xref: /openbmc/linux/arch/arm64/mm/mmu.c (revision e65e175b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Based on arch/arm/mm/mmu.c
4  *
5  * Copyright (C) 1995-2005 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  */
8 
9 #include <linux/cache.h>
10 #include <linux/export.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/ioport.h>
15 #include <linux/kexec.h>
16 #include <linux/libfdt.h>
17 #include <linux/mman.h>
18 #include <linux/nodemask.h>
19 #include <linux/memblock.h>
20 #include <linux/memremap.h>
21 #include <linux/memory.h>
22 #include <linux/fs.h>
23 #include <linux/io.h>
24 #include <linux/mm.h>
25 #include <linux/vmalloc.h>
26 #include <linux/set_memory.h>
27 
28 #include <asm/barrier.h>
29 #include <asm/cputype.h>
30 #include <asm/fixmap.h>
31 #include <asm/kasan.h>
32 #include <asm/kernel-pgtable.h>
33 #include <asm/sections.h>
34 #include <asm/setup.h>
35 #include <linux/sizes.h>
36 #include <asm/tlb.h>
37 #include <asm/mmu_context.h>
38 #include <asm/ptdump.h>
39 #include <asm/tlbflush.h>
40 #include <asm/pgalloc.h>
41 
42 #define NO_BLOCK_MAPPINGS	BIT(0)
43 #define NO_CONT_MAPPINGS	BIT(1)
44 #define NO_EXEC_MAPPINGS	BIT(2)	/* assumes FEAT_HPDS is not used */
45 
46 int idmap_t0sz __ro_after_init;
47 
48 #if VA_BITS > 48
49 u64 vabits_actual __ro_after_init = VA_BITS_MIN;
50 EXPORT_SYMBOL(vabits_actual);
51 #endif
52 
53 u64 kimage_vaddr __ro_after_init = (u64)&_text;
54 EXPORT_SYMBOL(kimage_vaddr);
55 
56 u64 kimage_voffset __ro_after_init;
57 EXPORT_SYMBOL(kimage_voffset);
58 
59 u32 __boot_cpu_mode[] = { BOOT_CPU_MODE_EL2, BOOT_CPU_MODE_EL1 };
60 
61 /*
62  * The booting CPU updates the failed status @__early_cpu_boot_status,
63  * with MMU turned off.
64  */
65 long __section(".mmuoff.data.write") __early_cpu_boot_status;
66 
67 /*
68  * Empty_zero_page is a special page that is used for zero-initialized data
69  * and COW.
70  */
71 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
72 EXPORT_SYMBOL(empty_zero_page);
73 
74 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
75 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
76 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
77 
78 static DEFINE_SPINLOCK(swapper_pgdir_lock);
79 static DEFINE_MUTEX(fixmap_lock);
80 
81 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
82 {
83 	pgd_t *fixmap_pgdp;
84 
85 	spin_lock(&swapper_pgdir_lock);
86 	fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
87 	WRITE_ONCE(*fixmap_pgdp, pgd);
88 	/*
89 	 * We need dsb(ishst) here to ensure the page-table-walker sees
90 	 * our new entry before set_p?d() returns. The fixmap's
91 	 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
92 	 */
93 	pgd_clear_fixmap();
94 	spin_unlock(&swapper_pgdir_lock);
95 }
96 
97 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
98 			      unsigned long size, pgprot_t vma_prot)
99 {
100 	if (!pfn_is_map_memory(pfn))
101 		return pgprot_noncached(vma_prot);
102 	else if (file->f_flags & O_SYNC)
103 		return pgprot_writecombine(vma_prot);
104 	return vma_prot;
105 }
106 EXPORT_SYMBOL(phys_mem_access_prot);
107 
108 static phys_addr_t __init early_pgtable_alloc(int shift)
109 {
110 	phys_addr_t phys;
111 	void *ptr;
112 
113 	phys = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0,
114 					 MEMBLOCK_ALLOC_NOLEAKTRACE);
115 	if (!phys)
116 		panic("Failed to allocate page table page\n");
117 
118 	/*
119 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
120 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
121 	 * any level of table.
122 	 */
123 	ptr = pte_set_fixmap(phys);
124 
125 	memset(ptr, 0, PAGE_SIZE);
126 
127 	/*
128 	 * Implicit barriers also ensure the zeroed page is visible to the page
129 	 * table walker
130 	 */
131 	pte_clear_fixmap();
132 
133 	return phys;
134 }
135 
136 static bool pgattr_change_is_safe(u64 old, u64 new)
137 {
138 	/*
139 	 * The following mapping attributes may be updated in live
140 	 * kernel mappings without the need for break-before-make.
141 	 */
142 	pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
143 
144 	/* creating or taking down mappings is always safe */
145 	if (old == 0 || new == 0)
146 		return true;
147 
148 	/* live contiguous mappings may not be manipulated at all */
149 	if ((old | new) & PTE_CONT)
150 		return false;
151 
152 	/* Transitioning from Non-Global to Global is unsafe */
153 	if (old & ~new & PTE_NG)
154 		return false;
155 
156 	/*
157 	 * Changing the memory type between Normal and Normal-Tagged is safe
158 	 * since Tagged is considered a permission attribute from the
159 	 * mismatched attribute aliases perspective.
160 	 */
161 	if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
162 	     (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
163 	    ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
164 	     (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
165 		mask |= PTE_ATTRINDX_MASK;
166 
167 	return ((old ^ new) & ~mask) == 0;
168 }
169 
170 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
171 		     phys_addr_t phys, pgprot_t prot)
172 {
173 	pte_t *ptep;
174 
175 	ptep = pte_set_fixmap_offset(pmdp, addr);
176 	do {
177 		pte_t old_pte = READ_ONCE(*ptep);
178 
179 		set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
180 
181 		/*
182 		 * After the PTE entry has been populated once, we
183 		 * only allow updates to the permission attributes.
184 		 */
185 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
186 					      READ_ONCE(pte_val(*ptep))));
187 
188 		phys += PAGE_SIZE;
189 	} while (ptep++, addr += PAGE_SIZE, addr != end);
190 
191 	pte_clear_fixmap();
192 }
193 
194 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
195 				unsigned long end, phys_addr_t phys,
196 				pgprot_t prot,
197 				phys_addr_t (*pgtable_alloc)(int),
198 				int flags)
199 {
200 	unsigned long next;
201 	pmd_t pmd = READ_ONCE(*pmdp);
202 
203 	BUG_ON(pmd_sect(pmd));
204 	if (pmd_none(pmd)) {
205 		pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
206 		phys_addr_t pte_phys;
207 
208 		if (flags & NO_EXEC_MAPPINGS)
209 			pmdval |= PMD_TABLE_PXN;
210 		BUG_ON(!pgtable_alloc);
211 		pte_phys = pgtable_alloc(PAGE_SHIFT);
212 		__pmd_populate(pmdp, pte_phys, pmdval);
213 		pmd = READ_ONCE(*pmdp);
214 	}
215 	BUG_ON(pmd_bad(pmd));
216 
217 	do {
218 		pgprot_t __prot = prot;
219 
220 		next = pte_cont_addr_end(addr, end);
221 
222 		/* use a contiguous mapping if the range is suitably aligned */
223 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
224 		    (flags & NO_CONT_MAPPINGS) == 0)
225 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
226 
227 		init_pte(pmdp, addr, next, phys, __prot);
228 
229 		phys += next - addr;
230 	} while (addr = next, addr != end);
231 }
232 
233 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
234 		     phys_addr_t phys, pgprot_t prot,
235 		     phys_addr_t (*pgtable_alloc)(int), int flags)
236 {
237 	unsigned long next;
238 	pmd_t *pmdp;
239 
240 	pmdp = pmd_set_fixmap_offset(pudp, addr);
241 	do {
242 		pmd_t old_pmd = READ_ONCE(*pmdp);
243 
244 		next = pmd_addr_end(addr, end);
245 
246 		/* try section mapping first */
247 		if (((addr | next | phys) & ~PMD_MASK) == 0 &&
248 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
249 			pmd_set_huge(pmdp, phys, prot);
250 
251 			/*
252 			 * After the PMD entry has been populated once, we
253 			 * only allow updates to the permission attributes.
254 			 */
255 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
256 						      READ_ONCE(pmd_val(*pmdp))));
257 		} else {
258 			alloc_init_cont_pte(pmdp, addr, next, phys, prot,
259 					    pgtable_alloc, flags);
260 
261 			BUG_ON(pmd_val(old_pmd) != 0 &&
262 			       pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
263 		}
264 		phys += next - addr;
265 	} while (pmdp++, addr = next, addr != end);
266 
267 	pmd_clear_fixmap();
268 }
269 
270 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
271 				unsigned long end, phys_addr_t phys,
272 				pgprot_t prot,
273 				phys_addr_t (*pgtable_alloc)(int), int flags)
274 {
275 	unsigned long next;
276 	pud_t pud = READ_ONCE(*pudp);
277 
278 	/*
279 	 * Check for initial section mappings in the pgd/pud.
280 	 */
281 	BUG_ON(pud_sect(pud));
282 	if (pud_none(pud)) {
283 		pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
284 		phys_addr_t pmd_phys;
285 
286 		if (flags & NO_EXEC_MAPPINGS)
287 			pudval |= PUD_TABLE_PXN;
288 		BUG_ON(!pgtable_alloc);
289 		pmd_phys = pgtable_alloc(PMD_SHIFT);
290 		__pud_populate(pudp, pmd_phys, pudval);
291 		pud = READ_ONCE(*pudp);
292 	}
293 	BUG_ON(pud_bad(pud));
294 
295 	do {
296 		pgprot_t __prot = prot;
297 
298 		next = pmd_cont_addr_end(addr, end);
299 
300 		/* use a contiguous mapping if the range is suitably aligned */
301 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
302 		    (flags & NO_CONT_MAPPINGS) == 0)
303 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
304 
305 		init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
306 
307 		phys += next - addr;
308 	} while (addr = next, addr != end);
309 }
310 
311 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
312 			   phys_addr_t phys, pgprot_t prot,
313 			   phys_addr_t (*pgtable_alloc)(int),
314 			   int flags)
315 {
316 	unsigned long next;
317 	pud_t *pudp;
318 	p4d_t *p4dp = p4d_offset(pgdp, addr);
319 	p4d_t p4d = READ_ONCE(*p4dp);
320 
321 	if (p4d_none(p4d)) {
322 		p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
323 		phys_addr_t pud_phys;
324 
325 		if (flags & NO_EXEC_MAPPINGS)
326 			p4dval |= P4D_TABLE_PXN;
327 		BUG_ON(!pgtable_alloc);
328 		pud_phys = pgtable_alloc(PUD_SHIFT);
329 		__p4d_populate(p4dp, pud_phys, p4dval);
330 		p4d = READ_ONCE(*p4dp);
331 	}
332 	BUG_ON(p4d_bad(p4d));
333 
334 	pudp = pud_set_fixmap_offset(p4dp, addr);
335 	do {
336 		pud_t old_pud = READ_ONCE(*pudp);
337 
338 		next = pud_addr_end(addr, end);
339 
340 		/*
341 		 * For 4K granule only, attempt to put down a 1GB block
342 		 */
343 		if (pud_sect_supported() &&
344 		   ((addr | next | phys) & ~PUD_MASK) == 0 &&
345 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
346 			pud_set_huge(pudp, phys, prot);
347 
348 			/*
349 			 * After the PUD entry has been populated once, we
350 			 * only allow updates to the permission attributes.
351 			 */
352 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
353 						      READ_ONCE(pud_val(*pudp))));
354 		} else {
355 			alloc_init_cont_pmd(pudp, addr, next, phys, prot,
356 					    pgtable_alloc, flags);
357 
358 			BUG_ON(pud_val(old_pud) != 0 &&
359 			       pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
360 		}
361 		phys += next - addr;
362 	} while (pudp++, addr = next, addr != end);
363 
364 	pud_clear_fixmap();
365 }
366 
367 static void __create_pgd_mapping_locked(pgd_t *pgdir, phys_addr_t phys,
368 					unsigned long virt, phys_addr_t size,
369 					pgprot_t prot,
370 					phys_addr_t (*pgtable_alloc)(int),
371 					int flags)
372 {
373 	unsigned long addr, end, next;
374 	pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
375 
376 	/*
377 	 * If the virtual and physical address don't have the same offset
378 	 * within a page, we cannot map the region as the caller expects.
379 	 */
380 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
381 		return;
382 
383 	phys &= PAGE_MASK;
384 	addr = virt & PAGE_MASK;
385 	end = PAGE_ALIGN(virt + size);
386 
387 	do {
388 		next = pgd_addr_end(addr, end);
389 		alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
390 			       flags);
391 		phys += next - addr;
392 	} while (pgdp++, addr = next, addr != end);
393 }
394 
395 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
396 				 unsigned long virt, phys_addr_t size,
397 				 pgprot_t prot,
398 				 phys_addr_t (*pgtable_alloc)(int),
399 				 int flags)
400 {
401 	mutex_lock(&fixmap_lock);
402 	__create_pgd_mapping_locked(pgdir, phys, virt, size, prot,
403 				    pgtable_alloc, flags);
404 	mutex_unlock(&fixmap_lock);
405 }
406 
407 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
408 extern __alias(__create_pgd_mapping_locked)
409 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
410 			     phys_addr_t size, pgprot_t prot,
411 			     phys_addr_t (*pgtable_alloc)(int), int flags);
412 #endif
413 
414 static phys_addr_t __pgd_pgtable_alloc(int shift)
415 {
416 	void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
417 	BUG_ON(!ptr);
418 
419 	/* Ensure the zeroed page is visible to the page table walker */
420 	dsb(ishst);
421 	return __pa(ptr);
422 }
423 
424 static phys_addr_t pgd_pgtable_alloc(int shift)
425 {
426 	phys_addr_t pa = __pgd_pgtable_alloc(shift);
427 
428 	/*
429 	 * Call proper page table ctor in case later we need to
430 	 * call core mm functions like apply_to_page_range() on
431 	 * this pre-allocated page table.
432 	 *
433 	 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
434 	 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
435 	 */
436 	if (shift == PAGE_SHIFT)
437 		BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
438 	else if (shift == PMD_SHIFT)
439 		BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
440 
441 	return pa;
442 }
443 
444 /*
445  * This function can only be used to modify existing table entries,
446  * without allocating new levels of table. Note that this permits the
447  * creation of new section or page entries.
448  */
449 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
450 				  phys_addr_t size, pgprot_t prot)
451 {
452 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
453 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
454 			&phys, virt);
455 		return;
456 	}
457 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
458 			     NO_CONT_MAPPINGS);
459 }
460 
461 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
462 			       unsigned long virt, phys_addr_t size,
463 			       pgprot_t prot, bool page_mappings_only)
464 {
465 	int flags = 0;
466 
467 	BUG_ON(mm == &init_mm);
468 
469 	if (page_mappings_only)
470 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
471 
472 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
473 			     pgd_pgtable_alloc, flags);
474 }
475 
476 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
477 				phys_addr_t size, pgprot_t prot)
478 {
479 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
480 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
481 			&phys, virt);
482 		return;
483 	}
484 
485 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
486 			     NO_CONT_MAPPINGS);
487 
488 	/* flush the TLBs after updating live kernel mappings */
489 	flush_tlb_kernel_range(virt, virt + size);
490 }
491 
492 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
493 				  phys_addr_t end, pgprot_t prot, int flags)
494 {
495 	__create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
496 			     prot, early_pgtable_alloc, flags);
497 }
498 
499 void __init mark_linear_text_alias_ro(void)
500 {
501 	/*
502 	 * Remove the write permissions from the linear alias of .text/.rodata
503 	 */
504 	update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
505 			    (unsigned long)__init_begin - (unsigned long)_stext,
506 			    PAGE_KERNEL_RO);
507 }
508 
509 static bool crash_mem_map __initdata;
510 
511 static int __init enable_crash_mem_map(char *arg)
512 {
513 	/*
514 	 * Proper parameter parsing is done by reserve_crashkernel(). We only
515 	 * need to know if the linear map has to avoid block mappings so that
516 	 * the crashkernel reservations can be unmapped later.
517 	 */
518 	crash_mem_map = true;
519 
520 	return 0;
521 }
522 early_param("crashkernel", enable_crash_mem_map);
523 
524 static void __init map_mem(pgd_t *pgdp)
525 {
526 	static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
527 	phys_addr_t kernel_start = __pa_symbol(_stext);
528 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
529 	phys_addr_t start, end;
530 	int flags = NO_EXEC_MAPPINGS;
531 	u64 i;
532 
533 	/*
534 	 * Setting hierarchical PXNTable attributes on table entries covering
535 	 * the linear region is only possible if it is guaranteed that no table
536 	 * entries at any level are being shared between the linear region and
537 	 * the vmalloc region. Check whether this is true for the PGD level, in
538 	 * which case it is guaranteed to be true for all other levels as well.
539 	 */
540 	BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
541 
542 	if (can_set_direct_map())
543 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
544 
545 	/*
546 	 * Take care not to create a writable alias for the
547 	 * read-only text and rodata sections of the kernel image.
548 	 * So temporarily mark them as NOMAP to skip mappings in
549 	 * the following for-loop
550 	 */
551 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
552 
553 #ifdef CONFIG_KEXEC_CORE
554 	if (crash_mem_map) {
555 		if (defer_reserve_crashkernel())
556 			flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
557 		else if (crashk_res.end)
558 			memblock_mark_nomap(crashk_res.start,
559 			    resource_size(&crashk_res));
560 	}
561 #endif
562 
563 	/* map all the memory banks */
564 	for_each_mem_range(i, &start, &end) {
565 		if (start >= end)
566 			break;
567 		/*
568 		 * The linear map must allow allocation tags reading/writing
569 		 * if MTE is present. Otherwise, it has the same attributes as
570 		 * PAGE_KERNEL.
571 		 */
572 		__map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
573 			       flags);
574 	}
575 
576 	/*
577 	 * Map the linear alias of the [_stext, __init_begin) interval
578 	 * as non-executable now, and remove the write permission in
579 	 * mark_linear_text_alias_ro() below (which will be called after
580 	 * alternative patching has completed). This makes the contents
581 	 * of the region accessible to subsystems such as hibernate,
582 	 * but protects it from inadvertent modification or execution.
583 	 * Note that contiguous mappings cannot be remapped in this way,
584 	 * so we should avoid them here.
585 	 */
586 	__map_memblock(pgdp, kernel_start, kernel_end,
587 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
588 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
589 
590 	/*
591 	 * Use page-level mappings here so that we can shrink the region
592 	 * in page granularity and put back unused memory to buddy system
593 	 * through /sys/kernel/kexec_crash_size interface.
594 	 */
595 #ifdef CONFIG_KEXEC_CORE
596 	if (crash_mem_map && !defer_reserve_crashkernel()) {
597 		if (crashk_res.end) {
598 			__map_memblock(pgdp, crashk_res.start,
599 				       crashk_res.end + 1,
600 				       PAGE_KERNEL,
601 				       NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
602 			memblock_clear_nomap(crashk_res.start,
603 					     resource_size(&crashk_res));
604 		}
605 	}
606 #endif
607 }
608 
609 void mark_rodata_ro(void)
610 {
611 	unsigned long section_size;
612 
613 	/*
614 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
615 	 * to cover NOTES and EXCEPTION_TABLE.
616 	 */
617 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
618 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
619 			    section_size, PAGE_KERNEL_RO);
620 
621 	debug_checkwx();
622 }
623 
624 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
625 				      pgprot_t prot, struct vm_struct *vma,
626 				      int flags, unsigned long vm_flags)
627 {
628 	phys_addr_t pa_start = __pa_symbol(va_start);
629 	unsigned long size = va_end - va_start;
630 
631 	BUG_ON(!PAGE_ALIGNED(pa_start));
632 	BUG_ON(!PAGE_ALIGNED(size));
633 
634 	__create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
635 			     early_pgtable_alloc, flags);
636 
637 	if (!(vm_flags & VM_NO_GUARD))
638 		size += PAGE_SIZE;
639 
640 	vma->addr	= va_start;
641 	vma->phys_addr	= pa_start;
642 	vma->size	= size;
643 	vma->flags	= VM_MAP | vm_flags;
644 	vma->caller	= __builtin_return_address(0);
645 
646 	vm_area_add_early(vma);
647 }
648 
649 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
650 static int __init map_entry_trampoline(void)
651 {
652 	int i;
653 
654 	pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
655 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
656 
657 	/* The trampoline is always mapped and can therefore be global */
658 	pgprot_val(prot) &= ~PTE_NG;
659 
660 	/* Map only the text into the trampoline page table */
661 	memset(tramp_pg_dir, 0, PGD_SIZE);
662 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
663 			     entry_tramp_text_size(), prot,
664 			     __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
665 
666 	/* Map both the text and data into the kernel page table */
667 	for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
668 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
669 			     pa_start + i * PAGE_SIZE, prot);
670 
671 	if (IS_ENABLED(CONFIG_RELOCATABLE))
672 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
673 			     pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
674 
675 	return 0;
676 }
677 core_initcall(map_entry_trampoline);
678 #endif
679 
680 /*
681  * Open coded check for BTI, only for use to determine configuration
682  * for early mappings for before the cpufeature code has run.
683  */
684 static bool arm64_early_this_cpu_has_bti(void)
685 {
686 	u64 pfr1;
687 
688 	if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
689 		return false;
690 
691 	pfr1 = __read_sysreg_by_encoding(SYS_ID_AA64PFR1_EL1);
692 	return cpuid_feature_extract_unsigned_field(pfr1,
693 						    ID_AA64PFR1_EL1_BT_SHIFT);
694 }
695 
696 /*
697  * Create fine-grained mappings for the kernel.
698  */
699 static void __init map_kernel(pgd_t *pgdp)
700 {
701 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
702 				vmlinux_initdata, vmlinux_data;
703 
704 	/*
705 	 * External debuggers may need to write directly to the text
706 	 * mapping to install SW breakpoints. Allow this (only) when
707 	 * explicitly requested with rodata=off.
708 	 */
709 	pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
710 
711 	/*
712 	 * If we have a CPU that supports BTI and a kernel built for
713 	 * BTI then mark the kernel executable text as guarded pages
714 	 * now so we don't have to rewrite the page tables later.
715 	 */
716 	if (arm64_early_this_cpu_has_bti())
717 		text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
718 
719 	/*
720 	 * Only rodata will be remapped with different permissions later on,
721 	 * all other segments are allowed to use contiguous mappings.
722 	 */
723 	map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0,
724 			   VM_NO_GUARD);
725 	map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
726 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
727 	map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
728 			   &vmlinux_inittext, 0, VM_NO_GUARD);
729 	map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
730 			   &vmlinux_initdata, 0, VM_NO_GUARD);
731 	map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
732 
733 	if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_START)))) {
734 		/*
735 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
736 		 * live in the carveout for the swapper_pg_dir. We can simply
737 		 * re-use the existing dir for the fixmap.
738 		 */
739 		set_pgd(pgd_offset_pgd(pgdp, FIXADDR_START),
740 			READ_ONCE(*pgd_offset_k(FIXADDR_START)));
741 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
742 		pgd_t *bm_pgdp;
743 		p4d_t *bm_p4dp;
744 		pud_t *bm_pudp;
745 		/*
746 		 * The fixmap shares its top level pgd entry with the kernel
747 		 * mapping. This can really only occur when we are running
748 		 * with 16k/4 levels, so we can simply reuse the pud level
749 		 * entry instead.
750 		 */
751 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
752 		bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_START);
753 		bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_START);
754 		bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_START);
755 		pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
756 		pud_clear_fixmap();
757 	} else {
758 		BUG();
759 	}
760 
761 	kasan_copy_shadow(pgdp);
762 }
763 
764 static void __init create_idmap(void)
765 {
766 	u64 start = __pa_symbol(__idmap_text_start);
767 	u64 size = __pa_symbol(__idmap_text_end) - start;
768 	pgd_t *pgd = idmap_pg_dir;
769 	u64 pgd_phys;
770 
771 	/* check if we need an additional level of translation */
772 	if (VA_BITS < 48 && idmap_t0sz < (64 - VA_BITS_MIN)) {
773 		pgd_phys = early_pgtable_alloc(PAGE_SHIFT);
774 		set_pgd(&idmap_pg_dir[start >> VA_BITS],
775 			__pgd(pgd_phys | P4D_TYPE_TABLE));
776 		pgd = __va(pgd_phys);
777 	}
778 	__create_pgd_mapping(pgd, start, start, size, PAGE_KERNEL_ROX,
779 			     early_pgtable_alloc, 0);
780 
781 	if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0)) {
782 		extern u32 __idmap_kpti_flag;
783 		u64 pa = __pa_symbol(&__idmap_kpti_flag);
784 
785 		/*
786 		 * The KPTI G-to-nG conversion code needs a read-write mapping
787 		 * of its synchronization flag in the ID map.
788 		 */
789 		__create_pgd_mapping(pgd, pa, pa, sizeof(u32), PAGE_KERNEL,
790 				     early_pgtable_alloc, 0);
791 	}
792 }
793 
794 void __init paging_init(void)
795 {
796 	pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
797 	extern pgd_t init_idmap_pg_dir[];
798 
799 	idmap_t0sz = 63UL - __fls(__pa_symbol(_end) | GENMASK(VA_BITS_MIN - 1, 0));
800 
801 	map_kernel(pgdp);
802 	map_mem(pgdp);
803 
804 	pgd_clear_fixmap();
805 
806 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir), init_idmap_pg_dir);
807 	init_mm.pgd = swapper_pg_dir;
808 
809 	memblock_phys_free(__pa_symbol(init_pg_dir),
810 			   __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
811 
812 	memblock_allow_resize();
813 
814 	create_idmap();
815 }
816 
817 #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 = READ_ONCE(*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 = READ_ONCE(*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(READ_ONCE(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 (CONFIG_PGTABLE_LEVELS <= 3)
1081 		return;
1082 
1083 	if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_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 	unsigned long next;
1107 	p4d_t *p4dp, p4d;
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 
1121 static void free_empty_tables(unsigned long addr, unsigned long end,
1122 			      unsigned long floor, unsigned long ceiling)
1123 {
1124 	unsigned long next;
1125 	pgd_t *pgdp, pgd;
1126 
1127 	do {
1128 		next = pgd_addr_end(addr, end);
1129 		pgdp = pgd_offset_k(addr);
1130 		pgd = READ_ONCE(*pgdp);
1131 		if (pgd_none(pgd))
1132 			continue;
1133 
1134 		WARN_ON(!pgd_present(pgd));
1135 		free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1136 	} while (addr = next, addr < end);
1137 }
1138 #endif
1139 
1140 void __meminit vmemmap_set_pmd(pmd_t *pmdp, void *p, int node,
1141 			       unsigned long addr, unsigned long next)
1142 {
1143 	pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1144 }
1145 
1146 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1147 				unsigned long addr, unsigned long next)
1148 {
1149 	vmemmap_verify((pte_t *)pmdp, node, addr, next);
1150 	return 1;
1151 }
1152 
1153 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1154 		struct vmem_altmap *altmap)
1155 {
1156 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1157 
1158 	if (!IS_ENABLED(CONFIG_ARM64_4K_PAGES))
1159 		return vmemmap_populate_basepages(start, end, node, altmap);
1160 	else
1161 		return vmemmap_populate_hugepages(start, end, node, altmap);
1162 }
1163 
1164 #ifdef CONFIG_MEMORY_HOTPLUG
1165 void vmemmap_free(unsigned long start, unsigned long end,
1166 		struct vmem_altmap *altmap)
1167 {
1168 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1169 
1170 	unmap_hotplug_range(start, end, true, altmap);
1171 	free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1172 }
1173 #endif /* CONFIG_MEMORY_HOTPLUG */
1174 
1175 static inline pud_t *fixmap_pud(unsigned long addr)
1176 {
1177 	pgd_t *pgdp = pgd_offset_k(addr);
1178 	p4d_t *p4dp = p4d_offset(pgdp, addr);
1179 	p4d_t p4d = READ_ONCE(*p4dp);
1180 
1181 	BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
1182 
1183 	return pud_offset_kimg(p4dp, addr);
1184 }
1185 
1186 static inline pmd_t *fixmap_pmd(unsigned long addr)
1187 {
1188 	pud_t *pudp = fixmap_pud(addr);
1189 	pud_t pud = READ_ONCE(*pudp);
1190 
1191 	BUG_ON(pud_none(pud) || pud_bad(pud));
1192 
1193 	return pmd_offset_kimg(pudp, addr);
1194 }
1195 
1196 static inline pte_t *fixmap_pte(unsigned long addr)
1197 {
1198 	return &bm_pte[pte_index(addr)];
1199 }
1200 
1201 /*
1202  * The p*d_populate functions call virt_to_phys implicitly so they can't be used
1203  * directly on kernel symbols (bm_p*d). This function is called too early to use
1204  * lm_alias so __p*d_populate functions must be used to populate with the
1205  * physical address from __pa_symbol.
1206  */
1207 void __init early_fixmap_init(void)
1208 {
1209 	pgd_t *pgdp;
1210 	p4d_t *p4dp, p4d;
1211 	pud_t *pudp;
1212 	pmd_t *pmdp;
1213 	unsigned long addr = FIXADDR_START;
1214 
1215 	pgdp = pgd_offset_k(addr);
1216 	p4dp = p4d_offset(pgdp, addr);
1217 	p4d = READ_ONCE(*p4dp);
1218 	if (CONFIG_PGTABLE_LEVELS > 3 &&
1219 	    !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
1220 		/*
1221 		 * We only end up here if the kernel mapping and the fixmap
1222 		 * share the top level pgd entry, which should only happen on
1223 		 * 16k/4 levels configurations.
1224 		 */
1225 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
1226 		pudp = pud_offset_kimg(p4dp, addr);
1227 	} else {
1228 		if (p4d_none(p4d))
1229 			__p4d_populate(p4dp, __pa_symbol(bm_pud), P4D_TYPE_TABLE);
1230 		pudp = fixmap_pud(addr);
1231 	}
1232 	if (pud_none(READ_ONCE(*pudp)))
1233 		__pud_populate(pudp, __pa_symbol(bm_pmd), PUD_TYPE_TABLE);
1234 	pmdp = fixmap_pmd(addr);
1235 	__pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
1236 
1237 	/*
1238 	 * The boot-ioremap range spans multiple pmds, for which
1239 	 * we are not prepared:
1240 	 */
1241 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1242 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1243 
1244 	if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
1245 	     || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
1246 		WARN_ON(1);
1247 		pr_warn("pmdp %p != %p, %p\n",
1248 			pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
1249 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
1250 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1251 			fix_to_virt(FIX_BTMAP_BEGIN));
1252 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1253 			fix_to_virt(FIX_BTMAP_END));
1254 
1255 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1256 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1257 	}
1258 }
1259 
1260 /*
1261  * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
1262  * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
1263  */
1264 void __set_fixmap(enum fixed_addresses idx,
1265 			       phys_addr_t phys, pgprot_t flags)
1266 {
1267 	unsigned long addr = __fix_to_virt(idx);
1268 	pte_t *ptep;
1269 
1270 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
1271 
1272 	ptep = fixmap_pte(addr);
1273 
1274 	if (pgprot_val(flags)) {
1275 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
1276 	} else {
1277 		pte_clear(&init_mm, addr, ptep);
1278 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
1279 	}
1280 }
1281 
1282 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
1283 {
1284 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
1285 	int offset;
1286 	void *dt_virt;
1287 
1288 	/*
1289 	 * Check whether the physical FDT address is set and meets the minimum
1290 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
1291 	 * at least 8 bytes so that we can always access the magic and size
1292 	 * fields of the FDT header after mapping the first chunk, double check
1293 	 * here if that is indeed the case.
1294 	 */
1295 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
1296 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
1297 		return NULL;
1298 
1299 	/*
1300 	 * Make sure that the FDT region can be mapped without the need to
1301 	 * allocate additional translation table pages, so that it is safe
1302 	 * to call create_mapping_noalloc() this early.
1303 	 *
1304 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
1305 	 * be in the same PMD as the rest of the fixmap.
1306 	 * On 4k pages, we'll use section mappings for the FDT so we only
1307 	 * have to be in the same PUD.
1308 	 */
1309 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
1310 
1311 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
1312 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
1313 
1314 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
1315 	dt_virt = (void *)dt_virt_base + offset;
1316 
1317 	/* map the first chunk so we can read the size from the header */
1318 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
1319 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
1320 
1321 	if (fdt_magic(dt_virt) != FDT_MAGIC)
1322 		return NULL;
1323 
1324 	*size = fdt_totalsize(dt_virt);
1325 	if (*size > MAX_FDT_SIZE)
1326 		return NULL;
1327 
1328 	if (offset + *size > SWAPPER_BLOCK_SIZE)
1329 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
1330 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
1331 
1332 	return dt_virt;
1333 }
1334 
1335 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1336 {
1337 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1338 
1339 	/* Only allow permission changes for now */
1340 	if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1341 				   pud_val(new_pud)))
1342 		return 0;
1343 
1344 	VM_BUG_ON(phys & ~PUD_MASK);
1345 	set_pud(pudp, new_pud);
1346 	return 1;
1347 }
1348 
1349 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1350 {
1351 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1352 
1353 	/* Only allow permission changes for now */
1354 	if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1355 				   pmd_val(new_pmd)))
1356 		return 0;
1357 
1358 	VM_BUG_ON(phys & ~PMD_MASK);
1359 	set_pmd(pmdp, new_pmd);
1360 	return 1;
1361 }
1362 
1363 int pud_clear_huge(pud_t *pudp)
1364 {
1365 	if (!pud_sect(READ_ONCE(*pudp)))
1366 		return 0;
1367 	pud_clear(pudp);
1368 	return 1;
1369 }
1370 
1371 int pmd_clear_huge(pmd_t *pmdp)
1372 {
1373 	if (!pmd_sect(READ_ONCE(*pmdp)))
1374 		return 0;
1375 	pmd_clear(pmdp);
1376 	return 1;
1377 }
1378 
1379 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1380 {
1381 	pte_t *table;
1382 	pmd_t pmd;
1383 
1384 	pmd = READ_ONCE(*pmdp);
1385 
1386 	if (!pmd_table(pmd)) {
1387 		VM_WARN_ON(1);
1388 		return 1;
1389 	}
1390 
1391 	table = pte_offset_kernel(pmdp, addr);
1392 	pmd_clear(pmdp);
1393 	__flush_tlb_kernel_pgtable(addr);
1394 	pte_free_kernel(NULL, table);
1395 	return 1;
1396 }
1397 
1398 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1399 {
1400 	pmd_t *table;
1401 	pmd_t *pmdp;
1402 	pud_t pud;
1403 	unsigned long next, end;
1404 
1405 	pud = READ_ONCE(*pudp);
1406 
1407 	if (!pud_table(pud)) {
1408 		VM_WARN_ON(1);
1409 		return 1;
1410 	}
1411 
1412 	table = pmd_offset(pudp, addr);
1413 	pmdp = table;
1414 	next = addr;
1415 	end = addr + PUD_SIZE;
1416 	do {
1417 		pmd_free_pte_page(pmdp, next);
1418 	} while (pmdp++, next += PMD_SIZE, next != end);
1419 
1420 	pud_clear(pudp);
1421 	__flush_tlb_kernel_pgtable(addr);
1422 	pmd_free(NULL, table);
1423 	return 1;
1424 }
1425 
1426 #ifdef CONFIG_MEMORY_HOTPLUG
1427 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1428 {
1429 	unsigned long end = start + size;
1430 
1431 	WARN_ON(pgdir != init_mm.pgd);
1432 	WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1433 
1434 	unmap_hotplug_range(start, end, false, NULL);
1435 	free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1436 }
1437 
1438 struct range arch_get_mappable_range(void)
1439 {
1440 	struct range mhp_range;
1441 	u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1442 	u64 end_linear_pa = __pa(PAGE_END - 1);
1443 
1444 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1445 		/*
1446 		 * Check for a wrap, it is possible because of randomized linear
1447 		 * mapping the start physical address is actually bigger than
1448 		 * the end physical address. In this case set start to zero
1449 		 * because [0, end_linear_pa] range must still be able to cover
1450 		 * all addressable physical addresses.
1451 		 */
1452 		if (start_linear_pa > end_linear_pa)
1453 			start_linear_pa = 0;
1454 	}
1455 
1456 	WARN_ON(start_linear_pa > end_linear_pa);
1457 
1458 	/*
1459 	 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1460 	 * accommodating both its ends but excluding PAGE_END. Max physical
1461 	 * range which can be mapped inside this linear mapping range, must
1462 	 * also be derived from its end points.
1463 	 */
1464 	mhp_range.start = start_linear_pa;
1465 	mhp_range.end =  end_linear_pa;
1466 
1467 	return mhp_range;
1468 }
1469 
1470 int arch_add_memory(int nid, u64 start, u64 size,
1471 		    struct mhp_params *params)
1472 {
1473 	int ret, flags = NO_EXEC_MAPPINGS;
1474 
1475 	VM_BUG_ON(!mhp_range_allowed(start, size, true));
1476 
1477 	if (can_set_direct_map())
1478 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1479 
1480 	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1481 			     size, params->pgprot, __pgd_pgtable_alloc,
1482 			     flags);
1483 
1484 	memblock_clear_nomap(start, size);
1485 
1486 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1487 			   params);
1488 	if (ret)
1489 		__remove_pgd_mapping(swapper_pg_dir,
1490 				     __phys_to_virt(start), size);
1491 	else {
1492 		max_pfn = PFN_UP(start + size);
1493 		max_low_pfn = max_pfn;
1494 	}
1495 
1496 	return ret;
1497 }
1498 
1499 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1500 {
1501 	unsigned long start_pfn = start >> PAGE_SHIFT;
1502 	unsigned long nr_pages = size >> PAGE_SHIFT;
1503 
1504 	__remove_pages(start_pfn, nr_pages, altmap);
1505 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1506 }
1507 
1508 /*
1509  * This memory hotplug notifier helps prevent boot memory from being
1510  * inadvertently removed as it blocks pfn range offlining process in
1511  * __offline_pages(). Hence this prevents both offlining as well as
1512  * removal process for boot memory which is initially always online.
1513  * In future if and when boot memory could be removed, this notifier
1514  * should be dropped and free_hotplug_page_range() should handle any
1515  * reserved pages allocated during boot.
1516  */
1517 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1518 					   unsigned long action, void *data)
1519 {
1520 	struct mem_section *ms;
1521 	struct memory_notify *arg = data;
1522 	unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1523 	unsigned long pfn = arg->start_pfn;
1524 
1525 	if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1526 		return NOTIFY_OK;
1527 
1528 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1529 		unsigned long start = PFN_PHYS(pfn);
1530 		unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1531 
1532 		ms = __pfn_to_section(pfn);
1533 		if (!early_section(ms))
1534 			continue;
1535 
1536 		if (action == MEM_GOING_OFFLINE) {
1537 			/*
1538 			 * Boot memory removal is not supported. Prevent
1539 			 * it via blocking any attempted offline request
1540 			 * for the boot memory and just report it.
1541 			 */
1542 			pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1543 			return NOTIFY_BAD;
1544 		} else if (action == MEM_OFFLINE) {
1545 			/*
1546 			 * This should have never happened. Boot memory
1547 			 * offlining should have been prevented by this
1548 			 * very notifier. Probably some memory removal
1549 			 * procedure might have changed which would then
1550 			 * require further debug.
1551 			 */
1552 			pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1553 
1554 			/*
1555 			 * Core memory hotplug does not process a return
1556 			 * code from the notifier for MEM_OFFLINE events.
1557 			 * The error condition has been reported. Return
1558 			 * from here as if ignored.
1559 			 */
1560 			return NOTIFY_DONE;
1561 		}
1562 	}
1563 	return NOTIFY_OK;
1564 }
1565 
1566 static struct notifier_block prevent_bootmem_remove_nb = {
1567 	.notifier_call = prevent_bootmem_remove_notifier,
1568 };
1569 
1570 /*
1571  * This ensures that boot memory sections on the platform are online
1572  * from early boot. Memory sections could not be prevented from being
1573  * offlined, unless for some reason they are not online to begin with.
1574  * This helps validate the basic assumption on which the above memory
1575  * event notifier works to prevent boot memory section offlining and
1576  * its possible removal.
1577  */
1578 static void validate_bootmem_online(void)
1579 {
1580 	phys_addr_t start, end, addr;
1581 	struct mem_section *ms;
1582 	u64 i;
1583 
1584 	/*
1585 	 * Scanning across all memblock might be expensive
1586 	 * on some big memory systems. Hence enable this
1587 	 * validation only with DEBUG_VM.
1588 	 */
1589 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
1590 		return;
1591 
1592 	for_each_mem_range(i, &start, &end) {
1593 		for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1594 			ms = __pfn_to_section(PHYS_PFN(addr));
1595 
1596 			/*
1597 			 * All memory ranges in the system at this point
1598 			 * should have been marked as early sections.
1599 			 */
1600 			WARN_ON(!early_section(ms));
1601 
1602 			/*
1603 			 * Memory notifier mechanism here to prevent boot
1604 			 * memory offlining depends on the fact that each
1605 			 * early section memory on the system is initially
1606 			 * online. Otherwise a given memory section which
1607 			 * is already offline will be overlooked and can
1608 			 * be removed completely. Call out such sections.
1609 			 */
1610 			if (!online_section(ms))
1611 				pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1612 					addr, addr + (1UL << PA_SECTION_SHIFT));
1613 		}
1614 	}
1615 }
1616 
1617 static int __init prevent_bootmem_remove_init(void)
1618 {
1619 	int ret = 0;
1620 
1621 	if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1622 		return ret;
1623 
1624 	validate_bootmem_online();
1625 	ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1626 	if (ret)
1627 		pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1628 
1629 	return ret;
1630 }
1631 early_initcall(prevent_bootmem_remove_init);
1632 #endif
1633