xref: /linux/arch/arm64/mm/mmu.c (revision 74ce1896c6c65b2f8cccbf59162d542988835835)
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/ioport.h>
26 #include <linux/kexec.h>
27 #include <linux/libfdt.h>
28 #include <linux/mman.h>
29 #include <linux/nodemask.h>
30 #include <linux/memblock.h>
31 #include <linux/fs.h>
32 #include <linux/io.h>
33 #include <linux/mm.h>
34 #include <linux/vmalloc.h>
35 
36 #include <asm/barrier.h>
37 #include <asm/cputype.h>
38 #include <asm/fixmap.h>
39 #include <asm/kasan.h>
40 #include <asm/kernel-pgtable.h>
41 #include <asm/sections.h>
42 #include <asm/setup.h>
43 #include <asm/sizes.h>
44 #include <asm/tlb.h>
45 #include <asm/memblock.h>
46 #include <asm/mmu_context.h>
47 #include <asm/ptdump.h>
48 
49 #define NO_BLOCK_MAPPINGS	BIT(0)
50 #define NO_CONT_MAPPINGS	BIT(1)
51 
52 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
53 
54 u64 kimage_voffset __ro_after_init;
55 EXPORT_SYMBOL(kimage_voffset);
56 
57 /*
58  * Empty_zero_page is a special page that is used for zero-initialized data
59  * and COW.
60  */
61 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
62 EXPORT_SYMBOL(empty_zero_page);
63 
64 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
65 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
66 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
67 
68 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
69 			      unsigned long size, pgprot_t vma_prot)
70 {
71 	if (!pfn_valid(pfn))
72 		return pgprot_noncached(vma_prot);
73 	else if (file->f_flags & O_SYNC)
74 		return pgprot_writecombine(vma_prot);
75 	return vma_prot;
76 }
77 EXPORT_SYMBOL(phys_mem_access_prot);
78 
79 static phys_addr_t __init early_pgtable_alloc(void)
80 {
81 	phys_addr_t phys;
82 	void *ptr;
83 
84 	phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
85 
86 	/*
87 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
88 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
89 	 * any level of table.
90 	 */
91 	ptr = pte_set_fixmap(phys);
92 
93 	memset(ptr, 0, PAGE_SIZE);
94 
95 	/*
96 	 * Implicit barriers also ensure the zeroed page is visible to the page
97 	 * table walker
98 	 */
99 	pte_clear_fixmap();
100 
101 	return phys;
102 }
103 
104 static bool pgattr_change_is_safe(u64 old, u64 new)
105 {
106 	/*
107 	 * The following mapping attributes may be updated in live
108 	 * kernel mappings without the need for break-before-make.
109 	 */
110 	static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE;
111 
112 	/* creating or taking down mappings is always safe */
113 	if (old == 0 || new == 0)
114 		return true;
115 
116 	/* live contiguous mappings may not be manipulated at all */
117 	if ((old | new) & PTE_CONT)
118 		return false;
119 
120 	return ((old ^ new) & ~mask) == 0;
121 }
122 
123 static void init_pte(pmd_t *pmd, unsigned long addr, unsigned long end,
124 		     phys_addr_t phys, pgprot_t prot)
125 {
126 	pte_t *pte;
127 
128 	pte = pte_set_fixmap_offset(pmd, addr);
129 	do {
130 		pte_t old_pte = *pte;
131 
132 		set_pte(pte, pfn_pte(__phys_to_pfn(phys), prot));
133 
134 		/*
135 		 * After the PTE entry has been populated once, we
136 		 * only allow updates to the permission attributes.
137 		 */
138 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte)));
139 
140 		phys += PAGE_SIZE;
141 	} while (pte++, addr += PAGE_SIZE, addr != end);
142 
143 	pte_clear_fixmap();
144 }
145 
146 static void alloc_init_cont_pte(pmd_t *pmd, unsigned long addr,
147 				unsigned long end, phys_addr_t phys,
148 				pgprot_t prot,
149 				phys_addr_t (*pgtable_alloc)(void),
150 				int flags)
151 {
152 	unsigned long next;
153 
154 	BUG_ON(pmd_sect(*pmd));
155 	if (pmd_none(*pmd)) {
156 		phys_addr_t pte_phys;
157 		BUG_ON(!pgtable_alloc);
158 		pte_phys = pgtable_alloc();
159 		__pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
160 	}
161 	BUG_ON(pmd_bad(*pmd));
162 
163 	do {
164 		pgprot_t __prot = prot;
165 
166 		next = pte_cont_addr_end(addr, end);
167 
168 		/* use a contiguous mapping if the range is suitably aligned */
169 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
170 		    (flags & NO_CONT_MAPPINGS) == 0)
171 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
172 
173 		init_pte(pmd, addr, next, phys, __prot);
174 
175 		phys += next - addr;
176 	} while (addr = next, addr != end);
177 }
178 
179 static void init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
180 		     phys_addr_t phys, pgprot_t prot,
181 		     phys_addr_t (*pgtable_alloc)(void), int flags)
182 {
183 	unsigned long next;
184 	pmd_t *pmd;
185 
186 	pmd = pmd_set_fixmap_offset(pud, addr);
187 	do {
188 		pmd_t old_pmd = *pmd;
189 
190 		next = pmd_addr_end(addr, end);
191 
192 		/* try section mapping first */
193 		if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
194 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
195 			pmd_set_huge(pmd, phys, prot);
196 
197 			/*
198 			 * After the PMD entry has been populated once, we
199 			 * only allow updates to the permission attributes.
200 			 */
201 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
202 						      pmd_val(*pmd)));
203 		} else {
204 			alloc_init_cont_pte(pmd, addr, next, phys, prot,
205 					    pgtable_alloc, flags);
206 
207 			BUG_ON(pmd_val(old_pmd) != 0 &&
208 			       pmd_val(old_pmd) != pmd_val(*pmd));
209 		}
210 		phys += next - addr;
211 	} while (pmd++, addr = next, addr != end);
212 
213 	pmd_clear_fixmap();
214 }
215 
216 static void alloc_init_cont_pmd(pud_t *pud, unsigned long addr,
217 				unsigned long end, phys_addr_t phys,
218 				pgprot_t prot,
219 				phys_addr_t (*pgtable_alloc)(void), int flags)
220 {
221 	unsigned long next;
222 
223 	/*
224 	 * Check for initial section mappings in the pgd/pud.
225 	 */
226 	BUG_ON(pud_sect(*pud));
227 	if (pud_none(*pud)) {
228 		phys_addr_t pmd_phys;
229 		BUG_ON(!pgtable_alloc);
230 		pmd_phys = pgtable_alloc();
231 		__pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
232 	}
233 	BUG_ON(pud_bad(*pud));
234 
235 	do {
236 		pgprot_t __prot = prot;
237 
238 		next = pmd_cont_addr_end(addr, end);
239 
240 		/* use a contiguous mapping if the range is suitably aligned */
241 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
242 		    (flags & NO_CONT_MAPPINGS) == 0)
243 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
244 
245 		init_pmd(pud, addr, next, phys, __prot, pgtable_alloc, flags);
246 
247 		phys += next - addr;
248 	} while (addr = next, addr != end);
249 }
250 
251 static inline bool use_1G_block(unsigned long addr, unsigned long next,
252 			unsigned long phys)
253 {
254 	if (PAGE_SHIFT != 12)
255 		return false;
256 
257 	if (((addr | next | phys) & ~PUD_MASK) != 0)
258 		return false;
259 
260 	return true;
261 }
262 
263 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
264 				  phys_addr_t phys, pgprot_t prot,
265 				  phys_addr_t (*pgtable_alloc)(void),
266 				  int flags)
267 {
268 	pud_t *pud;
269 	unsigned long next;
270 
271 	if (pgd_none(*pgd)) {
272 		phys_addr_t pud_phys;
273 		BUG_ON(!pgtable_alloc);
274 		pud_phys = pgtable_alloc();
275 		__pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
276 	}
277 	BUG_ON(pgd_bad(*pgd));
278 
279 	pud = pud_set_fixmap_offset(pgd, addr);
280 	do {
281 		pud_t old_pud = *pud;
282 
283 		next = pud_addr_end(addr, end);
284 
285 		/*
286 		 * For 4K granule only, attempt to put down a 1GB block
287 		 */
288 		if (use_1G_block(addr, next, phys) &&
289 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
290 			pud_set_huge(pud, phys, prot);
291 
292 			/*
293 			 * After the PUD entry has been populated once, we
294 			 * only allow updates to the permission attributes.
295 			 */
296 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
297 						      pud_val(*pud)));
298 		} else {
299 			alloc_init_cont_pmd(pud, addr, next, phys, prot,
300 					    pgtable_alloc, flags);
301 
302 			BUG_ON(pud_val(old_pud) != 0 &&
303 			       pud_val(old_pud) != pud_val(*pud));
304 		}
305 		phys += next - addr;
306 	} while (pud++, addr = next, addr != end);
307 
308 	pud_clear_fixmap();
309 }
310 
311 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
312 				 unsigned long virt, phys_addr_t size,
313 				 pgprot_t prot,
314 				 phys_addr_t (*pgtable_alloc)(void),
315 				 int flags)
316 {
317 	unsigned long addr, length, end, next;
318 	pgd_t *pgd = pgd_offset_raw(pgdir, virt);
319 
320 	/*
321 	 * If the virtual and physical address don't have the same offset
322 	 * within a page, we cannot map the region as the caller expects.
323 	 */
324 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
325 		return;
326 
327 	phys &= PAGE_MASK;
328 	addr = virt & PAGE_MASK;
329 	length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
330 
331 	end = addr + length;
332 	do {
333 		next = pgd_addr_end(addr, end);
334 		alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
335 			       flags);
336 		phys += next - addr;
337 	} while (pgd++, addr = next, addr != end);
338 }
339 
340 static phys_addr_t pgd_pgtable_alloc(void)
341 {
342 	void *ptr = (void *)__get_free_page(PGALLOC_GFP);
343 	if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
344 		BUG();
345 
346 	/* Ensure the zeroed page is visible to the page table walker */
347 	dsb(ishst);
348 	return __pa(ptr);
349 }
350 
351 /*
352  * This function can only be used to modify existing table entries,
353  * without allocating new levels of table. Note that this permits the
354  * creation of new section or page entries.
355  */
356 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
357 				  phys_addr_t size, pgprot_t prot)
358 {
359 	if (virt < VMALLOC_START) {
360 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
361 			&phys, virt);
362 		return;
363 	}
364 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
365 			     NO_CONT_MAPPINGS);
366 }
367 
368 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
369 			       unsigned long virt, phys_addr_t size,
370 			       pgprot_t prot, bool page_mappings_only)
371 {
372 	int flags = 0;
373 
374 	BUG_ON(mm == &init_mm);
375 
376 	if (page_mappings_only)
377 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
378 
379 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
380 			     pgd_pgtable_alloc, flags);
381 }
382 
383 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
384 				phys_addr_t size, pgprot_t prot)
385 {
386 	if (virt < VMALLOC_START) {
387 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
388 			&phys, virt);
389 		return;
390 	}
391 
392 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
393 			     NO_CONT_MAPPINGS);
394 
395 	/* flush the TLBs after updating live kernel mappings */
396 	flush_tlb_kernel_range(virt, virt + size);
397 }
398 
399 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start,
400 				  phys_addr_t end, pgprot_t prot, int flags)
401 {
402 	__create_pgd_mapping(pgd, start, __phys_to_virt(start), end - start,
403 			     prot, early_pgtable_alloc, flags);
404 }
405 
406 void __init mark_linear_text_alias_ro(void)
407 {
408 	/*
409 	 * Remove the write permissions from the linear alias of .text/.rodata
410 	 */
411 	update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
412 			    (unsigned long)__init_begin - (unsigned long)_text,
413 			    PAGE_KERNEL_RO);
414 }
415 
416 static void __init map_mem(pgd_t *pgd)
417 {
418 	phys_addr_t kernel_start = __pa_symbol(_text);
419 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
420 	struct memblock_region *reg;
421 	int flags = 0;
422 
423 	if (debug_pagealloc_enabled())
424 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
425 
426 	/*
427 	 * Take care not to create a writable alias for the
428 	 * read-only text and rodata sections of the kernel image.
429 	 * So temporarily mark them as NOMAP to skip mappings in
430 	 * the following for-loop
431 	 */
432 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
433 #ifdef CONFIG_KEXEC_CORE
434 	if (crashk_res.end)
435 		memblock_mark_nomap(crashk_res.start,
436 				    resource_size(&crashk_res));
437 #endif
438 
439 	/* map all the memory banks */
440 	for_each_memblock(memory, reg) {
441 		phys_addr_t start = reg->base;
442 		phys_addr_t end = start + reg->size;
443 
444 		if (start >= end)
445 			break;
446 		if (memblock_is_nomap(reg))
447 			continue;
448 
449 		__map_memblock(pgd, start, end, PAGE_KERNEL, flags);
450 	}
451 
452 	/*
453 	 * Map the linear alias of the [_text, __init_begin) interval
454 	 * as non-executable now, and remove the write permission in
455 	 * mark_linear_text_alias_ro() below (which will be called after
456 	 * alternative patching has completed). This makes the contents
457 	 * of the region accessible to subsystems such as hibernate,
458 	 * but protects it from inadvertent modification or execution.
459 	 * Note that contiguous mappings cannot be remapped in this way,
460 	 * so we should avoid them here.
461 	 */
462 	__map_memblock(pgd, kernel_start, kernel_end,
463 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
464 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
465 
466 #ifdef CONFIG_KEXEC_CORE
467 	/*
468 	 * Use page-level mappings here so that we can shrink the region
469 	 * in page granularity and put back unused memory to buddy system
470 	 * through /sys/kernel/kexec_crash_size interface.
471 	 */
472 	if (crashk_res.end) {
473 		__map_memblock(pgd, crashk_res.start, crashk_res.end + 1,
474 			       PAGE_KERNEL,
475 			       NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
476 		memblock_clear_nomap(crashk_res.start,
477 				     resource_size(&crashk_res));
478 	}
479 #endif
480 }
481 
482 void mark_rodata_ro(void)
483 {
484 	unsigned long section_size;
485 
486 	/*
487 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
488 	 * to cover NOTES and EXCEPTION_TABLE.
489 	 */
490 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
491 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
492 			    section_size, PAGE_KERNEL_RO);
493 
494 	debug_checkwx();
495 }
496 
497 static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
498 				      pgprot_t prot, struct vm_struct *vma,
499 				      int flags, unsigned long vm_flags)
500 {
501 	phys_addr_t pa_start = __pa_symbol(va_start);
502 	unsigned long size = va_end - va_start;
503 
504 	BUG_ON(!PAGE_ALIGNED(pa_start));
505 	BUG_ON(!PAGE_ALIGNED(size));
506 
507 	__create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
508 			     early_pgtable_alloc, flags);
509 
510 	if (!(vm_flags & VM_NO_GUARD))
511 		size += PAGE_SIZE;
512 
513 	vma->addr	= va_start;
514 	vma->phys_addr	= pa_start;
515 	vma->size	= size;
516 	vma->flags	= VM_MAP | vm_flags;
517 	vma->caller	= __builtin_return_address(0);
518 
519 	vm_area_add_early(vma);
520 }
521 
522 static int __init parse_rodata(char *arg)
523 {
524 	return strtobool(arg, &rodata_enabled);
525 }
526 early_param("rodata", parse_rodata);
527 
528 /*
529  * Create fine-grained mappings for the kernel.
530  */
531 static void __init map_kernel(pgd_t *pgd)
532 {
533 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
534 				vmlinux_initdata, vmlinux_data;
535 
536 	/*
537 	 * External debuggers may need to write directly to the text
538 	 * mapping to install SW breakpoints. Allow this (only) when
539 	 * explicitly requested with rodata=off.
540 	 */
541 	pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
542 
543 	/*
544 	 * Only rodata will be remapped with different permissions later on,
545 	 * all other segments are allowed to use contiguous mappings.
546 	 */
547 	map_kernel_segment(pgd, _text, _etext, text_prot, &vmlinux_text, 0,
548 			   VM_NO_GUARD);
549 	map_kernel_segment(pgd, __start_rodata, __inittext_begin, PAGE_KERNEL,
550 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
551 	map_kernel_segment(pgd, __inittext_begin, __inittext_end, text_prot,
552 			   &vmlinux_inittext, 0, VM_NO_GUARD);
553 	map_kernel_segment(pgd, __initdata_begin, __initdata_end, PAGE_KERNEL,
554 			   &vmlinux_initdata, 0, VM_NO_GUARD);
555 	map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
556 
557 	if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
558 		/*
559 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
560 		 * live in the carveout for the swapper_pg_dir. We can simply
561 		 * re-use the existing dir for the fixmap.
562 		 */
563 		set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
564 			*pgd_offset_k(FIXADDR_START));
565 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
566 		/*
567 		 * The fixmap shares its top level pgd entry with the kernel
568 		 * mapping. This can really only occur when we are running
569 		 * with 16k/4 levels, so we can simply reuse the pud level
570 		 * entry instead.
571 		 */
572 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
573 		set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
574 			__pud(__pa_symbol(bm_pmd) | PUD_TYPE_TABLE));
575 		pud_clear_fixmap();
576 	} else {
577 		BUG();
578 	}
579 
580 	kasan_copy_shadow(pgd);
581 }
582 
583 /*
584  * paging_init() sets up the page tables, initialises the zone memory
585  * maps and sets up the zero page.
586  */
587 void __init paging_init(void)
588 {
589 	phys_addr_t pgd_phys = early_pgtable_alloc();
590 	pgd_t *pgd = pgd_set_fixmap(pgd_phys);
591 
592 	map_kernel(pgd);
593 	map_mem(pgd);
594 
595 	/*
596 	 * We want to reuse the original swapper_pg_dir so we don't have to
597 	 * communicate the new address to non-coherent secondaries in
598 	 * secondary_entry, and so cpu_switch_mm can generate the address with
599 	 * adrp+add rather than a load from some global variable.
600 	 *
601 	 * To do this we need to go via a temporary pgd.
602 	 */
603 	cpu_replace_ttbr1(__va(pgd_phys));
604 	memcpy(swapper_pg_dir, pgd, PGD_SIZE);
605 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
606 
607 	pgd_clear_fixmap();
608 	memblock_free(pgd_phys, PAGE_SIZE);
609 
610 	/*
611 	 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
612 	 * allocated with it.
613 	 */
614 	memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE,
615 		      SWAPPER_DIR_SIZE - PAGE_SIZE);
616 }
617 
618 /*
619  * Check whether a kernel address is valid (derived from arch/x86/).
620  */
621 int kern_addr_valid(unsigned long addr)
622 {
623 	pgd_t *pgd;
624 	pud_t *pud;
625 	pmd_t *pmd;
626 	pte_t *pte;
627 
628 	if ((((long)addr) >> VA_BITS) != -1UL)
629 		return 0;
630 
631 	pgd = pgd_offset_k(addr);
632 	if (pgd_none(*pgd))
633 		return 0;
634 
635 	pud = pud_offset(pgd, addr);
636 	if (pud_none(*pud))
637 		return 0;
638 
639 	if (pud_sect(*pud))
640 		return pfn_valid(pud_pfn(*pud));
641 
642 	pmd = pmd_offset(pud, addr);
643 	if (pmd_none(*pmd))
644 		return 0;
645 
646 	if (pmd_sect(*pmd))
647 		return pfn_valid(pmd_pfn(*pmd));
648 
649 	pte = pte_offset_kernel(pmd, addr);
650 	if (pte_none(*pte))
651 		return 0;
652 
653 	return pfn_valid(pte_pfn(*pte));
654 }
655 #ifdef CONFIG_SPARSEMEM_VMEMMAP
656 #if !ARM64_SWAPPER_USES_SECTION_MAPS
657 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
658 {
659 	return vmemmap_populate_basepages(start, end, node);
660 }
661 #else	/* !ARM64_SWAPPER_USES_SECTION_MAPS */
662 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
663 {
664 	unsigned long addr = start;
665 	unsigned long next;
666 	pgd_t *pgd;
667 	pud_t *pud;
668 	pmd_t *pmd;
669 
670 	do {
671 		next = pmd_addr_end(addr, end);
672 
673 		pgd = vmemmap_pgd_populate(addr, node);
674 		if (!pgd)
675 			return -ENOMEM;
676 
677 		pud = vmemmap_pud_populate(pgd, addr, node);
678 		if (!pud)
679 			return -ENOMEM;
680 
681 		pmd = pmd_offset(pud, addr);
682 		if (pmd_none(*pmd)) {
683 			void *p = NULL;
684 
685 			p = vmemmap_alloc_block_buf(PMD_SIZE, node);
686 			if (!p)
687 				return -ENOMEM;
688 
689 			set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
690 		} else
691 			vmemmap_verify((pte_t *)pmd, node, addr, next);
692 	} while (addr = next, addr != end);
693 
694 	return 0;
695 }
696 #endif	/* CONFIG_ARM64_64K_PAGES */
697 void vmemmap_free(unsigned long start, unsigned long end)
698 {
699 }
700 #endif	/* CONFIG_SPARSEMEM_VMEMMAP */
701 
702 static inline pud_t * fixmap_pud(unsigned long addr)
703 {
704 	pgd_t *pgd = pgd_offset_k(addr);
705 
706 	BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
707 
708 	return pud_offset_kimg(pgd, addr);
709 }
710 
711 static inline pmd_t * fixmap_pmd(unsigned long addr)
712 {
713 	pud_t *pud = fixmap_pud(addr);
714 
715 	BUG_ON(pud_none(*pud) || pud_bad(*pud));
716 
717 	return pmd_offset_kimg(pud, addr);
718 }
719 
720 static inline pte_t * fixmap_pte(unsigned long addr)
721 {
722 	return &bm_pte[pte_index(addr)];
723 }
724 
725 /*
726  * The p*d_populate functions call virt_to_phys implicitly so they can't be used
727  * directly on kernel symbols (bm_p*d). This function is called too early to use
728  * lm_alias so __p*d_populate functions must be used to populate with the
729  * physical address from __pa_symbol.
730  */
731 void __init early_fixmap_init(void)
732 {
733 	pgd_t *pgd;
734 	pud_t *pud;
735 	pmd_t *pmd;
736 	unsigned long addr = FIXADDR_START;
737 
738 	pgd = pgd_offset_k(addr);
739 	if (CONFIG_PGTABLE_LEVELS > 3 &&
740 	    !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa_symbol(bm_pud))) {
741 		/*
742 		 * We only end up here if the kernel mapping and the fixmap
743 		 * share the top level pgd entry, which should only happen on
744 		 * 16k/4 levels configurations.
745 		 */
746 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
747 		pud = pud_offset_kimg(pgd, addr);
748 	} else {
749 		if (pgd_none(*pgd))
750 			__pgd_populate(pgd, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
751 		pud = fixmap_pud(addr);
752 	}
753 	if (pud_none(*pud))
754 		__pud_populate(pud, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
755 	pmd = fixmap_pmd(addr);
756 	__pmd_populate(pmd, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
757 
758 	/*
759 	 * The boot-ioremap range spans multiple pmds, for which
760 	 * we are not prepared:
761 	 */
762 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
763 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
764 
765 	if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
766 	     || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
767 		WARN_ON(1);
768 		pr_warn("pmd %p != %p, %p\n",
769 			pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
770 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
771 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
772 			fix_to_virt(FIX_BTMAP_BEGIN));
773 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
774 			fix_to_virt(FIX_BTMAP_END));
775 
776 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
777 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
778 	}
779 }
780 
781 void __set_fixmap(enum fixed_addresses idx,
782 			       phys_addr_t phys, pgprot_t flags)
783 {
784 	unsigned long addr = __fix_to_virt(idx);
785 	pte_t *pte;
786 
787 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
788 
789 	pte = fixmap_pte(addr);
790 
791 	if (pgprot_val(flags)) {
792 		set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
793 	} else {
794 		pte_clear(&init_mm, addr, pte);
795 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
796 	}
797 }
798 
799 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
800 {
801 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
802 	int offset;
803 	void *dt_virt;
804 
805 	/*
806 	 * Check whether the physical FDT address is set and meets the minimum
807 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
808 	 * at least 8 bytes so that we can always access the magic and size
809 	 * fields of the FDT header after mapping the first chunk, double check
810 	 * here if that is indeed the case.
811 	 */
812 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
813 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
814 		return NULL;
815 
816 	/*
817 	 * Make sure that the FDT region can be mapped without the need to
818 	 * allocate additional translation table pages, so that it is safe
819 	 * to call create_mapping_noalloc() this early.
820 	 *
821 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
822 	 * be in the same PMD as the rest of the fixmap.
823 	 * On 4k pages, we'll use section mappings for the FDT so we only
824 	 * have to be in the same PUD.
825 	 */
826 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
827 
828 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
829 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
830 
831 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
832 	dt_virt = (void *)dt_virt_base + offset;
833 
834 	/* map the first chunk so we can read the size from the header */
835 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
836 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
837 
838 	if (fdt_magic(dt_virt) != FDT_MAGIC)
839 		return NULL;
840 
841 	*size = fdt_totalsize(dt_virt);
842 	if (*size > MAX_FDT_SIZE)
843 		return NULL;
844 
845 	if (offset + *size > SWAPPER_BLOCK_SIZE)
846 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
847 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
848 
849 	return dt_virt;
850 }
851 
852 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
853 {
854 	void *dt_virt;
855 	int size;
856 
857 	dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
858 	if (!dt_virt)
859 		return NULL;
860 
861 	memblock_reserve(dt_phys, size);
862 	return dt_virt;
863 }
864 
865 int __init arch_ioremap_pud_supported(void)
866 {
867 	/* only 4k granule supports level 1 block mappings */
868 	return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
869 }
870 
871 int __init arch_ioremap_pmd_supported(void)
872 {
873 	return 1;
874 }
875 
876 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
877 {
878 	BUG_ON(phys & ~PUD_MASK);
879 	set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
880 	return 1;
881 }
882 
883 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
884 {
885 	BUG_ON(phys & ~PMD_MASK);
886 	set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
887 	return 1;
888 }
889 
890 int pud_clear_huge(pud_t *pud)
891 {
892 	if (!pud_sect(*pud))
893 		return 0;
894 	pud_clear(pud);
895 	return 1;
896 }
897 
898 int pmd_clear_huge(pmd_t *pmd)
899 {
900 	if (!pmd_sect(*pmd))
901 		return 0;
902 	pmd_clear(pmd);
903 	return 1;
904 }
905