xref: /linux/arch/arm64/mm/mmu.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
1 /*
2  * Based on arch/arm/mm/mmu.c
3  *
4  * Copyright (C) 1995-2005 Russell King
5  * Copyright (C) 2012 ARM Ltd.
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include <linux/cache.h>
21 #include <linux/export.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/init.h>
25 #include <linux/libfdt.h>
26 #include <linux/mman.h>
27 #include <linux/nodemask.h>
28 #include <linux/memblock.h>
29 #include <linux/fs.h>
30 #include <linux/io.h>
31 #include <linux/slab.h>
32 #include <linux/stop_machine.h>
33 
34 #include <asm/barrier.h>
35 #include <asm/cputype.h>
36 #include <asm/fixmap.h>
37 #include <asm/kasan.h>
38 #include <asm/kernel-pgtable.h>
39 #include <asm/sections.h>
40 #include <asm/setup.h>
41 #include <asm/sizes.h>
42 #include <asm/tlb.h>
43 #include <asm/memblock.h>
44 #include <asm/mmu_context.h>
45 
46 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
47 
48 u64 kimage_voffset __ro_after_init;
49 EXPORT_SYMBOL(kimage_voffset);
50 
51 /*
52  * Empty_zero_page is a special page that is used for zero-initialized data
53  * and COW.
54  */
55 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
56 EXPORT_SYMBOL(empty_zero_page);
57 
58 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
59 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
60 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
61 
62 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
63 			      unsigned long size, pgprot_t vma_prot)
64 {
65 	if (!pfn_valid(pfn))
66 		return pgprot_noncached(vma_prot);
67 	else if (file->f_flags & O_SYNC)
68 		return pgprot_writecombine(vma_prot);
69 	return vma_prot;
70 }
71 EXPORT_SYMBOL(phys_mem_access_prot);
72 
73 static phys_addr_t __init early_pgtable_alloc(void)
74 {
75 	phys_addr_t phys;
76 	void *ptr;
77 
78 	phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
79 
80 	/*
81 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
82 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
83 	 * any level of table.
84 	 */
85 	ptr = pte_set_fixmap(phys);
86 
87 	memset(ptr, 0, PAGE_SIZE);
88 
89 	/*
90 	 * Implicit barriers also ensure the zeroed page is visible to the page
91 	 * table walker
92 	 */
93 	pte_clear_fixmap();
94 
95 	return phys;
96 }
97 
98 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
99 				  unsigned long end, unsigned long pfn,
100 				  pgprot_t prot,
101 				  phys_addr_t (*pgtable_alloc)(void))
102 {
103 	pte_t *pte;
104 
105 	BUG_ON(pmd_sect(*pmd));
106 	if (pmd_none(*pmd)) {
107 		phys_addr_t pte_phys;
108 		BUG_ON(!pgtable_alloc);
109 		pte_phys = pgtable_alloc();
110 		pte = pte_set_fixmap(pte_phys);
111 		__pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
112 		pte_clear_fixmap();
113 	}
114 	BUG_ON(pmd_bad(*pmd));
115 
116 	pte = pte_set_fixmap_offset(pmd, addr);
117 	do {
118 		set_pte(pte, pfn_pte(pfn, prot));
119 		pfn++;
120 	} while (pte++, addr += PAGE_SIZE, addr != end);
121 
122 	pte_clear_fixmap();
123 }
124 
125 static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
126 				  phys_addr_t phys, pgprot_t prot,
127 				  phys_addr_t (*pgtable_alloc)(void),
128 				  bool allow_block_mappings)
129 {
130 	pmd_t *pmd;
131 	unsigned long next;
132 
133 	/*
134 	 * Check for initial section mappings in the pgd/pud and remove them.
135 	 */
136 	BUG_ON(pud_sect(*pud));
137 	if (pud_none(*pud)) {
138 		phys_addr_t pmd_phys;
139 		BUG_ON(!pgtable_alloc);
140 		pmd_phys = pgtable_alloc();
141 		pmd = pmd_set_fixmap(pmd_phys);
142 		__pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
143 		pmd_clear_fixmap();
144 	}
145 	BUG_ON(pud_bad(*pud));
146 
147 	pmd = pmd_set_fixmap_offset(pud, addr);
148 	do {
149 		next = pmd_addr_end(addr, end);
150 		/* try section mapping first */
151 		if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
152 		      allow_block_mappings) {
153 			pmd_t old_pmd =*pmd;
154 			pmd_set_huge(pmd, phys, prot);
155 			/*
156 			 * Check for previous table entries created during
157 			 * boot (__create_page_tables) and flush them.
158 			 */
159 			if (!pmd_none(old_pmd)) {
160 				flush_tlb_all();
161 				if (pmd_table(old_pmd)) {
162 					phys_addr_t table = pmd_page_paddr(old_pmd);
163 					if (!WARN_ON_ONCE(slab_is_available()))
164 						memblock_free(table, PAGE_SIZE);
165 				}
166 			}
167 		} else {
168 			alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
169 				       prot, pgtable_alloc);
170 		}
171 		phys += next - addr;
172 	} while (pmd++, addr = next, addr != end);
173 
174 	pmd_clear_fixmap();
175 }
176 
177 static inline bool use_1G_block(unsigned long addr, unsigned long next,
178 			unsigned long phys)
179 {
180 	if (PAGE_SHIFT != 12)
181 		return false;
182 
183 	if (((addr | next | phys) & ~PUD_MASK) != 0)
184 		return false;
185 
186 	return true;
187 }
188 
189 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
190 				  phys_addr_t phys, pgprot_t prot,
191 				  phys_addr_t (*pgtable_alloc)(void),
192 				  bool allow_block_mappings)
193 {
194 	pud_t *pud;
195 	unsigned long next;
196 
197 	if (pgd_none(*pgd)) {
198 		phys_addr_t pud_phys;
199 		BUG_ON(!pgtable_alloc);
200 		pud_phys = pgtable_alloc();
201 		__pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
202 	}
203 	BUG_ON(pgd_bad(*pgd));
204 
205 	pud = pud_set_fixmap_offset(pgd, addr);
206 	do {
207 		next = pud_addr_end(addr, end);
208 
209 		/*
210 		 * For 4K granule only, attempt to put down a 1GB block
211 		 */
212 		if (use_1G_block(addr, next, phys) && allow_block_mappings) {
213 			pud_t old_pud = *pud;
214 			pud_set_huge(pud, phys, prot);
215 
216 			/*
217 			 * If we have an old value for a pud, it will
218 			 * be pointing to a pmd table that we no longer
219 			 * need (from swapper_pg_dir).
220 			 *
221 			 * Look up the old pmd table and free it.
222 			 */
223 			if (!pud_none(old_pud)) {
224 				flush_tlb_all();
225 				if (pud_table(old_pud)) {
226 					phys_addr_t table = pud_page_paddr(old_pud);
227 					if (!WARN_ON_ONCE(slab_is_available()))
228 						memblock_free(table, PAGE_SIZE);
229 				}
230 			}
231 		} else {
232 			alloc_init_pmd(pud, addr, next, phys, prot,
233 				       pgtable_alloc, allow_block_mappings);
234 		}
235 		phys += next - addr;
236 	} while (pud++, addr = next, addr != end);
237 
238 	pud_clear_fixmap();
239 }
240 
241 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
242 				 unsigned long virt, phys_addr_t size,
243 				 pgprot_t prot,
244 				 phys_addr_t (*pgtable_alloc)(void),
245 				 bool allow_block_mappings)
246 {
247 	unsigned long addr, length, end, next;
248 	pgd_t *pgd = pgd_offset_raw(pgdir, virt);
249 
250 	/*
251 	 * If the virtual and physical address don't have the same offset
252 	 * within a page, we cannot map the region as the caller expects.
253 	 */
254 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
255 		return;
256 
257 	phys &= PAGE_MASK;
258 	addr = virt & PAGE_MASK;
259 	length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
260 
261 	end = addr + length;
262 	do {
263 		next = pgd_addr_end(addr, end);
264 		alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
265 			       allow_block_mappings);
266 		phys += next - addr;
267 	} while (pgd++, addr = next, addr != end);
268 }
269 
270 static phys_addr_t pgd_pgtable_alloc(void)
271 {
272 	void *ptr = (void *)__get_free_page(PGALLOC_GFP);
273 	if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
274 		BUG();
275 
276 	/* Ensure the zeroed page is visible to the page table walker */
277 	dsb(ishst);
278 	return __pa(ptr);
279 }
280 
281 /*
282  * This function can only be used to modify existing table entries,
283  * without allocating new levels of table. Note that this permits the
284  * creation of new section or page entries.
285  */
286 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
287 				  phys_addr_t size, pgprot_t prot)
288 {
289 	if (virt < VMALLOC_START) {
290 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
291 			&phys, virt);
292 		return;
293 	}
294 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, true);
295 }
296 
297 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
298 			       unsigned long virt, phys_addr_t size,
299 			       pgprot_t prot, bool allow_block_mappings)
300 {
301 	BUG_ON(mm == &init_mm);
302 
303 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
304 			     pgd_pgtable_alloc, allow_block_mappings);
305 }
306 
307 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
308 				  phys_addr_t size, pgprot_t prot)
309 {
310 	if (virt < VMALLOC_START) {
311 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
312 			&phys, virt);
313 		return;
314 	}
315 
316 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
317 			     NULL, !debug_pagealloc_enabled());
318 }
319 
320 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
321 {
322 	unsigned long kernel_start = __pa(_text);
323 	unsigned long kernel_end = __pa(__init_begin);
324 
325 	/*
326 	 * Take care not to create a writable alias for the
327 	 * read-only text and rodata sections of the kernel image.
328 	 */
329 
330 	/* No overlap with the kernel text/rodata */
331 	if (end < kernel_start || start >= kernel_end) {
332 		__create_pgd_mapping(pgd, start, __phys_to_virt(start),
333 				     end - start, PAGE_KERNEL,
334 				     early_pgtable_alloc,
335 				     !debug_pagealloc_enabled());
336 		return;
337 	}
338 
339 	/*
340 	 * This block overlaps the kernel text/rodata mappings.
341 	 * Map the portion(s) which don't overlap.
342 	 */
343 	if (start < kernel_start)
344 		__create_pgd_mapping(pgd, start,
345 				     __phys_to_virt(start),
346 				     kernel_start - start, PAGE_KERNEL,
347 				     early_pgtable_alloc,
348 				     !debug_pagealloc_enabled());
349 	if (kernel_end < end)
350 		__create_pgd_mapping(pgd, kernel_end,
351 				     __phys_to_virt(kernel_end),
352 				     end - kernel_end, PAGE_KERNEL,
353 				     early_pgtable_alloc,
354 				     !debug_pagealloc_enabled());
355 
356 	/*
357 	 * Map the linear alias of the [_text, __init_begin) interval as
358 	 * read-only/non-executable. This makes the contents of the
359 	 * region accessible to subsystems such as hibernate, but
360 	 * protects it from inadvertent modification or execution.
361 	 */
362 	__create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
363 			     kernel_end - kernel_start, PAGE_KERNEL_RO,
364 			     early_pgtable_alloc, !debug_pagealloc_enabled());
365 }
366 
367 static void __init map_mem(pgd_t *pgd)
368 {
369 	struct memblock_region *reg;
370 
371 	/* map all the memory banks */
372 	for_each_memblock(memory, reg) {
373 		phys_addr_t start = reg->base;
374 		phys_addr_t end = start + reg->size;
375 
376 		if (start >= end)
377 			break;
378 		if (memblock_is_nomap(reg))
379 			continue;
380 
381 		__map_memblock(pgd, start, end);
382 	}
383 }
384 
385 void mark_rodata_ro(void)
386 {
387 	unsigned long section_size;
388 
389 	section_size = (unsigned long)_etext - (unsigned long)_text;
390 	create_mapping_late(__pa(_text), (unsigned long)_text,
391 			    section_size, PAGE_KERNEL_ROX);
392 	/*
393 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
394 	 * to cover NOTES and EXCEPTION_TABLE.
395 	 */
396 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
397 	create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
398 			    section_size, PAGE_KERNEL_RO);
399 }
400 
401 static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
402 				      pgprot_t prot, struct vm_struct *vma)
403 {
404 	phys_addr_t pa_start = __pa(va_start);
405 	unsigned long size = va_end - va_start;
406 
407 	BUG_ON(!PAGE_ALIGNED(pa_start));
408 	BUG_ON(!PAGE_ALIGNED(size));
409 
410 	__create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
411 			     early_pgtable_alloc, !debug_pagealloc_enabled());
412 
413 	vma->addr	= va_start;
414 	vma->phys_addr	= pa_start;
415 	vma->size	= size;
416 	vma->flags	= VM_MAP;
417 	vma->caller	= __builtin_return_address(0);
418 
419 	vm_area_add_early(vma);
420 }
421 
422 /*
423  * Create fine-grained mappings for the kernel.
424  */
425 static void __init map_kernel(pgd_t *pgd)
426 {
427 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
428 
429 	map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
430 	map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
431 	map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
432 			   &vmlinux_init);
433 	map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
434 
435 	if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
436 		/*
437 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
438 		 * live in the carveout for the swapper_pg_dir. We can simply
439 		 * re-use the existing dir for the fixmap.
440 		 */
441 		set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
442 			*pgd_offset_k(FIXADDR_START));
443 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
444 		/*
445 		 * The fixmap shares its top level pgd entry with the kernel
446 		 * mapping. This can really only occur when we are running
447 		 * with 16k/4 levels, so we can simply reuse the pud level
448 		 * entry instead.
449 		 */
450 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
451 		set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
452 			__pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
453 		pud_clear_fixmap();
454 	} else {
455 		BUG();
456 	}
457 
458 	kasan_copy_shadow(pgd);
459 }
460 
461 /*
462  * paging_init() sets up the page tables, initialises the zone memory
463  * maps and sets up the zero page.
464  */
465 void __init paging_init(void)
466 {
467 	phys_addr_t pgd_phys = early_pgtable_alloc();
468 	pgd_t *pgd = pgd_set_fixmap(pgd_phys);
469 
470 	map_kernel(pgd);
471 	map_mem(pgd);
472 
473 	/*
474 	 * We want to reuse the original swapper_pg_dir so we don't have to
475 	 * communicate the new address to non-coherent secondaries in
476 	 * secondary_entry, and so cpu_switch_mm can generate the address with
477 	 * adrp+add rather than a load from some global variable.
478 	 *
479 	 * To do this we need to go via a temporary pgd.
480 	 */
481 	cpu_replace_ttbr1(__va(pgd_phys));
482 	memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
483 	cpu_replace_ttbr1(swapper_pg_dir);
484 
485 	pgd_clear_fixmap();
486 	memblock_free(pgd_phys, PAGE_SIZE);
487 
488 	/*
489 	 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
490 	 * allocated with it.
491 	 */
492 	memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
493 		      SWAPPER_DIR_SIZE - PAGE_SIZE);
494 }
495 
496 /*
497  * Check whether a kernel address is valid (derived from arch/x86/).
498  */
499 int kern_addr_valid(unsigned long addr)
500 {
501 	pgd_t *pgd;
502 	pud_t *pud;
503 	pmd_t *pmd;
504 	pte_t *pte;
505 
506 	if ((((long)addr) >> VA_BITS) != -1UL)
507 		return 0;
508 
509 	pgd = pgd_offset_k(addr);
510 	if (pgd_none(*pgd))
511 		return 0;
512 
513 	pud = pud_offset(pgd, addr);
514 	if (pud_none(*pud))
515 		return 0;
516 
517 	if (pud_sect(*pud))
518 		return pfn_valid(pud_pfn(*pud));
519 
520 	pmd = pmd_offset(pud, addr);
521 	if (pmd_none(*pmd))
522 		return 0;
523 
524 	if (pmd_sect(*pmd))
525 		return pfn_valid(pmd_pfn(*pmd));
526 
527 	pte = pte_offset_kernel(pmd, addr);
528 	if (pte_none(*pte))
529 		return 0;
530 
531 	return pfn_valid(pte_pfn(*pte));
532 }
533 #ifdef CONFIG_SPARSEMEM_VMEMMAP
534 #if !ARM64_SWAPPER_USES_SECTION_MAPS
535 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
536 {
537 	return vmemmap_populate_basepages(start, end, node);
538 }
539 #else	/* !ARM64_SWAPPER_USES_SECTION_MAPS */
540 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
541 {
542 	unsigned long addr = start;
543 	unsigned long next;
544 	pgd_t *pgd;
545 	pud_t *pud;
546 	pmd_t *pmd;
547 
548 	do {
549 		next = pmd_addr_end(addr, end);
550 
551 		pgd = vmemmap_pgd_populate(addr, node);
552 		if (!pgd)
553 			return -ENOMEM;
554 
555 		pud = vmemmap_pud_populate(pgd, addr, node);
556 		if (!pud)
557 			return -ENOMEM;
558 
559 		pmd = pmd_offset(pud, addr);
560 		if (pmd_none(*pmd)) {
561 			void *p = NULL;
562 
563 			p = vmemmap_alloc_block_buf(PMD_SIZE, node);
564 			if (!p)
565 				return -ENOMEM;
566 
567 			set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
568 		} else
569 			vmemmap_verify((pte_t *)pmd, node, addr, next);
570 	} while (addr = next, addr != end);
571 
572 	return 0;
573 }
574 #endif	/* CONFIG_ARM64_64K_PAGES */
575 void vmemmap_free(unsigned long start, unsigned long end)
576 {
577 }
578 #endif	/* CONFIG_SPARSEMEM_VMEMMAP */
579 
580 static inline pud_t * fixmap_pud(unsigned long addr)
581 {
582 	pgd_t *pgd = pgd_offset_k(addr);
583 
584 	BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
585 
586 	return pud_offset_kimg(pgd, addr);
587 }
588 
589 static inline pmd_t * fixmap_pmd(unsigned long addr)
590 {
591 	pud_t *pud = fixmap_pud(addr);
592 
593 	BUG_ON(pud_none(*pud) || pud_bad(*pud));
594 
595 	return pmd_offset_kimg(pud, addr);
596 }
597 
598 static inline pte_t * fixmap_pte(unsigned long addr)
599 {
600 	return &bm_pte[pte_index(addr)];
601 }
602 
603 void __init early_fixmap_init(void)
604 {
605 	pgd_t *pgd;
606 	pud_t *pud;
607 	pmd_t *pmd;
608 	unsigned long addr = FIXADDR_START;
609 
610 	pgd = pgd_offset_k(addr);
611 	if (CONFIG_PGTABLE_LEVELS > 3 &&
612 	    !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
613 		/*
614 		 * We only end up here if the kernel mapping and the fixmap
615 		 * share the top level pgd entry, which should only happen on
616 		 * 16k/4 levels configurations.
617 		 */
618 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
619 		pud = pud_offset_kimg(pgd, addr);
620 	} else {
621 		pgd_populate(&init_mm, pgd, bm_pud);
622 		pud = fixmap_pud(addr);
623 	}
624 	pud_populate(&init_mm, pud, bm_pmd);
625 	pmd = fixmap_pmd(addr);
626 	pmd_populate_kernel(&init_mm, pmd, bm_pte);
627 
628 	/*
629 	 * The boot-ioremap range spans multiple pmds, for which
630 	 * we are not prepared:
631 	 */
632 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
633 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
634 
635 	if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
636 	     || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
637 		WARN_ON(1);
638 		pr_warn("pmd %p != %p, %p\n",
639 			pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
640 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
641 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
642 			fix_to_virt(FIX_BTMAP_BEGIN));
643 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
644 			fix_to_virt(FIX_BTMAP_END));
645 
646 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
647 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
648 	}
649 }
650 
651 void __set_fixmap(enum fixed_addresses idx,
652 			       phys_addr_t phys, pgprot_t flags)
653 {
654 	unsigned long addr = __fix_to_virt(idx);
655 	pte_t *pte;
656 
657 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
658 
659 	pte = fixmap_pte(addr);
660 
661 	if (pgprot_val(flags)) {
662 		set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
663 	} else {
664 		pte_clear(&init_mm, addr, pte);
665 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
666 	}
667 }
668 
669 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
670 {
671 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
672 	int offset;
673 	void *dt_virt;
674 
675 	/*
676 	 * Check whether the physical FDT address is set and meets the minimum
677 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
678 	 * at least 8 bytes so that we can always access the magic and size
679 	 * fields of the FDT header after mapping the first chunk, double check
680 	 * here if that is indeed the case.
681 	 */
682 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
683 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
684 		return NULL;
685 
686 	/*
687 	 * Make sure that the FDT region can be mapped without the need to
688 	 * allocate additional translation table pages, so that it is safe
689 	 * to call create_mapping_noalloc() this early.
690 	 *
691 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
692 	 * be in the same PMD as the rest of the fixmap.
693 	 * On 4k pages, we'll use section mappings for the FDT so we only
694 	 * have to be in the same PUD.
695 	 */
696 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
697 
698 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
699 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
700 
701 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
702 	dt_virt = (void *)dt_virt_base + offset;
703 
704 	/* map the first chunk so we can read the size from the header */
705 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
706 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
707 
708 	if (fdt_magic(dt_virt) != FDT_MAGIC)
709 		return NULL;
710 
711 	*size = fdt_totalsize(dt_virt);
712 	if (*size > MAX_FDT_SIZE)
713 		return NULL;
714 
715 	if (offset + *size > SWAPPER_BLOCK_SIZE)
716 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
717 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
718 
719 	return dt_virt;
720 }
721 
722 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
723 {
724 	void *dt_virt;
725 	int size;
726 
727 	dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
728 	if (!dt_virt)
729 		return NULL;
730 
731 	memblock_reserve(dt_phys, size);
732 	return dt_virt;
733 }
734 
735 int __init arch_ioremap_pud_supported(void)
736 {
737 	/* only 4k granule supports level 1 block mappings */
738 	return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
739 }
740 
741 int __init arch_ioremap_pmd_supported(void)
742 {
743 	return 1;
744 }
745 
746 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
747 {
748 	BUG_ON(phys & ~PUD_MASK);
749 	set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
750 	return 1;
751 }
752 
753 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
754 {
755 	BUG_ON(phys & ~PMD_MASK);
756 	set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
757 	return 1;
758 }
759 
760 int pud_clear_huge(pud_t *pud)
761 {
762 	if (!pud_sect(*pud))
763 		return 0;
764 	pud_clear(pud);
765 	return 1;
766 }
767 
768 int pmd_clear_huge(pmd_t *pmd)
769 {
770 	if (!pmd_sect(*pmd))
771 		return 0;
772 	pmd_clear(pmd);
773 	return 1;
774 }
775