xref: /linux/arch/arm64/mm/mmu.c (revision 55f3538c4923e9dfca132e99ebec370e8094afda)
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 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD;
54 
55 u64 kimage_voffset __ro_after_init;
56 EXPORT_SYMBOL(kimage_voffset);
57 
58 /*
59  * Empty_zero_page is a special page that is used for zero-initialized data
60  * and COW.
61  */
62 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
63 EXPORT_SYMBOL(empty_zero_page);
64 
65 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
66 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
67 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
68 
69 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
70 			      unsigned long size, pgprot_t vma_prot)
71 {
72 	if (!pfn_valid(pfn))
73 		return pgprot_noncached(vma_prot);
74 	else if (file->f_flags & O_SYNC)
75 		return pgprot_writecombine(vma_prot);
76 	return vma_prot;
77 }
78 EXPORT_SYMBOL(phys_mem_access_prot);
79 
80 static phys_addr_t __init early_pgtable_alloc(void)
81 {
82 	phys_addr_t phys;
83 	void *ptr;
84 
85 	phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
86 
87 	/*
88 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
89 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
90 	 * any level of table.
91 	 */
92 	ptr = pte_set_fixmap(phys);
93 
94 	memset(ptr, 0, PAGE_SIZE);
95 
96 	/*
97 	 * Implicit barriers also ensure the zeroed page is visible to the page
98 	 * table walker
99 	 */
100 	pte_clear_fixmap();
101 
102 	return phys;
103 }
104 
105 static bool pgattr_change_is_safe(u64 old, u64 new)
106 {
107 	/*
108 	 * The following mapping attributes may be updated in live
109 	 * kernel mappings without the need for break-before-make.
110 	 */
111 	static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE;
112 
113 	/* creating or taking down mappings is always safe */
114 	if (old == 0 || new == 0)
115 		return true;
116 
117 	/* live contiguous mappings may not be manipulated at all */
118 	if ((old | new) & PTE_CONT)
119 		return false;
120 
121 	return ((old ^ new) & ~mask) == 0;
122 }
123 
124 static void init_pte(pmd_t *pmd, unsigned long addr, unsigned long end,
125 		     phys_addr_t phys, pgprot_t prot)
126 {
127 	pte_t *pte;
128 
129 	pte = pte_set_fixmap_offset(pmd, addr);
130 	do {
131 		pte_t old_pte = *pte;
132 
133 		set_pte(pte, pfn_pte(__phys_to_pfn(phys), prot));
134 
135 		/*
136 		 * After the PTE entry has been populated once, we
137 		 * only allow updates to the permission attributes.
138 		 */
139 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte)));
140 
141 		phys += PAGE_SIZE;
142 	} while (pte++, addr += PAGE_SIZE, addr != end);
143 
144 	pte_clear_fixmap();
145 }
146 
147 static void alloc_init_cont_pte(pmd_t *pmd, unsigned long addr,
148 				unsigned long end, phys_addr_t phys,
149 				pgprot_t prot,
150 				phys_addr_t (*pgtable_alloc)(void),
151 				int flags)
152 {
153 	unsigned long next;
154 
155 	BUG_ON(pmd_sect(*pmd));
156 	if (pmd_none(*pmd)) {
157 		phys_addr_t pte_phys;
158 		BUG_ON(!pgtable_alloc);
159 		pte_phys = pgtable_alloc();
160 		__pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
161 	}
162 	BUG_ON(pmd_bad(*pmd));
163 
164 	do {
165 		pgprot_t __prot = prot;
166 
167 		next = pte_cont_addr_end(addr, end);
168 
169 		/* use a contiguous mapping if the range is suitably aligned */
170 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
171 		    (flags & NO_CONT_MAPPINGS) == 0)
172 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
173 
174 		init_pte(pmd, addr, next, phys, __prot);
175 
176 		phys += next - addr;
177 	} while (addr = next, addr != end);
178 }
179 
180 static void init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
181 		     phys_addr_t phys, pgprot_t prot,
182 		     phys_addr_t (*pgtable_alloc)(void), int flags)
183 {
184 	unsigned long next;
185 	pmd_t *pmd;
186 
187 	pmd = pmd_set_fixmap_offset(pud, addr);
188 	do {
189 		pmd_t old_pmd = *pmd;
190 
191 		next = pmd_addr_end(addr, end);
192 
193 		/* try section mapping first */
194 		if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
195 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
196 			pmd_set_huge(pmd, phys, prot);
197 
198 			/*
199 			 * After the PMD entry has been populated once, we
200 			 * only allow updates to the permission attributes.
201 			 */
202 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
203 						      pmd_val(*pmd)));
204 		} else {
205 			alloc_init_cont_pte(pmd, addr, next, phys, prot,
206 					    pgtable_alloc, flags);
207 
208 			BUG_ON(pmd_val(old_pmd) != 0 &&
209 			       pmd_val(old_pmd) != pmd_val(*pmd));
210 		}
211 		phys += next - addr;
212 	} while (pmd++, addr = next, addr != end);
213 
214 	pmd_clear_fixmap();
215 }
216 
217 static void alloc_init_cont_pmd(pud_t *pud, unsigned long addr,
218 				unsigned long end, phys_addr_t phys,
219 				pgprot_t prot,
220 				phys_addr_t (*pgtable_alloc)(void), int flags)
221 {
222 	unsigned long next;
223 
224 	/*
225 	 * Check for initial section mappings in the pgd/pud.
226 	 */
227 	BUG_ON(pud_sect(*pud));
228 	if (pud_none(*pud)) {
229 		phys_addr_t pmd_phys;
230 		BUG_ON(!pgtable_alloc);
231 		pmd_phys = pgtable_alloc();
232 		__pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
233 	}
234 	BUG_ON(pud_bad(*pud));
235 
236 	do {
237 		pgprot_t __prot = prot;
238 
239 		next = pmd_cont_addr_end(addr, end);
240 
241 		/* use a contiguous mapping if the range is suitably aligned */
242 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
243 		    (flags & NO_CONT_MAPPINGS) == 0)
244 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
245 
246 		init_pmd(pud, addr, next, phys, __prot, pgtable_alloc, flags);
247 
248 		phys += next - addr;
249 	} while (addr = next, addr != end);
250 }
251 
252 static inline bool use_1G_block(unsigned long addr, unsigned long next,
253 			unsigned long phys)
254 {
255 	if (PAGE_SHIFT != 12)
256 		return false;
257 
258 	if (((addr | next | phys) & ~PUD_MASK) != 0)
259 		return false;
260 
261 	return true;
262 }
263 
264 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
265 				  phys_addr_t phys, pgprot_t prot,
266 				  phys_addr_t (*pgtable_alloc)(void),
267 				  int flags)
268 {
269 	pud_t *pud;
270 	unsigned long next;
271 
272 	if (pgd_none(*pgd)) {
273 		phys_addr_t pud_phys;
274 		BUG_ON(!pgtable_alloc);
275 		pud_phys = pgtable_alloc();
276 		__pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
277 	}
278 	BUG_ON(pgd_bad(*pgd));
279 
280 	pud = pud_set_fixmap_offset(pgd, addr);
281 	do {
282 		pud_t old_pud = *pud;
283 
284 		next = pud_addr_end(addr, end);
285 
286 		/*
287 		 * For 4K granule only, attempt to put down a 1GB block
288 		 */
289 		if (use_1G_block(addr, next, phys) &&
290 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
291 			pud_set_huge(pud, phys, prot);
292 
293 			/*
294 			 * After the PUD entry has been populated once, we
295 			 * only allow updates to the permission attributes.
296 			 */
297 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
298 						      pud_val(*pud)));
299 		} else {
300 			alloc_init_cont_pmd(pud, addr, next, phys, prot,
301 					    pgtable_alloc, flags);
302 
303 			BUG_ON(pud_val(old_pud) != 0 &&
304 			       pud_val(old_pud) != pud_val(*pud));
305 		}
306 		phys += next - addr;
307 	} while (pud++, addr = next, addr != end);
308 
309 	pud_clear_fixmap();
310 }
311 
312 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
313 				 unsigned long virt, phys_addr_t size,
314 				 pgprot_t prot,
315 				 phys_addr_t (*pgtable_alloc)(void),
316 				 int flags)
317 {
318 	unsigned long addr, length, end, next;
319 	pgd_t *pgd = pgd_offset_raw(pgdir, virt);
320 
321 	/*
322 	 * If the virtual and physical address don't have the same offset
323 	 * within a page, we cannot map the region as the caller expects.
324 	 */
325 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
326 		return;
327 
328 	phys &= PAGE_MASK;
329 	addr = virt & PAGE_MASK;
330 	length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
331 
332 	end = addr + length;
333 	do {
334 		next = pgd_addr_end(addr, end);
335 		alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
336 			       flags);
337 		phys += next - addr;
338 	} while (pgd++, addr = next, addr != end);
339 }
340 
341 static phys_addr_t pgd_pgtable_alloc(void)
342 {
343 	void *ptr = (void *)__get_free_page(PGALLOC_GFP);
344 	if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
345 		BUG();
346 
347 	/* Ensure the zeroed page is visible to the page table walker */
348 	dsb(ishst);
349 	return __pa(ptr);
350 }
351 
352 /*
353  * This function can only be used to modify existing table entries,
354  * without allocating new levels of table. Note that this permits the
355  * creation of new section or page entries.
356  */
357 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
358 				  phys_addr_t size, pgprot_t prot)
359 {
360 	if (virt < VMALLOC_START) {
361 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
362 			&phys, virt);
363 		return;
364 	}
365 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
366 			     NO_CONT_MAPPINGS);
367 }
368 
369 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
370 			       unsigned long virt, phys_addr_t size,
371 			       pgprot_t prot, bool page_mappings_only)
372 {
373 	int flags = 0;
374 
375 	BUG_ON(mm == &init_mm);
376 
377 	if (page_mappings_only)
378 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
379 
380 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
381 			     pgd_pgtable_alloc, flags);
382 }
383 
384 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
385 				phys_addr_t size, pgprot_t prot)
386 {
387 	if (virt < VMALLOC_START) {
388 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
389 			&phys, virt);
390 		return;
391 	}
392 
393 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
394 			     NO_CONT_MAPPINGS);
395 
396 	/* flush the TLBs after updating live kernel mappings */
397 	flush_tlb_kernel_range(virt, virt + size);
398 }
399 
400 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start,
401 				  phys_addr_t end, pgprot_t prot, int flags)
402 {
403 	__create_pgd_mapping(pgd, start, __phys_to_virt(start), end - start,
404 			     prot, early_pgtable_alloc, flags);
405 }
406 
407 void __init mark_linear_text_alias_ro(void)
408 {
409 	/*
410 	 * Remove the write permissions from the linear alias of .text/.rodata
411 	 */
412 	update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
413 			    (unsigned long)__init_begin - (unsigned long)_text,
414 			    PAGE_KERNEL_RO);
415 }
416 
417 static void __init map_mem(pgd_t *pgd)
418 {
419 	phys_addr_t kernel_start = __pa_symbol(_text);
420 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
421 	struct memblock_region *reg;
422 	int flags = 0;
423 
424 	if (debug_pagealloc_enabled())
425 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
426 
427 	/*
428 	 * Take care not to create a writable alias for the
429 	 * read-only text and rodata sections of the kernel image.
430 	 * So temporarily mark them as NOMAP to skip mappings in
431 	 * the following for-loop
432 	 */
433 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
434 #ifdef CONFIG_KEXEC_CORE
435 	if (crashk_res.end)
436 		memblock_mark_nomap(crashk_res.start,
437 				    resource_size(&crashk_res));
438 #endif
439 
440 	/* map all the memory banks */
441 	for_each_memblock(memory, reg) {
442 		phys_addr_t start = reg->base;
443 		phys_addr_t end = start + reg->size;
444 
445 		if (start >= end)
446 			break;
447 		if (memblock_is_nomap(reg))
448 			continue;
449 
450 		__map_memblock(pgd, start, end, PAGE_KERNEL, flags);
451 	}
452 
453 	/*
454 	 * Map the linear alias of the [_text, __init_begin) interval
455 	 * as non-executable now, and remove the write permission in
456 	 * mark_linear_text_alias_ro() below (which will be called after
457 	 * alternative patching has completed). This makes the contents
458 	 * of the region accessible to subsystems such as hibernate,
459 	 * but protects it from inadvertent modification or execution.
460 	 * Note that contiguous mappings cannot be remapped in this way,
461 	 * so we should avoid them here.
462 	 */
463 	__map_memblock(pgd, kernel_start, kernel_end,
464 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
465 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
466 
467 #ifdef CONFIG_KEXEC_CORE
468 	/*
469 	 * Use page-level mappings here so that we can shrink the region
470 	 * in page granularity and put back unused memory to buddy system
471 	 * through /sys/kernel/kexec_crash_size interface.
472 	 */
473 	if (crashk_res.end) {
474 		__map_memblock(pgd, crashk_res.start, crashk_res.end + 1,
475 			       PAGE_KERNEL,
476 			       NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
477 		memblock_clear_nomap(crashk_res.start,
478 				     resource_size(&crashk_res));
479 	}
480 #endif
481 }
482 
483 void mark_rodata_ro(void)
484 {
485 	unsigned long section_size;
486 
487 	/*
488 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
489 	 * to cover NOTES and EXCEPTION_TABLE.
490 	 */
491 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
492 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
493 			    section_size, PAGE_KERNEL_RO);
494 
495 	debug_checkwx();
496 }
497 
498 static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
499 				      pgprot_t prot, struct vm_struct *vma,
500 				      int flags, unsigned long vm_flags)
501 {
502 	phys_addr_t pa_start = __pa_symbol(va_start);
503 	unsigned long size = va_end - va_start;
504 
505 	BUG_ON(!PAGE_ALIGNED(pa_start));
506 	BUG_ON(!PAGE_ALIGNED(size));
507 
508 	__create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
509 			     early_pgtable_alloc, flags);
510 
511 	if (!(vm_flags & VM_NO_GUARD))
512 		size += PAGE_SIZE;
513 
514 	vma->addr	= va_start;
515 	vma->phys_addr	= pa_start;
516 	vma->size	= size;
517 	vma->flags	= VM_MAP | vm_flags;
518 	vma->caller	= __builtin_return_address(0);
519 
520 	vm_area_add_early(vma);
521 }
522 
523 static int __init parse_rodata(char *arg)
524 {
525 	return strtobool(arg, &rodata_enabled);
526 }
527 early_param("rodata", parse_rodata);
528 
529 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
530 static int __init map_entry_trampoline(void)
531 {
532 	pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
533 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
534 
535 	/* The trampoline is always mapped and can therefore be global */
536 	pgprot_val(prot) &= ~PTE_NG;
537 
538 	/* Map only the text into the trampoline page table */
539 	memset(tramp_pg_dir, 0, PGD_SIZE);
540 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
541 			     prot, pgd_pgtable_alloc, 0);
542 
543 	/* Map both the text and data into the kernel page table */
544 	__set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
545 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
546 		extern char __entry_tramp_data_start[];
547 
548 		__set_fixmap(FIX_ENTRY_TRAMP_DATA,
549 			     __pa_symbol(__entry_tramp_data_start),
550 			     PAGE_KERNEL_RO);
551 	}
552 
553 	return 0;
554 }
555 core_initcall(map_entry_trampoline);
556 #endif
557 
558 /*
559  * Create fine-grained mappings for the kernel.
560  */
561 static void __init map_kernel(pgd_t *pgd)
562 {
563 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
564 				vmlinux_initdata, vmlinux_data;
565 
566 	/*
567 	 * External debuggers may need to write directly to the text
568 	 * mapping to install SW breakpoints. Allow this (only) when
569 	 * explicitly requested with rodata=off.
570 	 */
571 	pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
572 
573 	/*
574 	 * Only rodata will be remapped with different permissions later on,
575 	 * all other segments are allowed to use contiguous mappings.
576 	 */
577 	map_kernel_segment(pgd, _text, _etext, text_prot, &vmlinux_text, 0,
578 			   VM_NO_GUARD);
579 	map_kernel_segment(pgd, __start_rodata, __inittext_begin, PAGE_KERNEL,
580 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
581 	map_kernel_segment(pgd, __inittext_begin, __inittext_end, text_prot,
582 			   &vmlinux_inittext, 0, VM_NO_GUARD);
583 	map_kernel_segment(pgd, __initdata_begin, __initdata_end, PAGE_KERNEL,
584 			   &vmlinux_initdata, 0, VM_NO_GUARD);
585 	map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
586 
587 	if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
588 		/*
589 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
590 		 * live in the carveout for the swapper_pg_dir. We can simply
591 		 * re-use the existing dir for the fixmap.
592 		 */
593 		set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
594 			*pgd_offset_k(FIXADDR_START));
595 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
596 		/*
597 		 * The fixmap shares its top level pgd entry with the kernel
598 		 * mapping. This can really only occur when we are running
599 		 * with 16k/4 levels, so we can simply reuse the pud level
600 		 * entry instead.
601 		 */
602 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
603 		pud_populate(&init_mm, pud_set_fixmap_offset(pgd, FIXADDR_START),
604 			     lm_alias(bm_pmd));
605 		pud_clear_fixmap();
606 	} else {
607 		BUG();
608 	}
609 
610 	kasan_copy_shadow(pgd);
611 }
612 
613 /*
614  * paging_init() sets up the page tables, initialises the zone memory
615  * maps and sets up the zero page.
616  */
617 void __init paging_init(void)
618 {
619 	phys_addr_t pgd_phys = early_pgtable_alloc();
620 	pgd_t *pgd = pgd_set_fixmap(pgd_phys);
621 
622 	map_kernel(pgd);
623 	map_mem(pgd);
624 
625 	/*
626 	 * We want to reuse the original swapper_pg_dir so we don't have to
627 	 * communicate the new address to non-coherent secondaries in
628 	 * secondary_entry, and so cpu_switch_mm can generate the address with
629 	 * adrp+add rather than a load from some global variable.
630 	 *
631 	 * To do this we need to go via a temporary pgd.
632 	 */
633 	cpu_replace_ttbr1(__va(pgd_phys));
634 	memcpy(swapper_pg_dir, pgd, PGD_SIZE);
635 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
636 
637 	pgd_clear_fixmap();
638 	memblock_free(pgd_phys, PAGE_SIZE);
639 
640 	/*
641 	 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
642 	 * allocated with it.
643 	 */
644 	memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE,
645 		      __pa_symbol(swapper_pg_end) - __pa_symbol(swapper_pg_dir)
646 		      - PAGE_SIZE);
647 }
648 
649 /*
650  * Check whether a kernel address is valid (derived from arch/x86/).
651  */
652 int kern_addr_valid(unsigned long addr)
653 {
654 	pgd_t *pgd;
655 	pud_t *pud;
656 	pmd_t *pmd;
657 	pte_t *pte;
658 
659 	if ((((long)addr) >> VA_BITS) != -1UL)
660 		return 0;
661 
662 	pgd = pgd_offset_k(addr);
663 	if (pgd_none(*pgd))
664 		return 0;
665 
666 	pud = pud_offset(pgd, addr);
667 	if (pud_none(*pud))
668 		return 0;
669 
670 	if (pud_sect(*pud))
671 		return pfn_valid(pud_pfn(*pud));
672 
673 	pmd = pmd_offset(pud, addr);
674 	if (pmd_none(*pmd))
675 		return 0;
676 
677 	if (pmd_sect(*pmd))
678 		return pfn_valid(pmd_pfn(*pmd));
679 
680 	pte = pte_offset_kernel(pmd, addr);
681 	if (pte_none(*pte))
682 		return 0;
683 
684 	return pfn_valid(pte_pfn(*pte));
685 }
686 #ifdef CONFIG_SPARSEMEM_VMEMMAP
687 #if !ARM64_SWAPPER_USES_SECTION_MAPS
688 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
689 {
690 	return vmemmap_populate_basepages(start, end, node);
691 }
692 #else	/* !ARM64_SWAPPER_USES_SECTION_MAPS */
693 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
694 {
695 	unsigned long addr = start;
696 	unsigned long next;
697 	pgd_t *pgd;
698 	pud_t *pud;
699 	pmd_t *pmd;
700 
701 	do {
702 		next = pmd_addr_end(addr, end);
703 
704 		pgd = vmemmap_pgd_populate(addr, node);
705 		if (!pgd)
706 			return -ENOMEM;
707 
708 		pud = vmemmap_pud_populate(pgd, addr, node);
709 		if (!pud)
710 			return -ENOMEM;
711 
712 		pmd = pmd_offset(pud, addr);
713 		if (pmd_none(*pmd)) {
714 			void *p = NULL;
715 
716 			p = vmemmap_alloc_block_buf(PMD_SIZE, node);
717 			if (!p)
718 				return -ENOMEM;
719 
720 			pmd_set_huge(pmd, __pa(p), __pgprot(PROT_SECT_NORMAL));
721 		} else
722 			vmemmap_verify((pte_t *)pmd, node, addr, next);
723 	} while (addr = next, addr != end);
724 
725 	return 0;
726 }
727 #endif	/* CONFIG_ARM64_64K_PAGES */
728 void vmemmap_free(unsigned long start, unsigned long end)
729 {
730 }
731 #endif	/* CONFIG_SPARSEMEM_VMEMMAP */
732 
733 static inline pud_t * fixmap_pud(unsigned long addr)
734 {
735 	pgd_t *pgd = pgd_offset_k(addr);
736 
737 	BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
738 
739 	return pud_offset_kimg(pgd, addr);
740 }
741 
742 static inline pmd_t * fixmap_pmd(unsigned long addr)
743 {
744 	pud_t *pud = fixmap_pud(addr);
745 
746 	BUG_ON(pud_none(*pud) || pud_bad(*pud));
747 
748 	return pmd_offset_kimg(pud, addr);
749 }
750 
751 static inline pte_t * fixmap_pte(unsigned long addr)
752 {
753 	return &bm_pte[pte_index(addr)];
754 }
755 
756 /*
757  * The p*d_populate functions call virt_to_phys implicitly so they can't be used
758  * directly on kernel symbols (bm_p*d). This function is called too early to use
759  * lm_alias so __p*d_populate functions must be used to populate with the
760  * physical address from __pa_symbol.
761  */
762 void __init early_fixmap_init(void)
763 {
764 	pgd_t *pgd;
765 	pud_t *pud;
766 	pmd_t *pmd;
767 	unsigned long addr = FIXADDR_START;
768 
769 	pgd = pgd_offset_k(addr);
770 	if (CONFIG_PGTABLE_LEVELS > 3 &&
771 	    !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa_symbol(bm_pud))) {
772 		/*
773 		 * We only end up here if the kernel mapping and the fixmap
774 		 * share the top level pgd entry, which should only happen on
775 		 * 16k/4 levels configurations.
776 		 */
777 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
778 		pud = pud_offset_kimg(pgd, addr);
779 	} else {
780 		if (pgd_none(*pgd))
781 			__pgd_populate(pgd, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
782 		pud = fixmap_pud(addr);
783 	}
784 	if (pud_none(*pud))
785 		__pud_populate(pud, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
786 	pmd = fixmap_pmd(addr);
787 	__pmd_populate(pmd, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
788 
789 	/*
790 	 * The boot-ioremap range spans multiple pmds, for which
791 	 * we are not prepared:
792 	 */
793 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
794 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
795 
796 	if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
797 	     || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
798 		WARN_ON(1);
799 		pr_warn("pmd %p != %p, %p\n",
800 			pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
801 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
802 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
803 			fix_to_virt(FIX_BTMAP_BEGIN));
804 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
805 			fix_to_virt(FIX_BTMAP_END));
806 
807 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
808 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
809 	}
810 }
811 
812 /*
813  * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
814  * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
815  */
816 void __set_fixmap(enum fixed_addresses idx,
817 			       phys_addr_t phys, pgprot_t flags)
818 {
819 	unsigned long addr = __fix_to_virt(idx);
820 	pte_t *pte;
821 
822 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
823 
824 	pte = fixmap_pte(addr);
825 
826 	if (pgprot_val(flags)) {
827 		set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
828 	} else {
829 		pte_clear(&init_mm, addr, pte);
830 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
831 	}
832 }
833 
834 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
835 {
836 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
837 	int offset;
838 	void *dt_virt;
839 
840 	/*
841 	 * Check whether the physical FDT address is set and meets the minimum
842 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
843 	 * at least 8 bytes so that we can always access the magic and size
844 	 * fields of the FDT header after mapping the first chunk, double check
845 	 * here if that is indeed the case.
846 	 */
847 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
848 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
849 		return NULL;
850 
851 	/*
852 	 * Make sure that the FDT region can be mapped without the need to
853 	 * allocate additional translation table pages, so that it is safe
854 	 * to call create_mapping_noalloc() this early.
855 	 *
856 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
857 	 * be in the same PMD as the rest of the fixmap.
858 	 * On 4k pages, we'll use section mappings for the FDT so we only
859 	 * have to be in the same PUD.
860 	 */
861 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
862 
863 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
864 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
865 
866 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
867 	dt_virt = (void *)dt_virt_base + offset;
868 
869 	/* map the first chunk so we can read the size from the header */
870 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
871 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
872 
873 	if (fdt_magic(dt_virt) != FDT_MAGIC)
874 		return NULL;
875 
876 	*size = fdt_totalsize(dt_virt);
877 	if (*size > MAX_FDT_SIZE)
878 		return NULL;
879 
880 	if (offset + *size > SWAPPER_BLOCK_SIZE)
881 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
882 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
883 
884 	return dt_virt;
885 }
886 
887 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
888 {
889 	void *dt_virt;
890 	int size;
891 
892 	dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
893 	if (!dt_virt)
894 		return NULL;
895 
896 	memblock_reserve(dt_phys, size);
897 	return dt_virt;
898 }
899 
900 int __init arch_ioremap_pud_supported(void)
901 {
902 	/* only 4k granule supports level 1 block mappings */
903 	return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
904 }
905 
906 int __init arch_ioremap_pmd_supported(void)
907 {
908 	return 1;
909 }
910 
911 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
912 {
913 	pgprot_t sect_prot = __pgprot(PUD_TYPE_SECT |
914 					pgprot_val(mk_sect_prot(prot)));
915 	BUG_ON(phys & ~PUD_MASK);
916 	set_pud(pud, pfn_pud(__phys_to_pfn(phys), sect_prot));
917 	return 1;
918 }
919 
920 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
921 {
922 	pgprot_t sect_prot = __pgprot(PMD_TYPE_SECT |
923 					pgprot_val(mk_sect_prot(prot)));
924 	BUG_ON(phys & ~PMD_MASK);
925 	set_pmd(pmd, pfn_pmd(__phys_to_pfn(phys), sect_prot));
926 	return 1;
927 }
928 
929 int pud_clear_huge(pud_t *pud)
930 {
931 	if (!pud_sect(*pud))
932 		return 0;
933 	pud_clear(pud);
934 	return 1;
935 }
936 
937 int pmd_clear_huge(pmd_t *pmd)
938 {
939 	if (!pmd_sect(*pmd))
940 		return 0;
941 	pmd_clear(pmd);
942 	return 1;
943 }
944