xref: /openbmc/linux/arch/arm64/mm/mmu.c (revision 2daf7fab)
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 u64 idmap_t0sz = TCR_T0SZ(VA_BITS_MIN);
47 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD;
48 
49 u64 __section(".mmuoff.data.write") vabits_actual;
50 EXPORT_SYMBOL(vabits_actual);
51 
52 u64 kimage_voffset __ro_after_init;
53 EXPORT_SYMBOL(kimage_voffset);
54 
55 /*
56  * Empty_zero_page is a special page that is used for zero-initialized data
57  * and COW.
58  */
59 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
60 EXPORT_SYMBOL(empty_zero_page);
61 
62 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
63 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
64 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
65 
66 static DEFINE_SPINLOCK(swapper_pgdir_lock);
67 static DEFINE_MUTEX(fixmap_lock);
68 
69 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
70 {
71 	pgd_t *fixmap_pgdp;
72 
73 	spin_lock(&swapper_pgdir_lock);
74 	fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
75 	WRITE_ONCE(*fixmap_pgdp, pgd);
76 	/*
77 	 * We need dsb(ishst) here to ensure the page-table-walker sees
78 	 * our new entry before set_p?d() returns. The fixmap's
79 	 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
80 	 */
81 	pgd_clear_fixmap();
82 	spin_unlock(&swapper_pgdir_lock);
83 }
84 
85 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
86 			      unsigned long size, pgprot_t vma_prot)
87 {
88 	if (!pfn_is_map_memory(pfn))
89 		return pgprot_noncached(vma_prot);
90 	else if (file->f_flags & O_SYNC)
91 		return pgprot_writecombine(vma_prot);
92 	return vma_prot;
93 }
94 EXPORT_SYMBOL(phys_mem_access_prot);
95 
96 static phys_addr_t __init early_pgtable_alloc(int shift)
97 {
98 	phys_addr_t phys;
99 	void *ptr;
100 
101 	phys = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0,
102 					 MEMBLOCK_ALLOC_NOLEAKTRACE);
103 	if (!phys)
104 		panic("Failed to allocate page table page\n");
105 
106 	/*
107 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
108 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
109 	 * any level of table.
110 	 */
111 	ptr = pte_set_fixmap(phys);
112 
113 	memset(ptr, 0, PAGE_SIZE);
114 
115 	/*
116 	 * Implicit barriers also ensure the zeroed page is visible to the page
117 	 * table walker
118 	 */
119 	pte_clear_fixmap();
120 
121 	return phys;
122 }
123 
124 static bool pgattr_change_is_safe(u64 old, u64 new)
125 {
126 	/*
127 	 * The following mapping attributes may be updated in live
128 	 * kernel mappings without the need for break-before-make.
129 	 */
130 	pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
131 
132 	/* creating or taking down mappings is always safe */
133 	if (old == 0 || new == 0)
134 		return true;
135 
136 	/* live contiguous mappings may not be manipulated at all */
137 	if ((old | new) & PTE_CONT)
138 		return false;
139 
140 	/* Transitioning from Non-Global to Global is unsafe */
141 	if (old & ~new & PTE_NG)
142 		return false;
143 
144 	/*
145 	 * Changing the memory type between Normal and Normal-Tagged is safe
146 	 * since Tagged is considered a permission attribute from the
147 	 * mismatched attribute aliases perspective.
148 	 */
149 	if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
150 	     (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
151 	    ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
152 	     (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
153 		mask |= PTE_ATTRINDX_MASK;
154 
155 	return ((old ^ new) & ~mask) == 0;
156 }
157 
158 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
159 		     phys_addr_t phys, pgprot_t prot)
160 {
161 	pte_t *ptep;
162 
163 	ptep = pte_set_fixmap_offset(pmdp, addr);
164 	do {
165 		pte_t old_pte = READ_ONCE(*ptep);
166 
167 		set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
168 
169 		/*
170 		 * After the PTE entry has been populated once, we
171 		 * only allow updates to the permission attributes.
172 		 */
173 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
174 					      READ_ONCE(pte_val(*ptep))));
175 
176 		phys += PAGE_SIZE;
177 	} while (ptep++, addr += PAGE_SIZE, addr != end);
178 
179 	pte_clear_fixmap();
180 }
181 
182 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
183 				unsigned long end, phys_addr_t phys,
184 				pgprot_t prot,
185 				phys_addr_t (*pgtable_alloc)(int),
186 				int flags)
187 {
188 	unsigned long next;
189 	pmd_t pmd = READ_ONCE(*pmdp);
190 
191 	BUG_ON(pmd_sect(pmd));
192 	if (pmd_none(pmd)) {
193 		pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
194 		phys_addr_t pte_phys;
195 
196 		if (flags & NO_EXEC_MAPPINGS)
197 			pmdval |= PMD_TABLE_PXN;
198 		BUG_ON(!pgtable_alloc);
199 		pte_phys = pgtable_alloc(PAGE_SHIFT);
200 		__pmd_populate(pmdp, pte_phys, pmdval);
201 		pmd = READ_ONCE(*pmdp);
202 	}
203 	BUG_ON(pmd_bad(pmd));
204 
205 	do {
206 		pgprot_t __prot = prot;
207 
208 		next = pte_cont_addr_end(addr, end);
209 
210 		/* use a contiguous mapping if the range is suitably aligned */
211 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
212 		    (flags & NO_CONT_MAPPINGS) == 0)
213 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
214 
215 		init_pte(pmdp, addr, next, phys, __prot);
216 
217 		phys += next - addr;
218 	} while (addr = next, addr != end);
219 }
220 
221 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
222 		     phys_addr_t phys, pgprot_t prot,
223 		     phys_addr_t (*pgtable_alloc)(int), int flags)
224 {
225 	unsigned long next;
226 	pmd_t *pmdp;
227 
228 	pmdp = pmd_set_fixmap_offset(pudp, addr);
229 	do {
230 		pmd_t old_pmd = READ_ONCE(*pmdp);
231 
232 		next = pmd_addr_end(addr, end);
233 
234 		/* try section mapping first */
235 		if (((addr | next | phys) & ~PMD_MASK) == 0 &&
236 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
237 			pmd_set_huge(pmdp, phys, prot);
238 
239 			/*
240 			 * After the PMD entry has been populated once, we
241 			 * only allow updates to the permission attributes.
242 			 */
243 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
244 						      READ_ONCE(pmd_val(*pmdp))));
245 		} else {
246 			alloc_init_cont_pte(pmdp, addr, next, phys, prot,
247 					    pgtable_alloc, flags);
248 
249 			BUG_ON(pmd_val(old_pmd) != 0 &&
250 			       pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
251 		}
252 		phys += next - addr;
253 	} while (pmdp++, addr = next, addr != end);
254 
255 	pmd_clear_fixmap();
256 }
257 
258 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
259 				unsigned long end, phys_addr_t phys,
260 				pgprot_t prot,
261 				phys_addr_t (*pgtable_alloc)(int), int flags)
262 {
263 	unsigned long next;
264 	pud_t pud = READ_ONCE(*pudp);
265 
266 	/*
267 	 * Check for initial section mappings in the pgd/pud.
268 	 */
269 	BUG_ON(pud_sect(pud));
270 	if (pud_none(pud)) {
271 		pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
272 		phys_addr_t pmd_phys;
273 
274 		if (flags & NO_EXEC_MAPPINGS)
275 			pudval |= PUD_TABLE_PXN;
276 		BUG_ON(!pgtable_alloc);
277 		pmd_phys = pgtable_alloc(PMD_SHIFT);
278 		__pud_populate(pudp, pmd_phys, pudval);
279 		pud = READ_ONCE(*pudp);
280 	}
281 	BUG_ON(pud_bad(pud));
282 
283 	do {
284 		pgprot_t __prot = prot;
285 
286 		next = pmd_cont_addr_end(addr, end);
287 
288 		/* use a contiguous mapping if the range is suitably aligned */
289 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
290 		    (flags & NO_CONT_MAPPINGS) == 0)
291 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
292 
293 		init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
294 
295 		phys += next - addr;
296 	} while (addr = next, addr != end);
297 }
298 
299 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
300 			   phys_addr_t phys, pgprot_t prot,
301 			   phys_addr_t (*pgtable_alloc)(int),
302 			   int flags)
303 {
304 	unsigned long next;
305 	pud_t *pudp;
306 	p4d_t *p4dp = p4d_offset(pgdp, addr);
307 	p4d_t p4d = READ_ONCE(*p4dp);
308 
309 	if (p4d_none(p4d)) {
310 		p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
311 		phys_addr_t pud_phys;
312 
313 		if (flags & NO_EXEC_MAPPINGS)
314 			p4dval |= P4D_TABLE_PXN;
315 		BUG_ON(!pgtable_alloc);
316 		pud_phys = pgtable_alloc(PUD_SHIFT);
317 		__p4d_populate(p4dp, pud_phys, p4dval);
318 		p4d = READ_ONCE(*p4dp);
319 	}
320 	BUG_ON(p4d_bad(p4d));
321 
322 	/*
323 	 * No need for locking during early boot. And it doesn't work as
324 	 * expected with KASLR enabled.
325 	 */
326 	if (system_state != SYSTEM_BOOTING)
327 		mutex_lock(&fixmap_lock);
328 	pudp = pud_set_fixmap_offset(p4dp, addr);
329 	do {
330 		pud_t old_pud = READ_ONCE(*pudp);
331 
332 		next = pud_addr_end(addr, end);
333 
334 		/*
335 		 * For 4K granule only, attempt to put down a 1GB block
336 		 */
337 		if (pud_sect_supported() &&
338 		   ((addr | next | phys) & ~PUD_MASK) == 0 &&
339 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
340 			pud_set_huge(pudp, phys, prot);
341 
342 			/*
343 			 * After the PUD entry has been populated once, we
344 			 * only allow updates to the permission attributes.
345 			 */
346 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
347 						      READ_ONCE(pud_val(*pudp))));
348 		} else {
349 			alloc_init_cont_pmd(pudp, addr, next, phys, prot,
350 					    pgtable_alloc, flags);
351 
352 			BUG_ON(pud_val(old_pud) != 0 &&
353 			       pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
354 		}
355 		phys += next - addr;
356 	} while (pudp++, addr = next, addr != end);
357 
358 	pud_clear_fixmap();
359 	if (system_state != SYSTEM_BOOTING)
360 		mutex_unlock(&fixmap_lock);
361 }
362 
363 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
364 				 unsigned long virt, phys_addr_t size,
365 				 pgprot_t prot,
366 				 phys_addr_t (*pgtable_alloc)(int),
367 				 int flags)
368 {
369 	unsigned long addr, end, next;
370 	pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
371 
372 	/*
373 	 * If the virtual and physical address don't have the same offset
374 	 * within a page, we cannot map the region as the caller expects.
375 	 */
376 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
377 		return;
378 
379 	phys &= PAGE_MASK;
380 	addr = virt & PAGE_MASK;
381 	end = PAGE_ALIGN(virt + size);
382 
383 	do {
384 		next = pgd_addr_end(addr, end);
385 		alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
386 			       flags);
387 		phys += next - addr;
388 	} while (pgdp++, addr = next, addr != end);
389 }
390 
391 static phys_addr_t __pgd_pgtable_alloc(int shift)
392 {
393 	void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
394 	BUG_ON(!ptr);
395 
396 	/* Ensure the zeroed page is visible to the page table walker */
397 	dsb(ishst);
398 	return __pa(ptr);
399 }
400 
401 static phys_addr_t pgd_pgtable_alloc(int shift)
402 {
403 	phys_addr_t pa = __pgd_pgtable_alloc(shift);
404 
405 	/*
406 	 * Call proper page table ctor in case later we need to
407 	 * call core mm functions like apply_to_page_range() on
408 	 * this pre-allocated page table.
409 	 *
410 	 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
411 	 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
412 	 */
413 	if (shift == PAGE_SHIFT)
414 		BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
415 	else if (shift == PMD_SHIFT)
416 		BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
417 
418 	return pa;
419 }
420 
421 /*
422  * This function can only be used to modify existing table entries,
423  * without allocating new levels of table. Note that this permits the
424  * creation of new section or page entries.
425  */
426 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
427 				  phys_addr_t size, pgprot_t prot)
428 {
429 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
430 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
431 			&phys, virt);
432 		return;
433 	}
434 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
435 			     NO_CONT_MAPPINGS);
436 }
437 
438 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
439 			       unsigned long virt, phys_addr_t size,
440 			       pgprot_t prot, bool page_mappings_only)
441 {
442 	int flags = 0;
443 
444 	BUG_ON(mm == &init_mm);
445 
446 	if (page_mappings_only)
447 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
448 
449 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
450 			     pgd_pgtable_alloc, flags);
451 }
452 
453 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
454 				phys_addr_t size, pgprot_t prot)
455 {
456 	if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
457 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
458 			&phys, virt);
459 		return;
460 	}
461 
462 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
463 			     NO_CONT_MAPPINGS);
464 
465 	/* flush the TLBs after updating live kernel mappings */
466 	flush_tlb_kernel_range(virt, virt + size);
467 }
468 
469 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
470 				  phys_addr_t end, pgprot_t prot, int flags)
471 {
472 	__create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
473 			     prot, early_pgtable_alloc, flags);
474 }
475 
476 void __init mark_linear_text_alias_ro(void)
477 {
478 	/*
479 	 * Remove the write permissions from the linear alias of .text/.rodata
480 	 */
481 	update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
482 			    (unsigned long)__init_begin - (unsigned long)_stext,
483 			    PAGE_KERNEL_RO);
484 }
485 
486 static bool crash_mem_map __initdata;
487 
488 static int __init enable_crash_mem_map(char *arg)
489 {
490 	/*
491 	 * Proper parameter parsing is done by reserve_crashkernel(). We only
492 	 * need to know if the linear map has to avoid block mappings so that
493 	 * the crashkernel reservations can be unmapped later.
494 	 */
495 	crash_mem_map = true;
496 
497 	return 0;
498 }
499 early_param("crashkernel", enable_crash_mem_map);
500 
501 static void __init map_mem(pgd_t *pgdp)
502 {
503 	static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
504 	phys_addr_t kernel_start = __pa_symbol(_stext);
505 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
506 	phys_addr_t start, end;
507 	int flags = NO_EXEC_MAPPINGS;
508 	u64 i;
509 
510 	/*
511 	 * Setting hierarchical PXNTable attributes on table entries covering
512 	 * the linear region is only possible if it is guaranteed that no table
513 	 * entries at any level are being shared between the linear region and
514 	 * the vmalloc region. Check whether this is true for the PGD level, in
515 	 * which case it is guaranteed to be true for all other levels as well.
516 	 */
517 	BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
518 
519 	if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
520 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
521 
522 	/*
523 	 * Take care not to create a writable alias for the
524 	 * read-only text and rodata sections of the kernel image.
525 	 * So temporarily mark them as NOMAP to skip mappings in
526 	 * the following for-loop
527 	 */
528 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
529 
530 #ifdef CONFIG_KEXEC_CORE
531 	if (crash_mem_map) {
532 		if (IS_ENABLED(CONFIG_ZONE_DMA) ||
533 		    IS_ENABLED(CONFIG_ZONE_DMA32))
534 			flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
535 		else if (crashk_res.end)
536 			memblock_mark_nomap(crashk_res.start,
537 			    resource_size(&crashk_res));
538 	}
539 #endif
540 
541 	/* map all the memory banks */
542 	for_each_mem_range(i, &start, &end) {
543 		if (start >= end)
544 			break;
545 		/*
546 		 * The linear map must allow allocation tags reading/writing
547 		 * if MTE is present. Otherwise, it has the same attributes as
548 		 * PAGE_KERNEL.
549 		 */
550 		__map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
551 			       flags);
552 	}
553 
554 	/*
555 	 * Map the linear alias of the [_stext, __init_begin) interval
556 	 * as non-executable now, and remove the write permission in
557 	 * mark_linear_text_alias_ro() below (which will be called after
558 	 * alternative patching has completed). This makes the contents
559 	 * of the region accessible to subsystems such as hibernate,
560 	 * but protects it from inadvertent modification or execution.
561 	 * Note that contiguous mappings cannot be remapped in this way,
562 	 * so we should avoid them here.
563 	 */
564 	__map_memblock(pgdp, kernel_start, kernel_end,
565 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
566 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
567 
568 	/*
569 	 * Use page-level mappings here so that we can shrink the region
570 	 * in page granularity and put back unused memory to buddy system
571 	 * through /sys/kernel/kexec_crash_size interface.
572 	 */
573 #ifdef CONFIG_KEXEC_CORE
574 	if (crash_mem_map &&
575 	    !IS_ENABLED(CONFIG_ZONE_DMA) && !IS_ENABLED(CONFIG_ZONE_DMA32)) {
576 		if (crashk_res.end) {
577 			__map_memblock(pgdp, crashk_res.start,
578 				       crashk_res.end + 1,
579 				       PAGE_KERNEL,
580 				       NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
581 			memblock_clear_nomap(crashk_res.start,
582 					     resource_size(&crashk_res));
583 		}
584 	}
585 #endif
586 }
587 
588 void mark_rodata_ro(void)
589 {
590 	unsigned long section_size;
591 
592 	/*
593 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
594 	 * to cover NOTES and EXCEPTION_TABLE.
595 	 */
596 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
597 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
598 			    section_size, PAGE_KERNEL_RO);
599 
600 	debug_checkwx();
601 }
602 
603 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
604 				      pgprot_t prot, struct vm_struct *vma,
605 				      int flags, unsigned long vm_flags)
606 {
607 	phys_addr_t pa_start = __pa_symbol(va_start);
608 	unsigned long size = va_end - va_start;
609 
610 	BUG_ON(!PAGE_ALIGNED(pa_start));
611 	BUG_ON(!PAGE_ALIGNED(size));
612 
613 	__create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
614 			     early_pgtable_alloc, flags);
615 
616 	if (!(vm_flags & VM_NO_GUARD))
617 		size += PAGE_SIZE;
618 
619 	vma->addr	= va_start;
620 	vma->phys_addr	= pa_start;
621 	vma->size	= size;
622 	vma->flags	= VM_MAP | vm_flags;
623 	vma->caller	= __builtin_return_address(0);
624 
625 	vm_area_add_early(vma);
626 }
627 
628 static int __init parse_rodata(char *arg)
629 {
630 	int ret = strtobool(arg, &rodata_enabled);
631 	if (!ret) {
632 		rodata_full = false;
633 		return 0;
634 	}
635 
636 	/* permit 'full' in addition to boolean options */
637 	if (strcmp(arg, "full"))
638 		return -EINVAL;
639 
640 	rodata_enabled = true;
641 	rodata_full = true;
642 	return 0;
643 }
644 early_param("rodata", parse_rodata);
645 
646 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
647 static int __init map_entry_trampoline(void)
648 {
649 	int i;
650 
651 	pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
652 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
653 
654 	/* The trampoline is always mapped and can therefore be global */
655 	pgprot_val(prot) &= ~PTE_NG;
656 
657 	/* Map only the text into the trampoline page table */
658 	memset(tramp_pg_dir, 0, PGD_SIZE);
659 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
660 			     entry_tramp_text_size(), prot,
661 			     __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
662 
663 	/* Map both the text and data into the kernel page table */
664 	for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
665 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
666 			     pa_start + i * PAGE_SIZE, prot);
667 
668 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
669 		extern char __entry_tramp_data_start[];
670 
671 		__set_fixmap(FIX_ENTRY_TRAMP_DATA,
672 			     __pa_symbol(__entry_tramp_data_start),
673 			     PAGE_KERNEL_RO);
674 	}
675 
676 	return 0;
677 }
678 core_initcall(map_entry_trampoline);
679 #endif
680 
681 /*
682  * Open coded check for BTI, only for use to determine configuration
683  * for early mappings for before the cpufeature code has run.
684  */
685 static bool arm64_early_this_cpu_has_bti(void)
686 {
687 	u64 pfr1;
688 
689 	if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
690 		return false;
691 
692 	pfr1 = __read_sysreg_by_encoding(SYS_ID_AA64PFR1_EL1);
693 	return cpuid_feature_extract_unsigned_field(pfr1,
694 						    ID_AA64PFR1_BT_SHIFT);
695 }
696 
697 /*
698  * Create fine-grained mappings for the kernel.
699  */
700 static void __init map_kernel(pgd_t *pgdp)
701 {
702 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
703 				vmlinux_initdata, vmlinux_data;
704 
705 	/*
706 	 * External debuggers may need to write directly to the text
707 	 * mapping to install SW breakpoints. Allow this (only) when
708 	 * explicitly requested with rodata=off.
709 	 */
710 	pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
711 
712 	/*
713 	 * If we have a CPU that supports BTI and a kernel built for
714 	 * BTI then mark the kernel executable text as guarded pages
715 	 * now so we don't have to rewrite the page tables later.
716 	 */
717 	if (arm64_early_this_cpu_has_bti())
718 		text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
719 
720 	/*
721 	 * Only rodata will be remapped with different permissions later on,
722 	 * all other segments are allowed to use contiguous mappings.
723 	 */
724 	map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0,
725 			   VM_NO_GUARD);
726 	map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
727 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
728 	map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
729 			   &vmlinux_inittext, 0, VM_NO_GUARD);
730 	map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
731 			   &vmlinux_initdata, 0, VM_NO_GUARD);
732 	map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
733 
734 	if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_START)))) {
735 		/*
736 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
737 		 * live in the carveout for the swapper_pg_dir. We can simply
738 		 * re-use the existing dir for the fixmap.
739 		 */
740 		set_pgd(pgd_offset_pgd(pgdp, FIXADDR_START),
741 			READ_ONCE(*pgd_offset_k(FIXADDR_START)));
742 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
743 		pgd_t *bm_pgdp;
744 		p4d_t *bm_p4dp;
745 		pud_t *bm_pudp;
746 		/*
747 		 * The fixmap shares its top level pgd entry with the kernel
748 		 * mapping. This can really only occur when we are running
749 		 * with 16k/4 levels, so we can simply reuse the pud level
750 		 * entry instead.
751 		 */
752 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
753 		bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_START);
754 		bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_START);
755 		bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_START);
756 		pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
757 		pud_clear_fixmap();
758 	} else {
759 		BUG();
760 	}
761 
762 	kasan_copy_shadow(pgdp);
763 }
764 
765 void __init paging_init(void)
766 {
767 	pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
768 
769 	map_kernel(pgdp);
770 	map_mem(pgdp);
771 
772 	pgd_clear_fixmap();
773 
774 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
775 	init_mm.pgd = swapper_pg_dir;
776 
777 	memblock_phys_free(__pa_symbol(init_pg_dir),
778 			   __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
779 
780 	memblock_allow_resize();
781 }
782 
783 /*
784  * Check whether a kernel address is valid (derived from arch/x86/).
785  */
786 int kern_addr_valid(unsigned long addr)
787 {
788 	pgd_t *pgdp;
789 	p4d_t *p4dp;
790 	pud_t *pudp, pud;
791 	pmd_t *pmdp, pmd;
792 	pte_t *ptep, pte;
793 
794 	addr = arch_kasan_reset_tag(addr);
795 	if ((((long)addr) >> VA_BITS) != -1UL)
796 		return 0;
797 
798 	pgdp = pgd_offset_k(addr);
799 	if (pgd_none(READ_ONCE(*pgdp)))
800 		return 0;
801 
802 	p4dp = p4d_offset(pgdp, addr);
803 	if (p4d_none(READ_ONCE(*p4dp)))
804 		return 0;
805 
806 	pudp = pud_offset(p4dp, addr);
807 	pud = READ_ONCE(*pudp);
808 	if (pud_none(pud))
809 		return 0;
810 
811 	if (pud_sect(pud))
812 		return pfn_valid(pud_pfn(pud));
813 
814 	pmdp = pmd_offset(pudp, addr);
815 	pmd = READ_ONCE(*pmdp);
816 	if (pmd_none(pmd))
817 		return 0;
818 
819 	if (pmd_sect(pmd))
820 		return pfn_valid(pmd_pfn(pmd));
821 
822 	ptep = pte_offset_kernel(pmdp, addr);
823 	pte = READ_ONCE(*ptep);
824 	if (pte_none(pte))
825 		return 0;
826 
827 	return pfn_valid(pte_pfn(pte));
828 }
829 
830 #ifdef CONFIG_MEMORY_HOTPLUG
831 static void free_hotplug_page_range(struct page *page, size_t size,
832 				    struct vmem_altmap *altmap)
833 {
834 	if (altmap) {
835 		vmem_altmap_free(altmap, size >> PAGE_SHIFT);
836 	} else {
837 		WARN_ON(PageReserved(page));
838 		free_pages((unsigned long)page_address(page), get_order(size));
839 	}
840 }
841 
842 static void free_hotplug_pgtable_page(struct page *page)
843 {
844 	free_hotplug_page_range(page, PAGE_SIZE, NULL);
845 }
846 
847 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
848 				  unsigned long floor, unsigned long ceiling,
849 				  unsigned long mask)
850 {
851 	start &= mask;
852 	if (start < floor)
853 		return false;
854 
855 	if (ceiling) {
856 		ceiling &= mask;
857 		if (!ceiling)
858 			return false;
859 	}
860 
861 	if (end - 1 > ceiling - 1)
862 		return false;
863 	return true;
864 }
865 
866 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
867 				    unsigned long end, bool free_mapped,
868 				    struct vmem_altmap *altmap)
869 {
870 	pte_t *ptep, pte;
871 
872 	do {
873 		ptep = pte_offset_kernel(pmdp, addr);
874 		pte = READ_ONCE(*ptep);
875 		if (pte_none(pte))
876 			continue;
877 
878 		WARN_ON(!pte_present(pte));
879 		pte_clear(&init_mm, addr, ptep);
880 		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
881 		if (free_mapped)
882 			free_hotplug_page_range(pte_page(pte),
883 						PAGE_SIZE, altmap);
884 	} while (addr += PAGE_SIZE, addr < end);
885 }
886 
887 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
888 				    unsigned long end, bool free_mapped,
889 				    struct vmem_altmap *altmap)
890 {
891 	unsigned long next;
892 	pmd_t *pmdp, pmd;
893 
894 	do {
895 		next = pmd_addr_end(addr, end);
896 		pmdp = pmd_offset(pudp, addr);
897 		pmd = READ_ONCE(*pmdp);
898 		if (pmd_none(pmd))
899 			continue;
900 
901 		WARN_ON(!pmd_present(pmd));
902 		if (pmd_sect(pmd)) {
903 			pmd_clear(pmdp);
904 
905 			/*
906 			 * One TLBI should be sufficient here as the PMD_SIZE
907 			 * range is mapped with a single block entry.
908 			 */
909 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
910 			if (free_mapped)
911 				free_hotplug_page_range(pmd_page(pmd),
912 							PMD_SIZE, altmap);
913 			continue;
914 		}
915 		WARN_ON(!pmd_table(pmd));
916 		unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
917 	} while (addr = next, addr < end);
918 }
919 
920 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
921 				    unsigned long end, bool free_mapped,
922 				    struct vmem_altmap *altmap)
923 {
924 	unsigned long next;
925 	pud_t *pudp, pud;
926 
927 	do {
928 		next = pud_addr_end(addr, end);
929 		pudp = pud_offset(p4dp, addr);
930 		pud = READ_ONCE(*pudp);
931 		if (pud_none(pud))
932 			continue;
933 
934 		WARN_ON(!pud_present(pud));
935 		if (pud_sect(pud)) {
936 			pud_clear(pudp);
937 
938 			/*
939 			 * One TLBI should be sufficient here as the PUD_SIZE
940 			 * range is mapped with a single block entry.
941 			 */
942 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
943 			if (free_mapped)
944 				free_hotplug_page_range(pud_page(pud),
945 							PUD_SIZE, altmap);
946 			continue;
947 		}
948 		WARN_ON(!pud_table(pud));
949 		unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
950 	} while (addr = next, addr < end);
951 }
952 
953 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
954 				    unsigned long end, bool free_mapped,
955 				    struct vmem_altmap *altmap)
956 {
957 	unsigned long next;
958 	p4d_t *p4dp, p4d;
959 
960 	do {
961 		next = p4d_addr_end(addr, end);
962 		p4dp = p4d_offset(pgdp, addr);
963 		p4d = READ_ONCE(*p4dp);
964 		if (p4d_none(p4d))
965 			continue;
966 
967 		WARN_ON(!p4d_present(p4d));
968 		unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
969 	} while (addr = next, addr < end);
970 }
971 
972 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
973 				bool free_mapped, struct vmem_altmap *altmap)
974 {
975 	unsigned long next;
976 	pgd_t *pgdp, pgd;
977 
978 	/*
979 	 * altmap can only be used as vmemmap mapping backing memory.
980 	 * In case the backing memory itself is not being freed, then
981 	 * altmap is irrelevant. Warn about this inconsistency when
982 	 * encountered.
983 	 */
984 	WARN_ON(!free_mapped && altmap);
985 
986 	do {
987 		next = pgd_addr_end(addr, end);
988 		pgdp = pgd_offset_k(addr);
989 		pgd = READ_ONCE(*pgdp);
990 		if (pgd_none(pgd))
991 			continue;
992 
993 		WARN_ON(!pgd_present(pgd));
994 		unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
995 	} while (addr = next, addr < end);
996 }
997 
998 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
999 				 unsigned long end, unsigned long floor,
1000 				 unsigned long ceiling)
1001 {
1002 	pte_t *ptep, pte;
1003 	unsigned long i, start = addr;
1004 
1005 	do {
1006 		ptep = pte_offset_kernel(pmdp, addr);
1007 		pte = READ_ONCE(*ptep);
1008 
1009 		/*
1010 		 * This is just a sanity check here which verifies that
1011 		 * pte clearing has been done by earlier unmap loops.
1012 		 */
1013 		WARN_ON(!pte_none(pte));
1014 	} while (addr += PAGE_SIZE, addr < end);
1015 
1016 	if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
1017 		return;
1018 
1019 	/*
1020 	 * Check whether we can free the pte page if the rest of the
1021 	 * entries are empty. Overlap with other regions have been
1022 	 * handled by the floor/ceiling check.
1023 	 */
1024 	ptep = pte_offset_kernel(pmdp, 0UL);
1025 	for (i = 0; i < PTRS_PER_PTE; i++) {
1026 		if (!pte_none(READ_ONCE(ptep[i])))
1027 			return;
1028 	}
1029 
1030 	pmd_clear(pmdp);
1031 	__flush_tlb_kernel_pgtable(start);
1032 	free_hotplug_pgtable_page(virt_to_page(ptep));
1033 }
1034 
1035 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1036 				 unsigned long end, unsigned long floor,
1037 				 unsigned long ceiling)
1038 {
1039 	pmd_t *pmdp, pmd;
1040 	unsigned long i, next, start = addr;
1041 
1042 	do {
1043 		next = pmd_addr_end(addr, end);
1044 		pmdp = pmd_offset(pudp, addr);
1045 		pmd = READ_ONCE(*pmdp);
1046 		if (pmd_none(pmd))
1047 			continue;
1048 
1049 		WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1050 		free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1051 	} while (addr = next, addr < end);
1052 
1053 	if (CONFIG_PGTABLE_LEVELS <= 2)
1054 		return;
1055 
1056 	if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1057 		return;
1058 
1059 	/*
1060 	 * Check whether we can free the pmd page if the rest of the
1061 	 * entries are empty. Overlap with other regions have been
1062 	 * handled by the floor/ceiling check.
1063 	 */
1064 	pmdp = pmd_offset(pudp, 0UL);
1065 	for (i = 0; i < PTRS_PER_PMD; i++) {
1066 		if (!pmd_none(READ_ONCE(pmdp[i])))
1067 			return;
1068 	}
1069 
1070 	pud_clear(pudp);
1071 	__flush_tlb_kernel_pgtable(start);
1072 	free_hotplug_pgtable_page(virt_to_page(pmdp));
1073 }
1074 
1075 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1076 				 unsigned long end, unsigned long floor,
1077 				 unsigned long ceiling)
1078 {
1079 	pud_t *pudp, pud;
1080 	unsigned long i, next, start = addr;
1081 
1082 	do {
1083 		next = pud_addr_end(addr, end);
1084 		pudp = pud_offset(p4dp, addr);
1085 		pud = READ_ONCE(*pudp);
1086 		if (pud_none(pud))
1087 			continue;
1088 
1089 		WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1090 		free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1091 	} while (addr = next, addr < end);
1092 
1093 	if (CONFIG_PGTABLE_LEVELS <= 3)
1094 		return;
1095 
1096 	if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1097 		return;
1098 
1099 	/*
1100 	 * Check whether we can free the pud page if the rest of the
1101 	 * entries are empty. Overlap with other regions have been
1102 	 * handled by the floor/ceiling check.
1103 	 */
1104 	pudp = pud_offset(p4dp, 0UL);
1105 	for (i = 0; i < PTRS_PER_PUD; i++) {
1106 		if (!pud_none(READ_ONCE(pudp[i])))
1107 			return;
1108 	}
1109 
1110 	p4d_clear(p4dp);
1111 	__flush_tlb_kernel_pgtable(start);
1112 	free_hotplug_pgtable_page(virt_to_page(pudp));
1113 }
1114 
1115 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1116 				 unsigned long end, unsigned long floor,
1117 				 unsigned long ceiling)
1118 {
1119 	unsigned long next;
1120 	p4d_t *p4dp, p4d;
1121 
1122 	do {
1123 		next = p4d_addr_end(addr, end);
1124 		p4dp = p4d_offset(pgdp, addr);
1125 		p4d = READ_ONCE(*p4dp);
1126 		if (p4d_none(p4d))
1127 			continue;
1128 
1129 		WARN_ON(!p4d_present(p4d));
1130 		free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1131 	} while (addr = next, addr < end);
1132 }
1133 
1134 static void free_empty_tables(unsigned long addr, unsigned long end,
1135 			      unsigned long floor, unsigned long ceiling)
1136 {
1137 	unsigned long next;
1138 	pgd_t *pgdp, pgd;
1139 
1140 	do {
1141 		next = pgd_addr_end(addr, end);
1142 		pgdp = pgd_offset_k(addr);
1143 		pgd = READ_ONCE(*pgdp);
1144 		if (pgd_none(pgd))
1145 			continue;
1146 
1147 		WARN_ON(!pgd_present(pgd));
1148 		free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1149 	} while (addr = next, addr < end);
1150 }
1151 #endif
1152 
1153 #if !ARM64_KERNEL_USES_PMD_MAPS
1154 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1155 		struct vmem_altmap *altmap)
1156 {
1157 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1158 	return vmemmap_populate_basepages(start, end, node, altmap);
1159 }
1160 #else	/* !ARM64_KERNEL_USES_PMD_MAPS */
1161 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1162 		struct vmem_altmap *altmap)
1163 {
1164 	unsigned long addr = start;
1165 	unsigned long next;
1166 	pgd_t *pgdp;
1167 	p4d_t *p4dp;
1168 	pud_t *pudp;
1169 	pmd_t *pmdp;
1170 
1171 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1172 	do {
1173 		next = pmd_addr_end(addr, end);
1174 
1175 		pgdp = vmemmap_pgd_populate(addr, node);
1176 		if (!pgdp)
1177 			return -ENOMEM;
1178 
1179 		p4dp = vmemmap_p4d_populate(pgdp, addr, node);
1180 		if (!p4dp)
1181 			return -ENOMEM;
1182 
1183 		pudp = vmemmap_pud_populate(p4dp, addr, node);
1184 		if (!pudp)
1185 			return -ENOMEM;
1186 
1187 		pmdp = pmd_offset(pudp, addr);
1188 		if (pmd_none(READ_ONCE(*pmdp))) {
1189 			void *p = NULL;
1190 
1191 			p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1192 			if (!p) {
1193 				if (vmemmap_populate_basepages(addr, next, node, altmap))
1194 					return -ENOMEM;
1195 				continue;
1196 			}
1197 
1198 			pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1199 		} else
1200 			vmemmap_verify((pte_t *)pmdp, node, addr, next);
1201 	} while (addr = next, addr != end);
1202 
1203 	return 0;
1204 }
1205 #endif	/* !ARM64_KERNEL_USES_PMD_MAPS */
1206 
1207 #ifdef CONFIG_MEMORY_HOTPLUG
1208 void vmemmap_free(unsigned long start, unsigned long end,
1209 		struct vmem_altmap *altmap)
1210 {
1211 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1212 
1213 	unmap_hotplug_range(start, end, true, altmap);
1214 	free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1215 }
1216 #endif /* CONFIG_MEMORY_HOTPLUG */
1217 
1218 static inline pud_t *fixmap_pud(unsigned long addr)
1219 {
1220 	pgd_t *pgdp = pgd_offset_k(addr);
1221 	p4d_t *p4dp = p4d_offset(pgdp, addr);
1222 	p4d_t p4d = READ_ONCE(*p4dp);
1223 
1224 	BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
1225 
1226 	return pud_offset_kimg(p4dp, addr);
1227 }
1228 
1229 static inline pmd_t *fixmap_pmd(unsigned long addr)
1230 {
1231 	pud_t *pudp = fixmap_pud(addr);
1232 	pud_t pud = READ_ONCE(*pudp);
1233 
1234 	BUG_ON(pud_none(pud) || pud_bad(pud));
1235 
1236 	return pmd_offset_kimg(pudp, addr);
1237 }
1238 
1239 static inline pte_t *fixmap_pte(unsigned long addr)
1240 {
1241 	return &bm_pte[pte_index(addr)];
1242 }
1243 
1244 /*
1245  * The p*d_populate functions call virt_to_phys implicitly so they can't be used
1246  * directly on kernel symbols (bm_p*d). This function is called too early to use
1247  * lm_alias so __p*d_populate functions must be used to populate with the
1248  * physical address from __pa_symbol.
1249  */
1250 void __init early_fixmap_init(void)
1251 {
1252 	pgd_t *pgdp;
1253 	p4d_t *p4dp, p4d;
1254 	pud_t *pudp;
1255 	pmd_t *pmdp;
1256 	unsigned long addr = FIXADDR_START;
1257 
1258 	pgdp = pgd_offset_k(addr);
1259 	p4dp = p4d_offset(pgdp, addr);
1260 	p4d = READ_ONCE(*p4dp);
1261 	if (CONFIG_PGTABLE_LEVELS > 3 &&
1262 	    !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
1263 		/*
1264 		 * We only end up here if the kernel mapping and the fixmap
1265 		 * share the top level pgd entry, which should only happen on
1266 		 * 16k/4 levels configurations.
1267 		 */
1268 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
1269 		pudp = pud_offset_kimg(p4dp, addr);
1270 	} else {
1271 		if (p4d_none(p4d))
1272 			__p4d_populate(p4dp, __pa_symbol(bm_pud), P4D_TYPE_TABLE);
1273 		pudp = fixmap_pud(addr);
1274 	}
1275 	if (pud_none(READ_ONCE(*pudp)))
1276 		__pud_populate(pudp, __pa_symbol(bm_pmd), PUD_TYPE_TABLE);
1277 	pmdp = fixmap_pmd(addr);
1278 	__pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
1279 
1280 	/*
1281 	 * The boot-ioremap range spans multiple pmds, for which
1282 	 * we are not prepared:
1283 	 */
1284 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1285 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1286 
1287 	if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
1288 	     || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
1289 		WARN_ON(1);
1290 		pr_warn("pmdp %p != %p, %p\n",
1291 			pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
1292 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
1293 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1294 			fix_to_virt(FIX_BTMAP_BEGIN));
1295 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1296 			fix_to_virt(FIX_BTMAP_END));
1297 
1298 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1299 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1300 	}
1301 }
1302 
1303 /*
1304  * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
1305  * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
1306  */
1307 void __set_fixmap(enum fixed_addresses idx,
1308 			       phys_addr_t phys, pgprot_t flags)
1309 {
1310 	unsigned long addr = __fix_to_virt(idx);
1311 	pte_t *ptep;
1312 
1313 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
1314 
1315 	ptep = fixmap_pte(addr);
1316 
1317 	if (pgprot_val(flags)) {
1318 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
1319 	} else {
1320 		pte_clear(&init_mm, addr, ptep);
1321 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
1322 	}
1323 }
1324 
1325 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
1326 {
1327 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
1328 	int offset;
1329 	void *dt_virt;
1330 
1331 	/*
1332 	 * Check whether the physical FDT address is set and meets the minimum
1333 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
1334 	 * at least 8 bytes so that we can always access the magic and size
1335 	 * fields of the FDT header after mapping the first chunk, double check
1336 	 * here if that is indeed the case.
1337 	 */
1338 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
1339 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
1340 		return NULL;
1341 
1342 	/*
1343 	 * Make sure that the FDT region can be mapped without the need to
1344 	 * allocate additional translation table pages, so that it is safe
1345 	 * to call create_mapping_noalloc() this early.
1346 	 *
1347 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
1348 	 * be in the same PMD as the rest of the fixmap.
1349 	 * On 4k pages, we'll use section mappings for the FDT so we only
1350 	 * have to be in the same PUD.
1351 	 */
1352 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
1353 
1354 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
1355 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
1356 
1357 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
1358 	dt_virt = (void *)dt_virt_base + offset;
1359 
1360 	/* map the first chunk so we can read the size from the header */
1361 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
1362 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
1363 
1364 	if (fdt_magic(dt_virt) != FDT_MAGIC)
1365 		return NULL;
1366 
1367 	*size = fdt_totalsize(dt_virt);
1368 	if (*size > MAX_FDT_SIZE)
1369 		return NULL;
1370 
1371 	if (offset + *size > SWAPPER_BLOCK_SIZE)
1372 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
1373 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
1374 
1375 	return dt_virt;
1376 }
1377 
1378 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1379 {
1380 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1381 
1382 	/* Only allow permission changes for now */
1383 	if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1384 				   pud_val(new_pud)))
1385 		return 0;
1386 
1387 	VM_BUG_ON(phys & ~PUD_MASK);
1388 	set_pud(pudp, new_pud);
1389 	return 1;
1390 }
1391 
1392 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1393 {
1394 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1395 
1396 	/* Only allow permission changes for now */
1397 	if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1398 				   pmd_val(new_pmd)))
1399 		return 0;
1400 
1401 	VM_BUG_ON(phys & ~PMD_MASK);
1402 	set_pmd(pmdp, new_pmd);
1403 	return 1;
1404 }
1405 
1406 int pud_clear_huge(pud_t *pudp)
1407 {
1408 	if (!pud_sect(READ_ONCE(*pudp)))
1409 		return 0;
1410 	pud_clear(pudp);
1411 	return 1;
1412 }
1413 
1414 int pmd_clear_huge(pmd_t *pmdp)
1415 {
1416 	if (!pmd_sect(READ_ONCE(*pmdp)))
1417 		return 0;
1418 	pmd_clear(pmdp);
1419 	return 1;
1420 }
1421 
1422 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1423 {
1424 	pte_t *table;
1425 	pmd_t pmd;
1426 
1427 	pmd = READ_ONCE(*pmdp);
1428 
1429 	if (!pmd_table(pmd)) {
1430 		VM_WARN_ON(1);
1431 		return 1;
1432 	}
1433 
1434 	table = pte_offset_kernel(pmdp, addr);
1435 	pmd_clear(pmdp);
1436 	__flush_tlb_kernel_pgtable(addr);
1437 	pte_free_kernel(NULL, table);
1438 	return 1;
1439 }
1440 
1441 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1442 {
1443 	pmd_t *table;
1444 	pmd_t *pmdp;
1445 	pud_t pud;
1446 	unsigned long next, end;
1447 
1448 	pud = READ_ONCE(*pudp);
1449 
1450 	if (!pud_table(pud)) {
1451 		VM_WARN_ON(1);
1452 		return 1;
1453 	}
1454 
1455 	table = pmd_offset(pudp, addr);
1456 	pmdp = table;
1457 	next = addr;
1458 	end = addr + PUD_SIZE;
1459 	do {
1460 		pmd_free_pte_page(pmdp, next);
1461 	} while (pmdp++, next += PMD_SIZE, next != end);
1462 
1463 	pud_clear(pudp);
1464 	__flush_tlb_kernel_pgtable(addr);
1465 	pmd_free(NULL, table);
1466 	return 1;
1467 }
1468 
1469 #ifdef CONFIG_MEMORY_HOTPLUG
1470 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1471 {
1472 	unsigned long end = start + size;
1473 
1474 	WARN_ON(pgdir != init_mm.pgd);
1475 	WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1476 
1477 	unmap_hotplug_range(start, end, false, NULL);
1478 	free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1479 }
1480 
1481 struct range arch_get_mappable_range(void)
1482 {
1483 	struct range mhp_range;
1484 	u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1485 	u64 end_linear_pa = __pa(PAGE_END - 1);
1486 
1487 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1488 		/*
1489 		 * Check for a wrap, it is possible because of randomized linear
1490 		 * mapping the start physical address is actually bigger than
1491 		 * the end physical address. In this case set start to zero
1492 		 * because [0, end_linear_pa] range must still be able to cover
1493 		 * all addressable physical addresses.
1494 		 */
1495 		if (start_linear_pa > end_linear_pa)
1496 			start_linear_pa = 0;
1497 	}
1498 
1499 	WARN_ON(start_linear_pa > end_linear_pa);
1500 
1501 	/*
1502 	 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1503 	 * accommodating both its ends but excluding PAGE_END. Max physical
1504 	 * range which can be mapped inside this linear mapping range, must
1505 	 * also be derived from its end points.
1506 	 */
1507 	mhp_range.start = start_linear_pa;
1508 	mhp_range.end =  end_linear_pa;
1509 
1510 	return mhp_range;
1511 }
1512 
1513 int arch_add_memory(int nid, u64 start, u64 size,
1514 		    struct mhp_params *params)
1515 {
1516 	int ret, flags = NO_EXEC_MAPPINGS;
1517 
1518 	VM_BUG_ON(!mhp_range_allowed(start, size, true));
1519 
1520 	/*
1521 	 * KFENCE requires linear map to be mapped at page granularity, so that
1522 	 * it is possible to protect/unprotect single pages in the KFENCE pool.
1523 	 */
1524 	if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
1525 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1526 
1527 	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1528 			     size, params->pgprot, __pgd_pgtable_alloc,
1529 			     flags);
1530 
1531 	memblock_clear_nomap(start, size);
1532 
1533 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1534 			   params);
1535 	if (ret)
1536 		__remove_pgd_mapping(swapper_pg_dir,
1537 				     __phys_to_virt(start), size);
1538 	else {
1539 		max_pfn = PFN_UP(start + size);
1540 		max_low_pfn = max_pfn;
1541 	}
1542 
1543 	return ret;
1544 }
1545 
1546 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1547 {
1548 	unsigned long start_pfn = start >> PAGE_SHIFT;
1549 	unsigned long nr_pages = size >> PAGE_SHIFT;
1550 
1551 	__remove_pages(start_pfn, nr_pages, altmap);
1552 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1553 }
1554 
1555 /*
1556  * This memory hotplug notifier helps prevent boot memory from being
1557  * inadvertently removed as it blocks pfn range offlining process in
1558  * __offline_pages(). Hence this prevents both offlining as well as
1559  * removal process for boot memory which is initially always online.
1560  * In future if and when boot memory could be removed, this notifier
1561  * should be dropped and free_hotplug_page_range() should handle any
1562  * reserved pages allocated during boot.
1563  */
1564 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1565 					   unsigned long action, void *data)
1566 {
1567 	struct mem_section *ms;
1568 	struct memory_notify *arg = data;
1569 	unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1570 	unsigned long pfn = arg->start_pfn;
1571 
1572 	if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1573 		return NOTIFY_OK;
1574 
1575 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1576 		unsigned long start = PFN_PHYS(pfn);
1577 		unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1578 
1579 		ms = __pfn_to_section(pfn);
1580 		if (!early_section(ms))
1581 			continue;
1582 
1583 		if (action == MEM_GOING_OFFLINE) {
1584 			/*
1585 			 * Boot memory removal is not supported. Prevent
1586 			 * it via blocking any attempted offline request
1587 			 * for the boot memory and just report it.
1588 			 */
1589 			pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1590 			return NOTIFY_BAD;
1591 		} else if (action == MEM_OFFLINE) {
1592 			/*
1593 			 * This should have never happened. Boot memory
1594 			 * offlining should have been prevented by this
1595 			 * very notifier. Probably some memory removal
1596 			 * procedure might have changed which would then
1597 			 * require further debug.
1598 			 */
1599 			pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1600 
1601 			/*
1602 			 * Core memory hotplug does not process a return
1603 			 * code from the notifier for MEM_OFFLINE events.
1604 			 * The error condition has been reported. Return
1605 			 * from here as if ignored.
1606 			 */
1607 			return NOTIFY_DONE;
1608 		}
1609 	}
1610 	return NOTIFY_OK;
1611 }
1612 
1613 static struct notifier_block prevent_bootmem_remove_nb = {
1614 	.notifier_call = prevent_bootmem_remove_notifier,
1615 };
1616 
1617 /*
1618  * This ensures that boot memory sections on the platform are online
1619  * from early boot. Memory sections could not be prevented from being
1620  * offlined, unless for some reason they are not online to begin with.
1621  * This helps validate the basic assumption on which the above memory
1622  * event notifier works to prevent boot memory section offlining and
1623  * its possible removal.
1624  */
1625 static void validate_bootmem_online(void)
1626 {
1627 	phys_addr_t start, end, addr;
1628 	struct mem_section *ms;
1629 	u64 i;
1630 
1631 	/*
1632 	 * Scanning across all memblock might be expensive
1633 	 * on some big memory systems. Hence enable this
1634 	 * validation only with DEBUG_VM.
1635 	 */
1636 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
1637 		return;
1638 
1639 	for_each_mem_range(i, &start, &end) {
1640 		for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1641 			ms = __pfn_to_section(PHYS_PFN(addr));
1642 
1643 			/*
1644 			 * All memory ranges in the system at this point
1645 			 * should have been marked as early sections.
1646 			 */
1647 			WARN_ON(!early_section(ms));
1648 
1649 			/*
1650 			 * Memory notifier mechanism here to prevent boot
1651 			 * memory offlining depends on the fact that each
1652 			 * early section memory on the system is initially
1653 			 * online. Otherwise a given memory section which
1654 			 * is already offline will be overlooked and can
1655 			 * be removed completely. Call out such sections.
1656 			 */
1657 			if (!online_section(ms))
1658 				pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1659 					addr, addr + (1UL << PA_SECTION_SHIFT));
1660 		}
1661 	}
1662 }
1663 
1664 static int __init prevent_bootmem_remove_init(void)
1665 {
1666 	int ret = 0;
1667 
1668 	if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1669 		return ret;
1670 
1671 	validate_bootmem_online();
1672 	ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1673 	if (ret)
1674 		pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1675 
1676 	return ret;
1677 }
1678 early_initcall(prevent_bootmem_remove_init);
1679 #endif
1680