• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/memory.h>
31 #include <linux/memory_hotplug.h>
32 #include <linux/memremap.h>
33 #include <linux/nmi.h>
34 #include <linux/gfp.h>
35 #include <linux/kcore.h>
36 
37 #include <asm/processor.h>
38 #include <asm/bios_ebda.h>
39 #include <asm/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
42 #include <asm/dma.h>
43 #include <asm/fixmap.h>
44 #include <asm/e820.h>
45 #include <asm/apic.h>
46 #include <asm/tlb.h>
47 #include <asm/mmu_context.h>
48 #include <asm/proto.h>
49 #include <asm/smp.h>
50 #include <asm/sections.h>
51 #include <asm/kdebug.h>
52 #include <asm/numa.h>
53 #include <asm/cacheflush.h>
54 #include <asm/init.h>
55 #include <asm/uv/uv.h>
56 #include <asm/setup.h>
57 
58 #include "mm_internal.h"
59 
60 #include "ident_map.c"
61 
62 /*
63  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
64  * physical space so we can cache the place of the first one and move
65  * around without checking the pgd every time.
66  */
67 
68 pteval_t __supported_pte_mask __read_mostly = ~0;
69 EXPORT_SYMBOL_GPL(__supported_pte_mask);
70 
71 int force_personality32;
72 
73 /*
74  * noexec32=on|off
75  * Control non executable heap for 32bit processes.
76  * To control the stack too use noexec=off
77  *
78  * on	PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
79  * off	PROT_READ implies PROT_EXEC
80  */
nonx32_setup(char * str)81 static int __init nonx32_setup(char *str)
82 {
83 	if (!strcmp(str, "on"))
84 		force_personality32 &= ~READ_IMPLIES_EXEC;
85 	else if (!strcmp(str, "off"))
86 		force_personality32 |= READ_IMPLIES_EXEC;
87 	return 1;
88 }
89 __setup("noexec32=", nonx32_setup);
90 
91 /*
92  * When memory was added/removed make sure all the processes MM have
93  * suitable PGD entries in the local PGD level page.
94  */
sync_global_pgds(unsigned long start,unsigned long end,int removed)95 void sync_global_pgds(unsigned long start, unsigned long end, int removed)
96 {
97 	unsigned long addr;
98 
99 	for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
100 		const pgd_t *pgd_ref = pgd_offset_k(addr);
101 		struct page *page;
102 
103 		/*
104 		 * When it is called after memory hot remove, pgd_none()
105 		 * returns true. In this case (removed == 1), we must clear
106 		 * the PGD entries in the local PGD level page.
107 		 */
108 		if (pgd_none(*pgd_ref) && !removed)
109 			continue;
110 
111 		spin_lock(&pgd_lock);
112 		list_for_each_entry(page, &pgd_list, lru) {
113 			pgd_t *pgd;
114 			spinlock_t *pgt_lock;
115 
116 			pgd = (pgd_t *)page_address(page) + pgd_index(addr);
117 			/* the pgt_lock only for Xen */
118 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
119 			spin_lock(pgt_lock);
120 
121 			if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
122 				BUG_ON(pgd_page_vaddr(*pgd)
123 				       != pgd_page_vaddr(*pgd_ref));
124 
125 			if (removed) {
126 				if (pgd_none(*pgd_ref) && !pgd_none(*pgd))
127 					pgd_clear(pgd);
128 			} else {
129 				if (pgd_none(*pgd))
130 					set_pgd(pgd, *pgd_ref);
131 			}
132 
133 			spin_unlock(pgt_lock);
134 		}
135 		spin_unlock(&pgd_lock);
136 	}
137 }
138 
139 /*
140  * NOTE: This function is marked __ref because it calls __init function
141  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
142  */
spp_getpage(void)143 static __ref void *spp_getpage(void)
144 {
145 	void *ptr;
146 
147 	if (after_bootmem)
148 		ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
149 	else
150 		ptr = alloc_bootmem_pages(PAGE_SIZE);
151 
152 	if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
153 		panic("set_pte_phys: cannot allocate page data %s\n",
154 			after_bootmem ? "after bootmem" : "");
155 	}
156 
157 	pr_debug("spp_getpage %p\n", ptr);
158 
159 	return ptr;
160 }
161 
fill_pud(pgd_t * pgd,unsigned long vaddr)162 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
163 {
164 	if (pgd_none(*pgd)) {
165 		pud_t *pud = (pud_t *)spp_getpage();
166 		pgd_populate(&init_mm, pgd, pud);
167 		if (pud != pud_offset(pgd, 0))
168 			printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
169 			       pud, pud_offset(pgd, 0));
170 	}
171 	return pud_offset(pgd, vaddr);
172 }
173 
fill_pmd(pud_t * pud,unsigned long vaddr)174 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
175 {
176 	if (pud_none(*pud)) {
177 		pmd_t *pmd = (pmd_t *) spp_getpage();
178 		pud_populate(&init_mm, pud, pmd);
179 		if (pmd != pmd_offset(pud, 0))
180 			printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
181 			       pmd, pmd_offset(pud, 0));
182 	}
183 	return pmd_offset(pud, vaddr);
184 }
185 
fill_pte(pmd_t * pmd,unsigned long vaddr)186 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
187 {
188 	if (pmd_none(*pmd)) {
189 		pte_t *pte = (pte_t *) spp_getpage();
190 		pmd_populate_kernel(&init_mm, pmd, pte);
191 		if (pte != pte_offset_kernel(pmd, 0))
192 			printk(KERN_ERR "PAGETABLE BUG #02!\n");
193 	}
194 	return pte_offset_kernel(pmd, vaddr);
195 }
196 
set_pte_vaddr_pud(pud_t * pud_page,unsigned long vaddr,pte_t new_pte)197 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
198 {
199 	pud_t *pud;
200 	pmd_t *pmd;
201 	pte_t *pte;
202 
203 	pud = pud_page + pud_index(vaddr);
204 	pmd = fill_pmd(pud, vaddr);
205 	pte = fill_pte(pmd, vaddr);
206 
207 	set_pte(pte, new_pte);
208 
209 	/*
210 	 * It's enough to flush this one mapping.
211 	 * (PGE mappings get flushed as well)
212 	 */
213 	__flush_tlb_one(vaddr);
214 }
215 
set_pte_vaddr(unsigned long vaddr,pte_t pteval)216 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
217 {
218 	pgd_t *pgd;
219 	pud_t *pud_page;
220 
221 	pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
222 
223 	pgd = pgd_offset_k(vaddr);
224 	if (pgd_none(*pgd)) {
225 		printk(KERN_ERR
226 			"PGD FIXMAP MISSING, it should be setup in head.S!\n");
227 		return;
228 	}
229 	pud_page = (pud_t*)pgd_page_vaddr(*pgd);
230 	set_pte_vaddr_pud(pud_page, vaddr, pteval);
231 }
232 
populate_extra_pmd(unsigned long vaddr)233 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
234 {
235 	pgd_t *pgd;
236 	pud_t *pud;
237 
238 	pgd = pgd_offset_k(vaddr);
239 	pud = fill_pud(pgd, vaddr);
240 	return fill_pmd(pud, vaddr);
241 }
242 
populate_extra_pte(unsigned long vaddr)243 pte_t * __init populate_extra_pte(unsigned long vaddr)
244 {
245 	pmd_t *pmd;
246 
247 	pmd = populate_extra_pmd(vaddr);
248 	return fill_pte(pmd, vaddr);
249 }
250 
251 /*
252  * Create large page table mappings for a range of physical addresses.
253  */
__init_extra_mapping(unsigned long phys,unsigned long size,enum page_cache_mode cache)254 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
255 					enum page_cache_mode cache)
256 {
257 	pgd_t *pgd;
258 	pud_t *pud;
259 	pmd_t *pmd;
260 	pgprot_t prot;
261 
262 	pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
263 		pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache)));
264 	BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
265 	for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
266 		pgd = pgd_offset_k((unsigned long)__va(phys));
267 		if (pgd_none(*pgd)) {
268 			pud = (pud_t *) spp_getpage();
269 			set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
270 						_PAGE_USER));
271 		}
272 		pud = pud_offset(pgd, (unsigned long)__va(phys));
273 		if (pud_none(*pud)) {
274 			pmd = (pmd_t *) spp_getpage();
275 			set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
276 						_PAGE_USER));
277 		}
278 		pmd = pmd_offset(pud, phys);
279 		BUG_ON(!pmd_none(*pmd));
280 		set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
281 	}
282 }
283 
init_extra_mapping_wb(unsigned long phys,unsigned long size)284 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
285 {
286 	__init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
287 }
288 
init_extra_mapping_uc(unsigned long phys,unsigned long size)289 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
290 {
291 	__init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
292 }
293 
294 /*
295  * The head.S code sets up the kernel high mapping:
296  *
297  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
298  *
299  * phys_base holds the negative offset to the kernel, which is added
300  * to the compile time generated pmds. This results in invalid pmds up
301  * to the point where we hit the physaddr 0 mapping.
302  *
303  * We limit the mappings to the region from _text to _brk_end.  _brk_end
304  * is rounded up to the 2MB boundary. This catches the invalid pmds as
305  * well, as they are located before _text:
306  */
cleanup_highmap(void)307 void __init cleanup_highmap(void)
308 {
309 	unsigned long vaddr = __START_KERNEL_map;
310 	unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
311 	unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
312 	pmd_t *pmd = level2_kernel_pgt;
313 
314 	/*
315 	 * Native path, max_pfn_mapped is not set yet.
316 	 * Xen has valid max_pfn_mapped set in
317 	 *	arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
318 	 */
319 	if (max_pfn_mapped)
320 		vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
321 
322 	for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
323 		if (pmd_none(*pmd))
324 			continue;
325 		if (vaddr < (unsigned long) _text || vaddr > end)
326 			set_pmd(pmd, __pmd(0));
327 		else if (kaiser_enabled) {
328 			/*
329 			 * level2_kernel_pgt is initialized with _PAGE_GLOBAL:
330 			 * clear that now.  This is not important, so long as
331 			 * CR4.PGE remains clear, but it removes an anomaly.
332 			 * Physical mapping setup below avoids _PAGE_GLOBAL
333 			 * by use of massage_pgprot() inside pfn_pte() etc.
334 			 */
335 			set_pmd(pmd, pmd_clear_flags(*pmd, _PAGE_GLOBAL));
336 		}
337 	}
338 }
339 
340 /*
341  * Create PTE level page table mapping for physical addresses.
342  * It returns the last physical address mapped.
343  */
344 static unsigned long __meminit
phys_pte_init(pte_t * pte_page,unsigned long paddr,unsigned long paddr_end,pgprot_t prot)345 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
346 	      pgprot_t prot)
347 {
348 	unsigned long pages = 0, paddr_next;
349 	unsigned long paddr_last = paddr_end;
350 	pte_t *pte;
351 	int i;
352 
353 	pte = pte_page + pte_index(paddr);
354 	i = pte_index(paddr);
355 
356 	for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
357 		paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
358 		if (paddr >= paddr_end) {
359 			if (!after_bootmem &&
360 			    !e820_any_mapped(paddr & PAGE_MASK, paddr_next,
361 					     E820_RAM) &&
362 			    !e820_any_mapped(paddr & PAGE_MASK, paddr_next,
363 					     E820_RESERVED_KERN))
364 				set_pte(pte, __pte(0));
365 			continue;
366 		}
367 
368 		/*
369 		 * We will re-use the existing mapping.
370 		 * Xen for example has some special requirements, like mapping
371 		 * pagetable pages as RO. So assume someone who pre-setup
372 		 * these mappings are more intelligent.
373 		 */
374 		if (!pte_none(*pte)) {
375 			if (!after_bootmem)
376 				pages++;
377 			continue;
378 		}
379 
380 		if (0)
381 			pr_info("   pte=%p addr=%lx pte=%016lx\n", pte, paddr,
382 				pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
383 		pages++;
384 		set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
385 		paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
386 	}
387 
388 	update_page_count(PG_LEVEL_4K, pages);
389 
390 	return paddr_last;
391 }
392 
393 /*
394  * Create PMD level page table mapping for physical addresses. The virtual
395  * and physical address have to be aligned at this level.
396  * It returns the last physical address mapped.
397  */
398 static unsigned long __meminit
phys_pmd_init(pmd_t * pmd_page,unsigned long paddr,unsigned long paddr_end,unsigned long page_size_mask,pgprot_t prot)399 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
400 	      unsigned long page_size_mask, pgprot_t prot)
401 {
402 	unsigned long pages = 0, paddr_next;
403 	unsigned long paddr_last = paddr_end;
404 
405 	int i = pmd_index(paddr);
406 
407 	for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
408 		pmd_t *pmd = pmd_page + pmd_index(paddr);
409 		pte_t *pte;
410 		pgprot_t new_prot = prot;
411 
412 		paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
413 		if (paddr >= paddr_end) {
414 			if (!after_bootmem &&
415 			    !e820_any_mapped(paddr & PMD_MASK, paddr_next,
416 					     E820_RAM) &&
417 			    !e820_any_mapped(paddr & PMD_MASK, paddr_next,
418 					     E820_RESERVED_KERN))
419 				set_pmd(pmd, __pmd(0));
420 			continue;
421 		}
422 
423 		if (!pmd_none(*pmd)) {
424 			if (!pmd_large(*pmd)) {
425 				spin_lock(&init_mm.page_table_lock);
426 				pte = (pte_t *)pmd_page_vaddr(*pmd);
427 				paddr_last = phys_pte_init(pte, paddr,
428 							   paddr_end, prot);
429 				spin_unlock(&init_mm.page_table_lock);
430 				continue;
431 			}
432 			/*
433 			 * If we are ok with PG_LEVEL_2M mapping, then we will
434 			 * use the existing mapping,
435 			 *
436 			 * Otherwise, we will split the large page mapping but
437 			 * use the same existing protection bits except for
438 			 * large page, so that we don't violate Intel's TLB
439 			 * Application note (317080) which says, while changing
440 			 * the page sizes, new and old translations should
441 			 * not differ with respect to page frame and
442 			 * attributes.
443 			 */
444 			if (page_size_mask & (1 << PG_LEVEL_2M)) {
445 				if (!after_bootmem)
446 					pages++;
447 				paddr_last = paddr_next;
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((paddr & PMD_MASK) >> PAGE_SHIFT,
458 					__pgprot(pgprot_val(prot) | _PAGE_PSE)));
459 			spin_unlock(&init_mm.page_table_lock);
460 			paddr_last = paddr_next;
461 			continue;
462 		}
463 
464 		pte = alloc_low_page();
465 		paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot);
466 
467 		spin_lock(&init_mm.page_table_lock);
468 		pmd_populate_kernel(&init_mm, pmd, pte);
469 		spin_unlock(&init_mm.page_table_lock);
470 	}
471 	update_page_count(PG_LEVEL_2M, pages);
472 	return paddr_last;
473 }
474 
475 /*
476  * Create PUD level page table mapping for physical addresses. The virtual
477  * and physical address do not have to be aligned at this level. KASLR can
478  * randomize virtual addresses up to this level.
479  * It returns the last physical address mapped.
480  */
481 static unsigned long __meminit
phys_pud_init(pud_t * pud_page,unsigned long paddr,unsigned long paddr_end,unsigned long page_size_mask)482 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
483 	      unsigned long page_size_mask)
484 {
485 	unsigned long pages = 0, paddr_next;
486 	unsigned long paddr_last = paddr_end;
487 	unsigned long vaddr = (unsigned long)__va(paddr);
488 	int i = pud_index(vaddr);
489 
490 	for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
491 		pud_t *pud;
492 		pmd_t *pmd;
493 		pgprot_t prot = PAGE_KERNEL;
494 
495 		vaddr = (unsigned long)__va(paddr);
496 		pud = pud_page + pud_index(vaddr);
497 		paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
498 
499 		if (paddr >= paddr_end) {
500 			if (!after_bootmem &&
501 			    !e820_any_mapped(paddr & PUD_MASK, paddr_next,
502 					     E820_RAM) &&
503 			    !e820_any_mapped(paddr & PUD_MASK, paddr_next,
504 					     E820_RESERVED_KERN))
505 				set_pud(pud, __pud(0));
506 			continue;
507 		}
508 
509 		if (!pud_none(*pud)) {
510 			if (!pud_large(*pud)) {
511 				pmd = pmd_offset(pud, 0);
512 				paddr_last = phys_pmd_init(pmd, paddr,
513 							   paddr_end,
514 							   page_size_mask,
515 							   prot);
516 				__flush_tlb_all();
517 				continue;
518 			}
519 			/*
520 			 * If we are ok with PG_LEVEL_1G mapping, then we will
521 			 * use the existing mapping.
522 			 *
523 			 * Otherwise, we will split the gbpage mapping but use
524 			 * the same existing protection  bits except for large
525 			 * page, so that we don't violate Intel's TLB
526 			 * Application note (317080) which says, while changing
527 			 * the page sizes, new and old translations should
528 			 * not differ with respect to page frame and
529 			 * attributes.
530 			 */
531 			if (page_size_mask & (1 << PG_LEVEL_1G)) {
532 				if (!after_bootmem)
533 					pages++;
534 				paddr_last = paddr_next;
535 				continue;
536 			}
537 			prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
538 		}
539 
540 		if (page_size_mask & (1<<PG_LEVEL_1G)) {
541 			pages++;
542 			spin_lock(&init_mm.page_table_lock);
543 			set_pte((pte_t *)pud,
544 				pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
545 					PAGE_KERNEL_LARGE));
546 			spin_unlock(&init_mm.page_table_lock);
547 			paddr_last = paddr_next;
548 			continue;
549 		}
550 
551 		pmd = alloc_low_page();
552 		paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
553 					   page_size_mask, prot);
554 
555 		spin_lock(&init_mm.page_table_lock);
556 		pud_populate(&init_mm, pud, pmd);
557 		spin_unlock(&init_mm.page_table_lock);
558 	}
559 	__flush_tlb_all();
560 
561 	update_page_count(PG_LEVEL_1G, pages);
562 
563 	return paddr_last;
564 }
565 
566 /*
567  * Create page table mapping for the physical memory for specific physical
568  * addresses. The virtual and physical addresses have to be aligned on PMD level
569  * down. It returns the last physical address mapped.
570  */
571 unsigned long __meminit
kernel_physical_mapping_init(unsigned long paddr_start,unsigned long paddr_end,unsigned long page_size_mask)572 kernel_physical_mapping_init(unsigned long paddr_start,
573 			     unsigned long paddr_end,
574 			     unsigned long page_size_mask)
575 {
576 	bool pgd_changed = false;
577 	unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
578 
579 	paddr_last = paddr_end;
580 	vaddr = (unsigned long)__va(paddr_start);
581 	vaddr_end = (unsigned long)__va(paddr_end);
582 	vaddr_start = vaddr;
583 
584 	for (; vaddr < vaddr_end; vaddr = vaddr_next) {
585 		pgd_t *pgd = pgd_offset_k(vaddr);
586 		pud_t *pud;
587 
588 		vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
589 
590 		if (pgd_val(*pgd)) {
591 			pud = (pud_t *)pgd_page_vaddr(*pgd);
592 			paddr_last = phys_pud_init(pud, __pa(vaddr),
593 						   __pa(vaddr_end),
594 						   page_size_mask);
595 			continue;
596 		}
597 
598 		pud = alloc_low_page();
599 		paddr_last = phys_pud_init(pud, __pa(vaddr), __pa(vaddr_end),
600 					   page_size_mask);
601 
602 		spin_lock(&init_mm.page_table_lock);
603 		pgd_populate(&init_mm, pgd, pud);
604 		spin_unlock(&init_mm.page_table_lock);
605 		pgd_changed = true;
606 	}
607 
608 	if (pgd_changed)
609 		sync_global_pgds(vaddr_start, vaddr_end - 1, 0);
610 
611 	__flush_tlb_all();
612 
613 	return paddr_last;
614 }
615 
616 #ifndef CONFIG_NUMA
initmem_init(void)617 void __init initmem_init(void)
618 {
619 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
620 }
621 #endif
622 
paging_init(void)623 void __init paging_init(void)
624 {
625 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
626 	sparse_init();
627 
628 	/*
629 	 * clear the default setting with node 0
630 	 * note: don't use nodes_clear here, that is really clearing when
631 	 *	 numa support is not compiled in, and later node_set_state
632 	 *	 will not set it back.
633 	 */
634 	node_clear_state(0, N_MEMORY);
635 	if (N_MEMORY != N_NORMAL_MEMORY)
636 		node_clear_state(0, N_NORMAL_MEMORY);
637 
638 	zone_sizes_init();
639 }
640 
641 /*
642  * Memory hotplug specific functions
643  */
644 #ifdef CONFIG_MEMORY_HOTPLUG
645 /*
646  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
647  * updating.
648  */
update_end_of_memory_vars(u64 start,u64 size)649 static void  update_end_of_memory_vars(u64 start, u64 size)
650 {
651 	unsigned long end_pfn = PFN_UP(start + size);
652 
653 	if (end_pfn > max_pfn) {
654 		max_pfn = end_pfn;
655 		max_low_pfn = end_pfn;
656 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
657 	}
658 }
659 
660 /*
661  * Memory is added always to NORMAL zone. This means you will never get
662  * additional DMA/DMA32 memory.
663  */
arch_add_memory(int nid,u64 start,u64 size,bool for_device)664 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
665 {
666 	struct pglist_data *pgdat = NODE_DATA(nid);
667 	struct zone *zone = pgdat->node_zones +
668 		zone_for_memory(nid, start, size, ZONE_NORMAL, for_device);
669 	unsigned long start_pfn = start >> PAGE_SHIFT;
670 	unsigned long nr_pages = size >> PAGE_SHIFT;
671 	int ret;
672 
673 	init_memory_mapping(start, start + size);
674 
675 	ret = __add_pages(nid, zone, start_pfn, nr_pages);
676 	WARN_ON_ONCE(ret);
677 
678 	/* update max_pfn, max_low_pfn and high_memory */
679 	update_end_of_memory_vars(start, size);
680 
681 	return ret;
682 }
683 EXPORT_SYMBOL_GPL(arch_add_memory);
684 
685 #define PAGE_INUSE 0xFD
686 
free_pagetable(struct page * page,int order)687 static void __meminit free_pagetable(struct page *page, int order)
688 {
689 	unsigned long magic;
690 	unsigned int nr_pages = 1 << order;
691 	struct vmem_altmap *altmap = to_vmem_altmap((unsigned long) page);
692 
693 	if (altmap) {
694 		vmem_altmap_free(altmap, nr_pages);
695 		return;
696 	}
697 
698 	/* bootmem page has reserved flag */
699 	if (PageReserved(page)) {
700 		__ClearPageReserved(page);
701 
702 		magic = (unsigned long)page->freelist;
703 		if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
704 			while (nr_pages--)
705 				put_page_bootmem(page++);
706 		} else
707 			while (nr_pages--)
708 				free_reserved_page(page++);
709 	} else
710 		free_pages((unsigned long)page_address(page), order);
711 }
712 
free_pte_table(pte_t * pte_start,pmd_t * pmd)713 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
714 {
715 	pte_t *pte;
716 	int i;
717 
718 	for (i = 0; i < PTRS_PER_PTE; i++) {
719 		pte = pte_start + i;
720 		if (!pte_none(*pte))
721 			return;
722 	}
723 
724 	/* free a pte talbe */
725 	free_pagetable(pmd_page(*pmd), 0);
726 	spin_lock(&init_mm.page_table_lock);
727 	pmd_clear(pmd);
728 	spin_unlock(&init_mm.page_table_lock);
729 }
730 
free_pmd_table(pmd_t * pmd_start,pud_t * pud)731 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
732 {
733 	pmd_t *pmd;
734 	int i;
735 
736 	for (i = 0; i < PTRS_PER_PMD; i++) {
737 		pmd = pmd_start + i;
738 		if (!pmd_none(*pmd))
739 			return;
740 	}
741 
742 	/* free a pmd talbe */
743 	free_pagetable(pud_page(*pud), 0);
744 	spin_lock(&init_mm.page_table_lock);
745 	pud_clear(pud);
746 	spin_unlock(&init_mm.page_table_lock);
747 }
748 
749 static void __meminit
remove_pte_table(pte_t * pte_start,unsigned long addr,unsigned long end,bool direct)750 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
751 		 bool direct)
752 {
753 	unsigned long next, pages = 0;
754 	pte_t *pte;
755 	void *page_addr;
756 	phys_addr_t phys_addr;
757 
758 	pte = pte_start + pte_index(addr);
759 	for (; addr < end; addr = next, pte++) {
760 		next = (addr + PAGE_SIZE) & PAGE_MASK;
761 		if (next > end)
762 			next = end;
763 
764 		if (!pte_present(*pte))
765 			continue;
766 
767 		/*
768 		 * We mapped [0,1G) memory as identity mapping when
769 		 * initializing, in arch/x86/kernel/head_64.S. These
770 		 * pagetables cannot be removed.
771 		 */
772 		phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
773 		if (phys_addr < (phys_addr_t)0x40000000)
774 			return;
775 
776 		if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
777 			/*
778 			 * Do not free direct mapping pages since they were
779 			 * freed when offlining, or simplely not in use.
780 			 */
781 			if (!direct)
782 				free_pagetable(pte_page(*pte), 0);
783 
784 			spin_lock(&init_mm.page_table_lock);
785 			pte_clear(&init_mm, addr, pte);
786 			spin_unlock(&init_mm.page_table_lock);
787 
788 			/* For non-direct mapping, pages means nothing. */
789 			pages++;
790 		} else {
791 			/*
792 			 * If we are here, we are freeing vmemmap pages since
793 			 * direct mapped memory ranges to be freed are aligned.
794 			 *
795 			 * If we are not removing the whole page, it means
796 			 * other page structs in this page are being used and
797 			 * we canot remove them. So fill the unused page_structs
798 			 * with 0xFD, and remove the page when it is wholly
799 			 * filled with 0xFD.
800 			 */
801 			memset((void *)addr, PAGE_INUSE, next - addr);
802 
803 			page_addr = page_address(pte_page(*pte));
804 			if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
805 				free_pagetable(pte_page(*pte), 0);
806 
807 				spin_lock(&init_mm.page_table_lock);
808 				pte_clear(&init_mm, addr, pte);
809 				spin_unlock(&init_mm.page_table_lock);
810 			}
811 		}
812 	}
813 
814 	/* Call free_pte_table() in remove_pmd_table(). */
815 	flush_tlb_all();
816 	if (direct)
817 		update_page_count(PG_LEVEL_4K, -pages);
818 }
819 
820 static void __meminit
remove_pmd_table(pmd_t * pmd_start,unsigned long addr,unsigned long end,bool direct)821 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
822 		 bool direct)
823 {
824 	unsigned long next, pages = 0;
825 	pte_t *pte_base;
826 	pmd_t *pmd;
827 	void *page_addr;
828 
829 	pmd = pmd_start + pmd_index(addr);
830 	for (; addr < end; addr = next, pmd++) {
831 		next = pmd_addr_end(addr, end);
832 
833 		if (!pmd_present(*pmd))
834 			continue;
835 
836 		if (pmd_large(*pmd)) {
837 			if (IS_ALIGNED(addr, PMD_SIZE) &&
838 			    IS_ALIGNED(next, PMD_SIZE)) {
839 				if (!direct)
840 					free_pagetable(pmd_page(*pmd),
841 						       get_order(PMD_SIZE));
842 
843 				spin_lock(&init_mm.page_table_lock);
844 				pmd_clear(pmd);
845 				spin_unlock(&init_mm.page_table_lock);
846 				pages++;
847 			} else {
848 				/* If here, we are freeing vmemmap pages. */
849 				memset((void *)addr, PAGE_INUSE, next - addr);
850 
851 				page_addr = page_address(pmd_page(*pmd));
852 				if (!memchr_inv(page_addr, PAGE_INUSE,
853 						PMD_SIZE)) {
854 					free_pagetable(pmd_page(*pmd),
855 						       get_order(PMD_SIZE));
856 
857 					spin_lock(&init_mm.page_table_lock);
858 					pmd_clear(pmd);
859 					spin_unlock(&init_mm.page_table_lock);
860 				}
861 			}
862 
863 			continue;
864 		}
865 
866 		pte_base = (pte_t *)pmd_page_vaddr(*pmd);
867 		remove_pte_table(pte_base, addr, next, direct);
868 		free_pte_table(pte_base, pmd);
869 	}
870 
871 	/* Call free_pmd_table() in remove_pud_table(). */
872 	if (direct)
873 		update_page_count(PG_LEVEL_2M, -pages);
874 }
875 
876 static void __meminit
remove_pud_table(pud_t * pud_start,unsigned long addr,unsigned long end,bool direct)877 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
878 		 bool direct)
879 {
880 	unsigned long next, pages = 0;
881 	pmd_t *pmd_base;
882 	pud_t *pud;
883 	void *page_addr;
884 
885 	pud = pud_start + pud_index(addr);
886 	for (; addr < end; addr = next, pud++) {
887 		next = pud_addr_end(addr, end);
888 
889 		if (!pud_present(*pud))
890 			continue;
891 
892 		if (pud_large(*pud)) {
893 			if (IS_ALIGNED(addr, PUD_SIZE) &&
894 			    IS_ALIGNED(next, PUD_SIZE)) {
895 				if (!direct)
896 					free_pagetable(pud_page(*pud),
897 						       get_order(PUD_SIZE));
898 
899 				spin_lock(&init_mm.page_table_lock);
900 				pud_clear(pud);
901 				spin_unlock(&init_mm.page_table_lock);
902 				pages++;
903 			} else {
904 				/* If here, we are freeing vmemmap pages. */
905 				memset((void *)addr, PAGE_INUSE, next - addr);
906 
907 				page_addr = page_address(pud_page(*pud));
908 				if (!memchr_inv(page_addr, PAGE_INUSE,
909 						PUD_SIZE)) {
910 					free_pagetable(pud_page(*pud),
911 						       get_order(PUD_SIZE));
912 
913 					spin_lock(&init_mm.page_table_lock);
914 					pud_clear(pud);
915 					spin_unlock(&init_mm.page_table_lock);
916 				}
917 			}
918 
919 			continue;
920 		}
921 
922 		pmd_base = (pmd_t *)pud_page_vaddr(*pud);
923 		remove_pmd_table(pmd_base, addr, next, direct);
924 		free_pmd_table(pmd_base, pud);
925 	}
926 
927 	if (direct)
928 		update_page_count(PG_LEVEL_1G, -pages);
929 }
930 
931 /* start and end are both virtual address. */
932 static void __meminit
remove_pagetable(unsigned long start,unsigned long end,bool direct)933 remove_pagetable(unsigned long start, unsigned long end, bool direct)
934 {
935 	unsigned long next;
936 	unsigned long addr;
937 	pgd_t *pgd;
938 	pud_t *pud;
939 
940 	for (addr = start; addr < end; addr = next) {
941 		next = pgd_addr_end(addr, end);
942 
943 		pgd = pgd_offset_k(addr);
944 		if (!pgd_present(*pgd))
945 			continue;
946 
947 		pud = (pud_t *)pgd_page_vaddr(*pgd);
948 		remove_pud_table(pud, addr, next, direct);
949 	}
950 
951 	flush_tlb_all();
952 }
953 
vmemmap_free(unsigned long start,unsigned long end)954 void __ref vmemmap_free(unsigned long start, unsigned long end)
955 {
956 	remove_pagetable(start, end, false);
957 }
958 
959 #ifdef CONFIG_MEMORY_HOTREMOVE
960 static void __meminit
kernel_physical_mapping_remove(unsigned long start,unsigned long end)961 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
962 {
963 	start = (unsigned long)__va(start);
964 	end = (unsigned long)__va(end);
965 
966 	remove_pagetable(start, end, true);
967 }
968 
arch_remove_memory(u64 start,u64 size)969 int __ref arch_remove_memory(u64 start, u64 size)
970 {
971 	unsigned long start_pfn = start >> PAGE_SHIFT;
972 	unsigned long nr_pages = size >> PAGE_SHIFT;
973 	struct page *page = pfn_to_page(start_pfn);
974 	struct vmem_altmap *altmap;
975 	struct zone *zone;
976 	int ret;
977 
978 	/* With altmap the first mapped page is offset from @start */
979 	altmap = to_vmem_altmap((unsigned long) page);
980 	if (altmap)
981 		page += vmem_altmap_offset(altmap);
982 	zone = page_zone(page);
983 	ret = __remove_pages(zone, start_pfn, nr_pages);
984 	WARN_ON_ONCE(ret);
985 	kernel_physical_mapping_remove(start, start + size);
986 
987 	return ret;
988 }
989 #endif
990 #endif /* CONFIG_MEMORY_HOTPLUG */
991 
992 static struct kcore_list kcore_vsyscall;
993 
register_page_bootmem_info(void)994 static void __init register_page_bootmem_info(void)
995 {
996 #ifdef CONFIG_NUMA
997 	int i;
998 
999 	for_each_online_node(i)
1000 		register_page_bootmem_info_node(NODE_DATA(i));
1001 #endif
1002 }
1003 
mem_init(void)1004 void __init mem_init(void)
1005 {
1006 	pci_iommu_alloc();
1007 
1008 	/* clear_bss() already clear the empty_zero_page */
1009 
1010 	register_page_bootmem_info();
1011 
1012 	/* this will put all memory onto the freelists */
1013 	free_all_bootmem();
1014 	after_bootmem = 1;
1015 
1016 	/* Register memory areas for /proc/kcore */
1017 	kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR,
1018 			 PAGE_SIZE, KCORE_OTHER);
1019 
1020 	mem_init_print_info(NULL);
1021 }
1022 
1023 const int rodata_test_data = 0xC3;
1024 EXPORT_SYMBOL_GPL(rodata_test_data);
1025 
1026 int kernel_set_to_readonly;
1027 
set_kernel_text_rw(void)1028 void set_kernel_text_rw(void)
1029 {
1030 	unsigned long start = PFN_ALIGN(_text);
1031 	unsigned long end = PFN_ALIGN(__stop___ex_table);
1032 
1033 	if (!kernel_set_to_readonly)
1034 		return;
1035 
1036 	pr_debug("Set kernel text: %lx - %lx for read write\n",
1037 		 start, end);
1038 
1039 	/*
1040 	 * Make the kernel identity mapping for text RW. Kernel text
1041 	 * mapping will always be RO. Refer to the comment in
1042 	 * static_protections() in pageattr.c
1043 	 */
1044 	set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1045 }
1046 
set_kernel_text_ro(void)1047 void set_kernel_text_ro(void)
1048 {
1049 	unsigned long start = PFN_ALIGN(_text);
1050 	unsigned long end = PFN_ALIGN(__stop___ex_table);
1051 
1052 	if (!kernel_set_to_readonly)
1053 		return;
1054 
1055 	pr_debug("Set kernel text: %lx - %lx for read only\n",
1056 		 start, end);
1057 
1058 	/*
1059 	 * Set the kernel identity mapping for text RO.
1060 	 */
1061 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1062 }
1063 
mark_rodata_ro(void)1064 void mark_rodata_ro(void)
1065 {
1066 	unsigned long start = PFN_ALIGN(_text);
1067 	unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1068 	unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1069 	unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1070 	unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1071 	unsigned long all_end;
1072 
1073 	printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1074 	       (end - start) >> 10);
1075 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1076 
1077 	kernel_set_to_readonly = 1;
1078 
1079 	/*
1080 	 * The rodata/data/bss/brk section (but not the kernel text!)
1081 	 * should also be not-executable.
1082 	 *
1083 	 * We align all_end to PMD_SIZE because the existing mapping
1084 	 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1085 	 * split the PMD and the reminder between _brk_end and the end
1086 	 * of the PMD will remain mapped executable.
1087 	 *
1088 	 * Any PMD which was setup after the one which covers _brk_end
1089 	 * has been zapped already via cleanup_highmem().
1090 	 */
1091 	all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1092 	set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1093 
1094 	rodata_test();
1095 
1096 #ifdef CONFIG_CPA_DEBUG
1097 	printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1098 	set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1099 
1100 	printk(KERN_INFO "Testing CPA: again\n");
1101 	set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1102 #endif
1103 
1104 	free_init_pages("unused kernel",
1105 			(unsigned long) __va(__pa_symbol(text_end)),
1106 			(unsigned long) __va(__pa_symbol(rodata_start)));
1107 	free_init_pages("unused kernel",
1108 			(unsigned long) __va(__pa_symbol(rodata_end)),
1109 			(unsigned long) __va(__pa_symbol(_sdata)));
1110 
1111 	debug_checkwx();
1112 }
1113 
kern_addr_valid(unsigned long addr)1114 int kern_addr_valid(unsigned long addr)
1115 {
1116 	unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1117 	pgd_t *pgd;
1118 	pud_t *pud;
1119 	pmd_t *pmd;
1120 	pte_t *pte;
1121 
1122 	if (above != 0 && above != -1UL)
1123 		return 0;
1124 
1125 	pgd = pgd_offset_k(addr);
1126 	if (pgd_none(*pgd))
1127 		return 0;
1128 
1129 	pud = pud_offset(pgd, addr);
1130 	if (pud_none(*pud))
1131 		return 0;
1132 
1133 	if (pud_large(*pud))
1134 		return pfn_valid(pud_pfn(*pud));
1135 
1136 	pmd = pmd_offset(pud, addr);
1137 	if (pmd_none(*pmd))
1138 		return 0;
1139 
1140 	if (pmd_large(*pmd))
1141 		return pfn_valid(pmd_pfn(*pmd));
1142 
1143 	pte = pte_offset_kernel(pmd, addr);
1144 	if (pte_none(*pte))
1145 		return 0;
1146 
1147 	return pfn_valid(pte_pfn(*pte));
1148 }
1149 
probe_memory_block_size(void)1150 static unsigned long probe_memory_block_size(void)
1151 {
1152 	unsigned long bz = MIN_MEMORY_BLOCK_SIZE;
1153 
1154 	/* if system is UV or has 64GB of RAM or more, use large blocks */
1155 	if (is_uv_system() || ((max_pfn << PAGE_SHIFT) >= (64UL << 30)))
1156 		bz = 2UL << 30; /* 2GB */
1157 
1158 	pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1159 
1160 	return bz;
1161 }
1162 
1163 static unsigned long memory_block_size_probed;
memory_block_size_bytes(void)1164 unsigned long memory_block_size_bytes(void)
1165 {
1166 	if (!memory_block_size_probed)
1167 		memory_block_size_probed = probe_memory_block_size();
1168 
1169 	return memory_block_size_probed;
1170 }
1171 
1172 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1173 /*
1174  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1175  */
1176 static long __meminitdata addr_start, addr_end;
1177 static void __meminitdata *p_start, *p_end;
1178 static int __meminitdata node_start;
1179 
vmemmap_populate_hugepages(unsigned long start,unsigned long end,int node,struct vmem_altmap * altmap)1180 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1181 		unsigned long end, int node, struct vmem_altmap *altmap)
1182 {
1183 	unsigned long addr;
1184 	unsigned long next;
1185 	pgd_t *pgd;
1186 	pud_t *pud;
1187 	pmd_t *pmd;
1188 
1189 	for (addr = start; addr < end; addr = next) {
1190 		next = pmd_addr_end(addr, end);
1191 
1192 		pgd = vmemmap_pgd_populate(addr, node);
1193 		if (!pgd)
1194 			return -ENOMEM;
1195 
1196 		pud = vmemmap_pud_populate(pgd, addr, node);
1197 		if (!pud)
1198 			return -ENOMEM;
1199 
1200 		pmd = pmd_offset(pud, addr);
1201 		if (pmd_none(*pmd)) {
1202 			void *p;
1203 
1204 			p = __vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1205 			if (p) {
1206 				pte_t entry;
1207 
1208 				entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1209 						PAGE_KERNEL_LARGE);
1210 				set_pmd(pmd, __pmd(pte_val(entry)));
1211 
1212 				/* check to see if we have contiguous blocks */
1213 				if (p_end != p || node_start != node) {
1214 					if (p_start)
1215 						pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1216 						       addr_start, addr_end-1, p_start, p_end-1, node_start);
1217 					addr_start = addr;
1218 					node_start = node;
1219 					p_start = p;
1220 				}
1221 
1222 				addr_end = addr + PMD_SIZE;
1223 				p_end = p + PMD_SIZE;
1224 				continue;
1225 			} else if (altmap)
1226 				return -ENOMEM; /* no fallback */
1227 		} else if (pmd_large(*pmd)) {
1228 			vmemmap_verify((pte_t *)pmd, node, addr, next);
1229 			continue;
1230 		}
1231 		pr_warn_once("vmemmap: falling back to regular page backing\n");
1232 		if (vmemmap_populate_basepages(addr, next, node))
1233 			return -ENOMEM;
1234 	}
1235 	return 0;
1236 }
1237 
vmemmap_populate(unsigned long start,unsigned long end,int node)1238 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
1239 {
1240 	struct vmem_altmap *altmap = to_vmem_altmap(start);
1241 	int err;
1242 
1243 	if (boot_cpu_has(X86_FEATURE_PSE))
1244 		err = vmemmap_populate_hugepages(start, end, node, altmap);
1245 	else if (altmap) {
1246 		pr_err_once("%s: no cpu support for altmap allocations\n",
1247 				__func__);
1248 		err = -ENOMEM;
1249 	} else
1250 		err = vmemmap_populate_basepages(start, end, node);
1251 	if (!err)
1252 		sync_global_pgds(start, end - 1, 0);
1253 	return err;
1254 }
1255 
1256 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
register_page_bootmem_memmap(unsigned long section_nr,struct page * start_page,unsigned long size)1257 void register_page_bootmem_memmap(unsigned long section_nr,
1258 				  struct page *start_page, unsigned long size)
1259 {
1260 	unsigned long addr = (unsigned long)start_page;
1261 	unsigned long end = (unsigned long)(start_page + size);
1262 	unsigned long next;
1263 	pgd_t *pgd;
1264 	pud_t *pud;
1265 	pmd_t *pmd;
1266 	unsigned int nr_pages;
1267 	struct page *page;
1268 
1269 	for (; addr < end; addr = next) {
1270 		pte_t *pte = NULL;
1271 
1272 		pgd = pgd_offset_k(addr);
1273 		if (pgd_none(*pgd)) {
1274 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1275 			continue;
1276 		}
1277 		get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1278 
1279 		pud = pud_offset(pgd, addr);
1280 		if (pud_none(*pud)) {
1281 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1282 			continue;
1283 		}
1284 		get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1285 
1286 		if (!boot_cpu_has(X86_FEATURE_PSE)) {
1287 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1288 			pmd = pmd_offset(pud, addr);
1289 			if (pmd_none(*pmd))
1290 				continue;
1291 			get_page_bootmem(section_nr, pmd_page(*pmd),
1292 					 MIX_SECTION_INFO);
1293 
1294 			pte = pte_offset_kernel(pmd, addr);
1295 			if (pte_none(*pte))
1296 				continue;
1297 			get_page_bootmem(section_nr, pte_page(*pte),
1298 					 SECTION_INFO);
1299 		} else {
1300 			next = pmd_addr_end(addr, end);
1301 
1302 			pmd = pmd_offset(pud, addr);
1303 			if (pmd_none(*pmd))
1304 				continue;
1305 
1306 			nr_pages = 1 << (get_order(PMD_SIZE));
1307 			page = pmd_page(*pmd);
1308 			while (nr_pages--)
1309 				get_page_bootmem(section_nr, page++,
1310 						 SECTION_INFO);
1311 		}
1312 	}
1313 }
1314 #endif
1315 
vmemmap_populate_print_last(void)1316 void __meminit vmemmap_populate_print_last(void)
1317 {
1318 	if (p_start) {
1319 		pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1320 			addr_start, addr_end-1, p_start, p_end-1, node_start);
1321 		p_start = NULL;
1322 		p_end = NULL;
1323 		node_start = 0;
1324 	}
1325 }
1326 #endif
1327