 | return; | |
510 | ||
511 | printk(KERN_INFO "early_memtest: pattern num %d", memtest_pattern); | |
272b9cad YL |
512 | for (pattern = 0; pattern < memtest_pattern; pattern++) { |
513 | t_start = start; | |
514 | t_size = 0; | |
515 | while (t_start < end) { | |
516 | t_start = find_e820_area_size(t_start, &t_size, 1); | |
517 | ||
518 | /* done ? */ | |
519 | if (t_start >= end) | |
520 | break; | |
521 | if (t_start + t_size > end) | |
522 | t_size = end - t_start; | |
523 | ||
524 | printk(KERN_CONT "\n %016lx - %016lx pattern %d", | |
525 | t_start, t_start + t_size, pattern); | |
526 | ||
527 | memtest(t_start, t_size, pattern); | |
528 | ||
529 | t_start += t_size; | |
530 | } | |
531 | } | |
c64df707 | 532 | printk(KERN_CONT "\n"); |
272b9cad | 533 | } |
c64df707 YL |
534 | #else |
535 | static void __init early_memtest(unsigned long start, unsigned long end) | |
536 | { | |
537 | } | |
538 | #endif | |
272b9cad | 539 | |
14a62c34 TG |
540 | /* |
541 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
542 | * This runs before bootmem is initialized and gets pages directly from | |
543 | * the physical memory. To access them they are temporarily mapped. | |
544 | */ | |
b6fd6ecb | 545 | void __init_refok init_memory_mapping(unsigned long start, unsigned long end) |
14a62c34 TG |
546 | { |
547 | unsigned long next; | |
272b9cad | 548 | unsigned long start_phys = start, end_phys = end; |
1da177e4 | 549 | |
272b9cad | 550 | printk(KERN_INFO "init_memory_mapping\n"); |
1da177e4 | 551 | |
14a62c34 | 552 | /* |
1da177e4 | 553 | * Find space for the kernel direct mapping tables. |
14a62c34 TG |
554 | * |
555 | * Later we should allocate these tables in the local node of the | |
556 | * memory mapped. Unfortunately this is done currently before the | |
557 | * nodes are discovered. | |
1da177e4 | 558 | */ |
ef925766 AK |
559 | if (!after_bootmem) { |
560 | init_gbpages(); | |
44df75e6 | 561 | find_early_table_space(end); |
ef925766 | 562 | } |
1da177e4 LT |
563 | |
564 | start = (unsigned long)__va(start); | |
565 | end = (unsigned long)__va(end); | |
566 | ||
567 | for (; start < end; start = next) { | |
44df75e6 | 568 | pgd_t *pgd = pgd_offset_k(start); |
14a62c34 | 569 | unsigned long pud_phys; |
44df75e6 MT |
570 | pud_t *pud; |
571 | ||
572 | if (after_bootmem) | |
d2ae5b5f | 573 | pud = pud_offset(pgd, start & PGDIR_MASK); |
44df75e6 | 574 | else |
dafe41ee | 575 | pud = alloc_low_page(&pud_phys); |
44df75e6 | 576 | |
1da177e4 | 577 | next = start + PGDIR_SIZE; |
14a62c34 TG |
578 | if (next > end) |
579 | next = end; | |
1da177e4 | 580 | phys_pud_init(pud, __pa(start), __pa(next)); |
44df75e6 MT |
581 | if (!after_bootmem) |
582 | set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys)); | |
dafe41ee | 583 | unmap_low_page(pud); |
14a62c34 | 584 | } |
1da177e4 | 585 | |
44df75e6 | 586 | if (!after_bootmem) |
f51c9452 | 587 | mmu_cr4_features = read_cr4(); |
1da177e4 | 588 | __flush_tlb_all(); |
75175278 | 589 | |
24a5da73 YL |
590 | if (!after_bootmem) |
591 | reserve_early(table_start << PAGE_SHIFT, | |
592 | table_end << PAGE_SHIFT, "PGTABLE"); | |
272b9cad YL |
593 | |
594 | if (!after_bootmem) | |
595 | early_memtest(start_phys, end_phys); | |
1da177e4 LT |
596 | } |
597 | ||
2b97690f | 598 | #ifndef CONFIG_NUMA |
1da177e4 LT |
599 | void __init paging_init(void) |
600 | { | |
6391af17 | 601 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
14a62c34 | 602 | |
6391af17 MG |
603 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
604 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
605 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
606 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
607 | ||
44df75e6 MT |
608 | memory_present(0, 0, end_pfn); |
609 | sparse_init(); | |
5cb248ab | 610 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
611 | } |
612 | #endif | |
613 | ||
44df75e6 MT |
614 | /* |
615 | * Memory hotplug specific functions | |
44df75e6 | 616 | */ |
44df75e6 MT |
617 | void online_page(struct page *page) |
618 | { | |
619 | ClearPageReserved(page); | |
7835e98b | 620 | init_page_count(page); |
44df75e6 MT |
621 | __free_page(page); |
622 | totalram_pages++; | |
623 | num_physpages++; | |
624 | } | |
625 | ||
bc02af93 | 626 | #ifdef CONFIG_MEMORY_HOTPLUG |
9d99aaa3 AK |
627 | /* |
628 | * Memory is added always to NORMAL zone. This means you will never get | |
629 | * additional DMA/DMA32 memory. | |
630 | */ | |
bc02af93 | 631 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 632 | { |
bc02af93 | 633 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 634 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
44df75e6 MT |
635 | unsigned long start_pfn = start >> PAGE_SHIFT; |
636 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
637 | int ret; | |
638 | ||
14a62c34 | 639 | init_memory_mapping(start, start + size-1); |
45e0b78b | 640 | |
44df75e6 | 641 | ret = __add_pages(zone, start_pfn, nr_pages); |
10f22dde | 642 | WARN_ON(1); |
44df75e6 | 643 | |
44df75e6 | 644 | return ret; |
44df75e6 | 645 | } |
bc02af93 | 646 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 | 647 | |
8243229f | 648 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
4942e998 KM |
649 | int memory_add_physaddr_to_nid(u64 start) |
650 | { | |
651 | return 0; | |
652 | } | |
8c2676a5 | 653 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
4942e998 KM |
654 | #endif |
655 | ||
45e0b78b KM |
656 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
657 | ||
14a62c34 TG |
658 | static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, |
659 | kcore_modules, kcore_vsyscall; | |
1da177e4 LT |
660 | |
661 | void __init mem_init(void) | |
662 | { | |
0a43e4bf | 663 | long codesize, reservedpages, datasize, initsize; |
1da177e4 | 664 | |
0dc243ae | 665 | pci_iommu_alloc(); |
1da177e4 | 666 | |
48ddb154 | 667 | /* clear_bss() already clear the empty_zero_page */ |
1da177e4 LT |
668 | |
669 | reservedpages = 0; | |
670 | ||
671 | /* this will put all low memory onto the freelists */ | |
2b97690f | 672 | #ifdef CONFIG_NUMA |
0a43e4bf | 673 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 674 | #else |
0a43e4bf | 675 | totalram_pages = free_all_bootmem(); |
1da177e4 | 676 | #endif |
5cb248ab MG |
677 | reservedpages = end_pfn - totalram_pages - |
678 | absent_pages_in_range(0, end_pfn); | |
1da177e4 LT |
679 | after_bootmem = 1; |
680 | ||
681 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
682 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
683 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
684 | ||
685 | /* Register memory areas for /proc/kcore */ | |
14a62c34 TG |
686 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
687 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
1da177e4 LT |
688 | VMALLOC_END-VMALLOC_START); |
689 | kclist_add(&kcore_kernel, &_stext, _end - _stext); | |
690 | kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); | |
14a62c34 | 691 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
1da177e4 LT |
692 | VSYSCALL_END - VSYSCALL_START); |
693 | ||
10f22dde | 694 | printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " |
14a62c34 | 695 | "%ldk reserved, %ldk data, %ldk init)\n", |
1da177e4 LT |
696 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
697 | end_pfn << (PAGE_SHIFT-10), | |
698 | codesize >> 10, | |
699 | reservedpages << (PAGE_SHIFT-10), | |
700 | datasize >> 10, | |
701 | initsize >> 10); | |
76ebd054 TG |
702 | |
703 | cpa_init(); | |
1da177e4 LT |
704 | } |
705 | ||
d167a518 | 706 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
1da177e4 | 707 | { |
bfc734b2 | 708 | unsigned long addr = begin; |
1da177e4 | 709 | |
bfc734b2 | 710 | if (addr >= end) |
d167a518 GH |
711 | return; |
712 | ||
ee01f112 IM |
713 | /* |
714 | * If debugging page accesses then do not free this memory but | |
715 | * mark them not present - any buggy init-section access will | |
716 | * create a kernel page fault: | |
717 | */ | |
718 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
719 | printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", | |
720 | begin, PAGE_ALIGN(end)); | |
721 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
722 | #else | |
6fb14755 | 723 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
14a62c34 | 724 | |
bfc734b2 | 725 | for (; addr < end; addr += PAGE_SIZE) { |
e3ebadd9 LT |
726 | ClearPageReserved(virt_to_page(addr)); |
727 | init_page_count(virt_to_page(addr)); | |
728 | memset((void *)(addr & ~(PAGE_SIZE-1)), | |
729 | POISON_FREE_INITMEM, PAGE_SIZE); | |
e3ebadd9 | 730 | free_page(addr); |
1da177e4 LT |
731 | totalram_pages++; |
732 | } | |
ee01f112 | 733 | #endif |
d167a518 GH |
734 | } |
735 | ||
736 | void free_initmem(void) | |
737 | { | |
d167a518 | 738 | free_init_pages("unused kernel memory", |
e3ebadd9 LT |
739 | (unsigned long)(&__init_begin), |
740 | (unsigned long)(&__init_end)); | |
1da177e4 LT |
741 | } |
742 | ||
67df197b | 743 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
744 | const int rodata_test_data = 0xC3; |
745 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
67df197b | 746 | |
67df197b AV |
747 | void mark_rodata_ro(void) |
748 | { | |
4e4eee0e | 749 | unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata); |
67df197b | 750 | |
6fb14755 | 751 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 752 | (end - start) >> 10); |
984bb80d AV |
753 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
754 | ||
755 | /* | |
756 | * The rodata section (but not the kernel text!) should also be | |
757 | * not-executable. | |
758 | */ | |
759 | start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; | |
760 | set_memory_nx(start, (end - start) >> PAGE_SHIFT); | |
67df197b | 761 | |
1a487252 AV |
762 | rodata_test(); |
763 | ||
0c42f392 | 764 | #ifdef CONFIG_CPA_DEBUG |
10f22dde | 765 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); |
6d238cc4 | 766 | set_memory_rw(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 767 | |
10f22dde | 768 | printk(KERN_INFO "Testing CPA: again\n"); |
6d238cc4 | 769 | set_memory_ro(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 770 | #endif |
67df197b | 771 | } |
4e4eee0e | 772 | |
67df197b AV |
773 | #endif |
774 | ||
1da177e4 LT |
775 | #ifdef CONFIG_BLK_DEV_INITRD |
776 | void free_initrd_mem(unsigned long start, unsigned long end) | |
777 | { | |
e3ebadd9 | 778 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
779 | } |
780 | #endif | |
781 | ||
14a62c34 TG |
782 | void __init reserve_bootmem_generic(unsigned long phys, unsigned len) |
783 | { | |
2b97690f | 784 | #ifdef CONFIG_NUMA |
1da177e4 | 785 | int nid = phys_to_nid(phys); |
5e58a02a AK |
786 | #endif |
787 | unsigned long pfn = phys >> PAGE_SHIFT; | |
14a62c34 | 788 | |
5e58a02a | 789 | if (pfn >= end_pfn) { |
14a62c34 TG |
790 | /* |
791 | * This can happen with kdump kernels when accessing | |
792 | * firmware tables: | |
793 | */ | |
5e58a02a AK |
794 | if (pfn < end_pfn_map) |
795 | return; | |
14a62c34 | 796 | |
5e58a02a AK |
797 | printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n", |
798 | phys, len); | |
799 | return; | |
800 | } | |
801 | ||
802 | /* Should check here against the e820 map to avoid double free */ | |
803 | #ifdef CONFIG_NUMA | |
72a7fe39 | 804 | reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT); |
14a62c34 | 805 | #else |
72a7fe39 | 806 | reserve_bootmem(phys, len, BOOTMEM_DEFAULT); |
1da177e4 | 807 | #endif |
0e0b864e | 808 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
e18c6874 | 809 | dma_reserve += len / PAGE_SIZE; |
0e0b864e MG |
810 | set_dma_reserve(dma_reserve); |
811 | } | |
1da177e4 LT |
812 | } |
813 | ||
14a62c34 TG |
814 | int kern_addr_valid(unsigned long addr) |
815 | { | |
1da177e4 | 816 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
14a62c34 TG |
817 | pgd_t *pgd; |
818 | pud_t *pud; | |
819 | pmd_t *pmd; | |
820 | pte_t *pte; | |
1da177e4 LT |
821 | |
822 | if (above != 0 && above != -1UL) | |
14a62c34 TG |
823 | return 0; |
824 | ||
1da177e4 LT |
825 | pgd = pgd_offset_k(addr); |
826 | if (pgd_none(*pgd)) | |
827 | return 0; | |
828 | ||
829 | pud = pud_offset(pgd, addr); | |
830 | if (pud_none(*pud)) | |
14a62c34 | 831 | return 0; |
1da177e4 LT |
832 | |
833 | pmd = pmd_offset(pud, addr); | |
834 | if (pmd_none(*pmd)) | |
835 | return 0; | |
14a62c34 | 836 | |
1da177e4 LT |
837 | if (pmd_large(*pmd)) |
838 | return pfn_valid(pmd_pfn(*pmd)); | |
839 | ||
840 | pte = pte_offset_kernel(pmd, addr); | |
841 | if (pte_none(*pte)) | |
842 | return 0; | |
14a62c34 | 843 | |
1da177e4 LT |
844 | return pfn_valid(pte_pfn(*pte)); |
845 | } | |
846 | ||
14a62c34 TG |
847 | /* |
848 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
849 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
850 | * not need special handling anymore: | |
851 | */ | |
1da177e4 | 852 | static struct vm_area_struct gate_vma = { |
14a62c34 TG |
853 | .vm_start = VSYSCALL_START, |
854 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), | |
855 | .vm_page_prot = PAGE_READONLY_EXEC, | |
856 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
857 | }; |
858 | ||
1da177e4 LT |
859 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
860 | { | |
861 | #ifdef CONFIG_IA32_EMULATION | |
1e014410 AK |
862 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
863 | return NULL; | |
1da177e4 LT |
864 | #endif |
865 | return &gate_vma; | |
866 | } | |
867 | ||
868 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
869 | { | |
870 | struct vm_area_struct *vma = get_gate_vma(task); | |
14a62c34 | 871 | |
1e014410 AK |
872 | if (!vma) |
873 | return 0; | |
14a62c34 | 874 | |
1da177e4 LT |
875 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
876 | } | |
877 | ||
14a62c34 TG |
878 | /* |
879 | * Use this when you have no reliable task/vma, typically from interrupt | |
880 | * context. It is less reliable than using the task's vma and may give | |
881 | * false positives: | |
1da177e4 LT |
882 | */ |
883 | int in_gate_area_no_task(unsigned long addr) | |
884 | { | |
1e014410 | 885 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 886 | } |
2e1c49db | 887 | |
2aae950b AK |
888 | const char *arch_vma_name(struct vm_area_struct *vma) |
889 | { | |
890 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
891 | return "[vdso]"; | |
892 | if (vma == &gate_vma) | |
893 | return "[vsyscall]"; | |
894 | return NULL; | |
895 | } | |
0889eba5 CL |
896 | |
897 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
898 | /* | |
899 | * Initialise the sparsemem vmemmap using huge-pages at the PMD level. | |
900 | */ | |
14a62c34 TG |
901 | int __meminit |
902 | vmemmap_populate(struct page *start_page, unsigned long size, int node) | |
0889eba5 CL |
903 | { |
904 | unsigned long addr = (unsigned long)start_page; | |
905 | unsigned long end = (unsigned long)(start_page + size); | |
906 | unsigned long next; | |
907 | pgd_t *pgd; | |
908 | pud_t *pud; | |
909 | pmd_t *pmd; | |
910 | ||
911 | for (; addr < end; addr = next) { | |
912 | next = pmd_addr_end(addr, end); | |
913 | ||
914 | pgd = vmemmap_pgd_populate(addr, node); | |
915 | if (!pgd) | |
916 | return -ENOMEM; | |
14a62c34 | 917 | |
0889eba5 CL |
918 | pud = vmemmap_pud_populate(pgd, addr, node); |
919 | if (!pud) | |
920 | return -ENOMEM; | |
921 | ||
922 | pmd = pmd_offset(pud, addr); | |
923 | if (pmd_none(*pmd)) { | |
924 | pte_t entry; | |
14a62c34 TG |
925 | void *p; |
926 | ||
927 | p = vmemmap_alloc_block(PMD_SIZE, node); | |
0889eba5 CL |
928 | if (!p) |
929 | return -ENOMEM; | |
930 | ||
14a62c34 TG |
931 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, |
932 | PAGE_KERNEL_LARGE); | |
0889eba5 CL |
933 | set_pmd(pmd, __pmd(pte_val(entry))); |
934 | ||
935 | printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n", | |
936 | addr, addr + PMD_SIZE - 1, p, node); | |
14a62c34 | 937 | } else { |
0889eba5 | 938 | vmemmap_verify((pte_t *)pmd, node, addr, next); |
14a62c34 | 939 | } |
0889eba5 | 940 | } |
0889eba5 CL |
941 | return 0; |
942 | } | |
943 | #endif |