xref: /linux/arch/arm64/mm/mmu.c (revision cea0f76a483d1270ac6f6513964e3e75193dda48)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Based on arch/arm/mm/mmu.c
4  *
5  * Copyright (C) 1995-2005 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  */
8 
9 #include <linux/cache.h>
10 #include <linux/export.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/ioport.h>
15 #include <linux/kexec.h>
16 #include <linux/libfdt.h>
17 #include <linux/mman.h>
18 #include <linux/nodemask.h>
19 #include <linux/memblock.h>
20 #include <linux/memory.h>
21 #include <linux/fs.h>
22 #include <linux/io.h>
23 #include <linux/mm.h>
24 #include <linux/vmalloc.h>
25 
26 #include <asm/barrier.h>
27 #include <asm/cputype.h>
28 #include <asm/fixmap.h>
29 #include <asm/kasan.h>
30 #include <asm/kernel-pgtable.h>
31 #include <asm/sections.h>
32 #include <asm/setup.h>
33 #include <linux/sizes.h>
34 #include <asm/tlb.h>
35 #include <asm/mmu_context.h>
36 #include <asm/ptdump.h>
37 #include <asm/tlbflush.h>
38 
39 #define NO_BLOCK_MAPPINGS	BIT(0)
40 #define NO_CONT_MAPPINGS	BIT(1)
41 
42 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
43 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD;
44 
45 u64 __section(".mmuoff.data.write") vabits_actual;
46 EXPORT_SYMBOL(vabits_actual);
47 
48 u64 kimage_voffset __ro_after_init;
49 EXPORT_SYMBOL(kimage_voffset);
50 
51 /*
52  * Empty_zero_page is a special page that is used for zero-initialized data
53  * and COW.
54  */
55 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
56 EXPORT_SYMBOL(empty_zero_page);
57 
58 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
59 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
60 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
61 
62 static DEFINE_SPINLOCK(swapper_pgdir_lock);
63 
64 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
65 {
66 	pgd_t *fixmap_pgdp;
67 
68 	spin_lock(&swapper_pgdir_lock);
69 	fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
70 	WRITE_ONCE(*fixmap_pgdp, pgd);
71 	/*
72 	 * We need dsb(ishst) here to ensure the page-table-walker sees
73 	 * our new entry before set_p?d() returns. The fixmap's
74 	 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
75 	 */
76 	pgd_clear_fixmap();
77 	spin_unlock(&swapper_pgdir_lock);
78 }
79 
80 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
81 			      unsigned long size, pgprot_t vma_prot)
82 {
83 	if (!pfn_valid(pfn))
84 		return pgprot_noncached(vma_prot);
85 	else if (file->f_flags & O_SYNC)
86 		return pgprot_writecombine(vma_prot);
87 	return vma_prot;
88 }
89 EXPORT_SYMBOL(phys_mem_access_prot);
90 
91 static phys_addr_t __init early_pgtable_alloc(int shift)
92 {
93 	phys_addr_t phys;
94 	void *ptr;
95 
96 	phys = memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
97 	if (!phys)
98 		panic("Failed to allocate page table page\n");
99 
100 	/*
101 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
102 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
103 	 * any level of table.
104 	 */
105 	ptr = pte_set_fixmap(phys);
106 
107 	memset(ptr, 0, PAGE_SIZE);
108 
109 	/*
110 	 * Implicit barriers also ensure the zeroed page is visible to the page
111 	 * table walker
112 	 */
113 	pte_clear_fixmap();
114 
115 	return phys;
116 }
117 
118 static bool pgattr_change_is_safe(u64 old, u64 new)
119 {
120 	/*
121 	 * The following mapping attributes may be updated in live
122 	 * kernel mappings without the need for break-before-make.
123 	 */
124 	static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
125 
126 	/* creating or taking down mappings is always safe */
127 	if (old == 0 || new == 0)
128 		return true;
129 
130 	/* live contiguous mappings may not be manipulated at all */
131 	if ((old | new) & PTE_CONT)
132 		return false;
133 
134 	/* Transitioning from Non-Global to Global is unsafe */
135 	if (old & ~new & PTE_NG)
136 		return false;
137 
138 	return ((old ^ new) & ~mask) == 0;
139 }
140 
141 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
142 		     phys_addr_t phys, pgprot_t prot)
143 {
144 	pte_t *ptep;
145 
146 	ptep = pte_set_fixmap_offset(pmdp, addr);
147 	do {
148 		pte_t old_pte = READ_ONCE(*ptep);
149 
150 		set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
151 
152 		/*
153 		 * After the PTE entry has been populated once, we
154 		 * only allow updates to the permission attributes.
155 		 */
156 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
157 					      READ_ONCE(pte_val(*ptep))));
158 
159 		phys += PAGE_SIZE;
160 	} while (ptep++, addr += PAGE_SIZE, addr != end);
161 
162 	pte_clear_fixmap();
163 }
164 
165 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
166 				unsigned long end, phys_addr_t phys,
167 				pgprot_t prot,
168 				phys_addr_t (*pgtable_alloc)(int),
169 				int flags)
170 {
171 	unsigned long next;
172 	pmd_t pmd = READ_ONCE(*pmdp);
173 
174 	BUG_ON(pmd_sect(pmd));
175 	if (pmd_none(pmd)) {
176 		phys_addr_t pte_phys;
177 		BUG_ON(!pgtable_alloc);
178 		pte_phys = pgtable_alloc(PAGE_SHIFT);
179 		__pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE);
180 		pmd = READ_ONCE(*pmdp);
181 	}
182 	BUG_ON(pmd_bad(pmd));
183 
184 	do {
185 		pgprot_t __prot = prot;
186 
187 		next = pte_cont_addr_end(addr, end);
188 
189 		/* use a contiguous mapping if the range is suitably aligned */
190 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
191 		    (flags & NO_CONT_MAPPINGS) == 0)
192 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
193 
194 		init_pte(pmdp, addr, next, phys, __prot);
195 
196 		phys += next - addr;
197 	} while (addr = next, addr != end);
198 }
199 
200 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
201 		     phys_addr_t phys, pgprot_t prot,
202 		     phys_addr_t (*pgtable_alloc)(int), int flags)
203 {
204 	unsigned long next;
205 	pmd_t *pmdp;
206 
207 	pmdp = pmd_set_fixmap_offset(pudp, addr);
208 	do {
209 		pmd_t old_pmd = READ_ONCE(*pmdp);
210 
211 		next = pmd_addr_end(addr, end);
212 
213 		/* try section mapping first */
214 		if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
215 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
216 			pmd_set_huge(pmdp, phys, prot);
217 
218 			/*
219 			 * After the PMD entry has been populated once, we
220 			 * only allow updates to the permission attributes.
221 			 */
222 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
223 						      READ_ONCE(pmd_val(*pmdp))));
224 		} else {
225 			alloc_init_cont_pte(pmdp, addr, next, phys, prot,
226 					    pgtable_alloc, flags);
227 
228 			BUG_ON(pmd_val(old_pmd) != 0 &&
229 			       pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
230 		}
231 		phys += next - addr;
232 	} while (pmdp++, addr = next, addr != end);
233 
234 	pmd_clear_fixmap();
235 }
236 
237 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
238 				unsigned long end, phys_addr_t phys,
239 				pgprot_t prot,
240 				phys_addr_t (*pgtable_alloc)(int), int flags)
241 {
242 	unsigned long next;
243 	pud_t pud = READ_ONCE(*pudp);
244 
245 	/*
246 	 * Check for initial section mappings in the pgd/pud.
247 	 */
248 	BUG_ON(pud_sect(pud));
249 	if (pud_none(pud)) {
250 		phys_addr_t pmd_phys;
251 		BUG_ON(!pgtable_alloc);
252 		pmd_phys = pgtable_alloc(PMD_SHIFT);
253 		__pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE);
254 		pud = READ_ONCE(*pudp);
255 	}
256 	BUG_ON(pud_bad(pud));
257 
258 	do {
259 		pgprot_t __prot = prot;
260 
261 		next = pmd_cont_addr_end(addr, end);
262 
263 		/* use a contiguous mapping if the range is suitably aligned */
264 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
265 		    (flags & NO_CONT_MAPPINGS) == 0)
266 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
267 
268 		init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
269 
270 		phys += next - addr;
271 	} while (addr = next, addr != end);
272 }
273 
274 static inline bool use_1G_block(unsigned long addr, unsigned long next,
275 			unsigned long phys)
276 {
277 	if (PAGE_SHIFT != 12)
278 		return false;
279 
280 	if (((addr | next | phys) & ~PUD_MASK) != 0)
281 		return false;
282 
283 	return true;
284 }
285 
286 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
287 			   phys_addr_t phys, pgprot_t prot,
288 			   phys_addr_t (*pgtable_alloc)(int),
289 			   int flags)
290 {
291 	unsigned long next;
292 	pud_t *pudp;
293 	p4d_t *p4dp = p4d_offset(pgdp, addr);
294 	p4d_t p4d = READ_ONCE(*p4dp);
295 
296 	if (p4d_none(p4d)) {
297 		phys_addr_t pud_phys;
298 		BUG_ON(!pgtable_alloc);
299 		pud_phys = pgtable_alloc(PUD_SHIFT);
300 		__p4d_populate(p4dp, pud_phys, PUD_TYPE_TABLE);
301 		p4d = READ_ONCE(*p4dp);
302 	}
303 	BUG_ON(p4d_bad(p4d));
304 
305 	pudp = pud_set_fixmap_offset(p4dp, addr);
306 	do {
307 		pud_t old_pud = READ_ONCE(*pudp);
308 
309 		next = pud_addr_end(addr, end);
310 
311 		/*
312 		 * For 4K granule only, attempt to put down a 1GB block
313 		 */
314 		if (use_1G_block(addr, next, phys) &&
315 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
316 			pud_set_huge(pudp, phys, prot);
317 
318 			/*
319 			 * After the PUD entry has been populated once, we
320 			 * only allow updates to the permission attributes.
321 			 */
322 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
323 						      READ_ONCE(pud_val(*pudp))));
324 		} else {
325 			alloc_init_cont_pmd(pudp, addr, next, phys, prot,
326 					    pgtable_alloc, flags);
327 
328 			BUG_ON(pud_val(old_pud) != 0 &&
329 			       pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
330 		}
331 		phys += next - addr;
332 	} while (pudp++, addr = next, addr != end);
333 
334 	pud_clear_fixmap();
335 }
336 
337 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
338 				 unsigned long virt, phys_addr_t size,
339 				 pgprot_t prot,
340 				 phys_addr_t (*pgtable_alloc)(int),
341 				 int flags)
342 {
343 	unsigned long addr, end, next;
344 	pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
345 
346 	/*
347 	 * If the virtual and physical address don't have the same offset
348 	 * within a page, we cannot map the region as the caller expects.
349 	 */
350 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
351 		return;
352 
353 	phys &= PAGE_MASK;
354 	addr = virt & PAGE_MASK;
355 	end = PAGE_ALIGN(virt + size);
356 
357 	do {
358 		next = pgd_addr_end(addr, end);
359 		alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
360 			       flags);
361 		phys += next - addr;
362 	} while (pgdp++, addr = next, addr != end);
363 }
364 
365 static phys_addr_t __pgd_pgtable_alloc(int shift)
366 {
367 	void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
368 	BUG_ON(!ptr);
369 
370 	/* Ensure the zeroed page is visible to the page table walker */
371 	dsb(ishst);
372 	return __pa(ptr);
373 }
374 
375 static phys_addr_t pgd_pgtable_alloc(int shift)
376 {
377 	phys_addr_t pa = __pgd_pgtable_alloc(shift);
378 
379 	/*
380 	 * Call proper page table ctor in case later we need to
381 	 * call core mm functions like apply_to_page_range() on
382 	 * this pre-allocated page table.
383 	 *
384 	 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
385 	 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
386 	 */
387 	if (shift == PAGE_SHIFT)
388 		BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
389 	else if (shift == PMD_SHIFT)
390 		BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
391 
392 	return pa;
393 }
394 
395 /*
396  * This function can only be used to modify existing table entries,
397  * without allocating new levels of table. Note that this permits the
398  * creation of new section or page entries.
399  */
400 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
401 				  phys_addr_t size, pgprot_t prot)
402 {
403 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
404 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
405 			&phys, virt);
406 		return;
407 	}
408 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
409 			     NO_CONT_MAPPINGS);
410 }
411 
412 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
413 			       unsigned long virt, phys_addr_t size,
414 			       pgprot_t prot, bool page_mappings_only)
415 {
416 	int flags = 0;
417 
418 	BUG_ON(mm == &init_mm);
419 
420 	if (page_mappings_only)
421 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
422 
423 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
424 			     pgd_pgtable_alloc, flags);
425 }
426 
427 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
428 				phys_addr_t size, pgprot_t prot)
429 {
430 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
431 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
432 			&phys, virt);
433 		return;
434 	}
435 
436 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
437 			     NO_CONT_MAPPINGS);
438 
439 	/* flush the TLBs after updating live kernel mappings */
440 	flush_tlb_kernel_range(virt, virt + size);
441 }
442 
443 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
444 				  phys_addr_t end, pgprot_t prot, int flags)
445 {
446 	__create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
447 			     prot, early_pgtable_alloc, flags);
448 }
449 
450 void __init mark_linear_text_alias_ro(void)
451 {
452 	/*
453 	 * Remove the write permissions from the linear alias of .text/.rodata
454 	 */
455 	update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
456 			    (unsigned long)__init_begin - (unsigned long)_text,
457 			    PAGE_KERNEL_RO);
458 }
459 
460 static void __init map_mem(pgd_t *pgdp)
461 {
462 	phys_addr_t kernel_start = __pa_symbol(_text);
463 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
464 	struct memblock_region *reg;
465 	int flags = 0;
466 
467 	if (rodata_full || debug_pagealloc_enabled())
468 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
469 
470 	/*
471 	 * Take care not to create a writable alias for the
472 	 * read-only text and rodata sections of the kernel image.
473 	 * So temporarily mark them as NOMAP to skip mappings in
474 	 * the following for-loop
475 	 */
476 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
477 #ifdef CONFIG_KEXEC_CORE
478 	if (crashk_res.end)
479 		memblock_mark_nomap(crashk_res.start,
480 				    resource_size(&crashk_res));
481 #endif
482 
483 	/* map all the memory banks */
484 	for_each_memblock(memory, reg) {
485 		phys_addr_t start = reg->base;
486 		phys_addr_t end = start + reg->size;
487 
488 		if (start >= end)
489 			break;
490 		if (memblock_is_nomap(reg))
491 			continue;
492 
493 		__map_memblock(pgdp, start, end, PAGE_KERNEL, flags);
494 	}
495 
496 	/*
497 	 * Map the linear alias of the [_text, __init_begin) interval
498 	 * as non-executable now, and remove the write permission in
499 	 * mark_linear_text_alias_ro() below (which will be called after
500 	 * alternative patching has completed). This makes the contents
501 	 * of the region accessible to subsystems such as hibernate,
502 	 * but protects it from inadvertent modification or execution.
503 	 * Note that contiguous mappings cannot be remapped in this way,
504 	 * so we should avoid them here.
505 	 */
506 	__map_memblock(pgdp, kernel_start, kernel_end,
507 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
508 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
509 
510 #ifdef CONFIG_KEXEC_CORE
511 	/*
512 	 * Use page-level mappings here so that we can shrink the region
513 	 * in page granularity and put back unused memory to buddy system
514 	 * through /sys/kernel/kexec_crash_size interface.
515 	 */
516 	if (crashk_res.end) {
517 		__map_memblock(pgdp, crashk_res.start, crashk_res.end + 1,
518 			       PAGE_KERNEL,
519 			       NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
520 		memblock_clear_nomap(crashk_res.start,
521 				     resource_size(&crashk_res));
522 	}
523 #endif
524 }
525 
526 void mark_rodata_ro(void)
527 {
528 	unsigned long section_size;
529 
530 	/*
531 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
532 	 * to cover NOTES and EXCEPTION_TABLE.
533 	 */
534 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
535 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
536 			    section_size, PAGE_KERNEL_RO);
537 
538 	debug_checkwx();
539 }
540 
541 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
542 				      pgprot_t prot, struct vm_struct *vma,
543 				      int flags, unsigned long vm_flags)
544 {
545 	phys_addr_t pa_start = __pa_symbol(va_start);
546 	unsigned long size = va_end - va_start;
547 
548 	BUG_ON(!PAGE_ALIGNED(pa_start));
549 	BUG_ON(!PAGE_ALIGNED(size));
550 
551 	__create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
552 			     early_pgtable_alloc, flags);
553 
554 	if (!(vm_flags & VM_NO_GUARD))
555 		size += PAGE_SIZE;
556 
557 	vma->addr	= va_start;
558 	vma->phys_addr	= pa_start;
559 	vma->size	= size;
560 	vma->flags	= VM_MAP | vm_flags;
561 	vma->caller	= __builtin_return_address(0);
562 
563 	vm_area_add_early(vma);
564 }
565 
566 static int __init parse_rodata(char *arg)
567 {
568 	int ret = strtobool(arg, &rodata_enabled);
569 	if (!ret) {
570 		rodata_full = false;
571 		return 0;
572 	}
573 
574 	/* permit 'full' in addition to boolean options */
575 	if (strcmp(arg, "full"))
576 		return -EINVAL;
577 
578 	rodata_enabled = true;
579 	rodata_full = true;
580 	return 0;
581 }
582 early_param("rodata", parse_rodata);
583 
584 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
585 static int __init map_entry_trampoline(void)
586 {
587 	pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
588 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
589 
590 	/* The trampoline is always mapped and can therefore be global */
591 	pgprot_val(prot) &= ~PTE_NG;
592 
593 	/* Map only the text into the trampoline page table */
594 	memset(tramp_pg_dir, 0, PGD_SIZE);
595 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
596 			     prot, __pgd_pgtable_alloc, 0);
597 
598 	/* Map both the text and data into the kernel page table */
599 	__set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
600 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
601 		extern char __entry_tramp_data_start[];
602 
603 		__set_fixmap(FIX_ENTRY_TRAMP_DATA,
604 			     __pa_symbol(__entry_tramp_data_start),
605 			     PAGE_KERNEL_RO);
606 	}
607 
608 	return 0;
609 }
610 core_initcall(map_entry_trampoline);
611 #endif
612 
613 /*
614  * Open coded check for BTI, only for use to determine configuration
615  * for early mappings for before the cpufeature code has run.
616  */
617 static bool arm64_early_this_cpu_has_bti(void)
618 {
619 	u64 pfr1;
620 
621 	if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
622 		return false;
623 
624 	pfr1 = read_sysreg_s(SYS_ID_AA64PFR1_EL1);
625 	return cpuid_feature_extract_unsigned_field(pfr1,
626 						    ID_AA64PFR1_BT_SHIFT);
627 }
628 
629 /*
630  * Create fine-grained mappings for the kernel.
631  */
632 static void __init map_kernel(pgd_t *pgdp)
633 {
634 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
635 				vmlinux_initdata, vmlinux_data;
636 
637 	/*
638 	 * External debuggers may need to write directly to the text
639 	 * mapping to install SW breakpoints. Allow this (only) when
640 	 * explicitly requested with rodata=off.
641 	 */
642 	pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
643 
644 	/*
645 	 * If we have a CPU that supports BTI and a kernel built for
646 	 * BTI then mark the kernel executable text as guarded pages
647 	 * now so we don't have to rewrite the page tables later.
648 	 */
649 	if (arm64_early_this_cpu_has_bti())
650 		text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
651 
652 	/*
653 	 * Only rodata will be remapped with different permissions later on,
654 	 * all other segments are allowed to use contiguous mappings.
655 	 */
656 	map_kernel_segment(pgdp, _text, _etext, text_prot, &vmlinux_text, 0,
657 			   VM_NO_GUARD);
658 	map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
659 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
660 	map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
661 			   &vmlinux_inittext, 0, VM_NO_GUARD);
662 	map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
663 			   &vmlinux_initdata, 0, VM_NO_GUARD);
664 	map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
665 
666 	if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_START)))) {
667 		/*
668 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
669 		 * live in the carveout for the swapper_pg_dir. We can simply
670 		 * re-use the existing dir for the fixmap.
671 		 */
672 		set_pgd(pgd_offset_pgd(pgdp, FIXADDR_START),
673 			READ_ONCE(*pgd_offset_k(FIXADDR_START)));
674 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
675 		pgd_t *bm_pgdp;
676 		p4d_t *bm_p4dp;
677 		pud_t *bm_pudp;
678 		/*
679 		 * The fixmap shares its top level pgd entry with the kernel
680 		 * mapping. This can really only occur when we are running
681 		 * with 16k/4 levels, so we can simply reuse the pud level
682 		 * entry instead.
683 		 */
684 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
685 		bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_START);
686 		bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_START);
687 		bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_START);
688 		pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
689 		pud_clear_fixmap();
690 	} else {
691 		BUG();
692 	}
693 
694 	kasan_copy_shadow(pgdp);
695 }
696 
697 void __init paging_init(void)
698 {
699 	pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
700 
701 	map_kernel(pgdp);
702 	map_mem(pgdp);
703 
704 	pgd_clear_fixmap();
705 
706 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
707 	init_mm.pgd = swapper_pg_dir;
708 
709 	memblock_free(__pa_symbol(init_pg_dir),
710 		      __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
711 
712 	memblock_allow_resize();
713 }
714 
715 /*
716  * Check whether a kernel address is valid (derived from arch/x86/).
717  */
718 int kern_addr_valid(unsigned long addr)
719 {
720 	pgd_t *pgdp;
721 	p4d_t *p4dp;
722 	pud_t *pudp, pud;
723 	pmd_t *pmdp, pmd;
724 	pte_t *ptep, pte;
725 
726 	if ((((long)addr) >> VA_BITS) != -1UL)
727 		return 0;
728 
729 	pgdp = pgd_offset_k(addr);
730 	if (pgd_none(READ_ONCE(*pgdp)))
731 		return 0;
732 
733 	p4dp = p4d_offset(pgdp, addr);
734 	if (p4d_none(READ_ONCE(*p4dp)))
735 		return 0;
736 
737 	pudp = pud_offset(p4dp, addr);
738 	pud = READ_ONCE(*pudp);
739 	if (pud_none(pud))
740 		return 0;
741 
742 	if (pud_sect(pud))
743 		return pfn_valid(pud_pfn(pud));
744 
745 	pmdp = pmd_offset(pudp, addr);
746 	pmd = READ_ONCE(*pmdp);
747 	if (pmd_none(pmd))
748 		return 0;
749 
750 	if (pmd_sect(pmd))
751 		return pfn_valid(pmd_pfn(pmd));
752 
753 	ptep = pte_offset_kernel(pmdp, addr);
754 	pte = READ_ONCE(*ptep);
755 	if (pte_none(pte))
756 		return 0;
757 
758 	return pfn_valid(pte_pfn(pte));
759 }
760 
761 #ifdef CONFIG_MEMORY_HOTPLUG
762 static void free_hotplug_page_range(struct page *page, size_t size)
763 {
764 	WARN_ON(PageReserved(page));
765 	free_pages((unsigned long)page_address(page), get_order(size));
766 }
767 
768 static void free_hotplug_pgtable_page(struct page *page)
769 {
770 	free_hotplug_page_range(page, PAGE_SIZE);
771 }
772 
773 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
774 				  unsigned long floor, unsigned long ceiling,
775 				  unsigned long mask)
776 {
777 	start &= mask;
778 	if (start < floor)
779 		return false;
780 
781 	if (ceiling) {
782 		ceiling &= mask;
783 		if (!ceiling)
784 			return false;
785 	}
786 
787 	if (end - 1 > ceiling - 1)
788 		return false;
789 	return true;
790 }
791 
792 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
793 				    unsigned long end, bool free_mapped)
794 {
795 	pte_t *ptep, pte;
796 
797 	do {
798 		ptep = pte_offset_kernel(pmdp, addr);
799 		pte = READ_ONCE(*ptep);
800 		if (pte_none(pte))
801 			continue;
802 
803 		WARN_ON(!pte_present(pte));
804 		pte_clear(&init_mm, addr, ptep);
805 		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
806 		if (free_mapped)
807 			free_hotplug_page_range(pte_page(pte), PAGE_SIZE);
808 	} while (addr += PAGE_SIZE, addr < end);
809 }
810 
811 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
812 				    unsigned long end, bool free_mapped)
813 {
814 	unsigned long next;
815 	pmd_t *pmdp, pmd;
816 
817 	do {
818 		next = pmd_addr_end(addr, end);
819 		pmdp = pmd_offset(pudp, addr);
820 		pmd = READ_ONCE(*pmdp);
821 		if (pmd_none(pmd))
822 			continue;
823 
824 		WARN_ON(!pmd_present(pmd));
825 		if (pmd_sect(pmd)) {
826 			pmd_clear(pmdp);
827 
828 			/*
829 			 * One TLBI should be sufficient here as the PMD_SIZE
830 			 * range is mapped with a single block entry.
831 			 */
832 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
833 			if (free_mapped)
834 				free_hotplug_page_range(pmd_page(pmd),
835 							PMD_SIZE);
836 			continue;
837 		}
838 		WARN_ON(!pmd_table(pmd));
839 		unmap_hotplug_pte_range(pmdp, addr, next, free_mapped);
840 	} while (addr = next, addr < end);
841 }
842 
843 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
844 				    unsigned long end, bool free_mapped)
845 {
846 	unsigned long next;
847 	pud_t *pudp, pud;
848 
849 	do {
850 		next = pud_addr_end(addr, end);
851 		pudp = pud_offset(p4dp, addr);
852 		pud = READ_ONCE(*pudp);
853 		if (pud_none(pud))
854 			continue;
855 
856 		WARN_ON(!pud_present(pud));
857 		if (pud_sect(pud)) {
858 			pud_clear(pudp);
859 
860 			/*
861 			 * One TLBI should be sufficient here as the PUD_SIZE
862 			 * range is mapped with a single block entry.
863 			 */
864 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
865 			if (free_mapped)
866 				free_hotplug_page_range(pud_page(pud),
867 							PUD_SIZE);
868 			continue;
869 		}
870 		WARN_ON(!pud_table(pud));
871 		unmap_hotplug_pmd_range(pudp, addr, next, free_mapped);
872 	} while (addr = next, addr < end);
873 }
874 
875 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
876 				    unsigned long end, bool free_mapped)
877 {
878 	unsigned long next;
879 	p4d_t *p4dp, p4d;
880 
881 	do {
882 		next = p4d_addr_end(addr, end);
883 		p4dp = p4d_offset(pgdp, addr);
884 		p4d = READ_ONCE(*p4dp);
885 		if (p4d_none(p4d))
886 			continue;
887 
888 		WARN_ON(!p4d_present(p4d));
889 		unmap_hotplug_pud_range(p4dp, addr, next, free_mapped);
890 	} while (addr = next, addr < end);
891 }
892 
893 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
894 				bool free_mapped)
895 {
896 	unsigned long next;
897 	pgd_t *pgdp, pgd;
898 
899 	do {
900 		next = pgd_addr_end(addr, end);
901 		pgdp = pgd_offset_k(addr);
902 		pgd = READ_ONCE(*pgdp);
903 		if (pgd_none(pgd))
904 			continue;
905 
906 		WARN_ON(!pgd_present(pgd));
907 		unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
908 	} while (addr = next, addr < end);
909 }
910 
911 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
912 				 unsigned long end, unsigned long floor,
913 				 unsigned long ceiling)
914 {
915 	pte_t *ptep, pte;
916 	unsigned long i, start = addr;
917 
918 	do {
919 		ptep = pte_offset_kernel(pmdp, addr);
920 		pte = READ_ONCE(*ptep);
921 
922 		/*
923 		 * This is just a sanity check here which verifies that
924 		 * pte clearing has been done by earlier unmap loops.
925 		 */
926 		WARN_ON(!pte_none(pte));
927 	} while (addr += PAGE_SIZE, addr < end);
928 
929 	if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
930 		return;
931 
932 	/*
933 	 * Check whether we can free the pte page if the rest of the
934 	 * entries are empty. Overlap with other regions have been
935 	 * handled by the floor/ceiling check.
936 	 */
937 	ptep = pte_offset_kernel(pmdp, 0UL);
938 	for (i = 0; i < PTRS_PER_PTE; i++) {
939 		if (!pte_none(READ_ONCE(ptep[i])))
940 			return;
941 	}
942 
943 	pmd_clear(pmdp);
944 	__flush_tlb_kernel_pgtable(start);
945 	free_hotplug_pgtable_page(virt_to_page(ptep));
946 }
947 
948 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
949 				 unsigned long end, unsigned long floor,
950 				 unsigned long ceiling)
951 {
952 	pmd_t *pmdp, pmd;
953 	unsigned long i, next, start = addr;
954 
955 	do {
956 		next = pmd_addr_end(addr, end);
957 		pmdp = pmd_offset(pudp, addr);
958 		pmd = READ_ONCE(*pmdp);
959 		if (pmd_none(pmd))
960 			continue;
961 
962 		WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
963 		free_empty_pte_table(pmdp, addr, next, floor, ceiling);
964 	} while (addr = next, addr < end);
965 
966 	if (CONFIG_PGTABLE_LEVELS <= 2)
967 		return;
968 
969 	if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
970 		return;
971 
972 	/*
973 	 * Check whether we can free the pmd page if the rest of the
974 	 * entries are empty. Overlap with other regions have been
975 	 * handled by the floor/ceiling check.
976 	 */
977 	pmdp = pmd_offset(pudp, 0UL);
978 	for (i = 0; i < PTRS_PER_PMD; i++) {
979 		if (!pmd_none(READ_ONCE(pmdp[i])))
980 			return;
981 	}
982 
983 	pud_clear(pudp);
984 	__flush_tlb_kernel_pgtable(start);
985 	free_hotplug_pgtable_page(virt_to_page(pmdp));
986 }
987 
988 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
989 				 unsigned long end, unsigned long floor,
990 				 unsigned long ceiling)
991 {
992 	pud_t *pudp, pud;
993 	unsigned long i, next, start = addr;
994 
995 	do {
996 		next = pud_addr_end(addr, end);
997 		pudp = pud_offset(p4dp, addr);
998 		pud = READ_ONCE(*pudp);
999 		if (pud_none(pud))
1000 			continue;
1001 
1002 		WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1003 		free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1004 	} while (addr = next, addr < end);
1005 
1006 	if (CONFIG_PGTABLE_LEVELS <= 3)
1007 		return;
1008 
1009 	if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1010 		return;
1011 
1012 	/*
1013 	 * Check whether we can free the pud page if the rest of the
1014 	 * entries are empty. Overlap with other regions have been
1015 	 * handled by the floor/ceiling check.
1016 	 */
1017 	pudp = pud_offset(p4dp, 0UL);
1018 	for (i = 0; i < PTRS_PER_PUD; i++) {
1019 		if (!pud_none(READ_ONCE(pudp[i])))
1020 			return;
1021 	}
1022 
1023 	p4d_clear(p4dp);
1024 	__flush_tlb_kernel_pgtable(start);
1025 	free_hotplug_pgtable_page(virt_to_page(pudp));
1026 }
1027 
1028 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1029 				 unsigned long end, unsigned long floor,
1030 				 unsigned long ceiling)
1031 {
1032 	unsigned long next;
1033 	p4d_t *p4dp, p4d;
1034 
1035 	do {
1036 		next = p4d_addr_end(addr, end);
1037 		p4dp = p4d_offset(pgdp, addr);
1038 		p4d = READ_ONCE(*p4dp);
1039 		if (p4d_none(p4d))
1040 			continue;
1041 
1042 		WARN_ON(!p4d_present(p4d));
1043 		free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1044 	} while (addr = next, addr < end);
1045 }
1046 
1047 static void free_empty_tables(unsigned long addr, unsigned long end,
1048 			      unsigned long floor, unsigned long ceiling)
1049 {
1050 	unsigned long next;
1051 	pgd_t *pgdp, pgd;
1052 
1053 	do {
1054 		next = pgd_addr_end(addr, end);
1055 		pgdp = pgd_offset_k(addr);
1056 		pgd = READ_ONCE(*pgdp);
1057 		if (pgd_none(pgd))
1058 			continue;
1059 
1060 		WARN_ON(!pgd_present(pgd));
1061 		free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1062 	} while (addr = next, addr < end);
1063 }
1064 #endif
1065 
1066 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1067 #if !ARM64_SWAPPER_USES_SECTION_MAPS
1068 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1069 		struct vmem_altmap *altmap)
1070 {
1071 	return vmemmap_populate_basepages(start, end, node);
1072 }
1073 #else	/* !ARM64_SWAPPER_USES_SECTION_MAPS */
1074 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1075 		struct vmem_altmap *altmap)
1076 {
1077 	unsigned long addr = start;
1078 	unsigned long next;
1079 	pgd_t *pgdp;
1080 	p4d_t *p4dp;
1081 	pud_t *pudp;
1082 	pmd_t *pmdp;
1083 
1084 	do {
1085 		next = pmd_addr_end(addr, end);
1086 
1087 		pgdp = vmemmap_pgd_populate(addr, node);
1088 		if (!pgdp)
1089 			return -ENOMEM;
1090 
1091 		p4dp = vmemmap_p4d_populate(pgdp, addr, node);
1092 		if (!p4dp)
1093 			return -ENOMEM;
1094 
1095 		pudp = vmemmap_pud_populate(p4dp, addr, node);
1096 		if (!pudp)
1097 			return -ENOMEM;
1098 
1099 		pmdp = pmd_offset(pudp, addr);
1100 		if (pmd_none(READ_ONCE(*pmdp))) {
1101 			void *p = NULL;
1102 
1103 			p = vmemmap_alloc_block_buf(PMD_SIZE, node);
1104 			if (!p)
1105 				return -ENOMEM;
1106 
1107 			pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1108 		} else
1109 			vmemmap_verify((pte_t *)pmdp, node, addr, next);
1110 	} while (addr = next, addr != end);
1111 
1112 	return 0;
1113 }
1114 #endif	/* !ARM64_SWAPPER_USES_SECTION_MAPS */
1115 void vmemmap_free(unsigned long start, unsigned long end,
1116 		struct vmem_altmap *altmap)
1117 {
1118 #ifdef CONFIG_MEMORY_HOTPLUG
1119 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1120 
1121 	unmap_hotplug_range(start, end, true);
1122 	free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1123 #endif
1124 }
1125 #endif	/* CONFIG_SPARSEMEM_VMEMMAP */
1126 
1127 static inline pud_t * fixmap_pud(unsigned long addr)
1128 {
1129 	pgd_t *pgdp = pgd_offset_k(addr);
1130 	p4d_t *p4dp = p4d_offset(pgdp, addr);
1131 	p4d_t p4d = READ_ONCE(*p4dp);
1132 
1133 	BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
1134 
1135 	return pud_offset_kimg(p4dp, addr);
1136 }
1137 
1138 static inline pmd_t * fixmap_pmd(unsigned long addr)
1139 {
1140 	pud_t *pudp = fixmap_pud(addr);
1141 	pud_t pud = READ_ONCE(*pudp);
1142 
1143 	BUG_ON(pud_none(pud) || pud_bad(pud));
1144 
1145 	return pmd_offset_kimg(pudp, addr);
1146 }
1147 
1148 static inline pte_t * fixmap_pte(unsigned long addr)
1149 {
1150 	return &bm_pte[pte_index(addr)];
1151 }
1152 
1153 /*
1154  * The p*d_populate functions call virt_to_phys implicitly so they can't be used
1155  * directly on kernel symbols (bm_p*d). This function is called too early to use
1156  * lm_alias so __p*d_populate functions must be used to populate with the
1157  * physical address from __pa_symbol.
1158  */
1159 void __init early_fixmap_init(void)
1160 {
1161 	pgd_t *pgdp;
1162 	p4d_t *p4dp, p4d;
1163 	pud_t *pudp;
1164 	pmd_t *pmdp;
1165 	unsigned long addr = FIXADDR_START;
1166 
1167 	pgdp = pgd_offset_k(addr);
1168 	p4dp = p4d_offset(pgdp, addr);
1169 	p4d = READ_ONCE(*p4dp);
1170 	if (CONFIG_PGTABLE_LEVELS > 3 &&
1171 	    !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
1172 		/*
1173 		 * We only end up here if the kernel mapping and the fixmap
1174 		 * share the top level pgd entry, which should only happen on
1175 		 * 16k/4 levels configurations.
1176 		 */
1177 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
1178 		pudp = pud_offset_kimg(p4dp, addr);
1179 	} else {
1180 		if (p4d_none(p4d))
1181 			__p4d_populate(p4dp, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
1182 		pudp = fixmap_pud(addr);
1183 	}
1184 	if (pud_none(READ_ONCE(*pudp)))
1185 		__pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
1186 	pmdp = fixmap_pmd(addr);
1187 	__pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
1188 
1189 	/*
1190 	 * The boot-ioremap range spans multiple pmds, for which
1191 	 * we are not prepared:
1192 	 */
1193 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1194 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1195 
1196 	if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
1197 	     || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
1198 		WARN_ON(1);
1199 		pr_warn("pmdp %p != %p, %p\n",
1200 			pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
1201 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
1202 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1203 			fix_to_virt(FIX_BTMAP_BEGIN));
1204 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1205 			fix_to_virt(FIX_BTMAP_END));
1206 
1207 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1208 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1209 	}
1210 }
1211 
1212 /*
1213  * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
1214  * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
1215  */
1216 void __set_fixmap(enum fixed_addresses idx,
1217 			       phys_addr_t phys, pgprot_t flags)
1218 {
1219 	unsigned long addr = __fix_to_virt(idx);
1220 	pte_t *ptep;
1221 
1222 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
1223 
1224 	ptep = fixmap_pte(addr);
1225 
1226 	if (pgprot_val(flags)) {
1227 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
1228 	} else {
1229 		pte_clear(&init_mm, addr, ptep);
1230 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
1231 	}
1232 }
1233 
1234 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
1235 {
1236 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
1237 	int offset;
1238 	void *dt_virt;
1239 
1240 	/*
1241 	 * Check whether the physical FDT address is set and meets the minimum
1242 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
1243 	 * at least 8 bytes so that we can always access the magic and size
1244 	 * fields of the FDT header after mapping the first chunk, double check
1245 	 * here if that is indeed the case.
1246 	 */
1247 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
1248 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
1249 		return NULL;
1250 
1251 	/*
1252 	 * Make sure that the FDT region can be mapped without the need to
1253 	 * allocate additional translation table pages, so that it is safe
1254 	 * to call create_mapping_noalloc() this early.
1255 	 *
1256 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
1257 	 * be in the same PMD as the rest of the fixmap.
1258 	 * On 4k pages, we'll use section mappings for the FDT so we only
1259 	 * have to be in the same PUD.
1260 	 */
1261 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
1262 
1263 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
1264 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
1265 
1266 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
1267 	dt_virt = (void *)dt_virt_base + offset;
1268 
1269 	/* map the first chunk so we can read the size from the header */
1270 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
1271 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
1272 
1273 	if (fdt_magic(dt_virt) != FDT_MAGIC)
1274 		return NULL;
1275 
1276 	*size = fdt_totalsize(dt_virt);
1277 	if (*size > MAX_FDT_SIZE)
1278 		return NULL;
1279 
1280 	if (offset + *size > SWAPPER_BLOCK_SIZE)
1281 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
1282 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
1283 
1284 	return dt_virt;
1285 }
1286 
1287 int __init arch_ioremap_p4d_supported(void)
1288 {
1289 	return 0;
1290 }
1291 
1292 int __init arch_ioremap_pud_supported(void)
1293 {
1294 	/*
1295 	 * Only 4k granule supports level 1 block mappings.
1296 	 * SW table walks can't handle removal of intermediate entries.
1297 	 */
1298 	return IS_ENABLED(CONFIG_ARM64_4K_PAGES) &&
1299 	       !IS_ENABLED(CONFIG_PTDUMP_DEBUGFS);
1300 }
1301 
1302 int __init arch_ioremap_pmd_supported(void)
1303 {
1304 	/* See arch_ioremap_pud_supported() */
1305 	return !IS_ENABLED(CONFIG_PTDUMP_DEBUGFS);
1306 }
1307 
1308 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1309 {
1310 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1311 
1312 	/* Only allow permission changes for now */
1313 	if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1314 				   pud_val(new_pud)))
1315 		return 0;
1316 
1317 	VM_BUG_ON(phys & ~PUD_MASK);
1318 	set_pud(pudp, new_pud);
1319 	return 1;
1320 }
1321 
1322 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1323 {
1324 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1325 
1326 	/* Only allow permission changes for now */
1327 	if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1328 				   pmd_val(new_pmd)))
1329 		return 0;
1330 
1331 	VM_BUG_ON(phys & ~PMD_MASK);
1332 	set_pmd(pmdp, new_pmd);
1333 	return 1;
1334 }
1335 
1336 int pud_clear_huge(pud_t *pudp)
1337 {
1338 	if (!pud_sect(READ_ONCE(*pudp)))
1339 		return 0;
1340 	pud_clear(pudp);
1341 	return 1;
1342 }
1343 
1344 int pmd_clear_huge(pmd_t *pmdp)
1345 {
1346 	if (!pmd_sect(READ_ONCE(*pmdp)))
1347 		return 0;
1348 	pmd_clear(pmdp);
1349 	return 1;
1350 }
1351 
1352 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1353 {
1354 	pte_t *table;
1355 	pmd_t pmd;
1356 
1357 	pmd = READ_ONCE(*pmdp);
1358 
1359 	if (!pmd_table(pmd)) {
1360 		VM_WARN_ON(1);
1361 		return 1;
1362 	}
1363 
1364 	table = pte_offset_kernel(pmdp, addr);
1365 	pmd_clear(pmdp);
1366 	__flush_tlb_kernel_pgtable(addr);
1367 	pte_free_kernel(NULL, table);
1368 	return 1;
1369 }
1370 
1371 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1372 {
1373 	pmd_t *table;
1374 	pmd_t *pmdp;
1375 	pud_t pud;
1376 	unsigned long next, end;
1377 
1378 	pud = READ_ONCE(*pudp);
1379 
1380 	if (!pud_table(pud)) {
1381 		VM_WARN_ON(1);
1382 		return 1;
1383 	}
1384 
1385 	table = pmd_offset(pudp, addr);
1386 	pmdp = table;
1387 	next = addr;
1388 	end = addr + PUD_SIZE;
1389 	do {
1390 		pmd_free_pte_page(pmdp, next);
1391 	} while (pmdp++, next += PMD_SIZE, next != end);
1392 
1393 	pud_clear(pudp);
1394 	__flush_tlb_kernel_pgtable(addr);
1395 	pmd_free(NULL, table);
1396 	return 1;
1397 }
1398 
1399 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1400 {
1401 	return 0;	/* Don't attempt a block mapping */
1402 }
1403 
1404 #ifdef CONFIG_MEMORY_HOTPLUG
1405 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1406 {
1407 	unsigned long end = start + size;
1408 
1409 	WARN_ON(pgdir != init_mm.pgd);
1410 	WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1411 
1412 	unmap_hotplug_range(start, end, false);
1413 	free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1414 }
1415 
1416 int arch_add_memory(int nid, u64 start, u64 size,
1417 		    struct mhp_params *params)
1418 {
1419 	int ret, flags = 0;
1420 
1421 	if (rodata_full || debug_pagealloc_enabled())
1422 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1423 
1424 	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1425 			     size, params->pgprot, __pgd_pgtable_alloc,
1426 			     flags);
1427 
1428 	memblock_clear_nomap(start, size);
1429 
1430 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1431 			   params);
1432 	if (ret)
1433 		__remove_pgd_mapping(swapper_pg_dir,
1434 				     __phys_to_virt(start), size);
1435 	return ret;
1436 }
1437 
1438 void arch_remove_memory(int nid, u64 start, u64 size,
1439 			struct vmem_altmap *altmap)
1440 {
1441 	unsigned long start_pfn = start >> PAGE_SHIFT;
1442 	unsigned long nr_pages = size >> PAGE_SHIFT;
1443 
1444 	__remove_pages(start_pfn, nr_pages, altmap);
1445 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1446 }
1447 
1448 /*
1449  * This memory hotplug notifier helps prevent boot memory from being
1450  * inadvertently removed as it blocks pfn range offlining process in
1451  * __offline_pages(). Hence this prevents both offlining as well as
1452  * removal process for boot memory which is initially always online.
1453  * In future if and when boot memory could be removed, this notifier
1454  * should be dropped and free_hotplug_page_range() should handle any
1455  * reserved pages allocated during boot.
1456  */
1457 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1458 					   unsigned long action, void *data)
1459 {
1460 	struct mem_section *ms;
1461 	struct memory_notify *arg = data;
1462 	unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1463 	unsigned long pfn = arg->start_pfn;
1464 
1465 	if (action != MEM_GOING_OFFLINE)
1466 		return NOTIFY_OK;
1467 
1468 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1469 		ms = __pfn_to_section(pfn);
1470 		if (early_section(ms))
1471 			return NOTIFY_BAD;
1472 	}
1473 	return NOTIFY_OK;
1474 }
1475 
1476 static struct notifier_block prevent_bootmem_remove_nb = {
1477 	.notifier_call = prevent_bootmem_remove_notifier,
1478 };
1479 
1480 static int __init prevent_bootmem_remove_init(void)
1481 {
1482 	return register_memory_notifier(&prevent_bootmem_remove_nb);
1483 }
1484 device_initcall(prevent_bootmem_remove_init);
1485 #endif
1486