xref: /openbmc/linux/arch/x86/mm/init_64.c (revision 48c926cd)
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 <linux/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/api.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/set_memory.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  */
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 make sure all the processes MM have
93  * suitable PGD entries in the local PGD level page.
94  */
95 #ifdef CONFIG_X86_5LEVEL
96 void sync_global_pgds(unsigned long start, unsigned long end)
97 {
98 	unsigned long addr;
99 
100 	for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
101 		const pgd_t *pgd_ref = pgd_offset_k(addr);
102 		struct page *page;
103 
104 		/* Check for overflow */
105 		if (addr < start)
106 			break;
107 
108 		if (pgd_none(*pgd_ref))
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) != pgd_page_vaddr(*pgd_ref));
123 
124 			if (pgd_none(*pgd))
125 				set_pgd(pgd, *pgd_ref);
126 
127 			spin_unlock(pgt_lock);
128 		}
129 		spin_unlock(&pgd_lock);
130 	}
131 }
132 #else
133 void sync_global_pgds(unsigned long start, unsigned long end)
134 {
135 	unsigned long addr;
136 
137 	for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
138 		pgd_t *pgd_ref = pgd_offset_k(addr);
139 		const p4d_t *p4d_ref;
140 		struct page *page;
141 
142 		/*
143 		 * With folded p4d, pgd_none() is always false, we need to
144 		 * handle synchonization on p4d level.
145 		 */
146 		BUILD_BUG_ON(pgd_none(*pgd_ref));
147 		p4d_ref = p4d_offset(pgd_ref, addr);
148 
149 		if (p4d_none(*p4d_ref))
150 			continue;
151 
152 		spin_lock(&pgd_lock);
153 		list_for_each_entry(page, &pgd_list, lru) {
154 			pgd_t *pgd;
155 			p4d_t *p4d;
156 			spinlock_t *pgt_lock;
157 
158 			pgd = (pgd_t *)page_address(page) + pgd_index(addr);
159 			p4d = p4d_offset(pgd, addr);
160 			/* the pgt_lock only for Xen */
161 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
162 			spin_lock(pgt_lock);
163 
164 			if (!p4d_none(*p4d_ref) && !p4d_none(*p4d))
165 				BUG_ON(p4d_page_vaddr(*p4d)
166 				       != p4d_page_vaddr(*p4d_ref));
167 
168 			if (p4d_none(*p4d))
169 				set_p4d(p4d, *p4d_ref);
170 
171 			spin_unlock(pgt_lock);
172 		}
173 		spin_unlock(&pgd_lock);
174 	}
175 }
176 #endif
177 
178 /*
179  * NOTE: This function is marked __ref because it calls __init function
180  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
181  */
182 static __ref void *spp_getpage(void)
183 {
184 	void *ptr;
185 
186 	if (after_bootmem)
187 		ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
188 	else
189 		ptr = alloc_bootmem_pages(PAGE_SIZE);
190 
191 	if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
192 		panic("set_pte_phys: cannot allocate page data %s\n",
193 			after_bootmem ? "after bootmem" : "");
194 	}
195 
196 	pr_debug("spp_getpage %p\n", ptr);
197 
198 	return ptr;
199 }
200 
201 static p4d_t *fill_p4d(pgd_t *pgd, unsigned long vaddr)
202 {
203 	if (pgd_none(*pgd)) {
204 		p4d_t *p4d = (p4d_t *)spp_getpage();
205 		pgd_populate(&init_mm, pgd, p4d);
206 		if (p4d != p4d_offset(pgd, 0))
207 			printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
208 			       p4d, p4d_offset(pgd, 0));
209 	}
210 	return p4d_offset(pgd, vaddr);
211 }
212 
213 static pud_t *fill_pud(p4d_t *p4d, unsigned long vaddr)
214 {
215 	if (p4d_none(*p4d)) {
216 		pud_t *pud = (pud_t *)spp_getpage();
217 		p4d_populate(&init_mm, p4d, pud);
218 		if (pud != pud_offset(p4d, 0))
219 			printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
220 			       pud, pud_offset(p4d, 0));
221 	}
222 	return pud_offset(p4d, vaddr);
223 }
224 
225 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
226 {
227 	if (pud_none(*pud)) {
228 		pmd_t *pmd = (pmd_t *) spp_getpage();
229 		pud_populate(&init_mm, pud, pmd);
230 		if (pmd != pmd_offset(pud, 0))
231 			printk(KERN_ERR "PAGETABLE BUG #02! %p <-> %p\n",
232 			       pmd, pmd_offset(pud, 0));
233 	}
234 	return pmd_offset(pud, vaddr);
235 }
236 
237 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
238 {
239 	if (pmd_none(*pmd)) {
240 		pte_t *pte = (pte_t *) spp_getpage();
241 		pmd_populate_kernel(&init_mm, pmd, pte);
242 		if (pte != pte_offset_kernel(pmd, 0))
243 			printk(KERN_ERR "PAGETABLE BUG #03!\n");
244 	}
245 	return pte_offset_kernel(pmd, vaddr);
246 }
247 
248 static void __set_pte_vaddr(pud_t *pud, unsigned long vaddr, pte_t new_pte)
249 {
250 	pmd_t *pmd = fill_pmd(pud, vaddr);
251 	pte_t *pte = fill_pte(pmd, vaddr);
252 
253 	set_pte(pte, new_pte);
254 
255 	/*
256 	 * It's enough to flush this one mapping.
257 	 * (PGE mappings get flushed as well)
258 	 */
259 	__flush_tlb_one(vaddr);
260 }
261 
262 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte)
263 {
264 	p4d_t *p4d = p4d_page + p4d_index(vaddr);
265 	pud_t *pud = fill_pud(p4d, vaddr);
266 
267 	__set_pte_vaddr(pud, vaddr, new_pte);
268 }
269 
270 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
271 {
272 	pud_t *pud = pud_page + pud_index(vaddr);
273 
274 	__set_pte_vaddr(pud, vaddr, new_pte);
275 }
276 
277 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
278 {
279 	pgd_t *pgd;
280 	p4d_t *p4d_page;
281 
282 	pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
283 
284 	pgd = pgd_offset_k(vaddr);
285 	if (pgd_none(*pgd)) {
286 		printk(KERN_ERR
287 			"PGD FIXMAP MISSING, it should be setup in head.S!\n");
288 		return;
289 	}
290 
291 	p4d_page = p4d_offset(pgd, 0);
292 	set_pte_vaddr_p4d(p4d_page, vaddr, pteval);
293 }
294 
295 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
296 {
297 	pgd_t *pgd;
298 	p4d_t *p4d;
299 	pud_t *pud;
300 
301 	pgd = pgd_offset_k(vaddr);
302 	p4d = fill_p4d(pgd, vaddr);
303 	pud = fill_pud(p4d, vaddr);
304 	return fill_pmd(pud, vaddr);
305 }
306 
307 pte_t * __init populate_extra_pte(unsigned long vaddr)
308 {
309 	pmd_t *pmd;
310 
311 	pmd = populate_extra_pmd(vaddr);
312 	return fill_pte(pmd, vaddr);
313 }
314 
315 /*
316  * Create large page table mappings for a range of physical addresses.
317  */
318 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
319 					enum page_cache_mode cache)
320 {
321 	pgd_t *pgd;
322 	p4d_t *p4d;
323 	pud_t *pud;
324 	pmd_t *pmd;
325 	pgprot_t prot;
326 
327 	pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
328 		pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache)));
329 	BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
330 	for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
331 		pgd = pgd_offset_k((unsigned long)__va(phys));
332 		if (pgd_none(*pgd)) {
333 			p4d = (p4d_t *) spp_getpage();
334 			set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE |
335 						_PAGE_USER));
336 		}
337 		p4d = p4d_offset(pgd, (unsigned long)__va(phys));
338 		if (p4d_none(*p4d)) {
339 			pud = (pud_t *) spp_getpage();
340 			set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE |
341 						_PAGE_USER));
342 		}
343 		pud = pud_offset(p4d, (unsigned long)__va(phys));
344 		if (pud_none(*pud)) {
345 			pmd = (pmd_t *) spp_getpage();
346 			set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
347 						_PAGE_USER));
348 		}
349 		pmd = pmd_offset(pud, phys);
350 		BUG_ON(!pmd_none(*pmd));
351 		set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
352 	}
353 }
354 
355 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
356 {
357 	__init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
358 }
359 
360 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
361 {
362 	__init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
363 }
364 
365 /*
366  * The head.S code sets up the kernel high mapping:
367  *
368  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
369  *
370  * phys_base holds the negative offset to the kernel, which is added
371  * to the compile time generated pmds. This results in invalid pmds up
372  * to the point where we hit the physaddr 0 mapping.
373  *
374  * We limit the mappings to the region from _text to _brk_end.  _brk_end
375  * is rounded up to the 2MB boundary. This catches the invalid pmds as
376  * well, as they are located before _text:
377  */
378 void __init cleanup_highmap(void)
379 {
380 	unsigned long vaddr = __START_KERNEL_map;
381 	unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
382 	unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
383 	pmd_t *pmd = level2_kernel_pgt;
384 
385 	/*
386 	 * Native path, max_pfn_mapped is not set yet.
387 	 * Xen has valid max_pfn_mapped set in
388 	 *	arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
389 	 */
390 	if (max_pfn_mapped)
391 		vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
392 
393 	for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
394 		if (pmd_none(*pmd))
395 			continue;
396 		if (vaddr < (unsigned long) _text || vaddr > end)
397 			set_pmd(pmd, __pmd(0));
398 	}
399 }
400 
401 /*
402  * Create PTE level page table mapping for physical addresses.
403  * It returns the last physical address mapped.
404  */
405 static unsigned long __meminit
406 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
407 	      pgprot_t prot)
408 {
409 	unsigned long pages = 0, paddr_next;
410 	unsigned long paddr_last = paddr_end;
411 	pte_t *pte;
412 	int i;
413 
414 	pte = pte_page + pte_index(paddr);
415 	i = pte_index(paddr);
416 
417 	for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
418 		paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
419 		if (paddr >= paddr_end) {
420 			if (!after_bootmem &&
421 			    !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
422 					     E820_TYPE_RAM) &&
423 			    !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
424 					     E820_TYPE_RESERVED_KERN))
425 				set_pte(pte, __pte(0));
426 			continue;
427 		}
428 
429 		/*
430 		 * We will re-use the existing mapping.
431 		 * Xen for example has some special requirements, like mapping
432 		 * pagetable pages as RO. So assume someone who pre-setup
433 		 * these mappings are more intelligent.
434 		 */
435 		if (!pte_none(*pte)) {
436 			if (!after_bootmem)
437 				pages++;
438 			continue;
439 		}
440 
441 		if (0)
442 			pr_info("   pte=%p addr=%lx pte=%016lx\n", pte, paddr,
443 				pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
444 		pages++;
445 		set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
446 		paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
447 	}
448 
449 	update_page_count(PG_LEVEL_4K, pages);
450 
451 	return paddr_last;
452 }
453 
454 /*
455  * Create PMD level page table mapping for physical addresses. The virtual
456  * and physical address have to be aligned at this level.
457  * It returns the last physical address mapped.
458  */
459 static unsigned long __meminit
460 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
461 	      unsigned long page_size_mask, pgprot_t prot)
462 {
463 	unsigned long pages = 0, paddr_next;
464 	unsigned long paddr_last = paddr_end;
465 
466 	int i = pmd_index(paddr);
467 
468 	for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
469 		pmd_t *pmd = pmd_page + pmd_index(paddr);
470 		pte_t *pte;
471 		pgprot_t new_prot = prot;
472 
473 		paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
474 		if (paddr >= paddr_end) {
475 			if (!after_bootmem &&
476 			    !e820__mapped_any(paddr & PMD_MASK, paddr_next,
477 					     E820_TYPE_RAM) &&
478 			    !e820__mapped_any(paddr & PMD_MASK, paddr_next,
479 					     E820_TYPE_RESERVED_KERN))
480 				set_pmd(pmd, __pmd(0));
481 			continue;
482 		}
483 
484 		if (!pmd_none(*pmd)) {
485 			if (!pmd_large(*pmd)) {
486 				spin_lock(&init_mm.page_table_lock);
487 				pte = (pte_t *)pmd_page_vaddr(*pmd);
488 				paddr_last = phys_pte_init(pte, paddr,
489 							   paddr_end, prot);
490 				spin_unlock(&init_mm.page_table_lock);
491 				continue;
492 			}
493 			/*
494 			 * If we are ok with PG_LEVEL_2M mapping, then we will
495 			 * use the existing mapping,
496 			 *
497 			 * Otherwise, we will split the large page mapping but
498 			 * use the same existing protection bits except for
499 			 * large page, so that we don't violate Intel's TLB
500 			 * Application note (317080) which says, while changing
501 			 * the page sizes, new and old translations should
502 			 * not differ with respect to page frame and
503 			 * attributes.
504 			 */
505 			if (page_size_mask & (1 << PG_LEVEL_2M)) {
506 				if (!after_bootmem)
507 					pages++;
508 				paddr_last = paddr_next;
509 				continue;
510 			}
511 			new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
512 		}
513 
514 		if (page_size_mask & (1<<PG_LEVEL_2M)) {
515 			pages++;
516 			spin_lock(&init_mm.page_table_lock);
517 			set_pte((pte_t *)pmd,
518 				pfn_pte((paddr & PMD_MASK) >> PAGE_SHIFT,
519 					__pgprot(pgprot_val(prot) | _PAGE_PSE)));
520 			spin_unlock(&init_mm.page_table_lock);
521 			paddr_last = paddr_next;
522 			continue;
523 		}
524 
525 		pte = alloc_low_page();
526 		paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot);
527 
528 		spin_lock(&init_mm.page_table_lock);
529 		pmd_populate_kernel(&init_mm, pmd, pte);
530 		spin_unlock(&init_mm.page_table_lock);
531 	}
532 	update_page_count(PG_LEVEL_2M, pages);
533 	return paddr_last;
534 }
535 
536 /*
537  * Create PUD level page table mapping for physical addresses. The virtual
538  * and physical address do not have to be aligned at this level. KASLR can
539  * randomize virtual addresses up to this level.
540  * It returns the last physical address mapped.
541  */
542 static unsigned long __meminit
543 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
544 	      unsigned long page_size_mask)
545 {
546 	unsigned long pages = 0, paddr_next;
547 	unsigned long paddr_last = paddr_end;
548 	unsigned long vaddr = (unsigned long)__va(paddr);
549 	int i = pud_index(vaddr);
550 
551 	for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
552 		pud_t *pud;
553 		pmd_t *pmd;
554 		pgprot_t prot = PAGE_KERNEL;
555 
556 		vaddr = (unsigned long)__va(paddr);
557 		pud = pud_page + pud_index(vaddr);
558 		paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
559 
560 		if (paddr >= paddr_end) {
561 			if (!after_bootmem &&
562 			    !e820__mapped_any(paddr & PUD_MASK, paddr_next,
563 					     E820_TYPE_RAM) &&
564 			    !e820__mapped_any(paddr & PUD_MASK, paddr_next,
565 					     E820_TYPE_RESERVED_KERN))
566 				set_pud(pud, __pud(0));
567 			continue;
568 		}
569 
570 		if (!pud_none(*pud)) {
571 			if (!pud_large(*pud)) {
572 				pmd = pmd_offset(pud, 0);
573 				paddr_last = phys_pmd_init(pmd, paddr,
574 							   paddr_end,
575 							   page_size_mask,
576 							   prot);
577 				__flush_tlb_all();
578 				continue;
579 			}
580 			/*
581 			 * If we are ok with PG_LEVEL_1G mapping, then we will
582 			 * use the existing mapping.
583 			 *
584 			 * Otherwise, we will split the gbpage mapping but use
585 			 * the same existing protection  bits except for large
586 			 * page, so that we don't violate Intel's TLB
587 			 * Application note (317080) which says, while changing
588 			 * the page sizes, new and old translations should
589 			 * not differ with respect to page frame and
590 			 * attributes.
591 			 */
592 			if (page_size_mask & (1 << PG_LEVEL_1G)) {
593 				if (!after_bootmem)
594 					pages++;
595 				paddr_last = paddr_next;
596 				continue;
597 			}
598 			prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
599 		}
600 
601 		if (page_size_mask & (1<<PG_LEVEL_1G)) {
602 			pages++;
603 			spin_lock(&init_mm.page_table_lock);
604 			set_pte((pte_t *)pud,
605 				pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
606 					PAGE_KERNEL_LARGE));
607 			spin_unlock(&init_mm.page_table_lock);
608 			paddr_last = paddr_next;
609 			continue;
610 		}
611 
612 		pmd = alloc_low_page();
613 		paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
614 					   page_size_mask, prot);
615 
616 		spin_lock(&init_mm.page_table_lock);
617 		pud_populate(&init_mm, pud, pmd);
618 		spin_unlock(&init_mm.page_table_lock);
619 	}
620 	__flush_tlb_all();
621 
622 	update_page_count(PG_LEVEL_1G, pages);
623 
624 	return paddr_last;
625 }
626 
627 static unsigned long __meminit
628 phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
629 	      unsigned long page_size_mask)
630 {
631 	unsigned long paddr_next, paddr_last = paddr_end;
632 	unsigned long vaddr = (unsigned long)__va(paddr);
633 	int i = p4d_index(vaddr);
634 
635 	if (!IS_ENABLED(CONFIG_X86_5LEVEL))
636 		return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end, page_size_mask);
637 
638 	for (; i < PTRS_PER_P4D; i++, paddr = paddr_next) {
639 		p4d_t *p4d;
640 		pud_t *pud;
641 
642 		vaddr = (unsigned long)__va(paddr);
643 		p4d = p4d_page + p4d_index(vaddr);
644 		paddr_next = (paddr & P4D_MASK) + P4D_SIZE;
645 
646 		if (paddr >= paddr_end) {
647 			if (!after_bootmem &&
648 			    !e820__mapped_any(paddr & P4D_MASK, paddr_next,
649 					     E820_TYPE_RAM) &&
650 			    !e820__mapped_any(paddr & P4D_MASK, paddr_next,
651 					     E820_TYPE_RESERVED_KERN))
652 				set_p4d(p4d, __p4d(0));
653 			continue;
654 		}
655 
656 		if (!p4d_none(*p4d)) {
657 			pud = pud_offset(p4d, 0);
658 			paddr_last = phys_pud_init(pud, paddr,
659 					paddr_end,
660 					page_size_mask);
661 			__flush_tlb_all();
662 			continue;
663 		}
664 
665 		pud = alloc_low_page();
666 		paddr_last = phys_pud_init(pud, paddr, paddr_end,
667 					   page_size_mask);
668 
669 		spin_lock(&init_mm.page_table_lock);
670 		p4d_populate(&init_mm, p4d, pud);
671 		spin_unlock(&init_mm.page_table_lock);
672 	}
673 	__flush_tlb_all();
674 
675 	return paddr_last;
676 }
677 
678 /*
679  * Create page table mapping for the physical memory for specific physical
680  * addresses. The virtual and physical addresses have to be aligned on PMD level
681  * down. It returns the last physical address mapped.
682  */
683 unsigned long __meminit
684 kernel_physical_mapping_init(unsigned long paddr_start,
685 			     unsigned long paddr_end,
686 			     unsigned long page_size_mask)
687 {
688 	bool pgd_changed = false;
689 	unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
690 
691 	paddr_last = paddr_end;
692 	vaddr = (unsigned long)__va(paddr_start);
693 	vaddr_end = (unsigned long)__va(paddr_end);
694 	vaddr_start = vaddr;
695 
696 	for (; vaddr < vaddr_end; vaddr = vaddr_next) {
697 		pgd_t *pgd = pgd_offset_k(vaddr);
698 		p4d_t *p4d;
699 
700 		vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
701 
702 		if (pgd_val(*pgd)) {
703 			p4d = (p4d_t *)pgd_page_vaddr(*pgd);
704 			paddr_last = phys_p4d_init(p4d, __pa(vaddr),
705 						   __pa(vaddr_end),
706 						   page_size_mask);
707 			continue;
708 		}
709 
710 		p4d = alloc_low_page();
711 		paddr_last = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
712 					   page_size_mask);
713 
714 		spin_lock(&init_mm.page_table_lock);
715 		if (IS_ENABLED(CONFIG_X86_5LEVEL))
716 			pgd_populate(&init_mm, pgd, p4d);
717 		else
718 			p4d_populate(&init_mm, p4d_offset(pgd, vaddr), (pud_t *) p4d);
719 		spin_unlock(&init_mm.page_table_lock);
720 		pgd_changed = true;
721 	}
722 
723 	if (pgd_changed)
724 		sync_global_pgds(vaddr_start, vaddr_end - 1);
725 
726 	__flush_tlb_all();
727 
728 	return paddr_last;
729 }
730 
731 #ifndef CONFIG_NUMA
732 void __init initmem_init(void)
733 {
734 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
735 }
736 #endif
737 
738 void __init paging_init(void)
739 {
740 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
741 	sparse_init();
742 
743 	/*
744 	 * clear the default setting with node 0
745 	 * note: don't use nodes_clear here, that is really clearing when
746 	 *	 numa support is not compiled in, and later node_set_state
747 	 *	 will not set it back.
748 	 */
749 	node_clear_state(0, N_MEMORY);
750 	if (N_MEMORY != N_NORMAL_MEMORY)
751 		node_clear_state(0, N_NORMAL_MEMORY);
752 
753 	zone_sizes_init();
754 }
755 
756 /*
757  * Memory hotplug specific functions
758  */
759 #ifdef CONFIG_MEMORY_HOTPLUG
760 /*
761  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
762  * updating.
763  */
764 static void update_end_of_memory_vars(u64 start, u64 size)
765 {
766 	unsigned long end_pfn = PFN_UP(start + size);
767 
768 	if (end_pfn > max_pfn) {
769 		max_pfn = end_pfn;
770 		max_low_pfn = end_pfn;
771 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
772 	}
773 }
774 
775 int add_pages(int nid, unsigned long start_pfn,
776 	      unsigned long nr_pages, bool want_memblock)
777 {
778 	int ret;
779 
780 	ret = __add_pages(nid, start_pfn, nr_pages, want_memblock);
781 	WARN_ON_ONCE(ret);
782 
783 	/* update max_pfn, max_low_pfn and high_memory */
784 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
785 				  nr_pages << PAGE_SHIFT);
786 
787 	return ret;
788 }
789 
790 int arch_add_memory(int nid, u64 start, u64 size, bool want_memblock)
791 {
792 	unsigned long start_pfn = start >> PAGE_SHIFT;
793 	unsigned long nr_pages = size >> PAGE_SHIFT;
794 
795 	init_memory_mapping(start, start + size);
796 
797 	return add_pages(nid, start_pfn, nr_pages, want_memblock);
798 }
799 EXPORT_SYMBOL_GPL(arch_add_memory);
800 
801 #define PAGE_INUSE 0xFD
802 
803 static void __meminit free_pagetable(struct page *page, int order)
804 {
805 	unsigned long magic;
806 	unsigned int nr_pages = 1 << order;
807 	struct vmem_altmap *altmap = to_vmem_altmap((unsigned long) page);
808 
809 	if (altmap) {
810 		vmem_altmap_free(altmap, nr_pages);
811 		return;
812 	}
813 
814 	/* bootmem page has reserved flag */
815 	if (PageReserved(page)) {
816 		__ClearPageReserved(page);
817 
818 		magic = (unsigned long)page->freelist;
819 		if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
820 			while (nr_pages--)
821 				put_page_bootmem(page++);
822 		} else
823 			while (nr_pages--)
824 				free_reserved_page(page++);
825 	} else
826 		free_pages((unsigned long)page_address(page), order);
827 }
828 
829 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
830 {
831 	pte_t *pte;
832 	int i;
833 
834 	for (i = 0; i < PTRS_PER_PTE; i++) {
835 		pte = pte_start + i;
836 		if (!pte_none(*pte))
837 			return;
838 	}
839 
840 	/* free a pte talbe */
841 	free_pagetable(pmd_page(*pmd), 0);
842 	spin_lock(&init_mm.page_table_lock);
843 	pmd_clear(pmd);
844 	spin_unlock(&init_mm.page_table_lock);
845 }
846 
847 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
848 {
849 	pmd_t *pmd;
850 	int i;
851 
852 	for (i = 0; i < PTRS_PER_PMD; i++) {
853 		pmd = pmd_start + i;
854 		if (!pmd_none(*pmd))
855 			return;
856 	}
857 
858 	/* free a pmd talbe */
859 	free_pagetable(pud_page(*pud), 0);
860 	spin_lock(&init_mm.page_table_lock);
861 	pud_clear(pud);
862 	spin_unlock(&init_mm.page_table_lock);
863 }
864 
865 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
866 {
867 	pud_t *pud;
868 	int i;
869 
870 	for (i = 0; i < PTRS_PER_PUD; i++) {
871 		pud = pud_start + i;
872 		if (!pud_none(*pud))
873 			return;
874 	}
875 
876 	/* free a pud talbe */
877 	free_pagetable(p4d_page(*p4d), 0);
878 	spin_lock(&init_mm.page_table_lock);
879 	p4d_clear(p4d);
880 	spin_unlock(&init_mm.page_table_lock);
881 }
882 
883 static void __meminit
884 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
885 		 bool direct)
886 {
887 	unsigned long next, pages = 0;
888 	pte_t *pte;
889 	void *page_addr;
890 	phys_addr_t phys_addr;
891 
892 	pte = pte_start + pte_index(addr);
893 	for (; addr < end; addr = next, pte++) {
894 		next = (addr + PAGE_SIZE) & PAGE_MASK;
895 		if (next > end)
896 			next = end;
897 
898 		if (!pte_present(*pte))
899 			continue;
900 
901 		/*
902 		 * We mapped [0,1G) memory as identity mapping when
903 		 * initializing, in arch/x86/kernel/head_64.S. These
904 		 * pagetables cannot be removed.
905 		 */
906 		phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
907 		if (phys_addr < (phys_addr_t)0x40000000)
908 			return;
909 
910 		if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
911 			/*
912 			 * Do not free direct mapping pages since they were
913 			 * freed when offlining, or simplely not in use.
914 			 */
915 			if (!direct)
916 				free_pagetable(pte_page(*pte), 0);
917 
918 			spin_lock(&init_mm.page_table_lock);
919 			pte_clear(&init_mm, addr, pte);
920 			spin_unlock(&init_mm.page_table_lock);
921 
922 			/* For non-direct mapping, pages means nothing. */
923 			pages++;
924 		} else {
925 			/*
926 			 * If we are here, we are freeing vmemmap pages since
927 			 * direct mapped memory ranges to be freed are aligned.
928 			 *
929 			 * If we are not removing the whole page, it means
930 			 * other page structs in this page are being used and
931 			 * we canot remove them. So fill the unused page_structs
932 			 * with 0xFD, and remove the page when it is wholly
933 			 * filled with 0xFD.
934 			 */
935 			memset((void *)addr, PAGE_INUSE, next - addr);
936 
937 			page_addr = page_address(pte_page(*pte));
938 			if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
939 				free_pagetable(pte_page(*pte), 0);
940 
941 				spin_lock(&init_mm.page_table_lock);
942 				pte_clear(&init_mm, addr, pte);
943 				spin_unlock(&init_mm.page_table_lock);
944 			}
945 		}
946 	}
947 
948 	/* Call free_pte_table() in remove_pmd_table(). */
949 	flush_tlb_all();
950 	if (direct)
951 		update_page_count(PG_LEVEL_4K, -pages);
952 }
953 
954 static void __meminit
955 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
956 		 bool direct)
957 {
958 	unsigned long next, pages = 0;
959 	pte_t *pte_base;
960 	pmd_t *pmd;
961 	void *page_addr;
962 
963 	pmd = pmd_start + pmd_index(addr);
964 	for (; addr < end; addr = next, pmd++) {
965 		next = pmd_addr_end(addr, end);
966 
967 		if (!pmd_present(*pmd))
968 			continue;
969 
970 		if (pmd_large(*pmd)) {
971 			if (IS_ALIGNED(addr, PMD_SIZE) &&
972 			    IS_ALIGNED(next, PMD_SIZE)) {
973 				if (!direct)
974 					free_pagetable(pmd_page(*pmd),
975 						       get_order(PMD_SIZE));
976 
977 				spin_lock(&init_mm.page_table_lock);
978 				pmd_clear(pmd);
979 				spin_unlock(&init_mm.page_table_lock);
980 				pages++;
981 			} else {
982 				/* If here, we are freeing vmemmap pages. */
983 				memset((void *)addr, PAGE_INUSE, next - addr);
984 
985 				page_addr = page_address(pmd_page(*pmd));
986 				if (!memchr_inv(page_addr, PAGE_INUSE,
987 						PMD_SIZE)) {
988 					free_pagetable(pmd_page(*pmd),
989 						       get_order(PMD_SIZE));
990 
991 					spin_lock(&init_mm.page_table_lock);
992 					pmd_clear(pmd);
993 					spin_unlock(&init_mm.page_table_lock);
994 				}
995 			}
996 
997 			continue;
998 		}
999 
1000 		pte_base = (pte_t *)pmd_page_vaddr(*pmd);
1001 		remove_pte_table(pte_base, addr, next, direct);
1002 		free_pte_table(pte_base, pmd);
1003 	}
1004 
1005 	/* Call free_pmd_table() in remove_pud_table(). */
1006 	if (direct)
1007 		update_page_count(PG_LEVEL_2M, -pages);
1008 }
1009 
1010 static void __meminit
1011 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
1012 		 bool direct)
1013 {
1014 	unsigned long next, pages = 0;
1015 	pmd_t *pmd_base;
1016 	pud_t *pud;
1017 	void *page_addr;
1018 
1019 	pud = pud_start + pud_index(addr);
1020 	for (; addr < end; addr = next, pud++) {
1021 		next = pud_addr_end(addr, end);
1022 
1023 		if (!pud_present(*pud))
1024 			continue;
1025 
1026 		if (pud_large(*pud)) {
1027 			if (IS_ALIGNED(addr, PUD_SIZE) &&
1028 			    IS_ALIGNED(next, PUD_SIZE)) {
1029 				if (!direct)
1030 					free_pagetable(pud_page(*pud),
1031 						       get_order(PUD_SIZE));
1032 
1033 				spin_lock(&init_mm.page_table_lock);
1034 				pud_clear(pud);
1035 				spin_unlock(&init_mm.page_table_lock);
1036 				pages++;
1037 			} else {
1038 				/* If here, we are freeing vmemmap pages. */
1039 				memset((void *)addr, PAGE_INUSE, next - addr);
1040 
1041 				page_addr = page_address(pud_page(*pud));
1042 				if (!memchr_inv(page_addr, PAGE_INUSE,
1043 						PUD_SIZE)) {
1044 					free_pagetable(pud_page(*pud),
1045 						       get_order(PUD_SIZE));
1046 
1047 					spin_lock(&init_mm.page_table_lock);
1048 					pud_clear(pud);
1049 					spin_unlock(&init_mm.page_table_lock);
1050 				}
1051 			}
1052 
1053 			continue;
1054 		}
1055 
1056 		pmd_base = pmd_offset(pud, 0);
1057 		remove_pmd_table(pmd_base, addr, next, direct);
1058 		free_pmd_table(pmd_base, pud);
1059 	}
1060 
1061 	if (direct)
1062 		update_page_count(PG_LEVEL_1G, -pages);
1063 }
1064 
1065 static void __meminit
1066 remove_p4d_table(p4d_t *p4d_start, unsigned long addr, unsigned long end,
1067 		 bool direct)
1068 {
1069 	unsigned long next, pages = 0;
1070 	pud_t *pud_base;
1071 	p4d_t *p4d;
1072 
1073 	p4d = p4d_start + p4d_index(addr);
1074 	for (; addr < end; addr = next, p4d++) {
1075 		next = p4d_addr_end(addr, end);
1076 
1077 		if (!p4d_present(*p4d))
1078 			continue;
1079 
1080 		BUILD_BUG_ON(p4d_large(*p4d));
1081 
1082 		pud_base = pud_offset(p4d, 0);
1083 		remove_pud_table(pud_base, addr, next, direct);
1084 		/*
1085 		 * For 4-level page tables we do not want to free PUDs, but in the
1086 		 * 5-level case we should free them. This code will have to change
1087 		 * to adapt for boot-time switching between 4 and 5 level page tables.
1088 		 */
1089 		if (CONFIG_PGTABLE_LEVELS == 5)
1090 			free_pud_table(pud_base, p4d);
1091 	}
1092 
1093 	if (direct)
1094 		update_page_count(PG_LEVEL_512G, -pages);
1095 }
1096 
1097 /* start and end are both virtual address. */
1098 static void __meminit
1099 remove_pagetable(unsigned long start, unsigned long end, bool direct)
1100 {
1101 	unsigned long next;
1102 	unsigned long addr;
1103 	pgd_t *pgd;
1104 	p4d_t *p4d;
1105 
1106 	for (addr = start; addr < end; addr = next) {
1107 		next = pgd_addr_end(addr, end);
1108 
1109 		pgd = pgd_offset_k(addr);
1110 		if (!pgd_present(*pgd))
1111 			continue;
1112 
1113 		p4d = p4d_offset(pgd, 0);
1114 		remove_p4d_table(p4d, addr, next, direct);
1115 	}
1116 
1117 	flush_tlb_all();
1118 }
1119 
1120 void __ref vmemmap_free(unsigned long start, unsigned long end)
1121 {
1122 	remove_pagetable(start, end, false);
1123 }
1124 
1125 #ifdef CONFIG_MEMORY_HOTREMOVE
1126 static void __meminit
1127 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1128 {
1129 	start = (unsigned long)__va(start);
1130 	end = (unsigned long)__va(end);
1131 
1132 	remove_pagetable(start, end, true);
1133 }
1134 
1135 int __ref arch_remove_memory(u64 start, u64 size)
1136 {
1137 	unsigned long start_pfn = start >> PAGE_SHIFT;
1138 	unsigned long nr_pages = size >> PAGE_SHIFT;
1139 	struct page *page = pfn_to_page(start_pfn);
1140 	struct vmem_altmap *altmap;
1141 	struct zone *zone;
1142 	int ret;
1143 
1144 	/* With altmap the first mapped page is offset from @start */
1145 	altmap = to_vmem_altmap((unsigned long) page);
1146 	if (altmap)
1147 		page += vmem_altmap_offset(altmap);
1148 	zone = page_zone(page);
1149 	ret = __remove_pages(zone, start_pfn, nr_pages);
1150 	WARN_ON_ONCE(ret);
1151 	kernel_physical_mapping_remove(start, start + size);
1152 
1153 	return ret;
1154 }
1155 #endif
1156 #endif /* CONFIG_MEMORY_HOTPLUG */
1157 
1158 static struct kcore_list kcore_vsyscall;
1159 
1160 static void __init register_page_bootmem_info(void)
1161 {
1162 #ifdef CONFIG_NUMA
1163 	int i;
1164 
1165 	for_each_online_node(i)
1166 		register_page_bootmem_info_node(NODE_DATA(i));
1167 #endif
1168 }
1169 
1170 void __init mem_init(void)
1171 {
1172 	pci_iommu_alloc();
1173 
1174 	/* clear_bss() already clear the empty_zero_page */
1175 
1176 	register_page_bootmem_info();
1177 
1178 	/* this will put all memory onto the freelists */
1179 	free_all_bootmem();
1180 	after_bootmem = 1;
1181 
1182 	/* Register memory areas for /proc/kcore */
1183 	kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR,
1184 			 PAGE_SIZE, KCORE_OTHER);
1185 
1186 	mem_init_print_info(NULL);
1187 }
1188 
1189 int kernel_set_to_readonly;
1190 
1191 void set_kernel_text_rw(void)
1192 {
1193 	unsigned long start = PFN_ALIGN(_text);
1194 	unsigned long end = PFN_ALIGN(__stop___ex_table);
1195 
1196 	if (!kernel_set_to_readonly)
1197 		return;
1198 
1199 	pr_debug("Set kernel text: %lx - %lx for read write\n",
1200 		 start, end);
1201 
1202 	/*
1203 	 * Make the kernel identity mapping for text RW. Kernel text
1204 	 * mapping will always be RO. Refer to the comment in
1205 	 * static_protections() in pageattr.c
1206 	 */
1207 	set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1208 }
1209 
1210 void set_kernel_text_ro(void)
1211 {
1212 	unsigned long start = PFN_ALIGN(_text);
1213 	unsigned long end = PFN_ALIGN(__stop___ex_table);
1214 
1215 	if (!kernel_set_to_readonly)
1216 		return;
1217 
1218 	pr_debug("Set kernel text: %lx - %lx for read only\n",
1219 		 start, end);
1220 
1221 	/*
1222 	 * Set the kernel identity mapping for text RO.
1223 	 */
1224 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1225 }
1226 
1227 void mark_rodata_ro(void)
1228 {
1229 	unsigned long start = PFN_ALIGN(_text);
1230 	unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1231 	unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1232 	unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1233 	unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1234 	unsigned long all_end;
1235 
1236 	printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1237 	       (end - start) >> 10);
1238 	set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1239 
1240 	kernel_set_to_readonly = 1;
1241 
1242 	/*
1243 	 * The rodata/data/bss/brk section (but not the kernel text!)
1244 	 * should also be not-executable.
1245 	 *
1246 	 * We align all_end to PMD_SIZE because the existing mapping
1247 	 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1248 	 * split the PMD and the reminder between _brk_end and the end
1249 	 * of the PMD will remain mapped executable.
1250 	 *
1251 	 * Any PMD which was setup after the one which covers _brk_end
1252 	 * has been zapped already via cleanup_highmem().
1253 	 */
1254 	all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1255 	set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1256 
1257 #ifdef CONFIG_CPA_DEBUG
1258 	printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1259 	set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1260 
1261 	printk(KERN_INFO "Testing CPA: again\n");
1262 	set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1263 #endif
1264 
1265 	free_init_pages("unused kernel",
1266 			(unsigned long) __va(__pa_symbol(text_end)),
1267 			(unsigned long) __va(__pa_symbol(rodata_start)));
1268 	free_init_pages("unused kernel",
1269 			(unsigned long) __va(__pa_symbol(rodata_end)),
1270 			(unsigned long) __va(__pa_symbol(_sdata)));
1271 
1272 	debug_checkwx();
1273 }
1274 
1275 int kern_addr_valid(unsigned long addr)
1276 {
1277 	unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1278 	pgd_t *pgd;
1279 	p4d_t *p4d;
1280 	pud_t *pud;
1281 	pmd_t *pmd;
1282 	pte_t *pte;
1283 
1284 	if (above != 0 && above != -1UL)
1285 		return 0;
1286 
1287 	pgd = pgd_offset_k(addr);
1288 	if (pgd_none(*pgd))
1289 		return 0;
1290 
1291 	p4d = p4d_offset(pgd, addr);
1292 	if (p4d_none(*p4d))
1293 		return 0;
1294 
1295 	pud = pud_offset(p4d, addr);
1296 	if (pud_none(*pud))
1297 		return 0;
1298 
1299 	if (pud_large(*pud))
1300 		return pfn_valid(pud_pfn(*pud));
1301 
1302 	pmd = pmd_offset(pud, addr);
1303 	if (pmd_none(*pmd))
1304 		return 0;
1305 
1306 	if (pmd_large(*pmd))
1307 		return pfn_valid(pmd_pfn(*pmd));
1308 
1309 	pte = pte_offset_kernel(pmd, addr);
1310 	if (pte_none(*pte))
1311 		return 0;
1312 
1313 	return pfn_valid(pte_pfn(*pte));
1314 }
1315 
1316 static unsigned long probe_memory_block_size(void)
1317 {
1318 	unsigned long bz = MIN_MEMORY_BLOCK_SIZE;
1319 
1320 	/* if system is UV or has 64GB of RAM or more, use large blocks */
1321 	if (is_uv_system() || ((max_pfn << PAGE_SHIFT) >= (64UL << 30)))
1322 		bz = 2UL << 30; /* 2GB */
1323 
1324 	pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1325 
1326 	return bz;
1327 }
1328 
1329 static unsigned long memory_block_size_probed;
1330 unsigned long memory_block_size_bytes(void)
1331 {
1332 	if (!memory_block_size_probed)
1333 		memory_block_size_probed = probe_memory_block_size();
1334 
1335 	return memory_block_size_probed;
1336 }
1337 
1338 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1339 /*
1340  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1341  */
1342 static long __meminitdata addr_start, addr_end;
1343 static void __meminitdata *p_start, *p_end;
1344 static int __meminitdata node_start;
1345 
1346 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1347 		unsigned long end, int node, struct vmem_altmap *altmap)
1348 {
1349 	unsigned long addr;
1350 	unsigned long next;
1351 	pgd_t *pgd;
1352 	p4d_t *p4d;
1353 	pud_t *pud;
1354 	pmd_t *pmd;
1355 
1356 	for (addr = start; addr < end; addr = next) {
1357 		next = pmd_addr_end(addr, end);
1358 
1359 		pgd = vmemmap_pgd_populate(addr, node);
1360 		if (!pgd)
1361 			return -ENOMEM;
1362 
1363 		p4d = vmemmap_p4d_populate(pgd, addr, node);
1364 		if (!p4d)
1365 			return -ENOMEM;
1366 
1367 		pud = vmemmap_pud_populate(p4d, addr, node);
1368 		if (!pud)
1369 			return -ENOMEM;
1370 
1371 		pmd = pmd_offset(pud, addr);
1372 		if (pmd_none(*pmd)) {
1373 			void *p;
1374 
1375 			p = __vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1376 			if (p) {
1377 				pte_t entry;
1378 
1379 				entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1380 						PAGE_KERNEL_LARGE);
1381 				set_pmd(pmd, __pmd(pte_val(entry)));
1382 
1383 				/* check to see if we have contiguous blocks */
1384 				if (p_end != p || node_start != node) {
1385 					if (p_start)
1386 						pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1387 						       addr_start, addr_end-1, p_start, p_end-1, node_start);
1388 					addr_start = addr;
1389 					node_start = node;
1390 					p_start = p;
1391 				}
1392 
1393 				addr_end = addr + PMD_SIZE;
1394 				p_end = p + PMD_SIZE;
1395 				continue;
1396 			} else if (altmap)
1397 				return -ENOMEM; /* no fallback */
1398 		} else if (pmd_large(*pmd)) {
1399 			vmemmap_verify((pte_t *)pmd, node, addr, next);
1400 			continue;
1401 		}
1402 		pr_warn_once("vmemmap: falling back to regular page backing\n");
1403 		if (vmemmap_populate_basepages(addr, next, node))
1404 			return -ENOMEM;
1405 	}
1406 	return 0;
1407 }
1408 
1409 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
1410 {
1411 	struct vmem_altmap *altmap = to_vmem_altmap(start);
1412 	int err;
1413 
1414 	if (boot_cpu_has(X86_FEATURE_PSE))
1415 		err = vmemmap_populate_hugepages(start, end, node, altmap);
1416 	else if (altmap) {
1417 		pr_err_once("%s: no cpu support for altmap allocations\n",
1418 				__func__);
1419 		err = -ENOMEM;
1420 	} else
1421 		err = vmemmap_populate_basepages(start, end, node);
1422 	if (!err)
1423 		sync_global_pgds(start, end - 1);
1424 	return err;
1425 }
1426 
1427 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1428 void register_page_bootmem_memmap(unsigned long section_nr,
1429 				  struct page *start_page, unsigned long size)
1430 {
1431 	unsigned long addr = (unsigned long)start_page;
1432 	unsigned long end = (unsigned long)(start_page + size);
1433 	unsigned long next;
1434 	pgd_t *pgd;
1435 	p4d_t *p4d;
1436 	pud_t *pud;
1437 	pmd_t *pmd;
1438 	unsigned int nr_pages;
1439 	struct page *page;
1440 
1441 	for (; addr < end; addr = next) {
1442 		pte_t *pte = NULL;
1443 
1444 		pgd = pgd_offset_k(addr);
1445 		if (pgd_none(*pgd)) {
1446 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1447 			continue;
1448 		}
1449 		get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1450 
1451 		p4d = p4d_offset(pgd, addr);
1452 		if (p4d_none(*p4d)) {
1453 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1454 			continue;
1455 		}
1456 		get_page_bootmem(section_nr, p4d_page(*p4d), MIX_SECTION_INFO);
1457 
1458 		pud = pud_offset(p4d, addr);
1459 		if (pud_none(*pud)) {
1460 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1461 			continue;
1462 		}
1463 		get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1464 
1465 		if (!boot_cpu_has(X86_FEATURE_PSE)) {
1466 			next = (addr + PAGE_SIZE) & PAGE_MASK;
1467 			pmd = pmd_offset(pud, addr);
1468 			if (pmd_none(*pmd))
1469 				continue;
1470 			get_page_bootmem(section_nr, pmd_page(*pmd),
1471 					 MIX_SECTION_INFO);
1472 
1473 			pte = pte_offset_kernel(pmd, addr);
1474 			if (pte_none(*pte))
1475 				continue;
1476 			get_page_bootmem(section_nr, pte_page(*pte),
1477 					 SECTION_INFO);
1478 		} else {
1479 			next = pmd_addr_end(addr, end);
1480 
1481 			pmd = pmd_offset(pud, addr);
1482 			if (pmd_none(*pmd))
1483 				continue;
1484 
1485 			nr_pages = 1 << (get_order(PMD_SIZE));
1486 			page = pmd_page(*pmd);
1487 			while (nr_pages--)
1488 				get_page_bootmem(section_nr, page++,
1489 						 SECTION_INFO);
1490 		}
1491 	}
1492 }
1493 #endif
1494 
1495 void __meminit vmemmap_populate_print_last(void)
1496 {
1497 	if (p_start) {
1498 		pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1499 			addr_start, addr_end-1, p_start, p_end-1, node_start);
1500 		p_start = NULL;
1501 		p_end = NULL;
1502 		node_start = 0;
1503 	}
1504 }
1505 #endif
1506