]>
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 MT |
30 | #include <linux/module.h> |
31 | #include <linux/memory_hotplug.h> | |
ae32b129 | 32 | #include <linux/nmi.h> |
5a0e3ad6 | 33 | #include <linux/gfp.h> |
1da177e4 LT |
34 | |
35 | #include <asm/processor.h> | |
46eaa670 | 36 | #include <asm/bios_ebda.h> |
1da177e4 LT |
37 | #include <asm/system.h> |
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> |
1da177e4 | 55 | |
00d1c5e0 IM |
56 | static int __init parse_direct_gbpages_off(char *arg) |
57 | { | |
58 | direct_gbpages = 0; | |
59 | return 0; | |
60 | } | |
61 | early_param("nogbpages", parse_direct_gbpages_off); | |
62 | ||
63 | static int __init parse_direct_gbpages_on(char *arg) | |
64 | { | |
65 | direct_gbpages = 1; | |
66 | return 0; | |
67 | } | |
68 | early_param("gbpages", parse_direct_gbpages_on); | |
69 | ||
1da177e4 LT |
70 | /* |
71 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the | |
72 | * physical space so we can cache the place of the first one and move | |
73 | * around without checking the pgd every time. | |
74 | */ | |
75 | ||
be43d728 | 76 | pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP; |
bd220a24 YL |
77 | EXPORT_SYMBOL_GPL(__supported_pte_mask); |
78 | ||
bd220a24 YL |
79 | int force_personality32; |
80 | ||
deed05b7 IM |
81 | /* |
82 | * noexec32=on|off | |
83 | * Control non executable heap for 32bit processes. | |
84 | * To control the stack too use noexec=off | |
85 | * | |
86 | * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default) | |
87 | * off PROT_READ implies PROT_EXEC | |
88 | */ | |
bd220a24 YL |
89 | static int __init nonx32_setup(char *str) |
90 | { | |
91 | if (!strcmp(str, "on")) | |
92 | force_personality32 &= ~READ_IMPLIES_EXEC; | |
93 | else if (!strcmp(str, "off")) | |
94 | force_personality32 |= READ_IMPLIES_EXEC; | |
95 | return 1; | |
96 | } | |
97 | __setup("noexec32=", nonx32_setup); | |
98 | ||
6afb5157 HL |
99 | /* |
100 | * When memory was added/removed make sure all the processes MM have | |
101 | * suitable PGD entries in the local PGD level page. | |
102 | */ | |
103 | void sync_global_pgds(unsigned long start, unsigned long end) | |
104 | { | |
44235dcd JF |
105 | unsigned long address; |
106 | ||
107 | for (address = start; address <= end; address += PGDIR_SIZE) { | |
108 | const pgd_t *pgd_ref = pgd_offset_k(address); | |
44235dcd JF |
109 | struct page *page; |
110 | ||
111 | if (pgd_none(*pgd_ref)) | |
112 | continue; | |
113 | ||
a79e53d8 | 114 | spin_lock(&pgd_lock); |
44235dcd JF |
115 | list_for_each_entry(page, &pgd_list, lru) { |
116 | pgd_t *pgd; | |
617d34d9 JF |
117 | spinlock_t *pgt_lock; |
118 | ||
44235dcd | 119 | pgd = (pgd_t *)page_address(page) + pgd_index(address); |
a79e53d8 | 120 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
121 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
122 | spin_lock(pgt_lock); | |
123 | ||
44235dcd JF |
124 | if (pgd_none(*pgd)) |
125 | set_pgd(pgd, *pgd_ref); | |
126 | else | |
127 | BUG_ON(pgd_page_vaddr(*pgd) | |
128 | != pgd_page_vaddr(*pgd_ref)); | |
617d34d9 JF |
129 | |
130 | spin_unlock(pgt_lock); | |
44235dcd | 131 | } |
a79e53d8 | 132 | spin_unlock(&pgd_lock); |
44235dcd | 133 | } |
6afb5157 HL |
134 | } |
135 | ||
8d6ea967 MS |
136 | /* |
137 | * NOTE: This function is marked __ref because it calls __init function | |
138 | * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0. | |
139 | */ | |
140 | static __ref void *spp_getpage(void) | |
14a62c34 | 141 | { |
1da177e4 | 142 | void *ptr; |
14a62c34 | 143 | |
1da177e4 | 144 | if (after_bootmem) |
9e730237 | 145 | ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); |
1da177e4 LT |
146 | else |
147 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
14a62c34 TG |
148 | |
149 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { | |
150 | panic("set_pte_phys: cannot allocate page data %s\n", | |
151 | after_bootmem ? "after bootmem" : ""); | |
152 | } | |
1da177e4 | 153 | |
10f22dde | 154 | pr_debug("spp_getpage %p\n", ptr); |
14a62c34 | 155 | |
1da177e4 | 156 | return ptr; |
14a62c34 | 157 | } |
1da177e4 | 158 | |
f254f390 | 159 | static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr) |
1da177e4 | 160 | { |
458a3e64 TH |
161 | if (pgd_none(*pgd)) { |
162 | pud_t *pud = (pud_t *)spp_getpage(); | |
163 | pgd_populate(&init_mm, pgd, pud); | |
164 | if (pud != pud_offset(pgd, 0)) | |
165 | printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n", | |
166 | pud, pud_offset(pgd, 0)); | |
167 | } | |
168 | return pud_offset(pgd, vaddr); | |
169 | } | |
1da177e4 | 170 | |
f254f390 | 171 | static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr) |
458a3e64 | 172 | { |
1da177e4 | 173 | if (pud_none(*pud)) { |
458a3e64 | 174 | pmd_t *pmd = (pmd_t *) spp_getpage(); |
bb23e403 | 175 | pud_populate(&init_mm, pud, pmd); |
458a3e64 | 176 | if (pmd != pmd_offset(pud, 0)) |
10f22dde | 177 | printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", |
458a3e64 | 178 | pmd, pmd_offset(pud, 0)); |
1da177e4 | 179 | } |
458a3e64 TH |
180 | return pmd_offset(pud, vaddr); |
181 | } | |
182 | ||
f254f390 | 183 | static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr) |
458a3e64 | 184 | { |
1da177e4 | 185 | if (pmd_none(*pmd)) { |
458a3e64 | 186 | pte_t *pte = (pte_t *) spp_getpage(); |
bb23e403 | 187 | pmd_populate_kernel(&init_mm, pmd, pte); |
458a3e64 | 188 | if (pte != pte_offset_kernel(pmd, 0)) |
10f22dde | 189 | printk(KERN_ERR "PAGETABLE BUG #02!\n"); |
1da177e4 | 190 | } |
458a3e64 TH |
191 | return pte_offset_kernel(pmd, vaddr); |
192 | } | |
193 | ||
194 | void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte) | |
195 | { | |
196 | pud_t *pud; | |
197 | pmd_t *pmd; | |
198 | pte_t *pte; | |
199 | ||
200 | pud = pud_page + pud_index(vaddr); | |
201 | pmd = fill_pmd(pud, vaddr); | |
202 | pte = fill_pte(pmd, vaddr); | |
1da177e4 | 203 | |
1da177e4 LT |
204 | set_pte(pte, new_pte); |
205 | ||
206 | /* | |
207 | * It's enough to flush this one mapping. | |
208 | * (PGE mappings get flushed as well) | |
209 | */ | |
210 | __flush_tlb_one(vaddr); | |
211 | } | |
212 | ||
458a3e64 | 213 | void set_pte_vaddr(unsigned long vaddr, pte_t pteval) |
0814e0ba EH |
214 | { |
215 | pgd_t *pgd; | |
216 | pud_t *pud_page; | |
217 | ||
218 | pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval)); | |
219 | ||
220 | pgd = pgd_offset_k(vaddr); | |
221 | if (pgd_none(*pgd)) { | |
222 | printk(KERN_ERR | |
223 | "PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
224 | return; | |
225 | } | |
226 | pud_page = (pud_t*)pgd_page_vaddr(*pgd); | |
227 | set_pte_vaddr_pud(pud_page, vaddr, pteval); | |
228 | } | |
229 | ||
458a3e64 | 230 | pmd_t * __init populate_extra_pmd(unsigned long vaddr) |
11124411 TH |
231 | { |
232 | pgd_t *pgd; | |
233 | pud_t *pud; | |
234 | ||
235 | pgd = pgd_offset_k(vaddr); | |
458a3e64 TH |
236 | pud = fill_pud(pgd, vaddr); |
237 | return fill_pmd(pud, vaddr); | |
238 | } | |
239 | ||
240 | pte_t * __init populate_extra_pte(unsigned long vaddr) | |
241 | { | |
242 | pmd_t *pmd; | |
11124411 | 243 | |
458a3e64 TH |
244 | pmd = populate_extra_pmd(vaddr); |
245 | return fill_pte(pmd, vaddr); | |
11124411 TH |
246 | } |
247 | ||
3a9e189d JS |
248 | /* |
249 | * Create large page table mappings for a range of physical addresses. | |
250 | */ | |
251 | static void __init __init_extra_mapping(unsigned long phys, unsigned long size, | |
252 | pgprot_t prot) | |
253 | { | |
254 | pgd_t *pgd; | |
255 | pud_t *pud; | |
256 | pmd_t *pmd; | |
257 | ||
258 | BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK)); | |
259 | for (; size; phys += PMD_SIZE, size -= PMD_SIZE) { | |
260 | pgd = pgd_offset_k((unsigned long)__va(phys)); | |
261 | if (pgd_none(*pgd)) { | |
262 | pud = (pud_t *) spp_getpage(); | |
263 | set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE | | |
264 | _PAGE_USER)); | |
265 | } | |
266 | pud = pud_offset(pgd, (unsigned long)__va(phys)); | |
267 | if (pud_none(*pud)) { | |
268 | pmd = (pmd_t *) spp_getpage(); | |
269 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | | |
270 | _PAGE_USER)); | |
271 | } | |
272 | pmd = pmd_offset(pud, phys); | |
273 | BUG_ON(!pmd_none(*pmd)); | |
274 | set_pmd(pmd, __pmd(phys | pgprot_val(prot))); | |
275 | } | |
276 | } | |
277 | ||
278 | void __init init_extra_mapping_wb(unsigned long phys, unsigned long size) | |
279 | { | |
280 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE); | |
281 | } | |
282 | ||
283 | void __init init_extra_mapping_uc(unsigned long phys, unsigned long size) | |
284 | { | |
285 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE); | |
286 | } | |
287 | ||
31eedd82 | 288 | /* |
88f3aec7 IM |
289 | * The head.S code sets up the kernel high mapping: |
290 | * | |
291 | * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) | |
31eedd82 TG |
292 | * |
293 | * phys_addr holds the negative offset to the kernel, which is added | |
294 | * to the compile time generated pmds. This results in invalid pmds up | |
295 | * to the point where we hit the physaddr 0 mapping. | |
296 | * | |
297 | * We limit the mappings to the region from _text to _end. _end is | |
298 | * rounded up to the 2MB boundary. This catches the invalid pmds as | |
299 | * well, as they are located before _text: | |
300 | */ | |
301 | void __init cleanup_highmap(void) | |
302 | { | |
303 | unsigned long vaddr = __START_KERNEL_map; | |
d86bb0da | 304 | unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1; |
31eedd82 TG |
305 | pmd_t *pmd = level2_kernel_pgt; |
306 | pmd_t *last_pmd = pmd + PTRS_PER_PMD; | |
307 | ||
308 | for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) { | |
2884f110 | 309 | if (pmd_none(*pmd)) |
31eedd82 TG |
310 | continue; |
311 | if (vaddr < (unsigned long) _text || vaddr > end) | |
312 | set_pmd(pmd, __pmd(0)); | |
313 | } | |
314 | } | |
315 | ||
9482ac6e | 316 | static __ref void *alloc_low_page(unsigned long *phys) |
14a62c34 | 317 | { |
d1b19426 | 318 | unsigned long pfn = pgt_buf_end++; |
1da177e4 LT |
319 | void *adr; |
320 | ||
44df75e6 | 321 | if (after_bootmem) { |
9e730237 | 322 | adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); |
44df75e6 | 323 | *phys = __pa(adr); |
14a62c34 | 324 | |
44df75e6 MT |
325 | return adr; |
326 | } | |
327 | ||
d1b19426 | 328 | if (pfn >= pgt_buf_top) |
14a62c34 | 329 | panic("alloc_low_page: ran out of memory"); |
dafe41ee | 330 | |
14941779 | 331 | adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE); |
234bb549 | 332 | clear_page(adr); |
dafe41ee VG |
333 | *phys = pfn * PAGE_SIZE; |
334 | return adr; | |
335 | } | |
1da177e4 | 336 | |
4b239f45 YL |
337 | static __ref void *map_low_page(void *virt) |
338 | { | |
339 | void *adr; | |
340 | unsigned long phys, left; | |
341 | ||
342 | if (after_bootmem) | |
343 | return virt; | |
344 | ||
345 | phys = __pa(virt); | |
346 | left = phys & (PAGE_SIZE - 1); | |
347 | adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE); | |
348 | adr = (void *)(((unsigned long)adr) | left); | |
349 | ||
350 | return adr; | |
351 | } | |
352 | ||
9482ac6e | 353 | static __ref void unmap_low_page(void *adr) |
14a62c34 | 354 | { |
44df75e6 MT |
355 | if (after_bootmem) |
356 | return; | |
357 | ||
4b239f45 | 358 | early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE); |
14a62c34 | 359 | } |
1da177e4 | 360 | |
7b16eb89 | 361 | static unsigned long __meminit |
b27a43c1 SS |
362 | phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end, |
363 | pgprot_t prot) | |
4f9c11dd JF |
364 | { |
365 | unsigned pages = 0; | |
7b16eb89 | 366 | unsigned long last_map_addr = end; |
4f9c11dd | 367 | int i; |
7b16eb89 | 368 | |
4f9c11dd JF |
369 | pte_t *pte = pte_page + pte_index(addr); |
370 | ||
371 | for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) { | |
372 | ||
373 | if (addr >= end) { | |
374 | if (!after_bootmem) { | |
375 | for(; i < PTRS_PER_PTE; i++, pte++) | |
376 | set_pte(pte, __pte(0)); | |
377 | } | |
378 | break; | |
379 | } | |
380 | ||
b27a43c1 SS |
381 | /* |
382 | * We will re-use the existing mapping. | |
383 | * Xen for example has some special requirements, like mapping | |
384 | * pagetable pages as RO. So assume someone who pre-setup | |
385 | * these mappings are more intelligent. | |
386 | */ | |
3afa3949 YL |
387 | if (pte_val(*pte)) { |
388 | pages++; | |
4f9c11dd | 389 | continue; |
3afa3949 | 390 | } |
4f9c11dd JF |
391 | |
392 | if (0) | |
393 | printk(" pte=%p addr=%lx pte=%016lx\n", | |
394 | pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte); | |
4f9c11dd | 395 | pages++; |
b27a43c1 | 396 | set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot)); |
7b16eb89 | 397 | last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE; |
4f9c11dd | 398 | } |
a2699e47 | 399 | |
4f9c11dd | 400 | update_page_count(PG_LEVEL_4K, pages); |
7b16eb89 YL |
401 | |
402 | return last_map_addr; | |
4f9c11dd JF |
403 | } |
404 | ||
cc615032 | 405 | static unsigned long __meminit |
b50efd2a | 406 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end, |
b27a43c1 | 407 | unsigned long page_size_mask, pgprot_t prot) |
44df75e6 | 408 | { |
ce0c0e50 | 409 | unsigned long pages = 0; |
7b16eb89 | 410 | unsigned long last_map_addr = end; |
ce0c0e50 | 411 | |
6ad91658 | 412 | int i = pmd_index(address); |
44df75e6 | 413 | |
6ad91658 | 414 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
4f9c11dd | 415 | unsigned long pte_phys; |
6ad91658 | 416 | pmd_t *pmd = pmd_page + pmd_index(address); |
4f9c11dd | 417 | pte_t *pte; |
b27a43c1 | 418 | pgprot_t new_prot = prot; |
44df75e6 | 419 | |
5f51e139 | 420 | if (address >= end) { |
14a62c34 | 421 | if (!after_bootmem) { |
5f51e139 JB |
422 | for (; i < PTRS_PER_PMD; i++, pmd++) |
423 | set_pmd(pmd, __pmd(0)); | |
14a62c34 | 424 | } |
44df75e6 MT |
425 | break; |
426 | } | |
6ad91658 | 427 | |
4f9c11dd | 428 | if (pmd_val(*pmd)) { |
8ae3a5a8 JB |
429 | if (!pmd_large(*pmd)) { |
430 | spin_lock(&init_mm.page_table_lock); | |
4b239f45 YL |
431 | pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd)); |
432 | last_map_addr = phys_pte_init(pte, address, | |
b27a43c1 | 433 | end, prot); |
4b239f45 | 434 | unmap_low_page(pte); |
8ae3a5a8 | 435 | spin_unlock(&init_mm.page_table_lock); |
a2699e47 | 436 | continue; |
8ae3a5a8 | 437 | } |
b27a43c1 SS |
438 | /* |
439 | * If we are ok with PG_LEVEL_2M mapping, then we will | |
440 | * use the existing mapping, | |
441 | * | |
442 | * Otherwise, we will split the large page mapping but | |
443 | * use the same existing protection bits except for | |
444 | * large page, so that we don't violate Intel's TLB | |
445 | * Application note (317080) which says, while changing | |
446 | * the page sizes, new and old translations should | |
447 | * not differ with respect to page frame and | |
448 | * attributes. | |
449 | */ | |
3afa3949 YL |
450 | if (page_size_mask & (1 << PG_LEVEL_2M)) { |
451 | pages++; | |
b27a43c1 | 452 | continue; |
3afa3949 | 453 | } |
b27a43c1 | 454 | new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd)); |
4f9c11dd JF |
455 | } |
456 | ||
b50efd2a | 457 | if (page_size_mask & (1<<PG_LEVEL_2M)) { |
4f9c11dd | 458 | pages++; |
8ae3a5a8 | 459 | spin_lock(&init_mm.page_table_lock); |
4f9c11dd | 460 | set_pte((pte_t *)pmd, |
b27a43c1 SS |
461 | pfn_pte(address >> PAGE_SHIFT, |
462 | __pgprot(pgprot_val(prot) | _PAGE_PSE))); | |
8ae3a5a8 | 463 | spin_unlock(&init_mm.page_table_lock); |
7b16eb89 | 464 | last_map_addr = (address & PMD_MASK) + PMD_SIZE; |
6ad91658 | 465 | continue; |
4f9c11dd | 466 | } |
6ad91658 | 467 | |
4f9c11dd | 468 | pte = alloc_low_page(&pte_phys); |
b27a43c1 | 469 | last_map_addr = phys_pte_init(pte, address, end, new_prot); |
4f9c11dd JF |
470 | unmap_low_page(pte); |
471 | ||
8ae3a5a8 | 472 | spin_lock(&init_mm.page_table_lock); |
4f9c11dd | 473 | pmd_populate_kernel(&init_mm, pmd, __va(pte_phys)); |
8ae3a5a8 | 474 | spin_unlock(&init_mm.page_table_lock); |
44df75e6 | 475 | } |
ce0c0e50 | 476 | update_page_count(PG_LEVEL_2M, pages); |
7b16eb89 | 477 | return last_map_addr; |
44df75e6 MT |
478 | } |
479 | ||
cc615032 | 480 | static unsigned long __meminit |
b50efd2a YL |
481 | phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end, |
482 | unsigned long page_size_mask) | |
14a62c34 | 483 | { |
ce0c0e50 | 484 | unsigned long pages = 0; |
cc615032 | 485 | unsigned long last_map_addr = end; |
6ad91658 | 486 | int i = pud_index(addr); |
44df75e6 | 487 | |
14a62c34 | 488 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { |
6ad91658 KM |
489 | unsigned long pmd_phys; |
490 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 | 491 | pmd_t *pmd; |
b27a43c1 | 492 | pgprot_t prot = PAGE_KERNEL; |
1da177e4 | 493 | |
6ad91658 | 494 | if (addr >= end) |
1da177e4 | 495 | break; |
1da177e4 | 496 | |
14a62c34 TG |
497 | if (!after_bootmem && |
498 | !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { | |
499 | set_pud(pud, __pud(0)); | |
1da177e4 | 500 | continue; |
14a62c34 | 501 | } |
1da177e4 | 502 | |
6ad91658 | 503 | if (pud_val(*pud)) { |
a2699e47 | 504 | if (!pud_large(*pud)) { |
4b239f45 YL |
505 | pmd = map_low_page(pmd_offset(pud, 0)); |
506 | last_map_addr = phys_pmd_init(pmd, addr, end, | |
b27a43c1 | 507 | page_size_mask, prot); |
4b239f45 YL |
508 | unmap_low_page(pmd); |
509 | __flush_tlb_all(); | |
a2699e47 SS |
510 | continue; |
511 | } | |
b27a43c1 SS |
512 | /* |
513 | * If we are ok with PG_LEVEL_1G mapping, then we will | |
514 | * use the existing mapping. | |
515 | * | |
516 | * Otherwise, we will split the gbpage mapping but use | |
517 | * the same existing protection bits except for large | |
518 | * page, so that we don't violate Intel's TLB | |
519 | * Application note (317080) which says, while changing | |
520 | * the page sizes, new and old translations should | |
521 | * not differ with respect to page frame and | |
522 | * attributes. | |
523 | */ | |
3afa3949 YL |
524 | if (page_size_mask & (1 << PG_LEVEL_1G)) { |
525 | pages++; | |
b27a43c1 | 526 | continue; |
3afa3949 | 527 | } |
b27a43c1 | 528 | prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud)); |
ef925766 AK |
529 | } |
530 | ||
b50efd2a | 531 | if (page_size_mask & (1<<PG_LEVEL_1G)) { |
ce0c0e50 | 532 | pages++; |
8ae3a5a8 | 533 | spin_lock(&init_mm.page_table_lock); |
ef925766 AK |
534 | set_pte((pte_t *)pud, |
535 | pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
8ae3a5a8 | 536 | spin_unlock(&init_mm.page_table_lock); |
cc615032 | 537 | last_map_addr = (addr & PUD_MASK) + PUD_SIZE; |
6ad91658 KM |
538 | continue; |
539 | } | |
540 | ||
dafe41ee | 541 | pmd = alloc_low_page(&pmd_phys); |
b27a43c1 SS |
542 | last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask, |
543 | prot); | |
4f9c11dd | 544 | unmap_low_page(pmd); |
8ae3a5a8 JB |
545 | |
546 | spin_lock(&init_mm.page_table_lock); | |
4f9c11dd | 547 | pud_populate(&init_mm, pud, __va(pmd_phys)); |
44df75e6 | 548 | spin_unlock(&init_mm.page_table_lock); |
1da177e4 | 549 | } |
1a2b4412 | 550 | __flush_tlb_all(); |
a2699e47 | 551 | |
ce0c0e50 | 552 | update_page_count(PG_LEVEL_1G, pages); |
cc615032 | 553 | |
1a0db38e | 554 | return last_map_addr; |
14a62c34 | 555 | } |
1da177e4 | 556 | |
41d840e2 | 557 | unsigned long __meminit |
f765090a PE |
558 | kernel_physical_mapping_init(unsigned long start, |
559 | unsigned long end, | |
560 | unsigned long page_size_mask) | |
14a62c34 | 561 | { |
9b861528 | 562 | bool pgd_changed = false; |
b50efd2a | 563 | unsigned long next, last_map_addr = end; |
9b861528 | 564 | unsigned long addr; |
1da177e4 LT |
565 | |
566 | start = (unsigned long)__va(start); | |
567 | end = (unsigned long)__va(end); | |
1c5f50ee | 568 | addr = start; |
1da177e4 LT |
569 | |
570 | for (; start < end; start = next) { | |
44df75e6 | 571 | pgd_t *pgd = pgd_offset_k(start); |
14a62c34 | 572 | unsigned long pud_phys; |
44df75e6 MT |
573 | pud_t *pud; |
574 | ||
e22146e6 | 575 | next = (start + PGDIR_SIZE) & PGDIR_MASK; |
4f9c11dd JF |
576 | if (next > end) |
577 | next = end; | |
578 | ||
579 | if (pgd_val(*pgd)) { | |
4b239f45 YL |
580 | pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd)); |
581 | last_map_addr = phys_pud_init(pud, __pa(start), | |
b50efd2a | 582 | __pa(end), page_size_mask); |
4b239f45 | 583 | unmap_low_page(pud); |
4f9c11dd JF |
584 | continue; |
585 | } | |
586 | ||
8ae3a5a8 | 587 | pud = alloc_low_page(&pud_phys); |
b50efd2a YL |
588 | last_map_addr = phys_pud_init(pud, __pa(start), __pa(next), |
589 | page_size_mask); | |
4f9c11dd | 590 | unmap_low_page(pud); |
8ae3a5a8 JB |
591 | |
592 | spin_lock(&init_mm.page_table_lock); | |
593 | pgd_populate(&init_mm, pgd, __va(pud_phys)); | |
594 | spin_unlock(&init_mm.page_table_lock); | |
9b861528 | 595 | pgd_changed = true; |
14a62c34 | 596 | } |
9b861528 HL |
597 | |
598 | if (pgd_changed) | |
599 | sync_global_pgds(addr, end); | |
600 | ||
a2699e47 | 601 | __flush_tlb_all(); |
1da177e4 | 602 | |
b50efd2a YL |
603 | return last_map_addr; |
604 | } | |
7b16eb89 | 605 | |
2b97690f | 606 | #ifndef CONFIG_NUMA |
d8fc3afc | 607 | void __init initmem_init(void) |
1f75d7e3 | 608 | { |
86ef4dbf | 609 | memblock_x86_register_active_regions(0, 0, max_pfn); |
1f75d7e3 | 610 | } |
3551f88f | 611 | #endif |
1f75d7e3 | 612 | |
1da177e4 LT |
613 | void __init paging_init(void) |
614 | { | |
6391af17 | 615 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
14a62c34 | 616 | |
6391af17 MG |
617 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
618 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
619 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
c987d12f | 620 | max_zone_pfns[ZONE_NORMAL] = max_pfn; |
6391af17 | 621 | |
3551f88f | 622 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
44df75e6 | 623 | sparse_init(); |
44b57280 YL |
624 | |
625 | /* | |
626 | * clear the default setting with node 0 | |
627 | * note: don't use nodes_clear here, that is really clearing when | |
628 | * numa support is not compiled in, and later node_set_state | |
629 | * will not set it back. | |
630 | */ | |
631 | node_clear_state(0, N_NORMAL_MEMORY); | |
632 | ||
5cb248ab | 633 | free_area_init_nodes(max_zone_pfns); |
1da177e4 | 634 | } |
1da177e4 | 635 | |
44df75e6 MT |
636 | /* |
637 | * Memory hotplug specific functions | |
44df75e6 | 638 | */ |
bc02af93 | 639 | #ifdef CONFIG_MEMORY_HOTPLUG |
ea085417 SZ |
640 | /* |
641 | * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need | |
642 | * updating. | |
643 | */ | |
644 | static void update_end_of_memory_vars(u64 start, u64 size) | |
645 | { | |
646 | unsigned long end_pfn = PFN_UP(start + size); | |
647 | ||
648 | if (end_pfn > max_pfn) { | |
649 | max_pfn = end_pfn; | |
650 | max_low_pfn = end_pfn; | |
651 | high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; | |
652 | } | |
653 | } | |
654 | ||
9d99aaa3 AK |
655 | /* |
656 | * Memory is added always to NORMAL zone. This means you will never get | |
657 | * additional DMA/DMA32 memory. | |
658 | */ | |
bc02af93 | 659 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 660 | { |
bc02af93 | 661 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 662 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
cc615032 | 663 | unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT; |
44df75e6 MT |
664 | unsigned long nr_pages = size >> PAGE_SHIFT; |
665 | int ret; | |
666 | ||
60817c9b | 667 | last_mapped_pfn = init_memory_mapping(start, start + size); |
cc615032 AK |
668 | if (last_mapped_pfn > max_pfn_mapped) |
669 | max_pfn_mapped = last_mapped_pfn; | |
45e0b78b | 670 | |
c04fc586 | 671 | ret = __add_pages(nid, zone, start_pfn, nr_pages); |
fe8b868e | 672 | WARN_ON_ONCE(ret); |
44df75e6 | 673 | |
ea085417 SZ |
674 | /* update max_pfn, max_low_pfn and high_memory */ |
675 | update_end_of_memory_vars(start, size); | |
676 | ||
44df75e6 | 677 | return ret; |
44df75e6 | 678 | } |
bc02af93 | 679 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 | 680 | |
8243229f | 681 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
4942e998 KM |
682 | int memory_add_physaddr_to_nid(u64 start) |
683 | { | |
684 | return 0; | |
685 | } | |
8c2676a5 | 686 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
4942e998 KM |
687 | #endif |
688 | ||
45e0b78b KM |
689 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
690 | ||
81ac3ad9 | 691 | static struct kcore_list kcore_vsyscall; |
1da177e4 LT |
692 | |
693 | void __init mem_init(void) | |
694 | { | |
0a43e4bf | 695 | long codesize, reservedpages, datasize, initsize; |
11a6b0c9 | 696 | unsigned long absent_pages; |
1da177e4 | 697 | |
0dc243ae | 698 | pci_iommu_alloc(); |
1da177e4 | 699 | |
48ddb154 | 700 | /* clear_bss() already clear the empty_zero_page */ |
1da177e4 LT |
701 | |
702 | reservedpages = 0; | |
703 | ||
704 | /* this will put all low memory onto the freelists */ | |
2b97690f | 705 | #ifdef CONFIG_NUMA |
0a43e4bf | 706 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 707 | #else |
0a43e4bf | 708 | totalram_pages = free_all_bootmem(); |
1da177e4 | 709 | #endif |
11a6b0c9 YL |
710 | |
711 | absent_pages = absent_pages_in_range(0, max_pfn); | |
712 | reservedpages = max_pfn - totalram_pages - absent_pages; | |
1da177e4 LT |
713 | after_bootmem = 1; |
714 | ||
715 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
716 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
717 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
718 | ||
719 | /* Register memory areas for /proc/kcore */ | |
14a62c34 | 720 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
c30bb2a2 | 721 | VSYSCALL_END - VSYSCALL_START, KCORE_OTHER); |
1da177e4 | 722 | |
10f22dde | 723 | printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " |
11a6b0c9 | 724 | "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n", |
cc013a88 | 725 | nr_free_pages() << (PAGE_SHIFT-10), |
c987d12f | 726 | max_pfn << (PAGE_SHIFT-10), |
1da177e4 | 727 | codesize >> 10, |
11a6b0c9 | 728 | absent_pages << (PAGE_SHIFT-10), |
1da177e4 LT |
729 | reservedpages << (PAGE_SHIFT-10), |
730 | datasize >> 10, | |
731 | initsize >> 10); | |
1da177e4 LT |
732 | } |
733 | ||
67df197b | 734 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
735 | const int rodata_test_data = 0xC3; |
736 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
67df197b | 737 | |
502f6604 | 738 | int kernel_set_to_readonly; |
16239630 SR |
739 | |
740 | void set_kernel_text_rw(void) | |
741 | { | |
b9af7c0d | 742 | unsigned long start = PFN_ALIGN(_text); |
e7d23dde | 743 | unsigned long end = PFN_ALIGN(__stop___ex_table); |
16239630 SR |
744 | |
745 | if (!kernel_set_to_readonly) | |
746 | return; | |
747 | ||
748 | pr_debug("Set kernel text: %lx - %lx for read write\n", | |
749 | start, end); | |
750 | ||
e7d23dde SS |
751 | /* |
752 | * Make the kernel identity mapping for text RW. Kernel text | |
753 | * mapping will always be RO. Refer to the comment in | |
754 | * static_protections() in pageattr.c | |
755 | */ | |
16239630 SR |
756 | set_memory_rw(start, (end - start) >> PAGE_SHIFT); |
757 | } | |
758 | ||
759 | void set_kernel_text_ro(void) | |
760 | { | |
b9af7c0d | 761 | unsigned long start = PFN_ALIGN(_text); |
e7d23dde | 762 | unsigned long end = PFN_ALIGN(__stop___ex_table); |
16239630 SR |
763 | |
764 | if (!kernel_set_to_readonly) | |
765 | return; | |
766 | ||
767 | pr_debug("Set kernel text: %lx - %lx for read only\n", | |
768 | start, end); | |
769 | ||
e7d23dde SS |
770 | /* |
771 | * Set the kernel identity mapping for text RO. | |
772 | */ | |
16239630 SR |
773 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
774 | } | |
775 | ||
67df197b AV |
776 | void mark_rodata_ro(void) |
777 | { | |
74e08179 | 778 | unsigned long start = PFN_ALIGN(_text); |
8f0f996e SR |
779 | unsigned long rodata_start = |
780 | ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; | |
74e08179 SS |
781 | unsigned long end = (unsigned long) &__end_rodata_hpage_align; |
782 | unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table); | |
783 | unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata); | |
784 | unsigned long data_start = (unsigned long) &_sdata; | |
8f0f996e | 785 | |
6fb14755 | 786 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 787 | (end - start) >> 10); |
984bb80d AV |
788 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
789 | ||
16239630 SR |
790 | kernel_set_to_readonly = 1; |
791 | ||
984bb80d AV |
792 | /* |
793 | * The rodata section (but not the kernel text!) should also be | |
794 | * not-executable. | |
795 | */ | |
72b59d67 | 796 | set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT); |
67df197b | 797 | |
1a487252 AV |
798 | rodata_test(); |
799 | ||
0c42f392 | 800 | #ifdef CONFIG_CPA_DEBUG |
10f22dde | 801 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); |
6d238cc4 | 802 | set_memory_rw(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 803 | |
10f22dde | 804 | printk(KERN_INFO "Testing CPA: again\n"); |
6d238cc4 | 805 | set_memory_ro(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 806 | #endif |
74e08179 SS |
807 | |
808 | free_init_pages("unused kernel memory", | |
809 | (unsigned long) page_address(virt_to_page(text_end)), | |
810 | (unsigned long) | |
811 | page_address(virt_to_page(rodata_start))); | |
812 | free_init_pages("unused kernel memory", | |
813 | (unsigned long) page_address(virt_to_page(rodata_end)), | |
814 | (unsigned long) page_address(virt_to_page(data_start))); | |
67df197b | 815 | } |
4e4eee0e | 816 | |
67df197b AV |
817 | #endif |
818 | ||
14a62c34 TG |
819 | int kern_addr_valid(unsigned long addr) |
820 | { | |
1da177e4 | 821 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
14a62c34 TG |
822 | pgd_t *pgd; |
823 | pud_t *pud; | |
824 | pmd_t *pmd; | |
825 | pte_t *pte; | |
1da177e4 LT |
826 | |
827 | if (above != 0 && above != -1UL) | |
14a62c34 TG |
828 | return 0; |
829 | ||
1da177e4 LT |
830 | pgd = pgd_offset_k(addr); |
831 | if (pgd_none(*pgd)) | |
832 | return 0; | |
833 | ||
834 | pud = pud_offset(pgd, addr); | |
835 | if (pud_none(*pud)) | |
14a62c34 | 836 | return 0; |
1da177e4 LT |
837 | |
838 | pmd = pmd_offset(pud, addr); | |
839 | if (pmd_none(*pmd)) | |
840 | return 0; | |
14a62c34 | 841 | |
1da177e4 LT |
842 | if (pmd_large(*pmd)) |
843 | return pfn_valid(pmd_pfn(*pmd)); | |
844 | ||
845 | pte = pte_offset_kernel(pmd, addr); | |
846 | if (pte_none(*pte)) | |
847 | return 0; | |
14a62c34 | 848 | |
1da177e4 LT |
849 | return pfn_valid(pte_pfn(*pte)); |
850 | } | |
851 | ||
14a62c34 TG |
852 | /* |
853 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
854 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
855 | * not need special handling anymore: | |
856 | */ | |
1da177e4 | 857 | static struct vm_area_struct gate_vma = { |
14a62c34 TG |
858 | .vm_start = VSYSCALL_START, |
859 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), | |
860 | .vm_page_prot = PAGE_READONLY_EXEC, | |
861 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
862 | }; |
863 | ||
1da177e4 LT |
864 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
865 | { | |
866 | #ifdef CONFIG_IA32_EMULATION | |
1e014410 AK |
867 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
868 | return NULL; | |
1da177e4 LT |
869 | #endif |
870 | return &gate_vma; | |
871 | } | |
872 | ||
873 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
874 | { | |
875 | struct vm_area_struct *vma = get_gate_vma(task); | |
14a62c34 | 876 | |
1e014410 AK |
877 | if (!vma) |
878 | return 0; | |
14a62c34 | 879 | |
1da177e4 LT |
880 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
881 | } | |
882 | ||
14a62c34 TG |
883 | /* |
884 | * Use this when you have no reliable task/vma, typically from interrupt | |
885 | * context. It is less reliable than using the task's vma and may give | |
886 | * false positives: | |
1da177e4 LT |
887 | */ |
888 | int in_gate_area_no_task(unsigned long addr) | |
889 | { | |
1e014410 | 890 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 891 | } |
2e1c49db | 892 | |
2aae950b AK |
893 | const char *arch_vma_name(struct vm_area_struct *vma) |
894 | { | |
895 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
896 | return "[vdso]"; | |
897 | if (vma == &gate_vma) | |
898 | return "[vsyscall]"; | |
899 | return NULL; | |
900 | } | |
0889eba5 | 901 | |
1dc41aa6 NF |
902 | #ifdef CONFIG_X86_UV |
903 | #define MIN_MEMORY_BLOCK_SIZE (1 << SECTION_SIZE_BITS) | |
904 | ||
905 | unsigned long memory_block_size_bytes(void) | |
906 | { | |
907 | if (is_uv_system()) { | |
908 | printk(KERN_INFO "UV: memory block size 2GB\n"); | |
909 | return 2UL * 1024 * 1024 * 1024; | |
910 | } | |
911 | return MIN_MEMORY_BLOCK_SIZE; | |
912 | } | |
913 | #endif | |
914 | ||
0889eba5 CL |
915 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
916 | /* | |
917 | * Initialise the sparsemem vmemmap using huge-pages at the PMD level. | |
918 | */ | |
c2b91e2e YL |
919 | static long __meminitdata addr_start, addr_end; |
920 | static void __meminitdata *p_start, *p_end; | |
921 | static int __meminitdata node_start; | |
922 | ||
14a62c34 TG |
923 | int __meminit |
924 | vmemmap_populate(struct page *start_page, unsigned long size, int node) | |
0889eba5 CL |
925 | { |
926 | unsigned long addr = (unsigned long)start_page; | |
927 | unsigned long end = (unsigned long)(start_page + size); | |
928 | unsigned long next; | |
929 | pgd_t *pgd; | |
930 | pud_t *pud; | |
931 | pmd_t *pmd; | |
932 | ||
933 | for (; addr < end; addr = next) { | |
7c934d39 | 934 | void *p = NULL; |
0889eba5 CL |
935 | |
936 | pgd = vmemmap_pgd_populate(addr, node); | |
937 | if (!pgd) | |
938 | return -ENOMEM; | |
14a62c34 | 939 | |
0889eba5 CL |
940 | pud = vmemmap_pud_populate(pgd, addr, node); |
941 | if (!pud) | |
942 | return -ENOMEM; | |
943 | ||
7c934d39 JF |
944 | if (!cpu_has_pse) { |
945 | next = (addr + PAGE_SIZE) & PAGE_MASK; | |
946 | pmd = vmemmap_pmd_populate(pud, addr, node); | |
947 | ||
948 | if (!pmd) | |
949 | return -ENOMEM; | |
950 | ||
951 | p = vmemmap_pte_populate(pmd, addr, node); | |
14a62c34 | 952 | |
0889eba5 CL |
953 | if (!p) |
954 | return -ENOMEM; | |
955 | ||
7c934d39 JF |
956 | addr_end = addr + PAGE_SIZE; |
957 | p_end = p + PAGE_SIZE; | |
14a62c34 | 958 | } else { |
7c934d39 JF |
959 | next = pmd_addr_end(addr, end); |
960 | ||
961 | pmd = pmd_offset(pud, addr); | |
962 | if (pmd_none(*pmd)) { | |
963 | pte_t entry; | |
964 | ||
9bdac914 | 965 | p = vmemmap_alloc_block_buf(PMD_SIZE, node); |
7c934d39 JF |
966 | if (!p) |
967 | return -ENOMEM; | |
968 | ||
969 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, | |
970 | PAGE_KERNEL_LARGE); | |
971 | set_pmd(pmd, __pmd(pte_val(entry))); | |
972 | ||
7c934d39 JF |
973 | /* check to see if we have contiguous blocks */ |
974 | if (p_end != p || node_start != node) { | |
975 | if (p_start) | |
976 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
977 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
978 | addr_start = addr; | |
979 | node_start = node; | |
980 | p_start = p; | |
981 | } | |
49c980df YL |
982 | |
983 | addr_end = addr + PMD_SIZE; | |
984 | p_end = p + PMD_SIZE; | |
7c934d39 JF |
985 | } else |
986 | vmemmap_verify((pte_t *)pmd, node, addr, next); | |
14a62c34 | 987 | } |
7c934d39 | 988 | |
0889eba5 | 989 | } |
9b861528 | 990 | sync_global_pgds((unsigned long)start_page, end); |
0889eba5 CL |
991 | return 0; |
992 | } | |
c2b91e2e YL |
993 | |
994 | void __meminit vmemmap_populate_print_last(void) | |
995 | { | |
996 | if (p_start) { | |
997 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
998 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
999 | p_start = NULL; | |
1000 | p_end = NULL; | |
1001 | node_start = 0; | |
1002 | } | |
1003 | } | |
0889eba5 | 1004 | #endif |