Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/x86_64/mm/init.c | |
3 | * | |
4 | * Copyright (C) 1995 Linus Torvalds | |
a2531293 | 5 | * Copyright (C) 2000 Pavel Machek <pavel@ucw.cz> |
1da177e4 LT |
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> | |
a9ce6bc1 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/proc_fs.h> |
59170891 | 26 | #include <linux/pci.h> |
6fb14755 | 27 | #include <linux/pfn.h> |
c9cf5528 | 28 | #include <linux/poison.h> |
17a941d8 | 29 | #include <linux/dma-mapping.h> |
44df75e6 | 30 | #include <linux/module.h> |
a63fdc51 | 31 | #include <linux/memory.h> |
44df75e6 | 32 | #include <linux/memory_hotplug.h> |
ae32b129 | 33 | #include <linux/nmi.h> |
5a0e3ad6 | 34 | #include <linux/gfp.h> |
1da177e4 LT |
35 | |
36 | #include <asm/processor.h> | |
46eaa670 | 37 | #include <asm/bios_ebda.h> |
1da177e4 LT |
38 | #include <asm/uaccess.h> |
39 | #include <asm/pgtable.h> | |
40 | #include <asm/pgalloc.h> | |
41 | #include <asm/dma.h> | |
42 | #include <asm/fixmap.h> | |
43 | #include <asm/e820.h> | |
44 | #include <asm/apic.h> | |
45 | #include <asm/tlb.h> | |
46 | #include <asm/mmu_context.h> | |
47 | #include <asm/proto.h> | |
48 | #include <asm/smp.h> | |
2bc0414e | 49 | #include <asm/sections.h> |
718fc13b | 50 | #include <asm/kdebug.h> |
aaa64e04 | 51 | #include <asm/numa.h> |
7bfeab9a | 52 | #include <asm/cacheflush.h> |
4fcb2083 | 53 | #include <asm/init.h> |
1dc41aa6 | 54 | #include <asm/uv/uv.h> |
e5f15b45 | 55 | #include <asm/setup.h> |
1da177e4 | 56 | |
00d1c5e0 IM |
57 | static int __init parse_direct_gbpages_off(char *arg) |
58 | { | |
59 | direct_gbpages = 0; | |
60 | return 0; | |
61 | } | |
62 | early_param("nogbpages", parse_direct_gbpages_off); | |
63 | ||
64 | static int __init parse_direct_gbpages_on(char *arg) | |
65 | { | |
66 | direct_gbpages = 1; | |
67 | return 0; | |
68 | } | |
69 | early_param("gbpages", parse_direct_gbpages_on); | |
70 | ||
1da177e4 LT |
71 | /* |
72 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the | |
73 | * physical space so we can cache the place of the first one and move | |
74 | * around without checking the pgd every time. | |
75 | */ | |
76 | ||
be43d728 | 77 | pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP; |
bd220a24 YL |
78 | EXPORT_SYMBOL_GPL(__supported_pte_mask); |
79 | ||
bd220a24 YL |
80 | int force_personality32; |
81 | ||
deed05b7 IM |
82 | /* |
83 | * noexec32=on|off | |
84 | * Control non executable heap for 32bit processes. | |
85 | * To control the stack too use noexec=off | |
86 | * | |
87 | * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default) | |
88 | * off PROT_READ implies PROT_EXEC | |
89 | */ | |
bd220a24 YL |
90 | static int __init nonx32_setup(char *str) |
91 | { | |
92 | if (!strcmp(str, "on")) | |
93 | force_personality32 &= ~READ_IMPLIES_EXEC; | |
94 | else if (!strcmp(str, "off")) | |
95 | force_personality32 |= READ_IMPLIES_EXEC; | |
96 | return 1; | |
97 | } | |
98 | __setup("noexec32=", nonx32_setup); | |
99 | ||
6afb5157 HL |
100 | /* |
101 | * When memory was added/removed make sure all the processes MM have | |
102 | * suitable PGD entries in the local PGD level page. | |
103 | */ | |
104 | void sync_global_pgds(unsigned long start, unsigned long end) | |
105 | { | |
44235dcd JF |
106 | unsigned long address; |
107 | ||
108 | for (address = start; address <= end; address += PGDIR_SIZE) { | |
109 | const pgd_t *pgd_ref = pgd_offset_k(address); | |
44235dcd | 110 | struct page *page; |
53b87cf0 | 111 | pgd_t *pgd; |
44235dcd JF |
112 | |
113 | if (pgd_none(*pgd_ref)) | |
114 | continue; | |
115 | ||
a79e53d8 | 116 | spin_lock(&pgd_lock); |
44235dcd | 117 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 JF |
118 | spinlock_t *pgt_lock; |
119 | ||
44235dcd | 120 | pgd = (pgd_t *)page_address(page) + pgd_index(address); |
a79e53d8 | 121 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
122 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
123 | spin_lock(pgt_lock); | |
124 | ||
44235dcd JF |
125 | if (pgd_none(*pgd)) |
126 | set_pgd(pgd, *pgd_ref); | |
127 | else | |
128 | BUG_ON(pgd_page_vaddr(*pgd) | |
129 | != pgd_page_vaddr(*pgd_ref)); | |
617d34d9 JF |
130 | |
131 | spin_unlock(pgt_lock); | |
44235dcd | 132 | } |
53b87cf0 MF |
133 | |
134 | pgd = __va(real_mode_header->trampoline_pgd); | |
135 | pgd += pgd_index(address); | |
136 | ||
137 | if (pgd_none(*pgd)) | |
138 | set_pgd(pgd, *pgd_ref); | |
139 | ||
a79e53d8 | 140 | spin_unlock(&pgd_lock); |
44235dcd | 141 | } |
6afb5157 HL |
142 | } |
143 | ||
8d6ea967 MS |
144 | /* |
145 | * NOTE: This function is marked __ref because it calls __init function | |
146 | * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0. | |
147 | */ | |
148 | static __ref void *spp_getpage(void) | |
14a62c34 | 149 | { |
1da177e4 | 150 | void *ptr; |
14a62c34 | 151 | |
1da177e4 | 152 | if (after_bootmem) |
9e730237 | 153 | ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); |
1da177e4 LT |
154 | else |
155 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
14a62c34 TG |
156 | |
157 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { | |
158 | panic("set_pte_phys: cannot allocate page data %s\n", | |
159 | after_bootmem ? "after bootmem" : ""); | |
160 | } | |
1da177e4 | 161 | |
10f22dde | 162 | pr_debug("spp_getpage %p\n", ptr); |
14a62c34 | 163 | |
1da177e4 | 164 | return ptr; |
14a62c34 | 165 | } |
1da177e4 | 166 | |
f254f390 | 167 | static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr) |
1da177e4 | 168 | { |
458a3e64 TH |
169 | if (pgd_none(*pgd)) { |
170 | pud_t *pud = (pud_t *)spp_getpage(); | |
171 | pgd_populate(&init_mm, pgd, pud); | |
172 | if (pud != pud_offset(pgd, 0)) | |
173 | printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n", | |
174 | pud, pud_offset(pgd, 0)); | |
175 | } | |
176 | return pud_offset(pgd, vaddr); | |
177 | } | |
1da177e4 | 178 | |
f254f390 | 179 | static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr) |
458a3e64 | 180 | { |
1da177e4 | 181 | if (pud_none(*pud)) { |
458a3e64 | 182 | pmd_t *pmd = (pmd_t *) spp_getpage(); |
bb23e403 | 183 | pud_populate(&init_mm, pud, pmd); |
458a3e64 | 184 | if (pmd != pmd_offset(pud, 0)) |
10f22dde | 185 | printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", |
458a3e64 | 186 | pmd, pmd_offset(pud, 0)); |
1da177e4 | 187 | } |
458a3e64 TH |
188 | return pmd_offset(pud, vaddr); |
189 | } | |
190 | ||
f254f390 | 191 | static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr) |
458a3e64 | 192 | { |
1da177e4 | 193 | if (pmd_none(*pmd)) { |
458a3e64 | 194 | pte_t *pte = (pte_t *) spp_getpage(); |
bb23e403 | 195 | pmd_populate_kernel(&init_mm, pmd, pte); |
458a3e64 | 196 | if (pte != pte_offset_kernel(pmd, 0)) |
10f22dde | 197 | printk(KERN_ERR "PAGETABLE BUG #02!\n"); |
1da177e4 | 198 | } |
458a3e64 TH |
199 | return pte_offset_kernel(pmd, vaddr); |
200 | } | |
201 | ||
202 | void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte) | |
203 | { | |
204 | pud_t *pud; | |
205 | pmd_t *pmd; | |
206 | pte_t *pte; | |
207 | ||
208 | pud = pud_page + pud_index(vaddr); | |
209 | pmd = fill_pmd(pud, vaddr); | |
210 | pte = fill_pte(pmd, vaddr); | |
1da177e4 | 211 | |
1da177e4 LT |
212 | set_pte(pte, new_pte); |
213 | ||
214 | /* | |
215 | * It's enough to flush this one mapping. | |
216 | * (PGE mappings get flushed as well) | |
217 | */ | |
218 | __flush_tlb_one(vaddr); | |
219 | } | |
220 | ||
458a3e64 | 221 | void set_pte_vaddr(unsigned long vaddr, pte_t pteval) |
0814e0ba EH |
222 | { |
223 | pgd_t *pgd; | |
224 | pud_t *pud_page; | |
225 | ||
226 | pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval)); | |
227 | ||
228 | pgd = pgd_offset_k(vaddr); | |
229 | if (pgd_none(*pgd)) { | |
230 | printk(KERN_ERR | |
231 | "PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
232 | return; | |
233 | } | |
234 | pud_page = (pud_t*)pgd_page_vaddr(*pgd); | |
235 | set_pte_vaddr_pud(pud_page, vaddr, pteval); | |
236 | } | |
237 | ||
458a3e64 | 238 | pmd_t * __init populate_extra_pmd(unsigned long vaddr) |
11124411 TH |
239 | { |
240 | pgd_t *pgd; | |
241 | pud_t *pud; | |
242 | ||
243 | pgd = pgd_offset_k(vaddr); | |
458a3e64 TH |
244 | pud = fill_pud(pgd, vaddr); |
245 | return fill_pmd(pud, vaddr); | |
246 | } | |
247 | ||
248 | pte_t * __init populate_extra_pte(unsigned long vaddr) | |
249 | { | |
250 | pmd_t *pmd; | |
11124411 | 251 | |
458a3e64 TH |
252 | pmd = populate_extra_pmd(vaddr); |
253 | return fill_pte(pmd, vaddr); | |
11124411 TH |
254 | } |
255 | ||
3a9e189d JS |
256 | /* |
257 | * Create large page table mappings for a range of physical addresses. | |
258 | */ | |
259 | static void __init __init_extra_mapping(unsigned long phys, unsigned long size, | |
260 | pgprot_t prot) | |
261 | { | |
262 | pgd_t *pgd; | |
263 | pud_t *pud; | |
264 | pmd_t *pmd; | |
265 | ||
266 | BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK)); | |
267 | for (; size; phys += PMD_SIZE, size -= PMD_SIZE) { | |
268 | pgd = pgd_offset_k((unsigned long)__va(phys)); | |
269 | if (pgd_none(*pgd)) { | |
270 | pud = (pud_t *) spp_getpage(); | |
271 | set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE | | |
272 | _PAGE_USER)); | |
273 | } | |
274 | pud = pud_offset(pgd, (unsigned long)__va(phys)); | |
275 | if (pud_none(*pud)) { | |
276 | pmd = (pmd_t *) spp_getpage(); | |
277 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | | |
278 | _PAGE_USER)); | |
279 | } | |
280 | pmd = pmd_offset(pud, phys); | |
281 | BUG_ON(!pmd_none(*pmd)); | |
282 | set_pmd(pmd, __pmd(phys | pgprot_val(prot))); | |
283 | } | |
284 | } | |
285 | ||
286 | void __init init_extra_mapping_wb(unsigned long phys, unsigned long size) | |
287 | { | |
288 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE); | |
289 | } | |
290 | ||
291 | void __init init_extra_mapping_uc(unsigned long phys, unsigned long size) | |
292 | { | |
293 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE); | |
294 | } | |
295 | ||
31eedd82 | 296 | /* |
88f3aec7 IM |
297 | * The head.S code sets up the kernel high mapping: |
298 | * | |
299 | * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) | |
31eedd82 TG |
300 | * |
301 | * phys_addr holds the negative offset to the kernel, which is added | |
302 | * to the compile time generated pmds. This results in invalid pmds up | |
303 | * to the point where we hit the physaddr 0 mapping. | |
304 | * | |
e5f15b45 YL |
305 | * We limit the mappings to the region from _text to _brk_end. _brk_end |
306 | * is rounded up to the 2MB boundary. This catches the invalid pmds as | |
31eedd82 TG |
307 | * well, as they are located before _text: |
308 | */ | |
309 | void __init cleanup_highmap(void) | |
310 | { | |
311 | unsigned long vaddr = __START_KERNEL_map; | |
e5f15b45 YL |
312 | unsigned long vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT); |
313 | unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1; | |
31eedd82 | 314 | pmd_t *pmd = level2_kernel_pgt; |
31eedd82 | 315 | |
e5f15b45 | 316 | for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) { |
2884f110 | 317 | if (pmd_none(*pmd)) |
31eedd82 TG |
318 | continue; |
319 | if (vaddr < (unsigned long) _text || vaddr > end) | |
320 | set_pmd(pmd, __pmd(0)); | |
321 | } | |
322 | } | |
323 | ||
9482ac6e | 324 | static __ref void *alloc_low_page(unsigned long *phys) |
14a62c34 | 325 | { |
d1b19426 | 326 | unsigned long pfn = pgt_buf_end++; |
1da177e4 LT |
327 | void *adr; |
328 | ||
44df75e6 | 329 | if (after_bootmem) { |
9e730237 | 330 | adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); |
44df75e6 | 331 | *phys = __pa(adr); |
14a62c34 | 332 | |
44df75e6 MT |
333 | return adr; |
334 | } | |
335 | ||
d1b19426 | 336 | if (pfn >= pgt_buf_top) |
14a62c34 | 337 | panic("alloc_low_page: ran out of memory"); |
dafe41ee | 338 | |
14941779 | 339 | adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE); |
234bb549 | 340 | clear_page(adr); |
dafe41ee VG |
341 | *phys = pfn * PAGE_SIZE; |
342 | return adr; | |
343 | } | |
1da177e4 | 344 | |
4b239f45 YL |
345 | static __ref void *map_low_page(void *virt) |
346 | { | |
347 | void *adr; | |
348 | unsigned long phys, left; | |
349 | ||
350 | if (after_bootmem) | |
351 | return virt; | |
352 | ||
353 | phys = __pa(virt); | |
354 | left = phys & (PAGE_SIZE - 1); | |
355 | adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE); | |
356 | adr = (void *)(((unsigned long)adr) | left); | |
357 | ||
358 | return adr; | |
359 | } | |
360 | ||
9482ac6e | 361 | static __ref void unmap_low_page(void *adr) |
14a62c34 | 362 | { |
44df75e6 MT |
363 | if (after_bootmem) |
364 | return; | |
365 | ||
4b239f45 | 366 | early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE); |
14a62c34 | 367 | } |
1da177e4 | 368 | |
7b16eb89 | 369 | static unsigned long __meminit |
b27a43c1 SS |
370 | phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end, |
371 | pgprot_t prot) | |
4f9c11dd JF |
372 | { |
373 | unsigned pages = 0; | |
7b16eb89 | 374 | unsigned long last_map_addr = end; |
4f9c11dd | 375 | int i; |
7b16eb89 | 376 | |
4f9c11dd JF |
377 | pte_t *pte = pte_page + pte_index(addr); |
378 | ||
379 | for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) { | |
380 | ||
381 | if (addr >= end) { | |
382 | if (!after_bootmem) { | |
383 | for(; i < PTRS_PER_PTE; i++, pte++) | |
384 | set_pte(pte, __pte(0)); | |
385 | } | |
386 | break; | |
387 | } | |
388 | ||
b27a43c1 SS |
389 | /* |
390 | * We will re-use the existing mapping. | |
391 | * Xen for example has some special requirements, like mapping | |
392 | * pagetable pages as RO. So assume someone who pre-setup | |
393 | * these mappings are more intelligent. | |
394 | */ | |
3afa3949 | 395 | if (pte_val(*pte)) { |
876ee61a JB |
396 | if (!after_bootmem) |
397 | pages++; | |
4f9c11dd | 398 | continue; |
3afa3949 | 399 | } |
4f9c11dd JF |
400 | |
401 | if (0) | |
402 | printk(" pte=%p addr=%lx pte=%016lx\n", | |
403 | pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte); | |
4f9c11dd | 404 | pages++; |
b27a43c1 | 405 | set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot)); |
7b16eb89 | 406 | last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE; |
4f9c11dd | 407 | } |
a2699e47 | 408 | |
4f9c11dd | 409 | update_page_count(PG_LEVEL_4K, pages); |
7b16eb89 YL |
410 | |
411 | return last_map_addr; | |
4f9c11dd JF |
412 | } |
413 | ||
cc615032 | 414 | static unsigned long __meminit |
b50efd2a | 415 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end, |
b27a43c1 | 416 | unsigned long page_size_mask, pgprot_t prot) |
44df75e6 | 417 | { |
20167d34 | 418 | unsigned long pages = 0, next; |
7b16eb89 | 419 | unsigned long last_map_addr = end; |
ce0c0e50 | 420 | |
6ad91658 | 421 | int i = pmd_index(address); |
44df75e6 | 422 | |
20167d34 | 423 | for (; i < PTRS_PER_PMD; i++, address = next) { |
4f9c11dd | 424 | unsigned long pte_phys; |
6ad91658 | 425 | pmd_t *pmd = pmd_page + pmd_index(address); |
4f9c11dd | 426 | pte_t *pte; |
b27a43c1 | 427 | pgprot_t new_prot = prot; |
44df75e6 | 428 | |
5f51e139 | 429 | if (address >= end) { |
14a62c34 | 430 | if (!after_bootmem) { |
5f51e139 JB |
431 | for (; i < PTRS_PER_PMD; i++, pmd++) |
432 | set_pmd(pmd, __pmd(0)); | |
14a62c34 | 433 | } |
44df75e6 MT |
434 | break; |
435 | } | |
6ad91658 | 436 | |
20167d34 JB |
437 | next = (address & PMD_MASK) + PMD_SIZE; |
438 | ||
4f9c11dd | 439 | if (pmd_val(*pmd)) { |
8ae3a5a8 JB |
440 | if (!pmd_large(*pmd)) { |
441 | spin_lock(&init_mm.page_table_lock); | |
4b239f45 YL |
442 | pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd)); |
443 | last_map_addr = phys_pte_init(pte, address, | |
b27a43c1 | 444 | end, prot); |
4b239f45 | 445 | unmap_low_page(pte); |
8ae3a5a8 | 446 | spin_unlock(&init_mm.page_table_lock); |
a2699e47 | 447 | continue; |
8ae3a5a8 | 448 | } |
b27a43c1 SS |
449 | /* |
450 | * If we are ok with PG_LEVEL_2M mapping, then we will | |
451 | * use the existing mapping, | |
452 | * | |
453 | * Otherwise, we will split the large page mapping but | |
454 | * use the same existing protection bits except for | |
455 | * large page, so that we don't violate Intel's TLB | |
456 | * Application note (317080) which says, while changing | |
457 | * the page sizes, new and old translations should | |
458 | * not differ with respect to page frame and | |
459 | * attributes. | |
460 | */ | |
3afa3949 | 461 | if (page_size_mask & (1 << PG_LEVEL_2M)) { |
876ee61a JB |
462 | if (!after_bootmem) |
463 | pages++; | |
20167d34 | 464 | last_map_addr = next; |
b27a43c1 | 465 | continue; |
3afa3949 | 466 | } |
b27a43c1 | 467 | new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd)); |
4f9c11dd JF |
468 | } |
469 | ||
b50efd2a | 470 | if (page_size_mask & (1<<PG_LEVEL_2M)) { |
4f9c11dd | 471 | pages++; |
8ae3a5a8 | 472 | spin_lock(&init_mm.page_table_lock); |
4f9c11dd | 473 | set_pte((pte_t *)pmd, |
b27a43c1 SS |
474 | pfn_pte(address >> PAGE_SHIFT, |
475 | __pgprot(pgprot_val(prot) | _PAGE_PSE))); | |
8ae3a5a8 | 476 | spin_unlock(&init_mm.page_table_lock); |
20167d34 | 477 | last_map_addr = next; |
6ad91658 | 478 | continue; |
4f9c11dd | 479 | } |
6ad91658 | 480 | |
4f9c11dd | 481 | pte = alloc_low_page(&pte_phys); |
b27a43c1 | 482 | last_map_addr = phys_pte_init(pte, address, end, new_prot); |
4f9c11dd JF |
483 | unmap_low_page(pte); |
484 | ||
8ae3a5a8 | 485 | spin_lock(&init_mm.page_table_lock); |
4f9c11dd | 486 | pmd_populate_kernel(&init_mm, pmd, __va(pte_phys)); |
8ae3a5a8 | 487 | spin_unlock(&init_mm.page_table_lock); |
44df75e6 | 488 | } |
ce0c0e50 | 489 | update_page_count(PG_LEVEL_2M, pages); |
7b16eb89 | 490 | return last_map_addr; |
44df75e6 MT |
491 | } |
492 | ||
cc615032 | 493 | static unsigned long __meminit |
b50efd2a YL |
494 | phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end, |
495 | unsigned long page_size_mask) | |
14a62c34 | 496 | { |
20167d34 | 497 | unsigned long pages = 0, next; |
cc615032 | 498 | unsigned long last_map_addr = end; |
6ad91658 | 499 | int i = pud_index(addr); |
44df75e6 | 500 | |
20167d34 | 501 | for (; i < PTRS_PER_PUD; i++, addr = next) { |
6ad91658 KM |
502 | unsigned long pmd_phys; |
503 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 | 504 | pmd_t *pmd; |
b27a43c1 | 505 | pgprot_t prot = PAGE_KERNEL; |
1da177e4 | 506 | |
6ad91658 | 507 | if (addr >= end) |
1da177e4 | 508 | break; |
1da177e4 | 509 | |
20167d34 JB |
510 | next = (addr & PUD_MASK) + PUD_SIZE; |
511 | ||
512 | if (!after_bootmem && !e820_any_mapped(addr, next, 0)) { | |
14a62c34 | 513 | set_pud(pud, __pud(0)); |
1da177e4 | 514 | continue; |
14a62c34 | 515 | } |
1da177e4 | 516 | |
6ad91658 | 517 | if (pud_val(*pud)) { |
a2699e47 | 518 | if (!pud_large(*pud)) { |
4b239f45 YL |
519 | pmd = map_low_page(pmd_offset(pud, 0)); |
520 | last_map_addr = phys_pmd_init(pmd, addr, end, | |
b27a43c1 | 521 | page_size_mask, prot); |
4b239f45 YL |
522 | unmap_low_page(pmd); |
523 | __flush_tlb_all(); | |
a2699e47 SS |
524 | continue; |
525 | } | |
b27a43c1 SS |
526 | /* |
527 | * If we are ok with PG_LEVEL_1G mapping, then we will | |
528 | * use the existing mapping. | |
529 | * | |
530 | * Otherwise, we will split the gbpage mapping but use | |
531 | * the same existing protection bits except for large | |
532 | * page, so that we don't violate Intel's TLB | |
533 | * Application note (317080) which says, while changing | |
534 | * the page sizes, new and old translations should | |
535 | * not differ with respect to page frame and | |
536 | * attributes. | |
537 | */ | |
3afa3949 | 538 | if (page_size_mask & (1 << PG_LEVEL_1G)) { |
876ee61a JB |
539 | if (!after_bootmem) |
540 | pages++; | |
20167d34 | 541 | last_map_addr = next; |
b27a43c1 | 542 | continue; |
3afa3949 | 543 | } |
b27a43c1 | 544 | prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud)); |
ef925766 AK |
545 | } |
546 | ||
b50efd2a | 547 | if (page_size_mask & (1<<PG_LEVEL_1G)) { |
ce0c0e50 | 548 | pages++; |
8ae3a5a8 | 549 | spin_lock(&init_mm.page_table_lock); |
ef925766 AK |
550 | set_pte((pte_t *)pud, |
551 | pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
8ae3a5a8 | 552 | spin_unlock(&init_mm.page_table_lock); |
20167d34 | 553 | last_map_addr = next; |
6ad91658 KM |
554 | continue; |
555 | } | |
556 | ||
dafe41ee | 557 | pmd = alloc_low_page(&pmd_phys); |
b27a43c1 SS |
558 | last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask, |
559 | prot); | |
4f9c11dd | 560 | unmap_low_page(pmd); |
8ae3a5a8 JB |
561 | |
562 | spin_lock(&init_mm.page_table_lock); | |
4f9c11dd | 563 | pud_populate(&init_mm, pud, __va(pmd_phys)); |
44df75e6 | 564 | spin_unlock(&init_mm.page_table_lock); |
1da177e4 | 565 | } |
1a2b4412 | 566 | __flush_tlb_all(); |
a2699e47 | 567 | |
ce0c0e50 | 568 | update_page_count(PG_LEVEL_1G, pages); |
cc615032 | 569 | |
1a0db38e | 570 | return last_map_addr; |
14a62c34 | 571 | } |
1da177e4 | 572 | |
41d840e2 | 573 | unsigned long __meminit |
f765090a PE |
574 | kernel_physical_mapping_init(unsigned long start, |
575 | unsigned long end, | |
576 | unsigned long page_size_mask) | |
14a62c34 | 577 | { |
9b861528 | 578 | bool pgd_changed = false; |
b50efd2a | 579 | unsigned long next, last_map_addr = end; |
9b861528 | 580 | unsigned long addr; |
1da177e4 LT |
581 | |
582 | start = (unsigned long)__va(start); | |
583 | end = (unsigned long)__va(end); | |
1c5f50ee | 584 | addr = start; |
1da177e4 LT |
585 | |
586 | for (; start < end; start = next) { | |
44df75e6 | 587 | pgd_t *pgd = pgd_offset_k(start); |
14a62c34 | 588 | unsigned long pud_phys; |
44df75e6 MT |
589 | pud_t *pud; |
590 | ||
e22146e6 | 591 | next = (start + PGDIR_SIZE) & PGDIR_MASK; |
4f9c11dd JF |
592 | if (next > end) |
593 | next = end; | |
594 | ||
595 | if (pgd_val(*pgd)) { | |
4b239f45 YL |
596 | pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd)); |
597 | last_map_addr = phys_pud_init(pud, __pa(start), | |
b50efd2a | 598 | __pa(end), page_size_mask); |
4b239f45 | 599 | unmap_low_page(pud); |
4f9c11dd JF |
600 | continue; |
601 | } | |
602 | ||
8ae3a5a8 | 603 | pud = alloc_low_page(&pud_phys); |
b50efd2a YL |
604 | last_map_addr = phys_pud_init(pud, __pa(start), __pa(next), |
605 | page_size_mask); | |
4f9c11dd | 606 | unmap_low_page(pud); |
8ae3a5a8 JB |
607 | |
608 | spin_lock(&init_mm.page_table_lock); | |
609 | pgd_populate(&init_mm, pgd, __va(pud_phys)); | |
610 | spin_unlock(&init_mm.page_table_lock); | |
9b861528 | 611 | pgd_changed = true; |
14a62c34 | 612 | } |
9b861528 HL |
613 | |
614 | if (pgd_changed) | |
615 | sync_global_pgds(addr, end); | |
616 | ||
a2699e47 | 617 | __flush_tlb_all(); |
1da177e4 | 618 | |
b50efd2a YL |
619 | return last_map_addr; |
620 | } | |
7b16eb89 | 621 | |
2b97690f | 622 | #ifndef CONFIG_NUMA |
d8fc3afc | 623 | void __init initmem_init(void) |
1f75d7e3 | 624 | { |
0608f70c | 625 | memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0); |
1f75d7e3 | 626 | } |
3551f88f | 627 | #endif |
1f75d7e3 | 628 | |
1da177e4 LT |
629 | void __init paging_init(void) |
630 | { | |
3551f88f | 631 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
44df75e6 | 632 | sparse_init(); |
44b57280 YL |
633 | |
634 | /* | |
635 | * clear the default setting with node 0 | |
636 | * note: don't use nodes_clear here, that is really clearing when | |
637 | * numa support is not compiled in, and later node_set_state | |
638 | * will not set it back. | |
639 | */ | |
4b0ef1fe LJ |
640 | node_clear_state(0, N_MEMORY); |
641 | if (N_MEMORY != N_NORMAL_MEMORY) | |
642 | node_clear_state(0, N_NORMAL_MEMORY); | |
44b57280 | 643 | |
4c0b2e5f | 644 | zone_sizes_init(); |
1da177e4 | 645 | } |
1da177e4 | 646 | |
44df75e6 MT |
647 | /* |
648 | * Memory hotplug specific functions | |
44df75e6 | 649 | */ |
bc02af93 | 650 | #ifdef CONFIG_MEMORY_HOTPLUG |
ea085417 SZ |
651 | /* |
652 | * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need | |
653 | * updating. | |
654 | */ | |
655 | static void update_end_of_memory_vars(u64 start, u64 size) | |
656 | { | |
657 | unsigned long end_pfn = PFN_UP(start + size); | |
658 | ||
659 | if (end_pfn > max_pfn) { | |
660 | max_pfn = end_pfn; | |
661 | max_low_pfn = end_pfn; | |
662 | high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; | |
663 | } | |
664 | } | |
665 | ||
9d99aaa3 AK |
666 | /* |
667 | * Memory is added always to NORMAL zone. This means you will never get | |
668 | * additional DMA/DMA32 memory. | |
669 | */ | |
bc02af93 | 670 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 671 | { |
bc02af93 | 672 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 673 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
cc615032 | 674 | unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT; |
44df75e6 MT |
675 | unsigned long nr_pages = size >> PAGE_SHIFT; |
676 | int ret; | |
677 | ||
60817c9b | 678 | last_mapped_pfn = init_memory_mapping(start, start + size); |
cc615032 AK |
679 | if (last_mapped_pfn > max_pfn_mapped) |
680 | max_pfn_mapped = last_mapped_pfn; | |
45e0b78b | 681 | |
c04fc586 | 682 | ret = __add_pages(nid, zone, start_pfn, nr_pages); |
fe8b868e | 683 | WARN_ON_ONCE(ret); |
44df75e6 | 684 | |
ea085417 SZ |
685 | /* update max_pfn, max_low_pfn and high_memory */ |
686 | update_end_of_memory_vars(start, size); | |
687 | ||
44df75e6 | 688 | return ret; |
44df75e6 | 689 | } |
bc02af93 | 690 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 | 691 | |
45e0b78b KM |
692 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
693 | ||
81ac3ad9 | 694 | static struct kcore_list kcore_vsyscall; |
1da177e4 LT |
695 | |
696 | void __init mem_init(void) | |
697 | { | |
0a43e4bf | 698 | long codesize, reservedpages, datasize, initsize; |
11a6b0c9 | 699 | unsigned long absent_pages; |
1da177e4 | 700 | |
0dc243ae | 701 | pci_iommu_alloc(); |
1da177e4 | 702 | |
48ddb154 | 703 | /* clear_bss() already clear the empty_zero_page */ |
1da177e4 LT |
704 | |
705 | reservedpages = 0; | |
706 | ||
707 | /* this will put all low memory onto the freelists */ | |
2b97690f | 708 | #ifdef CONFIG_NUMA |
0a43e4bf | 709 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 710 | #else |
0a43e4bf | 711 | totalram_pages = free_all_bootmem(); |
1da177e4 | 712 | #endif |
11a6b0c9 YL |
713 | |
714 | absent_pages = absent_pages_in_range(0, max_pfn); | |
715 | reservedpages = max_pfn - totalram_pages - absent_pages; | |
1da177e4 LT |
716 | after_bootmem = 1; |
717 | ||
718 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
719 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
720 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
721 | ||
722 | /* Register memory areas for /proc/kcore */ | |
14a62c34 | 723 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
c30bb2a2 | 724 | VSYSCALL_END - VSYSCALL_START, KCORE_OTHER); |
1da177e4 | 725 | |
10f22dde | 726 | printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " |
11a6b0c9 | 727 | "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n", |
cc013a88 | 728 | nr_free_pages() << (PAGE_SHIFT-10), |
c987d12f | 729 | max_pfn << (PAGE_SHIFT-10), |
1da177e4 | 730 | codesize >> 10, |
11a6b0c9 | 731 | absent_pages << (PAGE_SHIFT-10), |
1da177e4 LT |
732 | reservedpages << (PAGE_SHIFT-10), |
733 | datasize >> 10, | |
734 | initsize >> 10); | |
1da177e4 LT |
735 | } |
736 | ||
67df197b | 737 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
738 | const int rodata_test_data = 0xC3; |
739 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
67df197b | 740 | |
502f6604 | 741 | int kernel_set_to_readonly; |
16239630 SR |
742 | |
743 | void set_kernel_text_rw(void) | |
744 | { | |
b9af7c0d | 745 | unsigned long start = PFN_ALIGN(_text); |
e7d23dde | 746 | unsigned long end = PFN_ALIGN(__stop___ex_table); |
16239630 SR |
747 | |
748 | if (!kernel_set_to_readonly) | |
749 | return; | |
750 | ||
751 | pr_debug("Set kernel text: %lx - %lx for read write\n", | |
752 | start, end); | |
753 | ||
e7d23dde SS |
754 | /* |
755 | * Make the kernel identity mapping for text RW. Kernel text | |
756 | * mapping will always be RO. Refer to the comment in | |
757 | * static_protections() in pageattr.c | |
758 | */ | |
16239630 SR |
759 | set_memory_rw(start, (end - start) >> PAGE_SHIFT); |
760 | } | |
761 | ||
762 | void set_kernel_text_ro(void) | |
763 | { | |
b9af7c0d | 764 | unsigned long start = PFN_ALIGN(_text); |
e7d23dde | 765 | unsigned long end = PFN_ALIGN(__stop___ex_table); |
16239630 SR |
766 | |
767 | if (!kernel_set_to_readonly) | |
768 | return; | |
769 | ||
770 | pr_debug("Set kernel text: %lx - %lx for read only\n", | |
771 | start, end); | |
772 | ||
e7d23dde SS |
773 | /* |
774 | * Set the kernel identity mapping for text RO. | |
775 | */ | |
16239630 SR |
776 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
777 | } | |
778 | ||
67df197b AV |
779 | void mark_rodata_ro(void) |
780 | { | |
74e08179 | 781 | unsigned long start = PFN_ALIGN(_text); |
8f0f996e SR |
782 | unsigned long rodata_start = |
783 | ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; | |
74e08179 SS |
784 | unsigned long end = (unsigned long) &__end_rodata_hpage_align; |
785 | unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table); | |
786 | unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata); | |
787 | unsigned long data_start = (unsigned long) &_sdata; | |
8f0f996e | 788 | |
6fb14755 | 789 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 790 | (end - start) >> 10); |
984bb80d AV |
791 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
792 | ||
16239630 SR |
793 | kernel_set_to_readonly = 1; |
794 | ||
984bb80d AV |
795 | /* |
796 | * The rodata section (but not the kernel text!) should also be | |
797 | * not-executable. | |
798 | */ | |
72b59d67 | 799 | set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT); |
67df197b | 800 | |
1a487252 AV |
801 | rodata_test(); |
802 | ||
0c42f392 | 803 | #ifdef CONFIG_CPA_DEBUG |
10f22dde | 804 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); |
6d238cc4 | 805 | set_memory_rw(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 806 | |
10f22dde | 807 | printk(KERN_INFO "Testing CPA: again\n"); |
6d238cc4 | 808 | set_memory_ro(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 809 | #endif |
74e08179 SS |
810 | |
811 | free_init_pages("unused kernel memory", | |
812 | (unsigned long) page_address(virt_to_page(text_end)), | |
813 | (unsigned long) | |
814 | page_address(virt_to_page(rodata_start))); | |
815 | free_init_pages("unused kernel memory", | |
816 | (unsigned long) page_address(virt_to_page(rodata_end)), | |
817 | (unsigned long) page_address(virt_to_page(data_start))); | |
67df197b | 818 | } |
4e4eee0e | 819 | |
67df197b AV |
820 | #endif |
821 | ||
14a62c34 TG |
822 | int kern_addr_valid(unsigned long addr) |
823 | { | |
1da177e4 | 824 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
14a62c34 TG |
825 | pgd_t *pgd; |
826 | pud_t *pud; | |
827 | pmd_t *pmd; | |
828 | pte_t *pte; | |
1da177e4 LT |
829 | |
830 | if (above != 0 && above != -1UL) | |
14a62c34 TG |
831 | return 0; |
832 | ||
1da177e4 LT |
833 | pgd = pgd_offset_k(addr); |
834 | if (pgd_none(*pgd)) | |
835 | return 0; | |
836 | ||
837 | pud = pud_offset(pgd, addr); | |
838 | if (pud_none(*pud)) | |
14a62c34 | 839 | return 0; |
1da177e4 LT |
840 | |
841 | pmd = pmd_offset(pud, addr); | |
842 | if (pmd_none(*pmd)) | |
843 | return 0; | |
14a62c34 | 844 | |
1da177e4 LT |
845 | if (pmd_large(*pmd)) |
846 | return pfn_valid(pmd_pfn(*pmd)); | |
847 | ||
848 | pte = pte_offset_kernel(pmd, addr); | |
849 | if (pte_none(*pte)) | |
850 | return 0; | |
14a62c34 | 851 | |
1da177e4 LT |
852 | return pfn_valid(pte_pfn(*pte)); |
853 | } | |
854 | ||
14a62c34 TG |
855 | /* |
856 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
857 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
858 | * not need special handling anymore: | |
859 | */ | |
1da177e4 | 860 | static struct vm_area_struct gate_vma = { |
14a62c34 TG |
861 | .vm_start = VSYSCALL_START, |
862 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), | |
863 | .vm_page_prot = PAGE_READONLY_EXEC, | |
864 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
865 | }; |
866 | ||
31db58b3 | 867 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) |
1da177e4 LT |
868 | { |
869 | #ifdef CONFIG_IA32_EMULATION | |
31db58b3 | 870 | if (!mm || mm->context.ia32_compat) |
1e014410 | 871 | return NULL; |
1da177e4 LT |
872 | #endif |
873 | return &gate_vma; | |
874 | } | |
875 | ||
83b964bb | 876 | int in_gate_area(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 877 | { |
83b964bb | 878 | struct vm_area_struct *vma = get_gate_vma(mm); |
14a62c34 | 879 | |
1e014410 AK |
880 | if (!vma) |
881 | return 0; | |
14a62c34 | 882 | |
1da177e4 LT |
883 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
884 | } | |
885 | ||
14a62c34 | 886 | /* |
cae5d390 SW |
887 | * Use this when you have no reliable mm, typically from interrupt |
888 | * context. It is less reliable than using a task's mm and may give | |
889 | * false positives. | |
1da177e4 | 890 | */ |
cae5d390 | 891 | int in_gate_area_no_mm(unsigned long addr) |
1da177e4 | 892 | { |
1e014410 | 893 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 894 | } |
2e1c49db | 895 | |
2aae950b AK |
896 | const char *arch_vma_name(struct vm_area_struct *vma) |
897 | { | |
898 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
899 | return "[vdso]"; | |
900 | if (vma == &gate_vma) | |
901 | return "[vsyscall]"; | |
902 | return NULL; | |
903 | } | |
0889eba5 | 904 | |
1dc41aa6 | 905 | #ifdef CONFIG_X86_UV |
1dc41aa6 NF |
906 | unsigned long memory_block_size_bytes(void) |
907 | { | |
908 | if (is_uv_system()) { | |
909 | printk(KERN_INFO "UV: memory block size 2GB\n"); | |
910 | return 2UL * 1024 * 1024 * 1024; | |
911 | } | |
912 | return MIN_MEMORY_BLOCK_SIZE; | |
913 | } | |
914 | #endif | |
915 | ||
0889eba5 CL |
916 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
917 | /* | |
918 | * Initialise the sparsemem vmemmap using huge-pages at the PMD level. | |
919 | */ | |
c2b91e2e YL |
920 | static long __meminitdata addr_start, addr_end; |
921 | static void __meminitdata *p_start, *p_end; | |
922 | static int __meminitdata node_start; | |
923 | ||
14a62c34 TG |
924 | int __meminit |
925 | vmemmap_populate(struct page *start_page, unsigned long size, int node) | |
0889eba5 CL |
926 | { |
927 | unsigned long addr = (unsigned long)start_page; | |
928 | unsigned long end = (unsigned long)(start_page + size); | |
929 | unsigned long next; | |
930 | pgd_t *pgd; | |
931 | pud_t *pud; | |
932 | pmd_t *pmd; | |
933 | ||
934 | for (; addr < end; addr = next) { | |
7c934d39 | 935 | void *p = NULL; |
0889eba5 CL |
936 | |
937 | pgd = vmemmap_pgd_populate(addr, node); | |
938 | if (!pgd) | |
939 | return -ENOMEM; | |
14a62c34 | 940 | |
0889eba5 CL |
941 | pud = vmemmap_pud_populate(pgd, addr, node); |
942 | if (!pud) | |
943 | return -ENOMEM; | |
944 | ||
7c934d39 JF |
945 | if (!cpu_has_pse) { |
946 | next = (addr + PAGE_SIZE) & PAGE_MASK; | |
947 | pmd = vmemmap_pmd_populate(pud, addr, node); | |
948 | ||
949 | if (!pmd) | |
950 | return -ENOMEM; | |
951 | ||
952 | p = vmemmap_pte_populate(pmd, addr, node); | |
14a62c34 | 953 | |
0889eba5 CL |
954 | if (!p) |
955 | return -ENOMEM; | |
956 | ||
7c934d39 JF |
957 | addr_end = addr + PAGE_SIZE; |
958 | p_end = p + PAGE_SIZE; | |
14a62c34 | 959 | } else { |
7c934d39 JF |
960 | next = pmd_addr_end(addr, end); |
961 | ||
962 | pmd = pmd_offset(pud, addr); | |
963 | if (pmd_none(*pmd)) { | |
964 | pte_t entry; | |
965 | ||
9bdac914 | 966 | p = vmemmap_alloc_block_buf(PMD_SIZE, node); |
7c934d39 JF |
967 | if (!p) |
968 | return -ENOMEM; | |
969 | ||
970 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, | |
971 | PAGE_KERNEL_LARGE); | |
972 | set_pmd(pmd, __pmd(pte_val(entry))); | |
973 | ||
7c934d39 JF |
974 | /* check to see if we have contiguous blocks */ |
975 | if (p_end != p || node_start != node) { | |
976 | if (p_start) | |
977 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
978 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
979 | addr_start = addr; | |
980 | node_start = node; | |
981 | p_start = p; | |
982 | } | |
49c980df YL |
983 | |
984 | addr_end = addr + PMD_SIZE; | |
985 | p_end = p + PMD_SIZE; | |
7c934d39 JF |
986 | } else |
987 | vmemmap_verify((pte_t *)pmd, node, addr, next); | |
14a62c34 | 988 | } |
7c934d39 | 989 | |
0889eba5 | 990 | } |
9b861528 | 991 | sync_global_pgds((unsigned long)start_page, end); |
0889eba5 CL |
992 | return 0; |
993 | } | |
c2b91e2e YL |
994 | |
995 | void __meminit vmemmap_populate_print_last(void) | |
996 | { | |
997 | if (p_start) { | |
998 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
999 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
1000 | p_start = NULL; | |
1001 | p_end = NULL; | |
1002 | node_start = 0; | |
1003 | } | |
1004 | } | |
0889eba5 | 1005 | #endif |