]>
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> | |
46eaa670 | 34 | #include <asm/bios_ebda.h> |
1da177e4 LT |
35 | #include <asm/system.h> |
36 | #include <asm/uaccess.h> | |
37 | #include <asm/pgtable.h> | |
38 | #include <asm/pgalloc.h> | |
39 | #include <asm/dma.h> | |
40 | #include <asm/fixmap.h> | |
41 | #include <asm/e820.h> | |
42 | #include <asm/apic.h> | |
43 | #include <asm/tlb.h> | |
44 | #include <asm/mmu_context.h> | |
45 | #include <asm/proto.h> | |
46 | #include <asm/smp.h> | |
2bc0414e | 47 | #include <asm/sections.h> |
718fc13b | 48 | #include <asm/kdebug.h> |
aaa64e04 | 49 | #include <asm/numa.h> |
7bfeab9a | 50 | #include <asm/cacheflush.h> |
4fcb2083 | 51 | #include <asm/init.h> |
ea085417 | 52 | #include <linux/bootmem.h> |
1da177e4 | 53 | |
e18c6874 AK |
54 | static unsigned long dma_reserve __initdata; |
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 | ||
8d6ea967 MS |
99 | /* |
100 | * NOTE: This function is marked __ref because it calls __init function | |
101 | * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0. | |
102 | */ | |
103 | static __ref void *spp_getpage(void) | |
14a62c34 | 104 | { |
1da177e4 | 105 | void *ptr; |
14a62c34 | 106 | |
1da177e4 | 107 | if (after_bootmem) |
9e730237 | 108 | ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); |
1da177e4 LT |
109 | else |
110 | ptr = alloc_bootmem_pages(PAGE_SIZE); | |
14a62c34 TG |
111 | |
112 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { | |
113 | panic("set_pte_phys: cannot allocate page data %s\n", | |
114 | after_bootmem ? "after bootmem" : ""); | |
115 | } | |
1da177e4 | 116 | |
10f22dde | 117 | pr_debug("spp_getpage %p\n", ptr); |
14a62c34 | 118 | |
1da177e4 | 119 | return ptr; |
14a62c34 | 120 | } |
1da177e4 | 121 | |
f254f390 | 122 | static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr) |
1da177e4 | 123 | { |
458a3e64 TH |
124 | if (pgd_none(*pgd)) { |
125 | pud_t *pud = (pud_t *)spp_getpage(); | |
126 | pgd_populate(&init_mm, pgd, pud); | |
127 | if (pud != pud_offset(pgd, 0)) | |
128 | printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n", | |
129 | pud, pud_offset(pgd, 0)); | |
130 | } | |
131 | return pud_offset(pgd, vaddr); | |
132 | } | |
1da177e4 | 133 | |
f254f390 | 134 | static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr) |
458a3e64 | 135 | { |
1da177e4 | 136 | if (pud_none(*pud)) { |
458a3e64 | 137 | pmd_t *pmd = (pmd_t *) spp_getpage(); |
bb23e403 | 138 | pud_populate(&init_mm, pud, pmd); |
458a3e64 | 139 | if (pmd != pmd_offset(pud, 0)) |
10f22dde | 140 | printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", |
458a3e64 | 141 | pmd, pmd_offset(pud, 0)); |
1da177e4 | 142 | } |
458a3e64 TH |
143 | return pmd_offset(pud, vaddr); |
144 | } | |
145 | ||
f254f390 | 146 | static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr) |
458a3e64 | 147 | { |
1da177e4 | 148 | if (pmd_none(*pmd)) { |
458a3e64 | 149 | pte_t *pte = (pte_t *) spp_getpage(); |
bb23e403 | 150 | pmd_populate_kernel(&init_mm, pmd, pte); |
458a3e64 | 151 | if (pte != pte_offset_kernel(pmd, 0)) |
10f22dde | 152 | printk(KERN_ERR "PAGETABLE BUG #02!\n"); |
1da177e4 | 153 | } |
458a3e64 TH |
154 | return pte_offset_kernel(pmd, vaddr); |
155 | } | |
156 | ||
157 | void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte) | |
158 | { | |
159 | pud_t *pud; | |
160 | pmd_t *pmd; | |
161 | pte_t *pte; | |
162 | ||
163 | pud = pud_page + pud_index(vaddr); | |
164 | pmd = fill_pmd(pud, vaddr); | |
165 | pte = fill_pte(pmd, vaddr); | |
1da177e4 | 166 | |
1da177e4 LT |
167 | set_pte(pte, new_pte); |
168 | ||
169 | /* | |
170 | * It's enough to flush this one mapping. | |
171 | * (PGE mappings get flushed as well) | |
172 | */ | |
173 | __flush_tlb_one(vaddr); | |
174 | } | |
175 | ||
458a3e64 | 176 | void set_pte_vaddr(unsigned long vaddr, pte_t pteval) |
0814e0ba EH |
177 | { |
178 | pgd_t *pgd; | |
179 | pud_t *pud_page; | |
180 | ||
181 | pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval)); | |
182 | ||
183 | pgd = pgd_offset_k(vaddr); | |
184 | if (pgd_none(*pgd)) { | |
185 | printk(KERN_ERR | |
186 | "PGD FIXMAP MISSING, it should be setup in head.S!\n"); | |
187 | return; | |
188 | } | |
189 | pud_page = (pud_t*)pgd_page_vaddr(*pgd); | |
190 | set_pte_vaddr_pud(pud_page, vaddr, pteval); | |
191 | } | |
192 | ||
458a3e64 | 193 | pmd_t * __init populate_extra_pmd(unsigned long vaddr) |
11124411 TH |
194 | { |
195 | pgd_t *pgd; | |
196 | pud_t *pud; | |
197 | ||
198 | pgd = pgd_offset_k(vaddr); | |
458a3e64 TH |
199 | pud = fill_pud(pgd, vaddr); |
200 | return fill_pmd(pud, vaddr); | |
201 | } | |
202 | ||
203 | pte_t * __init populate_extra_pte(unsigned long vaddr) | |
204 | { | |
205 | pmd_t *pmd; | |
11124411 | 206 | |
458a3e64 TH |
207 | pmd = populate_extra_pmd(vaddr); |
208 | return fill_pte(pmd, vaddr); | |
11124411 TH |
209 | } |
210 | ||
3a9e189d JS |
211 | /* |
212 | * Create large page table mappings for a range of physical addresses. | |
213 | */ | |
214 | static void __init __init_extra_mapping(unsigned long phys, unsigned long size, | |
215 | pgprot_t prot) | |
216 | { | |
217 | pgd_t *pgd; | |
218 | pud_t *pud; | |
219 | pmd_t *pmd; | |
220 | ||
221 | BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK)); | |
222 | for (; size; phys += PMD_SIZE, size -= PMD_SIZE) { | |
223 | pgd = pgd_offset_k((unsigned long)__va(phys)); | |
224 | if (pgd_none(*pgd)) { | |
225 | pud = (pud_t *) spp_getpage(); | |
226 | set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE | | |
227 | _PAGE_USER)); | |
228 | } | |
229 | pud = pud_offset(pgd, (unsigned long)__va(phys)); | |
230 | if (pud_none(*pud)) { | |
231 | pmd = (pmd_t *) spp_getpage(); | |
232 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | | |
233 | _PAGE_USER)); | |
234 | } | |
235 | pmd = pmd_offset(pud, phys); | |
236 | BUG_ON(!pmd_none(*pmd)); | |
237 | set_pmd(pmd, __pmd(phys | pgprot_val(prot))); | |
238 | } | |
239 | } | |
240 | ||
241 | void __init init_extra_mapping_wb(unsigned long phys, unsigned long size) | |
242 | { | |
243 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE); | |
244 | } | |
245 | ||
246 | void __init init_extra_mapping_uc(unsigned long phys, unsigned long size) | |
247 | { | |
248 | __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE); | |
249 | } | |
250 | ||
31eedd82 | 251 | /* |
88f3aec7 IM |
252 | * The head.S code sets up the kernel high mapping: |
253 | * | |
254 | * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) | |
31eedd82 TG |
255 | * |
256 | * phys_addr holds the negative offset to the kernel, which is added | |
257 | * to the compile time generated pmds. This results in invalid pmds up | |
258 | * to the point where we hit the physaddr 0 mapping. | |
259 | * | |
260 | * We limit the mappings to the region from _text to _end. _end is | |
261 | * rounded up to the 2MB boundary. This catches the invalid pmds as | |
262 | * well, as they are located before _text: | |
263 | */ | |
264 | void __init cleanup_highmap(void) | |
265 | { | |
266 | unsigned long vaddr = __START_KERNEL_map; | |
d86bb0da | 267 | unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1; |
31eedd82 TG |
268 | pmd_t *pmd = level2_kernel_pgt; |
269 | pmd_t *last_pmd = pmd + PTRS_PER_PMD; | |
270 | ||
271 | for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) { | |
2884f110 | 272 | if (pmd_none(*pmd)) |
31eedd82 TG |
273 | continue; |
274 | if (vaddr < (unsigned long) _text || vaddr > end) | |
275 | set_pmd(pmd, __pmd(0)); | |
276 | } | |
277 | } | |
278 | ||
9482ac6e | 279 | static __ref void *alloc_low_page(unsigned long *phys) |
14a62c34 | 280 | { |
298af9d8 | 281 | unsigned long pfn = e820_table_end++; |
1da177e4 LT |
282 | void *adr; |
283 | ||
44df75e6 | 284 | if (after_bootmem) { |
9e730237 | 285 | adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); |
44df75e6 | 286 | *phys = __pa(adr); |
14a62c34 | 287 | |
44df75e6 MT |
288 | return adr; |
289 | } | |
290 | ||
298af9d8 | 291 | if (pfn >= e820_table_top) |
14a62c34 | 292 | panic("alloc_low_page: ran out of memory"); |
dafe41ee | 293 | |
14941779 | 294 | adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE); |
44df75e6 | 295 | memset(adr, 0, PAGE_SIZE); |
dafe41ee VG |
296 | *phys = pfn * PAGE_SIZE; |
297 | return adr; | |
298 | } | |
1da177e4 | 299 | |
9482ac6e | 300 | static __ref void unmap_low_page(void *adr) |
14a62c34 | 301 | { |
44df75e6 MT |
302 | if (after_bootmem) |
303 | return; | |
304 | ||
dafe41ee | 305 | early_iounmap(adr, PAGE_SIZE); |
14a62c34 | 306 | } |
1da177e4 | 307 | |
7b16eb89 | 308 | static unsigned long __meminit |
b27a43c1 SS |
309 | phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end, |
310 | pgprot_t prot) | |
4f9c11dd JF |
311 | { |
312 | unsigned pages = 0; | |
7b16eb89 | 313 | unsigned long last_map_addr = end; |
4f9c11dd | 314 | int i; |
7b16eb89 | 315 | |
4f9c11dd JF |
316 | pte_t *pte = pte_page + pte_index(addr); |
317 | ||
318 | for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) { | |
319 | ||
320 | if (addr >= end) { | |
321 | if (!after_bootmem) { | |
322 | for(; i < PTRS_PER_PTE; i++, pte++) | |
323 | set_pte(pte, __pte(0)); | |
324 | } | |
325 | break; | |
326 | } | |
327 | ||
b27a43c1 SS |
328 | /* |
329 | * We will re-use the existing mapping. | |
330 | * Xen for example has some special requirements, like mapping | |
331 | * pagetable pages as RO. So assume someone who pre-setup | |
332 | * these mappings are more intelligent. | |
333 | */ | |
3afa3949 YL |
334 | if (pte_val(*pte)) { |
335 | pages++; | |
4f9c11dd | 336 | continue; |
3afa3949 | 337 | } |
4f9c11dd JF |
338 | |
339 | if (0) | |
340 | printk(" pte=%p addr=%lx pte=%016lx\n", | |
341 | pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte); | |
4f9c11dd | 342 | pages++; |
b27a43c1 | 343 | set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot)); |
7b16eb89 | 344 | last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE; |
4f9c11dd | 345 | } |
a2699e47 | 346 | |
4f9c11dd | 347 | update_page_count(PG_LEVEL_4K, pages); |
7b16eb89 YL |
348 | |
349 | return last_map_addr; | |
4f9c11dd JF |
350 | } |
351 | ||
7b16eb89 | 352 | static unsigned long __meminit |
b27a43c1 SS |
353 | phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end, |
354 | pgprot_t prot) | |
4f9c11dd JF |
355 | { |
356 | pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd); | |
357 | ||
b27a43c1 | 358 | return phys_pte_init(pte, address, end, prot); |
4f9c11dd JF |
359 | } |
360 | ||
cc615032 | 361 | static unsigned long __meminit |
b50efd2a | 362 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end, |
b27a43c1 | 363 | unsigned long page_size_mask, pgprot_t prot) |
44df75e6 | 364 | { |
ce0c0e50 | 365 | unsigned long pages = 0; |
7b16eb89 | 366 | unsigned long last_map_addr = end; |
ce0c0e50 | 367 | |
6ad91658 | 368 | int i = pmd_index(address); |
44df75e6 | 369 | |
6ad91658 | 370 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
4f9c11dd | 371 | unsigned long pte_phys; |
6ad91658 | 372 | pmd_t *pmd = pmd_page + pmd_index(address); |
4f9c11dd | 373 | pte_t *pte; |
b27a43c1 | 374 | pgprot_t new_prot = prot; |
44df75e6 | 375 | |
5f51e139 | 376 | if (address >= end) { |
14a62c34 | 377 | if (!after_bootmem) { |
5f51e139 JB |
378 | for (; i < PTRS_PER_PMD; i++, pmd++) |
379 | set_pmd(pmd, __pmd(0)); | |
14a62c34 | 380 | } |
44df75e6 MT |
381 | break; |
382 | } | |
6ad91658 | 383 | |
4f9c11dd | 384 | if (pmd_val(*pmd)) { |
8ae3a5a8 JB |
385 | if (!pmd_large(*pmd)) { |
386 | spin_lock(&init_mm.page_table_lock); | |
7b16eb89 | 387 | last_map_addr = phys_pte_update(pmd, address, |
b27a43c1 | 388 | end, prot); |
8ae3a5a8 | 389 | spin_unlock(&init_mm.page_table_lock); |
a2699e47 | 390 | continue; |
8ae3a5a8 | 391 | } |
b27a43c1 SS |
392 | /* |
393 | * If we are ok with PG_LEVEL_2M mapping, then we will | |
394 | * use the existing mapping, | |
395 | * | |
396 | * Otherwise, we will split the large page mapping but | |
397 | * use the same existing protection bits except for | |
398 | * large page, so that we don't violate Intel's TLB | |
399 | * Application note (317080) which says, while changing | |
400 | * the page sizes, new and old translations should | |
401 | * not differ with respect to page frame and | |
402 | * attributes. | |
403 | */ | |
3afa3949 YL |
404 | if (page_size_mask & (1 << PG_LEVEL_2M)) { |
405 | pages++; | |
b27a43c1 | 406 | continue; |
3afa3949 | 407 | } |
b27a43c1 | 408 | new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd)); |
4f9c11dd JF |
409 | } |
410 | ||
b50efd2a | 411 | if (page_size_mask & (1<<PG_LEVEL_2M)) { |
4f9c11dd | 412 | pages++; |
8ae3a5a8 | 413 | spin_lock(&init_mm.page_table_lock); |
4f9c11dd | 414 | set_pte((pte_t *)pmd, |
b27a43c1 SS |
415 | pfn_pte(address >> PAGE_SHIFT, |
416 | __pgprot(pgprot_val(prot) | _PAGE_PSE))); | |
8ae3a5a8 | 417 | spin_unlock(&init_mm.page_table_lock); |
7b16eb89 | 418 | last_map_addr = (address & PMD_MASK) + PMD_SIZE; |
6ad91658 | 419 | continue; |
4f9c11dd | 420 | } |
6ad91658 | 421 | |
4f9c11dd | 422 | pte = alloc_low_page(&pte_phys); |
b27a43c1 | 423 | last_map_addr = phys_pte_init(pte, address, end, new_prot); |
4f9c11dd JF |
424 | unmap_low_page(pte); |
425 | ||
8ae3a5a8 | 426 | spin_lock(&init_mm.page_table_lock); |
4f9c11dd | 427 | pmd_populate_kernel(&init_mm, pmd, __va(pte_phys)); |
8ae3a5a8 | 428 | spin_unlock(&init_mm.page_table_lock); |
44df75e6 | 429 | } |
ce0c0e50 | 430 | update_page_count(PG_LEVEL_2M, pages); |
7b16eb89 | 431 | return last_map_addr; |
44df75e6 MT |
432 | } |
433 | ||
cc615032 | 434 | static unsigned long __meminit |
b50efd2a | 435 | phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end, |
b27a43c1 | 436 | unsigned long page_size_mask, pgprot_t prot) |
44df75e6 | 437 | { |
14a62c34 | 438 | pmd_t *pmd = pmd_offset(pud, 0); |
cc615032 AK |
439 | unsigned long last_map_addr; |
440 | ||
b27a43c1 | 441 | last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot); |
6ad91658 | 442 | __flush_tlb_all(); |
cc615032 | 443 | return last_map_addr; |
44df75e6 MT |
444 | } |
445 | ||
cc615032 | 446 | static unsigned long __meminit |
b50efd2a YL |
447 | phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end, |
448 | unsigned long page_size_mask) | |
14a62c34 | 449 | { |
ce0c0e50 | 450 | unsigned long pages = 0; |
cc615032 | 451 | unsigned long last_map_addr = end; |
6ad91658 | 452 | int i = pud_index(addr); |
44df75e6 | 453 | |
14a62c34 | 454 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { |
6ad91658 KM |
455 | unsigned long pmd_phys; |
456 | pud_t *pud = pud_page + pud_index(addr); | |
1da177e4 | 457 | pmd_t *pmd; |
b27a43c1 | 458 | pgprot_t prot = PAGE_KERNEL; |
1da177e4 | 459 | |
6ad91658 | 460 | if (addr >= end) |
1da177e4 | 461 | break; |
1da177e4 | 462 | |
14a62c34 TG |
463 | if (!after_bootmem && |
464 | !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { | |
465 | set_pud(pud, __pud(0)); | |
1da177e4 | 466 | continue; |
14a62c34 | 467 | } |
1da177e4 | 468 | |
6ad91658 | 469 | if (pud_val(*pud)) { |
a2699e47 | 470 | if (!pud_large(*pud)) { |
b50efd2a | 471 | last_map_addr = phys_pmd_update(pud, addr, end, |
b27a43c1 | 472 | page_size_mask, prot); |
a2699e47 SS |
473 | continue; |
474 | } | |
b27a43c1 SS |
475 | /* |
476 | * If we are ok with PG_LEVEL_1G mapping, then we will | |
477 | * use the existing mapping. | |
478 | * | |
479 | * Otherwise, we will split the gbpage mapping but use | |
480 | * the same existing protection bits except for large | |
481 | * page, so that we don't violate Intel's TLB | |
482 | * Application note (317080) which says, while changing | |
483 | * the page sizes, new and old translations should | |
484 | * not differ with respect to page frame and | |
485 | * attributes. | |
486 | */ | |
3afa3949 YL |
487 | if (page_size_mask & (1 << PG_LEVEL_1G)) { |
488 | pages++; | |
b27a43c1 | 489 | continue; |
3afa3949 | 490 | } |
b27a43c1 | 491 | prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud)); |
ef925766 AK |
492 | } |
493 | ||
b50efd2a | 494 | if (page_size_mask & (1<<PG_LEVEL_1G)) { |
ce0c0e50 | 495 | pages++; |
8ae3a5a8 | 496 | spin_lock(&init_mm.page_table_lock); |
ef925766 AK |
497 | set_pte((pte_t *)pud, |
498 | pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); | |
8ae3a5a8 | 499 | spin_unlock(&init_mm.page_table_lock); |
cc615032 | 500 | last_map_addr = (addr & PUD_MASK) + PUD_SIZE; |
6ad91658 KM |
501 | continue; |
502 | } | |
503 | ||
dafe41ee | 504 | pmd = alloc_low_page(&pmd_phys); |
b27a43c1 SS |
505 | last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask, |
506 | prot); | |
4f9c11dd | 507 | unmap_low_page(pmd); |
8ae3a5a8 JB |
508 | |
509 | spin_lock(&init_mm.page_table_lock); | |
4f9c11dd | 510 | pud_populate(&init_mm, pud, __va(pmd_phys)); |
44df75e6 | 511 | spin_unlock(&init_mm.page_table_lock); |
1da177e4 | 512 | } |
1a2b4412 | 513 | __flush_tlb_all(); |
a2699e47 | 514 | |
ce0c0e50 | 515 | update_page_count(PG_LEVEL_1G, pages); |
cc615032 | 516 | |
1a0db38e | 517 | return last_map_addr; |
14a62c34 | 518 | } |
1da177e4 | 519 | |
4f9c11dd | 520 | static unsigned long __meminit |
b50efd2a YL |
521 | phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end, |
522 | unsigned long page_size_mask) | |
4f9c11dd JF |
523 | { |
524 | pud_t *pud; | |
525 | ||
526 | pud = (pud_t *)pgd_page_vaddr(*pgd); | |
527 | ||
b50efd2a | 528 | return phys_pud_init(pud, addr, end, page_size_mask); |
4f9c11dd JF |
529 | } |
530 | ||
41d840e2 | 531 | unsigned long __meminit |
f765090a PE |
532 | kernel_physical_mapping_init(unsigned long start, |
533 | unsigned long end, | |
534 | unsigned long page_size_mask) | |
14a62c34 | 535 | { |
1da177e4 | 536 | |
b50efd2a | 537 | unsigned long next, last_map_addr = end; |
1da177e4 LT |
538 | |
539 | start = (unsigned long)__va(start); | |
540 | end = (unsigned long)__va(end); | |
541 | ||
542 | for (; start < end; start = next) { | |
44df75e6 | 543 | pgd_t *pgd = pgd_offset_k(start); |
14a62c34 | 544 | unsigned long pud_phys; |
44df75e6 MT |
545 | pud_t *pud; |
546 | ||
e22146e6 | 547 | next = (start + PGDIR_SIZE) & PGDIR_MASK; |
4f9c11dd JF |
548 | if (next > end) |
549 | next = end; | |
550 | ||
551 | if (pgd_val(*pgd)) { | |
b50efd2a YL |
552 | last_map_addr = phys_pud_update(pgd, __pa(start), |
553 | __pa(end), page_size_mask); | |
4f9c11dd JF |
554 | continue; |
555 | } | |
556 | ||
8ae3a5a8 | 557 | pud = alloc_low_page(&pud_phys); |
b50efd2a YL |
558 | last_map_addr = phys_pud_init(pud, __pa(start), __pa(next), |
559 | page_size_mask); | |
4f9c11dd | 560 | unmap_low_page(pud); |
8ae3a5a8 JB |
561 | |
562 | spin_lock(&init_mm.page_table_lock); | |
563 | pgd_populate(&init_mm, pgd, __va(pud_phys)); | |
564 | spin_unlock(&init_mm.page_table_lock); | |
14a62c34 | 565 | } |
a2699e47 | 566 | __flush_tlb_all(); |
1da177e4 | 567 | |
b50efd2a YL |
568 | return last_map_addr; |
569 | } | |
7b16eb89 | 570 | |
2b97690f | 571 | #ifndef CONFIG_NUMA |
8ee2debc DR |
572 | void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn, |
573 | int acpi, int k8) | |
1f75d7e3 YL |
574 | { |
575 | unsigned long bootmap_size, bootmap; | |
576 | ||
577 | bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT; | |
578 | bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size, | |
579 | PAGE_SIZE); | |
580 | if (bootmap == -1L) | |
581 | panic("Cannot find bootmem map of size %ld\n", bootmap_size); | |
1842f90c | 582 | reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP"); |
346cafec YL |
583 | /* don't touch min_low_pfn */ |
584 | bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT, | |
585 | 0, end_pfn); | |
1f75d7e3 YL |
586 | e820_register_active_regions(0, start_pfn, end_pfn); |
587 | free_bootmem_with_active_regions(0, end_pfn); | |
1f75d7e3 | 588 | } |
3551f88f | 589 | #endif |
1f75d7e3 | 590 | |
1da177e4 LT |
591 | void __init paging_init(void) |
592 | { | |
6391af17 | 593 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
14a62c34 | 594 | |
6391af17 MG |
595 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
596 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
597 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
c987d12f | 598 | max_zone_pfns[ZONE_NORMAL] = max_pfn; |
6391af17 | 599 | |
3551f88f | 600 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
44df75e6 | 601 | sparse_init(); |
44b57280 YL |
602 | |
603 | /* | |
604 | * clear the default setting with node 0 | |
605 | * note: don't use nodes_clear here, that is really clearing when | |
606 | * numa support is not compiled in, and later node_set_state | |
607 | * will not set it back. | |
608 | */ | |
609 | node_clear_state(0, N_NORMAL_MEMORY); | |
610 | ||
5cb248ab | 611 | free_area_init_nodes(max_zone_pfns); |
1da177e4 | 612 | } |
1da177e4 | 613 | |
44df75e6 MT |
614 | /* |
615 | * Memory hotplug specific functions | |
44df75e6 | 616 | */ |
bc02af93 | 617 | #ifdef CONFIG_MEMORY_HOTPLUG |
ea085417 SZ |
618 | /* |
619 | * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need | |
620 | * updating. | |
621 | */ | |
622 | static void update_end_of_memory_vars(u64 start, u64 size) | |
623 | { | |
624 | unsigned long end_pfn = PFN_UP(start + size); | |
625 | ||
626 | if (end_pfn > max_pfn) { | |
627 | max_pfn = end_pfn; | |
628 | max_low_pfn = end_pfn; | |
629 | high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; | |
630 | } | |
631 | } | |
632 | ||
9d99aaa3 AK |
633 | /* |
634 | * Memory is added always to NORMAL zone. This means you will never get | |
635 | * additional DMA/DMA32 memory. | |
636 | */ | |
bc02af93 | 637 | int arch_add_memory(int nid, u64 start, u64 size) |
44df75e6 | 638 | { |
bc02af93 | 639 | struct pglist_data *pgdat = NODE_DATA(nid); |
776ed98b | 640 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
cc615032 | 641 | unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT; |
44df75e6 MT |
642 | unsigned long nr_pages = size >> PAGE_SHIFT; |
643 | int ret; | |
644 | ||
60817c9b | 645 | last_mapped_pfn = init_memory_mapping(start, start + size); |
cc615032 AK |
646 | if (last_mapped_pfn > max_pfn_mapped) |
647 | max_pfn_mapped = last_mapped_pfn; | |
45e0b78b | 648 | |
c04fc586 | 649 | ret = __add_pages(nid, zone, start_pfn, nr_pages); |
fe8b868e | 650 | WARN_ON_ONCE(ret); |
44df75e6 | 651 | |
ea085417 SZ |
652 | /* update max_pfn, max_low_pfn and high_memory */ |
653 | update_end_of_memory_vars(start, size); | |
654 | ||
44df75e6 | 655 | return ret; |
44df75e6 | 656 | } |
bc02af93 | 657 | EXPORT_SYMBOL_GPL(arch_add_memory); |
44df75e6 | 658 | |
8243229f | 659 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
4942e998 KM |
660 | int memory_add_physaddr_to_nid(u64 start) |
661 | { | |
662 | return 0; | |
663 | } | |
8c2676a5 | 664 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
4942e998 KM |
665 | #endif |
666 | ||
45e0b78b KM |
667 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
668 | ||
81ac3ad9 | 669 | static struct kcore_list kcore_vsyscall; |
1da177e4 LT |
670 | |
671 | void __init mem_init(void) | |
672 | { | |
0a43e4bf | 673 | long codesize, reservedpages, datasize, initsize; |
11a6b0c9 | 674 | unsigned long absent_pages; |
1da177e4 | 675 | |
0dc243ae | 676 | pci_iommu_alloc(); |
1da177e4 | 677 | |
48ddb154 | 678 | /* clear_bss() already clear the empty_zero_page */ |
1da177e4 LT |
679 | |
680 | reservedpages = 0; | |
681 | ||
682 | /* this will put all low memory onto the freelists */ | |
2b97690f | 683 | #ifdef CONFIG_NUMA |
0a43e4bf | 684 | totalram_pages = numa_free_all_bootmem(); |
1da177e4 | 685 | #else |
0a43e4bf | 686 | totalram_pages = free_all_bootmem(); |
1da177e4 | 687 | #endif |
11a6b0c9 YL |
688 | |
689 | absent_pages = absent_pages_in_range(0, max_pfn); | |
690 | reservedpages = max_pfn - totalram_pages - absent_pages; | |
1da177e4 LT |
691 | after_bootmem = 1; |
692 | ||
693 | codesize = (unsigned long) &_etext - (unsigned long) &_text; | |
694 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; | |
695 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; | |
696 | ||
697 | /* Register memory areas for /proc/kcore */ | |
14a62c34 | 698 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
c30bb2a2 | 699 | VSYSCALL_END - VSYSCALL_START, KCORE_OTHER); |
1da177e4 | 700 | |
10f22dde | 701 | printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " |
11a6b0c9 | 702 | "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n", |
cc013a88 | 703 | nr_free_pages() << (PAGE_SHIFT-10), |
c987d12f | 704 | max_pfn << (PAGE_SHIFT-10), |
1da177e4 | 705 | codesize >> 10, |
11a6b0c9 | 706 | absent_pages << (PAGE_SHIFT-10), |
1da177e4 LT |
707 | reservedpages << (PAGE_SHIFT-10), |
708 | datasize >> 10, | |
709 | initsize >> 10); | |
1da177e4 LT |
710 | } |
711 | ||
67df197b | 712 | #ifdef CONFIG_DEBUG_RODATA |
edeed305 AV |
713 | const int rodata_test_data = 0xC3; |
714 | EXPORT_SYMBOL_GPL(rodata_test_data); | |
67df197b | 715 | |
502f6604 | 716 | int kernel_set_to_readonly; |
16239630 SR |
717 | |
718 | void set_kernel_text_rw(void) | |
719 | { | |
b9af7c0d | 720 | unsigned long start = PFN_ALIGN(_text); |
e7d23dde | 721 | unsigned long end = PFN_ALIGN(__stop___ex_table); |
16239630 SR |
722 | |
723 | if (!kernel_set_to_readonly) | |
724 | return; | |
725 | ||
726 | pr_debug("Set kernel text: %lx - %lx for read write\n", | |
727 | start, end); | |
728 | ||
e7d23dde SS |
729 | /* |
730 | * Make the kernel identity mapping for text RW. Kernel text | |
731 | * mapping will always be RO. Refer to the comment in | |
732 | * static_protections() in pageattr.c | |
733 | */ | |
16239630 SR |
734 | set_memory_rw(start, (end - start) >> PAGE_SHIFT); |
735 | } | |
736 | ||
737 | void set_kernel_text_ro(void) | |
738 | { | |
b9af7c0d | 739 | unsigned long start = PFN_ALIGN(_text); |
e7d23dde | 740 | unsigned long end = PFN_ALIGN(__stop___ex_table); |
16239630 SR |
741 | |
742 | if (!kernel_set_to_readonly) | |
743 | return; | |
744 | ||
745 | pr_debug("Set kernel text: %lx - %lx for read only\n", | |
746 | start, end); | |
747 | ||
e7d23dde SS |
748 | /* |
749 | * Set the kernel identity mapping for text RO. | |
750 | */ | |
16239630 SR |
751 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
752 | } | |
753 | ||
67df197b AV |
754 | void mark_rodata_ro(void) |
755 | { | |
74e08179 | 756 | unsigned long start = PFN_ALIGN(_text); |
8f0f996e SR |
757 | unsigned long rodata_start = |
758 | ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; | |
74e08179 SS |
759 | unsigned long end = (unsigned long) &__end_rodata_hpage_align; |
760 | unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table); | |
761 | unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata); | |
762 | unsigned long data_start = (unsigned long) &_sdata; | |
8f0f996e | 763 | |
6fb14755 | 764 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
e3ebadd9 | 765 | (end - start) >> 10); |
984bb80d AV |
766 | set_memory_ro(start, (end - start) >> PAGE_SHIFT); |
767 | ||
16239630 SR |
768 | kernel_set_to_readonly = 1; |
769 | ||
984bb80d AV |
770 | /* |
771 | * The rodata section (but not the kernel text!) should also be | |
772 | * not-executable. | |
773 | */ | |
72b59d67 | 774 | set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT); |
67df197b | 775 | |
1a487252 AV |
776 | rodata_test(); |
777 | ||
0c42f392 | 778 | #ifdef CONFIG_CPA_DEBUG |
10f22dde | 779 | printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); |
6d238cc4 | 780 | set_memory_rw(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 781 | |
10f22dde | 782 | printk(KERN_INFO "Testing CPA: again\n"); |
6d238cc4 | 783 | set_memory_ro(start, (end-start) >> PAGE_SHIFT); |
0c42f392 | 784 | #endif |
74e08179 SS |
785 | |
786 | free_init_pages("unused kernel memory", | |
787 | (unsigned long) page_address(virt_to_page(text_end)), | |
788 | (unsigned long) | |
789 | page_address(virt_to_page(rodata_start))); | |
790 | free_init_pages("unused kernel memory", | |
791 | (unsigned long) page_address(virt_to_page(rodata_end)), | |
792 | (unsigned long) page_address(virt_to_page(data_start))); | |
67df197b | 793 | } |
4e4eee0e | 794 | |
67df197b AV |
795 | #endif |
796 | ||
d2dbf343 YL |
797 | int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, |
798 | int flags) | |
14a62c34 | 799 | { |
2b97690f | 800 | #ifdef CONFIG_NUMA |
8b3cd09e | 801 | int nid, next_nid; |
6a07a0ed | 802 | int ret; |
5e58a02a AK |
803 | #endif |
804 | unsigned long pfn = phys >> PAGE_SHIFT; | |
14a62c34 | 805 | |
c987d12f | 806 | if (pfn >= max_pfn) { |
14a62c34 TG |
807 | /* |
808 | * This can happen with kdump kernels when accessing | |
809 | * firmware tables: | |
810 | */ | |
67794292 | 811 | if (pfn < max_pfn_mapped) |
8b2ef1d7 | 812 | return -EFAULT; |
14a62c34 | 813 | |
6a07a0ed | 814 | printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n", |
5e58a02a | 815 | phys, len); |
8b2ef1d7 | 816 | return -EFAULT; |
5e58a02a AK |
817 | } |
818 | ||
819 | /* Should check here against the e820 map to avoid double free */ | |
820 | #ifdef CONFIG_NUMA | |
8b3cd09e YL |
821 | nid = phys_to_nid(phys); |
822 | next_nid = phys_to_nid(phys + len - 1); | |
823 | if (nid == next_nid) | |
8b2ef1d7 | 824 | ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags); |
8b3cd09e | 825 | else |
8b2ef1d7 BW |
826 | ret = reserve_bootmem(phys, len, flags); |
827 | ||
828 | if (ret != 0) | |
829 | return ret; | |
830 | ||
14a62c34 | 831 | #else |
a6a06f7b | 832 | reserve_bootmem(phys, len, flags); |
1da177e4 | 833 | #endif |
8b3cd09e | 834 | |
0e0b864e | 835 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
e18c6874 | 836 | dma_reserve += len / PAGE_SIZE; |
0e0b864e MG |
837 | set_dma_reserve(dma_reserve); |
838 | } | |
8b2ef1d7 BW |
839 | |
840 | return 0; | |
1da177e4 LT |
841 | } |
842 | ||
14a62c34 TG |
843 | int kern_addr_valid(unsigned long addr) |
844 | { | |
1da177e4 | 845 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
14a62c34 TG |
846 | pgd_t *pgd; |
847 | pud_t *pud; | |
848 | pmd_t *pmd; | |
849 | pte_t *pte; | |
1da177e4 LT |
850 | |
851 | if (above != 0 && above != -1UL) | |
14a62c34 TG |
852 | return 0; |
853 | ||
1da177e4 LT |
854 | pgd = pgd_offset_k(addr); |
855 | if (pgd_none(*pgd)) | |
856 | return 0; | |
857 | ||
858 | pud = pud_offset(pgd, addr); | |
859 | if (pud_none(*pud)) | |
14a62c34 | 860 | return 0; |
1da177e4 LT |
861 | |
862 | pmd = pmd_offset(pud, addr); | |
863 | if (pmd_none(*pmd)) | |
864 | return 0; | |
14a62c34 | 865 | |
1da177e4 LT |
866 | if (pmd_large(*pmd)) |
867 | return pfn_valid(pmd_pfn(*pmd)); | |
868 | ||
869 | pte = pte_offset_kernel(pmd, addr); | |
870 | if (pte_none(*pte)) | |
871 | return 0; | |
14a62c34 | 872 | |
1da177e4 LT |
873 | return pfn_valid(pte_pfn(*pte)); |
874 | } | |
875 | ||
14a62c34 TG |
876 | /* |
877 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
878 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
879 | * not need special handling anymore: | |
880 | */ | |
1da177e4 | 881 | static struct vm_area_struct gate_vma = { |
14a62c34 TG |
882 | .vm_start = VSYSCALL_START, |
883 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), | |
884 | .vm_page_prot = PAGE_READONLY_EXEC, | |
885 | .vm_flags = VM_READ | VM_EXEC | |
1da177e4 LT |
886 | }; |
887 | ||
1da177e4 LT |
888 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
889 | { | |
890 | #ifdef CONFIG_IA32_EMULATION | |
1e014410 AK |
891 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
892 | return NULL; | |
1da177e4 LT |
893 | #endif |
894 | return &gate_vma; | |
895 | } | |
896 | ||
897 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
898 | { | |
899 | struct vm_area_struct *vma = get_gate_vma(task); | |
14a62c34 | 900 | |
1e014410 AK |
901 | if (!vma) |
902 | return 0; | |
14a62c34 | 903 | |
1da177e4 LT |
904 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
905 | } | |
906 | ||
14a62c34 TG |
907 | /* |
908 | * Use this when you have no reliable task/vma, typically from interrupt | |
909 | * context. It is less reliable than using the task's vma and may give | |
910 | * false positives: | |
1da177e4 LT |
911 | */ |
912 | int in_gate_area_no_task(unsigned long addr) | |
913 | { | |
1e014410 | 914 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
1da177e4 | 915 | } |
2e1c49db | 916 | |
2aae950b AK |
917 | const char *arch_vma_name(struct vm_area_struct *vma) |
918 | { | |
919 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
920 | return "[vdso]"; | |
921 | if (vma == &gate_vma) | |
922 | return "[vsyscall]"; | |
923 | return NULL; | |
924 | } | |
0889eba5 CL |
925 | |
926 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
927 | /* | |
928 | * Initialise the sparsemem vmemmap using huge-pages at the PMD level. | |
929 | */ | |
c2b91e2e YL |
930 | static long __meminitdata addr_start, addr_end; |
931 | static void __meminitdata *p_start, *p_end; | |
932 | static int __meminitdata node_start; | |
933 | ||
14a62c34 TG |
934 | int __meminit |
935 | vmemmap_populate(struct page *start_page, unsigned long size, int node) | |
0889eba5 CL |
936 | { |
937 | unsigned long addr = (unsigned long)start_page; | |
938 | unsigned long end = (unsigned long)(start_page + size); | |
939 | unsigned long next; | |
940 | pgd_t *pgd; | |
941 | pud_t *pud; | |
942 | pmd_t *pmd; | |
943 | ||
944 | for (; addr < end; addr = next) { | |
7c934d39 | 945 | void *p = NULL; |
0889eba5 CL |
946 | |
947 | pgd = vmemmap_pgd_populate(addr, node); | |
948 | if (!pgd) | |
949 | return -ENOMEM; | |
14a62c34 | 950 | |
0889eba5 CL |
951 | pud = vmemmap_pud_populate(pgd, addr, node); |
952 | if (!pud) | |
953 | return -ENOMEM; | |
954 | ||
7c934d39 JF |
955 | if (!cpu_has_pse) { |
956 | next = (addr + PAGE_SIZE) & PAGE_MASK; | |
957 | pmd = vmemmap_pmd_populate(pud, addr, node); | |
958 | ||
959 | if (!pmd) | |
960 | return -ENOMEM; | |
961 | ||
962 | p = vmemmap_pte_populate(pmd, addr, node); | |
14a62c34 | 963 | |
0889eba5 CL |
964 | if (!p) |
965 | return -ENOMEM; | |
966 | ||
7c934d39 JF |
967 | addr_end = addr + PAGE_SIZE; |
968 | p_end = p + PAGE_SIZE; | |
14a62c34 | 969 | } else { |
7c934d39 JF |
970 | next = pmd_addr_end(addr, end); |
971 | ||
972 | pmd = pmd_offset(pud, addr); | |
973 | if (pmd_none(*pmd)) { | |
974 | pte_t entry; | |
975 | ||
976 | p = vmemmap_alloc_block(PMD_SIZE, node); | |
977 | if (!p) | |
978 | return -ENOMEM; | |
979 | ||
980 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, | |
981 | PAGE_KERNEL_LARGE); | |
982 | set_pmd(pmd, __pmd(pte_val(entry))); | |
983 | ||
7c934d39 JF |
984 | /* check to see if we have contiguous blocks */ |
985 | if (p_end != p || node_start != node) { | |
986 | if (p_start) | |
987 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
988 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
989 | addr_start = addr; | |
990 | node_start = node; | |
991 | p_start = p; | |
992 | } | |
49c980df YL |
993 | |
994 | addr_end = addr + PMD_SIZE; | |
995 | p_end = p + PMD_SIZE; | |
7c934d39 JF |
996 | } else |
997 | vmemmap_verify((pte_t *)pmd, node, addr, next); | |
14a62c34 | 998 | } |
7c934d39 | 999 | |
0889eba5 | 1000 | } |
0889eba5 CL |
1001 | return 0; |
1002 | } | |
c2b91e2e YL |
1003 | |
1004 | void __meminit vmemmap_populate_print_last(void) | |
1005 | { | |
1006 | if (p_start) { | |
1007 | printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", | |
1008 | addr_start, addr_end-1, p_start, p_end-1, node_start); | |
1009 | p_start = NULL; | |
1010 | p_end = NULL; | |
1011 | node_start = 0; | |
1012 | } | |
1013 | } | |
0889eba5 | 1014 | #endif |