• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/export.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/init.h>
24 #include <linux/libfdt.h>
25 #include <linux/mman.h>
26 #include <linux/nodemask.h>
27 #include <linux/memblock.h>
28 #include <linux/fs.h>
29 #include <linux/io.h>
30 #include <linux/slab.h>
31 #include <linux/stop_machine.h>
32 #include <linux/mm.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 #include "mm.h"
47 
48 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
49 
50 u64 kimage_voffset __read_mostly;
51 EXPORT_SYMBOL(kimage_voffset);
52 
53 /*
54  * Empty_zero_page is a special page that is used for zero-initialized data
55  * and COW.
56  */
57 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
58 EXPORT_SYMBOL(empty_zero_page);
59 
60 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
61 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
62 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
63 
phys_mem_access_prot(struct file * file,unsigned long pfn,unsigned long size,pgprot_t vma_prot)64 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
65 			      unsigned long size, pgprot_t vma_prot)
66 {
67 	if (!pfn_valid(pfn))
68 		return pgprot_noncached(vma_prot);
69 	else if (file->f_flags & O_SYNC)
70 		return pgprot_writecombine(vma_prot);
71 	return vma_prot;
72 }
73 EXPORT_SYMBOL(phys_mem_access_prot);
74 
early_pgtable_alloc(void)75 static phys_addr_t __init early_pgtable_alloc(void)
76 {
77 	phys_addr_t phys;
78 	void *ptr;
79 
80 	phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
81 	BUG_ON(!phys);
82 
83 	/*
84 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
85 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
86 	 * any level of table.
87 	 */
88 	ptr = pte_set_fixmap(phys);
89 
90 	memset(ptr, 0, PAGE_SIZE);
91 
92 	/*
93 	 * Implicit barriers also ensure the zeroed page is visible to the page
94 	 * table walker
95 	 */
96 	pte_clear_fixmap();
97 
98 	return phys;
99 }
100 
101 /*
102  * remap a PMD into pages
103  */
split_pmd(pmd_t * pmd,pte_t * pte)104 static void split_pmd(pmd_t *pmd, pte_t *pte)
105 {
106 	unsigned long pfn = pmd_pfn(*pmd);
107 	int i = 0;
108 
109 	do {
110 		/*
111 		 * Need to have the least restrictive permissions available
112 		 * permissions will be fixed up later
113 		 */
114 		set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
115 		pfn++;
116 	} while (pte++, i++, i < PTRS_PER_PTE);
117 }
118 
alloc_init_pte(pmd_t * pmd,unsigned long addr,unsigned long end,unsigned long pfn,pgprot_t prot,phys_addr_t (* pgtable_alloc)(void))119 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
120 				  unsigned long end, unsigned long pfn,
121 				  pgprot_t prot,
122 				  phys_addr_t (*pgtable_alloc)(void))
123 {
124 	pte_t *pte;
125 
126 	if (pmd_none(*pmd) || pmd_sect(*pmd)) {
127 		phys_addr_t pte_phys;
128 		BUG_ON(!pgtable_alloc);
129 		pte_phys = pgtable_alloc();
130 		pte = pte_set_fixmap(pte_phys);
131 		if (pmd_sect(*pmd))
132 			split_pmd(pmd, pte);
133 		__pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
134 		flush_tlb_all();
135 		pte_clear_fixmap();
136 	}
137 	BUG_ON(pmd_bad(*pmd));
138 
139 	pte = pte_set_fixmap_offset(pmd, addr);
140 	do {
141 		set_pte(pte, pfn_pte(pfn, prot));
142 		pfn++;
143 	} while (pte++, addr += PAGE_SIZE, addr != end);
144 
145 	pte_clear_fixmap();
146 }
147 
split_pud(pud_t * old_pud,pmd_t * pmd)148 static void split_pud(pud_t *old_pud, pmd_t *pmd)
149 {
150 	unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
151 	pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
152 	int i = 0;
153 
154 	do {
155 		set_pmd(pmd, __pmd(addr | pgprot_val(prot)));
156 		addr += PMD_SIZE;
157 	} while (pmd++, i++, i < PTRS_PER_PMD);
158 }
159 
160 #ifdef CONFIG_DEBUG_PAGEALLOC
block_mappings_allowed(phys_addr_t (* pgtable_alloc)(void))161 static bool block_mappings_allowed(phys_addr_t (*pgtable_alloc)(void))
162 {
163 
164 	/*
165 	 * If debug_page_alloc is enabled we must map the linear map
166 	 * using pages. However, other mappings created by
167 	 * create_mapping_noalloc must use sections in some cases. Allow
168 	 * sections to be used in those cases, where no pgtable_alloc
169 	 * function is provided.
170 	 */
171 	return !pgtable_alloc || !debug_pagealloc_enabled();
172 }
173 #else
block_mappings_allowed(phys_addr_t (* pgtable_alloc)(void))174 static bool block_mappings_allowed(phys_addr_t (*pgtable_alloc)(void))
175 {
176 	return true;
177 }
178 #endif
179 
alloc_init_pmd(pud_t * pud,unsigned long addr,unsigned long end,phys_addr_t phys,pgprot_t prot,phys_addr_t (* pgtable_alloc)(void))180 static void alloc_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))
183 {
184 	pmd_t *pmd;
185 	unsigned long next;
186 
187 	/*
188 	 * Check for initial section mappings in the pgd/pud and remove them.
189 	 */
190 	if (pud_none(*pud) || pud_sect(*pud)) {
191 		phys_addr_t pmd_phys;
192 		BUG_ON(!pgtable_alloc);
193 		pmd_phys = pgtable_alloc();
194 		pmd = pmd_set_fixmap(pmd_phys);
195 		if (pud_sect(*pud)) {
196 			/*
197 			 * need to have the 1G of mappings continue to be
198 			 * present
199 			 */
200 			split_pud(pud, pmd);
201 		}
202 		__pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
203 		flush_tlb_all();
204 		pmd_clear_fixmap();
205 	}
206 	BUG_ON(pud_bad(*pud));
207 
208 	pmd = pmd_set_fixmap_offset(pud, addr);
209 	do {
210 		next = pmd_addr_end(addr, end);
211 		/* try section mapping first */
212 		if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
213 		      block_mappings_allowed(pgtable_alloc)) {
214 			pmd_t old_pmd =*pmd;
215 			pmd_set_huge(pmd, phys, prot);
216 			/*
217 			 * Check for previous table entries created during
218 			 * boot (__create_page_tables) and flush them.
219 			 */
220 			if (!pmd_none(old_pmd)) {
221 				flush_tlb_all();
222 				if (pmd_table(old_pmd)) {
223 					phys_addr_t table = pmd_page_paddr(old_pmd);
224 					if (!WARN_ON_ONCE(slab_is_available()))
225 						memblock_free(table, PAGE_SIZE);
226 				}
227 			}
228 		} else {
229 			alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
230 				       prot, pgtable_alloc);
231 		}
232 		phys += next - addr;
233 	} while (pmd++, addr = next, addr != end);
234 
235 	pmd_clear_fixmap();
236 }
237 
use_1G_block(unsigned long addr,unsigned long next,unsigned long phys)238 static inline bool use_1G_block(unsigned long addr, unsigned long next,
239 			unsigned long phys)
240 {
241 	if (PAGE_SHIFT != 12)
242 		return false;
243 
244 	if (((addr | next | phys) & ~PUD_MASK) != 0)
245 		return false;
246 
247 	return true;
248 }
249 
alloc_init_pud(pgd_t * pgd,unsigned long addr,unsigned long end,phys_addr_t phys,pgprot_t prot,phys_addr_t (* pgtable_alloc)(void))250 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
251 				  phys_addr_t phys, pgprot_t prot,
252 				  phys_addr_t (*pgtable_alloc)(void))
253 {
254 	pud_t *pud;
255 	unsigned long next;
256 
257 	if (pgd_none(*pgd)) {
258 		phys_addr_t pud_phys;
259 		BUG_ON(!pgtable_alloc);
260 		pud_phys = pgtable_alloc();
261 		__pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
262 	}
263 	BUG_ON(pgd_bad(*pgd));
264 
265 	pud = pud_set_fixmap_offset(pgd, addr);
266 	do {
267 		next = pud_addr_end(addr, end);
268 
269 		/*
270 		 * For 4K granule only, attempt to put down a 1GB block
271 		 */
272 		if (use_1G_block(addr, next, phys) &&
273 		    block_mappings_allowed(pgtable_alloc)) {
274 			pud_t old_pud = *pud;
275 			pud_set_huge(pud, phys, prot);
276 
277 			/*
278 			 * If we have an old value for a pud, it will
279 			 * be pointing to a pmd table that we no longer
280 			 * need (from swapper_pg_dir).
281 			 *
282 			 * Look up the old pmd table and free it.
283 			 */
284 			if (!pud_none(old_pud)) {
285 				flush_tlb_all();
286 				if (pud_table(old_pud)) {
287 					phys_addr_t table = pud_page_paddr(old_pud);
288 					if (!WARN_ON_ONCE(slab_is_available()))
289 						memblock_free(table, PAGE_SIZE);
290 				}
291 			}
292 		} else {
293 			alloc_init_pmd(pud, addr, next, phys, prot,
294 				       pgtable_alloc);
295 		}
296 		phys += next - addr;
297 	} while (pud++, addr = next, addr != end);
298 
299 	pud_clear_fixmap();
300 }
301 
302 /*
303  * Create the page directory entries and any necessary page tables for the
304  * mapping specified by 'md'.
305  */
init_pgd(pgd_t * pgd,phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot,phys_addr_t (* pgtable_alloc)(void))306 static void init_pgd(pgd_t *pgd, phys_addr_t phys, unsigned long virt,
307 				    phys_addr_t size, pgprot_t prot,
308 				    phys_addr_t (*pgtable_alloc)(void))
309 {
310 	unsigned long addr, length, end, next;
311 
312 	/*
313 	 * If the virtual and physical address don't have the same offset
314 	 * within a page, we cannot map the region as the caller expects.
315 	 */
316 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
317 		return;
318 
319 	phys &= PAGE_MASK;
320 	addr = virt & PAGE_MASK;
321 	length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
322 
323 	end = addr + length;
324 	do {
325 		next = pgd_addr_end(addr, end);
326 		alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc);
327 		phys += next - addr;
328 	} while (pgd++, addr = next, addr != end);
329 }
330 
late_pgtable_alloc(void)331 static phys_addr_t late_pgtable_alloc(void)
332 {
333 	void *ptr = (void *)__get_free_page(PGALLOC_GFP);
334 	BUG_ON(!ptr);
335 
336 	/* Ensure the zeroed page is visible to the page table walker */
337 	dsb(ishst);
338 	return __pa(ptr);
339 }
340 
__create_pgd_mapping(pgd_t * pgdir,phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot,phys_addr_t (* alloc)(void))341 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
342 				 unsigned long virt, phys_addr_t size,
343 				 pgprot_t prot,
344 				 phys_addr_t (*alloc)(void))
345 {
346 	init_pgd(pgd_offset_raw(pgdir, virt), phys, virt, size, prot, alloc);
347 }
348 
349 /*
350  * This function can only be used to modify existing table entries,
351  * without allocating new levels of table. Note that this permits the
352  * creation of new section or page entries.
353  */
create_mapping_noalloc(phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot)354 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
355 				  phys_addr_t size, pgprot_t prot)
356 {
357 	if (virt < VMALLOC_START) {
358 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
359 			&phys, virt);
360 		return;
361 	}
362 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
363 			     NULL);
364 }
365 
create_pgd_mapping(struct mm_struct * mm,phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot)366 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
367 			       unsigned long virt, phys_addr_t size,
368 			       pgprot_t prot)
369 {
370 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
371 			     late_pgtable_alloc);
372 }
373 
create_mapping_late(phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot)374 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
375 				  phys_addr_t size, pgprot_t prot)
376 {
377 	if (virt < VMALLOC_START) {
378 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
379 			&phys, virt);
380 		return;
381 	}
382 
383 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
384 			     late_pgtable_alloc);
385 }
386 
__map_memblock(pgd_t * pgd,phys_addr_t start,phys_addr_t end)387 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
388 {
389 	unsigned long kernel_start = __pa_symbol(_stext);
390 	unsigned long kernel_end = __pa_symbol(__init_begin);
391 
392 	/*
393 	 * Take care not to create a writable alias for the
394 	 * read-only text and rodata sections of the kernel image.
395 	 */
396 
397 	/* No overlap with the kernel text/rodata */
398 	if (end < kernel_start || start >= kernel_end) {
399 		__create_pgd_mapping(pgd, start, __phys_to_virt(start),
400 				     end - start, PAGE_KERNEL,
401 				     early_pgtable_alloc);
402 		return;
403 	}
404 
405 	/*
406 	 * This block overlaps the kernel text/rodata mapping.
407 	 * Map the portion(s) which don't overlap.
408 	 */
409 	if (start < kernel_start)
410 		__create_pgd_mapping(pgd, start,
411 				     __phys_to_virt(start),
412 				     kernel_start - start, PAGE_KERNEL,
413 				     early_pgtable_alloc);
414 	if (kernel_end < end)
415 		__create_pgd_mapping(pgd, kernel_end,
416 				     __phys_to_virt(kernel_end),
417 				     end - kernel_end, PAGE_KERNEL,
418 				     early_pgtable_alloc);
419 
420 	/*
421 	 * Map the linear alias of the [_stext, __init_begin) interval as
422 	 * read-only/non-executable. This makes the contents of the
423 	 * region accessible to subsystems such as hibernate, but
424 	 * protects it from inadvertent modification or execution.
425 	 */
426 	__create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
427 			     kernel_end - kernel_start, PAGE_KERNEL_RO,
428 			     early_pgtable_alloc);
429 }
430 
map_mem(pgd_t * pgd)431 static void __init map_mem(pgd_t *pgd)
432 {
433 	struct memblock_region *reg;
434 
435 	/* map all the memory banks */
436 	for_each_memblock(memory, reg) {
437 		phys_addr_t start = reg->base;
438 		phys_addr_t end = start + reg->size;
439 
440 		if (start >= end)
441 			break;
442 		if (memblock_is_nomap(reg))
443 			continue;
444 
445 		__map_memblock(pgd, start, end);
446 	}
447 }
448 
mark_rodata_ro(void)449 void mark_rodata_ro(void)
450 {
451 	unsigned long section_size;
452 
453 	section_size = (unsigned long)_etext - (unsigned long)_stext;
454 	create_mapping_late(__pa_symbol(_stext), (unsigned long)_stext,
455 			    section_size, PAGE_KERNEL_ROX);
456 	/*
457 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
458 	 * to cover NOTES and EXCEPTION_TABLE.
459 	 */
460 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
461 	create_mapping_late(__pa_symbol(__start_rodata),
462 			    (unsigned long)__start_rodata,
463 			    section_size, PAGE_KERNEL_RO);
464 }
465 
fixup_init(void)466 void fixup_init(void)
467 {
468 	/*
469 	 * Unmap the __init region but leave the VM area in place. This
470 	 * prevents the region from being reused for kernel modules, which
471 	 * is not supported by kallsyms.
472 	 */
473 	unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
474 }
475 
map_kernel_chunk(pgd_t * pgd,void * va_start,void * va_end,pgprot_t prot,struct vm_struct * vma)476 static void __init map_kernel_chunk(pgd_t *pgd, void *va_start, void *va_end,
477 				    pgprot_t prot, struct vm_struct *vma)
478 {
479 	phys_addr_t pa_start = __pa_symbol(va_start);
480 	unsigned long size = va_end - va_start;
481 
482 	BUG_ON(!PAGE_ALIGNED(pa_start));
483 	BUG_ON(!PAGE_ALIGNED(size));
484 
485 	__create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
486 			     early_pgtable_alloc);
487 
488 	vma->addr	= va_start;
489 	vma->phys_addr	= pa_start;
490 	vma->size	= size;
491 	vma->flags	= VM_MAP;
492 	vma->caller	= __builtin_return_address(0);
493 
494 	vm_area_add_early(vma);
495 }
496 
497 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
map_entry_trampoline(void)498 static int __init map_entry_trampoline(void)
499 {
500 	extern char __entry_tramp_text_start[];
501 
502 	pgprot_t prot = PAGE_KERNEL_ROX;
503 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
504 
505 	/* The trampoline is always mapped and can therefore be global */
506 	pgprot_val(prot) &= ~PTE_NG;
507 
508 	/* Map only the text into the trampoline page table */
509 	memset(tramp_pg_dir, 0, PGD_SIZE);
510 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
511 			     prot, late_pgtable_alloc);
512 
513 	/* Map both the text and data into the kernel page table */
514 	__set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
515 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
516 		extern char __entry_tramp_data_start[];
517 
518 		__set_fixmap(FIX_ENTRY_TRAMP_DATA,
519 			     __pa_symbol(__entry_tramp_data_start),
520 			     PAGE_KERNEL_RO);
521 	}
522 
523 	return 0;
524 }
525 core_initcall(map_entry_trampoline);
526 #endif
527 
528 /*
529  * Create fine-grained mappings for the kernel.
530  */
map_kernel(pgd_t * pgd)531 static void __init map_kernel(pgd_t *pgd)
532 {
533 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
534 
535 	map_kernel_chunk(pgd, _stext, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
536 	map_kernel_chunk(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
537 	map_kernel_chunk(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
538 			 &vmlinux_init);
539 	map_kernel_chunk(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
540 
541 	if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
542 		/*
543 		 * The fixmap falls in a separate pgd to the kernel, and doesn't
544 		 * live in the carveout for the swapper_pg_dir. We can simply
545 		 * re-use the existing dir for the fixmap.
546 		 */
547 		set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
548 			*pgd_offset_k(FIXADDR_START));
549 	} else if (CONFIG_PGTABLE_LEVELS > 3) {
550 		/*
551 		 * The fixmap shares its top level pgd entry with the kernel
552 		 * mapping. This can really only occur when we are running
553 		 * with 16k/4 levels, so we can simply reuse the pud level
554 		 * entry instead.
555 		 */
556 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
557 		set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
558 			__pud(__pa_symbol(bm_pmd) | PUD_TYPE_TABLE));
559 		pud_clear_fixmap();
560 	} else {
561 		BUG();
562 	}
563 
564 	kasan_copy_shadow(pgd);
565 }
566 
567 /*
568  * paging_init() sets up the page tables, initialises the zone memory
569  * maps and sets up the zero page.
570  */
paging_init(void)571 void __init paging_init(void)
572 {
573 	phys_addr_t pgd_phys = early_pgtable_alloc();
574 	pgd_t *pgd = pgd_set_fixmap(pgd_phys);
575 
576 	map_kernel(pgd);
577 	map_mem(pgd);
578 
579 	/*
580 	 * We want to reuse the original swapper_pg_dir so we don't have to
581 	 * communicate the new address to non-coherent secondaries in
582 	 * secondary_entry, and so cpu_switch_mm can generate the address with
583 	 * adrp+add rather than a load from some global variable.
584 	 *
585 	 * To do this we need to go via a temporary pgd.
586 	 */
587 	cpu_replace_ttbr1(__va(pgd_phys));
588 	memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
589 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
590 
591 	pgd_clear_fixmap();
592 	memblock_free(pgd_phys, PAGE_SIZE);
593 
594 	/* Ensure the zero page is visible to the page table walker */
595 	dsb(ishst);
596 
597 	/*
598 	 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
599 	 * allocated with it.
600 	 */
601 	memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE,
602 		      SWAPPER_DIR_SIZE - PAGE_SIZE);
603 
604 	bootmem_init();
605 }
606 
607 /*
608  * Check whether a kernel address is valid (derived from arch/x86/).
609  */
kern_addr_valid(unsigned long addr)610 int kern_addr_valid(unsigned long addr)
611 {
612 	pgd_t *pgd;
613 	pud_t *pud;
614 	pmd_t *pmd;
615 	pte_t *pte;
616 
617 	if ((((long)addr) >> VA_BITS) != -1UL)
618 		return 0;
619 
620 	pgd = pgd_offset_k(addr);
621 	if (pgd_none(*pgd))
622 		return 0;
623 
624 	pud = pud_offset(pgd, addr);
625 	if (pud_none(*pud))
626 		return 0;
627 
628 	if (pud_sect(*pud))
629 		return pfn_valid(pud_pfn(*pud));
630 
631 	pmd = pmd_offset(pud, addr);
632 	if (pmd_none(*pmd))
633 		return 0;
634 
635 	if (pmd_sect(*pmd))
636 		return pfn_valid(pmd_pfn(*pmd));
637 
638 	pte = pte_offset_kernel(pmd, addr);
639 	if (pte_none(*pte))
640 		return 0;
641 
642 	return pfn_valid(pte_pfn(*pte));
643 }
644 #ifdef CONFIG_SPARSEMEM_VMEMMAP
645 #if !ARM64_SWAPPER_USES_SECTION_MAPS
vmemmap_populate(unsigned long start,unsigned long end,int node)646 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
647 {
648 	return vmemmap_populate_basepages(start, end, node);
649 }
650 #else	/* !ARM64_SWAPPER_USES_SECTION_MAPS */
vmemmap_populate(unsigned long start,unsigned long end,int node)651 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
652 {
653 	unsigned long addr = start;
654 	unsigned long next;
655 	pgd_t *pgd;
656 	pud_t *pud;
657 	pmd_t *pmd;
658 
659 	do {
660 		next = pmd_addr_end(addr, end);
661 
662 		pgd = vmemmap_pgd_populate(addr, node);
663 		if (!pgd)
664 			return -ENOMEM;
665 
666 		pud = vmemmap_pud_populate(pgd, addr, node);
667 		if (!pud)
668 			return -ENOMEM;
669 
670 		pmd = pmd_offset(pud, addr);
671 		if (pmd_none(*pmd)) {
672 			void *p = NULL;
673 
674 			p = vmemmap_alloc_block_buf(PMD_SIZE, node);
675 			if (!p)
676 				return -ENOMEM;
677 
678 			set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
679 		} else
680 			vmemmap_verify((pte_t *)pmd, node, addr, next);
681 	} while (addr = next, addr != end);
682 
683 	return 0;
684 }
685 #endif	/* CONFIG_ARM64_64K_PAGES */
vmemmap_free(unsigned long start,unsigned long end)686 void vmemmap_free(unsigned long start, unsigned long end)
687 {
688 }
689 #endif	/* CONFIG_SPARSEMEM_VMEMMAP */
690 
fixmap_pud(unsigned long addr)691 static inline pud_t * fixmap_pud(unsigned long addr)
692 {
693 	pgd_t *pgd = pgd_offset_k(addr);
694 
695 	BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
696 
697 	return pud_offset_kimg(pgd, addr);
698 }
699 
fixmap_pmd(unsigned long addr)700 static inline pmd_t * fixmap_pmd(unsigned long addr)
701 {
702 	pud_t *pud = fixmap_pud(addr);
703 
704 	BUG_ON(pud_none(*pud) || pud_bad(*pud));
705 
706 	return pmd_offset_kimg(pud, addr);
707 }
708 
fixmap_pte(unsigned long addr)709 static inline pte_t * fixmap_pte(unsigned long addr)
710 {
711 	return &bm_pte[pte_index(addr)];
712 }
713 
714 /*
715  * The p*d_populate functions call virt_to_phys implicitly so they can't be used
716  * directly on kernel symbols (bm_p*d). This function is called too early to use
717  * lm_alias so __p*d_populate functions must be used to populate with the
718  * physical address from __pa_symbol.
719  */
early_fixmap_init(void)720 void __init early_fixmap_init(void)
721 {
722 	pgd_t *pgd;
723 	pud_t *pud;
724 	pmd_t *pmd;
725 	unsigned long addr = FIXADDR_START;
726 
727 	pgd = pgd_offset_k(addr);
728 	if (CONFIG_PGTABLE_LEVELS > 3 &&
729 	    !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa_symbol(bm_pud))) {
730 		/*
731 		 * We only end up here if the kernel mapping and the fixmap
732 		 * share the top level pgd entry, which should only happen on
733 		 * 16k/4 levels configurations.
734 		 */
735 		BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
736 		pud = pud_offset_kimg(pgd, addr);
737 	} else {
738 		if (pgd_none(*pgd))
739 			__pgd_populate(pgd, __pa_symbol(bm_pud),
740 				       PUD_TYPE_TABLE);
741 		pud = fixmap_pud(addr);
742 	}
743 	if (pud_none(*pud))
744 		__pud_populate(pud, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
745 	pmd = fixmap_pmd(addr);
746 	__pmd_populate(pmd, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
747 
748 	/*
749 	 * The boot-ioremap range spans multiple pmds, for which
750 	 * we are not prepared:
751 	 */
752 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
753 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
754 
755 	if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
756 	     || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
757 		WARN_ON(1);
758 		pr_warn("pmd %p != %p, %p\n",
759 			pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
760 			fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
761 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
762 			fix_to_virt(FIX_BTMAP_BEGIN));
763 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
764 			fix_to_virt(FIX_BTMAP_END));
765 
766 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
767 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
768 	}
769 }
770 
__set_fixmap(enum fixed_addresses idx,phys_addr_t phys,pgprot_t flags)771 void __set_fixmap(enum fixed_addresses idx,
772 			       phys_addr_t phys, pgprot_t flags)
773 {
774 	unsigned long addr = __fix_to_virt(idx);
775 	pte_t *pte;
776 
777 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
778 
779 	pte = fixmap_pte(addr);
780 
781 	if (pgprot_val(flags)) {
782 		set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
783 	} else {
784 		pte_clear(&init_mm, addr, pte);
785 		flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
786 	}
787 }
788 
__fixmap_remap_fdt(phys_addr_t dt_phys,int * size,pgprot_t prot)789 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
790 {
791 	const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
792 	int offset;
793 	void *dt_virt;
794 
795 	/*
796 	 * Check whether the physical FDT address is set and meets the minimum
797 	 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
798 	 * at least 8 bytes so that we can always access the magic and size
799 	 * fields of the FDT header after mapping the first chunk, double check
800 	 * here if that is indeed the case.
801 	 */
802 	BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
803 	if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
804 		return NULL;
805 
806 	/*
807 	 * Make sure that the FDT region can be mapped without the need to
808 	 * allocate additional translation table pages, so that it is safe
809 	 * to call create_mapping_noalloc() this early.
810 	 *
811 	 * On 64k pages, the FDT will be mapped using PTEs, so we need to
812 	 * be in the same PMD as the rest of the fixmap.
813 	 * On 4k pages, we'll use section mappings for the FDT so we only
814 	 * have to be in the same PUD.
815 	 */
816 	BUILD_BUG_ON(dt_virt_base % SZ_2M);
817 
818 	BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
819 		     __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
820 
821 	offset = dt_phys % SWAPPER_BLOCK_SIZE;
822 	dt_virt = (void *)dt_virt_base + offset;
823 
824 	/* map the first chunk so we can read the size from the header */
825 	create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
826 			dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
827 
828 	if (fdt_magic(dt_virt) != FDT_MAGIC)
829 		return NULL;
830 
831 	*size = fdt_totalsize(dt_virt);
832 	if (*size > MAX_FDT_SIZE)
833 		return NULL;
834 
835 	if (offset + *size > SWAPPER_BLOCK_SIZE)
836 		create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
837 			       round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
838 
839 	return dt_virt;
840 }
841 
fixmap_remap_fdt(phys_addr_t dt_phys)842 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
843 {
844 	void *dt_virt;
845 	int size;
846 
847 	dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
848 	if (!dt_virt)
849 		return NULL;
850 
851 	memblock_reserve(dt_phys, size);
852 	return dt_virt;
853 }
854 
arch_ioremap_pud_supported(void)855 int __init arch_ioremap_pud_supported(void)
856 {
857 	/*
858 	 * Only 4k granule supports level 1 block mappings.
859 	 * SW table walks can't handle removal of intermediate entries.
860 	 */
861 	return IS_ENABLED(CONFIG_ARM64_4K_PAGES) &&
862 	       !IS_ENABLED(CONFIG_ARM64_PTDUMP_DEBUGFS);
863 }
864 
arch_ioremap_pmd_supported(void)865 int __init arch_ioremap_pmd_supported(void)
866 {
867 	/* See arch_ioremap_pud_supported() */
868 	return !IS_ENABLED(CONFIG_ARM64_PTDUMP_DEBUGFS);
869 }
870 
pud_set_huge(pud_t * pud,phys_addr_t phys,pgprot_t prot)871 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
872 {
873 	BUG_ON(phys & ~PUD_MASK);
874 	set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
875 	return 1;
876 }
877 
pmd_set_huge(pmd_t * pmd,phys_addr_t phys,pgprot_t prot)878 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
879 {
880 	BUG_ON(phys & ~PMD_MASK);
881 	set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
882 	return 1;
883 }
884 
pud_clear_huge(pud_t * pud)885 int pud_clear_huge(pud_t *pud)
886 {
887 	if (!pud_sect(*pud))
888 		return 0;
889 	pud_clear(pud);
890 	return 1;
891 }
892 
pmd_clear_huge(pmd_t * pmd)893 int pmd_clear_huge(pmd_t *pmd)
894 {
895 	if (!pmd_sect(*pmd))
896 		return 0;
897 	pmd_clear(pmd);
898 	return 1;
899 }
900 
901 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
pud_free_pmd_page(pud_t * pud,unsigned long addr)902 int pud_free_pmd_page(pud_t *pud, unsigned long addr)
903 {
904 	return pud_none(*pud);
905 }
906 
pmd_free_pte_page(pmd_t * pmd,unsigned long addr)907 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
908 {
909 	return pmd_none(*pmd);
910 }
911 #endif
912