xref: /linux/arch/arm64/mm/mmu.c (revision 36ec807b627b4c0a0a382f0ae48eac7187d14b2b)
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 u64 kimage_voffset __ro_after_init;
49 EXPORT_SYMBOL(kimage_voffset);
50 
51 u32 __boot_cpu_mode[] = { BOOT_CPU_MODE_EL2, BOOT_CPU_MODE_EL1 };
52 
53 static bool rodata_is_rw __ro_after_init = true;
54 
55 /*
56  * The booting CPU updates the failed status @__early_cpu_boot_status,
57  * with MMU turned off.
58  */
59 long __section(".mmuoff.data.write") __early_cpu_boot_status;
60 
61 /*
62  * Empty_zero_page is a special page that is used for zero-initialized data
63  * and COW.
64  */
65 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
66 EXPORT_SYMBOL(empty_zero_page);
67 
68 static DEFINE_SPINLOCK(swapper_pgdir_lock);
69 static DEFINE_MUTEX(fixmap_lock);
70 
71 void noinstr set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
72 {
73 	pgd_t *fixmap_pgdp;
74 
75 	/*
76 	 * Don't bother with the fixmap if swapper_pg_dir is still mapped
77 	 * writable in the kernel mapping.
78 	 */
79 	if (rodata_is_rw) {
80 		WRITE_ONCE(*pgdp, pgd);
81 		dsb(ishst);
82 		isb();
83 		return;
84 	}
85 
86 	spin_lock(&swapper_pgdir_lock);
87 	fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
88 	WRITE_ONCE(*fixmap_pgdp, pgd);
89 	/*
90 	 * We need dsb(ishst) here to ensure the page-table-walker sees
91 	 * our new entry before set_p?d() returns. The fixmap's
92 	 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
93 	 */
94 	pgd_clear_fixmap();
95 	spin_unlock(&swapper_pgdir_lock);
96 }
97 
98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
99 			      unsigned long size, pgprot_t vma_prot)
100 {
101 	if (!pfn_is_map_memory(pfn))
102 		return pgprot_noncached(vma_prot);
103 	else if (file->f_flags & O_SYNC)
104 		return pgprot_writecombine(vma_prot);
105 	return vma_prot;
106 }
107 EXPORT_SYMBOL(phys_mem_access_prot);
108 
109 static phys_addr_t __init early_pgtable_alloc(int shift)
110 {
111 	phys_addr_t phys;
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 	return phys;
119 }
120 
121 bool pgattr_change_is_safe(u64 old, u64 new)
122 {
123 	/*
124 	 * The following mapping attributes may be updated in live
125 	 * kernel mappings without the need for break-before-make.
126 	 */
127 	pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG |
128 			PTE_SWBITS_MASK;
129 
130 	/* creating or taking down mappings is always safe */
131 	if (!pte_valid(__pte(old)) || !pte_valid(__pte(new)))
132 		return true;
133 
134 	/* A live entry's pfn should not change */
135 	if (pte_pfn(__pte(old)) != pte_pfn(__pte(new)))
136 		return false;
137 
138 	/* live contiguous mappings may not be manipulated at all */
139 	if ((old | new) & PTE_CONT)
140 		return false;
141 
142 	/* Transitioning from Non-Global to Global is unsafe */
143 	if (old & ~new & PTE_NG)
144 		return false;
145 
146 	/*
147 	 * Changing the memory type between Normal and Normal-Tagged is safe
148 	 * since Tagged is considered a permission attribute from the
149 	 * mismatched attribute aliases perspective.
150 	 */
151 	if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
152 	     (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
153 	    ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
154 	     (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
155 		mask |= PTE_ATTRINDX_MASK;
156 
157 	return ((old ^ new) & ~mask) == 0;
158 }
159 
160 static void init_clear_pgtable(void *table)
161 {
162 	clear_page(table);
163 
164 	/* Ensure the zeroing is observed by page table walks. */
165 	dsb(ishst);
166 }
167 
168 static void init_pte(pte_t *ptep, unsigned long addr, unsigned long end,
169 		     phys_addr_t phys, pgprot_t prot)
170 {
171 	do {
172 		pte_t old_pte = __ptep_get(ptep);
173 
174 		/*
175 		 * Required barriers to make this visible to the table walker
176 		 * are deferred to the end of alloc_init_cont_pte().
177 		 */
178 		__set_pte_nosync(ptep, pfn_pte(__phys_to_pfn(phys), prot));
179 
180 		/*
181 		 * After the PTE entry has been populated once, we
182 		 * only allow updates to the permission attributes.
183 		 */
184 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
185 					      pte_val(__ptep_get(ptep))));
186 
187 		phys += PAGE_SIZE;
188 	} while (ptep++, addr += PAGE_SIZE, addr != end);
189 }
190 
191 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
192 				unsigned long end, phys_addr_t phys,
193 				pgprot_t prot,
194 				phys_addr_t (*pgtable_alloc)(int),
195 				int flags)
196 {
197 	unsigned long next;
198 	pmd_t pmd = READ_ONCE(*pmdp);
199 	pte_t *ptep;
200 
201 	BUG_ON(pmd_sect(pmd));
202 	if (pmd_none(pmd)) {
203 		pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
204 		phys_addr_t pte_phys;
205 
206 		if (flags & NO_EXEC_MAPPINGS)
207 			pmdval |= PMD_TABLE_PXN;
208 		BUG_ON(!pgtable_alloc);
209 		pte_phys = pgtable_alloc(PAGE_SHIFT);
210 		ptep = pte_set_fixmap(pte_phys);
211 		init_clear_pgtable(ptep);
212 		ptep += pte_index(addr);
213 		__pmd_populate(pmdp, pte_phys, pmdval);
214 	} else {
215 		BUG_ON(pmd_bad(pmd));
216 		ptep = pte_set_fixmap_offset(pmdp, addr);
217 	}
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(ptep, addr, next, phys, __prot);
230 
231 		ptep += pte_index(next) - pte_index(addr);
232 		phys += next - addr;
233 	} while (addr = next, addr != end);
234 
235 	/*
236 	 * Note: barriers and maintenance necessary to clear the fixmap slot
237 	 * ensure that all previous pgtable writes are visible to the table
238 	 * walker.
239 	 */
240 	pte_clear_fixmap();
241 }
242 
243 static void init_pmd(pmd_t *pmdp, unsigned long addr, unsigned long end,
244 		     phys_addr_t phys, pgprot_t prot,
245 		     phys_addr_t (*pgtable_alloc)(int), int flags)
246 {
247 	unsigned long next;
248 
249 	do {
250 		pmd_t old_pmd = READ_ONCE(*pmdp);
251 
252 		next = pmd_addr_end(addr, end);
253 
254 		/* try section mapping first */
255 		if (((addr | next | phys) & ~PMD_MASK) == 0 &&
256 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
257 			pmd_set_huge(pmdp, phys, prot);
258 
259 			/*
260 			 * After the PMD entry has been populated once, we
261 			 * only allow updates to the permission attributes.
262 			 */
263 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
264 						      READ_ONCE(pmd_val(*pmdp))));
265 		} else {
266 			alloc_init_cont_pte(pmdp, addr, next, phys, prot,
267 					    pgtable_alloc, flags);
268 
269 			BUG_ON(pmd_val(old_pmd) != 0 &&
270 			       pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
271 		}
272 		phys += next - addr;
273 	} while (pmdp++, addr = next, addr != end);
274 }
275 
276 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
277 				unsigned long end, phys_addr_t phys,
278 				pgprot_t prot,
279 				phys_addr_t (*pgtable_alloc)(int), int flags)
280 {
281 	unsigned long next;
282 	pud_t pud = READ_ONCE(*pudp);
283 	pmd_t *pmdp;
284 
285 	/*
286 	 * Check for initial section mappings in the pgd/pud.
287 	 */
288 	BUG_ON(pud_sect(pud));
289 	if (pud_none(pud)) {
290 		pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
291 		phys_addr_t pmd_phys;
292 
293 		if (flags & NO_EXEC_MAPPINGS)
294 			pudval |= PUD_TABLE_PXN;
295 		BUG_ON(!pgtable_alloc);
296 		pmd_phys = pgtable_alloc(PMD_SHIFT);
297 		pmdp = pmd_set_fixmap(pmd_phys);
298 		init_clear_pgtable(pmdp);
299 		pmdp += pmd_index(addr);
300 		__pud_populate(pudp, pmd_phys, pudval);
301 	} else {
302 		BUG_ON(pud_bad(pud));
303 		pmdp = pmd_set_fixmap_offset(pudp, addr);
304 	}
305 
306 	do {
307 		pgprot_t __prot = prot;
308 
309 		next = pmd_cont_addr_end(addr, end);
310 
311 		/* use a contiguous mapping if the range is suitably aligned */
312 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
313 		    (flags & NO_CONT_MAPPINGS) == 0)
314 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
315 
316 		init_pmd(pmdp, addr, next, phys, __prot, pgtable_alloc, flags);
317 
318 		pmdp += pmd_index(next) - pmd_index(addr);
319 		phys += next - addr;
320 	} while (addr = next, addr != end);
321 
322 	pmd_clear_fixmap();
323 }
324 
325 static void alloc_init_pud(p4d_t *p4dp, unsigned long addr, unsigned long end,
326 			   phys_addr_t phys, pgprot_t prot,
327 			   phys_addr_t (*pgtable_alloc)(int),
328 			   int flags)
329 {
330 	unsigned long next;
331 	p4d_t p4d = READ_ONCE(*p4dp);
332 	pud_t *pudp;
333 
334 	if (p4d_none(p4d)) {
335 		p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
336 		phys_addr_t pud_phys;
337 
338 		if (flags & NO_EXEC_MAPPINGS)
339 			p4dval |= P4D_TABLE_PXN;
340 		BUG_ON(!pgtable_alloc);
341 		pud_phys = pgtable_alloc(PUD_SHIFT);
342 		pudp = pud_set_fixmap(pud_phys);
343 		init_clear_pgtable(pudp);
344 		pudp += pud_index(addr);
345 		__p4d_populate(p4dp, pud_phys, p4dval);
346 	} else {
347 		BUG_ON(p4d_bad(p4d));
348 		pudp = pud_set_fixmap_offset(p4dp, addr);
349 	}
350 
351 	do {
352 		pud_t old_pud = READ_ONCE(*pudp);
353 
354 		next = pud_addr_end(addr, end);
355 
356 		/*
357 		 * For 4K granule only, attempt to put down a 1GB block
358 		 */
359 		if (pud_sect_supported() &&
360 		   ((addr | next | phys) & ~PUD_MASK) == 0 &&
361 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
362 			pud_set_huge(pudp, phys, prot);
363 
364 			/*
365 			 * After the PUD entry has been populated once, we
366 			 * only allow updates to the permission attributes.
367 			 */
368 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
369 						      READ_ONCE(pud_val(*pudp))));
370 		} else {
371 			alloc_init_cont_pmd(pudp, addr, next, phys, prot,
372 					    pgtable_alloc, flags);
373 
374 			BUG_ON(pud_val(old_pud) != 0 &&
375 			       pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
376 		}
377 		phys += next - addr;
378 	} while (pudp++, addr = next, addr != end);
379 
380 	pud_clear_fixmap();
381 }
382 
383 static void alloc_init_p4d(pgd_t *pgdp, unsigned long addr, unsigned long end,
384 			   phys_addr_t phys, pgprot_t prot,
385 			   phys_addr_t (*pgtable_alloc)(int),
386 			   int flags)
387 {
388 	unsigned long next;
389 	pgd_t pgd = READ_ONCE(*pgdp);
390 	p4d_t *p4dp;
391 
392 	if (pgd_none(pgd)) {
393 		pgdval_t pgdval = PGD_TYPE_TABLE | PGD_TABLE_UXN;
394 		phys_addr_t p4d_phys;
395 
396 		if (flags & NO_EXEC_MAPPINGS)
397 			pgdval |= PGD_TABLE_PXN;
398 		BUG_ON(!pgtable_alloc);
399 		p4d_phys = pgtable_alloc(P4D_SHIFT);
400 		p4dp = p4d_set_fixmap(p4d_phys);
401 		init_clear_pgtable(p4dp);
402 		p4dp += p4d_index(addr);
403 		__pgd_populate(pgdp, p4d_phys, pgdval);
404 	} else {
405 		BUG_ON(pgd_bad(pgd));
406 		p4dp = p4d_set_fixmap_offset(pgdp, addr);
407 	}
408 
409 	do {
410 		p4d_t old_p4d = READ_ONCE(*p4dp);
411 
412 		next = p4d_addr_end(addr, end);
413 
414 		alloc_init_pud(p4dp, addr, next, phys, prot,
415 			       pgtable_alloc, flags);
416 
417 		BUG_ON(p4d_val(old_p4d) != 0 &&
418 		       p4d_val(old_p4d) != READ_ONCE(p4d_val(*p4dp)));
419 
420 		phys += next - addr;
421 	} while (p4dp++, addr = next, addr != end);
422 
423 	p4d_clear_fixmap();
424 }
425 
426 static void __create_pgd_mapping_locked(pgd_t *pgdir, phys_addr_t phys,
427 					unsigned long virt, phys_addr_t size,
428 					pgprot_t prot,
429 					phys_addr_t (*pgtable_alloc)(int),
430 					int flags)
431 {
432 	unsigned long addr, end, next;
433 	pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
434 
435 	/*
436 	 * If the virtual and physical address don't have the same offset
437 	 * within a page, we cannot map the region as the caller expects.
438 	 */
439 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
440 		return;
441 
442 	phys &= PAGE_MASK;
443 	addr = virt & PAGE_MASK;
444 	end = PAGE_ALIGN(virt + size);
445 
446 	do {
447 		next = pgd_addr_end(addr, end);
448 		alloc_init_p4d(pgdp, addr, next, phys, prot, pgtable_alloc,
449 			       flags);
450 		phys += next - addr;
451 	} while (pgdp++, addr = next, addr != end);
452 }
453 
454 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
455 				 unsigned long virt, phys_addr_t size,
456 				 pgprot_t prot,
457 				 phys_addr_t (*pgtable_alloc)(int),
458 				 int flags)
459 {
460 	mutex_lock(&fixmap_lock);
461 	__create_pgd_mapping_locked(pgdir, phys, virt, size, prot,
462 				    pgtable_alloc, flags);
463 	mutex_unlock(&fixmap_lock);
464 }
465 
466 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
467 extern __alias(__create_pgd_mapping_locked)
468 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
469 			     phys_addr_t size, pgprot_t prot,
470 			     phys_addr_t (*pgtable_alloc)(int), int flags);
471 #endif
472 
473 static phys_addr_t __pgd_pgtable_alloc(int shift)
474 {
475 	/* Page is zeroed by init_clear_pgtable() so don't duplicate effort. */
476 	void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL & ~__GFP_ZERO);
477 
478 	BUG_ON(!ptr);
479 	return __pa(ptr);
480 }
481 
482 static phys_addr_t pgd_pgtable_alloc(int shift)
483 {
484 	phys_addr_t pa = __pgd_pgtable_alloc(shift);
485 	struct ptdesc *ptdesc = page_ptdesc(phys_to_page(pa));
486 
487 	/*
488 	 * Call proper page table ctor in case later we need to
489 	 * call core mm functions like apply_to_page_range() on
490 	 * this pre-allocated page table.
491 	 *
492 	 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
493 	 * folded, and if so pagetable_pte_ctor() becomes nop.
494 	 */
495 	if (shift == PAGE_SHIFT)
496 		BUG_ON(!pagetable_pte_ctor(ptdesc));
497 	else if (shift == PMD_SHIFT)
498 		BUG_ON(!pagetable_pmd_ctor(ptdesc));
499 
500 	return pa;
501 }
502 
503 /*
504  * This function can only be used to modify existing table entries,
505  * without allocating new levels of table. Note that this permits the
506  * creation of new section or page entries.
507  */
508 void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
509 				   phys_addr_t size, pgprot_t prot)
510 {
511 	if (virt < PAGE_OFFSET) {
512 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
513 			&phys, virt);
514 		return;
515 	}
516 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
517 			     NO_CONT_MAPPINGS);
518 }
519 
520 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
521 			       unsigned long virt, phys_addr_t size,
522 			       pgprot_t prot, bool page_mappings_only)
523 {
524 	int flags = 0;
525 
526 	BUG_ON(mm == &init_mm);
527 
528 	if (page_mappings_only)
529 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
530 
531 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
532 			     pgd_pgtable_alloc, flags);
533 }
534 
535 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
536 				phys_addr_t size, pgprot_t prot)
537 {
538 	if (virt < PAGE_OFFSET) {
539 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
540 			&phys, virt);
541 		return;
542 	}
543 
544 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
545 			     NO_CONT_MAPPINGS);
546 
547 	/* flush the TLBs after updating live kernel mappings */
548 	flush_tlb_kernel_range(virt, virt + size);
549 }
550 
551 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
552 				  phys_addr_t end, pgprot_t prot, int flags)
553 {
554 	__create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
555 			     prot, early_pgtable_alloc, flags);
556 }
557 
558 void __init mark_linear_text_alias_ro(void)
559 {
560 	/*
561 	 * Remove the write permissions from the linear alias of .text/.rodata
562 	 */
563 	update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
564 			    (unsigned long)__init_begin - (unsigned long)_stext,
565 			    PAGE_KERNEL_RO);
566 }
567 
568 #ifdef CONFIG_KFENCE
569 
570 bool __ro_after_init kfence_early_init = !!CONFIG_KFENCE_SAMPLE_INTERVAL;
571 
572 /* early_param() will be parsed before map_mem() below. */
573 static int __init parse_kfence_early_init(char *arg)
574 {
575 	int val;
576 
577 	if (get_option(&arg, &val))
578 		kfence_early_init = !!val;
579 	return 0;
580 }
581 early_param("kfence.sample_interval", parse_kfence_early_init);
582 
583 static phys_addr_t __init arm64_kfence_alloc_pool(void)
584 {
585 	phys_addr_t kfence_pool;
586 
587 	if (!kfence_early_init)
588 		return 0;
589 
590 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
591 	if (!kfence_pool) {
592 		pr_err("failed to allocate kfence pool\n");
593 		kfence_early_init = false;
594 		return 0;
595 	}
596 
597 	/* Temporarily mark as NOMAP. */
598 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
599 
600 	return kfence_pool;
601 }
602 
603 static void __init arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp)
604 {
605 	if (!kfence_pool)
606 		return;
607 
608 	/* KFENCE pool needs page-level mapping. */
609 	__map_memblock(pgdp, kfence_pool, kfence_pool + KFENCE_POOL_SIZE,
610 			pgprot_tagged(PAGE_KERNEL),
611 			NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
612 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
613 	__kfence_pool = phys_to_virt(kfence_pool);
614 }
615 #else /* CONFIG_KFENCE */
616 
617 static inline phys_addr_t arm64_kfence_alloc_pool(void) { return 0; }
618 static inline void arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp) { }
619 
620 #endif /* CONFIG_KFENCE */
621 
622 static void __init map_mem(pgd_t *pgdp)
623 {
624 	static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
625 	phys_addr_t kernel_start = __pa_symbol(_stext);
626 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
627 	phys_addr_t start, end;
628 	phys_addr_t early_kfence_pool;
629 	int flags = NO_EXEC_MAPPINGS;
630 	u64 i;
631 
632 	/*
633 	 * Setting hierarchical PXNTable attributes on table entries covering
634 	 * the linear region is only possible if it is guaranteed that no table
635 	 * entries at any level are being shared between the linear region and
636 	 * the vmalloc region. Check whether this is true for the PGD level, in
637 	 * which case it is guaranteed to be true for all other levels as well.
638 	 * (Unless we are running with support for LPA2, in which case the
639 	 * entire reduced VA space is covered by a single pgd_t which will have
640 	 * been populated without the PXNTable attribute by the time we get here.)
641 	 */
642 	BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end) &&
643 		     pgd_index(_PAGE_OFFSET(VA_BITS_MIN)) != PTRS_PER_PGD - 1);
644 
645 	early_kfence_pool = arm64_kfence_alloc_pool();
646 
647 	if (can_set_direct_map())
648 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
649 
650 	/*
651 	 * Take care not to create a writable alias for the
652 	 * read-only text and rodata sections of the kernel image.
653 	 * So temporarily mark them as NOMAP to skip mappings in
654 	 * the following for-loop
655 	 */
656 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
657 
658 	/* map all the memory banks */
659 	for_each_mem_range(i, &start, &end) {
660 		if (start >= end)
661 			break;
662 		/*
663 		 * The linear map must allow allocation tags reading/writing
664 		 * if MTE is present. Otherwise, it has the same attributes as
665 		 * PAGE_KERNEL.
666 		 */
667 		__map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
668 			       flags);
669 	}
670 
671 	/*
672 	 * Map the linear alias of the [_stext, __init_begin) interval
673 	 * as non-executable now, and remove the write permission in
674 	 * mark_linear_text_alias_ro() below (which will be called after
675 	 * alternative patching has completed). This makes the contents
676 	 * of the region accessible to subsystems such as hibernate,
677 	 * but protects it from inadvertent modification or execution.
678 	 * Note that contiguous mappings cannot be remapped in this way,
679 	 * so we should avoid them here.
680 	 */
681 	__map_memblock(pgdp, kernel_start, kernel_end,
682 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
683 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
684 	arm64_kfence_map_pool(early_kfence_pool, pgdp);
685 }
686 
687 void mark_rodata_ro(void)
688 {
689 	unsigned long section_size;
690 
691 	/*
692 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
693 	 * to cover NOTES and EXCEPTION_TABLE.
694 	 */
695 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
696 	WRITE_ONCE(rodata_is_rw, false);
697 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
698 			    section_size, PAGE_KERNEL_RO);
699 }
700 
701 static void __init declare_vma(struct vm_struct *vma,
702 			       void *va_start, void *va_end,
703 			       unsigned long vm_flags)
704 {
705 	phys_addr_t pa_start = __pa_symbol(va_start);
706 	unsigned long size = va_end - va_start;
707 
708 	BUG_ON(!PAGE_ALIGNED(pa_start));
709 	BUG_ON(!PAGE_ALIGNED(size));
710 
711 	if (!(vm_flags & VM_NO_GUARD))
712 		size += PAGE_SIZE;
713 
714 	vma->addr	= va_start;
715 	vma->phys_addr	= pa_start;
716 	vma->size	= size;
717 	vma->flags	= VM_MAP | vm_flags;
718 	vma->caller	= __builtin_return_address(0);
719 
720 	vm_area_add_early(vma);
721 }
722 
723 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
724 static pgprot_t kernel_exec_prot(void)
725 {
726 	return rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
727 }
728 
729 static int __init map_entry_trampoline(void)
730 {
731 	int i;
732 
733 	if (!arm64_kernel_unmapped_at_el0())
734 		return 0;
735 
736 	pgprot_t prot = kernel_exec_prot();
737 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
738 
739 	/* The trampoline is always mapped and can therefore be global */
740 	pgprot_val(prot) &= ~PTE_NG;
741 
742 	/* Map only the text into the trampoline page table */
743 	memset(tramp_pg_dir, 0, PGD_SIZE);
744 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
745 			     entry_tramp_text_size(), prot,
746 			     __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
747 
748 	/* Map both the text and data into the kernel page table */
749 	for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
750 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
751 			     pa_start + i * PAGE_SIZE, prot);
752 
753 	if (IS_ENABLED(CONFIG_RELOCATABLE))
754 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
755 			     pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
756 
757 	return 0;
758 }
759 core_initcall(map_entry_trampoline);
760 #endif
761 
762 /*
763  * Declare the VMA areas for the kernel
764  */
765 static void __init declare_kernel_vmas(void)
766 {
767 	static struct vm_struct vmlinux_seg[KERNEL_SEGMENT_COUNT];
768 
769 	declare_vma(&vmlinux_seg[0], _stext, _etext, VM_NO_GUARD);
770 	declare_vma(&vmlinux_seg[1], __start_rodata, __inittext_begin, VM_NO_GUARD);
771 	declare_vma(&vmlinux_seg[2], __inittext_begin, __inittext_end, VM_NO_GUARD);
772 	declare_vma(&vmlinux_seg[3], __initdata_begin, __initdata_end, VM_NO_GUARD);
773 	declare_vma(&vmlinux_seg[4], _data, _end, 0);
774 }
775 
776 void __pi_map_range(u64 *pgd, u64 start, u64 end, u64 pa, pgprot_t prot,
777 		    int level, pte_t *tbl, bool may_use_cont, u64 va_offset);
778 
779 static u8 idmap_ptes[IDMAP_LEVELS - 1][PAGE_SIZE] __aligned(PAGE_SIZE) __ro_after_init,
780 	  kpti_ptes[IDMAP_LEVELS - 1][PAGE_SIZE] __aligned(PAGE_SIZE) __ro_after_init;
781 
782 static void __init create_idmap(void)
783 {
784 	u64 start = __pa_symbol(__idmap_text_start);
785 	u64 end   = __pa_symbol(__idmap_text_end);
786 	u64 ptep  = __pa_symbol(idmap_ptes);
787 
788 	__pi_map_range(&ptep, start, end, start, PAGE_KERNEL_ROX,
789 		       IDMAP_ROOT_LEVEL, (pte_t *)idmap_pg_dir, false,
790 		       __phys_to_virt(ptep) - ptep);
791 
792 	if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0) && !arm64_use_ng_mappings) {
793 		extern u32 __idmap_kpti_flag;
794 		u64 pa = __pa_symbol(&__idmap_kpti_flag);
795 
796 		/*
797 		 * The KPTI G-to-nG conversion code needs a read-write mapping
798 		 * of its synchronization flag in the ID map.
799 		 */
800 		ptep = __pa_symbol(kpti_ptes);
801 		__pi_map_range(&ptep, pa, pa + sizeof(u32), pa, PAGE_KERNEL,
802 			       IDMAP_ROOT_LEVEL, (pte_t *)idmap_pg_dir, false,
803 			       __phys_to_virt(ptep) - ptep);
804 	}
805 }
806 
807 void __init paging_init(void)
808 {
809 	map_mem(swapper_pg_dir);
810 
811 	memblock_allow_resize();
812 
813 	create_idmap();
814 	declare_kernel_vmas();
815 }
816 
817 #ifdef CONFIG_MEMORY_HOTPLUG
818 static void free_hotplug_page_range(struct page *page, size_t size,
819 				    struct vmem_altmap *altmap)
820 {
821 	if (altmap) {
822 		vmem_altmap_free(altmap, size >> PAGE_SHIFT);
823 	} else {
824 		WARN_ON(PageReserved(page));
825 		free_pages((unsigned long)page_address(page), get_order(size));
826 	}
827 }
828 
829 static void free_hotplug_pgtable_page(struct page *page)
830 {
831 	free_hotplug_page_range(page, PAGE_SIZE, NULL);
832 }
833 
834 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
835 				  unsigned long floor, unsigned long ceiling,
836 				  unsigned long mask)
837 {
838 	start &= mask;
839 	if (start < floor)
840 		return false;
841 
842 	if (ceiling) {
843 		ceiling &= mask;
844 		if (!ceiling)
845 			return false;
846 	}
847 
848 	if (end - 1 > ceiling - 1)
849 		return false;
850 	return true;
851 }
852 
853 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
854 				    unsigned long end, bool free_mapped,
855 				    struct vmem_altmap *altmap)
856 {
857 	pte_t *ptep, pte;
858 
859 	do {
860 		ptep = pte_offset_kernel(pmdp, addr);
861 		pte = __ptep_get(ptep);
862 		if (pte_none(pte))
863 			continue;
864 
865 		WARN_ON(!pte_present(pte));
866 		__pte_clear(&init_mm, addr, ptep);
867 		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
868 		if (free_mapped)
869 			free_hotplug_page_range(pte_page(pte),
870 						PAGE_SIZE, altmap);
871 	} while (addr += PAGE_SIZE, addr < end);
872 }
873 
874 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
875 				    unsigned long end, bool free_mapped,
876 				    struct vmem_altmap *altmap)
877 {
878 	unsigned long next;
879 	pmd_t *pmdp, pmd;
880 
881 	do {
882 		next = pmd_addr_end(addr, end);
883 		pmdp = pmd_offset(pudp, addr);
884 		pmd = READ_ONCE(*pmdp);
885 		if (pmd_none(pmd))
886 			continue;
887 
888 		WARN_ON(!pmd_present(pmd));
889 		if (pmd_sect(pmd)) {
890 			pmd_clear(pmdp);
891 
892 			/*
893 			 * One TLBI should be sufficient here as the PMD_SIZE
894 			 * range is mapped with a single block entry.
895 			 */
896 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
897 			if (free_mapped)
898 				free_hotplug_page_range(pmd_page(pmd),
899 							PMD_SIZE, altmap);
900 			continue;
901 		}
902 		WARN_ON(!pmd_table(pmd));
903 		unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
904 	} while (addr = next, addr < end);
905 }
906 
907 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
908 				    unsigned long end, bool free_mapped,
909 				    struct vmem_altmap *altmap)
910 {
911 	unsigned long next;
912 	pud_t *pudp, pud;
913 
914 	do {
915 		next = pud_addr_end(addr, end);
916 		pudp = pud_offset(p4dp, addr);
917 		pud = READ_ONCE(*pudp);
918 		if (pud_none(pud))
919 			continue;
920 
921 		WARN_ON(!pud_present(pud));
922 		if (pud_sect(pud)) {
923 			pud_clear(pudp);
924 
925 			/*
926 			 * One TLBI should be sufficient here as the PUD_SIZE
927 			 * range is mapped with a single block entry.
928 			 */
929 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
930 			if (free_mapped)
931 				free_hotplug_page_range(pud_page(pud),
932 							PUD_SIZE, altmap);
933 			continue;
934 		}
935 		WARN_ON(!pud_table(pud));
936 		unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
937 	} while (addr = next, addr < end);
938 }
939 
940 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
941 				    unsigned long end, bool free_mapped,
942 				    struct vmem_altmap *altmap)
943 {
944 	unsigned long next;
945 	p4d_t *p4dp, p4d;
946 
947 	do {
948 		next = p4d_addr_end(addr, end);
949 		p4dp = p4d_offset(pgdp, addr);
950 		p4d = READ_ONCE(*p4dp);
951 		if (p4d_none(p4d))
952 			continue;
953 
954 		WARN_ON(!p4d_present(p4d));
955 		unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
956 	} while (addr = next, addr < end);
957 }
958 
959 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
960 				bool free_mapped, struct vmem_altmap *altmap)
961 {
962 	unsigned long next;
963 	pgd_t *pgdp, pgd;
964 
965 	/*
966 	 * altmap can only be used as vmemmap mapping backing memory.
967 	 * In case the backing memory itself is not being freed, then
968 	 * altmap is irrelevant. Warn about this inconsistency when
969 	 * encountered.
970 	 */
971 	WARN_ON(!free_mapped && altmap);
972 
973 	do {
974 		next = pgd_addr_end(addr, end);
975 		pgdp = pgd_offset_k(addr);
976 		pgd = READ_ONCE(*pgdp);
977 		if (pgd_none(pgd))
978 			continue;
979 
980 		WARN_ON(!pgd_present(pgd));
981 		unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
982 	} while (addr = next, addr < end);
983 }
984 
985 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
986 				 unsigned long end, unsigned long floor,
987 				 unsigned long ceiling)
988 {
989 	pte_t *ptep, pte;
990 	unsigned long i, start = addr;
991 
992 	do {
993 		ptep = pte_offset_kernel(pmdp, addr);
994 		pte = __ptep_get(ptep);
995 
996 		/*
997 		 * This is just a sanity check here which verifies that
998 		 * pte clearing has been done by earlier unmap loops.
999 		 */
1000 		WARN_ON(!pte_none(pte));
1001 	} while (addr += PAGE_SIZE, addr < end);
1002 
1003 	if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
1004 		return;
1005 
1006 	/*
1007 	 * Check whether we can free the pte page if the rest of the
1008 	 * entries are empty. Overlap with other regions have been
1009 	 * handled by the floor/ceiling check.
1010 	 */
1011 	ptep = pte_offset_kernel(pmdp, 0UL);
1012 	for (i = 0; i < PTRS_PER_PTE; i++) {
1013 		if (!pte_none(__ptep_get(&ptep[i])))
1014 			return;
1015 	}
1016 
1017 	pmd_clear(pmdp);
1018 	__flush_tlb_kernel_pgtable(start);
1019 	free_hotplug_pgtable_page(virt_to_page(ptep));
1020 }
1021 
1022 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1023 				 unsigned long end, unsigned long floor,
1024 				 unsigned long ceiling)
1025 {
1026 	pmd_t *pmdp, pmd;
1027 	unsigned long i, next, start = addr;
1028 
1029 	do {
1030 		next = pmd_addr_end(addr, end);
1031 		pmdp = pmd_offset(pudp, addr);
1032 		pmd = READ_ONCE(*pmdp);
1033 		if (pmd_none(pmd))
1034 			continue;
1035 
1036 		WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1037 		free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1038 	} while (addr = next, addr < end);
1039 
1040 	if (CONFIG_PGTABLE_LEVELS <= 2)
1041 		return;
1042 
1043 	if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1044 		return;
1045 
1046 	/*
1047 	 * Check whether we can free the pmd page if the rest of the
1048 	 * entries are empty. Overlap with other regions have been
1049 	 * handled by the floor/ceiling check.
1050 	 */
1051 	pmdp = pmd_offset(pudp, 0UL);
1052 	for (i = 0; i < PTRS_PER_PMD; i++) {
1053 		if (!pmd_none(READ_ONCE(pmdp[i])))
1054 			return;
1055 	}
1056 
1057 	pud_clear(pudp);
1058 	__flush_tlb_kernel_pgtable(start);
1059 	free_hotplug_pgtable_page(virt_to_page(pmdp));
1060 }
1061 
1062 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1063 				 unsigned long end, unsigned long floor,
1064 				 unsigned long ceiling)
1065 {
1066 	pud_t *pudp, pud;
1067 	unsigned long i, next, start = addr;
1068 
1069 	do {
1070 		next = pud_addr_end(addr, end);
1071 		pudp = pud_offset(p4dp, addr);
1072 		pud = READ_ONCE(*pudp);
1073 		if (pud_none(pud))
1074 			continue;
1075 
1076 		WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1077 		free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1078 	} while (addr = next, addr < end);
1079 
1080 	if (!pgtable_l4_enabled())
1081 		return;
1082 
1083 	if (!pgtable_range_aligned(start, end, floor, ceiling, P4D_MASK))
1084 		return;
1085 
1086 	/*
1087 	 * Check whether we can free the pud page if the rest of the
1088 	 * entries are empty. Overlap with other regions have been
1089 	 * handled by the floor/ceiling check.
1090 	 */
1091 	pudp = pud_offset(p4dp, 0UL);
1092 	for (i = 0; i < PTRS_PER_PUD; i++) {
1093 		if (!pud_none(READ_ONCE(pudp[i])))
1094 			return;
1095 	}
1096 
1097 	p4d_clear(p4dp);
1098 	__flush_tlb_kernel_pgtable(start);
1099 	free_hotplug_pgtable_page(virt_to_page(pudp));
1100 }
1101 
1102 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1103 				 unsigned long end, unsigned long floor,
1104 				 unsigned long ceiling)
1105 {
1106 	p4d_t *p4dp, p4d;
1107 	unsigned long i, next, start = addr;
1108 
1109 	do {
1110 		next = p4d_addr_end(addr, end);
1111 		p4dp = p4d_offset(pgdp, addr);
1112 		p4d = READ_ONCE(*p4dp);
1113 		if (p4d_none(p4d))
1114 			continue;
1115 
1116 		WARN_ON(!p4d_present(p4d));
1117 		free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1118 	} while (addr = next, addr < end);
1119 
1120 	if (!pgtable_l5_enabled())
1121 		return;
1122 
1123 	if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1124 		return;
1125 
1126 	/*
1127 	 * Check whether we can free the p4d page if the rest of the
1128 	 * entries are empty. Overlap with other regions have been
1129 	 * handled by the floor/ceiling check.
1130 	 */
1131 	p4dp = p4d_offset(pgdp, 0UL);
1132 	for (i = 0; i < PTRS_PER_P4D; i++) {
1133 		if (!p4d_none(READ_ONCE(p4dp[i])))
1134 			return;
1135 	}
1136 
1137 	pgd_clear(pgdp);
1138 	__flush_tlb_kernel_pgtable(start);
1139 	free_hotplug_pgtable_page(virt_to_page(p4dp));
1140 }
1141 
1142 static void free_empty_tables(unsigned long addr, unsigned long end,
1143 			      unsigned long floor, unsigned long ceiling)
1144 {
1145 	unsigned long next;
1146 	pgd_t *pgdp, pgd;
1147 
1148 	do {
1149 		next = pgd_addr_end(addr, end);
1150 		pgdp = pgd_offset_k(addr);
1151 		pgd = READ_ONCE(*pgdp);
1152 		if (pgd_none(pgd))
1153 			continue;
1154 
1155 		WARN_ON(!pgd_present(pgd));
1156 		free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1157 	} while (addr = next, addr < end);
1158 }
1159 #endif
1160 
1161 void __meminit vmemmap_set_pmd(pmd_t *pmdp, void *p, int node,
1162 			       unsigned long addr, unsigned long next)
1163 {
1164 	pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1165 }
1166 
1167 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1168 				unsigned long addr, unsigned long next)
1169 {
1170 	vmemmap_verify((pte_t *)pmdp, node, addr, next);
1171 	return 1;
1172 }
1173 
1174 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1175 		struct vmem_altmap *altmap)
1176 {
1177 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1178 
1179 	if (!IS_ENABLED(CONFIG_ARM64_4K_PAGES))
1180 		return vmemmap_populate_basepages(start, end, node, altmap);
1181 	else
1182 		return vmemmap_populate_hugepages(start, end, node, altmap);
1183 }
1184 
1185 #ifdef CONFIG_MEMORY_HOTPLUG
1186 void vmemmap_free(unsigned long start, unsigned long end,
1187 		struct vmem_altmap *altmap)
1188 {
1189 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1190 
1191 	unmap_hotplug_range(start, end, true, altmap);
1192 	free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1193 }
1194 #endif /* CONFIG_MEMORY_HOTPLUG */
1195 
1196 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1197 {
1198 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1199 
1200 	/* Only allow permission changes for now */
1201 	if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1202 				   pud_val(new_pud)))
1203 		return 0;
1204 
1205 	VM_BUG_ON(phys & ~PUD_MASK);
1206 	set_pud(pudp, new_pud);
1207 	return 1;
1208 }
1209 
1210 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1211 {
1212 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1213 
1214 	/* Only allow permission changes for now */
1215 	if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1216 				   pmd_val(new_pmd)))
1217 		return 0;
1218 
1219 	VM_BUG_ON(phys & ~PMD_MASK);
1220 	set_pmd(pmdp, new_pmd);
1221 	return 1;
1222 }
1223 
1224 #ifndef __PAGETABLE_P4D_FOLDED
1225 void p4d_clear_huge(p4d_t *p4dp)
1226 {
1227 }
1228 #endif
1229 
1230 int pud_clear_huge(pud_t *pudp)
1231 {
1232 	if (!pud_sect(READ_ONCE(*pudp)))
1233 		return 0;
1234 	pud_clear(pudp);
1235 	return 1;
1236 }
1237 
1238 int pmd_clear_huge(pmd_t *pmdp)
1239 {
1240 	if (!pmd_sect(READ_ONCE(*pmdp)))
1241 		return 0;
1242 	pmd_clear(pmdp);
1243 	return 1;
1244 }
1245 
1246 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1247 {
1248 	pte_t *table;
1249 	pmd_t pmd;
1250 
1251 	pmd = READ_ONCE(*pmdp);
1252 
1253 	if (!pmd_table(pmd)) {
1254 		VM_WARN_ON(1);
1255 		return 1;
1256 	}
1257 
1258 	table = pte_offset_kernel(pmdp, addr);
1259 	pmd_clear(pmdp);
1260 	__flush_tlb_kernel_pgtable(addr);
1261 	pte_free_kernel(NULL, table);
1262 	return 1;
1263 }
1264 
1265 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1266 {
1267 	pmd_t *table;
1268 	pmd_t *pmdp;
1269 	pud_t pud;
1270 	unsigned long next, end;
1271 
1272 	pud = READ_ONCE(*pudp);
1273 
1274 	if (!pud_table(pud)) {
1275 		VM_WARN_ON(1);
1276 		return 1;
1277 	}
1278 
1279 	table = pmd_offset(pudp, addr);
1280 	pmdp = table;
1281 	next = addr;
1282 	end = addr + PUD_SIZE;
1283 	do {
1284 		pmd_free_pte_page(pmdp, next);
1285 	} while (pmdp++, next += PMD_SIZE, next != end);
1286 
1287 	pud_clear(pudp);
1288 	__flush_tlb_kernel_pgtable(addr);
1289 	pmd_free(NULL, table);
1290 	return 1;
1291 }
1292 
1293 #ifdef CONFIG_MEMORY_HOTPLUG
1294 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1295 {
1296 	unsigned long end = start + size;
1297 
1298 	WARN_ON(pgdir != init_mm.pgd);
1299 	WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1300 
1301 	unmap_hotplug_range(start, end, false, NULL);
1302 	free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1303 }
1304 
1305 struct range arch_get_mappable_range(void)
1306 {
1307 	struct range mhp_range;
1308 	u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1309 	u64 end_linear_pa = __pa(PAGE_END - 1);
1310 
1311 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1312 		/*
1313 		 * Check for a wrap, it is possible because of randomized linear
1314 		 * mapping the start physical address is actually bigger than
1315 		 * the end physical address. In this case set start to zero
1316 		 * because [0, end_linear_pa] range must still be able to cover
1317 		 * all addressable physical addresses.
1318 		 */
1319 		if (start_linear_pa > end_linear_pa)
1320 			start_linear_pa = 0;
1321 	}
1322 
1323 	WARN_ON(start_linear_pa > end_linear_pa);
1324 
1325 	/*
1326 	 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1327 	 * accommodating both its ends but excluding PAGE_END. Max physical
1328 	 * range which can be mapped inside this linear mapping range, must
1329 	 * also be derived from its end points.
1330 	 */
1331 	mhp_range.start = start_linear_pa;
1332 	mhp_range.end =  end_linear_pa;
1333 
1334 	return mhp_range;
1335 }
1336 
1337 int arch_add_memory(int nid, u64 start, u64 size,
1338 		    struct mhp_params *params)
1339 {
1340 	int ret, flags = NO_EXEC_MAPPINGS;
1341 
1342 	VM_BUG_ON(!mhp_range_allowed(start, size, true));
1343 
1344 	if (can_set_direct_map())
1345 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1346 
1347 	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1348 			     size, params->pgprot, __pgd_pgtable_alloc,
1349 			     flags);
1350 
1351 	memblock_clear_nomap(start, size);
1352 
1353 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1354 			   params);
1355 	if (ret)
1356 		__remove_pgd_mapping(swapper_pg_dir,
1357 				     __phys_to_virt(start), size);
1358 	else {
1359 		max_pfn = PFN_UP(start + size);
1360 		max_low_pfn = max_pfn;
1361 	}
1362 
1363 	return ret;
1364 }
1365 
1366 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1367 {
1368 	unsigned long start_pfn = start >> PAGE_SHIFT;
1369 	unsigned long nr_pages = size >> PAGE_SHIFT;
1370 
1371 	__remove_pages(start_pfn, nr_pages, altmap);
1372 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1373 }
1374 
1375 /*
1376  * This memory hotplug notifier helps prevent boot memory from being
1377  * inadvertently removed as it blocks pfn range offlining process in
1378  * __offline_pages(). Hence this prevents both offlining as well as
1379  * removal process for boot memory which is initially always online.
1380  * In future if and when boot memory could be removed, this notifier
1381  * should be dropped and free_hotplug_page_range() should handle any
1382  * reserved pages allocated during boot.
1383  */
1384 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1385 					   unsigned long action, void *data)
1386 {
1387 	struct mem_section *ms;
1388 	struct memory_notify *arg = data;
1389 	unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1390 	unsigned long pfn = arg->start_pfn;
1391 
1392 	if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1393 		return NOTIFY_OK;
1394 
1395 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1396 		unsigned long start = PFN_PHYS(pfn);
1397 		unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1398 
1399 		ms = __pfn_to_section(pfn);
1400 		if (!early_section(ms))
1401 			continue;
1402 
1403 		if (action == MEM_GOING_OFFLINE) {
1404 			/*
1405 			 * Boot memory removal is not supported. Prevent
1406 			 * it via blocking any attempted offline request
1407 			 * for the boot memory and just report it.
1408 			 */
1409 			pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1410 			return NOTIFY_BAD;
1411 		} else if (action == MEM_OFFLINE) {
1412 			/*
1413 			 * This should have never happened. Boot memory
1414 			 * offlining should have been prevented by this
1415 			 * very notifier. Probably some memory removal
1416 			 * procedure might have changed which would then
1417 			 * require further debug.
1418 			 */
1419 			pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1420 
1421 			/*
1422 			 * Core memory hotplug does not process a return
1423 			 * code from the notifier for MEM_OFFLINE events.
1424 			 * The error condition has been reported. Return
1425 			 * from here as if ignored.
1426 			 */
1427 			return NOTIFY_DONE;
1428 		}
1429 	}
1430 	return NOTIFY_OK;
1431 }
1432 
1433 static struct notifier_block prevent_bootmem_remove_nb = {
1434 	.notifier_call = prevent_bootmem_remove_notifier,
1435 };
1436 
1437 /*
1438  * This ensures that boot memory sections on the platform are online
1439  * from early boot. Memory sections could not be prevented from being
1440  * offlined, unless for some reason they are not online to begin with.
1441  * This helps validate the basic assumption on which the above memory
1442  * event notifier works to prevent boot memory section offlining and
1443  * its possible removal.
1444  */
1445 static void validate_bootmem_online(void)
1446 {
1447 	phys_addr_t start, end, addr;
1448 	struct mem_section *ms;
1449 	u64 i;
1450 
1451 	/*
1452 	 * Scanning across all memblock might be expensive
1453 	 * on some big memory systems. Hence enable this
1454 	 * validation only with DEBUG_VM.
1455 	 */
1456 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
1457 		return;
1458 
1459 	for_each_mem_range(i, &start, &end) {
1460 		for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1461 			ms = __pfn_to_section(PHYS_PFN(addr));
1462 
1463 			/*
1464 			 * All memory ranges in the system at this point
1465 			 * should have been marked as early sections.
1466 			 */
1467 			WARN_ON(!early_section(ms));
1468 
1469 			/*
1470 			 * Memory notifier mechanism here to prevent boot
1471 			 * memory offlining depends on the fact that each
1472 			 * early section memory on the system is initially
1473 			 * online. Otherwise a given memory section which
1474 			 * is already offline will be overlooked and can
1475 			 * be removed completely. Call out such sections.
1476 			 */
1477 			if (!online_section(ms))
1478 				pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1479 					addr, addr + (1UL << PA_SECTION_SHIFT));
1480 		}
1481 	}
1482 }
1483 
1484 static int __init prevent_bootmem_remove_init(void)
1485 {
1486 	int ret = 0;
1487 
1488 	if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1489 		return ret;
1490 
1491 	validate_bootmem_online();
1492 	ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1493 	if (ret)
1494 		pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1495 
1496 	return ret;
1497 }
1498 early_initcall(prevent_bootmem_remove_init);
1499 #endif
1500 
1501 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
1502 {
1503 	if (alternative_has_cap_unlikely(ARM64_WORKAROUND_2645198)) {
1504 		/*
1505 		 * Break-before-make (BBM) is required for all user space mappings
1506 		 * when the permission changes from executable to non-executable
1507 		 * in cases where cpu is affected with errata #2645198.
1508 		 */
1509 		if (pte_user_exec(ptep_get(ptep)))
1510 			return ptep_clear_flush(vma, addr, ptep);
1511 	}
1512 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1513 }
1514 
1515 void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1516 			     pte_t old_pte, pte_t pte)
1517 {
1518 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1519 }
1520 
1521 /*
1522  * Atomically replaces the active TTBR1_EL1 PGD with a new VA-compatible PGD,
1523  * avoiding the possibility of conflicting TLB entries being allocated.
1524  */
1525 void __cpu_replace_ttbr1(pgd_t *pgdp, bool cnp)
1526 {
1527 	typedef void (ttbr_replace_func)(phys_addr_t);
1528 	extern ttbr_replace_func idmap_cpu_replace_ttbr1;
1529 	ttbr_replace_func *replace_phys;
1530 	unsigned long daif;
1531 
1532 	/* phys_to_ttbr() zeros lower 2 bits of ttbr with 52-bit PA */
1533 	phys_addr_t ttbr1 = phys_to_ttbr(virt_to_phys(pgdp));
1534 
1535 	if (cnp)
1536 		ttbr1 |= TTBR_CNP_BIT;
1537 
1538 	replace_phys = (void *)__pa_symbol(idmap_cpu_replace_ttbr1);
1539 
1540 	cpu_install_idmap();
1541 
1542 	/*
1543 	 * We really don't want to take *any* exceptions while TTBR1 is
1544 	 * in the process of being replaced so mask everything.
1545 	 */
1546 	daif = local_daif_save();
1547 	replace_phys(ttbr1);
1548 	local_daif_restore(daif);
1549 
1550 	cpu_uninstall_idmap();
1551 }
1552