xref: /openbmc/linux/arch/arm64/mm/mmu.c (revision 46290c6b)
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 #include <linux/kfence.h>
28 
29 #include <asm/barrier.h>
30 #include <asm/cputype.h>
31 #include <asm/fixmap.h>
32 #include <asm/kasan.h>
33 #include <asm/kernel-pgtable.h>
34 #include <asm/sections.h>
35 #include <asm/setup.h>
36 #include <linux/sizes.h>
37 #include <asm/tlb.h>
38 #include <asm/mmu_context.h>
39 #include <asm/ptdump.h>
40 #include <asm/tlbflush.h>
41 #include <asm/pgalloc.h>
42 #include <asm/kfence.h>
43 
44 #define NO_BLOCK_MAPPINGS	BIT(0)
45 #define NO_CONT_MAPPINGS	BIT(1)
46 #define NO_EXEC_MAPPINGS	BIT(2)	/* assumes FEAT_HPDS is not used */
47 
48 int idmap_t0sz __ro_after_init;
49 
50 #if VA_BITS > 48
51 u64 vabits_actual __ro_after_init = VA_BITS_MIN;
52 EXPORT_SYMBOL(vabits_actual);
53 #endif
54 
55 u64 kimage_vaddr __ro_after_init = (u64)&_text;
56 EXPORT_SYMBOL(kimage_vaddr);
57 
58 u64 kimage_voffset __ro_after_init;
59 EXPORT_SYMBOL(kimage_voffset);
60 
61 u32 __boot_cpu_mode[] = { BOOT_CPU_MODE_EL2, BOOT_CPU_MODE_EL1 };
62 
63 /*
64  * The booting CPU updates the failed status @__early_cpu_boot_status,
65  * with MMU turned off.
66  */
67 long __section(".mmuoff.data.write") __early_cpu_boot_status;
68 
69 /*
70  * Empty_zero_page is a special page that is used for zero-initialized data
71  * and COW.
72  */
73 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
74 EXPORT_SYMBOL(empty_zero_page);
75 
76 static DEFINE_SPINLOCK(swapper_pgdir_lock);
77 static DEFINE_MUTEX(fixmap_lock);
78 
79 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
80 {
81 	pgd_t *fixmap_pgdp;
82 
83 	spin_lock(&swapper_pgdir_lock);
84 	fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
85 	WRITE_ONCE(*fixmap_pgdp, pgd);
86 	/*
87 	 * We need dsb(ishst) here to ensure the page-table-walker sees
88 	 * our new entry before set_p?d() returns. The fixmap's
89 	 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
90 	 */
91 	pgd_clear_fixmap();
92 	spin_unlock(&swapper_pgdir_lock);
93 }
94 
95 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
96 			      unsigned long size, pgprot_t vma_prot)
97 {
98 	if (!pfn_is_map_memory(pfn))
99 		return pgprot_noncached(vma_prot);
100 	else if (file->f_flags & O_SYNC)
101 		return pgprot_writecombine(vma_prot);
102 	return vma_prot;
103 }
104 EXPORT_SYMBOL(phys_mem_access_prot);
105 
106 static phys_addr_t __init early_pgtable_alloc(int shift)
107 {
108 	phys_addr_t phys;
109 	void *ptr;
110 
111 	phys = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0,
112 					 MEMBLOCK_ALLOC_NOLEAKTRACE);
113 	if (!phys)
114 		panic("Failed to allocate page table page\n");
115 
116 	/*
117 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
118 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
119 	 * any level of table.
120 	 */
121 	ptr = pte_set_fixmap(phys);
122 
123 	memset(ptr, 0, PAGE_SIZE);
124 
125 	/*
126 	 * Implicit barriers also ensure the zeroed page is visible to the page
127 	 * table walker
128 	 */
129 	pte_clear_fixmap();
130 
131 	return phys;
132 }
133 
134 bool pgattr_change_is_safe(u64 old, u64 new)
135 {
136 	/*
137 	 * The following mapping attributes may be updated in live
138 	 * kernel mappings without the need for break-before-make.
139 	 */
140 	pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
141 
142 	/* creating or taking down mappings is always safe */
143 	if (!pte_valid(__pte(old)) || !pte_valid(__pte(new)))
144 		return true;
145 
146 	/* A live entry's pfn should not change */
147 	if (pte_pfn(__pte(old)) != pte_pfn(__pte(new)))
148 		return false;
149 
150 	/* live contiguous mappings may not be manipulated at all */
151 	if ((old | new) & PTE_CONT)
152 		return false;
153 
154 	/* Transitioning from Non-Global to Global is unsafe */
155 	if (old & ~new & PTE_NG)
156 		return false;
157 
158 	/*
159 	 * Changing the memory type between Normal and Normal-Tagged is safe
160 	 * since Tagged is considered a permission attribute from the
161 	 * mismatched attribute aliases perspective.
162 	 */
163 	if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
164 	     (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
165 	    ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
166 	     (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
167 		mask |= PTE_ATTRINDX_MASK;
168 
169 	return ((old ^ new) & ~mask) == 0;
170 }
171 
172 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
173 		     phys_addr_t phys, pgprot_t prot)
174 {
175 	pte_t *ptep;
176 
177 	ptep = pte_set_fixmap_offset(pmdp, addr);
178 	do {
179 		pte_t old_pte = READ_ONCE(*ptep);
180 
181 		set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
182 
183 		/*
184 		 * After the PTE entry has been populated once, we
185 		 * only allow updates to the permission attributes.
186 		 */
187 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
188 					      READ_ONCE(pte_val(*ptep))));
189 
190 		phys += PAGE_SIZE;
191 	} while (ptep++, addr += PAGE_SIZE, addr != end);
192 
193 	pte_clear_fixmap();
194 }
195 
196 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
197 				unsigned long end, phys_addr_t phys,
198 				pgprot_t prot,
199 				phys_addr_t (*pgtable_alloc)(int),
200 				int flags)
201 {
202 	unsigned long next;
203 	pmd_t pmd = READ_ONCE(*pmdp);
204 
205 	BUG_ON(pmd_sect(pmd));
206 	if (pmd_none(pmd)) {
207 		pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
208 		phys_addr_t pte_phys;
209 
210 		if (flags & NO_EXEC_MAPPINGS)
211 			pmdval |= PMD_TABLE_PXN;
212 		BUG_ON(!pgtable_alloc);
213 		pte_phys = pgtable_alloc(PAGE_SHIFT);
214 		__pmd_populate(pmdp, pte_phys, pmdval);
215 		pmd = READ_ONCE(*pmdp);
216 	}
217 	BUG_ON(pmd_bad(pmd));
218 
219 	do {
220 		pgprot_t __prot = prot;
221 
222 		next = pte_cont_addr_end(addr, end);
223 
224 		/* use a contiguous mapping if the range is suitably aligned */
225 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
226 		    (flags & NO_CONT_MAPPINGS) == 0)
227 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
228 
229 		init_pte(pmdp, addr, next, phys, __prot);
230 
231 		phys += next - addr;
232 	} while (addr = next, addr != end);
233 }
234 
235 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
236 		     phys_addr_t phys, pgprot_t prot,
237 		     phys_addr_t (*pgtable_alloc)(int), int flags)
238 {
239 	unsigned long next;
240 	pmd_t *pmdp;
241 
242 	pmdp = pmd_set_fixmap_offset(pudp, addr);
243 	do {
244 		pmd_t old_pmd = READ_ONCE(*pmdp);
245 
246 		next = pmd_addr_end(addr, end);
247 
248 		/* try section mapping first */
249 		if (((addr | next | phys) & ~PMD_MASK) == 0 &&
250 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
251 			pmd_set_huge(pmdp, phys, prot);
252 
253 			/*
254 			 * After the PMD entry has been populated once, we
255 			 * only allow updates to the permission attributes.
256 			 */
257 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
258 						      READ_ONCE(pmd_val(*pmdp))));
259 		} else {
260 			alloc_init_cont_pte(pmdp, addr, next, phys, prot,
261 					    pgtable_alloc, flags);
262 
263 			BUG_ON(pmd_val(old_pmd) != 0 &&
264 			       pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
265 		}
266 		phys += next - addr;
267 	} while (pmdp++, addr = next, addr != end);
268 
269 	pmd_clear_fixmap();
270 }
271 
272 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
273 				unsigned long end, phys_addr_t phys,
274 				pgprot_t prot,
275 				phys_addr_t (*pgtable_alloc)(int), int flags)
276 {
277 	unsigned long next;
278 	pud_t pud = READ_ONCE(*pudp);
279 
280 	/*
281 	 * Check for initial section mappings in the pgd/pud.
282 	 */
283 	BUG_ON(pud_sect(pud));
284 	if (pud_none(pud)) {
285 		pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
286 		phys_addr_t pmd_phys;
287 
288 		if (flags & NO_EXEC_MAPPINGS)
289 			pudval |= PUD_TABLE_PXN;
290 		BUG_ON(!pgtable_alloc);
291 		pmd_phys = pgtable_alloc(PMD_SHIFT);
292 		__pud_populate(pudp, pmd_phys, pudval);
293 		pud = READ_ONCE(*pudp);
294 	}
295 	BUG_ON(pud_bad(pud));
296 
297 	do {
298 		pgprot_t __prot = prot;
299 
300 		next = pmd_cont_addr_end(addr, end);
301 
302 		/* use a contiguous mapping if the range is suitably aligned */
303 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
304 		    (flags & NO_CONT_MAPPINGS) == 0)
305 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
306 
307 		init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
308 
309 		phys += next - addr;
310 	} while (addr = next, addr != end);
311 }
312 
313 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
314 			   phys_addr_t phys, pgprot_t prot,
315 			   phys_addr_t (*pgtable_alloc)(int),
316 			   int flags)
317 {
318 	unsigned long next;
319 	pud_t *pudp;
320 	p4d_t *p4dp = p4d_offset(pgdp, addr);
321 	p4d_t p4d = READ_ONCE(*p4dp);
322 
323 	if (p4d_none(p4d)) {
324 		p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
325 		phys_addr_t pud_phys;
326 
327 		if (flags & NO_EXEC_MAPPINGS)
328 			p4dval |= P4D_TABLE_PXN;
329 		BUG_ON(!pgtable_alloc);
330 		pud_phys = pgtable_alloc(PUD_SHIFT);
331 		__p4d_populate(p4dp, pud_phys, p4dval);
332 		p4d = READ_ONCE(*p4dp);
333 	}
334 	BUG_ON(p4d_bad(p4d));
335 
336 	pudp = pud_set_fixmap_offset(p4dp, addr);
337 	do {
338 		pud_t old_pud = READ_ONCE(*pudp);
339 
340 		next = pud_addr_end(addr, end);
341 
342 		/*
343 		 * For 4K granule only, attempt to put down a 1GB block
344 		 */
345 		if (pud_sect_supported() &&
346 		   ((addr | next | phys) & ~PUD_MASK) == 0 &&
347 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
348 			pud_set_huge(pudp, phys, prot);
349 
350 			/*
351 			 * After the PUD entry has been populated once, we
352 			 * only allow updates to the permission attributes.
353 			 */
354 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
355 						      READ_ONCE(pud_val(*pudp))));
356 		} else {
357 			alloc_init_cont_pmd(pudp, addr, next, phys, prot,
358 					    pgtable_alloc, flags);
359 
360 			BUG_ON(pud_val(old_pud) != 0 &&
361 			       pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
362 		}
363 		phys += next - addr;
364 	} while (pudp++, addr = next, addr != end);
365 
366 	pud_clear_fixmap();
367 }
368 
369 static void __create_pgd_mapping_locked(pgd_t *pgdir, phys_addr_t phys,
370 					unsigned long virt, phys_addr_t size,
371 					pgprot_t prot,
372 					phys_addr_t (*pgtable_alloc)(int),
373 					int flags)
374 {
375 	unsigned long addr, end, next;
376 	pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
377 
378 	/*
379 	 * If the virtual and physical address don't have the same offset
380 	 * within a page, we cannot map the region as the caller expects.
381 	 */
382 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
383 		return;
384 
385 	phys &= PAGE_MASK;
386 	addr = virt & PAGE_MASK;
387 	end = PAGE_ALIGN(virt + size);
388 
389 	do {
390 		next = pgd_addr_end(addr, end);
391 		alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
392 			       flags);
393 		phys += next - addr;
394 	} while (pgdp++, addr = next, addr != end);
395 }
396 
397 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
398 				 unsigned long virt, phys_addr_t size,
399 				 pgprot_t prot,
400 				 phys_addr_t (*pgtable_alloc)(int),
401 				 int flags)
402 {
403 	mutex_lock(&fixmap_lock);
404 	__create_pgd_mapping_locked(pgdir, phys, virt, size, prot,
405 				    pgtable_alloc, flags);
406 	mutex_unlock(&fixmap_lock);
407 }
408 
409 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
410 extern __alias(__create_pgd_mapping_locked)
411 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
412 			     phys_addr_t size, pgprot_t prot,
413 			     phys_addr_t (*pgtable_alloc)(int), int flags);
414 #endif
415 
416 static phys_addr_t __pgd_pgtable_alloc(int shift)
417 {
418 	void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
419 	BUG_ON(!ptr);
420 
421 	/* Ensure the zeroed page is visible to the page table walker */
422 	dsb(ishst);
423 	return __pa(ptr);
424 }
425 
426 static phys_addr_t pgd_pgtable_alloc(int shift)
427 {
428 	phys_addr_t pa = __pgd_pgtable_alloc(shift);
429 
430 	/*
431 	 * Call proper page table ctor in case later we need to
432 	 * call core mm functions like apply_to_page_range() on
433 	 * this pre-allocated page table.
434 	 *
435 	 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
436 	 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
437 	 */
438 	if (shift == PAGE_SHIFT)
439 		BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
440 	else if (shift == PMD_SHIFT)
441 		BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
442 
443 	return pa;
444 }
445 
446 /*
447  * This function can only be used to modify existing table entries,
448  * without allocating new levels of table. Note that this permits the
449  * creation of new section or page entries.
450  */
451 void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
452 				   phys_addr_t size, pgprot_t prot)
453 {
454 	if (virt < PAGE_OFFSET) {
455 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
456 			&phys, virt);
457 		return;
458 	}
459 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
460 			     NO_CONT_MAPPINGS);
461 }
462 
463 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
464 			       unsigned long virt, phys_addr_t size,
465 			       pgprot_t prot, bool page_mappings_only)
466 {
467 	int flags = 0;
468 
469 	BUG_ON(mm == &init_mm);
470 
471 	if (page_mappings_only)
472 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
473 
474 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
475 			     pgd_pgtable_alloc, flags);
476 }
477 
478 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
479 				phys_addr_t size, pgprot_t prot)
480 {
481 	if (virt < PAGE_OFFSET) {
482 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
483 			&phys, virt);
484 		return;
485 	}
486 
487 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
488 			     NO_CONT_MAPPINGS);
489 
490 	/* flush the TLBs after updating live kernel mappings */
491 	flush_tlb_kernel_range(virt, virt + size);
492 }
493 
494 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
495 				  phys_addr_t end, pgprot_t prot, int flags)
496 {
497 	__create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
498 			     prot, early_pgtable_alloc, flags);
499 }
500 
501 void __init mark_linear_text_alias_ro(void)
502 {
503 	/*
504 	 * Remove the write permissions from the linear alias of .text/.rodata
505 	 */
506 	update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
507 			    (unsigned long)__init_begin - (unsigned long)_stext,
508 			    PAGE_KERNEL_RO);
509 }
510 
511 #ifdef CONFIG_KFENCE
512 
513 bool __ro_after_init kfence_early_init = !!CONFIG_KFENCE_SAMPLE_INTERVAL;
514 
515 /* early_param() will be parsed before map_mem() below. */
516 static int __init parse_kfence_early_init(char *arg)
517 {
518 	int val;
519 
520 	if (get_option(&arg, &val))
521 		kfence_early_init = !!val;
522 	return 0;
523 }
524 early_param("kfence.sample_interval", parse_kfence_early_init);
525 
526 static phys_addr_t __init arm64_kfence_alloc_pool(void)
527 {
528 	phys_addr_t kfence_pool;
529 
530 	if (!kfence_early_init)
531 		return 0;
532 
533 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
534 	if (!kfence_pool) {
535 		pr_err("failed to allocate kfence pool\n");
536 		kfence_early_init = false;
537 		return 0;
538 	}
539 
540 	/* Temporarily mark as NOMAP. */
541 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
542 
543 	return kfence_pool;
544 }
545 
546 static void __init arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp)
547 {
548 	if (!kfence_pool)
549 		return;
550 
551 	/* KFENCE pool needs page-level mapping. */
552 	__map_memblock(pgdp, kfence_pool, kfence_pool + KFENCE_POOL_SIZE,
553 			pgprot_tagged(PAGE_KERNEL),
554 			NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
555 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
556 	__kfence_pool = phys_to_virt(kfence_pool);
557 }
558 #else /* CONFIG_KFENCE */
559 
560 static inline phys_addr_t arm64_kfence_alloc_pool(void) { return 0; }
561 static inline void arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp) { }
562 
563 #endif /* CONFIG_KFENCE */
564 
565 static void __init map_mem(pgd_t *pgdp)
566 {
567 	static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
568 	phys_addr_t kernel_start = __pa_symbol(_stext);
569 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
570 	phys_addr_t start, end;
571 	phys_addr_t early_kfence_pool;
572 	int flags = NO_EXEC_MAPPINGS;
573 	u64 i;
574 
575 	/*
576 	 * Setting hierarchical PXNTable attributes on table entries covering
577 	 * the linear region is only possible if it is guaranteed that no table
578 	 * entries at any level are being shared between the linear region and
579 	 * the vmalloc region. Check whether this is true for the PGD level, in
580 	 * which case it is guaranteed to be true for all other levels as well.
581 	 */
582 	BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
583 
584 	early_kfence_pool = arm64_kfence_alloc_pool();
585 
586 	if (can_set_direct_map())
587 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
588 
589 	/*
590 	 * Take care not to create a writable alias for the
591 	 * read-only text and rodata sections of the kernel image.
592 	 * So temporarily mark them as NOMAP to skip mappings in
593 	 * the following for-loop
594 	 */
595 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
596 
597 	/* map all the memory banks */
598 	for_each_mem_range(i, &start, &end) {
599 		if (start >= end)
600 			break;
601 		/*
602 		 * The linear map must allow allocation tags reading/writing
603 		 * if MTE is present. Otherwise, it has the same attributes as
604 		 * PAGE_KERNEL.
605 		 */
606 		__map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
607 			       flags);
608 	}
609 
610 	/*
611 	 * Map the linear alias of the [_stext, __init_begin) interval
612 	 * as non-executable now, and remove the write permission in
613 	 * mark_linear_text_alias_ro() below (which will be called after
614 	 * alternative patching has completed). This makes the contents
615 	 * of the region accessible to subsystems such as hibernate,
616 	 * but protects it from inadvertent modification or execution.
617 	 * Note that contiguous mappings cannot be remapped in this way,
618 	 * so we should avoid them here.
619 	 */
620 	__map_memblock(pgdp, kernel_start, kernel_end,
621 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
622 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
623 	arm64_kfence_map_pool(early_kfence_pool, pgdp);
624 }
625 
626 void mark_rodata_ro(void)
627 {
628 	unsigned long section_size;
629 
630 	/*
631 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
632 	 * to cover NOTES and EXCEPTION_TABLE.
633 	 */
634 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
635 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
636 			    section_size, PAGE_KERNEL_RO);
637 
638 	debug_checkwx();
639 }
640 
641 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
642 				      pgprot_t prot, struct vm_struct *vma,
643 				      int flags, unsigned long vm_flags)
644 {
645 	phys_addr_t pa_start = __pa_symbol(va_start);
646 	unsigned long size = va_end - va_start;
647 
648 	BUG_ON(!PAGE_ALIGNED(pa_start));
649 	BUG_ON(!PAGE_ALIGNED(size));
650 
651 	__create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
652 			     early_pgtable_alloc, flags);
653 
654 	if (!(vm_flags & VM_NO_GUARD))
655 		size += PAGE_SIZE;
656 
657 	vma->addr	= va_start;
658 	vma->phys_addr	= pa_start;
659 	vma->size	= size;
660 	vma->flags	= VM_MAP | vm_flags;
661 	vma->caller	= __builtin_return_address(0);
662 
663 	vm_area_add_early(vma);
664 }
665 
666 static pgprot_t kernel_exec_prot(void)
667 {
668 	return rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
669 }
670 
671 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
672 static int __init map_entry_trampoline(void)
673 {
674 	int i;
675 
676 	pgprot_t prot = kernel_exec_prot();
677 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
678 
679 	/* The trampoline is always mapped and can therefore be global */
680 	pgprot_val(prot) &= ~PTE_NG;
681 
682 	/* Map only the text into the trampoline page table */
683 	memset(tramp_pg_dir, 0, PGD_SIZE);
684 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
685 			     entry_tramp_text_size(), prot,
686 			     __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
687 
688 	/* Map both the text and data into the kernel page table */
689 	for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
690 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
691 			     pa_start + i * PAGE_SIZE, prot);
692 
693 	if (IS_ENABLED(CONFIG_RELOCATABLE))
694 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
695 			     pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
696 
697 	return 0;
698 }
699 core_initcall(map_entry_trampoline);
700 #endif
701 
702 /*
703  * Open coded check for BTI, only for use to determine configuration
704  * for early mappings for before the cpufeature code has run.
705  */
706 static bool arm64_early_this_cpu_has_bti(void)
707 {
708 	u64 pfr1;
709 
710 	if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
711 		return false;
712 
713 	pfr1 = __read_sysreg_by_encoding(SYS_ID_AA64PFR1_EL1);
714 	return cpuid_feature_extract_unsigned_field(pfr1,
715 						    ID_AA64PFR1_EL1_BT_SHIFT);
716 }
717 
718 /*
719  * Create fine-grained mappings for the kernel.
720  */
721 static void __init map_kernel(pgd_t *pgdp)
722 {
723 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
724 				vmlinux_initdata, vmlinux_data;
725 
726 	/*
727 	 * External debuggers may need to write directly to the text
728 	 * mapping to install SW breakpoints. Allow this (only) when
729 	 * explicitly requested with rodata=off.
730 	 */
731 	pgprot_t text_prot = kernel_exec_prot();
732 
733 	/*
734 	 * If we have a CPU that supports BTI and a kernel built for
735 	 * BTI then mark the kernel executable text as guarded pages
736 	 * now so we don't have to rewrite the page tables later.
737 	 */
738 	if (arm64_early_this_cpu_has_bti())
739 		text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
740 
741 	/*
742 	 * Only rodata will be remapped with different permissions later on,
743 	 * all other segments are allowed to use contiguous mappings.
744 	 */
745 	map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0,
746 			   VM_NO_GUARD);
747 	map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
748 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
749 	map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
750 			   &vmlinux_inittext, 0, VM_NO_GUARD);
751 	map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
752 			   &vmlinux_initdata, 0, VM_NO_GUARD);
753 	map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
754 
755 	fixmap_copy(pgdp);
756 	kasan_copy_shadow(pgdp);
757 }
758 
759 static void __init create_idmap(void)
760 {
761 	u64 start = __pa_symbol(__idmap_text_start);
762 	u64 size = __pa_symbol(__idmap_text_end) - start;
763 	pgd_t *pgd = idmap_pg_dir;
764 	u64 pgd_phys;
765 
766 	/* check if we need an additional level of translation */
767 	if (VA_BITS < 48 && idmap_t0sz < (64 - VA_BITS_MIN)) {
768 		pgd_phys = early_pgtable_alloc(PAGE_SHIFT);
769 		set_pgd(&idmap_pg_dir[start >> VA_BITS],
770 			__pgd(pgd_phys | P4D_TYPE_TABLE));
771 		pgd = __va(pgd_phys);
772 	}
773 	__create_pgd_mapping(pgd, start, start, size, PAGE_KERNEL_ROX,
774 			     early_pgtable_alloc, 0);
775 
776 	if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0)) {
777 		extern u32 __idmap_kpti_flag;
778 		u64 pa = __pa_symbol(&__idmap_kpti_flag);
779 
780 		/*
781 		 * The KPTI G-to-nG conversion code needs a read-write mapping
782 		 * of its synchronization flag in the ID map.
783 		 */
784 		__create_pgd_mapping(pgd, pa, pa, sizeof(u32), PAGE_KERNEL,
785 				     early_pgtable_alloc, 0);
786 	}
787 }
788 
789 void __init paging_init(void)
790 {
791 	pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
792 	extern pgd_t init_idmap_pg_dir[];
793 
794 	idmap_t0sz = 63UL - __fls(__pa_symbol(_end) | GENMASK(VA_BITS_MIN - 1, 0));
795 
796 	map_kernel(pgdp);
797 	map_mem(pgdp);
798 
799 	pgd_clear_fixmap();
800 
801 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir), init_idmap_pg_dir);
802 	init_mm.pgd = swapper_pg_dir;
803 
804 	memblock_phys_free(__pa_symbol(init_pg_dir),
805 			   __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
806 
807 	memblock_allow_resize();
808 
809 	create_idmap();
810 }
811 
812 #ifdef CONFIG_MEMORY_HOTPLUG
813 static void free_hotplug_page_range(struct page *page, size_t size,
814 				    struct vmem_altmap *altmap)
815 {
816 	if (altmap) {
817 		vmem_altmap_free(altmap, size >> PAGE_SHIFT);
818 	} else {
819 		WARN_ON(PageReserved(page));
820 		free_pages((unsigned long)page_address(page), get_order(size));
821 	}
822 }
823 
824 static void free_hotplug_pgtable_page(struct page *page)
825 {
826 	free_hotplug_page_range(page, PAGE_SIZE, NULL);
827 }
828 
829 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
830 				  unsigned long floor, unsigned long ceiling,
831 				  unsigned long mask)
832 {
833 	start &= mask;
834 	if (start < floor)
835 		return false;
836 
837 	if (ceiling) {
838 		ceiling &= mask;
839 		if (!ceiling)
840 			return false;
841 	}
842 
843 	if (end - 1 > ceiling - 1)
844 		return false;
845 	return true;
846 }
847 
848 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
849 				    unsigned long end, bool free_mapped,
850 				    struct vmem_altmap *altmap)
851 {
852 	pte_t *ptep, pte;
853 
854 	do {
855 		ptep = pte_offset_kernel(pmdp, addr);
856 		pte = READ_ONCE(*ptep);
857 		if (pte_none(pte))
858 			continue;
859 
860 		WARN_ON(!pte_present(pte));
861 		pte_clear(&init_mm, addr, ptep);
862 		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
863 		if (free_mapped)
864 			free_hotplug_page_range(pte_page(pte),
865 						PAGE_SIZE, altmap);
866 	} while (addr += PAGE_SIZE, addr < end);
867 }
868 
869 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
870 				    unsigned long end, bool free_mapped,
871 				    struct vmem_altmap *altmap)
872 {
873 	unsigned long next;
874 	pmd_t *pmdp, pmd;
875 
876 	do {
877 		next = pmd_addr_end(addr, end);
878 		pmdp = pmd_offset(pudp, addr);
879 		pmd = READ_ONCE(*pmdp);
880 		if (pmd_none(pmd))
881 			continue;
882 
883 		WARN_ON(!pmd_present(pmd));
884 		if (pmd_sect(pmd)) {
885 			pmd_clear(pmdp);
886 
887 			/*
888 			 * One TLBI should be sufficient here as the PMD_SIZE
889 			 * range is mapped with a single block entry.
890 			 */
891 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
892 			if (free_mapped)
893 				free_hotplug_page_range(pmd_page(pmd),
894 							PMD_SIZE, altmap);
895 			continue;
896 		}
897 		WARN_ON(!pmd_table(pmd));
898 		unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
899 	} while (addr = next, addr < end);
900 }
901 
902 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
903 				    unsigned long end, bool free_mapped,
904 				    struct vmem_altmap *altmap)
905 {
906 	unsigned long next;
907 	pud_t *pudp, pud;
908 
909 	do {
910 		next = pud_addr_end(addr, end);
911 		pudp = pud_offset(p4dp, addr);
912 		pud = READ_ONCE(*pudp);
913 		if (pud_none(pud))
914 			continue;
915 
916 		WARN_ON(!pud_present(pud));
917 		if (pud_sect(pud)) {
918 			pud_clear(pudp);
919 
920 			/*
921 			 * One TLBI should be sufficient here as the PUD_SIZE
922 			 * range is mapped with a single block entry.
923 			 */
924 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
925 			if (free_mapped)
926 				free_hotplug_page_range(pud_page(pud),
927 							PUD_SIZE, altmap);
928 			continue;
929 		}
930 		WARN_ON(!pud_table(pud));
931 		unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
932 	} while (addr = next, addr < end);
933 }
934 
935 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
936 				    unsigned long end, bool free_mapped,
937 				    struct vmem_altmap *altmap)
938 {
939 	unsigned long next;
940 	p4d_t *p4dp, p4d;
941 
942 	do {
943 		next = p4d_addr_end(addr, end);
944 		p4dp = p4d_offset(pgdp, addr);
945 		p4d = READ_ONCE(*p4dp);
946 		if (p4d_none(p4d))
947 			continue;
948 
949 		WARN_ON(!p4d_present(p4d));
950 		unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
951 	} while (addr = next, addr < end);
952 }
953 
954 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
955 				bool free_mapped, struct vmem_altmap *altmap)
956 {
957 	unsigned long next;
958 	pgd_t *pgdp, pgd;
959 
960 	/*
961 	 * altmap can only be used as vmemmap mapping backing memory.
962 	 * In case the backing memory itself is not being freed, then
963 	 * altmap is irrelevant. Warn about this inconsistency when
964 	 * encountered.
965 	 */
966 	WARN_ON(!free_mapped && altmap);
967 
968 	do {
969 		next = pgd_addr_end(addr, end);
970 		pgdp = pgd_offset_k(addr);
971 		pgd = READ_ONCE(*pgdp);
972 		if (pgd_none(pgd))
973 			continue;
974 
975 		WARN_ON(!pgd_present(pgd));
976 		unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
977 	} while (addr = next, addr < end);
978 }
979 
980 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
981 				 unsigned long end, unsigned long floor,
982 				 unsigned long ceiling)
983 {
984 	pte_t *ptep, pte;
985 	unsigned long i, start = addr;
986 
987 	do {
988 		ptep = pte_offset_kernel(pmdp, addr);
989 		pte = READ_ONCE(*ptep);
990 
991 		/*
992 		 * This is just a sanity check here which verifies that
993 		 * pte clearing has been done by earlier unmap loops.
994 		 */
995 		WARN_ON(!pte_none(pte));
996 	} while (addr += PAGE_SIZE, addr < end);
997 
998 	if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
999 		return;
1000 
1001 	/*
1002 	 * Check whether we can free the pte page if the rest of the
1003 	 * entries are empty. Overlap with other regions have been
1004 	 * handled by the floor/ceiling check.
1005 	 */
1006 	ptep = pte_offset_kernel(pmdp, 0UL);
1007 	for (i = 0; i < PTRS_PER_PTE; i++) {
1008 		if (!pte_none(READ_ONCE(ptep[i])))
1009 			return;
1010 	}
1011 
1012 	pmd_clear(pmdp);
1013 	__flush_tlb_kernel_pgtable(start);
1014 	free_hotplug_pgtable_page(virt_to_page(ptep));
1015 }
1016 
1017 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1018 				 unsigned long end, unsigned long floor,
1019 				 unsigned long ceiling)
1020 {
1021 	pmd_t *pmdp, pmd;
1022 	unsigned long i, next, start = addr;
1023 
1024 	do {
1025 		next = pmd_addr_end(addr, end);
1026 		pmdp = pmd_offset(pudp, addr);
1027 		pmd = READ_ONCE(*pmdp);
1028 		if (pmd_none(pmd))
1029 			continue;
1030 
1031 		WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1032 		free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1033 	} while (addr = next, addr < end);
1034 
1035 	if (CONFIG_PGTABLE_LEVELS <= 2)
1036 		return;
1037 
1038 	if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1039 		return;
1040 
1041 	/*
1042 	 * Check whether we can free the pmd page if the rest of the
1043 	 * entries are empty. Overlap with other regions have been
1044 	 * handled by the floor/ceiling check.
1045 	 */
1046 	pmdp = pmd_offset(pudp, 0UL);
1047 	for (i = 0; i < PTRS_PER_PMD; i++) {
1048 		if (!pmd_none(READ_ONCE(pmdp[i])))
1049 			return;
1050 	}
1051 
1052 	pud_clear(pudp);
1053 	__flush_tlb_kernel_pgtable(start);
1054 	free_hotplug_pgtable_page(virt_to_page(pmdp));
1055 }
1056 
1057 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1058 				 unsigned long end, unsigned long floor,
1059 				 unsigned long ceiling)
1060 {
1061 	pud_t *pudp, pud;
1062 	unsigned long i, next, start = addr;
1063 
1064 	do {
1065 		next = pud_addr_end(addr, end);
1066 		pudp = pud_offset(p4dp, addr);
1067 		pud = READ_ONCE(*pudp);
1068 		if (pud_none(pud))
1069 			continue;
1070 
1071 		WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1072 		free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1073 	} while (addr = next, addr < end);
1074 
1075 	if (CONFIG_PGTABLE_LEVELS <= 3)
1076 		return;
1077 
1078 	if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1079 		return;
1080 
1081 	/*
1082 	 * Check whether we can free the pud page if the rest of the
1083 	 * entries are empty. Overlap with other regions have been
1084 	 * handled by the floor/ceiling check.
1085 	 */
1086 	pudp = pud_offset(p4dp, 0UL);
1087 	for (i = 0; i < PTRS_PER_PUD; i++) {
1088 		if (!pud_none(READ_ONCE(pudp[i])))
1089 			return;
1090 	}
1091 
1092 	p4d_clear(p4dp);
1093 	__flush_tlb_kernel_pgtable(start);
1094 	free_hotplug_pgtable_page(virt_to_page(pudp));
1095 }
1096 
1097 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1098 				 unsigned long end, unsigned long floor,
1099 				 unsigned long ceiling)
1100 {
1101 	unsigned long next;
1102 	p4d_t *p4dp, p4d;
1103 
1104 	do {
1105 		next = p4d_addr_end(addr, end);
1106 		p4dp = p4d_offset(pgdp, addr);
1107 		p4d = READ_ONCE(*p4dp);
1108 		if (p4d_none(p4d))
1109 			continue;
1110 
1111 		WARN_ON(!p4d_present(p4d));
1112 		free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1113 	} while (addr = next, addr < end);
1114 }
1115 
1116 static void free_empty_tables(unsigned long addr, unsigned long end,
1117 			      unsigned long floor, unsigned long ceiling)
1118 {
1119 	unsigned long next;
1120 	pgd_t *pgdp, pgd;
1121 
1122 	do {
1123 		next = pgd_addr_end(addr, end);
1124 		pgdp = pgd_offset_k(addr);
1125 		pgd = READ_ONCE(*pgdp);
1126 		if (pgd_none(pgd))
1127 			continue;
1128 
1129 		WARN_ON(!pgd_present(pgd));
1130 		free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1131 	} while (addr = next, addr < end);
1132 }
1133 #endif
1134 
1135 void __meminit vmemmap_set_pmd(pmd_t *pmdp, void *p, int node,
1136 			       unsigned long addr, unsigned long next)
1137 {
1138 	pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1139 }
1140 
1141 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1142 				unsigned long addr, unsigned long next)
1143 {
1144 	vmemmap_verify((pte_t *)pmdp, node, addr, next);
1145 	return 1;
1146 }
1147 
1148 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1149 		struct vmem_altmap *altmap)
1150 {
1151 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1152 
1153 	if (!IS_ENABLED(CONFIG_ARM64_4K_PAGES))
1154 		return vmemmap_populate_basepages(start, end, node, altmap);
1155 	else
1156 		return vmemmap_populate_hugepages(start, end, node, altmap);
1157 }
1158 
1159 #ifdef CONFIG_MEMORY_HOTPLUG
1160 void vmemmap_free(unsigned long start, unsigned long end,
1161 		struct vmem_altmap *altmap)
1162 {
1163 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1164 
1165 	unmap_hotplug_range(start, end, true, altmap);
1166 	free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1167 }
1168 #endif /* CONFIG_MEMORY_HOTPLUG */
1169 
1170 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1171 {
1172 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1173 
1174 	/* Only allow permission changes for now */
1175 	if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1176 				   pud_val(new_pud)))
1177 		return 0;
1178 
1179 	VM_BUG_ON(phys & ~PUD_MASK);
1180 	set_pud(pudp, new_pud);
1181 	return 1;
1182 }
1183 
1184 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1185 {
1186 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1187 
1188 	/* Only allow permission changes for now */
1189 	if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1190 				   pmd_val(new_pmd)))
1191 		return 0;
1192 
1193 	VM_BUG_ON(phys & ~PMD_MASK);
1194 	set_pmd(pmdp, new_pmd);
1195 	return 1;
1196 }
1197 
1198 int pud_clear_huge(pud_t *pudp)
1199 {
1200 	if (!pud_sect(READ_ONCE(*pudp)))
1201 		return 0;
1202 	pud_clear(pudp);
1203 	return 1;
1204 }
1205 
1206 int pmd_clear_huge(pmd_t *pmdp)
1207 {
1208 	if (!pmd_sect(READ_ONCE(*pmdp)))
1209 		return 0;
1210 	pmd_clear(pmdp);
1211 	return 1;
1212 }
1213 
1214 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1215 {
1216 	pte_t *table;
1217 	pmd_t pmd;
1218 
1219 	pmd = READ_ONCE(*pmdp);
1220 
1221 	if (!pmd_table(pmd)) {
1222 		VM_WARN_ON(1);
1223 		return 1;
1224 	}
1225 
1226 	table = pte_offset_kernel(pmdp, addr);
1227 	pmd_clear(pmdp);
1228 	__flush_tlb_kernel_pgtable(addr);
1229 	pte_free_kernel(NULL, table);
1230 	return 1;
1231 }
1232 
1233 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1234 {
1235 	pmd_t *table;
1236 	pmd_t *pmdp;
1237 	pud_t pud;
1238 	unsigned long next, end;
1239 
1240 	pud = READ_ONCE(*pudp);
1241 
1242 	if (!pud_table(pud)) {
1243 		VM_WARN_ON(1);
1244 		return 1;
1245 	}
1246 
1247 	table = pmd_offset(pudp, addr);
1248 	pmdp = table;
1249 	next = addr;
1250 	end = addr + PUD_SIZE;
1251 	do {
1252 		pmd_free_pte_page(pmdp, next);
1253 	} while (pmdp++, next += PMD_SIZE, next != end);
1254 
1255 	pud_clear(pudp);
1256 	__flush_tlb_kernel_pgtable(addr);
1257 	pmd_free(NULL, table);
1258 	return 1;
1259 }
1260 
1261 #ifdef CONFIG_MEMORY_HOTPLUG
1262 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1263 {
1264 	unsigned long end = start + size;
1265 
1266 	WARN_ON(pgdir != init_mm.pgd);
1267 	WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1268 
1269 	unmap_hotplug_range(start, end, false, NULL);
1270 	free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1271 }
1272 
1273 struct range arch_get_mappable_range(void)
1274 {
1275 	struct range mhp_range;
1276 	u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1277 	u64 end_linear_pa = __pa(PAGE_END - 1);
1278 
1279 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1280 		/*
1281 		 * Check for a wrap, it is possible because of randomized linear
1282 		 * mapping the start physical address is actually bigger than
1283 		 * the end physical address. In this case set start to zero
1284 		 * because [0, end_linear_pa] range must still be able to cover
1285 		 * all addressable physical addresses.
1286 		 */
1287 		if (start_linear_pa > end_linear_pa)
1288 			start_linear_pa = 0;
1289 	}
1290 
1291 	WARN_ON(start_linear_pa > end_linear_pa);
1292 
1293 	/*
1294 	 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1295 	 * accommodating both its ends but excluding PAGE_END. Max physical
1296 	 * range which can be mapped inside this linear mapping range, must
1297 	 * also be derived from its end points.
1298 	 */
1299 	mhp_range.start = start_linear_pa;
1300 	mhp_range.end =  end_linear_pa;
1301 
1302 	return mhp_range;
1303 }
1304 
1305 int arch_add_memory(int nid, u64 start, u64 size,
1306 		    struct mhp_params *params)
1307 {
1308 	int ret, flags = NO_EXEC_MAPPINGS;
1309 
1310 	VM_BUG_ON(!mhp_range_allowed(start, size, true));
1311 
1312 	if (can_set_direct_map())
1313 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1314 
1315 	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1316 			     size, params->pgprot, __pgd_pgtable_alloc,
1317 			     flags);
1318 
1319 	memblock_clear_nomap(start, size);
1320 
1321 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1322 			   params);
1323 	if (ret)
1324 		__remove_pgd_mapping(swapper_pg_dir,
1325 				     __phys_to_virt(start), size);
1326 	else {
1327 		max_pfn = PFN_UP(start + size);
1328 		max_low_pfn = max_pfn;
1329 	}
1330 
1331 	return ret;
1332 }
1333 
1334 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1335 {
1336 	unsigned long start_pfn = start >> PAGE_SHIFT;
1337 	unsigned long nr_pages = size >> PAGE_SHIFT;
1338 
1339 	__remove_pages(start_pfn, nr_pages, altmap);
1340 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1341 }
1342 
1343 /*
1344  * This memory hotplug notifier helps prevent boot memory from being
1345  * inadvertently removed as it blocks pfn range offlining process in
1346  * __offline_pages(). Hence this prevents both offlining as well as
1347  * removal process for boot memory which is initially always online.
1348  * In future if and when boot memory could be removed, this notifier
1349  * should be dropped and free_hotplug_page_range() should handle any
1350  * reserved pages allocated during boot.
1351  */
1352 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1353 					   unsigned long action, void *data)
1354 {
1355 	struct mem_section *ms;
1356 	struct memory_notify *arg = data;
1357 	unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1358 	unsigned long pfn = arg->start_pfn;
1359 
1360 	if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1361 		return NOTIFY_OK;
1362 
1363 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1364 		unsigned long start = PFN_PHYS(pfn);
1365 		unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1366 
1367 		ms = __pfn_to_section(pfn);
1368 		if (!early_section(ms))
1369 			continue;
1370 
1371 		if (action == MEM_GOING_OFFLINE) {
1372 			/*
1373 			 * Boot memory removal is not supported. Prevent
1374 			 * it via blocking any attempted offline request
1375 			 * for the boot memory and just report it.
1376 			 */
1377 			pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1378 			return NOTIFY_BAD;
1379 		} else if (action == MEM_OFFLINE) {
1380 			/*
1381 			 * This should have never happened. Boot memory
1382 			 * offlining should have been prevented by this
1383 			 * very notifier. Probably some memory removal
1384 			 * procedure might have changed which would then
1385 			 * require further debug.
1386 			 */
1387 			pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1388 
1389 			/*
1390 			 * Core memory hotplug does not process a return
1391 			 * code from the notifier for MEM_OFFLINE events.
1392 			 * The error condition has been reported. Return
1393 			 * from here as if ignored.
1394 			 */
1395 			return NOTIFY_DONE;
1396 		}
1397 	}
1398 	return NOTIFY_OK;
1399 }
1400 
1401 static struct notifier_block prevent_bootmem_remove_nb = {
1402 	.notifier_call = prevent_bootmem_remove_notifier,
1403 };
1404 
1405 /*
1406  * This ensures that boot memory sections on the platform are online
1407  * from early boot. Memory sections could not be prevented from being
1408  * offlined, unless for some reason they are not online to begin with.
1409  * This helps validate the basic assumption on which the above memory
1410  * event notifier works to prevent boot memory section offlining and
1411  * its possible removal.
1412  */
1413 static void validate_bootmem_online(void)
1414 {
1415 	phys_addr_t start, end, addr;
1416 	struct mem_section *ms;
1417 	u64 i;
1418 
1419 	/*
1420 	 * Scanning across all memblock might be expensive
1421 	 * on some big memory systems. Hence enable this
1422 	 * validation only with DEBUG_VM.
1423 	 */
1424 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
1425 		return;
1426 
1427 	for_each_mem_range(i, &start, &end) {
1428 		for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1429 			ms = __pfn_to_section(PHYS_PFN(addr));
1430 
1431 			/*
1432 			 * All memory ranges in the system at this point
1433 			 * should have been marked as early sections.
1434 			 */
1435 			WARN_ON(!early_section(ms));
1436 
1437 			/*
1438 			 * Memory notifier mechanism here to prevent boot
1439 			 * memory offlining depends on the fact that each
1440 			 * early section memory on the system is initially
1441 			 * online. Otherwise a given memory section which
1442 			 * is already offline will be overlooked and can
1443 			 * be removed completely. Call out such sections.
1444 			 */
1445 			if (!online_section(ms))
1446 				pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1447 					addr, addr + (1UL << PA_SECTION_SHIFT));
1448 		}
1449 	}
1450 }
1451 
1452 static int __init prevent_bootmem_remove_init(void)
1453 {
1454 	int ret = 0;
1455 
1456 	if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1457 		return ret;
1458 
1459 	validate_bootmem_online();
1460 	ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1461 	if (ret)
1462 		pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1463 
1464 	return ret;
1465 }
1466 early_initcall(prevent_bootmem_remove_init);
1467 #endif
1468 
1469 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
1470 {
1471 	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
1472 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
1473 		/*
1474 		 * Break-before-make (BBM) is required for all user space mappings
1475 		 * when the permission changes from executable to non-executable
1476 		 * in cases where cpu is affected with errata #2645198.
1477 		 */
1478 		if (pte_user_exec(READ_ONCE(*ptep)))
1479 			return ptep_clear_flush(vma, addr, ptep);
1480 	}
1481 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1482 }
1483 
1484 void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1485 			     pte_t old_pte, pte_t pte)
1486 {
1487 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1488 }
1489