xref: /linux/arch/x86/mm/init_64.c (revision b77e0ce62d63a761ffb7f7245a215a49f5921c2f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/arch/x86_64/mm/init.c
4  *
5  *  Copyright (C) 1995  Linus Torvalds
6  *  Copyright (C) 2000  Pavel Machek <pavel@ucw.cz>
7  *  Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
8  */
9 
10 #include <linux/signal.h>
11 #include <linux/sched.h>
12 #include <linux/kernel.h>
13 #include <linux/errno.h>
14 #include <linux/string.h>
15 #include <linux/types.h>
16 #include <linux/ptrace.h>
17 #include <linux/mman.h>
18 #include <linux/mm.h>
19 #include <linux/swap.h>
20 #include <linux/smp.h>
21 #include <linux/init.h>
22 #include <linux/initrd.h>
23 #include <linux/pagemap.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 <linux/uaccess.h>
40 #include <asm/pgalloc.h>
41 #include <asm/dma.h>
42 #include <asm/fixmap.h>
43 #include <asm/e820/api.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/set_memory.h>
53 #include <asm/init.h>
54 #include <asm/uv/uv.h>
55 #include <asm/setup.h>
56 #include <asm/ftrace.h>
57 
58 #include "mm_internal.h"
59 
60 #include "ident_map.c"
61 
62 #define DEFINE_POPULATE(fname, type1, type2, init)		\
63 static inline void fname##_init(struct mm_struct *mm,		\
64 		type1##_t *arg1, type2##_t *arg2, bool init)	\
65 {								\
66 	if (init)						\
67 		fname##_safe(mm, arg1, arg2);			\
68 	else							\
69 		fname(mm, arg1, arg2);				\
70 }
71 
72 DEFINE_POPULATE(p4d_populate, p4d, pud, init)
73 DEFINE_POPULATE(pgd_populate, pgd, p4d, init)
74 DEFINE_POPULATE(pud_populate, pud, pmd, init)
75 DEFINE_POPULATE(pmd_populate_kernel, pmd, pte, init)
76 
77 #define DEFINE_ENTRY(type1, type2, init)			\
78 static inline void set_##type1##_init(type1##_t *arg1,		\
79 			type2##_t arg2, bool init)		\
80 {								\
81 	if (init)						\
82 		set_##type1##_safe(arg1, arg2);			\
83 	else							\
84 		set_##type1(arg1, arg2);			\
85 }
86 
87 DEFINE_ENTRY(p4d, p4d, init)
88 DEFINE_ENTRY(pud, pud, init)
89 DEFINE_ENTRY(pmd, pmd, init)
90 DEFINE_ENTRY(pte, pte, init)
91 
92 
93 /*
94  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
95  * physical space so we can cache the place of the first one and move
96  * around without checking the pgd every time.
97  */
98 
99 /* Bits supported by the hardware: */
100 pteval_t __supported_pte_mask __read_mostly = ~0;
101 /* Bits allowed in normal kernel mappings: */
102 pteval_t __default_kernel_pte_mask __read_mostly = ~0;
103 EXPORT_SYMBOL_GPL(__supported_pte_mask);
104 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
105 EXPORT_SYMBOL(__default_kernel_pte_mask);
106 
107 int force_personality32;
108 
109 /*
110  * noexec32=on|off
111  * Control non executable heap for 32bit processes.
112  * To control the stack too use noexec=off
113  *
114  * on	PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
115  * off	PROT_READ implies PROT_EXEC
116  */
117 static int __init nonx32_setup(char *str)
118 {
119 	if (!strcmp(str, "on"))
120 		force_personality32 &= ~READ_IMPLIES_EXEC;
121 	else if (!strcmp(str, "off"))
122 		force_personality32 |= READ_IMPLIES_EXEC;
123 	return 1;
124 }
125 __setup("noexec32=", nonx32_setup);
126 
127 static void sync_global_pgds_l5(unsigned long start, unsigned long end)
128 {
129 	unsigned long addr;
130 
131 	for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
132 		const pgd_t *pgd_ref = pgd_offset_k(addr);
133 		struct page *page;
134 
135 		/* Check for overflow */
136 		if (addr < start)
137 			break;
138 
139 		if (pgd_none(*pgd_ref))
140 			continue;
141 
142 		spin_lock(&pgd_lock);
143 		list_for_each_entry(page, &pgd_list, lru) {
144 			pgd_t *pgd;
145 			spinlock_t *pgt_lock;
146 
147 			pgd = (pgd_t *)page_address(page) + pgd_index(addr);
148 			/* the pgt_lock only for Xen */
149 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
150 			spin_lock(pgt_lock);
151 
152 			if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
153 				BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
154 
155 			if (pgd_none(*pgd))
156 				set_pgd(pgd, *pgd_ref);
157 
158 			spin_unlock(pgt_lock);
159 		}
160 		spin_unlock(&pgd_lock);
161 	}
162 }
163 
164 static void sync_global_pgds_l4(unsigned long start, unsigned long end)
165 {
166 	unsigned long addr;
167 
168 	for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
169 		pgd_t *pgd_ref = pgd_offset_k(addr);
170 		const p4d_t *p4d_ref;
171 		struct page *page;
172 
173 		/*
174 		 * With folded p4d, pgd_none() is always false, we need to
175 		 * handle synchonization on p4d level.
176 		 */
177 		MAYBE_BUILD_BUG_ON(pgd_none(*pgd_ref));
178 		p4d_ref = p4d_offset(pgd_ref, addr);
179 
180 		if (p4d_none(*p4d_ref))
181 			continue;
182 
183 		spin_lock(&pgd_lock);
184 		list_for_each_entry(page, &pgd_list, lru) {
185 			pgd_t *pgd;
186 			p4d_t *p4d;
187 			spinlock_t *pgt_lock;
188 
189 			pgd = (pgd_t *)page_address(page) + pgd_index(addr);
190 			p4d = p4d_offset(pgd, addr);
191 			/* the pgt_lock only for Xen */
192 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
193 			spin_lock(pgt_lock);
194 
195 			if (!p4d_none(*p4d_ref) && !p4d_none(*p4d))
196 				BUG_ON(p4d_page_vaddr(*p4d)
197 				       != p4d_page_vaddr(*p4d_ref));
198 
199 			if (p4d_none(*p4d))
200 				set_p4d(p4d, *p4d_ref);
201 
202 			spin_unlock(pgt_lock);
203 		}
204 		spin_unlock(&pgd_lock);
205 	}
206 }
207 
208 /*
209  * When memory was added make sure all the processes MM have
210  * suitable PGD entries in the local PGD level page.
211  */
212 static void sync_global_pgds(unsigned long start, unsigned long end)
213 {
214 	if (pgtable_l5_enabled())
215 		sync_global_pgds_l5(start, end);
216 	else
217 		sync_global_pgds_l4(start, end);
218 }
219 
220 /*
221  * NOTE: This function is marked __ref because it calls __init function
222  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
223  */
224 static __ref void *spp_getpage(void)
225 {
226 	void *ptr;
227 
228 	if (after_bootmem)
229 		ptr = (void *) get_zeroed_page(GFP_ATOMIC);
230 	else
231 		ptr = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
232 
233 	if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
234 		panic("set_pte_phys: cannot allocate page data %s\n",
235 			after_bootmem ? "after bootmem" : "");
236 	}
237 
238 	pr_debug("spp_getpage %p\n", ptr);
239 
240 	return ptr;
241 }
242 
243 static p4d_t *fill_p4d(pgd_t *pgd, unsigned long vaddr)
244 {
245 	if (pgd_none(*pgd)) {
246 		p4d_t *p4d = (p4d_t *)spp_getpage();
247 		pgd_populate(&init_mm, pgd, p4d);
248 		if (p4d != p4d_offset(pgd, 0))
249 			printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
250 			       p4d, p4d_offset(pgd, 0));
251 	}
252 	return p4d_offset(pgd, vaddr);
253 }
254 
255 static pud_t *fill_pud(p4d_t *p4d, unsigned long vaddr)
256 {
257 	if (p4d_none(*p4d)) {
258 		pud_t *pud = (pud_t *)spp_getpage();
259 		p4d_populate(&init_mm, p4d, pud);
260 		if (pud != pud_offset(p4d, 0))
261 			printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
262 			       pud, pud_offset(p4d, 0));
263 	}
264 	return pud_offset(p4d, vaddr);
265 }
266 
267 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
268 {
269 	if (pud_none(*pud)) {
270 		pmd_t *pmd = (pmd_t *) spp_getpage();
271 		pud_populate(&init_mm, pud, pmd);
272 		if (pmd != pmd_offset(pud, 0))
273 			printk(KERN_ERR "PAGETABLE BUG #02! %p <-> %p\n",
274 			       pmd, pmd_offset(pud, 0));
275 	}
276 	return pmd_offset(pud, vaddr);
277 }
278 
279 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
280 {
281 	if (pmd_none(*pmd)) {
282 		pte_t *pte = (pte_t *) spp_getpage();
283 		pmd_populate_kernel(&init_mm, pmd, pte);
284 		if (pte != pte_offset_kernel(pmd, 0))
285 			printk(KERN_ERR "PAGETABLE BUG #03!\n");
286 	}
287 	return pte_offset_kernel(pmd, vaddr);
288 }
289 
290 static void __set_pte_vaddr(pud_t *pud, unsigned long vaddr, pte_t new_pte)
291 {
292 	pmd_t *pmd = fill_pmd(pud, vaddr);
293 	pte_t *pte = fill_pte(pmd, vaddr);
294 
295 	set_pte(pte, new_pte);
296 
297 	/*
298 	 * It's enough to flush this one mapping.
299 	 * (PGE mappings get flushed as well)
300 	 */
301 	flush_tlb_one_kernel(vaddr);
302 }
303 
304 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte)
305 {
306 	p4d_t *p4d = p4d_page + p4d_index(vaddr);
307 	pud_t *pud = fill_pud(p4d, vaddr);
308 
309 	__set_pte_vaddr(pud, vaddr, new_pte);
310 }
311 
312 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
313 {
314 	pud_t *pud = pud_page + pud_index(vaddr);
315 
316 	__set_pte_vaddr(pud, vaddr, new_pte);
317 }
318 
319 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
320 {
321 	pgd_t *pgd;
322 	p4d_t *p4d_page;
323 
324 	pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
325 
326 	pgd = pgd_offset_k(vaddr);
327 	if (pgd_none(*pgd)) {
328 		printk(KERN_ERR
329 			"PGD FIXMAP MISSING, it should be setup in head.S!\n");
330 		return;
331 	}
332 
333 	p4d_page = p4d_offset(pgd, 0);
334 	set_pte_vaddr_p4d(p4d_page, vaddr, pteval);
335 }
336 
337 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
338 {
339 	pgd_t *pgd;
340 	p4d_t *p4d;
341 	pud_t *pud;
342 
343 	pgd = pgd_offset_k(vaddr);
344 	p4d = fill_p4d(pgd, vaddr);
345 	pud = fill_pud(p4d, vaddr);
346 	return fill_pmd(pud, vaddr);
347 }
348 
349 pte_t * __init populate_extra_pte(unsigned long vaddr)
350 {
351 	pmd_t *pmd;
352 
353 	pmd = populate_extra_pmd(vaddr);
354 	return fill_pte(pmd, vaddr);
355 }
356 
357 /*
358  * Create large page table mappings for a range of physical addresses.
359  */
360 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
361 					enum page_cache_mode cache)
362 {
363 	pgd_t *pgd;
364 	p4d_t *p4d;
365 	pud_t *pud;
366 	pmd_t *pmd;
367 	pgprot_t prot;
368 
369 	pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
370 		protval_4k_2_large(cachemode2protval(cache));
371 	BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
372 	for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
373 		pgd = pgd_offset_k((unsigned long)__va(phys));
374 		if (pgd_none(*pgd)) {
375 			p4d = (p4d_t *) spp_getpage();
376 			set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE |
377 						_PAGE_USER));
378 		}
379 		p4d = p4d_offset(pgd, (unsigned long)__va(phys));
380 		if (p4d_none(*p4d)) {
381 			pud = (pud_t *) spp_getpage();
382 			set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE |
383 						_PAGE_USER));
384 		}
385 		pud = pud_offset(p4d, (unsigned long)__va(phys));
386 		if (pud_none(*pud)) {
387 			pmd = (pmd_t *) spp_getpage();
388 			set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
389 						_PAGE_USER));
390 		}
391 		pmd = pmd_offset(pud, phys);
392 		BUG_ON(!pmd_none(*pmd));
393 		set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
394 	}
395 }
396 
397 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
398 {
399 	__init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
400 }
401 
402 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
403 {
404 	__init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
405 }
406 
407 /*
408  * The head.S code sets up the kernel high mapping:
409  *
410  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
411  *
412  * phys_base holds the negative offset to the kernel, which is added
413  * to the compile time generated pmds. This results in invalid pmds up
414  * to the point where we hit the physaddr 0 mapping.
415  *
416  * We limit the mappings to the region from _text to _brk_end.  _brk_end
417  * is rounded up to the 2MB boundary. This catches the invalid pmds as
418  * well, as they are located before _text:
419  */
420 void __init cleanup_highmap(void)
421 {
422 	unsigned long vaddr = __START_KERNEL_map;
423 	unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
424 	unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
425 	pmd_t *pmd = level2_kernel_pgt;
426 
427 	/*
428 	 * Native path, max_pfn_mapped is not set yet.
429 	 * Xen has valid max_pfn_mapped set in
430 	 *	arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
431 	 */
432 	if (max_pfn_mapped)
433 		vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
434 
435 	for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
436 		if (pmd_none(*pmd))
437 			continue;
438 		if (vaddr < (unsigned long) _text || vaddr > end)
439 			set_pmd(pmd, __pmd(0));
440 	}
441 }
442 
443 /*
444  * Create PTE level page table mapping for physical addresses.
445  * It returns the last physical address mapped.
446  */
447 static unsigned long __meminit
448 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
449 	      pgprot_t prot, bool init)
450 {
451 	unsigned long pages = 0, paddr_next;
452 	unsigned long paddr_last = paddr_end;
453 	pte_t *pte;
454 	int i;
455 
456 	pte = pte_page + pte_index(paddr);
457 	i = pte_index(paddr);
458 
459 	for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
460 		paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
461 		if (paddr >= paddr_end) {
462 			if (!after_bootmem &&
463 			    !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
464 					     E820_TYPE_RAM) &&
465 			    !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
466 					     E820_TYPE_RESERVED_KERN))
467 				set_pte_init(pte, __pte(0), init);
468 			continue;
469 		}
470 
471 		/*
472 		 * We will re-use the existing mapping.
473 		 * Xen for example has some special requirements, like mapping
474 		 * pagetable pages as RO. So assume someone who pre-setup
475 		 * these mappings are more intelligent.
476 		 */
477 		if (!pte_none(*pte)) {
478 			if (!after_bootmem)
479 				pages++;
480 			continue;
481 		}
482 
483 		if (0)
484 			pr_info("   pte=%p addr=%lx pte=%016lx\n", pte, paddr,
485 				pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
486 		pages++;
487 		set_pte_init(pte, pfn_pte(paddr >> PAGE_SHIFT, prot), init);
488 		paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
489 	}
490 
491 	update_page_count(PG_LEVEL_4K, pages);
492 
493 	return paddr_last;
494 }
495 
496 /*
497  * Create PMD level page table mapping for physical addresses. The virtual
498  * and physical address have to be aligned at this level.
499  * It returns the last physical address mapped.
500  */
501 static unsigned long __meminit
502 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
503 	      unsigned long page_size_mask, pgprot_t prot, bool init)
504 {
505 	unsigned long pages = 0, paddr_next;
506 	unsigned long paddr_last = paddr_end;
507 
508 	int i = pmd_index(paddr);
509 
510 	for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
511 		pmd_t *pmd = pmd_page + pmd_index(paddr);
512 		pte_t *pte;
513 		pgprot_t new_prot = prot;
514 
515 		paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
516 		if (paddr >= paddr_end) {
517 			if (!after_bootmem &&
518 			    !e820__mapped_any(paddr & PMD_MASK, paddr_next,
519 					     E820_TYPE_RAM) &&
520 			    !e820__mapped_any(paddr & PMD_MASK, paddr_next,
521 					     E820_TYPE_RESERVED_KERN))
522 				set_pmd_init(pmd, __pmd(0), init);
523 			continue;
524 		}
525 
526 		if (!pmd_none(*pmd)) {
527 			if (!pmd_large(*pmd)) {
528 				spin_lock(&init_mm.page_table_lock);
529 				pte = (pte_t *)pmd_page_vaddr(*pmd);
530 				paddr_last = phys_pte_init(pte, paddr,
531 							   paddr_end, prot,
532 							   init);
533 				spin_unlock(&init_mm.page_table_lock);
534 				continue;
535 			}
536 			/*
537 			 * If we are ok with PG_LEVEL_2M mapping, then we will
538 			 * use the existing mapping,
539 			 *
540 			 * Otherwise, we will split the large page mapping but
541 			 * use the same existing protection bits except for
542 			 * large page, so that we don't violate Intel's TLB
543 			 * Application note (317080) which says, while changing
544 			 * the page sizes, new and old translations should
545 			 * not differ with respect to page frame and
546 			 * attributes.
547 			 */
548 			if (page_size_mask & (1 << PG_LEVEL_2M)) {
549 				if (!after_bootmem)
550 					pages++;
551 				paddr_last = paddr_next;
552 				continue;
553 			}
554 			new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
555 		}
556 
557 		if (page_size_mask & (1<<PG_LEVEL_2M)) {
558 			pages++;
559 			spin_lock(&init_mm.page_table_lock);
560 			set_pte_init((pte_t *)pmd,
561 				     pfn_pte((paddr & PMD_MASK) >> PAGE_SHIFT,
562 					     __pgprot(pgprot_val(prot) | _PAGE_PSE)),
563 				     init);
564 			spin_unlock(&init_mm.page_table_lock);
565 			paddr_last = paddr_next;
566 			continue;
567 		}
568 
569 		pte = alloc_low_page();
570 		paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot, init);
571 
572 		spin_lock(&init_mm.page_table_lock);
573 		pmd_populate_kernel_init(&init_mm, pmd, pte, init);
574 		spin_unlock(&init_mm.page_table_lock);
575 	}
576 	update_page_count(PG_LEVEL_2M, pages);
577 	return paddr_last;
578 }
579 
580 /*
581  * Create PUD level page table mapping for physical addresses. The virtual
582  * and physical address do not have to be aligned at this level. KASLR can
583  * randomize virtual addresses up to this level.
584  * It returns the last physical address mapped.
585  */
586 static unsigned long __meminit
587 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
588 	      unsigned long page_size_mask, pgprot_t _prot, bool init)
589 {
590 	unsigned long pages = 0, paddr_next;
591 	unsigned long paddr_last = paddr_end;
592 	unsigned long vaddr = (unsigned long)__va(paddr);
593 	int i = pud_index(vaddr);
594 
595 	for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
596 		pud_t *pud;
597 		pmd_t *pmd;
598 		pgprot_t prot = _prot;
599 
600 		vaddr = (unsigned long)__va(paddr);
601 		pud = pud_page + pud_index(vaddr);
602 		paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
603 
604 		if (paddr >= paddr_end) {
605 			if (!after_bootmem &&
606 			    !e820__mapped_any(paddr & PUD_MASK, paddr_next,
607 					     E820_TYPE_RAM) &&
608 			    !e820__mapped_any(paddr & PUD_MASK, paddr_next,
609 					     E820_TYPE_RESERVED_KERN))
610 				set_pud_init(pud, __pud(0), init);
611 			continue;
612 		}
613 
614 		if (!pud_none(*pud)) {
615 			if (!pud_large(*pud)) {
616 				pmd = pmd_offset(pud, 0);
617 				paddr_last = phys_pmd_init(pmd, paddr,
618 							   paddr_end,
619 							   page_size_mask,
620 							   prot, init);
621 				continue;
622 			}
623 			/*
624 			 * If we are ok with PG_LEVEL_1G mapping, then we will
625 			 * use the existing mapping.
626 			 *
627 			 * Otherwise, we will split the gbpage mapping but use
628 			 * the same existing protection  bits except for large
629 			 * page, so that we don't violate Intel's TLB
630 			 * Application note (317080) which says, while changing
631 			 * the page sizes, new and old translations should
632 			 * not differ with respect to page frame and
633 			 * attributes.
634 			 */
635 			if (page_size_mask & (1 << PG_LEVEL_1G)) {
636 				if (!after_bootmem)
637 					pages++;
638 				paddr_last = paddr_next;
639 				continue;
640 			}
641 			prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
642 		}
643 
644 		if (page_size_mask & (1<<PG_LEVEL_1G)) {
645 			pages++;
646 			spin_lock(&init_mm.page_table_lock);
647 
648 			prot = __pgprot(pgprot_val(prot) | __PAGE_KERNEL_LARGE);
649 
650 			set_pte_init((pte_t *)pud,
651 				     pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
652 					     prot),
653 				     init);
654 			spin_unlock(&init_mm.page_table_lock);
655 			paddr_last = paddr_next;
656 			continue;
657 		}
658 
659 		pmd = alloc_low_page();
660 		paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
661 					   page_size_mask, prot, init);
662 
663 		spin_lock(&init_mm.page_table_lock);
664 		pud_populate_init(&init_mm, pud, pmd, init);
665 		spin_unlock(&init_mm.page_table_lock);
666 	}
667 
668 	update_page_count(PG_LEVEL_1G, pages);
669 
670 	return paddr_last;
671 }
672 
673 static unsigned long __meminit
674 phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
675 	      unsigned long page_size_mask, pgprot_t prot, bool init)
676 {
677 	unsigned long vaddr, vaddr_end, vaddr_next, paddr_next, paddr_last;
678 
679 	paddr_last = paddr_end;
680 	vaddr = (unsigned long)__va(paddr);
681 	vaddr_end = (unsigned long)__va(paddr_end);
682 
683 	if (!pgtable_l5_enabled())
684 		return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
685 				     page_size_mask, prot, init);
686 
687 	for (; vaddr < vaddr_end; vaddr = vaddr_next) {
688 		p4d_t *p4d = p4d_page + p4d_index(vaddr);
689 		pud_t *pud;
690 
691 		vaddr_next = (vaddr & P4D_MASK) + P4D_SIZE;
692 		paddr = __pa(vaddr);
693 
694 		if (paddr >= paddr_end) {
695 			paddr_next = __pa(vaddr_next);
696 			if (!after_bootmem &&
697 			    !e820__mapped_any(paddr & P4D_MASK, paddr_next,
698 					     E820_TYPE_RAM) &&
699 			    !e820__mapped_any(paddr & P4D_MASK, paddr_next,
700 					     E820_TYPE_RESERVED_KERN))
701 				set_p4d_init(p4d, __p4d(0), init);
702 			continue;
703 		}
704 
705 		if (!p4d_none(*p4d)) {
706 			pud = pud_offset(p4d, 0);
707 			paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
708 					page_size_mask, prot, init);
709 			continue;
710 		}
711 
712 		pud = alloc_low_page();
713 		paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
714 					   page_size_mask, prot, init);
715 
716 		spin_lock(&init_mm.page_table_lock);
717 		p4d_populate_init(&init_mm, p4d, pud, init);
718 		spin_unlock(&init_mm.page_table_lock);
719 	}
720 
721 	return paddr_last;
722 }
723 
724 static unsigned long __meminit
725 __kernel_physical_mapping_init(unsigned long paddr_start,
726 			       unsigned long paddr_end,
727 			       unsigned long page_size_mask,
728 			       pgprot_t prot, bool init)
729 {
730 	bool pgd_changed = false;
731 	unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
732 
733 	paddr_last = paddr_end;
734 	vaddr = (unsigned long)__va(paddr_start);
735 	vaddr_end = (unsigned long)__va(paddr_end);
736 	vaddr_start = vaddr;
737 
738 	for (; vaddr < vaddr_end; vaddr = vaddr_next) {
739 		pgd_t *pgd = pgd_offset_k(vaddr);
740 		p4d_t *p4d;
741 
742 		vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
743 
744 		if (pgd_val(*pgd)) {
745 			p4d = (p4d_t *)pgd_page_vaddr(*pgd);
746 			paddr_last = phys_p4d_init(p4d, __pa(vaddr),
747 						   __pa(vaddr_end),
748 						   page_size_mask,
749 						   prot, init);
750 			continue;
751 		}
752 
753 		p4d = alloc_low_page();
754 		paddr_last = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
755 					   page_size_mask, prot, init);
756 
757 		spin_lock(&init_mm.page_table_lock);
758 		if (pgtable_l5_enabled())
759 			pgd_populate_init(&init_mm, pgd, p4d, init);
760 		else
761 			p4d_populate_init(&init_mm, p4d_offset(pgd, vaddr),
762 					  (pud_t *) p4d, init);
763 
764 		spin_unlock(&init_mm.page_table_lock);
765 		pgd_changed = true;
766 	}
767 
768 	if (pgd_changed)
769 		sync_global_pgds(vaddr_start, vaddr_end - 1);
770 
771 	return paddr_last;
772 }
773 
774 
775 /*
776  * Create page table mapping for the physical memory for specific physical
777  * addresses. Note that it can only be used to populate non-present entries.
778  * The virtual and physical addresses have to be aligned on PMD level
779  * down. It returns the last physical address mapped.
780  */
781 unsigned long __meminit
782 kernel_physical_mapping_init(unsigned long paddr_start,
783 			     unsigned long paddr_end,
784 			     unsigned long page_size_mask, pgprot_t prot)
785 {
786 	return __kernel_physical_mapping_init(paddr_start, paddr_end,
787 					      page_size_mask, prot, true);
788 }
789 
790 /*
791  * This function is similar to kernel_physical_mapping_init() above with the
792  * exception that it uses set_{pud,pmd}() instead of the set_{pud,pte}_safe()
793  * when updating the mapping. The caller is responsible to flush the TLBs after
794  * the function returns.
795  */
796 unsigned long __meminit
797 kernel_physical_mapping_change(unsigned long paddr_start,
798 			       unsigned long paddr_end,
799 			       unsigned long page_size_mask)
800 {
801 	return __kernel_physical_mapping_init(paddr_start, paddr_end,
802 					      page_size_mask, PAGE_KERNEL,
803 					      false);
804 }
805 
806 #ifndef CONFIG_NUMA
807 void __init initmem_init(void)
808 {
809 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
810 }
811 #endif
812 
813 void __init paging_init(void)
814 {
815 	sparse_init();
816 
817 	/*
818 	 * clear the default setting with node 0
819 	 * note: don't use nodes_clear here, that is really clearing when
820 	 *	 numa support is not compiled in, and later node_set_state
821 	 *	 will not set it back.
822 	 */
823 	node_clear_state(0, N_MEMORY);
824 	node_clear_state(0, N_NORMAL_MEMORY);
825 
826 	zone_sizes_init();
827 }
828 
829 /*
830  * Memory hotplug specific functions
831  */
832 #ifdef CONFIG_MEMORY_HOTPLUG
833 /*
834  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
835  * updating.
836  */
837 static void update_end_of_memory_vars(u64 start, u64 size)
838 {
839 	unsigned long end_pfn = PFN_UP(start + size);
840 
841 	if (end_pfn > max_pfn) {
842 		max_pfn = end_pfn;
843 		max_low_pfn = end_pfn;
844 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
845 	}
846 }
847 
848 int add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
849 	      struct mhp_params *params)
850 {
851 	int ret;
852 
853 	ret = __add_pages(nid, start_pfn, nr_pages, params);
854 	WARN_ON_ONCE(ret);
855 
856 	/* update max_pfn, max_low_pfn and high_memory */
857 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
858 				  nr_pages << PAGE_SHIFT);
859 
860 	return ret;
861 }
862 
863 int arch_add_memory(int nid, u64 start, u64 size,
864 		    struct mhp_params *params)
865 {
866 	unsigned long start_pfn = start >> PAGE_SHIFT;
867 	unsigned long nr_pages = size >> PAGE_SHIFT;
868 
869 	init_memory_mapping(start, start + size, params->pgprot);
870 
871 	return add_pages(nid, start_pfn, nr_pages, params);
872 }
873 
874 #define PAGE_INUSE 0xFD
875 
876 static void __meminit free_pagetable(struct page *page, int order)
877 {
878 	unsigned long magic;
879 	unsigned int nr_pages = 1 << order;
880 
881 	/* bootmem page has reserved flag */
882 	if (PageReserved(page)) {
883 		__ClearPageReserved(page);
884 
885 		magic = (unsigned long)page->freelist;
886 		if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
887 			while (nr_pages--)
888 				put_page_bootmem(page++);
889 		} else
890 			while (nr_pages--)
891 				free_reserved_page(page++);
892 	} else
893 		free_pages((unsigned long)page_address(page), order);
894 }
895 
896 static void __meminit free_hugepage_table(struct page *page,
897 		struct vmem_altmap *altmap)
898 {
899 	if (altmap)
900 		vmem_altmap_free(altmap, PMD_SIZE / PAGE_SIZE);
901 	else
902 		free_pagetable(page, get_order(PMD_SIZE));
903 }
904 
905 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
906 {
907 	pte_t *pte;
908 	int i;
909 
910 	for (i = 0; i < PTRS_PER_PTE; i++) {
911 		pte = pte_start + i;
912 		if (!pte_none(*pte))
913 			return;
914 	}
915 
916 	/* free a pte talbe */
917 	free_pagetable(pmd_page(*pmd), 0);
918 	spin_lock(&init_mm.page_table_lock);
919 	pmd_clear(pmd);
920 	spin_unlock(&init_mm.page_table_lock);
921 }
922 
923 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
924 {
925 	pmd_t *pmd;
926 	int i;
927 
928 	for (i = 0; i < PTRS_PER_PMD; i++) {
929 		pmd = pmd_start + i;
930 		if (!pmd_none(*pmd))
931 			return;
932 	}
933 
934 	/* free a pmd talbe */
935 	free_pagetable(pud_page(*pud), 0);
936 	spin_lock(&init_mm.page_table_lock);
937 	pud_clear(pud);
938 	spin_unlock(&init_mm.page_table_lock);
939 }
940 
941 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
942 {
943 	pud_t *pud;
944 	int i;
945 
946 	for (i = 0; i < PTRS_PER_PUD; i++) {
947 		pud = pud_start + i;
948 		if (!pud_none(*pud))
949 			return;
950 	}
951 
952 	/* free a pud talbe */
953 	free_pagetable(p4d_page(*p4d), 0);
954 	spin_lock(&init_mm.page_table_lock);
955 	p4d_clear(p4d);
956 	spin_unlock(&init_mm.page_table_lock);
957 }
958 
959 static void __meminit
960 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
961 		 bool direct)
962 {
963 	unsigned long next, pages = 0;
964 	pte_t *pte;
965 	void *page_addr;
966 	phys_addr_t phys_addr;
967 
968 	pte = pte_start + pte_index(addr);
969 	for (; addr < end; addr = next, pte++) {
970 		next = (addr + PAGE_SIZE) & PAGE_MASK;
971 		if (next > end)
972 			next = end;
973 
974 		if (!pte_present(*pte))
975 			continue;
976 
977 		/*
978 		 * We mapped [0,1G) memory as identity mapping when
979 		 * initializing, in arch/x86/kernel/head_64.S. These
980 		 * pagetables cannot be removed.
981 		 */
982 		phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
983 		if (phys_addr < (phys_addr_t)0x40000000)
984 			return;
985 
986 		if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
987 			/*
988 			 * Do not free direct mapping pages since they were
989 			 * freed when offlining, or simplely not in use.
990 			 */
991 			if (!direct)
992 				free_pagetable(pte_page(*pte), 0);
993 
994 			spin_lock(&init_mm.page_table_lock);
995 			pte_clear(&init_mm, addr, pte);
996 			spin_unlock(&init_mm.page_table_lock);
997 
998 			/* For non-direct mapping, pages means nothing. */
999 			pages++;
1000 		} else {
1001 			/*
1002 			 * If we are here, we are freeing vmemmap pages since
1003 			 * direct mapped memory ranges to be freed are aligned.
1004 			 *
1005 			 * If we are not removing the whole page, it means
1006 			 * other page structs in this page are being used and
1007 			 * we canot remove them. So fill the unused page_structs
1008 			 * with 0xFD, and remove the page when it is wholly
1009 			 * filled with 0xFD.
1010 			 */
1011 			memset((void *)addr, PAGE_INUSE, next - addr);
1012 
1013 			page_addr = page_address(pte_page(*pte));
1014 			if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
1015 				free_pagetable(pte_page(*pte), 0);
1016 
1017 				spin_lock(&init_mm.page_table_lock);
1018 				pte_clear(&init_mm, addr, pte);
1019 				spin_unlock(&init_mm.page_table_lock);
1020 			}
1021 		}
1022 	}
1023 
1024 	/* Call free_pte_table() in remove_pmd_table(). */
1025 	flush_tlb_all();
1026 	if (direct)
1027 		update_page_count(PG_LEVEL_4K, -pages);
1028 }
1029 
1030 static void __meminit
1031 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
1032 		 bool direct, struct vmem_altmap *altmap)
1033 {
1034 	unsigned long next, pages = 0;
1035 	pte_t *pte_base;
1036 	pmd_t *pmd;
1037 	void *page_addr;
1038 
1039 	pmd = pmd_start + pmd_index(addr);
1040 	for (; addr < end; addr = next, pmd++) {
1041 		next = pmd_addr_end(addr, end);
1042 
1043 		if (!pmd_present(*pmd))
1044 			continue;
1045 
1046 		if (pmd_large(*pmd)) {
1047 			if (IS_ALIGNED(addr, PMD_SIZE) &&
1048 			    IS_ALIGNED(next, PMD_SIZE)) {
1049 				if (!direct)
1050 					free_hugepage_table(pmd_page(*pmd),
1051 							    altmap);
1052 
1053 				spin_lock(&init_mm.page_table_lock);
1054 				pmd_clear(pmd);
1055 				spin_unlock(&init_mm.page_table_lock);
1056 				pages++;
1057 			} else {
1058 				/* If here, we are freeing vmemmap pages. */
1059 				memset((void *)addr, PAGE_INUSE, next - addr);
1060 
1061 				page_addr = page_address(pmd_page(*pmd));
1062 				if (!memchr_inv(page_addr, PAGE_INUSE,
1063 						PMD_SIZE)) {
1064 					free_hugepage_table(pmd_page(*pmd),
1065 							    altmap);
1066 
1067 					spin_lock(&init_mm.page_table_lock);
1068 					pmd_clear(pmd);
1069 					spin_unlock(&init_mm.page_table_lock);
1070 				}
1071 			}
1072 
1073 			continue;
1074 		}
1075 
1076 		pte_base = (pte_t *)pmd_page_vaddr(*pmd);
1077 		remove_pte_table(pte_base, addr, next, direct);
1078 		free_pte_table(pte_base, pmd);
1079 	}
1080 
1081 	/* Call free_pmd_table() in remove_pud_table(). */
1082 	if (direct)
1083 		update_page_count(PG_LEVEL_2M, -pages);
1084 }
1085 
1086 static void __meminit
1087 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
1088 		 struct vmem_altmap *altmap, bool direct)
1089 {
1090 	unsigned long next, pages = 0;
1091 	pmd_t *pmd_base;
1092 	pud_t *pud;
1093 	void *page_addr;
1094 
1095 	pud = pud_start + pud_index(addr);
1096 	for (; addr < end; addr = next, pud++) {
1097 		next = pud_addr_end(addr, end);
1098 
1099 		if (!pud_present(*pud))
1100 			continue;
1101 
1102 		if (pud_large(*pud)) {
1103 			if (IS_ALIGNED(addr, PUD_SIZE) &&
1104 			    IS_ALIGNED(next, PUD_SIZE)) {
1105 				if (!direct)
1106 					free_pagetable(pud_page(*pud),
1107 						       get_order(PUD_SIZE));
1108 
1109 				spin_lock(&init_mm.page_table_lock);
1110 				pud_clear(pud);
1111 				spin_unlock(&init_mm.page_table_lock);
1112 				pages++;
1113 			} else {
1114 				/* If here, we are freeing vmemmap pages. */
1115 				memset((void *)addr, PAGE_INUSE, next - addr);
1116 
1117 				page_addr = page_address(pud_page(*pud));
1118 				if (!memchr_inv(page_addr, PAGE_INUSE,
1119 						PUD_SIZE)) {
1120 					free_pagetable(pud_page(*pud),
1121 						       get_order(PUD_SIZE));
1122 
1123 					spin_lock(&init_mm.page_table_lock);
1124 					pud_clear(pud);
1125 					spin_unlock(&init_mm.page_table_lock);
1126 				}
1127 			}
1128 
1129 			continue;
1130 		}
1131 
1132 		pmd_base = pmd_offset(pud, 0);
1133 		remove_pmd_table(pmd_base, addr, next, direct, altmap);
1134 		free_pmd_table(pmd_base, pud);
1135 	}
1136 
1137 	if (direct)
1138 		update_page_count(PG_LEVEL_1G, -pages);
1139 }
1140 
1141 static void __meminit
1142 remove_p4d_table(p4d_t *p4d_start, unsigned long addr, unsigned long end,
1143 		 struct vmem_altmap *altmap, bool direct)
1144 {
1145 	unsigned long next, pages = 0;
1146 	pud_t *pud_base;
1147 	p4d_t *p4d;
1148 
1149 	p4d = p4d_start + p4d_index(addr);
1150 	for (; addr < end; addr = next, p4d++) {
1151 		next = p4d_addr_end(addr, end);
1152 
1153 		if (!p4d_present(*p4d))
1154 			continue;
1155 
1156 		BUILD_BUG_ON(p4d_large(*p4d));
1157 
1158 		pud_base = pud_offset(p4d, 0);
1159 		remove_pud_table(pud_base, addr, next, altmap, direct);
1160 		/*
1161 		 * For 4-level page tables we do not want to free PUDs, but in the
1162 		 * 5-level case we should free them. This code will have to change
1163 		 * to adapt for boot-time switching between 4 and 5 level page tables.
1164 		 */
1165 		if (pgtable_l5_enabled())
1166 			free_pud_table(pud_base, p4d);
1167 	}
1168 
1169 	if (direct)
1170 		update_page_count(PG_LEVEL_512G, -pages);
1171 }
1172 
1173 /* start and end are both virtual address. */
1174 static void __meminit
1175 remove_pagetable(unsigned long start, unsigned long end, bool direct,
1176 		struct vmem_altmap *altmap)
1177 {
1178 	unsigned long next;
1179 	unsigned long addr;
1180 	pgd_t *pgd;
1181 	p4d_t *p4d;
1182 
1183 	for (addr = start; addr < end; addr = next) {
1184 		next = pgd_addr_end(addr, end);
1185 
1186 		pgd = pgd_offset_k(addr);
1187 		if (!pgd_present(*pgd))
1188 			continue;
1189 
1190 		p4d = p4d_offset(pgd, 0);
1191 		remove_p4d_table(p4d, addr, next, altmap, direct);
1192 	}
1193 
1194 	flush_tlb_all();
1195 }
1196 
1197 void __ref vmemmap_free(unsigned long start, unsigned long end,
1198 		struct vmem_altmap *altmap)
1199 {
1200 	remove_pagetable(start, end, false, altmap);
1201 }
1202 
1203 static void __meminit
1204 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1205 {
1206 	start = (unsigned long)__va(start);
1207 	end = (unsigned long)__va(end);
1208 
1209 	remove_pagetable(start, end, true, NULL);
1210 }
1211 
1212 void __ref arch_remove_memory(int nid, u64 start, u64 size,
1213 			      struct vmem_altmap *altmap)
1214 {
1215 	unsigned long start_pfn = start >> PAGE_SHIFT;
1216 	unsigned long nr_pages = size >> PAGE_SHIFT;
1217 
1218 	__remove_pages(start_pfn, nr_pages, altmap);
1219 	kernel_physical_mapping_remove(start, start + size);
1220 }
1221 #endif /* CONFIG_MEMORY_HOTPLUG */
1222 
1223 static struct kcore_list kcore_vsyscall;
1224 
1225 static void __init register_page_bootmem_info(void)
1226 {
1227 #ifdef CONFIG_NUMA
1228 	int i;
1229 
1230 	for_each_online_node(i)
1231 		register_page_bootmem_info_node(NODE_DATA(i));
1232 #endif
1233 }
1234 
1235 /*
1236  * Pre-allocates page-table pages for the vmalloc area in the kernel page-table.
1237  * Only the level which needs to be synchronized between all page-tables is
1238  * allocated because the synchronization can be expensive.
1239  */
1240 static void __init preallocate_vmalloc_pages(void)
1241 {
1242 	unsigned long addr;
1243 	const char *lvl;
1244 
1245 	for (addr = VMALLOC_START; addr <= VMALLOC_END; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1246 		pgd_t *pgd = pgd_offset_k(addr);
1247 		p4d_t *p4d;
1248 		pud_t *pud;
1249 
1250 		lvl = "p4d";
1251 		p4d = p4d_alloc(&init_mm, pgd, addr);
1252 		if (!p4d)
1253 			goto failed;
1254 
1255 		if (pgtable_l5_enabled())
1256 			continue;
1257 
1258 		/*
1259 		 * The goal here is to allocate all possibly required
1260 		 * hardware page tables pointed to by the top hardware
1261 		 * level.
1262 		 *
1263 		 * On 4-level systems, the P4D layer is folded away and
1264 		 * the above code does no preallocation.  Below, go down
1265 		 * to the pud _software_ level to ensure the second
1266 		 * hardware level is allocated on 4-level systems too.
1267 		 */
1268 		lvl = "pud";
1269 		pud = pud_alloc(&init_mm, p4d, addr);
1270 		if (!pud)
1271 			goto failed;
1272 	}
1273 
1274 	return;
1275 
1276 failed:
1277 
1278 	/*
1279 	 * The pages have to be there now or they will be missing in
1280 	 * process page-tables later.
1281 	 */
1282 	panic("Failed to pre-allocate %s pages for vmalloc area\n", lvl);
1283 }
1284 
1285 void __init mem_init(void)
1286 {
1287 	pci_iommu_alloc();
1288 
1289 	/* clear_bss() already clear the empty_zero_page */
1290 
1291 	/* this will put all memory onto the freelists */
1292 	memblock_free_all();
1293 	after_bootmem = 1;
1294 	x86_init.hyper.init_after_bootmem();
1295 
1296 	/*
1297 	 * Must be done after boot memory is put on freelist, because here we
1298 	 * might set fields in deferred struct pages that have not yet been
1299 	 * initialized, and memblock_free_all() initializes all the reserved
1300 	 * deferred pages for us.
1301 	 */
1302 	register_page_bootmem_info();
1303 
1304 	/* Register memory areas for /proc/kcore */
1305 	if (get_gate_vma(&init_mm))
1306 		kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR, PAGE_SIZE, KCORE_USER);
1307 
1308 	preallocate_vmalloc_pages();
1309 
1310 	mem_init_print_info(NULL);
1311 }
1312 
1313 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1314 int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask)
1315 {
1316 	/*
1317 	 * More CPUs always led to greater speedups on tested systems, up to
1318 	 * all the nodes' CPUs.  Use all since the system is otherwise idle
1319 	 * now.
1320 	 */
1321 	return max_t(int, cpumask_weight(node_cpumask), 1);
1322 }
1323 #endif
1324 
1325 int kernel_set_to_readonly;
1326 
1327 void mark_rodata_ro(void)
1328 {
1329 	unsigned long start = PFN_ALIGN(_text);
1330 	unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1331 	unsigned long end = (unsigned long)__end_rodata_hpage_align;
1332 	unsigned long text_end = PFN_ALIGN(_etext);
1333 	unsigned long rodata_end = PFN_ALIGN(__end_rodata);
1334 	unsigned long all_end;
1335 
1336 	printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1337 	       (end - start) >> 10);
1338 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1339 
1340 	kernel_set_to_readonly = 1;
1341 
1342 	/*
1343 	 * The rodata/data/bss/brk section (but not the kernel text!)
1344 	 * should also be not-executable.
1345 	 *
1346 	 * We align all_end to PMD_SIZE because the existing mapping
1347 	 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1348 	 * split the PMD and the reminder between _brk_end and the end
1349 	 * of the PMD will remain mapped executable.
1350 	 *
1351 	 * Any PMD which was setup after the one which covers _brk_end
1352 	 * has been zapped already via cleanup_highmem().
1353 	 */
1354 	all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1355 	set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1356 
1357 	set_ftrace_ops_ro();
1358 
1359 #ifdef CONFIG_CPA_DEBUG
1360 	printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1361 	set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1362 
1363 	printk(KERN_INFO "Testing CPA: again\n");
1364 	set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1365 #endif
1366 
1367 	free_kernel_image_pages("unused kernel image (text/rodata gap)",
1368 				(void *)text_end, (void *)rodata_start);
1369 	free_kernel_image_pages("unused kernel image (rodata/data gap)",
1370 				(void *)rodata_end, (void *)_sdata);
1371 
1372 	debug_checkwx();
1373 }
1374 
1375 int kern_addr_valid(unsigned long addr)
1376 {
1377 	unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1378 	pgd_t *pgd;
1379 	p4d_t *p4d;
1380 	pud_t *pud;
1381 	pmd_t *pmd;
1382 	pte_t *pte;
1383 
1384 	if (above != 0 && above != -1UL)
1385 		return 0;
1386 
1387 	pgd = pgd_offset_k(addr);
1388 	if (pgd_none(*pgd))
1389 		return 0;
1390 
1391 	p4d = p4d_offset(pgd, addr);
1392 	if (p4d_none(*p4d))
1393 		return 0;
1394 
1395 	pud = pud_offset(p4d, addr);
1396 	if (pud_none(*pud))
1397 		return 0;
1398 
1399 	if (pud_large(*pud))
1400 		return pfn_valid(pud_pfn(*pud));
1401 
1402 	pmd = pmd_offset(pud, addr);
1403 	if (pmd_none(*pmd))
1404 		return 0;
1405 
1406 	if (pmd_large(*pmd))
1407 		return pfn_valid(pmd_pfn(*pmd));
1408 
1409 	pte = pte_offset_kernel(pmd, addr);
1410 	if (pte_none(*pte))
1411 		return 0;
1412 
1413 	return pfn_valid(pte_pfn(*pte));
1414 }
1415 
1416 /*
1417  * Block size is the minimum amount of memory which can be hotplugged or
1418  * hotremoved. It must be power of two and must be equal or larger than
1419  * MIN_MEMORY_BLOCK_SIZE.
1420  */
1421 #define MAX_BLOCK_SIZE (2UL << 30)
1422 
1423 /* Amount of ram needed to start using large blocks */
1424 #define MEM_SIZE_FOR_LARGE_BLOCK (64UL << 30)
1425 
1426 /* Adjustable memory block size */
1427 static unsigned long set_memory_block_size;
1428 int __init set_memory_block_size_order(unsigned int order)
1429 {
1430 	unsigned long size = 1UL << order;
1431 
1432 	if (size > MEM_SIZE_FOR_LARGE_BLOCK || size < MIN_MEMORY_BLOCK_SIZE)
1433 		return -EINVAL;
1434 
1435 	set_memory_block_size = size;
1436 	return 0;
1437 }
1438 
1439 static unsigned long probe_memory_block_size(void)
1440 {
1441 	unsigned long boot_mem_end = max_pfn << PAGE_SHIFT;
1442 	unsigned long bz;
1443 
1444 	/* If memory block size has been set, then use it */
1445 	bz = set_memory_block_size;
1446 	if (bz)
1447 		goto done;
1448 
1449 	/* Use regular block if RAM is smaller than MEM_SIZE_FOR_LARGE_BLOCK */
1450 	if (boot_mem_end < MEM_SIZE_FOR_LARGE_BLOCK) {
1451 		bz = MIN_MEMORY_BLOCK_SIZE;
1452 		goto done;
1453 	}
1454 
1455 	/*
1456 	 * Use max block size to minimize overhead on bare metal, where
1457 	 * alignment for memory hotplug isn't a concern.
1458 	 */
1459 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1460 		bz = MAX_BLOCK_SIZE;
1461 		goto done;
1462 	}
1463 
1464 	/* Find the largest allowed block size that aligns to memory end */
1465 	for (bz = MAX_BLOCK_SIZE; bz > MIN_MEMORY_BLOCK_SIZE; bz >>= 1) {
1466 		if (IS_ALIGNED(boot_mem_end, bz))
1467 			break;
1468 	}
1469 done:
1470 	pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1471 
1472 	return bz;
1473 }
1474 
1475 static unsigned long memory_block_size_probed;
1476 unsigned long memory_block_size_bytes(void)
1477 {
1478 	if (!memory_block_size_probed)
1479 		memory_block_size_probed = probe_memory_block_size();
1480 
1481 	return memory_block_size_probed;
1482 }
1483 
1484 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1485 /*
1486  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1487  */
1488 static long __meminitdata addr_start, addr_end;
1489 static void __meminitdata *p_start, *p_end;
1490 static int __meminitdata node_start;
1491 
1492 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1493 		unsigned long end, int node, struct vmem_altmap *altmap)
1494 {
1495 	unsigned long addr;
1496 	unsigned long next;
1497 	pgd_t *pgd;
1498 	p4d_t *p4d;
1499 	pud_t *pud;
1500 	pmd_t *pmd;
1501 
1502 	for (addr = start; addr < end; addr = next) {
1503 		next = pmd_addr_end(addr, end);
1504 
1505 		pgd = vmemmap_pgd_populate(addr, node);
1506 		if (!pgd)
1507 			return -ENOMEM;
1508 
1509 		p4d = vmemmap_p4d_populate(pgd, addr, node);
1510 		if (!p4d)
1511 			return -ENOMEM;
1512 
1513 		pud = vmemmap_pud_populate(p4d, addr, node);
1514 		if (!pud)
1515 			return -ENOMEM;
1516 
1517 		pmd = pmd_offset(pud, addr);
1518 		if (pmd_none(*pmd)) {
1519 			void *p;
1520 
1521 			p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1522 			if (p) {
1523 				pte_t entry;
1524 
1525 				entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1526 						PAGE_KERNEL_LARGE);
1527 				set_pmd(pmd, __pmd(pte_val(entry)));
1528 
1529 				/* check to see if we have contiguous blocks */
1530 				if (p_end != p || node_start != node) {
1531 					if (p_start)
1532 						pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1533 						       addr_start, addr_end-1, p_start, p_end-1, node_start);
1534 					addr_start = addr;
1535 					node_start = node;
1536 					p_start = p;
1537 				}
1538 
1539 				addr_end = addr + PMD_SIZE;
1540 				p_end = p + PMD_SIZE;
1541 				continue;
1542 			} else if (altmap)
1543 				return -ENOMEM; /* no fallback */
1544 		} else if (pmd_large(*pmd)) {
1545 			vmemmap_verify((pte_t *)pmd, node, addr, next);
1546 			continue;
1547 		}
1548 		if (vmemmap_populate_basepages(addr, next, node, NULL))
1549 			return -ENOMEM;
1550 	}
1551 	return 0;
1552 }
1553 
1554 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1555 		struct vmem_altmap *altmap)
1556 {
1557 	int err;
1558 
1559 	if (end - start < PAGES_PER_SECTION * sizeof(struct page))
1560 		err = vmemmap_populate_basepages(start, end, node, NULL);
1561 	else if (boot_cpu_has(X86_FEATURE_PSE))
1562 		err = vmemmap_populate_hugepages(start, end, node, altmap);
1563 	else if (altmap) {
1564 		pr_err_once("%s: no cpu support for altmap allocations\n",
1565 				__func__);
1566 		err = -ENOMEM;
1567 	} else
1568 		err = vmemmap_populate_basepages(start, end, node, NULL);
1569 	if (!err)
1570 		sync_global_pgds(start, end - 1);
1571 	return err;
1572 }
1573 
1574 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1575 void register_page_bootmem_memmap(unsigned long section_nr,
1576 				  struct page *start_page, unsigned long nr_pages)
1577 {
1578 	unsigned long addr = (unsigned long)start_page;
1579 	unsigned long end = (unsigned long)(start_page + nr_pages);
1580 	unsigned long next;
1581 	pgd_t *pgd;
1582 	p4d_t *p4d;
1583 	pud_t *pud;
1584 	pmd_t *pmd;
1585 	unsigned int nr_pmd_pages;
1586 	struct page *page;
1587 
1588 	for (; addr < end; addr = next) {
1589 		pte_t *pte = NULL;
1590 
1591 		pgd = pgd_offset_k(addr);
1592 		if (pgd_none(*pgd)) {
1593 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1594 			continue;
1595 		}
1596 		get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1597 
1598 		p4d = p4d_offset(pgd, addr);
1599 		if (p4d_none(*p4d)) {
1600 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1601 			continue;
1602 		}
1603 		get_page_bootmem(section_nr, p4d_page(*p4d), MIX_SECTION_INFO);
1604 
1605 		pud = pud_offset(p4d, addr);
1606 		if (pud_none(*pud)) {
1607 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1608 			continue;
1609 		}
1610 		get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1611 
1612 		if (!boot_cpu_has(X86_FEATURE_PSE)) {
1613 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1614 			pmd = pmd_offset(pud, addr);
1615 			if (pmd_none(*pmd))
1616 				continue;
1617 			get_page_bootmem(section_nr, pmd_page(*pmd),
1618 					 MIX_SECTION_INFO);
1619 
1620 			pte = pte_offset_kernel(pmd, addr);
1621 			if (pte_none(*pte))
1622 				continue;
1623 			get_page_bootmem(section_nr, pte_page(*pte),
1624 					 SECTION_INFO);
1625 		} else {
1626 			next = pmd_addr_end(addr, end);
1627 
1628 			pmd = pmd_offset(pud, addr);
1629 			if (pmd_none(*pmd))
1630 				continue;
1631 
1632 			nr_pmd_pages = 1 << get_order(PMD_SIZE);
1633 			page = pmd_page(*pmd);
1634 			while (nr_pmd_pages--)
1635 				get_page_bootmem(section_nr, page++,
1636 						 SECTION_INFO);
1637 		}
1638 	}
1639 }
1640 #endif
1641 
1642 void __meminit vmemmap_populate_print_last(void)
1643 {
1644 	if (p_start) {
1645 		pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1646 			addr_start, addr_end-1, p_start, p_end-1, node_start);
1647 		p_start = NULL;
1648 		p_end = NULL;
1649 		node_start = 0;
1650 	}
1651 }
1652 #endif
1653