xref: /linux/arch/arm64/mm/mmu.c (revision 3c4fc7bf4c9e66fe71abcbf93f62f4ddb89b7f15)
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/memremap.h>
21 #include <linux/memory.h>
22 #include <linux/fs.h>
23 #include <linux/io.h>
24 #include <linux/mm.h>
25 #include <linux/vmalloc.h>
26 #include <linux/set_memory.h>
27 
28 #include <asm/barrier.h>
29 #include <asm/cputype.h>
30 #include <asm/fixmap.h>
31 #include <asm/kasan.h>
32 #include <asm/kernel-pgtable.h>
33 #include <asm/sections.h>
34 #include <asm/setup.h>
35 #include <linux/sizes.h>
36 #include <asm/tlb.h>
37 #include <asm/mmu_context.h>
38 #include <asm/ptdump.h>
39 #include <asm/tlbflush.h>
40 #include <asm/pgalloc.h>
41 
42 #define NO_BLOCK_MAPPINGS	BIT(0)
43 #define NO_CONT_MAPPINGS	BIT(1)
44 #define NO_EXEC_MAPPINGS	BIT(2)	/* assumes FEAT_HPDS is not used */
45 
46 int idmap_t0sz __ro_after_init;
47 
48 #if VA_BITS > 48
49 u64 vabits_actual __ro_after_init = VA_BITS_MIN;
50 EXPORT_SYMBOL(vabits_actual);
51 #endif
52 
53 u64 kimage_vaddr __ro_after_init = (u64)&_text;
54 EXPORT_SYMBOL(kimage_vaddr);
55 
56 u64 kimage_voffset __ro_after_init;
57 EXPORT_SYMBOL(kimage_voffset);
58 
59 u32 __boot_cpu_mode[] = { BOOT_CPU_MODE_EL2, BOOT_CPU_MODE_EL1 };
60 
61 /*
62  * The booting CPU updates the failed status @__early_cpu_boot_status,
63  * with MMU turned off.
64  */
65 long __section(".mmuoff.data.write") __early_cpu_boot_status;
66 
67 /*
68  * Empty_zero_page is a special page that is used for zero-initialized data
69  * and COW.
70  */
71 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
72 EXPORT_SYMBOL(empty_zero_page);
73 
74 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
75 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
76 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
77 
78 static DEFINE_SPINLOCK(swapper_pgdir_lock);
79 static DEFINE_MUTEX(fixmap_lock);
80 
81 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
82 {
83 	pgd_t *fixmap_pgdp;
84 
85 	spin_lock(&swapper_pgdir_lock);
86 	fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
87 	WRITE_ONCE(*fixmap_pgdp, pgd);
88 	/*
89 	 * We need dsb(ishst) here to ensure the page-table-walker sees
90 	 * our new entry before set_p?d() returns. The fixmap's
91 	 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
92 	 */
93 	pgd_clear_fixmap();
94 	spin_unlock(&swapper_pgdir_lock);
95 }
96 
97 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
98 			      unsigned long size, pgprot_t vma_prot)
99 {
100 	if (!pfn_is_map_memory(pfn))
101 		return pgprot_noncached(vma_prot);
102 	else if (file->f_flags & O_SYNC)
103 		return pgprot_writecombine(vma_prot);
104 	return vma_prot;
105 }
106 EXPORT_SYMBOL(phys_mem_access_prot);
107 
108 static phys_addr_t __init early_pgtable_alloc(int shift)
109 {
110 	phys_addr_t phys;
111 	void *ptr;
112 
113 	phys = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0,
114 					 MEMBLOCK_ALLOC_NOLEAKTRACE);
115 	if (!phys)
116 		panic("Failed to allocate page table page\n");
117 
118 	/*
119 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
120 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
121 	 * any level of table.
122 	 */
123 	ptr = pte_set_fixmap(phys);
124 
125 	memset(ptr, 0, PAGE_SIZE);
126 
127 	/*
128 	 * Implicit barriers also ensure the zeroed page is visible to the page
129 	 * table walker
130 	 */
131 	pte_clear_fixmap();
132 
133 	return phys;
134 }
135 
136 static bool pgattr_change_is_safe(u64 old, u64 new)
137 {
138 	/*
139 	 * The following mapping attributes may be updated in live
140 	 * kernel mappings without the need for break-before-make.
141 	 */
142 	pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
143 
144 	/* creating or taking down mappings is always safe */
145 	if (old == 0 || new == 0)
146 		return true;
147 
148 	/* live contiguous mappings may not be manipulated at all */
149 	if ((old | new) & PTE_CONT)
150 		return false;
151 
152 	/* Transitioning from Non-Global to Global is unsafe */
153 	if (old & ~new & PTE_NG)
154 		return false;
155 
156 	/*
157 	 * Changing the memory type between Normal and Normal-Tagged is safe
158 	 * since Tagged is considered a permission attribute from the
159 	 * mismatched attribute aliases perspective.
160 	 */
161 	if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
162 	     (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
163 	    ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
164 	     (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
165 		mask |= PTE_ATTRINDX_MASK;
166 
167 	return ((old ^ new) & ~mask) == 0;
168 }
169 
170 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
171 		     phys_addr_t phys, pgprot_t prot)
172 {
173 	pte_t *ptep;
174 
175 	ptep = pte_set_fixmap_offset(pmdp, addr);
176 	do {
177 		pte_t old_pte = READ_ONCE(*ptep);
178 
179 		set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
180 
181 		/*
182 		 * After the PTE entry has been populated once, we
183 		 * only allow updates to the permission attributes.
184 		 */
185 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
186 					      READ_ONCE(pte_val(*ptep))));
187 
188 		phys += PAGE_SIZE;
189 	} while (ptep++, addr += PAGE_SIZE, addr != end);
190 
191 	pte_clear_fixmap();
192 }
193 
194 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
195 				unsigned long end, phys_addr_t phys,
196 				pgprot_t prot,
197 				phys_addr_t (*pgtable_alloc)(int),
198 				int flags)
199 {
200 	unsigned long next;
201 	pmd_t pmd = READ_ONCE(*pmdp);
202 
203 	BUG_ON(pmd_sect(pmd));
204 	if (pmd_none(pmd)) {
205 		pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
206 		phys_addr_t pte_phys;
207 
208 		if (flags & NO_EXEC_MAPPINGS)
209 			pmdval |= PMD_TABLE_PXN;
210 		BUG_ON(!pgtable_alloc);
211 		pte_phys = pgtable_alloc(PAGE_SHIFT);
212 		__pmd_populate(pmdp, pte_phys, pmdval);
213 		pmd = READ_ONCE(*pmdp);
214 	}
215 	BUG_ON(pmd_bad(pmd));
216 
217 	do {
218 		pgprot_t __prot = prot;
219 
220 		next = pte_cont_addr_end(addr, end);
221 
222 		/* use a contiguous mapping if the range is suitably aligned */
223 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
224 		    (flags & NO_CONT_MAPPINGS) == 0)
225 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
226 
227 		init_pte(pmdp, addr, next, phys, __prot);
228 
229 		phys += next - addr;
230 	} while (addr = next, addr != end);
231 }
232 
233 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
234 		     phys_addr_t phys, pgprot_t prot,
235 		     phys_addr_t (*pgtable_alloc)(int), int flags)
236 {
237 	unsigned long next;
238 	pmd_t *pmdp;
239 
240 	pmdp = pmd_set_fixmap_offset(pudp, addr);
241 	do {
242 		pmd_t old_pmd = READ_ONCE(*pmdp);
243 
244 		next = pmd_addr_end(addr, end);
245 
246 		/* try section mapping first */
247 		if (((addr | next | phys) & ~PMD_MASK) == 0 &&
248 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
249 			pmd_set_huge(pmdp, phys, prot);
250 
251 			/*
252 			 * After the PMD entry has been populated once, we
253 			 * only allow updates to the permission attributes.
254 			 */
255 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
256 						      READ_ONCE(pmd_val(*pmdp))));
257 		} else {
258 			alloc_init_cont_pte(pmdp, addr, next, phys, prot,
259 					    pgtable_alloc, flags);
260 
261 			BUG_ON(pmd_val(old_pmd) != 0 &&
262 			       pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
263 		}
264 		phys += next - addr;
265 	} while (pmdp++, addr = next, addr != end);
266 
267 	pmd_clear_fixmap();
268 }
269 
270 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
271 				unsigned long end, phys_addr_t phys,
272 				pgprot_t prot,
273 				phys_addr_t (*pgtable_alloc)(int), int flags)
274 {
275 	unsigned long next;
276 	pud_t pud = READ_ONCE(*pudp);
277 
278 	/*
279 	 * Check for initial section mappings in the pgd/pud.
280 	 */
281 	BUG_ON(pud_sect(pud));
282 	if (pud_none(pud)) {
283 		pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
284 		phys_addr_t pmd_phys;
285 
286 		if (flags & NO_EXEC_MAPPINGS)
287 			pudval |= PUD_TABLE_PXN;
288 		BUG_ON(!pgtable_alloc);
289 		pmd_phys = pgtable_alloc(PMD_SHIFT);
290 		__pud_populate(pudp, pmd_phys, pudval);
291 		pud = READ_ONCE(*pudp);
292 	}
293 	BUG_ON(pud_bad(pud));
294 
295 	do {
296 		pgprot_t __prot = prot;
297 
298 		next = pmd_cont_addr_end(addr, end);
299 
300 		/* use a contiguous mapping if the range is suitably aligned */
301 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
302 		    (flags & NO_CONT_MAPPINGS) == 0)
303 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
304 
305 		init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
306 
307 		phys += next - addr;
308 	} while (addr = next, addr != end);
309 }
310 
311 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
312 			   phys_addr_t phys, pgprot_t prot,
313 			   phys_addr_t (*pgtable_alloc)(int),
314 			   int flags)
315 {
316 	unsigned long next;
317 	pud_t *pudp;
318 	p4d_t *p4dp = p4d_offset(pgdp, addr);
319 	p4d_t p4d = READ_ONCE(*p4dp);
320 
321 	if (p4d_none(p4d)) {
322 		p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
323 		phys_addr_t pud_phys;
324 
325 		if (flags & NO_EXEC_MAPPINGS)
326 			p4dval |= P4D_TABLE_PXN;
327 		BUG_ON(!pgtable_alloc);
328 		pud_phys = pgtable_alloc(PUD_SHIFT);
329 		__p4d_populate(p4dp, pud_phys, p4dval);
330 		p4d = READ_ONCE(*p4dp);
331 	}
332 	BUG_ON(p4d_bad(p4d));
333 
334 	pudp = pud_set_fixmap_offset(p4dp, addr);
335 	do {
336 		pud_t old_pud = READ_ONCE(*pudp);
337 
338 		next = pud_addr_end(addr, end);
339 
340 		/*
341 		 * For 4K granule only, attempt to put down a 1GB block
342 		 */
343 		if (pud_sect_supported() &&
344 		   ((addr | next | phys) & ~PUD_MASK) == 0 &&
345 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
346 			pud_set_huge(pudp, phys, prot);
347 
348 			/*
349 			 * After the PUD entry has been populated once, we
350 			 * only allow updates to the permission attributes.
351 			 */
352 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
353 						      READ_ONCE(pud_val(*pudp))));
354 		} else {
355 			alloc_init_cont_pmd(pudp, addr, next, phys, prot,
356 					    pgtable_alloc, flags);
357 
358 			BUG_ON(pud_val(old_pud) != 0 &&
359 			       pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
360 		}
361 		phys += next - addr;
362 	} while (pudp++, addr = next, addr != end);
363 
364 	pud_clear_fixmap();
365 }
366 
367 static void __create_pgd_mapping_locked(pgd_t *pgdir, phys_addr_t phys,
368 					unsigned long virt, phys_addr_t size,
369 					pgprot_t prot,
370 					phys_addr_t (*pgtable_alloc)(int),
371 					int flags)
372 {
373 	unsigned long addr, end, next;
374 	pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
375 
376 	/*
377 	 * If the virtual and physical address don't have the same offset
378 	 * within a page, we cannot map the region as the caller expects.
379 	 */
380 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
381 		return;
382 
383 	phys &= PAGE_MASK;
384 	addr = virt & PAGE_MASK;
385 	end = PAGE_ALIGN(virt + size);
386 
387 	do {
388 		next = pgd_addr_end(addr, end);
389 		alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
390 			       flags);
391 		phys += next - addr;
392 	} while (pgdp++, addr = next, addr != end);
393 }
394 
395 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
396 				 unsigned long virt, phys_addr_t size,
397 				 pgprot_t prot,
398 				 phys_addr_t (*pgtable_alloc)(int),
399 				 int flags)
400 {
401 	mutex_lock(&fixmap_lock);
402 	__create_pgd_mapping_locked(pgdir, phys, virt, size, prot,
403 				    pgtable_alloc, flags);
404 	mutex_unlock(&fixmap_lock);
405 }
406 
407 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
408 extern __alias(__create_pgd_mapping_locked)
409 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
410 			     phys_addr_t size, pgprot_t prot,
411 			     phys_addr_t (*pgtable_alloc)(int), int flags);
412 #endif
413 
414 static phys_addr_t __pgd_pgtable_alloc(int shift)
415 {
416 	void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
417 	BUG_ON(!ptr);
418 
419 	/* Ensure the zeroed page is visible to the page table walker */
420 	dsb(ishst);
421 	return __pa(ptr);
422 }
423 
424 static phys_addr_t pgd_pgtable_alloc(int shift)
425 {
426 	phys_addr_t pa = __pgd_pgtable_alloc(shift);
427 
428 	/*
429 	 * Call proper page table ctor in case later we need to
430 	 * call core mm functions like apply_to_page_range() on
431 	 * this pre-allocated page table.
432 	 *
433 	 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
434 	 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
435 	 */
436 	if (shift == PAGE_SHIFT)
437 		BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
438 	else if (shift == PMD_SHIFT)
439 		BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
440 
441 	return pa;
442 }
443 
444 /*
445  * This function can only be used to modify existing table entries,
446  * without allocating new levels of table. Note that this permits the
447  * creation of new section or page entries.
448  */
449 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
450 				  phys_addr_t size, pgprot_t prot)
451 {
452 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
453 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
454 			&phys, virt);
455 		return;
456 	}
457 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
458 			     NO_CONT_MAPPINGS);
459 }
460 
461 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
462 			       unsigned long virt, phys_addr_t size,
463 			       pgprot_t prot, bool page_mappings_only)
464 {
465 	int flags = 0;
466 
467 	BUG_ON(mm == &init_mm);
468 
469 	if (page_mappings_only)
470 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
471 
472 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
473 			     pgd_pgtable_alloc, flags);
474 }
475 
476 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
477 				phys_addr_t size, pgprot_t prot)
478 {
479 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
480 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
481 			&phys, virt);
482 		return;
483 	}
484 
485 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
486 			     NO_CONT_MAPPINGS);
487 
488 	/* flush the TLBs after updating live kernel mappings */
489 	flush_tlb_kernel_range(virt, virt + size);
490 }
491 
492 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
493 				  phys_addr_t end, pgprot_t prot, int flags)
494 {
495 	__create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
496 			     prot, early_pgtable_alloc, flags);
497 }
498 
499 void __init mark_linear_text_alias_ro(void)
500 {
501 	/*
502 	 * Remove the write permissions from the linear alias of .text/.rodata
503 	 */
504 	update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
505 			    (unsigned long)__init_begin - (unsigned long)_stext,
506 			    PAGE_KERNEL_RO);
507 }
508 
509 static bool crash_mem_map __initdata;
510 
511 static int __init enable_crash_mem_map(char *arg)
512 {
513 	/*
514 	 * Proper parameter parsing is done by reserve_crashkernel(). We only
515 	 * need to know if the linear map has to avoid block mappings so that
516 	 * the crashkernel reservations can be unmapped later.
517 	 */
518 	crash_mem_map = true;
519 
520 	return 0;
521 }
522 early_param("crashkernel", enable_crash_mem_map);
523 
524 static void __init map_mem(pgd_t *pgdp)
525 {
526 	static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
527 	phys_addr_t kernel_start = __pa_symbol(_stext);
528 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
529 	phys_addr_t start, end;
530 	int flags = NO_EXEC_MAPPINGS;
531 	u64 i;
532 
533 	/*
534 	 * Setting hierarchical PXNTable attributes on table entries covering
535 	 * the linear region is only possible if it is guaranteed that no table
536 	 * entries at any level are being shared between the linear region and
537 	 * the vmalloc region. Check whether this is true for the PGD level, in
538 	 * which case it is guaranteed to be true for all other levels as well.
539 	 */
540 	BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
541 
542 	if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
543 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
544 
545 	/*
546 	 * Take care not to create a writable alias for the
547 	 * read-only text and rodata sections of the kernel image.
548 	 * So temporarily mark them as NOMAP to skip mappings in
549 	 * the following for-loop
550 	 */
551 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
552 
553 #ifdef CONFIG_KEXEC_CORE
554 	if (crash_mem_map) {
555 		if (defer_reserve_crashkernel())
556 			flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
557 		else if (crashk_res.end)
558 			memblock_mark_nomap(crashk_res.start,
559 			    resource_size(&crashk_res));
560 	}
561 #endif
562 
563 	/* map all the memory banks */
564 	for_each_mem_range(i, &start, &end) {
565 		if (start >= end)
566 			break;
567 		/*
568 		 * The linear map must allow allocation tags reading/writing
569 		 * if MTE is present. Otherwise, it has the same attributes as
570 		 * PAGE_KERNEL.
571 		 */
572 		__map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
573 			       flags);
574 	}
575 
576 	/*
577 	 * Map the linear alias of the [_stext, __init_begin) interval
578 	 * as non-executable now, and remove the write permission in
579 	 * mark_linear_text_alias_ro() below (which will be called after
580 	 * alternative patching has completed). This makes the contents
581 	 * of the region accessible to subsystems such as hibernate,
582 	 * but protects it from inadvertent modification or execution.
583 	 * Note that contiguous mappings cannot be remapped in this way,
584 	 * so we should avoid them here.
585 	 */
586 	__map_memblock(pgdp, kernel_start, kernel_end,
587 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
588 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
589 
590 	/*
591 	 * Use page-level mappings here so that we can shrink the region
592 	 * in page granularity and put back unused memory to buddy system
593 	 * through /sys/kernel/kexec_crash_size interface.
594 	 */
595 #ifdef CONFIG_KEXEC_CORE
596 	if (crash_mem_map && !defer_reserve_crashkernel()) {
597 		if (crashk_res.end) {
598 			__map_memblock(pgdp, crashk_res.start,
599 				       crashk_res.end + 1,
600 				       PAGE_KERNEL,
601 				       NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
602 			memblock_clear_nomap(crashk_res.start,
603 					     resource_size(&crashk_res));
604 		}
605 	}
606 #endif
607 }
608 
609 void mark_rodata_ro(void)
610 {
611 	unsigned long section_size;
612 
613 	/*
614 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
615 	 * to cover NOTES and EXCEPTION_TABLE.
616 	 */
617 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
618 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
619 			    section_size, PAGE_KERNEL_RO);
620 
621 	debug_checkwx();
622 }
623 
624 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
625 				      pgprot_t prot, struct vm_struct *vma,
626 				      int flags, unsigned long vm_flags)
627 {
628 	phys_addr_t pa_start = __pa_symbol(va_start);
629 	unsigned long size = va_end - va_start;
630 
631 	BUG_ON(!PAGE_ALIGNED(pa_start));
632 	BUG_ON(!PAGE_ALIGNED(size));
633 
634 	__create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
635 			     early_pgtable_alloc, flags);
636 
637 	if (!(vm_flags & VM_NO_GUARD))
638 		size += PAGE_SIZE;
639 
640 	vma->addr	= va_start;
641 	vma->phys_addr	= pa_start;
642 	vma->size	= size;
643 	vma->flags	= VM_MAP | vm_flags;
644 	vma->caller	= __builtin_return_address(0);
645 
646 	vm_area_add_early(vma);
647 }
648 
649 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
650 static int __init map_entry_trampoline(void)
651 {
652 	int i;
653 
654 	pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
655 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
656 
657 	/* The trampoline is always mapped and can therefore be global */
658 	pgprot_val(prot) &= ~PTE_NG;
659 
660 	/* Map only the text into the trampoline page table */
661 	memset(tramp_pg_dir, 0, PGD_SIZE);
662 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
663 			     entry_tramp_text_size(), prot,
664 			     __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
665 
666 	/* Map both the text and data into the kernel page table */
667 	for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
668 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
669 			     pa_start + i * PAGE_SIZE, prot);
670 
671 	if (IS_ENABLED(CONFIG_RELOCATABLE))
672 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
673 			     pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
674 
675 	return 0;
676 }
677 core_initcall(map_entry_trampoline);
678 #endif
679 
680 /*
681  * Open coded check for BTI, only for use to determine configuration
682  * for early mappings for before the cpufeature code has run.
683  */
684 static bool arm64_early_this_cpu_has_bti(void)
685 {
686 	u64 pfr1;
687 
688 	if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
689 		return false;
690 
691 	pfr1 = __read_sysreg_by_encoding(SYS_ID_AA64PFR1_EL1);
692 	return cpuid_feature_extract_unsigned_field(pfr1,
693 						    ID_AA64PFR1_BT_SHIFT);
694 }
695 
696 /*
697  * Create fine-grained mappings for the kernel.
698  */
699 static void __init map_kernel(pgd_t *pgdp)
700 {
701 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
702 				vmlinux_initdata, vmlinux_data;
703 
704 	/*
705 	 * External debuggers may need to write directly to the text
706 	 * mapping to install SW breakpoints. Allow this (only) when
707 	 * explicitly requested with rodata=off.
708 	 */
709 	pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
710 
711 	/*
712 	 * If we have a CPU that supports BTI and a kernel built for
713 	 * BTI then mark the kernel executable text as guarded pages
714 	 * now so we don't have to rewrite the page tables later.
715 	 */
716 	if (arm64_early_this_cpu_has_bti())
717 		text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
718 
719 	/*
720 	 * Only rodata will be remapped with different permissions later on,
721 	 * all other segments are allowed to use contiguous mappings.
722 	 */
723 	map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0,
724 			   VM_NO_GUARD);
725 	map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
726 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
727 	map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
728 			   &vmlinux_inittext, 0, VM_NO_GUARD);
729 	map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
730 			   &vmlinux_initdata, 0, VM_NO_GUARD);
731 	map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
732 
733 	if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_START)))) {
734 		/*
735 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
736 		 * live in the carveout for the swapper_pg_dir. We can simply
737 		 * re-use the existing dir for the fixmap.
738 		 */
739 		set_pgd(pgd_offset_pgd(pgdp, FIXADDR_START),
740 			READ_ONCE(*pgd_offset_k(FIXADDR_START)));
741 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
742 		pgd_t *bm_pgdp;
743 		p4d_t *bm_p4dp;
744 		pud_t *bm_pudp;
745 		/*
746 		 * The fixmap shares its top level pgd entry with the kernel
747 		 * mapping. This can really only occur when we are running
748 		 * with 16k/4 levels, so we can simply reuse the pud level
749 		 * entry instead.
750 		 */
751 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
752 		bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_START);
753 		bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_START);
754 		bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_START);
755 		pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
756 		pud_clear_fixmap();
757 	} else {
758 		BUG();
759 	}
760 
761 	kasan_copy_shadow(pgdp);
762 }
763 
764 static void __init create_idmap(void)
765 {
766 	u64 start = __pa_symbol(__idmap_text_start);
767 	u64 size = __pa_symbol(__idmap_text_end) - start;
768 	pgd_t *pgd = idmap_pg_dir;
769 	u64 pgd_phys;
770 
771 	/* check if we need an additional level of translation */
772 	if (VA_BITS < 48 && idmap_t0sz < (64 - VA_BITS_MIN)) {
773 		pgd_phys = early_pgtable_alloc(PAGE_SHIFT);
774 		set_pgd(&idmap_pg_dir[start >> VA_BITS],
775 			__pgd(pgd_phys | P4D_TYPE_TABLE));
776 		pgd = __va(pgd_phys);
777 	}
778 	__create_pgd_mapping(pgd, start, start, size, PAGE_KERNEL_ROX,
779 			     early_pgtable_alloc, 0);
780 
781 	if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0)) {
782 		extern u32 __idmap_kpti_flag;
783 		u64 pa = __pa_symbol(&__idmap_kpti_flag);
784 
785 		/*
786 		 * The KPTI G-to-nG conversion code needs a read-write mapping
787 		 * of its synchronization flag in the ID map.
788 		 */
789 		__create_pgd_mapping(pgd, pa, pa, sizeof(u32), PAGE_KERNEL,
790 				     early_pgtable_alloc, 0);
791 	}
792 }
793 
794 void __init paging_init(void)
795 {
796 	pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
797 	extern pgd_t init_idmap_pg_dir[];
798 
799 	idmap_t0sz = 63UL - __fls(__pa_symbol(_end) | GENMASK(VA_BITS_MIN - 1, 0));
800 
801 	map_kernel(pgdp);
802 	map_mem(pgdp);
803 
804 	pgd_clear_fixmap();
805 
806 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir), init_idmap_pg_dir);
807 	init_mm.pgd = swapper_pg_dir;
808 
809 	memblock_phys_free(__pa_symbol(init_pg_dir),
810 			   __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
811 
812 	memblock_allow_resize();
813 
814 	create_idmap();
815 }
816 
817 /*
818  * Check whether a kernel address is valid (derived from arch/x86/).
819  */
820 int kern_addr_valid(unsigned long addr)
821 {
822 	pgd_t *pgdp;
823 	p4d_t *p4dp;
824 	pud_t *pudp, pud;
825 	pmd_t *pmdp, pmd;
826 	pte_t *ptep, pte;
827 
828 	addr = arch_kasan_reset_tag(addr);
829 	if ((((long)addr) >> VA_BITS) != -1UL)
830 		return 0;
831 
832 	pgdp = pgd_offset_k(addr);
833 	if (pgd_none(READ_ONCE(*pgdp)))
834 		return 0;
835 
836 	p4dp = p4d_offset(pgdp, addr);
837 	if (p4d_none(READ_ONCE(*p4dp)))
838 		return 0;
839 
840 	pudp = pud_offset(p4dp, addr);
841 	pud = READ_ONCE(*pudp);
842 	if (pud_none(pud))
843 		return 0;
844 
845 	if (pud_sect(pud))
846 		return pfn_valid(pud_pfn(pud));
847 
848 	pmdp = pmd_offset(pudp, addr);
849 	pmd = READ_ONCE(*pmdp);
850 	if (pmd_none(pmd))
851 		return 0;
852 
853 	if (pmd_sect(pmd))
854 		return pfn_valid(pmd_pfn(pmd));
855 
856 	ptep = pte_offset_kernel(pmdp, addr);
857 	pte = READ_ONCE(*ptep);
858 	if (pte_none(pte))
859 		return 0;
860 
861 	return pfn_valid(pte_pfn(pte));
862 }
863 
864 #ifdef CONFIG_MEMORY_HOTPLUG
865 static void free_hotplug_page_range(struct page *page, size_t size,
866 				    struct vmem_altmap *altmap)
867 {
868 	if (altmap) {
869 		vmem_altmap_free(altmap, size >> PAGE_SHIFT);
870 	} else {
871 		WARN_ON(PageReserved(page));
872 		free_pages((unsigned long)page_address(page), get_order(size));
873 	}
874 }
875 
876 static void free_hotplug_pgtable_page(struct page *page)
877 {
878 	free_hotplug_page_range(page, PAGE_SIZE, NULL);
879 }
880 
881 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
882 				  unsigned long floor, unsigned long ceiling,
883 				  unsigned long mask)
884 {
885 	start &= mask;
886 	if (start < floor)
887 		return false;
888 
889 	if (ceiling) {
890 		ceiling &= mask;
891 		if (!ceiling)
892 			return false;
893 	}
894 
895 	if (end - 1 > ceiling - 1)
896 		return false;
897 	return true;
898 }
899 
900 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
901 				    unsigned long end, bool free_mapped,
902 				    struct vmem_altmap *altmap)
903 {
904 	pte_t *ptep, pte;
905 
906 	do {
907 		ptep = pte_offset_kernel(pmdp, addr);
908 		pte = READ_ONCE(*ptep);
909 		if (pte_none(pte))
910 			continue;
911 
912 		WARN_ON(!pte_present(pte));
913 		pte_clear(&init_mm, addr, ptep);
914 		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
915 		if (free_mapped)
916 			free_hotplug_page_range(pte_page(pte),
917 						PAGE_SIZE, altmap);
918 	} while (addr += PAGE_SIZE, addr < end);
919 }
920 
921 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
922 				    unsigned long end, bool free_mapped,
923 				    struct vmem_altmap *altmap)
924 {
925 	unsigned long next;
926 	pmd_t *pmdp, pmd;
927 
928 	do {
929 		next = pmd_addr_end(addr, end);
930 		pmdp = pmd_offset(pudp, addr);
931 		pmd = READ_ONCE(*pmdp);
932 		if (pmd_none(pmd))
933 			continue;
934 
935 		WARN_ON(!pmd_present(pmd));
936 		if (pmd_sect(pmd)) {
937 			pmd_clear(pmdp);
938 
939 			/*
940 			 * One TLBI should be sufficient here as the PMD_SIZE
941 			 * range is mapped with a single block entry.
942 			 */
943 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
944 			if (free_mapped)
945 				free_hotplug_page_range(pmd_page(pmd),
946 							PMD_SIZE, altmap);
947 			continue;
948 		}
949 		WARN_ON(!pmd_table(pmd));
950 		unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
951 	} while (addr = next, addr < end);
952 }
953 
954 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
955 				    unsigned long end, bool free_mapped,
956 				    struct vmem_altmap *altmap)
957 {
958 	unsigned long next;
959 	pud_t *pudp, pud;
960 
961 	do {
962 		next = pud_addr_end(addr, end);
963 		pudp = pud_offset(p4dp, addr);
964 		pud = READ_ONCE(*pudp);
965 		if (pud_none(pud))
966 			continue;
967 
968 		WARN_ON(!pud_present(pud));
969 		if (pud_sect(pud)) {
970 			pud_clear(pudp);
971 
972 			/*
973 			 * One TLBI should be sufficient here as the PUD_SIZE
974 			 * range is mapped with a single block entry.
975 			 */
976 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
977 			if (free_mapped)
978 				free_hotplug_page_range(pud_page(pud),
979 							PUD_SIZE, altmap);
980 			continue;
981 		}
982 		WARN_ON(!pud_table(pud));
983 		unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
984 	} while (addr = next, addr < end);
985 }
986 
987 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
988 				    unsigned long end, bool free_mapped,
989 				    struct vmem_altmap *altmap)
990 {
991 	unsigned long next;
992 	p4d_t *p4dp, p4d;
993 
994 	do {
995 		next = p4d_addr_end(addr, end);
996 		p4dp = p4d_offset(pgdp, addr);
997 		p4d = READ_ONCE(*p4dp);
998 		if (p4d_none(p4d))
999 			continue;
1000 
1001 		WARN_ON(!p4d_present(p4d));
1002 		unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
1003 	} while (addr = next, addr < end);
1004 }
1005 
1006 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
1007 				bool free_mapped, struct vmem_altmap *altmap)
1008 {
1009 	unsigned long next;
1010 	pgd_t *pgdp, pgd;
1011 
1012 	/*
1013 	 * altmap can only be used as vmemmap mapping backing memory.
1014 	 * In case the backing memory itself is not being freed, then
1015 	 * altmap is irrelevant. Warn about this inconsistency when
1016 	 * encountered.
1017 	 */
1018 	WARN_ON(!free_mapped && altmap);
1019 
1020 	do {
1021 		next = pgd_addr_end(addr, end);
1022 		pgdp = pgd_offset_k(addr);
1023 		pgd = READ_ONCE(*pgdp);
1024 		if (pgd_none(pgd))
1025 			continue;
1026 
1027 		WARN_ON(!pgd_present(pgd));
1028 		unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
1029 	} while (addr = next, addr < end);
1030 }
1031 
1032 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
1033 				 unsigned long end, unsigned long floor,
1034 				 unsigned long ceiling)
1035 {
1036 	pte_t *ptep, pte;
1037 	unsigned long i, start = addr;
1038 
1039 	do {
1040 		ptep = pte_offset_kernel(pmdp, addr);
1041 		pte = READ_ONCE(*ptep);
1042 
1043 		/*
1044 		 * This is just a sanity check here which verifies that
1045 		 * pte clearing has been done by earlier unmap loops.
1046 		 */
1047 		WARN_ON(!pte_none(pte));
1048 	} while (addr += PAGE_SIZE, addr < end);
1049 
1050 	if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
1051 		return;
1052 
1053 	/*
1054 	 * Check whether we can free the pte page if the rest of the
1055 	 * entries are empty. Overlap with other regions have been
1056 	 * handled by the floor/ceiling check.
1057 	 */
1058 	ptep = pte_offset_kernel(pmdp, 0UL);
1059 	for (i = 0; i < PTRS_PER_PTE; i++) {
1060 		if (!pte_none(READ_ONCE(ptep[i])))
1061 			return;
1062 	}
1063 
1064 	pmd_clear(pmdp);
1065 	__flush_tlb_kernel_pgtable(start);
1066 	free_hotplug_pgtable_page(virt_to_page(ptep));
1067 }
1068 
1069 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1070 				 unsigned long end, unsigned long floor,
1071 				 unsigned long ceiling)
1072 {
1073 	pmd_t *pmdp, pmd;
1074 	unsigned long i, next, start = addr;
1075 
1076 	do {
1077 		next = pmd_addr_end(addr, end);
1078 		pmdp = pmd_offset(pudp, addr);
1079 		pmd = READ_ONCE(*pmdp);
1080 		if (pmd_none(pmd))
1081 			continue;
1082 
1083 		WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1084 		free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1085 	} while (addr = next, addr < end);
1086 
1087 	if (CONFIG_PGTABLE_LEVELS <= 2)
1088 		return;
1089 
1090 	if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1091 		return;
1092 
1093 	/*
1094 	 * Check whether we can free the pmd page if the rest of the
1095 	 * entries are empty. Overlap with other regions have been
1096 	 * handled by the floor/ceiling check.
1097 	 */
1098 	pmdp = pmd_offset(pudp, 0UL);
1099 	for (i = 0; i < PTRS_PER_PMD; i++) {
1100 		if (!pmd_none(READ_ONCE(pmdp[i])))
1101 			return;
1102 	}
1103 
1104 	pud_clear(pudp);
1105 	__flush_tlb_kernel_pgtable(start);
1106 	free_hotplug_pgtable_page(virt_to_page(pmdp));
1107 }
1108 
1109 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1110 				 unsigned long end, unsigned long floor,
1111 				 unsigned long ceiling)
1112 {
1113 	pud_t *pudp, pud;
1114 	unsigned long i, next, start = addr;
1115 
1116 	do {
1117 		next = pud_addr_end(addr, end);
1118 		pudp = pud_offset(p4dp, addr);
1119 		pud = READ_ONCE(*pudp);
1120 		if (pud_none(pud))
1121 			continue;
1122 
1123 		WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1124 		free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1125 	} while (addr = next, addr < end);
1126 
1127 	if (CONFIG_PGTABLE_LEVELS <= 3)
1128 		return;
1129 
1130 	if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1131 		return;
1132 
1133 	/*
1134 	 * Check whether we can free the pud page if the rest of the
1135 	 * entries are empty. Overlap with other regions have been
1136 	 * handled by the floor/ceiling check.
1137 	 */
1138 	pudp = pud_offset(p4dp, 0UL);
1139 	for (i = 0; i < PTRS_PER_PUD; i++) {
1140 		if (!pud_none(READ_ONCE(pudp[i])))
1141 			return;
1142 	}
1143 
1144 	p4d_clear(p4dp);
1145 	__flush_tlb_kernel_pgtable(start);
1146 	free_hotplug_pgtable_page(virt_to_page(pudp));
1147 }
1148 
1149 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1150 				 unsigned long end, unsigned long floor,
1151 				 unsigned long ceiling)
1152 {
1153 	unsigned long next;
1154 	p4d_t *p4dp, p4d;
1155 
1156 	do {
1157 		next = p4d_addr_end(addr, end);
1158 		p4dp = p4d_offset(pgdp, addr);
1159 		p4d = READ_ONCE(*p4dp);
1160 		if (p4d_none(p4d))
1161 			continue;
1162 
1163 		WARN_ON(!p4d_present(p4d));
1164 		free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1165 	} while (addr = next, addr < end);
1166 }
1167 
1168 static void free_empty_tables(unsigned long addr, unsigned long end,
1169 			      unsigned long floor, unsigned long ceiling)
1170 {
1171 	unsigned long next;
1172 	pgd_t *pgdp, pgd;
1173 
1174 	do {
1175 		next = pgd_addr_end(addr, end);
1176 		pgdp = pgd_offset_k(addr);
1177 		pgd = READ_ONCE(*pgdp);
1178 		if (pgd_none(pgd))
1179 			continue;
1180 
1181 		WARN_ON(!pgd_present(pgd));
1182 		free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1183 	} while (addr = next, addr < end);
1184 }
1185 #endif
1186 
1187 #if !ARM64_KERNEL_USES_PMD_MAPS
1188 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1189 		struct vmem_altmap *altmap)
1190 {
1191 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1192 	return vmemmap_populate_basepages(start, end, node, altmap);
1193 }
1194 #else	/* !ARM64_KERNEL_USES_PMD_MAPS */
1195 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1196 		struct vmem_altmap *altmap)
1197 {
1198 	unsigned long addr = start;
1199 	unsigned long next;
1200 	pgd_t *pgdp;
1201 	p4d_t *p4dp;
1202 	pud_t *pudp;
1203 	pmd_t *pmdp;
1204 
1205 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1206 	do {
1207 		next = pmd_addr_end(addr, end);
1208 
1209 		pgdp = vmemmap_pgd_populate(addr, node);
1210 		if (!pgdp)
1211 			return -ENOMEM;
1212 
1213 		p4dp = vmemmap_p4d_populate(pgdp, addr, node);
1214 		if (!p4dp)
1215 			return -ENOMEM;
1216 
1217 		pudp = vmemmap_pud_populate(p4dp, addr, node);
1218 		if (!pudp)
1219 			return -ENOMEM;
1220 
1221 		pmdp = pmd_offset(pudp, addr);
1222 		if (pmd_none(READ_ONCE(*pmdp))) {
1223 			void *p = NULL;
1224 
1225 			p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1226 			if (!p) {
1227 				if (vmemmap_populate_basepages(addr, next, node, altmap))
1228 					return -ENOMEM;
1229 				continue;
1230 			}
1231 
1232 			pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1233 		} else
1234 			vmemmap_verify((pte_t *)pmdp, node, addr, next);
1235 	} while (addr = next, addr != end);
1236 
1237 	return 0;
1238 }
1239 #endif	/* !ARM64_KERNEL_USES_PMD_MAPS */
1240 
1241 #ifdef CONFIG_MEMORY_HOTPLUG
1242 void vmemmap_free(unsigned long start, unsigned long end,
1243 		struct vmem_altmap *altmap)
1244 {
1245 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1246 
1247 	unmap_hotplug_range(start, end, true, altmap);
1248 	free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1249 }
1250 #endif /* CONFIG_MEMORY_HOTPLUG */
1251 
1252 static inline pud_t *fixmap_pud(unsigned long addr)
1253 {
1254 	pgd_t *pgdp = pgd_offset_k(addr);
1255 	p4d_t *p4dp = p4d_offset(pgdp, addr);
1256 	p4d_t p4d = READ_ONCE(*p4dp);
1257 
1258 	BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
1259 
1260 	return pud_offset_kimg(p4dp, addr);
1261 }
1262 
1263 static inline pmd_t *fixmap_pmd(unsigned long addr)
1264 {
1265 	pud_t *pudp = fixmap_pud(addr);
1266 	pud_t pud = READ_ONCE(*pudp);
1267 
1268 	BUG_ON(pud_none(pud) || pud_bad(pud));
1269 
1270 	return pmd_offset_kimg(pudp, addr);
1271 }
1272 
1273 static inline pte_t *fixmap_pte(unsigned long addr)
1274 {
1275 	return &bm_pte[pte_index(addr)];
1276 }
1277 
1278 /*
1279  * The p*d_populate functions call virt_to_phys implicitly so they can't be used
1280  * directly on kernel symbols (bm_p*d). This function is called too early to use
1281  * lm_alias so __p*d_populate functions must be used to populate with the
1282  * physical address from __pa_symbol.
1283  */
1284 void __init early_fixmap_init(void)
1285 {
1286 	pgd_t *pgdp;
1287 	p4d_t *p4dp, p4d;
1288 	pud_t *pudp;
1289 	pmd_t *pmdp;
1290 	unsigned long addr = FIXADDR_START;
1291 
1292 	pgdp = pgd_offset_k(addr);
1293 	p4dp = p4d_offset(pgdp, addr);
1294 	p4d = READ_ONCE(*p4dp);
1295 	if (CONFIG_PGTABLE_LEVELS > 3 &&
1296 	    !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
1297 		/*
1298 		 * We only end up here if the kernel mapping and the fixmap
1299 		 * share the top level pgd entry, which should only happen on
1300 		 * 16k/4 levels configurations.
1301 		 */
1302 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
1303 		pudp = pud_offset_kimg(p4dp, addr);
1304 	} else {
1305 		if (p4d_none(p4d))
1306 			__p4d_populate(p4dp, __pa_symbol(bm_pud), P4D_TYPE_TABLE);
1307 		pudp = fixmap_pud(addr);
1308 	}
1309 	if (pud_none(READ_ONCE(*pudp)))
1310 		__pud_populate(pudp, __pa_symbol(bm_pmd), PUD_TYPE_TABLE);
1311 	pmdp = fixmap_pmd(addr);
1312 	__pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
1313 
1314 	/*
1315 	 * The boot-ioremap range spans multiple pmds, for which
1316 	 * we are not prepared:
1317 	 */
1318 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1319 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1320 
1321 	if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
1322 	     || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
1323 		WARN_ON(1);
1324 		pr_warn("pmdp %p != %p, %p\n",
1325 			pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
1326 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
1327 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1328 			fix_to_virt(FIX_BTMAP_BEGIN));
1329 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1330 			fix_to_virt(FIX_BTMAP_END));
1331 
1332 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1333 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1334 	}
1335 }
1336 
1337 /*
1338  * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
1339  * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
1340  */
1341 void __set_fixmap(enum fixed_addresses idx,
1342 			       phys_addr_t phys, pgprot_t flags)
1343 {
1344 	unsigned long addr = __fix_to_virt(idx);
1345 	pte_t *ptep;
1346 
1347 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
1348 
1349 	ptep = fixmap_pte(addr);
1350 
1351 	if (pgprot_val(flags)) {
1352 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
1353 	} else {
1354 		pte_clear(&init_mm, addr, ptep);
1355 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
1356 	}
1357 }
1358 
1359 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
1360 {
1361 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
1362 	int offset;
1363 	void *dt_virt;
1364 
1365 	/*
1366 	 * Check whether the physical FDT address is set and meets the minimum
1367 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
1368 	 * at least 8 bytes so that we can always access the magic and size
1369 	 * fields of the FDT header after mapping the first chunk, double check
1370 	 * here if that is indeed the case.
1371 	 */
1372 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
1373 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
1374 		return NULL;
1375 
1376 	/*
1377 	 * Make sure that the FDT region can be mapped without the need to
1378 	 * allocate additional translation table pages, so that it is safe
1379 	 * to call create_mapping_noalloc() this early.
1380 	 *
1381 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
1382 	 * be in the same PMD as the rest of the fixmap.
1383 	 * On 4k pages, we'll use section mappings for the FDT so we only
1384 	 * have to be in the same PUD.
1385 	 */
1386 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
1387 
1388 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
1389 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
1390 
1391 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
1392 	dt_virt = (void *)dt_virt_base + offset;
1393 
1394 	/* map the first chunk so we can read the size from the header */
1395 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
1396 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
1397 
1398 	if (fdt_magic(dt_virt) != FDT_MAGIC)
1399 		return NULL;
1400 
1401 	*size = fdt_totalsize(dt_virt);
1402 	if (*size > MAX_FDT_SIZE)
1403 		return NULL;
1404 
1405 	if (offset + *size > SWAPPER_BLOCK_SIZE)
1406 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
1407 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
1408 
1409 	return dt_virt;
1410 }
1411 
1412 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1413 {
1414 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1415 
1416 	/* Only allow permission changes for now */
1417 	if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1418 				   pud_val(new_pud)))
1419 		return 0;
1420 
1421 	VM_BUG_ON(phys & ~PUD_MASK);
1422 	set_pud(pudp, new_pud);
1423 	return 1;
1424 }
1425 
1426 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1427 {
1428 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1429 
1430 	/* Only allow permission changes for now */
1431 	if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1432 				   pmd_val(new_pmd)))
1433 		return 0;
1434 
1435 	VM_BUG_ON(phys & ~PMD_MASK);
1436 	set_pmd(pmdp, new_pmd);
1437 	return 1;
1438 }
1439 
1440 int pud_clear_huge(pud_t *pudp)
1441 {
1442 	if (!pud_sect(READ_ONCE(*pudp)))
1443 		return 0;
1444 	pud_clear(pudp);
1445 	return 1;
1446 }
1447 
1448 int pmd_clear_huge(pmd_t *pmdp)
1449 {
1450 	if (!pmd_sect(READ_ONCE(*pmdp)))
1451 		return 0;
1452 	pmd_clear(pmdp);
1453 	return 1;
1454 }
1455 
1456 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1457 {
1458 	pte_t *table;
1459 	pmd_t pmd;
1460 
1461 	pmd = READ_ONCE(*pmdp);
1462 
1463 	if (!pmd_table(pmd)) {
1464 		VM_WARN_ON(1);
1465 		return 1;
1466 	}
1467 
1468 	table = pte_offset_kernel(pmdp, addr);
1469 	pmd_clear(pmdp);
1470 	__flush_tlb_kernel_pgtable(addr);
1471 	pte_free_kernel(NULL, table);
1472 	return 1;
1473 }
1474 
1475 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1476 {
1477 	pmd_t *table;
1478 	pmd_t *pmdp;
1479 	pud_t pud;
1480 	unsigned long next, end;
1481 
1482 	pud = READ_ONCE(*pudp);
1483 
1484 	if (!pud_table(pud)) {
1485 		VM_WARN_ON(1);
1486 		return 1;
1487 	}
1488 
1489 	table = pmd_offset(pudp, addr);
1490 	pmdp = table;
1491 	next = addr;
1492 	end = addr + PUD_SIZE;
1493 	do {
1494 		pmd_free_pte_page(pmdp, next);
1495 	} while (pmdp++, next += PMD_SIZE, next != end);
1496 
1497 	pud_clear(pudp);
1498 	__flush_tlb_kernel_pgtable(addr);
1499 	pmd_free(NULL, table);
1500 	return 1;
1501 }
1502 
1503 #ifdef CONFIG_MEMORY_HOTPLUG
1504 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1505 {
1506 	unsigned long end = start + size;
1507 
1508 	WARN_ON(pgdir != init_mm.pgd);
1509 	WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1510 
1511 	unmap_hotplug_range(start, end, false, NULL);
1512 	free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1513 }
1514 
1515 struct range arch_get_mappable_range(void)
1516 {
1517 	struct range mhp_range;
1518 	u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1519 	u64 end_linear_pa = __pa(PAGE_END - 1);
1520 
1521 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1522 		/*
1523 		 * Check for a wrap, it is possible because of randomized linear
1524 		 * mapping the start physical address is actually bigger than
1525 		 * the end physical address. In this case set start to zero
1526 		 * because [0, end_linear_pa] range must still be able to cover
1527 		 * all addressable physical addresses.
1528 		 */
1529 		if (start_linear_pa > end_linear_pa)
1530 			start_linear_pa = 0;
1531 	}
1532 
1533 	WARN_ON(start_linear_pa > end_linear_pa);
1534 
1535 	/*
1536 	 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1537 	 * accommodating both its ends but excluding PAGE_END. Max physical
1538 	 * range which can be mapped inside this linear mapping range, must
1539 	 * also be derived from its end points.
1540 	 */
1541 	mhp_range.start = start_linear_pa;
1542 	mhp_range.end =  end_linear_pa;
1543 
1544 	return mhp_range;
1545 }
1546 
1547 int arch_add_memory(int nid, u64 start, u64 size,
1548 		    struct mhp_params *params)
1549 {
1550 	int ret, flags = NO_EXEC_MAPPINGS;
1551 
1552 	VM_BUG_ON(!mhp_range_allowed(start, size, true));
1553 
1554 	/*
1555 	 * KFENCE requires linear map to be mapped at page granularity, so that
1556 	 * it is possible to protect/unprotect single pages in the KFENCE pool.
1557 	 */
1558 	if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
1559 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1560 
1561 	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1562 			     size, params->pgprot, __pgd_pgtable_alloc,
1563 			     flags);
1564 
1565 	memblock_clear_nomap(start, size);
1566 
1567 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1568 			   params);
1569 	if (ret)
1570 		__remove_pgd_mapping(swapper_pg_dir,
1571 				     __phys_to_virt(start), size);
1572 	else {
1573 		max_pfn = PFN_UP(start + size);
1574 		max_low_pfn = max_pfn;
1575 	}
1576 
1577 	return ret;
1578 }
1579 
1580 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1581 {
1582 	unsigned long start_pfn = start >> PAGE_SHIFT;
1583 	unsigned long nr_pages = size >> PAGE_SHIFT;
1584 
1585 	__remove_pages(start_pfn, nr_pages, altmap);
1586 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1587 }
1588 
1589 /*
1590  * This memory hotplug notifier helps prevent boot memory from being
1591  * inadvertently removed as it blocks pfn range offlining process in
1592  * __offline_pages(). Hence this prevents both offlining as well as
1593  * removal process for boot memory which is initially always online.
1594  * In future if and when boot memory could be removed, this notifier
1595  * should be dropped and free_hotplug_page_range() should handle any
1596  * reserved pages allocated during boot.
1597  */
1598 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1599 					   unsigned long action, void *data)
1600 {
1601 	struct mem_section *ms;
1602 	struct memory_notify *arg = data;
1603 	unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1604 	unsigned long pfn = arg->start_pfn;
1605 
1606 	if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1607 		return NOTIFY_OK;
1608 
1609 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1610 		unsigned long start = PFN_PHYS(pfn);
1611 		unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1612 
1613 		ms = __pfn_to_section(pfn);
1614 		if (!early_section(ms))
1615 			continue;
1616 
1617 		if (action == MEM_GOING_OFFLINE) {
1618 			/*
1619 			 * Boot memory removal is not supported. Prevent
1620 			 * it via blocking any attempted offline request
1621 			 * for the boot memory and just report it.
1622 			 */
1623 			pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1624 			return NOTIFY_BAD;
1625 		} else if (action == MEM_OFFLINE) {
1626 			/*
1627 			 * This should have never happened. Boot memory
1628 			 * offlining should have been prevented by this
1629 			 * very notifier. Probably some memory removal
1630 			 * procedure might have changed which would then
1631 			 * require further debug.
1632 			 */
1633 			pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1634 
1635 			/*
1636 			 * Core memory hotplug does not process a return
1637 			 * code from the notifier for MEM_OFFLINE events.
1638 			 * The error condition has been reported. Return
1639 			 * from here as if ignored.
1640 			 */
1641 			return NOTIFY_DONE;
1642 		}
1643 	}
1644 	return NOTIFY_OK;
1645 }
1646 
1647 static struct notifier_block prevent_bootmem_remove_nb = {
1648 	.notifier_call = prevent_bootmem_remove_notifier,
1649 };
1650 
1651 /*
1652  * This ensures that boot memory sections on the platform are online
1653  * from early boot. Memory sections could not be prevented from being
1654  * offlined, unless for some reason they are not online to begin with.
1655  * This helps validate the basic assumption on which the above memory
1656  * event notifier works to prevent boot memory section offlining and
1657  * its possible removal.
1658  */
1659 static void validate_bootmem_online(void)
1660 {
1661 	phys_addr_t start, end, addr;
1662 	struct mem_section *ms;
1663 	u64 i;
1664 
1665 	/*
1666 	 * Scanning across all memblock might be expensive
1667 	 * on some big memory systems. Hence enable this
1668 	 * validation only with DEBUG_VM.
1669 	 */
1670 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
1671 		return;
1672 
1673 	for_each_mem_range(i, &start, &end) {
1674 		for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1675 			ms = __pfn_to_section(PHYS_PFN(addr));
1676 
1677 			/*
1678 			 * All memory ranges in the system at this point
1679 			 * should have been marked as early sections.
1680 			 */
1681 			WARN_ON(!early_section(ms));
1682 
1683 			/*
1684 			 * Memory notifier mechanism here to prevent boot
1685 			 * memory offlining depends on the fact that each
1686 			 * early section memory on the system is initially
1687 			 * online. Otherwise a given memory section which
1688 			 * is already offline will be overlooked and can
1689 			 * be removed completely. Call out such sections.
1690 			 */
1691 			if (!online_section(ms))
1692 				pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1693 					addr, addr + (1UL << PA_SECTION_SHIFT));
1694 		}
1695 	}
1696 }
1697 
1698 static int __init prevent_bootmem_remove_init(void)
1699 {
1700 	int ret = 0;
1701 
1702 	if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1703 		return ret;
1704 
1705 	validate_bootmem_online();
1706 	ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1707 	if (ret)
1708 		pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1709 
1710 	return ret;
1711 }
1712 early_initcall(prevent_bootmem_remove_init);
1713 #endif
1714