Loading...
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
33#include <asm/barrier.h>
34#include <asm/cputype.h>
35#include <asm/fixmap.h>
36#include <asm/kasan.h>
37#include <asm/kernel-pgtable.h>
38#include <asm/sections.h>
39#include <asm/setup.h>
40#include <asm/sizes.h>
41#include <asm/tlb.h>
42#include <asm/memblock.h>
43#include <asm/mmu_context.h>
44
45#include "mm.h"
46
47u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
48
49u64 kimage_voffset __read_mostly;
50EXPORT_SYMBOL(kimage_voffset);
51
52/*
53 * Empty_zero_page is a special page that is used for zero-initialized data
54 * and COW.
55 */
56unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
57EXPORT_SYMBOL(empty_zero_page);
58
59static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
60static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
61static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
62
63pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
64 unsigned long size, pgprot_t vma_prot)
65{
66 if (!pfn_valid(pfn))
67 return pgprot_noncached(vma_prot);
68 else if (file->f_flags & O_SYNC)
69 return pgprot_writecombine(vma_prot);
70 return vma_prot;
71}
72EXPORT_SYMBOL(phys_mem_access_prot);
73
74static phys_addr_t __init early_pgtable_alloc(void)
75{
76 phys_addr_t phys;
77 void *ptr;
78
79 phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
80 BUG_ON(!phys);
81
82 /*
83 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
84 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
85 * any level of table.
86 */
87 ptr = pte_set_fixmap(phys);
88
89 memset(ptr, 0, PAGE_SIZE);
90
91 /*
92 * Implicit barriers also ensure the zeroed page is visible to the page
93 * table walker
94 */
95 pte_clear_fixmap();
96
97 return phys;
98}
99
100/*
101 * remap a PMD into pages
102 */
103static void split_pmd(pmd_t *pmd, pte_t *pte)
104{
105 unsigned long pfn = pmd_pfn(*pmd);
106 int i = 0;
107
108 do {
109 /*
110 * Need to have the least restrictive permissions available
111 * permissions will be fixed up later
112 */
113 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
114 pfn++;
115 } while (pte++, i++, i < PTRS_PER_PTE);
116}
117
118static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
119 unsigned long end, unsigned long pfn,
120 pgprot_t prot,
121 phys_addr_t (*pgtable_alloc)(void))
122{
123 pte_t *pte;
124
125 if (pmd_none(*pmd) || pmd_sect(*pmd)) {
126 phys_addr_t pte_phys;
127 BUG_ON(!pgtable_alloc);
128 pte_phys = pgtable_alloc();
129 pte = pte_set_fixmap(pte_phys);
130 if (pmd_sect(*pmd))
131 split_pmd(pmd, pte);
132 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
133 flush_tlb_all();
134 pte_clear_fixmap();
135 }
136 BUG_ON(pmd_bad(*pmd));
137
138 pte = pte_set_fixmap_offset(pmd, addr);
139 do {
140 set_pte(pte, pfn_pte(pfn, prot));
141 pfn++;
142 } while (pte++, addr += PAGE_SIZE, addr != end);
143
144 pte_clear_fixmap();
145}
146
147static void split_pud(pud_t *old_pud, pmd_t *pmd)
148{
149 unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
150 pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
151 int i = 0;
152
153 do {
154 set_pmd(pmd, __pmd(addr | pgprot_val(prot)));
155 addr += PMD_SIZE;
156 } while (pmd++, i++, i < PTRS_PER_PMD);
157}
158
159#ifdef CONFIG_DEBUG_PAGEALLOC
160static bool block_mappings_allowed(phys_addr_t (*pgtable_alloc)(void))
161{
162
163 /*
164 * If debug_page_alloc is enabled we must map the linear map
165 * using pages. However, other mappings created by
166 * create_mapping_noalloc must use sections in some cases. Allow
167 * sections to be used in those cases, where no pgtable_alloc
168 * function is provided.
169 */
170 return !pgtable_alloc || !debug_pagealloc_enabled();
171}
172#else
173static bool block_mappings_allowed(phys_addr_t (*pgtable_alloc)(void))
174{
175 return true;
176}
177#endif
178
179static void alloc_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))
182{
183 pmd_t *pmd;
184 unsigned long next;
185
186 /*
187 * Check for initial section mappings in the pgd/pud and remove them.
188 */
189 if (pud_none(*pud) || pud_sect(*pud)) {
190 phys_addr_t pmd_phys;
191 BUG_ON(!pgtable_alloc);
192 pmd_phys = pgtable_alloc();
193 pmd = pmd_set_fixmap(pmd_phys);
194 if (pud_sect(*pud)) {
195 /*
196 * need to have the 1G of mappings continue to be
197 * present
198 */
199 split_pud(pud, pmd);
200 }
201 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
202 flush_tlb_all();
203 pmd_clear_fixmap();
204 }
205 BUG_ON(pud_bad(*pud));
206
207 pmd = pmd_set_fixmap_offset(pud, addr);
208 do {
209 next = pmd_addr_end(addr, end);
210 /* try section mapping first */
211 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
212 block_mappings_allowed(pgtable_alloc)) {
213 pmd_t old_pmd =*pmd;
214 pmd_set_huge(pmd, phys, prot);
215 /*
216 * Check for previous table entries created during
217 * boot (__create_page_tables) and flush them.
218 */
219 if (!pmd_none(old_pmd)) {
220 flush_tlb_all();
221 if (pmd_table(old_pmd)) {
222 phys_addr_t table = pmd_page_paddr(old_pmd);
223 if (!WARN_ON_ONCE(slab_is_available()))
224 memblock_free(table, PAGE_SIZE);
225 }
226 }
227 } else {
228 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
229 prot, pgtable_alloc);
230 }
231 phys += next - addr;
232 } while (pmd++, addr = next, addr != end);
233
234 pmd_clear_fixmap();
235}
236
237static inline bool use_1G_block(unsigned long addr, unsigned long next,
238 unsigned long phys)
239{
240 if (PAGE_SHIFT != 12)
241 return false;
242
243 if (((addr | next | phys) & ~PUD_MASK) != 0)
244 return false;
245
246 return true;
247}
248
249static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
250 phys_addr_t phys, pgprot_t prot,
251 phys_addr_t (*pgtable_alloc)(void))
252{
253 pud_t *pud;
254 unsigned long next;
255
256 if (pgd_none(*pgd)) {
257 phys_addr_t pud_phys;
258 BUG_ON(!pgtable_alloc);
259 pud_phys = pgtable_alloc();
260 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
261 }
262 BUG_ON(pgd_bad(*pgd));
263
264 pud = pud_set_fixmap_offset(pgd, addr);
265 do {
266 next = pud_addr_end(addr, end);
267
268 /*
269 * For 4K granule only, attempt to put down a 1GB block
270 */
271 if (use_1G_block(addr, next, phys) &&
272 block_mappings_allowed(pgtable_alloc)) {
273 pud_t old_pud = *pud;
274 pud_set_huge(pud, phys, prot);
275
276 /*
277 * If we have an old value for a pud, it will
278 * be pointing to a pmd table that we no longer
279 * need (from swapper_pg_dir).
280 *
281 * Look up the old pmd table and free it.
282 */
283 if (!pud_none(old_pud)) {
284 flush_tlb_all();
285 if (pud_table(old_pud)) {
286 phys_addr_t table = pud_page_paddr(old_pud);
287 if (!WARN_ON_ONCE(slab_is_available()))
288 memblock_free(table, PAGE_SIZE);
289 }
290 }
291 } else {
292 alloc_init_pmd(pud, addr, next, phys, prot,
293 pgtable_alloc);
294 }
295 phys += next - addr;
296 } while (pud++, addr = next, addr != end);
297
298 pud_clear_fixmap();
299}
300
301/*
302 * Create the page directory entries and any necessary page tables for the
303 * mapping specified by 'md'.
304 */
305static void init_pgd(pgd_t *pgd, phys_addr_t phys, unsigned long virt,
306 phys_addr_t size, pgprot_t prot,
307 phys_addr_t (*pgtable_alloc)(void))
308{
309 unsigned long addr, length, end, next;
310
311 /*
312 * If the virtual and physical address don't have the same offset
313 * within a page, we cannot map the region as the caller expects.
314 */
315 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
316 return;
317
318 phys &= PAGE_MASK;
319 addr = virt & PAGE_MASK;
320 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
321
322 end = addr + length;
323 do {
324 next = pgd_addr_end(addr, end);
325 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc);
326 phys += next - addr;
327 } while (pgd++, addr = next, addr != end);
328}
329
330static phys_addr_t late_pgtable_alloc(void)
331{
332 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
333 BUG_ON(!ptr);
334
335 /* Ensure the zeroed page is visible to the page table walker */
336 dsb(ishst);
337 return __pa(ptr);
338}
339
340static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
341 unsigned long virt, phys_addr_t size,
342 pgprot_t prot,
343 phys_addr_t (*alloc)(void))
344{
345 init_pgd(pgd_offset_raw(pgdir, virt), phys, virt, size, prot, alloc);
346}
347
348/*
349 * This function can only be used to modify existing table entries,
350 * without allocating new levels of table. Note that this permits the
351 * creation of new section or page entries.
352 */
353static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
354 phys_addr_t size, pgprot_t prot)
355{
356 if (virt < VMALLOC_START) {
357 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
358 &phys, virt);
359 return;
360 }
361 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
362 NULL);
363}
364
365void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
366 unsigned long virt, phys_addr_t size,
367 pgprot_t prot)
368{
369 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
370 late_pgtable_alloc);
371}
372
373static void create_mapping_late(phys_addr_t phys, unsigned long virt,
374 phys_addr_t size, pgprot_t prot)
375{
376 if (virt < VMALLOC_START) {
377 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
378 &phys, virt);
379 return;
380 }
381
382 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
383 late_pgtable_alloc);
384}
385
386static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
387{
388 unsigned long kernel_start = __pa(_stext);
389 unsigned long kernel_end = __pa(_etext);
390
391 /*
392 * Take care not to create a writable alias for the
393 * read-only text and rodata sections of the kernel image.
394 */
395
396 /* No overlap with the kernel text */
397 if (end < kernel_start || start >= kernel_end) {
398 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
399 end - start, PAGE_KERNEL,
400 early_pgtable_alloc);
401 return;
402 }
403
404 /*
405 * This block overlaps the kernel text mapping.
406 * Map the portion(s) which don't overlap.
407 */
408 if (start < kernel_start)
409 __create_pgd_mapping(pgd, start,
410 __phys_to_virt(start),
411 kernel_start - start, PAGE_KERNEL,
412 early_pgtable_alloc);
413 if (kernel_end < end)
414 __create_pgd_mapping(pgd, kernel_end,
415 __phys_to_virt(kernel_end),
416 end - kernel_end, PAGE_KERNEL,
417 early_pgtable_alloc);
418
419 /*
420 * Map the linear alias of the [_stext, _etext) interval as
421 * read-only/non-executable. This makes the contents of the
422 * region accessible to subsystems such as hibernate, but
423 * protects it from inadvertent modification or execution.
424 */
425 __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
426 kernel_end - kernel_start, PAGE_KERNEL_RO,
427 early_pgtable_alloc);
428}
429
430static void __init map_mem(pgd_t *pgd)
431{
432 struct memblock_region *reg;
433
434 /* map all the memory banks */
435 for_each_memblock(memory, reg) {
436 phys_addr_t start = reg->base;
437 phys_addr_t end = start + reg->size;
438
439 if (start >= end)
440 break;
441 if (memblock_is_nomap(reg))
442 continue;
443
444 __map_memblock(pgd, start, end);
445 }
446}
447
448void mark_rodata_ro(void)
449{
450 unsigned long section_size;
451
452 section_size = (unsigned long)__start_rodata - (unsigned long)_stext;
453 create_mapping_late(__pa(_stext), (unsigned long)_stext,
454 section_size, PAGE_KERNEL_ROX);
455 /*
456 * mark .rodata as read only. Use _etext rather than __end_rodata to
457 * cover NOTES and EXCEPTION_TABLE.
458 */
459 section_size = (unsigned long)_etext - (unsigned long)__start_rodata;
460 create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
461 section_size, PAGE_KERNEL_RO);
462}
463
464void fixup_init(void)
465{
466 /*
467 * Unmap the __init region but leave the VM area in place. This
468 * prevents the region from being reused for kernel modules, which
469 * is not supported by kallsyms.
470 */
471 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
472}
473
474static void __init map_kernel_chunk(pgd_t *pgd, void *va_start, void *va_end,
475 pgprot_t prot, struct vm_struct *vma)
476{
477 phys_addr_t pa_start = __pa(va_start);
478 unsigned long size = va_end - va_start;
479
480 BUG_ON(!PAGE_ALIGNED(pa_start));
481 BUG_ON(!PAGE_ALIGNED(size));
482
483 __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
484 early_pgtable_alloc);
485
486 vma->addr = va_start;
487 vma->phys_addr = pa_start;
488 vma->size = size;
489 vma->flags = VM_MAP;
490 vma->caller = __builtin_return_address(0);
491
492 vm_area_add_early(vma);
493}
494
495/*
496 * Create fine-grained mappings for the kernel.
497 */
498static void __init map_kernel(pgd_t *pgd)
499{
500 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
501
502 map_kernel_chunk(pgd, _stext, __start_rodata, PAGE_KERNEL_EXEC, &vmlinux_text);
503 map_kernel_chunk(pgd, __start_rodata, _etext, PAGE_KERNEL, &vmlinux_rodata);
504 map_kernel_chunk(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
505 &vmlinux_init);
506 map_kernel_chunk(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
507
508 if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
509 /*
510 * The fixmap falls in a separate pgd to the kernel, and doesn't
511 * live in the carveout for the swapper_pg_dir. We can simply
512 * re-use the existing dir for the fixmap.
513 */
514 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
515 *pgd_offset_k(FIXADDR_START));
516 } else if (CONFIG_PGTABLE_LEVELS > 3) {
517 /*
518 * The fixmap shares its top level pgd entry with the kernel
519 * mapping. This can really only occur when we are running
520 * with 16k/4 levels, so we can simply reuse the pud level
521 * entry instead.
522 */
523 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
524 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
525 __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
526 pud_clear_fixmap();
527 } else {
528 BUG();
529 }
530
531 kasan_copy_shadow(pgd);
532}
533
534/*
535 * paging_init() sets up the page tables, initialises the zone memory
536 * maps and sets up the zero page.
537 */
538void __init paging_init(void)
539{
540 phys_addr_t pgd_phys = early_pgtable_alloc();
541 pgd_t *pgd = pgd_set_fixmap(pgd_phys);
542
543 map_kernel(pgd);
544 map_mem(pgd);
545
546 /*
547 * We want to reuse the original swapper_pg_dir so we don't have to
548 * communicate the new address to non-coherent secondaries in
549 * secondary_entry, and so cpu_switch_mm can generate the address with
550 * adrp+add rather than a load from some global variable.
551 *
552 * To do this we need to go via a temporary pgd.
553 */
554 cpu_replace_ttbr1(__va(pgd_phys));
555 memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
556 cpu_replace_ttbr1(swapper_pg_dir);
557
558 pgd_clear_fixmap();
559 memblock_free(pgd_phys, PAGE_SIZE);
560
561 /*
562 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
563 * allocated with it.
564 */
565 memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
566 SWAPPER_DIR_SIZE - PAGE_SIZE);
567
568 bootmem_init();
569}
570
571/*
572 * Check whether a kernel address is valid (derived from arch/x86/).
573 */
574int kern_addr_valid(unsigned long addr)
575{
576 pgd_t *pgd;
577 pud_t *pud;
578 pmd_t *pmd;
579 pte_t *pte;
580
581 if ((((long)addr) >> VA_BITS) != -1UL)
582 return 0;
583
584 pgd = pgd_offset_k(addr);
585 if (pgd_none(*pgd))
586 return 0;
587
588 pud = pud_offset(pgd, addr);
589 if (pud_none(*pud))
590 return 0;
591
592 if (pud_sect(*pud))
593 return pfn_valid(pud_pfn(*pud));
594
595 pmd = pmd_offset(pud, addr);
596 if (pmd_none(*pmd))
597 return 0;
598
599 if (pmd_sect(*pmd))
600 return pfn_valid(pmd_pfn(*pmd));
601
602 pte = pte_offset_kernel(pmd, addr);
603 if (pte_none(*pte))
604 return 0;
605
606 return pfn_valid(pte_pfn(*pte));
607}
608#ifdef CONFIG_SPARSEMEM_VMEMMAP
609#if !ARM64_SWAPPER_USES_SECTION_MAPS
610int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
611{
612 return vmemmap_populate_basepages(start, end, node);
613}
614#else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
615int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
616{
617 unsigned long addr = start;
618 unsigned long next;
619 pgd_t *pgd;
620 pud_t *pud;
621 pmd_t *pmd;
622
623 do {
624 next = pmd_addr_end(addr, end);
625
626 pgd = vmemmap_pgd_populate(addr, node);
627 if (!pgd)
628 return -ENOMEM;
629
630 pud = vmemmap_pud_populate(pgd, addr, node);
631 if (!pud)
632 return -ENOMEM;
633
634 pmd = pmd_offset(pud, addr);
635 if (pmd_none(*pmd)) {
636 void *p = NULL;
637
638 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
639 if (!p)
640 return -ENOMEM;
641
642 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
643 } else
644 vmemmap_verify((pte_t *)pmd, node, addr, next);
645 } while (addr = next, addr != end);
646
647 return 0;
648}
649#endif /* CONFIG_ARM64_64K_PAGES */
650void vmemmap_free(unsigned long start, unsigned long end)
651{
652}
653#endif /* CONFIG_SPARSEMEM_VMEMMAP */
654
655static inline pud_t * fixmap_pud(unsigned long addr)
656{
657 pgd_t *pgd = pgd_offset_k(addr);
658
659 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
660
661 return pud_offset_kimg(pgd, addr);
662}
663
664static inline pmd_t * fixmap_pmd(unsigned long addr)
665{
666 pud_t *pud = fixmap_pud(addr);
667
668 BUG_ON(pud_none(*pud) || pud_bad(*pud));
669
670 return pmd_offset_kimg(pud, addr);
671}
672
673static inline pte_t * fixmap_pte(unsigned long addr)
674{
675 return &bm_pte[pte_index(addr)];
676}
677
678void __init early_fixmap_init(void)
679{
680 pgd_t *pgd;
681 pud_t *pud;
682 pmd_t *pmd;
683 unsigned long addr = FIXADDR_START;
684
685 pgd = pgd_offset_k(addr);
686 if (CONFIG_PGTABLE_LEVELS > 3 &&
687 !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
688 /*
689 * We only end up here if the kernel mapping and the fixmap
690 * share the top level pgd entry, which should only happen on
691 * 16k/4 levels configurations.
692 */
693 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
694 pud = pud_offset_kimg(pgd, addr);
695 } else {
696 pgd_populate(&init_mm, pgd, bm_pud);
697 pud = fixmap_pud(addr);
698 }
699 pud_populate(&init_mm, pud, bm_pmd);
700 pmd = fixmap_pmd(addr);
701 pmd_populate_kernel(&init_mm, pmd, bm_pte);
702
703 /*
704 * The boot-ioremap range spans multiple pmds, for which
705 * we are not prepared:
706 */
707 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
708 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
709
710 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
711 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
712 WARN_ON(1);
713 pr_warn("pmd %p != %p, %p\n",
714 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
715 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
716 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
717 fix_to_virt(FIX_BTMAP_BEGIN));
718 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
719 fix_to_virt(FIX_BTMAP_END));
720
721 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
722 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
723 }
724}
725
726void __set_fixmap(enum fixed_addresses idx,
727 phys_addr_t phys, pgprot_t flags)
728{
729 unsigned long addr = __fix_to_virt(idx);
730 pte_t *pte;
731
732 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
733
734 pte = fixmap_pte(addr);
735
736 if (pgprot_val(flags)) {
737 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
738 } else {
739 pte_clear(&init_mm, addr, pte);
740 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
741 }
742}
743
744void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
745{
746 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
747 int offset;
748 void *dt_virt;
749
750 /*
751 * Check whether the physical FDT address is set and meets the minimum
752 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
753 * at least 8 bytes so that we can always access the size field of the
754 * FDT header after mapping the first chunk, double check here if that
755 * is indeed the case.
756 */
757 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
758 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
759 return NULL;
760
761 /*
762 * Make sure that the FDT region can be mapped without the need to
763 * allocate additional translation table pages, so that it is safe
764 * to call create_mapping_noalloc() this early.
765 *
766 * On 64k pages, the FDT will be mapped using PTEs, so we need to
767 * be in the same PMD as the rest of the fixmap.
768 * On 4k pages, we'll use section mappings for the FDT so we only
769 * have to be in the same PUD.
770 */
771 BUILD_BUG_ON(dt_virt_base % SZ_2M);
772
773 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
774 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
775
776 offset = dt_phys % SWAPPER_BLOCK_SIZE;
777 dt_virt = (void *)dt_virt_base + offset;
778
779 /* map the first chunk so we can read the size from the header */
780 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
781 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
782
783 if (fdt_check_header(dt_virt) != 0)
784 return NULL;
785
786 *size = fdt_totalsize(dt_virt);
787 if (*size > MAX_FDT_SIZE)
788 return NULL;
789
790 if (offset + *size > SWAPPER_BLOCK_SIZE)
791 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
792 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
793
794 return dt_virt;
795}
796
797void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
798{
799 void *dt_virt;
800 int size;
801
802 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
803 if (!dt_virt)
804 return NULL;
805
806 memblock_reserve(dt_phys, size);
807 return dt_virt;
808}
809
810int __init arch_ioremap_pud_supported(void)
811{
812 /* only 4k granule supports level 1 block mappings */
813 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
814}
815
816int __init arch_ioremap_pmd_supported(void)
817{
818 return 1;
819}
820
821int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
822{
823 BUG_ON(phys & ~PUD_MASK);
824 set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
825 return 1;
826}
827
828int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
829{
830 BUG_ON(phys & ~PMD_MASK);
831 set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
832 return 1;
833}
834
835int pud_clear_huge(pud_t *pud)
836{
837 if (!pud_sect(*pud))
838 return 0;
839 pud_clear(pud);
840 return 1;
841}
842
843int pmd_clear_huge(pmd_t *pmd)
844{
845 if (!pmd_sect(*pmd))
846 return 0;
847 pmd_clear(pmd);
848 return 1;
849}
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
52u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
53u64 idmap_ptrs_per_pgd = PTRS_PER_PGD;
54
55u64 kimage_voffset __ro_after_init;
56EXPORT_SYMBOL(kimage_voffset);
57
58/*
59 * Empty_zero_page is a special page that is used for zero-initialized data
60 * and COW.
61 */
62unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
63EXPORT_SYMBOL(empty_zero_page);
64
65static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
66static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
67static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
68
69pgprot_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}
78EXPORT_SYMBOL(phys_mem_access_prot);
79
80static 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
105static 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 | PTE_NG;
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 /* Transitioning from Non-Global to Global is unsafe */
122 if (old & ~new & PTE_NG)
123 return false;
124
125 return ((old ^ new) & ~mask) == 0;
126}
127
128static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
129 phys_addr_t phys, pgprot_t prot)
130{
131 pte_t *ptep;
132
133 ptep = pte_set_fixmap_offset(pmdp, addr);
134 do {
135 pte_t old_pte = READ_ONCE(*ptep);
136
137 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
138
139 /*
140 * After the PTE entry has been populated once, we
141 * only allow updates to the permission attributes.
142 */
143 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
144 READ_ONCE(pte_val(*ptep))));
145
146 phys += PAGE_SIZE;
147 } while (ptep++, addr += PAGE_SIZE, addr != end);
148
149 pte_clear_fixmap();
150}
151
152static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
153 unsigned long end, phys_addr_t phys,
154 pgprot_t prot,
155 phys_addr_t (*pgtable_alloc)(void),
156 int flags)
157{
158 unsigned long next;
159 pmd_t pmd = READ_ONCE(*pmdp);
160
161 BUG_ON(pmd_sect(pmd));
162 if (pmd_none(pmd)) {
163 phys_addr_t pte_phys;
164 BUG_ON(!pgtable_alloc);
165 pte_phys = pgtable_alloc();
166 __pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE);
167 pmd = READ_ONCE(*pmdp);
168 }
169 BUG_ON(pmd_bad(pmd));
170
171 do {
172 pgprot_t __prot = prot;
173
174 next = pte_cont_addr_end(addr, end);
175
176 /* use a contiguous mapping if the range is suitably aligned */
177 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
178 (flags & NO_CONT_MAPPINGS) == 0)
179 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
180
181 init_pte(pmdp, addr, next, phys, __prot);
182
183 phys += next - addr;
184 } while (addr = next, addr != end);
185}
186
187static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
188 phys_addr_t phys, pgprot_t prot,
189 phys_addr_t (*pgtable_alloc)(void), int flags)
190{
191 unsigned long next;
192 pmd_t *pmdp;
193
194 pmdp = pmd_set_fixmap_offset(pudp, addr);
195 do {
196 pmd_t old_pmd = READ_ONCE(*pmdp);
197
198 next = pmd_addr_end(addr, end);
199
200 /* try section mapping first */
201 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
202 (flags & NO_BLOCK_MAPPINGS) == 0) {
203 pmd_set_huge(pmdp, phys, prot);
204
205 /*
206 * After the PMD entry has been populated once, we
207 * only allow updates to the permission attributes.
208 */
209 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
210 READ_ONCE(pmd_val(*pmdp))));
211 } else {
212 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
213 pgtable_alloc, flags);
214
215 BUG_ON(pmd_val(old_pmd) != 0 &&
216 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
217 }
218 phys += next - addr;
219 } while (pmdp++, addr = next, addr != end);
220
221 pmd_clear_fixmap();
222}
223
224static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
225 unsigned long end, phys_addr_t phys,
226 pgprot_t prot,
227 phys_addr_t (*pgtable_alloc)(void), int flags)
228{
229 unsigned long next;
230 pud_t pud = READ_ONCE(*pudp);
231
232 /*
233 * Check for initial section mappings in the pgd/pud.
234 */
235 BUG_ON(pud_sect(pud));
236 if (pud_none(pud)) {
237 phys_addr_t pmd_phys;
238 BUG_ON(!pgtable_alloc);
239 pmd_phys = pgtable_alloc();
240 __pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE);
241 pud = READ_ONCE(*pudp);
242 }
243 BUG_ON(pud_bad(pud));
244
245 do {
246 pgprot_t __prot = prot;
247
248 next = pmd_cont_addr_end(addr, end);
249
250 /* use a contiguous mapping if the range is suitably aligned */
251 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
252 (flags & NO_CONT_MAPPINGS) == 0)
253 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
254
255 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
256
257 phys += next - addr;
258 } while (addr = next, addr != end);
259}
260
261static inline bool use_1G_block(unsigned long addr, unsigned long next,
262 unsigned long phys)
263{
264 if (PAGE_SHIFT != 12)
265 return false;
266
267 if (((addr | next | phys) & ~PUD_MASK) != 0)
268 return false;
269
270 return true;
271}
272
273static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
274 phys_addr_t phys, pgprot_t prot,
275 phys_addr_t (*pgtable_alloc)(void),
276 int flags)
277{
278 unsigned long next;
279 pud_t *pudp;
280 pgd_t pgd = READ_ONCE(*pgdp);
281
282 if (pgd_none(pgd)) {
283 phys_addr_t pud_phys;
284 BUG_ON(!pgtable_alloc);
285 pud_phys = pgtable_alloc();
286 __pgd_populate(pgdp, pud_phys, PUD_TYPE_TABLE);
287 pgd = READ_ONCE(*pgdp);
288 }
289 BUG_ON(pgd_bad(pgd));
290
291 pudp = pud_set_fixmap_offset(pgdp, addr);
292 do {
293 pud_t old_pud = READ_ONCE(*pudp);
294
295 next = pud_addr_end(addr, end);
296
297 /*
298 * For 4K granule only, attempt to put down a 1GB block
299 */
300 if (use_1G_block(addr, next, phys) &&
301 (flags & NO_BLOCK_MAPPINGS) == 0) {
302 pud_set_huge(pudp, phys, prot);
303
304 /*
305 * After the PUD entry has been populated once, we
306 * only allow updates to the permission attributes.
307 */
308 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
309 READ_ONCE(pud_val(*pudp))));
310 } else {
311 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
312 pgtable_alloc, flags);
313
314 BUG_ON(pud_val(old_pud) != 0 &&
315 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
316 }
317 phys += next - addr;
318 } while (pudp++, addr = next, addr != end);
319
320 pud_clear_fixmap();
321}
322
323static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
324 unsigned long virt, phys_addr_t size,
325 pgprot_t prot,
326 phys_addr_t (*pgtable_alloc)(void),
327 int flags)
328{
329 unsigned long addr, length, end, next;
330 pgd_t *pgdp = pgd_offset_raw(pgdir, virt);
331
332 /*
333 * If the virtual and physical address don't have the same offset
334 * within a page, we cannot map the region as the caller expects.
335 */
336 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
337 return;
338
339 phys &= PAGE_MASK;
340 addr = virt & PAGE_MASK;
341 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
342
343 end = addr + length;
344 do {
345 next = pgd_addr_end(addr, end);
346 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
347 flags);
348 phys += next - addr;
349 } while (pgdp++, addr = next, addr != end);
350}
351
352static phys_addr_t pgd_pgtable_alloc(void)
353{
354 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
355 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
356 BUG();
357
358 /* Ensure the zeroed page is visible to the page table walker */
359 dsb(ishst);
360 return __pa(ptr);
361}
362
363/*
364 * This function can only be used to modify existing table entries,
365 * without allocating new levels of table. Note that this permits the
366 * creation of new section or page entries.
367 */
368static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
369 phys_addr_t size, pgprot_t prot)
370{
371 if (virt < VMALLOC_START) {
372 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
373 &phys, virt);
374 return;
375 }
376 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
377 NO_CONT_MAPPINGS);
378}
379
380void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
381 unsigned long virt, phys_addr_t size,
382 pgprot_t prot, bool page_mappings_only)
383{
384 int flags = 0;
385
386 BUG_ON(mm == &init_mm);
387
388 if (page_mappings_only)
389 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
390
391 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
392 pgd_pgtable_alloc, flags);
393}
394
395static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
396 phys_addr_t size, pgprot_t prot)
397{
398 if (virt < VMALLOC_START) {
399 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
400 &phys, virt);
401 return;
402 }
403
404 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
405 NO_CONT_MAPPINGS);
406
407 /* flush the TLBs after updating live kernel mappings */
408 flush_tlb_kernel_range(virt, virt + size);
409}
410
411static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
412 phys_addr_t end, pgprot_t prot, int flags)
413{
414 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
415 prot, early_pgtable_alloc, flags);
416}
417
418void __init mark_linear_text_alias_ro(void)
419{
420 /*
421 * Remove the write permissions from the linear alias of .text/.rodata
422 */
423 update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
424 (unsigned long)__init_begin - (unsigned long)_text,
425 PAGE_KERNEL_RO);
426}
427
428static void __init map_mem(pgd_t *pgdp)
429{
430 phys_addr_t kernel_start = __pa_symbol(_text);
431 phys_addr_t kernel_end = __pa_symbol(__init_begin);
432 struct memblock_region *reg;
433 int flags = 0;
434
435 if (debug_pagealloc_enabled())
436 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
437
438 /*
439 * Take care not to create a writable alias for the
440 * read-only text and rodata sections of the kernel image.
441 * So temporarily mark them as NOMAP to skip mappings in
442 * the following for-loop
443 */
444 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
445#ifdef CONFIG_KEXEC_CORE
446 if (crashk_res.end)
447 memblock_mark_nomap(crashk_res.start,
448 resource_size(&crashk_res));
449#endif
450
451 /* map all the memory banks */
452 for_each_memblock(memory, reg) {
453 phys_addr_t start = reg->base;
454 phys_addr_t end = start + reg->size;
455
456 if (start >= end)
457 break;
458 if (memblock_is_nomap(reg))
459 continue;
460
461 __map_memblock(pgdp, start, end, PAGE_KERNEL, flags);
462 }
463
464 /*
465 * Map the linear alias of the [_text, __init_begin) interval
466 * as non-executable now, and remove the write permission in
467 * mark_linear_text_alias_ro() below (which will be called after
468 * alternative patching has completed). This makes the contents
469 * of the region accessible to subsystems such as hibernate,
470 * but protects it from inadvertent modification or execution.
471 * Note that contiguous mappings cannot be remapped in this way,
472 * so we should avoid them here.
473 */
474 __map_memblock(pgdp, kernel_start, kernel_end,
475 PAGE_KERNEL, NO_CONT_MAPPINGS);
476 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
477
478#ifdef CONFIG_KEXEC_CORE
479 /*
480 * Use page-level mappings here so that we can shrink the region
481 * in page granularity and put back unused memory to buddy system
482 * through /sys/kernel/kexec_crash_size interface.
483 */
484 if (crashk_res.end) {
485 __map_memblock(pgdp, crashk_res.start, crashk_res.end + 1,
486 PAGE_KERNEL,
487 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
488 memblock_clear_nomap(crashk_res.start,
489 resource_size(&crashk_res));
490 }
491#endif
492}
493
494void mark_rodata_ro(void)
495{
496 unsigned long section_size;
497
498 /*
499 * mark .rodata as read only. Use __init_begin rather than __end_rodata
500 * to cover NOTES and EXCEPTION_TABLE.
501 */
502 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
503 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
504 section_size, PAGE_KERNEL_RO);
505
506 debug_checkwx();
507}
508
509static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
510 pgprot_t prot, struct vm_struct *vma,
511 int flags, unsigned long vm_flags)
512{
513 phys_addr_t pa_start = __pa_symbol(va_start);
514 unsigned long size = va_end - va_start;
515
516 BUG_ON(!PAGE_ALIGNED(pa_start));
517 BUG_ON(!PAGE_ALIGNED(size));
518
519 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
520 early_pgtable_alloc, flags);
521
522 if (!(vm_flags & VM_NO_GUARD))
523 size += PAGE_SIZE;
524
525 vma->addr = va_start;
526 vma->phys_addr = pa_start;
527 vma->size = size;
528 vma->flags = VM_MAP | vm_flags;
529 vma->caller = __builtin_return_address(0);
530
531 vm_area_add_early(vma);
532}
533
534static int __init parse_rodata(char *arg)
535{
536 return strtobool(arg, &rodata_enabled);
537}
538early_param("rodata", parse_rodata);
539
540#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
541static int __init map_entry_trampoline(void)
542{
543 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
544 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
545
546 /* The trampoline is always mapped and can therefore be global */
547 pgprot_val(prot) &= ~PTE_NG;
548
549 /* Map only the text into the trampoline page table */
550 memset(tramp_pg_dir, 0, PGD_SIZE);
551 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
552 prot, pgd_pgtable_alloc, 0);
553
554 /* Map both the text and data into the kernel page table */
555 __set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
556 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
557 extern char __entry_tramp_data_start[];
558
559 __set_fixmap(FIX_ENTRY_TRAMP_DATA,
560 __pa_symbol(__entry_tramp_data_start),
561 PAGE_KERNEL_RO);
562 }
563
564 return 0;
565}
566core_initcall(map_entry_trampoline);
567#endif
568
569/*
570 * Create fine-grained mappings for the kernel.
571 */
572static void __init map_kernel(pgd_t *pgdp)
573{
574 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
575 vmlinux_initdata, vmlinux_data;
576
577 /*
578 * External debuggers may need to write directly to the text
579 * mapping to install SW breakpoints. Allow this (only) when
580 * explicitly requested with rodata=off.
581 */
582 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
583
584 /*
585 * Only rodata will be remapped with different permissions later on,
586 * all other segments are allowed to use contiguous mappings.
587 */
588 map_kernel_segment(pgdp, _text, _etext, text_prot, &vmlinux_text, 0,
589 VM_NO_GUARD);
590 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
591 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
592 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
593 &vmlinux_inittext, 0, VM_NO_GUARD);
594 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
595 &vmlinux_initdata, 0, VM_NO_GUARD);
596 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
597
598 if (!READ_ONCE(pgd_val(*pgd_offset_raw(pgdp, FIXADDR_START)))) {
599 /*
600 * The fixmap falls in a separate pgd to the kernel, and doesn't
601 * live in the carveout for the swapper_pg_dir. We can simply
602 * re-use the existing dir for the fixmap.
603 */
604 set_pgd(pgd_offset_raw(pgdp, FIXADDR_START),
605 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
606 } else if (CONFIG_PGTABLE_LEVELS > 3) {
607 /*
608 * The fixmap shares its top level pgd entry with the kernel
609 * mapping. This can really only occur when we are running
610 * with 16k/4 levels, so we can simply reuse the pud level
611 * entry instead.
612 */
613 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
614 pud_populate(&init_mm,
615 pud_set_fixmap_offset(pgdp, FIXADDR_START),
616 lm_alias(bm_pmd));
617 pud_clear_fixmap();
618 } else {
619 BUG();
620 }
621
622 kasan_copy_shadow(pgdp);
623}
624
625/*
626 * paging_init() sets up the page tables, initialises the zone memory
627 * maps and sets up the zero page.
628 */
629void __init paging_init(void)
630{
631 phys_addr_t pgd_phys = early_pgtable_alloc();
632 pgd_t *pgdp = pgd_set_fixmap(pgd_phys);
633
634 map_kernel(pgdp);
635 map_mem(pgdp);
636
637 /*
638 * We want to reuse the original swapper_pg_dir so we don't have to
639 * communicate the new address to non-coherent secondaries in
640 * secondary_entry, and so cpu_switch_mm can generate the address with
641 * adrp+add rather than a load from some global variable.
642 *
643 * To do this we need to go via a temporary pgd.
644 */
645 cpu_replace_ttbr1(__va(pgd_phys));
646 memcpy(swapper_pg_dir, pgdp, PGD_SIZE);
647 cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
648
649 pgd_clear_fixmap();
650 memblock_free(pgd_phys, PAGE_SIZE);
651
652 /*
653 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
654 * allocated with it.
655 */
656 memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE,
657 __pa_symbol(swapper_pg_end) - __pa_symbol(swapper_pg_dir)
658 - PAGE_SIZE);
659}
660
661/*
662 * Check whether a kernel address is valid (derived from arch/x86/).
663 */
664int kern_addr_valid(unsigned long addr)
665{
666 pgd_t *pgdp;
667 pud_t *pudp, pud;
668 pmd_t *pmdp, pmd;
669 pte_t *ptep, pte;
670
671 if ((((long)addr) >> VA_BITS) != -1UL)
672 return 0;
673
674 pgdp = pgd_offset_k(addr);
675 if (pgd_none(READ_ONCE(*pgdp)))
676 return 0;
677
678 pudp = pud_offset(pgdp, addr);
679 pud = READ_ONCE(*pudp);
680 if (pud_none(pud))
681 return 0;
682
683 if (pud_sect(pud))
684 return pfn_valid(pud_pfn(pud));
685
686 pmdp = pmd_offset(pudp, addr);
687 pmd = READ_ONCE(*pmdp);
688 if (pmd_none(pmd))
689 return 0;
690
691 if (pmd_sect(pmd))
692 return pfn_valid(pmd_pfn(pmd));
693
694 ptep = pte_offset_kernel(pmdp, addr);
695 pte = READ_ONCE(*ptep);
696 if (pte_none(pte))
697 return 0;
698
699 return pfn_valid(pte_pfn(pte));
700}
701#ifdef CONFIG_SPARSEMEM_VMEMMAP
702#if !ARM64_SWAPPER_USES_SECTION_MAPS
703int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
704 struct vmem_altmap *altmap)
705{
706 return vmemmap_populate_basepages(start, end, node);
707}
708#else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
709int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
710 struct vmem_altmap *altmap)
711{
712 unsigned long addr = start;
713 unsigned long next;
714 pgd_t *pgdp;
715 pud_t *pudp;
716 pmd_t *pmdp;
717
718 do {
719 next = pmd_addr_end(addr, end);
720
721 pgdp = vmemmap_pgd_populate(addr, node);
722 if (!pgdp)
723 return -ENOMEM;
724
725 pudp = vmemmap_pud_populate(pgdp, addr, node);
726 if (!pudp)
727 return -ENOMEM;
728
729 pmdp = pmd_offset(pudp, addr);
730 if (pmd_none(READ_ONCE(*pmdp))) {
731 void *p = NULL;
732
733 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
734 if (!p)
735 return -ENOMEM;
736
737 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
738 } else
739 vmemmap_verify((pte_t *)pmdp, node, addr, next);
740 } while (addr = next, addr != end);
741
742 return 0;
743}
744#endif /* CONFIG_ARM64_64K_PAGES */
745void vmemmap_free(unsigned long start, unsigned long end,
746 struct vmem_altmap *altmap)
747{
748}
749#endif /* CONFIG_SPARSEMEM_VMEMMAP */
750
751static inline pud_t * fixmap_pud(unsigned long addr)
752{
753 pgd_t *pgdp = pgd_offset_k(addr);
754 pgd_t pgd = READ_ONCE(*pgdp);
755
756 BUG_ON(pgd_none(pgd) || pgd_bad(pgd));
757
758 return pud_offset_kimg(pgdp, addr);
759}
760
761static inline pmd_t * fixmap_pmd(unsigned long addr)
762{
763 pud_t *pudp = fixmap_pud(addr);
764 pud_t pud = READ_ONCE(*pudp);
765
766 BUG_ON(pud_none(pud) || pud_bad(pud));
767
768 return pmd_offset_kimg(pudp, addr);
769}
770
771static inline pte_t * fixmap_pte(unsigned long addr)
772{
773 return &bm_pte[pte_index(addr)];
774}
775
776/*
777 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
778 * directly on kernel symbols (bm_p*d). This function is called too early to use
779 * lm_alias so __p*d_populate functions must be used to populate with the
780 * physical address from __pa_symbol.
781 */
782void __init early_fixmap_init(void)
783{
784 pgd_t *pgdp, pgd;
785 pud_t *pudp;
786 pmd_t *pmdp;
787 unsigned long addr = FIXADDR_START;
788
789 pgdp = pgd_offset_k(addr);
790 pgd = READ_ONCE(*pgdp);
791 if (CONFIG_PGTABLE_LEVELS > 3 &&
792 !(pgd_none(pgd) || pgd_page_paddr(pgd) == __pa_symbol(bm_pud))) {
793 /*
794 * We only end up here if the kernel mapping and the fixmap
795 * share the top level pgd entry, which should only happen on
796 * 16k/4 levels configurations.
797 */
798 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
799 pudp = pud_offset_kimg(pgdp, addr);
800 } else {
801 if (pgd_none(pgd))
802 __pgd_populate(pgdp, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
803 pudp = fixmap_pud(addr);
804 }
805 if (pud_none(READ_ONCE(*pudp)))
806 __pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
807 pmdp = fixmap_pmd(addr);
808 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
809
810 /*
811 * The boot-ioremap range spans multiple pmds, for which
812 * we are not prepared:
813 */
814 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
815 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
816
817 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
818 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
819 WARN_ON(1);
820 pr_warn("pmdp %p != %p, %p\n",
821 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
822 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
823 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
824 fix_to_virt(FIX_BTMAP_BEGIN));
825 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
826 fix_to_virt(FIX_BTMAP_END));
827
828 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
829 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
830 }
831}
832
833/*
834 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
835 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
836 */
837void __set_fixmap(enum fixed_addresses idx,
838 phys_addr_t phys, pgprot_t flags)
839{
840 unsigned long addr = __fix_to_virt(idx);
841 pte_t *ptep;
842
843 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
844
845 ptep = fixmap_pte(addr);
846
847 if (pgprot_val(flags)) {
848 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
849 } else {
850 pte_clear(&init_mm, addr, ptep);
851 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
852 }
853}
854
855void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
856{
857 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
858 int offset;
859 void *dt_virt;
860
861 /*
862 * Check whether the physical FDT address is set and meets the minimum
863 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
864 * at least 8 bytes so that we can always access the magic and size
865 * fields of the FDT header after mapping the first chunk, double check
866 * here if that is indeed the case.
867 */
868 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
869 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
870 return NULL;
871
872 /*
873 * Make sure that the FDT region can be mapped without the need to
874 * allocate additional translation table pages, so that it is safe
875 * to call create_mapping_noalloc() this early.
876 *
877 * On 64k pages, the FDT will be mapped using PTEs, so we need to
878 * be in the same PMD as the rest of the fixmap.
879 * On 4k pages, we'll use section mappings for the FDT so we only
880 * have to be in the same PUD.
881 */
882 BUILD_BUG_ON(dt_virt_base % SZ_2M);
883
884 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
885 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
886
887 offset = dt_phys % SWAPPER_BLOCK_SIZE;
888 dt_virt = (void *)dt_virt_base + offset;
889
890 /* map the first chunk so we can read the size from the header */
891 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
892 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
893
894 if (fdt_magic(dt_virt) != FDT_MAGIC)
895 return NULL;
896
897 *size = fdt_totalsize(dt_virt);
898 if (*size > MAX_FDT_SIZE)
899 return NULL;
900
901 if (offset + *size > SWAPPER_BLOCK_SIZE)
902 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
903 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
904
905 return dt_virt;
906}
907
908void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
909{
910 void *dt_virt;
911 int size;
912
913 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
914 if (!dt_virt)
915 return NULL;
916
917 memblock_reserve(dt_phys, size);
918 return dt_virt;
919}
920
921int __init arch_ioremap_pud_supported(void)
922{
923 /* only 4k granule supports level 1 block mappings */
924 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
925}
926
927int __init arch_ioremap_pmd_supported(void)
928{
929 return 1;
930}
931
932int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
933{
934 pgprot_t sect_prot = __pgprot(PUD_TYPE_SECT |
935 pgprot_val(mk_sect_prot(prot)));
936 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), sect_prot);
937
938 /* Only allow permission changes for now */
939 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
940 pud_val(new_pud)))
941 return 0;
942
943 BUG_ON(phys & ~PUD_MASK);
944 set_pud(pudp, new_pud);
945 return 1;
946}
947
948int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
949{
950 pgprot_t sect_prot = __pgprot(PMD_TYPE_SECT |
951 pgprot_val(mk_sect_prot(prot)));
952 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), sect_prot);
953
954 /* Only allow permission changes for now */
955 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
956 pmd_val(new_pmd)))
957 return 0;
958
959 BUG_ON(phys & ~PMD_MASK);
960 set_pmd(pmdp, new_pmd);
961 return 1;
962}
963
964int pud_clear_huge(pud_t *pudp)
965{
966 if (!pud_sect(READ_ONCE(*pudp)))
967 return 0;
968 pud_clear(pudp);
969 return 1;
970}
971
972int pmd_clear_huge(pmd_t *pmdp)
973{
974 if (!pmd_sect(READ_ONCE(*pmdp)))
975 return 0;
976 pmd_clear(pmdp);
977 return 1;
978}
979
980int pud_free_pmd_page(pud_t *pud)
981{
982 return pud_none(*pud);
983}
984
985int pmd_free_pte_page(pmd_t *pmd)
986{
987 return pmd_none(*pmd);
988}