xref: /openbmc/linux/arch/x86/mm/init_64.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /*
2  *  linux/arch/x86_64/mm/init.c
3  *
4  *  Copyright (C) 1995  Linus Torvalds
5  *  Copyright (C) 2000  Pavel Machek <pavel@ucw.cz>
6  *  Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
7  */
8 
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
17 #include <linux/mm.h>
18 #include <linux/swap.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/initrd.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/proc_fs.h>
26 #include <linux/pci.h>
27 #include <linux/pfn.h>
28 #include <linux/poison.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/module.h>
31 #include <linux/memory_hotplug.h>
32 #include <linux/nmi.h>
33 #include <linux/gfp.h>
34 
35 #include <asm/processor.h>
36 #include <asm/bios_ebda.h>
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>
49 #include <asm/sections.h>
50 #include <asm/kdebug.h>
51 #include <asm/numa.h>
52 #include <asm/cacheflush.h>
53 #include <asm/init.h>
54 
55 static int __init parse_direct_gbpages_off(char *arg)
56 {
57 	direct_gbpages = 0;
58 	return 0;
59 }
60 early_param("nogbpages", parse_direct_gbpages_off);
61 
62 static int __init parse_direct_gbpages_on(char *arg)
63 {
64 	direct_gbpages = 1;
65 	return 0;
66 }
67 early_param("gbpages", parse_direct_gbpages_on);
68 
69 /*
70  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
71  * physical space so we can cache the place of the first one and move
72  * around without checking the pgd every time.
73  */
74 
75 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
76 EXPORT_SYMBOL_GPL(__supported_pte_mask);
77 
78 int force_personality32;
79 
80 /*
81  * noexec32=on|off
82  * Control non executable heap for 32bit processes.
83  * To control the stack too use noexec=off
84  *
85  * on	PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
86  * off	PROT_READ implies PROT_EXEC
87  */
88 static int __init nonx32_setup(char *str)
89 {
90 	if (!strcmp(str, "on"))
91 		force_personality32 &= ~READ_IMPLIES_EXEC;
92 	else if (!strcmp(str, "off"))
93 		force_personality32 |= READ_IMPLIES_EXEC;
94 	return 1;
95 }
96 __setup("noexec32=", nonx32_setup);
97 
98 /*
99  * When memory was added/removed make sure all the processes MM have
100  * suitable PGD entries in the local PGD level page.
101  */
102 void sync_global_pgds(unsigned long start, unsigned long end)
103 {
104 	unsigned long address;
105 
106 	for (address = start; address <= end; address += PGDIR_SIZE) {
107 		const pgd_t *pgd_ref = pgd_offset_k(address);
108 		unsigned long flags;
109 		struct page *page;
110 
111 		if (pgd_none(*pgd_ref))
112 			continue;
113 
114 		spin_lock_irqsave(&pgd_lock, flags);
115 		list_for_each_entry(page, &pgd_list, lru) {
116 			pgd_t *pgd;
117 			spinlock_t *pgt_lock;
118 
119 			pgd = (pgd_t *)page_address(page) + pgd_index(address);
120 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
121 			spin_lock(pgt_lock);
122 
123 			if (pgd_none(*pgd))
124 				set_pgd(pgd, *pgd_ref);
125 			else
126 				BUG_ON(pgd_page_vaddr(*pgd)
127 				       != pgd_page_vaddr(*pgd_ref));
128 
129 			spin_unlock(pgt_lock);
130 		}
131 		spin_unlock_irqrestore(&pgd_lock, flags);
132 	}
133 }
134 
135 /*
136  * NOTE: This function is marked __ref because it calls __init function
137  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
138  */
139 static __ref void *spp_getpage(void)
140 {
141 	void *ptr;
142 
143 	if (after_bootmem)
144 		ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
145 	else
146 		ptr = alloc_bootmem_pages(PAGE_SIZE);
147 
148 	if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
149 		panic("set_pte_phys: cannot allocate page data %s\n",
150 			after_bootmem ? "after bootmem" : "");
151 	}
152 
153 	pr_debug("spp_getpage %p\n", ptr);
154 
155 	return ptr;
156 }
157 
158 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
159 {
160 	if (pgd_none(*pgd)) {
161 		pud_t *pud = (pud_t *)spp_getpage();
162 		pgd_populate(&init_mm, pgd, pud);
163 		if (pud != pud_offset(pgd, 0))
164 			printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
165 			       pud, pud_offset(pgd, 0));
166 	}
167 	return pud_offset(pgd, vaddr);
168 }
169 
170 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
171 {
172 	if (pud_none(*pud)) {
173 		pmd_t *pmd = (pmd_t *) spp_getpage();
174 		pud_populate(&init_mm, pud, pmd);
175 		if (pmd != pmd_offset(pud, 0))
176 			printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
177 			       pmd, pmd_offset(pud, 0));
178 	}
179 	return pmd_offset(pud, vaddr);
180 }
181 
182 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
183 {
184 	if (pmd_none(*pmd)) {
185 		pte_t *pte = (pte_t *) spp_getpage();
186 		pmd_populate_kernel(&init_mm, pmd, pte);
187 		if (pte != pte_offset_kernel(pmd, 0))
188 			printk(KERN_ERR "PAGETABLE BUG #02!\n");
189 	}
190 	return pte_offset_kernel(pmd, vaddr);
191 }
192 
193 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
194 {
195 	pud_t *pud;
196 	pmd_t *pmd;
197 	pte_t *pte;
198 
199 	pud = pud_page + pud_index(vaddr);
200 	pmd = fill_pmd(pud, vaddr);
201 	pte = fill_pte(pmd, vaddr);
202 
203 	set_pte(pte, new_pte);
204 
205 	/*
206 	 * It's enough to flush this one mapping.
207 	 * (PGE mappings get flushed as well)
208 	 */
209 	__flush_tlb_one(vaddr);
210 }
211 
212 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
213 {
214 	pgd_t *pgd;
215 	pud_t *pud_page;
216 
217 	pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
218 
219 	pgd = pgd_offset_k(vaddr);
220 	if (pgd_none(*pgd)) {
221 		printk(KERN_ERR
222 			"PGD FIXMAP MISSING, it should be setup in head.S!\n");
223 		return;
224 	}
225 	pud_page = (pud_t*)pgd_page_vaddr(*pgd);
226 	set_pte_vaddr_pud(pud_page, vaddr, pteval);
227 }
228 
229 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
230 {
231 	pgd_t *pgd;
232 	pud_t *pud;
233 
234 	pgd = pgd_offset_k(vaddr);
235 	pud = fill_pud(pgd, vaddr);
236 	return fill_pmd(pud, vaddr);
237 }
238 
239 pte_t * __init populate_extra_pte(unsigned long vaddr)
240 {
241 	pmd_t *pmd;
242 
243 	pmd = populate_extra_pmd(vaddr);
244 	return fill_pte(pmd, vaddr);
245 }
246 
247 /*
248  * Create large page table mappings for a range of physical addresses.
249  */
250 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
251 						pgprot_t prot)
252 {
253 	pgd_t *pgd;
254 	pud_t *pud;
255 	pmd_t *pmd;
256 
257 	BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
258 	for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
259 		pgd = pgd_offset_k((unsigned long)__va(phys));
260 		if (pgd_none(*pgd)) {
261 			pud = (pud_t *) spp_getpage();
262 			set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
263 						_PAGE_USER));
264 		}
265 		pud = pud_offset(pgd, (unsigned long)__va(phys));
266 		if (pud_none(*pud)) {
267 			pmd = (pmd_t *) spp_getpage();
268 			set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
269 						_PAGE_USER));
270 		}
271 		pmd = pmd_offset(pud, phys);
272 		BUG_ON(!pmd_none(*pmd));
273 		set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
274 	}
275 }
276 
277 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
278 {
279 	__init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
280 }
281 
282 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
283 {
284 	__init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
285 }
286 
287 /*
288  * The head.S code sets up the kernel high mapping:
289  *
290  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
291  *
292  * phys_addr holds the negative offset to the kernel, which is added
293  * to the compile time generated pmds. This results in invalid pmds up
294  * to the point where we hit the physaddr 0 mapping.
295  *
296  * We limit the mappings to the region from _text to _end.  _end is
297  * rounded up to the 2MB boundary. This catches the invalid pmds as
298  * well, as they are located before _text:
299  */
300 void __init cleanup_highmap(void)
301 {
302 	unsigned long vaddr = __START_KERNEL_map;
303 	unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
304 	pmd_t *pmd = level2_kernel_pgt;
305 	pmd_t *last_pmd = pmd + PTRS_PER_PMD;
306 
307 	for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
308 		if (pmd_none(*pmd))
309 			continue;
310 		if (vaddr < (unsigned long) _text || vaddr > end)
311 			set_pmd(pmd, __pmd(0));
312 	}
313 }
314 
315 static __ref void *alloc_low_page(unsigned long *phys)
316 {
317 	unsigned long pfn = e820_table_end++;
318 	void *adr;
319 
320 	if (after_bootmem) {
321 		adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
322 		*phys = __pa(adr);
323 
324 		return adr;
325 	}
326 
327 	if (pfn >= e820_table_top)
328 		panic("alloc_low_page: ran out of memory");
329 
330 	adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
331 	clear_page(adr);
332 	*phys  = pfn * PAGE_SIZE;
333 	return adr;
334 }
335 
336 static __ref void unmap_low_page(void *adr)
337 {
338 	if (after_bootmem)
339 		return;
340 
341 	early_iounmap(adr, PAGE_SIZE);
342 }
343 
344 static unsigned long __meminit
345 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
346 	      pgprot_t prot)
347 {
348 	unsigned pages = 0;
349 	unsigned long last_map_addr = end;
350 	int i;
351 
352 	pte_t *pte = pte_page + pte_index(addr);
353 
354 	for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
355 
356 		if (addr >= end) {
357 			if (!after_bootmem) {
358 				for(; i < PTRS_PER_PTE; i++, pte++)
359 					set_pte(pte, __pte(0));
360 			}
361 			break;
362 		}
363 
364 		/*
365 		 * We will re-use the existing mapping.
366 		 * Xen for example has some special requirements, like mapping
367 		 * pagetable pages as RO. So assume someone who pre-setup
368 		 * these mappings are more intelligent.
369 		 */
370 		if (pte_val(*pte)) {
371 			pages++;
372 			continue;
373 		}
374 
375 		if (0)
376 			printk("   pte=%p addr=%lx pte=%016lx\n",
377 			       pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
378 		pages++;
379 		set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
380 		last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
381 	}
382 
383 	update_page_count(PG_LEVEL_4K, pages);
384 
385 	return last_map_addr;
386 }
387 
388 static unsigned long __meminit
389 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
390 		pgprot_t prot)
391 {
392 	pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
393 
394 	return phys_pte_init(pte, address, end, prot);
395 }
396 
397 static unsigned long __meminit
398 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
399 	      unsigned long page_size_mask, pgprot_t prot)
400 {
401 	unsigned long pages = 0;
402 	unsigned long last_map_addr = end;
403 
404 	int i = pmd_index(address);
405 
406 	for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
407 		unsigned long pte_phys;
408 		pmd_t *pmd = pmd_page + pmd_index(address);
409 		pte_t *pte;
410 		pgprot_t new_prot = prot;
411 
412 		if (address >= end) {
413 			if (!after_bootmem) {
414 				for (; i < PTRS_PER_PMD; i++, pmd++)
415 					set_pmd(pmd, __pmd(0));
416 			}
417 			break;
418 		}
419 
420 		if (pmd_val(*pmd)) {
421 			if (!pmd_large(*pmd)) {
422 				spin_lock(&init_mm.page_table_lock);
423 				last_map_addr = phys_pte_update(pmd, address,
424 								end, prot);
425 				spin_unlock(&init_mm.page_table_lock);
426 				continue;
427 			}
428 			/*
429 			 * If we are ok with PG_LEVEL_2M mapping, then we will
430 			 * use the existing mapping,
431 			 *
432 			 * Otherwise, we will split the large page mapping but
433 			 * use the same existing protection bits except for
434 			 * large page, so that we don't violate Intel's TLB
435 			 * Application note (317080) which says, while changing
436 			 * the page sizes, new and old translations should
437 			 * not differ with respect to page frame and
438 			 * attributes.
439 			 */
440 			if (page_size_mask & (1 << PG_LEVEL_2M)) {
441 				pages++;
442 				continue;
443 			}
444 			new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
445 		}
446 
447 		if (page_size_mask & (1<<PG_LEVEL_2M)) {
448 			pages++;
449 			spin_lock(&init_mm.page_table_lock);
450 			set_pte((pte_t *)pmd,
451 				pfn_pte(address >> PAGE_SHIFT,
452 					__pgprot(pgprot_val(prot) | _PAGE_PSE)));
453 			spin_unlock(&init_mm.page_table_lock);
454 			last_map_addr = (address & PMD_MASK) + PMD_SIZE;
455 			continue;
456 		}
457 
458 		pte = alloc_low_page(&pte_phys);
459 		last_map_addr = phys_pte_init(pte, address, end, new_prot);
460 		unmap_low_page(pte);
461 
462 		spin_lock(&init_mm.page_table_lock);
463 		pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
464 		spin_unlock(&init_mm.page_table_lock);
465 	}
466 	update_page_count(PG_LEVEL_2M, pages);
467 	return last_map_addr;
468 }
469 
470 static unsigned long __meminit
471 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
472 		unsigned long page_size_mask, pgprot_t prot)
473 {
474 	pmd_t *pmd = pmd_offset(pud, 0);
475 	unsigned long last_map_addr;
476 
477 	last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
478 	__flush_tlb_all();
479 	return last_map_addr;
480 }
481 
482 static unsigned long __meminit
483 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
484 			 unsigned long page_size_mask)
485 {
486 	unsigned long pages = 0;
487 	unsigned long last_map_addr = end;
488 	int i = pud_index(addr);
489 
490 	for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
491 		unsigned long pmd_phys;
492 		pud_t *pud = pud_page + pud_index(addr);
493 		pmd_t *pmd;
494 		pgprot_t prot = PAGE_KERNEL;
495 
496 		if (addr >= end)
497 			break;
498 
499 		if (!after_bootmem &&
500 				!e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
501 			set_pud(pud, __pud(0));
502 			continue;
503 		}
504 
505 		if (pud_val(*pud)) {
506 			if (!pud_large(*pud)) {
507 				last_map_addr = phys_pmd_update(pud, addr, end,
508 							 page_size_mask, prot);
509 				continue;
510 			}
511 			/*
512 			 * If we are ok with PG_LEVEL_1G mapping, then we will
513 			 * use the existing mapping.
514 			 *
515 			 * Otherwise, we will split the gbpage mapping but use
516 			 * the same existing protection  bits except for large
517 			 * page, so that we don't violate Intel's TLB
518 			 * Application note (317080) which says, while changing
519 			 * the page sizes, new and old translations should
520 			 * not differ with respect to page frame and
521 			 * attributes.
522 			 */
523 			if (page_size_mask & (1 << PG_LEVEL_1G)) {
524 				pages++;
525 				continue;
526 			}
527 			prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
528 		}
529 
530 		if (page_size_mask & (1<<PG_LEVEL_1G)) {
531 			pages++;
532 			spin_lock(&init_mm.page_table_lock);
533 			set_pte((pte_t *)pud,
534 				pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
535 			spin_unlock(&init_mm.page_table_lock);
536 			last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
537 			continue;
538 		}
539 
540 		pmd = alloc_low_page(&pmd_phys);
541 		last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
542 					      prot);
543 		unmap_low_page(pmd);
544 
545 		spin_lock(&init_mm.page_table_lock);
546 		pud_populate(&init_mm, pud, __va(pmd_phys));
547 		spin_unlock(&init_mm.page_table_lock);
548 	}
549 	__flush_tlb_all();
550 
551 	update_page_count(PG_LEVEL_1G, pages);
552 
553 	return last_map_addr;
554 }
555 
556 static unsigned long __meminit
557 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
558 		 unsigned long page_size_mask)
559 {
560 	pud_t *pud;
561 
562 	pud = (pud_t *)pgd_page_vaddr(*pgd);
563 
564 	return phys_pud_init(pud, addr, end, page_size_mask);
565 }
566 
567 unsigned long __meminit
568 kernel_physical_mapping_init(unsigned long start,
569 			     unsigned long end,
570 			     unsigned long page_size_mask)
571 {
572 	bool pgd_changed = false;
573 	unsigned long next, last_map_addr = end;
574 	unsigned long addr;
575 
576 	start = (unsigned long)__va(start);
577 	end = (unsigned long)__va(end);
578 	addr = start;
579 
580 	for (; start < end; start = next) {
581 		pgd_t *pgd = pgd_offset_k(start);
582 		unsigned long pud_phys;
583 		pud_t *pud;
584 
585 		next = (start + PGDIR_SIZE) & PGDIR_MASK;
586 		if (next > end)
587 			next = end;
588 
589 		if (pgd_val(*pgd)) {
590 			last_map_addr = phys_pud_update(pgd, __pa(start),
591 						 __pa(end), page_size_mask);
592 			continue;
593 		}
594 
595 		pud = alloc_low_page(&pud_phys);
596 		last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
597 						 page_size_mask);
598 		unmap_low_page(pud);
599 
600 		spin_lock(&init_mm.page_table_lock);
601 		pgd_populate(&init_mm, pgd, __va(pud_phys));
602 		spin_unlock(&init_mm.page_table_lock);
603 		pgd_changed = true;
604 	}
605 
606 	if (pgd_changed)
607 		sync_global_pgds(addr, end);
608 
609 	__flush_tlb_all();
610 
611 	return last_map_addr;
612 }
613 
614 #ifndef CONFIG_NUMA
615 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
616 				int acpi, int k8)
617 {
618 	memblock_x86_register_active_regions(0, start_pfn, end_pfn);
619 }
620 #endif
621 
622 void __init paging_init(void)
623 {
624 	unsigned long max_zone_pfns[MAX_NR_ZONES];
625 
626 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
627 	max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
628 	max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
629 	max_zone_pfns[ZONE_NORMAL] = max_pfn;
630 
631 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
632 	sparse_init();
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 	 */
640 	node_clear_state(0, N_NORMAL_MEMORY);
641 
642 	free_area_init_nodes(max_zone_pfns);
643 }
644 
645 /*
646  * Memory hotplug specific functions
647  */
648 #ifdef CONFIG_MEMORY_HOTPLUG
649 /*
650  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
651  * updating.
652  */
653 static void  update_end_of_memory_vars(u64 start, u64 size)
654 {
655 	unsigned long end_pfn = PFN_UP(start + size);
656 
657 	if (end_pfn > max_pfn) {
658 		max_pfn = end_pfn;
659 		max_low_pfn = end_pfn;
660 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
661 	}
662 }
663 
664 /*
665  * Memory is added always to NORMAL zone. This means you will never get
666  * additional DMA/DMA32 memory.
667  */
668 int arch_add_memory(int nid, u64 start, u64 size)
669 {
670 	struct pglist_data *pgdat = NODE_DATA(nid);
671 	struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
672 	unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
673 	unsigned long nr_pages = size >> PAGE_SHIFT;
674 	int ret;
675 
676 	last_mapped_pfn = init_memory_mapping(start, start + size);
677 	if (last_mapped_pfn > max_pfn_mapped)
678 		max_pfn_mapped = last_mapped_pfn;
679 
680 	ret = __add_pages(nid, zone, start_pfn, nr_pages);
681 	WARN_ON_ONCE(ret);
682 
683 	/* update max_pfn, max_low_pfn and high_memory */
684 	update_end_of_memory_vars(start, size);
685 
686 	return ret;
687 }
688 EXPORT_SYMBOL_GPL(arch_add_memory);
689 
690 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
691 int memory_add_physaddr_to_nid(u64 start)
692 {
693 	return 0;
694 }
695 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
696 #endif
697 
698 #endif /* CONFIG_MEMORY_HOTPLUG */
699 
700 static struct kcore_list kcore_vsyscall;
701 
702 void __init mem_init(void)
703 {
704 	long codesize, reservedpages, datasize, initsize;
705 	unsigned long absent_pages;
706 
707 	pci_iommu_alloc();
708 
709 	/* clear_bss() already clear the empty_zero_page */
710 
711 	reservedpages = 0;
712 
713 	/* this will put all low memory onto the freelists */
714 #ifdef CONFIG_NUMA
715 	totalram_pages = numa_free_all_bootmem();
716 #else
717 	totalram_pages = free_all_bootmem();
718 #endif
719 
720 	absent_pages = absent_pages_in_range(0, max_pfn);
721 	reservedpages = max_pfn - totalram_pages - absent_pages;
722 	after_bootmem = 1;
723 
724 	codesize =  (unsigned long) &_etext - (unsigned long) &_text;
725 	datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
726 	initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
727 
728 	/* Register memory areas for /proc/kcore */
729 	kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
730 			 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
731 
732 	printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
733 			 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
734 		nr_free_pages() << (PAGE_SHIFT-10),
735 		max_pfn << (PAGE_SHIFT-10),
736 		codesize >> 10,
737 		absent_pages << (PAGE_SHIFT-10),
738 		reservedpages << (PAGE_SHIFT-10),
739 		datasize >> 10,
740 		initsize >> 10);
741 }
742 
743 #ifdef CONFIG_DEBUG_RODATA
744 const int rodata_test_data = 0xC3;
745 EXPORT_SYMBOL_GPL(rodata_test_data);
746 
747 int kernel_set_to_readonly;
748 
749 void set_kernel_text_rw(void)
750 {
751 	unsigned long start = PFN_ALIGN(_text);
752 	unsigned long end = PFN_ALIGN(__stop___ex_table);
753 
754 	if (!kernel_set_to_readonly)
755 		return;
756 
757 	pr_debug("Set kernel text: %lx - %lx for read write\n",
758 		 start, end);
759 
760 	/*
761 	 * Make the kernel identity mapping for text RW. Kernel text
762 	 * mapping will always be RO. Refer to the comment in
763 	 * static_protections() in pageattr.c
764 	 */
765 	set_memory_rw(start, (end - start) >> PAGE_SHIFT);
766 }
767 
768 void set_kernel_text_ro(void)
769 {
770 	unsigned long start = PFN_ALIGN(_text);
771 	unsigned long end = PFN_ALIGN(__stop___ex_table);
772 
773 	if (!kernel_set_to_readonly)
774 		return;
775 
776 	pr_debug("Set kernel text: %lx - %lx for read only\n",
777 		 start, end);
778 
779 	/*
780 	 * Set the kernel identity mapping for text RO.
781 	 */
782 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
783 }
784 
785 void mark_rodata_ro(void)
786 {
787 	unsigned long start = PFN_ALIGN(_text);
788 	unsigned long rodata_start =
789 		((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
790 	unsigned long end = (unsigned long) &__end_rodata_hpage_align;
791 	unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
792 	unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
793 	unsigned long data_start = (unsigned long) &_sdata;
794 
795 	printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
796 	       (end - start) >> 10);
797 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
798 
799 	kernel_set_to_readonly = 1;
800 
801 	/*
802 	 * The rodata section (but not the kernel text!) should also be
803 	 * not-executable.
804 	 */
805 	set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
806 
807 	rodata_test();
808 
809 #ifdef CONFIG_CPA_DEBUG
810 	printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
811 	set_memory_rw(start, (end-start) >> PAGE_SHIFT);
812 
813 	printk(KERN_INFO "Testing CPA: again\n");
814 	set_memory_ro(start, (end-start) >> PAGE_SHIFT);
815 #endif
816 
817 	free_init_pages("unused kernel memory",
818 			(unsigned long) page_address(virt_to_page(text_end)),
819 			(unsigned long)
820 				 page_address(virt_to_page(rodata_start)));
821 	free_init_pages("unused kernel memory",
822 			(unsigned long) page_address(virt_to_page(rodata_end)),
823 			(unsigned long) page_address(virt_to_page(data_start)));
824 }
825 
826 #endif
827 
828 int kern_addr_valid(unsigned long addr)
829 {
830 	unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
831 	pgd_t *pgd;
832 	pud_t *pud;
833 	pmd_t *pmd;
834 	pte_t *pte;
835 
836 	if (above != 0 && above != -1UL)
837 		return 0;
838 
839 	pgd = pgd_offset_k(addr);
840 	if (pgd_none(*pgd))
841 		return 0;
842 
843 	pud = pud_offset(pgd, addr);
844 	if (pud_none(*pud))
845 		return 0;
846 
847 	pmd = pmd_offset(pud, addr);
848 	if (pmd_none(*pmd))
849 		return 0;
850 
851 	if (pmd_large(*pmd))
852 		return pfn_valid(pmd_pfn(*pmd));
853 
854 	pte = pte_offset_kernel(pmd, addr);
855 	if (pte_none(*pte))
856 		return 0;
857 
858 	return pfn_valid(pte_pfn(*pte));
859 }
860 
861 /*
862  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
863  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
864  * not need special handling anymore:
865  */
866 static struct vm_area_struct gate_vma = {
867 	.vm_start	= VSYSCALL_START,
868 	.vm_end		= VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
869 	.vm_page_prot	= PAGE_READONLY_EXEC,
870 	.vm_flags	= VM_READ | VM_EXEC
871 };
872 
873 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
874 {
875 #ifdef CONFIG_IA32_EMULATION
876 	if (test_tsk_thread_flag(tsk, TIF_IA32))
877 		return NULL;
878 #endif
879 	return &gate_vma;
880 }
881 
882 int in_gate_area(struct task_struct *task, unsigned long addr)
883 {
884 	struct vm_area_struct *vma = get_gate_vma(task);
885 
886 	if (!vma)
887 		return 0;
888 
889 	return (addr >= vma->vm_start) && (addr < vma->vm_end);
890 }
891 
892 /*
893  * Use this when you have no reliable task/vma, typically from interrupt
894  * context. It is less reliable than using the task's vma and may give
895  * false positives:
896  */
897 int in_gate_area_no_task(unsigned long addr)
898 {
899 	return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
900 }
901 
902 const char *arch_vma_name(struct vm_area_struct *vma)
903 {
904 	if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
905 		return "[vdso]";
906 	if (vma == &gate_vma)
907 		return "[vsyscall]";
908 	return NULL;
909 }
910 
911 #ifdef CONFIG_SPARSEMEM_VMEMMAP
912 /*
913  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
914  */
915 static long __meminitdata addr_start, addr_end;
916 static void __meminitdata *p_start, *p_end;
917 static int __meminitdata node_start;
918 
919 int __meminit
920 vmemmap_populate(struct page *start_page, unsigned long size, int node)
921 {
922 	unsigned long addr = (unsigned long)start_page;
923 	unsigned long end = (unsigned long)(start_page + size);
924 	unsigned long next;
925 	pgd_t *pgd;
926 	pud_t *pud;
927 	pmd_t *pmd;
928 
929 	for (; addr < end; addr = next) {
930 		void *p = NULL;
931 
932 		pgd = vmemmap_pgd_populate(addr, node);
933 		if (!pgd)
934 			return -ENOMEM;
935 
936 		pud = vmemmap_pud_populate(pgd, addr, node);
937 		if (!pud)
938 			return -ENOMEM;
939 
940 		if (!cpu_has_pse) {
941 			next = (addr + PAGE_SIZE) & PAGE_MASK;
942 			pmd = vmemmap_pmd_populate(pud, addr, node);
943 
944 			if (!pmd)
945 				return -ENOMEM;
946 
947 			p = vmemmap_pte_populate(pmd, addr, node);
948 
949 			if (!p)
950 				return -ENOMEM;
951 
952 			addr_end = addr + PAGE_SIZE;
953 			p_end = p + PAGE_SIZE;
954 		} else {
955 			next = pmd_addr_end(addr, end);
956 
957 			pmd = pmd_offset(pud, addr);
958 			if (pmd_none(*pmd)) {
959 				pte_t entry;
960 
961 				p = vmemmap_alloc_block_buf(PMD_SIZE, node);
962 				if (!p)
963 					return -ENOMEM;
964 
965 				entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
966 						PAGE_KERNEL_LARGE);
967 				set_pmd(pmd, __pmd(pte_val(entry)));
968 
969 				/* check to see if we have contiguous blocks */
970 				if (p_end != p || node_start != node) {
971 					if (p_start)
972 						printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
973 						       addr_start, addr_end-1, p_start, p_end-1, node_start);
974 					addr_start = addr;
975 					node_start = node;
976 					p_start = p;
977 				}
978 
979 				addr_end = addr + PMD_SIZE;
980 				p_end = p + PMD_SIZE;
981 			} else
982 				vmemmap_verify((pte_t *)pmd, node, addr, next);
983 		}
984 
985 	}
986 	sync_global_pgds((unsigned long)start_page, end);
987 	return 0;
988 }
989 
990 void __meminit vmemmap_populate_print_last(void)
991 {
992 	if (p_start) {
993 		printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
994 			addr_start, addr_end-1, p_start, p_end-1, node_start);
995 		p_start = NULL;
996 		p_end = NULL;
997 		node_start = 0;
998 	}
999 }
1000 #endif
1001