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 #include <asm/uv/uv.h>
55 #include <asm/setup.h>
56 
parse_direct_gbpages_off(char * arg)57 static int __init parse_direct_gbpages_off(char *arg)
58 {
59 	direct_gbpages = 0;
60 	return 0;
61 }
62 early_param("nogbpages", parse_direct_gbpages_off);
63 
parse_direct_gbpages_on(char * arg)64 static int __init parse_direct_gbpages_on(char *arg)
65 {
66 	direct_gbpages = 1;
67 	return 0;
68 }
69 early_param("gbpages", parse_direct_gbpages_on);
70 
71 /*
72  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
73  * physical space so we can cache the place of the first one and move
74  * around without checking the pgd every time.
75  */
76 
77 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
78 EXPORT_SYMBOL_GPL(__supported_pte_mask);
79 
80 int force_personality32;
81 
82 /*
83  * noexec32=on|off
84  * Control non executable heap for 32bit processes.
85  * To control the stack too use noexec=off
86  *
87  * on	PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
88  * off	PROT_READ implies PROT_EXEC
89  */
nonx32_setup(char * str)90 static int __init nonx32_setup(char *str)
91 {
92 	if (!strcmp(str, "on"))
93 		force_personality32 &= ~READ_IMPLIES_EXEC;
94 	else if (!strcmp(str, "off"))
95 		force_personality32 |= READ_IMPLIES_EXEC;
96 	return 1;
97 }
98 __setup("noexec32=", nonx32_setup);
99 
100 /*
101  * When memory was added/removed make sure all the processes MM have
102  * suitable PGD entries in the local PGD level page.
103  */
sync_global_pgds(unsigned long start,unsigned long end)104 void sync_global_pgds(unsigned long start, unsigned long end)
105 {
106 	unsigned long address;
107 
108 	for (address = start; address <= end; address += PGDIR_SIZE) {
109 		const pgd_t *pgd_ref = pgd_offset_k(address);
110 		struct page *page;
111 
112 		if (pgd_none(*pgd_ref))
113 			continue;
114 
115 		spin_lock(&pgd_lock);
116 		list_for_each_entry(page, &pgd_list, lru) {
117 			pgd_t *pgd;
118 			spinlock_t *pgt_lock;
119 
120 			pgd = (pgd_t *)page_address(page) + pgd_index(address);
121 			/* the pgt_lock only for Xen */
122 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
123 			spin_lock(pgt_lock);
124 
125 			if (pgd_none(*pgd))
126 				set_pgd(pgd, *pgd_ref);
127 			else
128 				BUG_ON(pgd_page_vaddr(*pgd)
129 				       != pgd_page_vaddr(*pgd_ref));
130 
131 			spin_unlock(pgt_lock);
132 		}
133 		spin_unlock(&pgd_lock);
134 	}
135 }
136 
137 /*
138  * NOTE: This function is marked __ref because it calls __init function
139  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
140  */
spp_getpage(void)141 static __ref void *spp_getpage(void)
142 {
143 	void *ptr;
144 
145 	if (after_bootmem)
146 		ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
147 	else
148 		ptr = alloc_bootmem_pages(PAGE_SIZE);
149 
150 	if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
151 		panic("set_pte_phys: cannot allocate page data %s\n",
152 			after_bootmem ? "after bootmem" : "");
153 	}
154 
155 	pr_debug("spp_getpage %p\n", ptr);
156 
157 	return ptr;
158 }
159 
fill_pud(pgd_t * pgd,unsigned long vaddr)160 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
161 {
162 	if (pgd_none(*pgd)) {
163 		pud_t *pud = (pud_t *)spp_getpage();
164 		pgd_populate(&init_mm, pgd, pud);
165 		if (pud != pud_offset(pgd, 0))
166 			printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
167 			       pud, pud_offset(pgd, 0));
168 	}
169 	return pud_offset(pgd, vaddr);
170 }
171 
fill_pmd(pud_t * pud,unsigned long vaddr)172 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
173 {
174 	if (pud_none(*pud)) {
175 		pmd_t *pmd = (pmd_t *) spp_getpage();
176 		pud_populate(&init_mm, pud, pmd);
177 		if (pmd != pmd_offset(pud, 0))
178 			printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
179 			       pmd, pmd_offset(pud, 0));
180 	}
181 	return pmd_offset(pud, vaddr);
182 }
183 
fill_pte(pmd_t * pmd,unsigned long vaddr)184 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
185 {
186 	if (pmd_none(*pmd)) {
187 		pte_t *pte = (pte_t *) spp_getpage();
188 		pmd_populate_kernel(&init_mm, pmd, pte);
189 		if (pte != pte_offset_kernel(pmd, 0))
190 			printk(KERN_ERR "PAGETABLE BUG #02!\n");
191 	}
192 	return pte_offset_kernel(pmd, vaddr);
193 }
194 
set_pte_vaddr_pud(pud_t * pud_page,unsigned long vaddr,pte_t new_pte)195 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
196 {
197 	pud_t *pud;
198 	pmd_t *pmd;
199 	pte_t *pte;
200 
201 	pud = pud_page + pud_index(vaddr);
202 	pmd = fill_pmd(pud, vaddr);
203 	pte = fill_pte(pmd, vaddr);
204 
205 	set_pte(pte, new_pte);
206 
207 	/*
208 	 * It's enough to flush this one mapping.
209 	 * (PGE mappings get flushed as well)
210 	 */
211 	__flush_tlb_one(vaddr);
212 }
213 
set_pte_vaddr(unsigned long vaddr,pte_t pteval)214 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
215 {
216 	pgd_t *pgd;
217 	pud_t *pud_page;
218 
219 	pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
220 
221 	pgd = pgd_offset_k(vaddr);
222 	if (pgd_none(*pgd)) {
223 		printk(KERN_ERR
224 			"PGD FIXMAP MISSING, it should be setup in head.S!\n");
225 		return;
226 	}
227 	pud_page = (pud_t*)pgd_page_vaddr(*pgd);
228 	set_pte_vaddr_pud(pud_page, vaddr, pteval);
229 }
230 
populate_extra_pmd(unsigned long vaddr)231 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
232 {
233 	pgd_t *pgd;
234 	pud_t *pud;
235 
236 	pgd = pgd_offset_k(vaddr);
237 	pud = fill_pud(pgd, vaddr);
238 	return fill_pmd(pud, vaddr);
239 }
240 
populate_extra_pte(unsigned long vaddr)241 pte_t * __init populate_extra_pte(unsigned long vaddr)
242 {
243 	pmd_t *pmd;
244 
245 	pmd = populate_extra_pmd(vaddr);
246 	return fill_pte(pmd, vaddr);
247 }
248 
249 /*
250  * Create large page table mappings for a range of physical addresses.
251  */
__init_extra_mapping(unsigned long phys,unsigned long size,pgprot_t prot)252 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
253 						pgprot_t prot)
254 {
255 	pgd_t *pgd;
256 	pud_t *pud;
257 	pmd_t *pmd;
258 
259 	BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
260 	for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
261 		pgd = pgd_offset_k((unsigned long)__va(phys));
262 		if (pgd_none(*pgd)) {
263 			pud = (pud_t *) spp_getpage();
264 			set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
265 						_PAGE_USER));
266 		}
267 		pud = pud_offset(pgd, (unsigned long)__va(phys));
268 		if (pud_none(*pud)) {
269 			pmd = (pmd_t *) spp_getpage();
270 			set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
271 						_PAGE_USER));
272 		}
273 		pmd = pmd_offset(pud, phys);
274 		BUG_ON(!pmd_none(*pmd));
275 		set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
276 	}
277 }
278 
init_extra_mapping_wb(unsigned long phys,unsigned long size)279 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
280 {
281 	__init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
282 }
283 
init_extra_mapping_uc(unsigned long phys,unsigned long size)284 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
285 {
286 	__init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
287 }
288 
289 /*
290  * The head.S code sets up the kernel high mapping:
291  *
292  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
293  *
294  * phys_addr holds the negative offset to the kernel, which is added
295  * to the compile time generated pmds. This results in invalid pmds up
296  * to the point where we hit the physaddr 0 mapping.
297  *
298  * We limit the mappings to the region from _text to _brk_end.  _brk_end
299  * is rounded up to the 2MB boundary. This catches the invalid pmds as
300  * well, as they are located before _text:
301  */
cleanup_highmap(void)302 void __init cleanup_highmap(void)
303 {
304 	unsigned long vaddr = __START_KERNEL_map;
305 	unsigned long vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
306 	unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
307 	pmd_t *pmd = level2_kernel_pgt;
308 
309 	for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
310 		if (pmd_none(*pmd))
311 			continue;
312 		if (vaddr < (unsigned long) _text || vaddr > end)
313 			set_pmd(pmd, __pmd(0));
314 	}
315 }
316 
alloc_low_page(unsigned long * phys)317 static __ref void *alloc_low_page(unsigned long *phys)
318 {
319 	unsigned long pfn = pgt_buf_end++;
320 	void *adr;
321 
322 	if (after_bootmem) {
323 		adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
324 		*phys = __pa(adr);
325 
326 		return adr;
327 	}
328 
329 	if (pfn >= pgt_buf_top)
330 		panic("alloc_low_page: ran out of memory");
331 
332 	adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
333 	clear_page(adr);
334 	*phys  = pfn * PAGE_SIZE;
335 	return adr;
336 }
337 
map_low_page(void * virt)338 static __ref void *map_low_page(void *virt)
339 {
340 	void *adr;
341 	unsigned long phys, left;
342 
343 	if (after_bootmem)
344 		return virt;
345 
346 	phys = __pa(virt);
347 	left = phys & (PAGE_SIZE - 1);
348 	adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE);
349 	adr = (void *)(((unsigned long)adr) | left);
350 
351 	return adr;
352 }
353 
unmap_low_page(void * adr)354 static __ref void unmap_low_page(void *adr)
355 {
356 	if (after_bootmem)
357 		return;
358 
359 	early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE);
360 }
361 
362 static unsigned long __meminit
phys_pte_init(pte_t * pte_page,unsigned long addr,unsigned long end,pgprot_t prot)363 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
364 	      pgprot_t prot)
365 {
366 	unsigned pages = 0;
367 	unsigned long last_map_addr = end;
368 	int i;
369 
370 	pte_t *pte = pte_page + pte_index(addr);
371 
372 	for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
373 
374 		if (addr >= end) {
375 			if (!after_bootmem) {
376 				for(; i < PTRS_PER_PTE; i++, pte++)
377 					set_pte(pte, __pte(0));
378 			}
379 			break;
380 		}
381 
382 		/*
383 		 * We will re-use the existing mapping.
384 		 * Xen for example has some special requirements, like mapping
385 		 * pagetable pages as RO. So assume someone who pre-setup
386 		 * these mappings are more intelligent.
387 		 */
388 		if (pte_val(*pte)) {
389 			pages++;
390 			continue;
391 		}
392 
393 		if (0)
394 			printk("   pte=%p addr=%lx pte=%016lx\n",
395 			       pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
396 		pages++;
397 		set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
398 		last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
399 	}
400 
401 	update_page_count(PG_LEVEL_4K, pages);
402 
403 	return last_map_addr;
404 }
405 
406 static unsigned long __meminit
phys_pmd_init(pmd_t * pmd_page,unsigned long address,unsigned long end,unsigned long page_size_mask,pgprot_t prot)407 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
408 	      unsigned long page_size_mask, pgprot_t prot)
409 {
410 	unsigned long pages = 0;
411 	unsigned long last_map_addr = end;
412 
413 	int i = pmd_index(address);
414 
415 	for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
416 		unsigned long pte_phys;
417 		pmd_t *pmd = pmd_page + pmd_index(address);
418 		pte_t *pte;
419 		pgprot_t new_prot = prot;
420 
421 		if (address >= end) {
422 			if (!after_bootmem) {
423 				for (; i < PTRS_PER_PMD; i++, pmd++)
424 					set_pmd(pmd, __pmd(0));
425 			}
426 			break;
427 		}
428 
429 		if (pmd_val(*pmd)) {
430 			if (!pmd_large(*pmd)) {
431 				spin_lock(&init_mm.page_table_lock);
432 				pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd));
433 				last_map_addr = phys_pte_init(pte, address,
434 								end, prot);
435 				unmap_low_page(pte);
436 				spin_unlock(&init_mm.page_table_lock);
437 				continue;
438 			}
439 			/*
440 			 * If we are ok with PG_LEVEL_2M mapping, then we will
441 			 * use the existing mapping,
442 			 *
443 			 * Otherwise, we will split the large page mapping but
444 			 * use the same existing protection bits except for
445 			 * large page, so that we don't violate Intel's TLB
446 			 * Application note (317080) which says, while changing
447 			 * the page sizes, new and old translations should
448 			 * not differ with respect to page frame and
449 			 * attributes.
450 			 */
451 			if (page_size_mask & (1 << PG_LEVEL_2M)) {
452 				pages++;
453 				continue;
454 			}
455 			new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
456 		}
457 
458 		if (page_size_mask & (1<<PG_LEVEL_2M)) {
459 			pages++;
460 			spin_lock(&init_mm.page_table_lock);
461 			set_pte((pte_t *)pmd,
462 				pfn_pte(address >> PAGE_SHIFT,
463 					__pgprot(pgprot_val(prot) | _PAGE_PSE)));
464 			spin_unlock(&init_mm.page_table_lock);
465 			last_map_addr = (address & PMD_MASK) + PMD_SIZE;
466 			continue;
467 		}
468 
469 		pte = alloc_low_page(&pte_phys);
470 		last_map_addr = phys_pte_init(pte, address, end, new_prot);
471 		unmap_low_page(pte);
472 
473 		spin_lock(&init_mm.page_table_lock);
474 		pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
475 		spin_unlock(&init_mm.page_table_lock);
476 	}
477 	update_page_count(PG_LEVEL_2M, pages);
478 	return last_map_addr;
479 }
480 
481 static unsigned long __meminit
phys_pud_init(pud_t * pud_page,unsigned long addr,unsigned long end,unsigned long page_size_mask)482 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
483 			 unsigned long page_size_mask)
484 {
485 	unsigned long pages = 0;
486 	unsigned long last_map_addr = end;
487 	int i = pud_index(addr);
488 
489 	for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
490 		unsigned long pmd_phys;
491 		pud_t *pud = pud_page + pud_index(addr);
492 		pmd_t *pmd;
493 		pgprot_t prot = PAGE_KERNEL;
494 
495 		if (addr >= end)
496 			break;
497 
498 		if (!after_bootmem &&
499 				!e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
500 			set_pud(pud, __pud(0));
501 			continue;
502 		}
503 
504 		if (pud_val(*pud)) {
505 			if (!pud_large(*pud)) {
506 				pmd = map_low_page(pmd_offset(pud, 0));
507 				last_map_addr = phys_pmd_init(pmd, addr, end,
508 							 page_size_mask, prot);
509 				unmap_low_page(pmd);
510 				__flush_tlb_all();
511 				continue;
512 			}
513 			/*
514 			 * If we are ok with PG_LEVEL_1G mapping, then we will
515 			 * use the existing mapping.
516 			 *
517 			 * Otherwise, we will split the gbpage mapping but use
518 			 * the same existing protection  bits except for large
519 			 * page, so that we don't violate Intel's TLB
520 			 * Application note (317080) which says, while changing
521 			 * the page sizes, new and old translations should
522 			 * not differ with respect to page frame and
523 			 * attributes.
524 			 */
525 			if (page_size_mask & (1 << PG_LEVEL_1G)) {
526 				pages++;
527 				continue;
528 			}
529 			prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
530 		}
531 
532 		if (page_size_mask & (1<<PG_LEVEL_1G)) {
533 			pages++;
534 			spin_lock(&init_mm.page_table_lock);
535 			set_pte((pte_t *)pud,
536 				pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
537 			spin_unlock(&init_mm.page_table_lock);
538 			last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
539 			continue;
540 		}
541 
542 		pmd = alloc_low_page(&pmd_phys);
543 		last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
544 					      prot);
545 		unmap_low_page(pmd);
546 
547 		spin_lock(&init_mm.page_table_lock);
548 		pud_populate(&init_mm, pud, __va(pmd_phys));
549 		spin_unlock(&init_mm.page_table_lock);
550 	}
551 	__flush_tlb_all();
552 
553 	update_page_count(PG_LEVEL_1G, pages);
554 
555 	return last_map_addr;
556 }
557 
558 unsigned long __meminit
kernel_physical_mapping_init(unsigned long start,unsigned long end,unsigned long page_size_mask)559 kernel_physical_mapping_init(unsigned long start,
560 			     unsigned long end,
561 			     unsigned long page_size_mask)
562 {
563 	bool pgd_changed = false;
564 	unsigned long next, last_map_addr = end;
565 	unsigned long addr;
566 
567 	start = (unsigned long)__va(start);
568 	end = (unsigned long)__va(end);
569 	addr = start;
570 
571 	for (; start < end; start = next) {
572 		pgd_t *pgd = pgd_offset_k(start);
573 		unsigned long pud_phys;
574 		pud_t *pud;
575 
576 		next = (start + PGDIR_SIZE) & PGDIR_MASK;
577 		if (next > end)
578 			next = end;
579 
580 		if (pgd_val(*pgd)) {
581 			pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd));
582 			last_map_addr = phys_pud_init(pud, __pa(start),
583 						 __pa(end), page_size_mask);
584 			unmap_low_page(pud);
585 			continue;
586 		}
587 
588 		pud = alloc_low_page(&pud_phys);
589 		last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
590 						 page_size_mask);
591 		unmap_low_page(pud);
592 
593 		spin_lock(&init_mm.page_table_lock);
594 		pgd_populate(&init_mm, pgd, __va(pud_phys));
595 		spin_unlock(&init_mm.page_table_lock);
596 		pgd_changed = true;
597 	}
598 
599 	if (pgd_changed)
600 		sync_global_pgds(addr, end);
601 
602 	__flush_tlb_all();
603 
604 	return last_map_addr;
605 }
606 
607 #ifndef CONFIG_NUMA
initmem_init(void)608 void __init initmem_init(void)
609 {
610 	memblock_x86_register_active_regions(0, 0, max_pfn);
611 }
612 #endif
613 
paging_init(void)614 void __init paging_init(void)
615 {
616 	unsigned long max_zone_pfns[MAX_NR_ZONES];
617 
618 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
619 	max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
620 	max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
621 	max_zone_pfns[ZONE_NORMAL] = max_pfn;
622 
623 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
624 	sparse_init();
625 
626 	/*
627 	 * clear the default setting with node 0
628 	 * note: don't use nodes_clear here, that is really clearing when
629 	 *	 numa support is not compiled in, and later node_set_state
630 	 *	 will not set it back.
631 	 */
632 	node_clear_state(0, N_NORMAL_MEMORY);
633 
634 	free_area_init_nodes(max_zone_pfns);
635 }
636 
637 /*
638  * Memory hotplug specific functions
639  */
640 #ifdef CONFIG_MEMORY_HOTPLUG
641 /*
642  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
643  * updating.
644  */
update_end_of_memory_vars(u64 start,u64 size)645 static void  update_end_of_memory_vars(u64 start, u64 size)
646 {
647 	unsigned long end_pfn = PFN_UP(start + size);
648 
649 	if (end_pfn > max_pfn) {
650 		max_pfn = end_pfn;
651 		max_low_pfn = end_pfn;
652 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
653 	}
654 }
655 
656 /*
657  * Memory is added always to NORMAL zone. This means you will never get
658  * additional DMA/DMA32 memory.
659  */
arch_add_memory(int nid,u64 start,u64 size)660 int arch_add_memory(int nid, u64 start, u64 size)
661 {
662 	struct pglist_data *pgdat = NODE_DATA(nid);
663 	struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
664 	unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
665 	unsigned long nr_pages = size >> PAGE_SHIFT;
666 	int ret;
667 
668 	last_mapped_pfn = init_memory_mapping(start, start + size);
669 	if (last_mapped_pfn > max_pfn_mapped)
670 		max_pfn_mapped = last_mapped_pfn;
671 
672 	ret = __add_pages(nid, zone, start_pfn, nr_pages);
673 	WARN_ON_ONCE(ret);
674 
675 	/* update max_pfn, max_low_pfn and high_memory */
676 	update_end_of_memory_vars(start, size);
677 
678 	return ret;
679 }
680 EXPORT_SYMBOL_GPL(arch_add_memory);
681 
682 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
memory_add_physaddr_to_nid(u64 start)683 int memory_add_physaddr_to_nid(u64 start)
684 {
685 	return 0;
686 }
687 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
688 #endif
689 
690 #endif /* CONFIG_MEMORY_HOTPLUG */
691 
692 static struct kcore_list kcore_vsyscall;
693 
mem_init(void)694 void __init mem_init(void)
695 {
696 	long codesize, reservedpages, datasize, initsize;
697 	unsigned long absent_pages;
698 
699 	pci_iommu_alloc();
700 
701 	/* clear_bss() already clear the empty_zero_page */
702 
703 	reservedpages = 0;
704 
705 	/* this will put all low memory onto the freelists */
706 #ifdef CONFIG_NUMA
707 	totalram_pages = numa_free_all_bootmem();
708 #else
709 	totalram_pages = free_all_bootmem();
710 #endif
711 
712 	absent_pages = absent_pages_in_range(0, max_pfn);
713 	reservedpages = max_pfn - totalram_pages - absent_pages;
714 	after_bootmem = 1;
715 
716 	codesize =  (unsigned long) &_etext - (unsigned long) &_text;
717 	datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
718 	initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
719 
720 	/* Register memory areas for /proc/kcore */
721 	kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
722 			 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
723 
724 	printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
725 			 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
726 		nr_free_pages() << (PAGE_SHIFT-10),
727 		max_pfn << (PAGE_SHIFT-10),
728 		codesize >> 10,
729 		absent_pages << (PAGE_SHIFT-10),
730 		reservedpages << (PAGE_SHIFT-10),
731 		datasize >> 10,
732 		initsize >> 10);
733 }
734 
735 #ifdef CONFIG_DEBUG_RODATA
736 const int rodata_test_data = 0xC3;
737 EXPORT_SYMBOL_GPL(rodata_test_data);
738 
739 int kernel_set_to_readonly;
740 
set_kernel_text_rw(void)741 void set_kernel_text_rw(void)
742 {
743 	unsigned long start = PFN_ALIGN(_text);
744 	unsigned long end = PFN_ALIGN(__stop___ex_table);
745 
746 	if (!kernel_set_to_readonly)
747 		return;
748 
749 	pr_debug("Set kernel text: %lx - %lx for read write\n",
750 		 start, end);
751 
752 	/*
753 	 * Make the kernel identity mapping for text RW. Kernel text
754 	 * mapping will always be RO. Refer to the comment in
755 	 * static_protections() in pageattr.c
756 	 */
757 	set_memory_rw(start, (end - start) >> PAGE_SHIFT);
758 }
759 
set_kernel_text_ro(void)760 void set_kernel_text_ro(void)
761 {
762 	unsigned long start = PFN_ALIGN(_text);
763 	unsigned long end = PFN_ALIGN(__stop___ex_table);
764 
765 	if (!kernel_set_to_readonly)
766 		return;
767 
768 	pr_debug("Set kernel text: %lx - %lx for read only\n",
769 		 start, end);
770 
771 	/*
772 	 * Set the kernel identity mapping for text RO.
773 	 */
774 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
775 }
776 
mark_rodata_ro(void)777 void mark_rodata_ro(void)
778 {
779 	unsigned long start = PFN_ALIGN(_text);
780 	unsigned long rodata_start =
781 		((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
782 	unsigned long end = (unsigned long) &__end_rodata_hpage_align;
783 	unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
784 	unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
785 	unsigned long data_start = (unsigned long) &_sdata;
786 
787 	printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
788 	       (end - start) >> 10);
789 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
790 
791 	kernel_set_to_readonly = 1;
792 
793 	/*
794 	 * The rodata section (but not the kernel text!) should also be
795 	 * not-executable.
796 	 */
797 	set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
798 
799 	rodata_test();
800 
801 #ifdef CONFIG_CPA_DEBUG
802 	printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
803 	set_memory_rw(start, (end-start) >> PAGE_SHIFT);
804 
805 	printk(KERN_INFO "Testing CPA: again\n");
806 	set_memory_ro(start, (end-start) >> PAGE_SHIFT);
807 #endif
808 
809 	free_init_pages("unused kernel memory",
810 			(unsigned long) page_address(virt_to_page(text_end)),
811 			(unsigned long)
812 				 page_address(virt_to_page(rodata_start)));
813 	free_init_pages("unused kernel memory",
814 			(unsigned long) page_address(virt_to_page(rodata_end)),
815 			(unsigned long) page_address(virt_to_page(data_start)));
816 }
817 
818 #endif
819 
kern_addr_valid(unsigned long addr)820 int kern_addr_valid(unsigned long addr)
821 {
822 	unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
823 	pgd_t *pgd;
824 	pud_t *pud;
825 	pmd_t *pmd;
826 	pte_t *pte;
827 
828 	if (above != 0 && above != -1UL)
829 		return 0;
830 
831 	pgd = pgd_offset_k(addr);
832 	if (pgd_none(*pgd))
833 		return 0;
834 
835 	pud = pud_offset(pgd, addr);
836 	if (pud_none(*pud))
837 		return 0;
838 
839 	pmd = pmd_offset(pud, addr);
840 	if (pmd_none(*pmd))
841 		return 0;
842 
843 	if (pmd_large(*pmd))
844 		return pfn_valid(pmd_pfn(*pmd));
845 
846 	pte = pte_offset_kernel(pmd, addr);
847 	if (pte_none(*pte))
848 		return 0;
849 
850 	return pfn_valid(pte_pfn(*pte));
851 }
852 
853 /*
854  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
855  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
856  * not need special handling anymore:
857  */
858 static struct vm_area_struct gate_vma = {
859 	.vm_start	= VSYSCALL_START,
860 	.vm_end		= VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
861 	.vm_page_prot	= PAGE_READONLY_EXEC,
862 	.vm_flags	= VM_READ | VM_EXEC
863 };
864 
get_gate_vma(struct mm_struct * mm)865 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
866 {
867 #ifdef CONFIG_IA32_EMULATION
868 	if (!mm || mm->context.ia32_compat)
869 		return NULL;
870 #endif
871 	return &gate_vma;
872 }
873 
in_gate_area(struct mm_struct * mm,unsigned long addr)874 int in_gate_area(struct mm_struct *mm, unsigned long addr)
875 {
876 	struct vm_area_struct *vma = get_gate_vma(mm);
877 
878 	if (!vma)
879 		return 0;
880 
881 	return (addr >= vma->vm_start) && (addr < vma->vm_end);
882 }
883 
884 /*
885  * Use this when you have no reliable mm, typically from interrupt
886  * context. It is less reliable than using a task's mm and may give
887  * false positives.
888  */
in_gate_area_no_mm(unsigned long addr)889 int in_gate_area_no_mm(unsigned long addr)
890 {
891 	return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
892 }
893 
arch_vma_name(struct vm_area_struct * vma)894 const char *arch_vma_name(struct vm_area_struct *vma)
895 {
896 	if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
897 		return "[vdso]";
898 	if (vma == &gate_vma)
899 		return "[vsyscall]";
900 	return NULL;
901 }
902 
903 #ifdef CONFIG_X86_UV
904 #define MIN_MEMORY_BLOCK_SIZE   (1 << SECTION_SIZE_BITS)
905 
memory_block_size_bytes(void)906 unsigned long memory_block_size_bytes(void)
907 {
908 	if (is_uv_system()) {
909 		printk(KERN_INFO "UV: memory block size 2GB\n");
910 		return 2UL * 1024 * 1024 * 1024;
911 	}
912 	return MIN_MEMORY_BLOCK_SIZE;
913 }
914 #endif
915 
916 #ifdef CONFIG_SPARSEMEM_VMEMMAP
917 /*
918  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
919  */
920 static long __meminitdata addr_start, addr_end;
921 static void __meminitdata *p_start, *p_end;
922 static int __meminitdata node_start;
923 
924 int __meminit
vmemmap_populate(struct page * start_page,unsigned long size,int node)925 vmemmap_populate(struct page *start_page, unsigned long size, int node)
926 {
927 	unsigned long addr = (unsigned long)start_page;
928 	unsigned long end = (unsigned long)(start_page + size);
929 	unsigned long next;
930 	pgd_t *pgd;
931 	pud_t *pud;
932 	pmd_t *pmd;
933 
934 	for (; addr < end; addr = next) {
935 		void *p = NULL;
936 
937 		pgd = vmemmap_pgd_populate(addr, node);
938 		if (!pgd)
939 			return -ENOMEM;
940 
941 		pud = vmemmap_pud_populate(pgd, addr, node);
942 		if (!pud)
943 			return -ENOMEM;
944 
945 		if (!cpu_has_pse) {
946 			next = (addr + PAGE_SIZE) & PAGE_MASK;
947 			pmd = vmemmap_pmd_populate(pud, addr, node);
948 
949 			if (!pmd)
950 				return -ENOMEM;
951 
952 			p = vmemmap_pte_populate(pmd, addr, node);
953 
954 			if (!p)
955 				return -ENOMEM;
956 
957 			addr_end = addr + PAGE_SIZE;
958 			p_end = p + PAGE_SIZE;
959 		} else {
960 			next = pmd_addr_end(addr, end);
961 
962 			pmd = pmd_offset(pud, addr);
963 			if (pmd_none(*pmd)) {
964 				pte_t entry;
965 
966 				p = vmemmap_alloc_block_buf(PMD_SIZE, node);
967 				if (!p)
968 					return -ENOMEM;
969 
970 				entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
971 						PAGE_KERNEL_LARGE);
972 				set_pmd(pmd, __pmd(pte_val(entry)));
973 
974 				/* check to see if we have contiguous blocks */
975 				if (p_end != p || node_start != node) {
976 					if (p_start)
977 						printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
978 						       addr_start, addr_end-1, p_start, p_end-1, node_start);
979 					addr_start = addr;
980 					node_start = node;
981 					p_start = p;
982 				}
983 
984 				addr_end = addr + PMD_SIZE;
985 				p_end = p + PMD_SIZE;
986 			} else
987 				vmemmap_verify((pte_t *)pmd, node, addr, next);
988 		}
989 
990 	}
991 	sync_global_pgds((unsigned long)start_page, end);
992 	return 0;
993 }
994 
vmemmap_populate_print_last(void)995 void __meminit vmemmap_populate_print_last(void)
996 {
997 	if (p_start) {
998 		printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
999 			addr_start, addr_end-1, p_start, p_end-1, node_start);
1000 		p_start = NULL;
1001 		p_end = NULL;
1002 		node_start = 0;
1003 	}
1004 }
1005 #endif
1006