]>
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 | * PFN of last memory page. | |
53 | */ | |
54 | unsigned long end_pfn; | |
55 | ||
56 | /* | |
57 | * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries. | |
58 | * The direct mapping extends to max_pfn_mapped, so that we can directly access | |
59 | * apertures, ACPI and other tables without having to play with fixmaps. | |
60 | */ | |
61 | unsigned long max_pfn_mapped; | |
62 | ||
e18c6874 AK |
63 | static unsigned long dma_reserve __initdata; |
64 | ||
1da177e4 LT |
65 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
66 | ||
00d1c5e0 IM |
67 | int direct_gbpages __meminitdata |
68 | #ifdef CONFIG_DIRECT_GBPAGES | |
69 | = 1 | |
70 | #endif | |
71 | ; | |
72 | ||
73 | static int __init parse_direct_gbpages_off(char *arg) | |
74 | { | |
75 | direct_gbpages = 0; | |
76 | return 0; | |
77 | } | |
78 | early_param("nogbpages", parse_direct_gbpages_off); | |
79 | ||
80 | static int __init parse_direct_gbpages_on(char *arg) | |
81 | { | |
82 | direct_gbpages = 1; | |
83 | return 0; | |
84 | } | |
85 | early_param("gbpages", parse_direct_gbpages_on); | |
86 | ||
1da177e4 LT |
87 | /* |
88 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the | |
89 | * physical space so we can cache the place of the first one and move | |
90 | * around without checking the pgd every time. | |
91 | */ | |
92 | ||
93 | void show_mem(void) | |
94 | { | |
e92343cc AK |
95 | long i, total = 0, reserved = 0; |
96 | long shared = 0, cached = 0; | |
1da177e4 | 97 | struct page *page; |
14a62c34 | 98 | pg_data_t *pgdat; |
1da177e4 | 99 | |
e92343cc | 100 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 101 | show_free_areas(); |
ec936fc5 | 102 | for_each_online_pgdat(pgdat) { |
14a62c34 TG |
103 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
104 | /* | |
105 | * This loop can take a while with 256 GB and | |
106 | * 4k pages so defer the NMI watchdog: | |
107 | */ | |
108 | if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) | |
ae32b129 | 109 | touch_nmi_watchdog(); |
14a62c34 | 110 | |
12710a56 BP |
111 | if (!pfn_valid(pgdat->node_start_pfn + i)) |
112 | continue; | |
14a62c34 | 113 | |
1da177e4 LT |
114 | page = pfn_to_page(pgdat->node_start_pfn + i); |
115 | total++; | |
e92343cc AK |
116 | if (PageReserved(page)) |
117 | reserved++; | |
118 | else if (PageSwapCache(page)) | |
119 | cached++; | |
120 | else if (page_count(page)) | |
121 | shared += page_count(page) - 1; | |
14a62c34 | 122 | } |
1da177e4 | 123 | } |
14a62c34 TG |
124 | printk(KERN_INFO "%lu pages of RAM\n", total); |
125 | printk(KERN_INFO "%lu reserved pages\n", reserved); | |
126 | printk(KERN_INFO "%lu pages shared\n", shared); | |
127 | printk(KERN_INFO "%lu pages swap cached\n", cached); | |
1da177e4 LT |
128 | } |
129 | ||
1da177e4 LT |
130 | int after_bootmem; |
131 | ||
5f44a669 | 132 | static __init void *spp_getpage(void) |
14a62c34 | 133 | { |
1da177e4 | 134 | void *ptr; |
14a62c34 | 135 | |
1da177e4 | 136 | if (after_bootmem) |
14a62c34 | 137 | ptr = (void *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
138 | else |
139 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
14a62c34 TG |
140 | |
141 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { | |
142 | panic("set_pte_phys: cannot allocate page data %s\n", | |
143 | after_bootmem ? "after bootmem" : ""); | |
144 | } | |
1da177e4 | 145 | |
10f22dde | 146 | pr_debug("spp_getpage %p\n", ptr); |
14a62c34 | 147 | |
1da177e4 | 148 | return ptr; |
14a62c34 | 149 | } |
1da177e4 | 150 | |
d494a961 JF |
151 | void |
152 | set_pte_vaddr(unsigned long vaddr, pte_t new_pte) | |
1da177e4 LT |
153 | { |
154 | pgd_t *pgd; | |
155 | pud_t *pud; | |
156 | pmd_t *pmd; | |
d494a961 | 157 | pte_t *pte; |
1da177e4 | 158 | |
d494a961 | 159 | pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(new_pte)); |
1da177e4 LT |
160 | |
161 | pgd = pgd_offset_k(vaddr); | |
162 | if (pgd_none(*pgd)) { | |
10f22dde IM |
163 | printk(KERN_ERR |
164 | "PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
1da177e4 LT |
165 | return; |
166 | } | |
167 | pud = pud_offset(pgd, vaddr); | |
168 | if (pud_none(*pud)) { | |
14a62c34 | 169 | pmd = (pmd_t *) spp_getpage(); |
1da177e4 LT |
170 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER)); |
171 | if (pmd != pmd_offset(pud, 0)) { | |
10f22dde | 172 | printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", |
14a62c34 | 173 | pmd, pmd_offset(pud, 0)); |
1da177e4 LT |
174 | return; |
175 | } | |
176 | } | |
177 | pmd = pmd_offset(pud, vaddr); | |
178 | if (pmd_none(*pmd)) { | |
179 | pte = (pte_t *) spp_getpage(); | |
180 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER)); | |
181 | if (pte != pte_offset_kernel(pmd, 0)) { | |
10f22dde | 182 | printk(KERN_ERR "PAGETABLE BUG #02!\n"); |
1da177e4 LT |
183 | return; |
184 | } | |
185 | } | |
1da177e4 LT |
186 | |
187 | pte = pte_offset_kernel(pmd, vaddr); | |
70c9f590 | 188 | if (!pte_none(*pte) && pte_val(new_pte) && |
1da177e4 LT |
189 | pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) |
190 | pte_ERROR(*pte); | |
191 | set_pte(pte, new_pte); | |
192 | ||
193 | /* | |
194 | * It's enough to flush this one mapping. | |
195 | * (PGE mappings get flushed as well) | |
196 | */ | |
197 | __flush_tlb_one(vaddr); | |
198 | } | |
199 | ||
31eedd82 | 200 | /* |
88f3aec7 IM |
201 | * The head.S code sets up the kernel high mapping: |
202 | * | |
203 | * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) | |
31eedd82 TG |
204 | * |
205 | * phys_addr holds the negative offset to the kernel, which is added | |
206 | * to the compile time generated pmds. This results in invalid pmds up | |
207 | * to the point where we hit the physaddr 0 mapping. | |
208 | * | |
209 | * We limit the mappings to the region from _text to _end. _end is | |
210 | * rounded up to the 2MB boundary. This catches the invalid pmds as | |
211 | * well, as they are located before _text: | |
212 | */ | |
213 | void __init cleanup_highmap(void) | |
214 | { | |
215 | unsigned long vaddr = __START_KERNEL_map; | |
216 | unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1; | |
217 | pmd_t *pmd = level2_kernel_pgt; | |
218 | pmd_t *last_pmd = pmd + PTRS_PER_PMD; | |
219 | ||
220 | for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) { | |
2884f110 | 221 | if (pmd_none(*pmd)) |
31eedd82 TG |
222 | continue; |
223 | if (vaddr < (unsigned long) _text || vaddr > end) | |
224 | set_pmd(pmd, __pmd(0)); | |
225 | } | |
226 | } | |
227 | ||
75175278 AK |
228 | static unsigned long __initdata table_start; |
229 | static unsigned long __meminitdata table_end; | |
1da177e4 | 230 | |
dafe41ee | 231 | static __meminit void *alloc_low_page(unsigned long *phys) |
14a62c34 | 232 | { |
dafe41ee | 233 | unsigned long pfn = table_end++; |
1da177e4 LT |
234 | void *adr; |
235 | ||
44df75e6 MT |
236 | if (after_bootmem) { |
237 | adr = (void *)get_zeroed_page(GFP_ATOMIC); | |
238 | *phys = __pa(adr); | |
14a62c34 | 239 | |
44df75e6 MT |
240 | return adr; |
241 | } | |
242 | ||
14a62c34 TG |
243 | if (pfn >= end_pfn) |
244 | panic("alloc_low_page: ran out of memory"); | |
dafe41ee VG |
245 | |
246 | adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); | |
44df75e6 | 247 | memset(adr, 0, PAGE_SIZE); |
dafe41ee VG |
248 | *phys = pfn * PAGE_SIZE; |
249 | return adr; | |
250 | } | |
1da177e4 | 251 | |
dafe41ee | 252 | static __meminit void unmap_low_page(void *adr) |
14a62c34 | 253 | { |
44df75e6 MT |
254 | if (after_bootmem) |
255 | return; | |
256 | ||
dafe41ee | 257 | early_iounmap(adr, PAGE_SIZE); |
14a62c34 | 258 | } |
1da177e4 | 259 | |
f2d3efed | 260 | /* Must run before zap_low_mappings */ |
a3142c8e | 261 | __meminit void *early_ioremap(unsigned long addr, unsigned long size) |
f2d3efed | 262 | { |
dafe41ee | 263 | pmd_t *pmd, *last_pmd; |
14a62c34 | 264 | unsigned long vaddr; |
dafe41ee VG |
265 | int i, pmds; |
266 | ||
267 | pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
268 | vaddr = __START_KERNEL_map; | |
269 | pmd = level2_kernel_pgt; | |
270 | last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1; | |
14a62c34 | 271 | |
dafe41ee VG |
272 | for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) { |
273 | for (i = 0; i < pmds; i++) { | |
274 | if (pmd_present(pmd[i])) | |
14a62c34 | 275 | goto continue_outer_loop; |
dafe41ee VG |
276 | } |
277 | vaddr += addr & ~PMD_MASK; | |
278 | addr &= PMD_MASK; | |
14a62c34 | 279 | |
dafe41ee | 280 | for (i = 0; i < pmds; i++, addr += PMD_SIZE) |
929fd589 | 281 | set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC)); |
1a2b4412 | 282 | __flush_tlb_all(); |
14a62c34 | 283 | |
dafe41ee | 284 | return (void *)vaddr; |
14a62c34 | 285 | continue_outer_loop: |
dafe41ee | 286 | ; |
f2d3efed | 287 | } |
10f22dde | 288 | printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size); |
14a62c34 | 289 | |
dafe41ee | 290 | return NULL; |
f2d3efed AK |
291 | } |
292 | ||
14a62c34 TG |
293 | /* |
294 | * To avoid virtual aliases later: | |
295 | */ | |
a3142c8e | 296 | __meminit void early_iounmap(void *addr, unsigned long size) |
f2d3efed | 297 | { |
dafe41ee VG |
298 | unsigned long vaddr; |
299 | pmd_t *pmd; | |
300 | int i, pmds; | |
301 | ||
302 | vaddr = (unsigned long)addr; | |
303 | pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; | |
304 | pmd = level2_kernel_pgt + pmd_index(vaddr); | |
14a62c34 | 305 | |
dafe41ee VG |
306 | for (i = 0; i < pmds; i++) |
307 | pmd_clear(pmd + i); | |
14a62c34 | 308 | |
1a2b4412 | 309 | __flush_tlb_all(); |
f2d3efed AK |
310 | } |
311 | ||
cc615032 | 312 | static unsigned long __meminit |
6ad91658 | 313 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end) |
44df75e6 | 314 | { |
ce0c0e50 AK |
315 | unsigned long pages = 0; |
316 | ||
6ad91658 | 317 | int i = pmd_index(address); |
44df75e6 | 318 | |
6ad91658 | 319 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
6ad91658 | 320 | pmd_t *pmd = pmd_page + pmd_index(address); |
44df75e6 | 321 | |
5f51e139 | 322 | if (address >= end) { |
14a62c34 | 323 | if (!after_bootmem) { |
5f51e139 JB |
324 | for (; i < PTRS_PER_PMD; i++, pmd++) |
325 | set_pmd(pmd, __pmd(0)); | |
14a62c34 | 326 | } |
44df75e6 MT |
327 | break; |
328 | } | |
6ad91658 KM |
329 | |
330 | if (pmd_val(*pmd)) | |
331 | continue; | |
332 | ||
ce0c0e50 | 333 | pages++; |
d4f71f79 AK |
334 | set_pte((pte_t *)pmd, |
335 | pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
44df75e6 | 336 | } |
ce0c0e50 | 337 | update_page_count(PG_LEVEL_2M, pages); |
cc615032 | 338 | return address; |
44df75e6 MT |
339 | } |
340 | ||
cc615032 | 341 | static unsigned long __meminit |
44df75e6 MT |
342 | phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end) |
343 | { | |
14a62c34 | 344 | pmd_t *pmd = pmd_offset(pud, 0); |
cc615032 AK |
345 | unsigned long last_map_addr; |
346 | ||
6ad91658 | 347 | spin_lock(&init_mm.page_table_lock); |
cc615032 | 348 | last_map_addr = phys_pmd_init(pmd, address, end); |
6ad91658 KM |
349 | spin_unlock(&init_mm.page_table_lock); |
350 | __flush_tlb_all(); | |
cc615032 | 351 | return last_map_addr; |
44df75e6 MT |
352 | } |
353 | ||
cc615032 | 354 | static unsigned long __meminit |
14a62c34 TG |
355 | phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end) |
356 | { | |
ce0c0e50 | 357 | unsigned long pages = 0; |
cc615032 | 358 | unsigned long last_map_addr = end; |
6ad91658 | 359 | int i = pud_index(addr); |
44df75e6 | 360 | |
14a62c34 | 361 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { |
6ad91658 KM |
362 | unsigned long pmd_phys; |
363 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 LT |
364 | pmd_t *pmd; |
365 | ||
6ad91658 | 366 | if (addr >= end) |
1da177e4 | 367 | break; |
1da177e4 | 368 | |
14a62c34 TG |
369 | if (!after_bootmem && |
370 | !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { | |
371 | set_pud(pud, __pud(0)); | |
1da177e4 | 372 | continue; |
14a62c34 | 373 | } |
1da177e4 | 374 | |
6ad91658 | 375 | if (pud_val(*pud)) { |
ef925766 | 376 | if (!pud_large(*pud)) |
cc615032 | 377 | last_map_addr = phys_pmd_update(pud, addr, end); |
ef925766 AK |
378 | continue; |
379 | } | |
380 | ||
381 | if (direct_gbpages) { | |
ce0c0e50 | 382 | pages++; |
ef925766 AK |
383 | set_pte((pte_t *)pud, |
384 | pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
cc615032 | 385 | last_map_addr = (addr & PUD_MASK) + PUD_SIZE; |
6ad91658 KM |
386 | continue; |
387 | } | |
388 | ||
dafe41ee | 389 | pmd = alloc_low_page(&pmd_phys); |
14a62c34 | 390 | |
44df75e6 | 391 | spin_lock(&init_mm.page_table_lock); |
1da177e4 | 392 | set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE)); |
cc615032 | 393 | last_map_addr = phys_pmd_init(pmd, addr, end); |
44df75e6 | 394 | spin_unlock(&init_mm.page_table_lock); |
14a62c34 | 395 | |
dafe41ee | 396 | unmap_low_page(pmd); |
1da177e4 | 397 | } |
1a2b4412 | 398 | __flush_tlb_all(); |
ce0c0e50 | 399 | update_page_count(PG_LEVEL_1G, pages); |
cc615032 AK |
400 | |
401 | return last_map_addr >> PAGE_SHIFT; | |
14a62c34 | 402 | } |
1da177e4 LT |
403 | |
404 | static void __init find_early_table_space(unsigned long end) | |
405 | { | |
6c5acd16 | 406 | unsigned long puds, pmds, tables, start; |
1da177e4 LT |
407 | |
408 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; | |
ef925766 AK |
409 | tables = round_up(puds * sizeof(pud_t), PAGE_SIZE); |
410 | if (!direct_gbpages) { | |
411 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; | |
412 | tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE); | |
413 | } | |
1da177e4 | 414 | |
14a62c34 TG |
415 | /* |
416 | * RED-PEN putting page tables only on node 0 could | |
417 | * cause a hotspot and fill up ZONE_DMA. The page tables | |
418 | * need roughly 0.5KB per GB. | |
419 | */ | |
420 | start = 0x8000; | |
24a5da73 | 421 | table_start = find_e820_area(start, end, tables, PAGE_SIZE); |
1da177e4 LT |
422 | if (table_start == -1UL) |
423 | panic("Cannot find space for the kernel page tables"); | |
424 | ||
425 | table_start >>= PAGE_SHIFT; | |
426 | table_end = table_start; | |
44df75e6 MT |
427 | |
428 | early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n", | |
5f51e139 JB |
429 | end, table_start << PAGE_SHIFT, |
430 | (table_start << PAGE_SHIFT) + tables); | |
1da177e4 LT |
431 | } |
432 | ||
ef925766 AK |
433 | static void __init init_gbpages(void) |
434 | { | |
435 | if (direct_gbpages && cpu_has_gbpages) | |
436 | printk(KERN_INFO "Using GB pages for direct mapping\n"); | |
437 | else | |
438 | direct_gbpages = 0; | |
439 | } | |
440 | ||
03273184 | 441 | #ifdef CONFIG_MEMTEST |
c64df707 YL |
442 | |
443 | static void __init memtest(unsigned long start_phys, unsigned long size, | |
444 | unsigned pattern) | |
272b9cad YL |
445 | { |
446 | unsigned long i; | |
447 | unsigned long *start; | |
448 | unsigned long start_bad; | |
449 | unsigned long last_bad; | |
450 | unsigned long val; | |
451 | unsigned long start_phys_aligned; | |
452 | unsigned long count; | |
453 | unsigned long incr; | |
454 | ||
455 | switch (pattern) { | |
456 | case 0: | |
457 | val = 0UL; | |
458 | break; | |
459 | case 1: | |
460 | val = -1UL; | |
461 | break; | |
462 | case 2: | |
463 | val = 0x5555555555555555UL; | |
464 | break; | |
465 | case 3: | |
466 | val = 0xaaaaaaaaaaaaaaaaUL; | |
467 | break; | |
468 | default: | |
469 | return; | |
470 | } | |
471 | ||
472 | incr = sizeof(unsigned long); | |
473 | start_phys_aligned = ALIGN(start_phys, incr); | |
474 | count = (size - (start_phys_aligned - start_phys))/incr; | |
475 | start = __va(start_phys_aligned); | |
476 | start_bad = 0; | |
477 | last_bad = 0; | |
478 | ||
479 | for (i = 0; i < count; i++) | |
480 | start[i] = val; | |
481 | for (i = 0; i < count; i++, start++, start_phys_aligned += incr) { | |
482 | if (*start != val) { | |
483 | if (start_phys_aligned == last_bad + incr) { | |
484 | last_bad += incr; | |
485 | } else { | |
486 | if (start_bad) { | |
dcfe9465 | 487 | printk(KERN_CONT "\n %016lx bad mem addr %016lx - %016lx reserved", |
272b9cad YL |
488 | val, start_bad, last_bad + incr); |
489 | reserve_early(start_bad, last_bad - start_bad, "BAD RAM"); | |
490 | } | |
491 | start_bad = last_bad = start_phys_aligned; | |
492 | } | |
493 | } | |
494 | } | |
495 | if (start_bad) { | |
dcfe9465 | 496 | printk(KERN_CONT "\n %016lx bad mem addr %016lx - %016lx reserved", |
272b9cad YL |
497 | val, start_bad, last_bad + incr); |
498 | reserve_early(start_bad, last_bad - start_bad, "BAD RAM"); | |
499 | } | |
500 | ||
501 | } | |
502 | ||
03273184 YL |
503 | /* default is disabled */ |
504 | static int memtest_pattern __initdata; | |
c64df707 | 505 | |
272b9cad YL |
506 | static int __init parse_memtest(char *arg) |
507 | { | |
508 | if (arg) | |
c64df707 | 509 | memtest_pattern = simple_strtoul(arg, NULL, 0); |
272b9cad YL |
510 | return 0; |
511 | } | |
512 | ||
513 | early_param("memtest", parse_memtest); | |
514 | ||
515 | static void __init early_memtest(unsigned long start, unsigned long end) | |
516 | { | |
27df66a4 | 517 | u64 t_start, t_size; |
272b9cad YL |
518 | unsigned pattern; |
519 | ||
c64df707 YL |
520 | if (!memtest_pattern) |
521 | return; | |
522 | ||
523 | printk(KERN_INFO "early_memtest: pattern num %d", memtest_pattern); | |
272b9cad YL |
524 | for (pattern = 0; pattern < memtest_pattern; pattern++) { |
525 | t_start = start; | |
526 | t_size = 0; | |
527 | while (t_start < end) { | |
528 | t_start = find_e820_area_size(t_start, &t_size, 1); | |
529 | ||
530 | /* done ? */ | |
531 | if (t_start >= end) | |
532 | break; | |
533 | if (t_start + t_size > end) | |
534 | t_size = end - t_start; | |
535 | ||
27df66a4 AM |
536 | printk(KERN_CONT "\n %016llx - %016llx pattern %d", |
537 | (unsigned long long)t_start, | |
538 | (unsigned long long)t_start + t_size, pattern); | |
272b9cad YL |
539 | |
540 | memtest(t_start, t_size, pattern); | |
541 | ||
542 | t_start += t_size; | |
543 | } | |
544 | } | |
c64df707 | 545 | printk(KERN_CONT "\n"); |
272b9cad | 546 | } |
c64df707 YL |
547 | #else |
548 | static void __init early_memtest(unsigned long start, unsigned long end) | |
549 | { | |
550 | } | |
551 | #endif | |
272b9cad | 552 | |
14a62c34 TG |
553 | /* |
554 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
555 | * This runs before bootmem is initialized and gets pages directly from | |
556 | * the physical memory. To access them they are temporarily mapped. | |
557 | */ | |
cc615032 | 558 | unsigned long __init_refok init_memory_mapping(unsigned long start, unsigned long end) |
14a62c34 | 559 | { |
cc615032 | 560 | unsigned long next, last_map_addr = end; |
272b9cad | 561 | unsigned long start_phys = start, end_phys = end; |
1da177e4 | 562 | |
272b9cad | 563 | printk(KERN_INFO "init_memory_mapping\n"); |
1da177e4 | 564 | |
14a62c34 | 565 | /* |
1da177e4 | 566 | * Find space for the kernel direct mapping tables. |
14a62c34 TG |
567 | * |
568 | * Later we should allocate these tables in the local node of the | |
569 | * memory mapped. Unfortunately this is done currently before the | |
570 | * nodes are discovered. | |
1da177e4 | 571 | */ |
ef925766 AK |
572 | if (!after_bootmem) { |
573 | init_gbpages(); | |
44df75e6 | 574 | find_early_table_space(end); |
ef925766 | 575 | } |
1da177e4 LT |
576 | |
577 | start = (unsigned long)__va(start); | |
578 | end = (unsigned long)__va(end); | |
579 | ||
580 | for (; start < end; start = next) { | |
44df75e6 | 581 | pgd_t *pgd = pgd_offset_k(start); |
14a62c34 | 582 | unsigned long pud_phys; |
44df75e6 MT |
583 | pud_t *pud; |
584 | ||
585 | if (after_bootmem) | |
d2ae5b5f | 586 | pud = pud_offset(pgd, start & PGDIR_MASK); |
44df75e6 | 587 | else |
dafe41ee | 588 | pud = alloc_low_page(&pud_phys); |
44df75e6 | 589 | |
1da177e4 | 590 | next = start + PGDIR_SIZE; |
14a62c34 TG |
591 | if (next > end) |
592 | next = end; | |
cc615032 | 593 | last_map_addr = phys_pud_init(pud, __pa(start), __pa(next)); |
44df75e6 MT |
594 | if (!after_bootmem) |
595 | set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys)); | |
dafe41ee | 596 | unmap_low_page(pud); |
14a62c34 | 597 | } |
1da177e4 | 598 | |
44df75e6 | 599 | if (!after_bootmem) |
f51c9452 | 600 | mmu_cr4_features = read_cr4(); |
1da177e4 | 601 | __flush_tlb_all(); |
75175278 | 602 | |
24a5da73 YL |
603 | if (!after_bootmem) |
604 | reserve_early(table_start << PAGE_SHIFT, | |
605 | table_end << PAGE_SHIFT, "PGTABLE"); | |
272b9cad YL |
606 | |
607 | if (!after_bootmem) | |
608 | early_memtest(start_phys, end_phys); | |
cc615032 AK |
609 | |
610 | return last_map_addr; | |
1da177e4 LT |
611 | } |
612 | ||
2b97690f | 613 | #ifndef CONFIG_NUMA |
1f75d7e3 YL |
614 | void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn) |
615 | { | |
616 | unsigned long bootmap_size, bootmap; | |
617 | ||
618 | bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT; | |
619 | bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size, | |
620 | PAGE_SIZE); | |
621 | if (bootmap == -1L) | |
622 | panic("Cannot find bootmem map of size %ld\n", bootmap_size); | |
346cafec YL |
623 | /* don't touch min_low_pfn */ |
624 | bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT, | |
625 | 0, end_pfn); | |
1f75d7e3 YL |
626 | e820_register_active_regions(0, start_pfn, end_pfn); |
627 | free_bootmem_with_active_regions(0, end_pfn); | |
628 | early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT); | |
629 | reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT); | |
630 | } | |
631 | ||
1da177e4 LT |
632 | void __init paging_init(void) |
633 | { | |
6391af17 | 634 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
14a62c34 | 635 | |
6391af17 MG |
636 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
637 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
638 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
639 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
640 | ||
44df75e6 MT |
641 | memory_present(0, 0, end_pfn); |
642 | sparse_init(); | |
5cb248ab | 643 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
644 | } |
645 | #endif | |
646 | ||
44df75e6 MT |
647 | /* |
648 | * Memory hotplug specific functions | |
44df75e6 | 649 | */ |
bc02af93 | 650 | #ifdef CONFIG_MEMORY_HOTPLUG |
9d99aaa3 AK |
651 | /* |
652 | * Memory is added always to NORMAL zone. This means you will never get | |
653 | * additional DMA/DMA32 memory. | |
654 | */ | |
bc02af93 | 655 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 656 | { |
bc02af93 | 657 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 658 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
cc615032 | 659 | unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT; |
44df75e6 MT |
660 | unsigned long nr_pages = size >> PAGE_SHIFT; |
661 | int ret; | |
662 | ||
cc615032 AK |
663 | last_mapped_pfn = init_memory_mapping(start, start + size-1); |
664 | if (last_mapped_pfn > max_pfn_mapped) | |
665 | max_pfn_mapped = last_mapped_pfn; | |
45e0b78b | 666 | |
44df75e6 | 667 | ret = __add_pages(zone, start_pfn, nr_pages); |
10f22dde | 668 | WARN_ON(1); |
44df75e6 | 669 | |
44df75e6 | 670 | return ret; |
44df75e6 | 671 | } |
bc02af93 | 672 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 | 673 | |
8243229f | 674 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
4942e998 KM |
675 | int memory_add_physaddr_to_nid(u64 start) |
676 | { | |
677 | return 0; | |
678 | } | |
8c2676a5 | 679 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
4942e998 KM |
680 | #endif |
681 | ||
45e0b78b KM |
682 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
683 | ||
ae531c26 AV |
684 | /* |
685 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
686 | * is valid. The argument is a physical page number. | |
687 | * | |
688 | * | |
689 | * On x86, access has to be given to the first megabyte of ram because that area | |
690 | * contains bios code and data regions used by X and dosemu and similar apps. | |
691 | * Access has to be given to non-kernel-ram areas as well, these contain the PCI | |
692 | * mmio resources as well as potential bios/acpi data regions. | |
693 | */ | |
694 | int devmem_is_allowed(unsigned long pagenr) | |
695 | { | |
696 | if (pagenr <= 256) | |
697 | return 1; | |
698 | if (!page_is_ram(pagenr)) | |
699 | return 1; | |
700 | return 0; | |
701 | } | |
702 | ||
703 | ||
14a62c34 TG |
704 | static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, |
705 | kcore_modules, kcore_vsyscall; | |
1da177e4 LT |
706 | |
707 | void __init mem_init(void) | |
708 | { | |
0a43e4bf | 709 | long codesize, reservedpages, datasize, initsize; |
1da177e4 | 710 | |
0dc243ae | 711 | pci_iommu_alloc(); |
1da177e4 | 712 | |
48ddb154 | 713 | /* clear_bss() already clear the empty_zero_page */ |
1da177e4 LT |
714 | |
715 | reservedpages = 0; | |
716 | ||
717 | /* this will put all low memory onto the freelists */ | |
2b97690f | 718 | #ifdef CONFIG_NUMA |
0a43e4bf | 719 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 720 | #else |
0a43e4bf | 721 | totalram_pages = free_all_bootmem(); |
1da177e4 | 722 | #endif |
5cb248ab MG |
723 | reservedpages = end_pfn - totalram_pages - |
724 | absent_pages_in_range(0, end_pfn); | |
1da177e4 LT |
725 | after_bootmem = 1; |
726 | ||
727 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
728 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
729 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
730 | ||
731 | /* Register memory areas for /proc/kcore */ | |
14a62c34 TG |
732 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
733 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, | |
1da177e4 LT |
734 | VMALLOC_END-VMALLOC_START); |
735 | kclist_add(&kcore_kernel, &_stext, _end - _stext); | |
736 | kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); | |
14a62c34 | 737 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
1da177e4 LT |
738 | VSYSCALL_END - VSYSCALL_START); |
739 | ||
10f22dde | 740 | printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " |
14a62c34 | 741 | "%ldk reserved, %ldk data, %ldk init)\n", |
1da177e4 LT |
742 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
743 | end_pfn << (PAGE_SHIFT-10), | |
744 | codesize >> 10, | |
745 | reservedpages << (PAGE_SHIFT-10), | |
746 | datasize >> 10, | |
747 | initsize >> 10); | |
76ebd054 TG |
748 | |
749 | cpa_init(); | |
1da177e4 LT |
750 | } |
751 | ||
d167a518 | 752 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
1da177e4 | 753 | { |
bfc734b2 | 754 | unsigned long addr = begin; |
1da177e4 | 755 | |
bfc734b2 | 756 | if (addr >= end) |
d167a518 GH |
757 | return; |
758 | ||
ee01f112 IM |
759 | /* |
760 | * If debugging page accesses then do not free this memory but | |
761 | * mark them not present - any buggy init-section access will | |
762 | * create a kernel page fault: | |
763 | */ | |
764 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
765 | printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", | |
766 | begin, PAGE_ALIGN(end)); | |
767 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
768 | #else | |
6fb14755 | 769 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
14a62c34 | 770 | |
bfc734b2 | 771 | for (; addr < end; addr += PAGE_SIZE) { |
e3ebadd9 LT |
772 | ClearPageReserved(virt_to_page(addr)); |
773 | init_page_count(virt_to_page(addr)); | |
774 | memset((void *)(addr & ~(PAGE_SIZE-1)), | |
775 | POISON_FREE_INITMEM, PAGE_SIZE); | |
e3ebadd9 | 776 | free_page(addr); |
1da177e4 LT |
777 | totalram_pages++; |
778 | } | |
ee01f112 | 779 | #endif |
d167a518 GH |
780 | } |
781 | ||
782 | void free_initmem(void) | |
783 | { | |
d167a518 | 784 | free_init_pages("unused kernel memory", |
e3ebadd9 LT |
785 | (unsigned long)(&__init_begin), |
786 | (unsigned long)(&__init_end)); | |
1da177e4 LT |
787 | } |
788 | ||
67df197b | 789 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
790 | const int rodata_test_data = 0xC3; |
791 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
67df197b | 792 | |
67df197b AV |
793 | void mark_rodata_ro(void) |
794 | { | |
4e4eee0e | 795 | unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata); |
67df197b | 796 | |
6fb14755 | 797 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 798 | (end - start) >> 10); |
984bb80d AV |
799 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
800 | ||
801 | /* | |
802 | * The rodata section (but not the kernel text!) should also be | |
803 | * not-executable. | |
804 | */ | |
805 | start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; | |
806 | set_memory_nx(start, (end - start) >> PAGE_SHIFT); | |
67df197b | 807 | |
1a487252 AV |
808 | rodata_test(); |
809 | ||
0c42f392 | 810 | #ifdef CONFIG_CPA_DEBUG |
10f22dde | 811 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); |
6d238cc4 | 812 | set_memory_rw(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 813 | |
10f22dde | 814 | printk(KERN_INFO "Testing CPA: again\n"); |
6d238cc4 | 815 | set_memory_ro(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 816 | #endif |
67df197b | 817 | } |
4e4eee0e | 818 | |
67df197b AV |
819 | #endif |
820 | ||
1da177e4 LT |
821 | #ifdef CONFIG_BLK_DEV_INITRD |
822 | void free_initrd_mem(unsigned long start, unsigned long end) | |
823 | { | |
e3ebadd9 | 824 | free_init_pages("initrd memory", start, end); |
1da177e4 LT |
825 | } |
826 | #endif | |
827 | ||
d2dbf343 YL |
828 | int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, |
829 | int flags) | |
14a62c34 | 830 | { |
2b97690f | 831 | #ifdef CONFIG_NUMA |
8b3cd09e | 832 | int nid, next_nid; |
6a07a0ed | 833 | int ret; |
5e58a02a AK |
834 | #endif |
835 | unsigned long pfn = phys >> PAGE_SHIFT; | |
14a62c34 | 836 | |
5e58a02a | 837 | if (pfn >= end_pfn) { |
14a62c34 TG |
838 | /* |
839 | * This can happen with kdump kernels when accessing | |
840 | * firmware tables: | |
841 | */ | |
67794292 | 842 | if (pfn < max_pfn_mapped) |
8b2ef1d7 | 843 | return -EFAULT; |
14a62c34 | 844 | |
6a07a0ed | 845 | printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n", |
5e58a02a | 846 | phys, len); |
8b2ef1d7 | 847 | return -EFAULT; |
5e58a02a AK |
848 | } |
849 | ||
850 | /* Should check here against the e820 map to avoid double free */ | |
851 | #ifdef CONFIG_NUMA | |
8b3cd09e YL |
852 | nid = phys_to_nid(phys); |
853 | next_nid = phys_to_nid(phys + len - 1); | |
854 | if (nid == next_nid) | |
8b2ef1d7 | 855 | ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags); |
8b3cd09e | 856 | else |
8b2ef1d7 BW |
857 | ret = reserve_bootmem(phys, len, flags); |
858 | ||
859 | if (ret != 0) | |
860 | return ret; | |
861 | ||
14a62c34 | 862 | #else |
72a7fe39 | 863 | reserve_bootmem(phys, len, BOOTMEM_DEFAULT); |
1da177e4 | 864 | #endif |
8b3cd09e | 865 | |
0e0b864e | 866 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
e18c6874 | 867 | dma_reserve += len / PAGE_SIZE; |
0e0b864e MG |
868 | set_dma_reserve(dma_reserve); |
869 | } | |
8b2ef1d7 BW |
870 | |
871 | return 0; | |
1da177e4 LT |
872 | } |
873 | ||
14a62c34 TG |
874 | int kern_addr_valid(unsigned long addr) |
875 | { | |
1da177e4 | 876 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
14a62c34 TG |
877 | pgd_t *pgd; |
878 | pud_t *pud; | |
879 | pmd_t *pmd; | |
880 | pte_t *pte; | |
1da177e4 LT |
881 | |
882 | if (above != 0 && above != -1UL) | |
14a62c34 TG |
883 | return 0; |
884 | ||
1da177e4 LT |
885 | pgd = pgd_offset_k(addr); |
886 | if (pgd_none(*pgd)) | |
887 | return 0; | |
888 | ||
889 | pud = pud_offset(pgd, addr); | |
890 | if (pud_none(*pud)) | |
14a62c34 | 891 | return 0; |
1da177e4 LT |
892 | |
893 | pmd = pmd_offset(pud, addr); | |
894 | if (pmd_none(*pmd)) | |
895 | return 0; | |
14a62c34 | 896 | |
1da177e4 LT |
897 | if (pmd_large(*pmd)) |
898 | return pfn_valid(pmd_pfn(*pmd)); | |
899 | ||
900 | pte = pte_offset_kernel(pmd, addr); | |
901 | if (pte_none(*pte)) | |
902 | return 0; | |
14a62c34 | 903 | |
1da177e4 LT |
904 | return pfn_valid(pte_pfn(*pte)); |
905 | } | |
906 | ||
14a62c34 TG |
907 | /* |
908 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
909 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
910 | * not need special handling anymore: | |
911 | */ | |
1da177e4 | 912 | static struct vm_area_struct gate_vma = { |
14a62c34 TG |
913 | .vm_start = VSYSCALL_START, |
914 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), | |
915 | .vm_page_prot = PAGE_READONLY_EXEC, | |
916 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
917 | }; |
918 | ||
1da177e4 LT |
919 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
920 | { | |
921 | #ifdef CONFIG_IA32_EMULATION | |
1e014410 AK |
922 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
923 | return NULL; | |
1da177e4 LT |
924 | #endif |
925 | return &gate_vma; | |
926 | } | |
927 | ||
928 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
929 | { | |
930 | struct vm_area_struct *vma = get_gate_vma(task); | |
14a62c34 | 931 | |
1e014410 AK |
932 | if (!vma) |
933 | return 0; | |
14a62c34 | 934 | |
1da177e4 LT |
935 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
936 | } | |
937 | ||
14a62c34 TG |
938 | /* |
939 | * Use this when you have no reliable task/vma, typically from interrupt | |
940 | * context. It is less reliable than using the task's vma and may give | |
941 | * false positives: | |
1da177e4 LT |
942 | */ |
943 | int in_gate_area_no_task(unsigned long addr) | |
944 | { | |
1e014410 | 945 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 946 | } |
2e1c49db | 947 | |
2aae950b AK |
948 | const char *arch_vma_name(struct vm_area_struct *vma) |
949 | { | |
950 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
951 | return "[vdso]"; | |
952 | if (vma == &gate_vma) | |
953 | return "[vsyscall]"; | |
954 | return NULL; | |
955 | } | |
0889eba5 CL |
956 | |
957 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
958 | /* | |
959 | * Initialise the sparsemem vmemmap using huge-pages at the PMD level. | |
960 | */ | |
c2b91e2e YL |
961 | static long __meminitdata addr_start, addr_end; |
962 | static void __meminitdata *p_start, *p_end; | |
963 | static int __meminitdata node_start; | |
964 | ||
14a62c34 TG |
965 | int __meminit |
966 | vmemmap_populate(struct page *start_page, unsigned long size, int node) | |
0889eba5 CL |
967 | { |
968 | unsigned long addr = (unsigned long)start_page; | |
969 | unsigned long end = (unsigned long)(start_page + size); | |
970 | unsigned long next; | |
971 | pgd_t *pgd; | |
972 | pud_t *pud; | |
973 | pmd_t *pmd; | |
974 | ||
975 | for (; addr < end; addr = next) { | |
976 | next = pmd_addr_end(addr, end); | |
977 | ||
978 | pgd = vmemmap_pgd_populate(addr, node); | |
979 | if (!pgd) | |
980 | return -ENOMEM; | |
14a62c34 | 981 | |
0889eba5 CL |
982 | pud = vmemmap_pud_populate(pgd, addr, node); |
983 | if (!pud) | |
984 | return -ENOMEM; | |
985 | ||
986 | pmd = pmd_offset(pud, addr); | |
987 | if (pmd_none(*pmd)) { | |
988 | pte_t entry; | |
14a62c34 TG |
989 | void *p; |
990 | ||
991 | p = vmemmap_alloc_block(PMD_SIZE, node); | |
0889eba5 CL |
992 | if (!p) |
993 | return -ENOMEM; | |
994 | ||
14a62c34 TG |
995 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, |
996 | PAGE_KERNEL_LARGE); | |
0889eba5 CL |
997 | set_pmd(pmd, __pmd(pte_val(entry))); |
998 | ||
c2b91e2e YL |
999 | /* check to see if we have contiguous blocks */ |
1000 | if (p_end != p || node_start != node) { | |
1001 | if (p_start) | |
1002 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
1003 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
1004 | addr_start = addr; | |
1005 | node_start = node; | |
1006 | p_start = p; | |
1007 | } | |
1008 | addr_end = addr + PMD_SIZE; | |
1009 | p_end = p + PMD_SIZE; | |
14a62c34 | 1010 | } else { |
0889eba5 | 1011 | vmemmap_verify((pte_t *)pmd, node, addr, next); |
14a62c34 | 1012 | } |
0889eba5 | 1013 | } |
0889eba5 CL |
1014 | return 0; |
1015 | } | |
c2b91e2e YL |
1016 | |
1017 | void __meminit vmemmap_populate_print_last(void) | |
1018 | { | |
1019 | if (p_start) { | |
1020 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
1021 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
1022 | p_start = NULL; | |
1023 | p_end = NULL; | |
1024 | node_start = 0; | |
1025 | } | |
1026 | } | |
0889eba5 | 1027 | #endif |