xref: /openbmc/linux/arch/x86/mm/init_64.c (revision 82ced6fd)
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 
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/proc_fs.h>
25 #include <linux/pci.h>
26 #include <linux/pfn.h>
27 #include <linux/poison.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/module.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/nmi.h>
32 
33 #include <asm/processor.h>
34 #include <asm/bios_ebda.h>
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>
47 #include <asm/sections.h>
48 #include <asm/kdebug.h>
49 #include <asm/numa.h>
50 #include <asm/cacheflush.h>
51 #include <asm/init.h>
52 
53 /*
54  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
55  * The direct mapping extends to max_pfn_mapped, so that we can directly access
56  * apertures, ACPI and other tables without having to play with fixmaps.
57  */
58 unsigned long max_low_pfn_mapped;
59 unsigned long max_pfn_mapped;
60 
61 static unsigned long dma_reserve __initdata;
62 
63 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
64 
65 static int __init parse_direct_gbpages_off(char *arg)
66 {
67 	direct_gbpages = 0;
68 	return 0;
69 }
70 early_param("nogbpages", parse_direct_gbpages_off);
71 
72 static int __init parse_direct_gbpages_on(char *arg)
73 {
74 	direct_gbpages = 1;
75 	return 0;
76 }
77 early_param("gbpages", parse_direct_gbpages_on);
78 
79 /*
80  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
81  * physical space so we can cache the place of the first one and move
82  * around without checking the pgd every time.
83  */
84 
85 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
86 EXPORT_SYMBOL_GPL(__supported_pte_mask);
87 
88 static int disable_nx __cpuinitdata;
89 
90 /*
91  * noexec=on|off
92  * Control non-executable mappings for 64-bit processes.
93  *
94  * on	Enable (default)
95  * off	Disable
96  */
97 static int __init nonx_setup(char *str)
98 {
99 	if (!str)
100 		return -EINVAL;
101 	if (!strncmp(str, "on", 2)) {
102 		__supported_pte_mask |= _PAGE_NX;
103 		disable_nx = 0;
104 	} else if (!strncmp(str, "off", 3)) {
105 		disable_nx = 1;
106 		__supported_pte_mask &= ~_PAGE_NX;
107 	}
108 	return 0;
109 }
110 early_param("noexec", nonx_setup);
111 
112 void __cpuinit check_efer(void)
113 {
114 	unsigned long efer;
115 
116 	rdmsrl(MSR_EFER, efer);
117 	if (!(efer & EFER_NX) || disable_nx)
118 		__supported_pte_mask &= ~_PAGE_NX;
119 }
120 
121 int force_personality32;
122 
123 /*
124  * noexec32=on|off
125  * Control non executable heap for 32bit processes.
126  * To control the stack too use noexec=off
127  *
128  * on	PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
129  * off	PROT_READ implies PROT_EXEC
130  */
131 static int __init nonx32_setup(char *str)
132 {
133 	if (!strcmp(str, "on"))
134 		force_personality32 &= ~READ_IMPLIES_EXEC;
135 	else if (!strcmp(str, "off"))
136 		force_personality32 |= READ_IMPLIES_EXEC;
137 	return 1;
138 }
139 __setup("noexec32=", nonx32_setup);
140 
141 /*
142  * NOTE: This function is marked __ref because it calls __init function
143  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
144  */
145 static __ref void *spp_getpage(void)
146 {
147 	void *ptr;
148 
149 	if (after_bootmem)
150 		ptr = (void *) get_zeroed_page(GFP_ATOMIC);
151 	else
152 		ptr = alloc_bootmem_pages(PAGE_SIZE);
153 
154 	if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
155 		panic("set_pte_phys: cannot allocate page data %s\n",
156 			after_bootmem ? "after bootmem" : "");
157 	}
158 
159 	pr_debug("spp_getpage %p\n", ptr);
160 
161 	return ptr;
162 }
163 
164 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
165 {
166 	if (pgd_none(*pgd)) {
167 		pud_t *pud = (pud_t *)spp_getpage();
168 		pgd_populate(&init_mm, pgd, pud);
169 		if (pud != pud_offset(pgd, 0))
170 			printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
171 			       pud, pud_offset(pgd, 0));
172 	}
173 	return pud_offset(pgd, vaddr);
174 }
175 
176 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
177 {
178 	if (pud_none(*pud)) {
179 		pmd_t *pmd = (pmd_t *) spp_getpage();
180 		pud_populate(&init_mm, pud, pmd);
181 		if (pmd != pmd_offset(pud, 0))
182 			printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
183 			       pmd, pmd_offset(pud, 0));
184 	}
185 	return pmd_offset(pud, vaddr);
186 }
187 
188 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
189 {
190 	if (pmd_none(*pmd)) {
191 		pte_t *pte = (pte_t *) spp_getpage();
192 		pmd_populate_kernel(&init_mm, pmd, pte);
193 		if (pte != pte_offset_kernel(pmd, 0))
194 			printk(KERN_ERR "PAGETABLE BUG #02!\n");
195 	}
196 	return pte_offset_kernel(pmd, vaddr);
197 }
198 
199 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
200 {
201 	pud_t *pud;
202 	pmd_t *pmd;
203 	pte_t *pte;
204 
205 	pud = pud_page + pud_index(vaddr);
206 	pmd = fill_pmd(pud, vaddr);
207 	pte = fill_pte(pmd, vaddr);
208 
209 	set_pte(pte, new_pte);
210 
211 	/*
212 	 * It's enough to flush this one mapping.
213 	 * (PGE mappings get flushed as well)
214 	 */
215 	__flush_tlb_one(vaddr);
216 }
217 
218 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
219 {
220 	pgd_t *pgd;
221 	pud_t *pud_page;
222 
223 	pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
224 
225 	pgd = pgd_offset_k(vaddr);
226 	if (pgd_none(*pgd)) {
227 		printk(KERN_ERR
228 			"PGD FIXMAP MISSING, it should be setup in head.S!\n");
229 		return;
230 	}
231 	pud_page = (pud_t*)pgd_page_vaddr(*pgd);
232 	set_pte_vaddr_pud(pud_page, vaddr, pteval);
233 }
234 
235 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
236 {
237 	pgd_t *pgd;
238 	pud_t *pud;
239 
240 	pgd = pgd_offset_k(vaddr);
241 	pud = fill_pud(pgd, vaddr);
242 	return fill_pmd(pud, vaddr);
243 }
244 
245 pte_t * __init populate_extra_pte(unsigned long vaddr)
246 {
247 	pmd_t *pmd;
248 
249 	pmd = populate_extra_pmd(vaddr);
250 	return fill_pte(pmd, vaddr);
251 }
252 
253 /*
254  * Create large page table mappings for a range of physical addresses.
255  */
256 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
257 						pgprot_t prot)
258 {
259 	pgd_t *pgd;
260 	pud_t *pud;
261 	pmd_t *pmd;
262 
263 	BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
264 	for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
265 		pgd = pgd_offset_k((unsigned long)__va(phys));
266 		if (pgd_none(*pgd)) {
267 			pud = (pud_t *) spp_getpage();
268 			set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
269 						_PAGE_USER));
270 		}
271 		pud = pud_offset(pgd, (unsigned long)__va(phys));
272 		if (pud_none(*pud)) {
273 			pmd = (pmd_t *) spp_getpage();
274 			set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
275 						_PAGE_USER));
276 		}
277 		pmd = pmd_offset(pud, phys);
278 		BUG_ON(!pmd_none(*pmd));
279 		set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
280 	}
281 }
282 
283 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
284 {
285 	__init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
286 }
287 
288 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
289 {
290 	__init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
291 }
292 
293 /*
294  * The head.S code sets up the kernel high mapping:
295  *
296  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
297  *
298  * phys_addr holds the negative offset to the kernel, which is added
299  * to the compile time generated pmds. This results in invalid pmds up
300  * to the point where we hit the physaddr 0 mapping.
301  *
302  * We limit the mappings to the region from _text to _end.  _end is
303  * rounded up to the 2MB boundary. This catches the invalid pmds as
304  * well, as they are located before _text:
305  */
306 void __init cleanup_highmap(void)
307 {
308 	unsigned long vaddr = __START_KERNEL_map;
309 	unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
310 	pmd_t *pmd = level2_kernel_pgt;
311 	pmd_t *last_pmd = pmd + PTRS_PER_PMD;
312 
313 	for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
314 		if (pmd_none(*pmd))
315 			continue;
316 		if (vaddr < (unsigned long) _text || vaddr > end)
317 			set_pmd(pmd, __pmd(0));
318 	}
319 }
320 
321 static __ref void *alloc_low_page(unsigned long *phys)
322 {
323 	unsigned long pfn = e820_table_end++;
324 	void *adr;
325 
326 	if (after_bootmem) {
327 		adr = (void *)get_zeroed_page(GFP_ATOMIC);
328 		*phys = __pa(adr);
329 
330 		return adr;
331 	}
332 
333 	if (pfn >= e820_table_top)
334 		panic("alloc_low_page: ran out of memory");
335 
336 	adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
337 	memset(adr, 0, PAGE_SIZE);
338 	*phys  = pfn * PAGE_SIZE;
339 	return adr;
340 }
341 
342 static __ref void unmap_low_page(void *adr)
343 {
344 	if (after_bootmem)
345 		return;
346 
347 	early_iounmap(adr, PAGE_SIZE);
348 }
349 
350 static unsigned long __meminit
351 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
352 	      pgprot_t prot)
353 {
354 	unsigned pages = 0;
355 	unsigned long last_map_addr = end;
356 	int i;
357 
358 	pte_t *pte = pte_page + pte_index(addr);
359 
360 	for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
361 
362 		if (addr >= end) {
363 			if (!after_bootmem) {
364 				for(; i < PTRS_PER_PTE; i++, pte++)
365 					set_pte(pte, __pte(0));
366 			}
367 			break;
368 		}
369 
370 		/*
371 		 * We will re-use the existing mapping.
372 		 * Xen for example has some special requirements, like mapping
373 		 * pagetable pages as RO. So assume someone who pre-setup
374 		 * these mappings are more intelligent.
375 		 */
376 		if (pte_val(*pte)) {
377 			pages++;
378 			continue;
379 		}
380 
381 		if (0)
382 			printk("   pte=%p addr=%lx pte=%016lx\n",
383 			       pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
384 		pages++;
385 		set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
386 		last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
387 	}
388 
389 	update_page_count(PG_LEVEL_4K, pages);
390 
391 	return last_map_addr;
392 }
393 
394 static unsigned long __meminit
395 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
396 		pgprot_t prot)
397 {
398 	pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
399 
400 	return phys_pte_init(pte, address, end, prot);
401 }
402 
403 static unsigned long __meminit
404 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
405 	      unsigned long page_size_mask, pgprot_t prot)
406 {
407 	unsigned long pages = 0;
408 	unsigned long last_map_addr = end;
409 
410 	int i = pmd_index(address);
411 
412 	for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
413 		unsigned long pte_phys;
414 		pmd_t *pmd = pmd_page + pmd_index(address);
415 		pte_t *pte;
416 		pgprot_t new_prot = prot;
417 
418 		if (address >= end) {
419 			if (!after_bootmem) {
420 				for (; i < PTRS_PER_PMD; i++, pmd++)
421 					set_pmd(pmd, __pmd(0));
422 			}
423 			break;
424 		}
425 
426 		if (pmd_val(*pmd)) {
427 			if (!pmd_large(*pmd)) {
428 				spin_lock(&init_mm.page_table_lock);
429 				last_map_addr = phys_pte_update(pmd, address,
430 								end, prot);
431 				spin_unlock(&init_mm.page_table_lock);
432 				continue;
433 			}
434 			/*
435 			 * If we are ok with PG_LEVEL_2M mapping, then we will
436 			 * use the existing mapping,
437 			 *
438 			 * Otherwise, we will split the large page mapping but
439 			 * use the same existing protection bits except for
440 			 * large page, so that we don't violate Intel's TLB
441 			 * Application note (317080) which says, while changing
442 			 * the page sizes, new and old translations should
443 			 * not differ with respect to page frame and
444 			 * attributes.
445 			 */
446 			if (page_size_mask & (1 << PG_LEVEL_2M)) {
447 				pages++;
448 				continue;
449 			}
450 			new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
451 		}
452 
453 		if (page_size_mask & (1<<PG_LEVEL_2M)) {
454 			pages++;
455 			spin_lock(&init_mm.page_table_lock);
456 			set_pte((pte_t *)pmd,
457 				pfn_pte(address >> PAGE_SHIFT,
458 					__pgprot(pgprot_val(prot) | _PAGE_PSE)));
459 			spin_unlock(&init_mm.page_table_lock);
460 			last_map_addr = (address & PMD_MASK) + PMD_SIZE;
461 			continue;
462 		}
463 
464 		pte = alloc_low_page(&pte_phys);
465 		last_map_addr = phys_pte_init(pte, address, end, new_prot);
466 		unmap_low_page(pte);
467 
468 		spin_lock(&init_mm.page_table_lock);
469 		pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
470 		spin_unlock(&init_mm.page_table_lock);
471 	}
472 	update_page_count(PG_LEVEL_2M, pages);
473 	return last_map_addr;
474 }
475 
476 static unsigned long __meminit
477 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
478 		unsigned long page_size_mask, pgprot_t prot)
479 {
480 	pmd_t *pmd = pmd_offset(pud, 0);
481 	unsigned long last_map_addr;
482 
483 	last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
484 	__flush_tlb_all();
485 	return last_map_addr;
486 }
487 
488 static unsigned long __meminit
489 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
490 			 unsigned long page_size_mask)
491 {
492 	unsigned long pages = 0;
493 	unsigned long last_map_addr = end;
494 	int i = pud_index(addr);
495 
496 	for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
497 		unsigned long pmd_phys;
498 		pud_t *pud = pud_page + pud_index(addr);
499 		pmd_t *pmd;
500 		pgprot_t prot = PAGE_KERNEL;
501 
502 		if (addr >= end)
503 			break;
504 
505 		if (!after_bootmem &&
506 				!e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
507 			set_pud(pud, __pud(0));
508 			continue;
509 		}
510 
511 		if (pud_val(*pud)) {
512 			if (!pud_large(*pud)) {
513 				last_map_addr = phys_pmd_update(pud, addr, end,
514 							 page_size_mask, prot);
515 				continue;
516 			}
517 			/*
518 			 * If we are ok with PG_LEVEL_1G mapping, then we will
519 			 * use the existing mapping.
520 			 *
521 			 * Otherwise, we will split the gbpage mapping but use
522 			 * the same existing protection  bits except for large
523 			 * page, so that we don't violate Intel's TLB
524 			 * Application note (317080) which says, while changing
525 			 * the page sizes, new and old translations should
526 			 * not differ with respect to page frame and
527 			 * attributes.
528 			 */
529 			if (page_size_mask & (1 << PG_LEVEL_1G)) {
530 				pages++;
531 				continue;
532 			}
533 			prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
534 		}
535 
536 		if (page_size_mask & (1<<PG_LEVEL_1G)) {
537 			pages++;
538 			spin_lock(&init_mm.page_table_lock);
539 			set_pte((pte_t *)pud,
540 				pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
541 			spin_unlock(&init_mm.page_table_lock);
542 			last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
543 			continue;
544 		}
545 
546 		pmd = alloc_low_page(&pmd_phys);
547 		last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
548 					      prot);
549 		unmap_low_page(pmd);
550 
551 		spin_lock(&init_mm.page_table_lock);
552 		pud_populate(&init_mm, pud, __va(pmd_phys));
553 		spin_unlock(&init_mm.page_table_lock);
554 	}
555 	__flush_tlb_all();
556 
557 	update_page_count(PG_LEVEL_1G, pages);
558 
559 	return last_map_addr;
560 }
561 
562 static unsigned long __meminit
563 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
564 		 unsigned long page_size_mask)
565 {
566 	pud_t *pud;
567 
568 	pud = (pud_t *)pgd_page_vaddr(*pgd);
569 
570 	return phys_pud_init(pud, addr, end, page_size_mask);
571 }
572 
573 unsigned long __init
574 kernel_physical_mapping_init(unsigned long start,
575 			     unsigned long end,
576 			     unsigned long page_size_mask)
577 {
578 
579 	unsigned long next, last_map_addr = end;
580 
581 	start = (unsigned long)__va(start);
582 	end = (unsigned long)__va(end);
583 
584 	for (; start < end; start = next) {
585 		pgd_t *pgd = pgd_offset_k(start);
586 		unsigned long pud_phys;
587 		pud_t *pud;
588 
589 		next = (start + PGDIR_SIZE) & PGDIR_MASK;
590 		if (next > end)
591 			next = end;
592 
593 		if (pgd_val(*pgd)) {
594 			last_map_addr = phys_pud_update(pgd, __pa(start),
595 						 __pa(end), page_size_mask);
596 			continue;
597 		}
598 
599 		pud = alloc_low_page(&pud_phys);
600 		last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
601 						 page_size_mask);
602 		unmap_low_page(pud);
603 
604 		spin_lock(&init_mm.page_table_lock);
605 		pgd_populate(&init_mm, pgd, __va(pud_phys));
606 		spin_unlock(&init_mm.page_table_lock);
607 	}
608 	__flush_tlb_all();
609 
610 	return last_map_addr;
611 }
612 
613 #ifndef CONFIG_NUMA
614 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn)
615 {
616 	unsigned long bootmap_size, bootmap;
617 
618 	bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
619 	bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
620 				 PAGE_SIZE);
621 	if (bootmap == -1L)
622 		panic("Cannot find bootmem map of size %ld\n", bootmap_size);
623 	/* don't touch min_low_pfn */
624 	bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
625 					 0, end_pfn);
626 	e820_register_active_regions(0, start_pfn, end_pfn);
627 	free_bootmem_with_active_regions(0, end_pfn);
628 	early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
629 	reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
630 }
631 
632 void __init paging_init(void)
633 {
634 	unsigned long max_zone_pfns[MAX_NR_ZONES];
635 
636 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
637 	max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
638 	max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
639 	max_zone_pfns[ZONE_NORMAL] = max_pfn;
640 
641 	memory_present(0, 0, max_pfn);
642 	sparse_init();
643 	free_area_init_nodes(max_zone_pfns);
644 }
645 #endif
646 
647 /*
648  * Memory hotplug specific functions
649  */
650 #ifdef CONFIG_MEMORY_HOTPLUG
651 /*
652  * Memory is added always to NORMAL zone. This means you will never get
653  * additional DMA/DMA32 memory.
654  */
655 int arch_add_memory(int nid, u64 start, u64 size)
656 {
657 	struct pglist_data *pgdat = NODE_DATA(nid);
658 	struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
659 	unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
660 	unsigned long nr_pages = size >> PAGE_SHIFT;
661 	int ret;
662 
663 	last_mapped_pfn = init_memory_mapping(start, start + size);
664 	if (last_mapped_pfn > max_pfn_mapped)
665 		max_pfn_mapped = last_mapped_pfn;
666 
667 	ret = __add_pages(nid, zone, start_pfn, nr_pages);
668 	WARN_ON_ONCE(ret);
669 
670 	return ret;
671 }
672 EXPORT_SYMBOL_GPL(arch_add_memory);
673 
674 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
675 int memory_add_physaddr_to_nid(u64 start)
676 {
677 	return 0;
678 }
679 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
680 #endif
681 
682 #endif /* CONFIG_MEMORY_HOTPLUG */
683 
684 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
685 			 kcore_modules, kcore_vsyscall;
686 
687 void __init mem_init(void)
688 {
689 	long codesize, reservedpages, datasize, initsize;
690 	unsigned long absent_pages;
691 
692 	pci_iommu_alloc();
693 
694 	/* clear_bss() already clear the empty_zero_page */
695 
696 	reservedpages = 0;
697 
698 	/* this will put all low memory onto the freelists */
699 #ifdef CONFIG_NUMA
700 	totalram_pages = numa_free_all_bootmem();
701 #else
702 	totalram_pages = free_all_bootmem();
703 #endif
704 
705 	absent_pages = absent_pages_in_range(0, max_pfn);
706 	reservedpages = max_pfn - totalram_pages - absent_pages;
707 	after_bootmem = 1;
708 
709 	codesize =  (unsigned long) &_etext - (unsigned long) &_text;
710 	datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
711 	initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
712 
713 	/* Register memory areas for /proc/kcore */
714 	kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
715 	kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
716 		   VMALLOC_END-VMALLOC_START);
717 	kclist_add(&kcore_kernel, &_stext, _end - _stext);
718 	kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
719 	kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
720 				 VSYSCALL_END - VSYSCALL_START);
721 
722 	printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
723 			 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
724 		(unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
725 		max_pfn << (PAGE_SHIFT-10),
726 		codesize >> 10,
727 		absent_pages << (PAGE_SHIFT-10),
728 		reservedpages << (PAGE_SHIFT-10),
729 		datasize >> 10,
730 		initsize >> 10);
731 }
732 
733 #ifdef CONFIG_DEBUG_RODATA
734 const int rodata_test_data = 0xC3;
735 EXPORT_SYMBOL_GPL(rodata_test_data);
736 
737 static int kernel_set_to_readonly;
738 
739 void set_kernel_text_rw(void)
740 {
741 	unsigned long start = PFN_ALIGN(_stext);
742 	unsigned long end = PFN_ALIGN(__start_rodata);
743 
744 	if (!kernel_set_to_readonly)
745 		return;
746 
747 	pr_debug("Set kernel text: %lx - %lx for read write\n",
748 		 start, end);
749 
750 	set_memory_rw(start, (end - start) >> PAGE_SHIFT);
751 }
752 
753 void set_kernel_text_ro(void)
754 {
755 	unsigned long start = PFN_ALIGN(_stext);
756 	unsigned long end = PFN_ALIGN(__start_rodata);
757 
758 	if (!kernel_set_to_readonly)
759 		return;
760 
761 	pr_debug("Set kernel text: %lx - %lx for read only\n",
762 		 start, end);
763 
764 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
765 }
766 
767 void mark_rodata_ro(void)
768 {
769 	unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
770 	unsigned long rodata_start =
771 		((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
772 
773 	printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
774 	       (end - start) >> 10);
775 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
776 
777 	kernel_set_to_readonly = 1;
778 
779 	/*
780 	 * The rodata section (but not the kernel text!) should also be
781 	 * not-executable.
782 	 */
783 	set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
784 
785 	rodata_test();
786 
787 #ifdef CONFIG_CPA_DEBUG
788 	printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
789 	set_memory_rw(start, (end-start) >> PAGE_SHIFT);
790 
791 	printk(KERN_INFO "Testing CPA: again\n");
792 	set_memory_ro(start, (end-start) >> PAGE_SHIFT);
793 #endif
794 }
795 
796 #endif
797 
798 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
799 				   int flags)
800 {
801 #ifdef CONFIG_NUMA
802 	int nid, next_nid;
803 	int ret;
804 #endif
805 	unsigned long pfn = phys >> PAGE_SHIFT;
806 
807 	if (pfn >= max_pfn) {
808 		/*
809 		 * This can happen with kdump kernels when accessing
810 		 * firmware tables:
811 		 */
812 		if (pfn < max_pfn_mapped)
813 			return -EFAULT;
814 
815 		printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
816 				phys, len);
817 		return -EFAULT;
818 	}
819 
820 	/* Should check here against the e820 map to avoid double free */
821 #ifdef CONFIG_NUMA
822 	nid = phys_to_nid(phys);
823 	next_nid = phys_to_nid(phys + len - 1);
824 	if (nid == next_nid)
825 		ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
826 	else
827 		ret = reserve_bootmem(phys, len, flags);
828 
829 	if (ret != 0)
830 		return ret;
831 
832 #else
833 	reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
834 #endif
835 
836 	if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
837 		dma_reserve += len / PAGE_SIZE;
838 		set_dma_reserve(dma_reserve);
839 	}
840 
841 	return 0;
842 }
843 
844 int kern_addr_valid(unsigned long addr)
845 {
846 	unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
847 	pgd_t *pgd;
848 	pud_t *pud;
849 	pmd_t *pmd;
850 	pte_t *pte;
851 
852 	if (above != 0 && above != -1UL)
853 		return 0;
854 
855 	pgd = pgd_offset_k(addr);
856 	if (pgd_none(*pgd))
857 		return 0;
858 
859 	pud = pud_offset(pgd, addr);
860 	if (pud_none(*pud))
861 		return 0;
862 
863 	pmd = pmd_offset(pud, addr);
864 	if (pmd_none(*pmd))
865 		return 0;
866 
867 	if (pmd_large(*pmd))
868 		return pfn_valid(pmd_pfn(*pmd));
869 
870 	pte = pte_offset_kernel(pmd, addr);
871 	if (pte_none(*pte))
872 		return 0;
873 
874 	return pfn_valid(pte_pfn(*pte));
875 }
876 
877 /*
878  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
879  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
880  * not need special handling anymore:
881  */
882 static struct vm_area_struct gate_vma = {
883 	.vm_start	= VSYSCALL_START,
884 	.vm_end		= VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
885 	.vm_page_prot	= PAGE_READONLY_EXEC,
886 	.vm_flags	= VM_READ | VM_EXEC
887 };
888 
889 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
890 {
891 #ifdef CONFIG_IA32_EMULATION
892 	if (test_tsk_thread_flag(tsk, TIF_IA32))
893 		return NULL;
894 #endif
895 	return &gate_vma;
896 }
897 
898 int in_gate_area(struct task_struct *task, unsigned long addr)
899 {
900 	struct vm_area_struct *vma = get_gate_vma(task);
901 
902 	if (!vma)
903 		return 0;
904 
905 	return (addr >= vma->vm_start) && (addr < vma->vm_end);
906 }
907 
908 /*
909  * Use this when you have no reliable task/vma, typically from interrupt
910  * context. It is less reliable than using the task's vma and may give
911  * false positives:
912  */
913 int in_gate_area_no_task(unsigned long addr)
914 {
915 	return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
916 }
917 
918 const char *arch_vma_name(struct vm_area_struct *vma)
919 {
920 	if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
921 		return "[vdso]";
922 	if (vma == &gate_vma)
923 		return "[vsyscall]";
924 	return NULL;
925 }
926 
927 #ifdef CONFIG_SPARSEMEM_VMEMMAP
928 /*
929  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
930  */
931 static long __meminitdata addr_start, addr_end;
932 static void __meminitdata *p_start, *p_end;
933 static int __meminitdata node_start;
934 
935 int __meminit
936 vmemmap_populate(struct page *start_page, unsigned long size, int node)
937 {
938 	unsigned long addr = (unsigned long)start_page;
939 	unsigned long end = (unsigned long)(start_page + size);
940 	unsigned long next;
941 	pgd_t *pgd;
942 	pud_t *pud;
943 	pmd_t *pmd;
944 
945 	for (; addr < end; addr = next) {
946 		void *p = NULL;
947 
948 		pgd = vmemmap_pgd_populate(addr, node);
949 		if (!pgd)
950 			return -ENOMEM;
951 
952 		pud = vmemmap_pud_populate(pgd, addr, node);
953 		if (!pud)
954 			return -ENOMEM;
955 
956 		if (!cpu_has_pse) {
957 			next = (addr + PAGE_SIZE) & PAGE_MASK;
958 			pmd = vmemmap_pmd_populate(pud, addr, node);
959 
960 			if (!pmd)
961 				return -ENOMEM;
962 
963 			p = vmemmap_pte_populate(pmd, addr, node);
964 
965 			if (!p)
966 				return -ENOMEM;
967 
968 			addr_end = addr + PAGE_SIZE;
969 			p_end = p + PAGE_SIZE;
970 		} else {
971 			next = pmd_addr_end(addr, end);
972 
973 			pmd = pmd_offset(pud, addr);
974 			if (pmd_none(*pmd)) {
975 				pte_t entry;
976 
977 				p = vmemmap_alloc_block(PMD_SIZE, node);
978 				if (!p)
979 					return -ENOMEM;
980 
981 				entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
982 						PAGE_KERNEL_LARGE);
983 				set_pmd(pmd, __pmd(pte_val(entry)));
984 
985 				/* check to see if we have contiguous blocks */
986 				if (p_end != p || node_start != node) {
987 					if (p_start)
988 						printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
989 						       addr_start, addr_end-1, p_start, p_end-1, node_start);
990 					addr_start = addr;
991 					node_start = node;
992 					p_start = p;
993 				}
994 
995 				addr_end = addr + PMD_SIZE;
996 				p_end = p + PMD_SIZE;
997 			} else
998 				vmemmap_verify((pte_t *)pmd, node, addr, next);
999 		}
1000 
1001 	}
1002 	return 0;
1003 }
1004 
1005 void __meminit vmemmap_populate_print_last(void)
1006 {
1007 	if (p_start) {
1008 		printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1009 			addr_start, addr_end-1, p_start, p_end-1, node_start);
1010 		p_start = NULL;
1011 		p_end = NULL;
1012 		node_start = 0;
1013 	}
1014 }
1015 #endif
1016