~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/arm64/mm/mmu.c

Version: ~ [ linux-5.19-rc8 ] ~ [ linux-5.18.14 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.57 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.133 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.207 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.253 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.289 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.324 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Based on arch/arm/mm/mmu.c
  3  *
  4  * Copyright (C) 1995-2005 Russell King
  5  * Copyright (C) 2012 ARM Ltd.
  6  *
  7  * This program is free software; you can redistribute it and/or modify
  8  * it under the terms of the GNU General Public License version 2 as
  9  * published by the Free Software Foundation.
 10  *
 11  * This program is distributed in the hope that it will be useful,
 12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 14  * GNU General Public License for more details.
 15  *
 16  * You should have received a copy of the GNU General Public License
 17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 18  */
 19 
 20 #include <linux/cache.h>
 21 #include <linux/export.h>
 22 #include <linux/kernel.h>
 23 #include <linux/errno.h>
 24 #include <linux/init.h>
 25 #include <linux/libfdt.h>
 26 #include <linux/mman.h>
 27 #include <linux/nodemask.h>
 28 #include <linux/memblock.h>
 29 #include <linux/fs.h>
 30 #include <linux/io.h>
 31 
 32 #include <asm/barrier.h>
 33 #include <asm/cputype.h>
 34 #include <asm/fixmap.h>
 35 #include <asm/kasan.h>
 36 #include <asm/kernel-pgtable.h>
 37 #include <asm/sections.h>
 38 #include <asm/setup.h>
 39 #include <asm/sizes.h>
 40 #include <asm/tlb.h>
 41 #include <asm/memblock.h>
 42 #include <asm/mmu_context.h>
 43 #include <asm/ptdump.h>
 44 
 45 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
 46 
 47 u64 kimage_voffset __ro_after_init;
 48 EXPORT_SYMBOL(kimage_voffset);
 49 
 50 /*
 51  * Empty_zero_page is a special page that is used for zero-initialized data
 52  * and COW.
 53  */
 54 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
 55 EXPORT_SYMBOL(empty_zero_page);
 56 
 57 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
 58 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
 59 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
 60 
 61 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 62                               unsigned long size, pgprot_t vma_prot)
 63 {
 64         if (!pfn_valid(pfn))
 65                 return pgprot_noncached(vma_prot);
 66         else if (file->f_flags & O_SYNC)
 67                 return pgprot_writecombine(vma_prot);
 68         return vma_prot;
 69 }
 70 EXPORT_SYMBOL(phys_mem_access_prot);
 71 
 72 static phys_addr_t __init early_pgtable_alloc(void)
 73 {
 74         phys_addr_t phys;
 75         void *ptr;
 76 
 77         phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
 78 
 79         /*
 80          * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
 81          * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
 82          * any level of table.
 83          */
 84         ptr = pte_set_fixmap(phys);
 85 
 86         memset(ptr, 0, PAGE_SIZE);
 87 
 88         /*
 89          * Implicit barriers also ensure the zeroed page is visible to the page
 90          * table walker
 91          */
 92         pte_clear_fixmap();
 93 
 94         return phys;
 95 }
 96 
 97 static bool pgattr_change_is_safe(u64 old, u64 new)
 98 {
 99         /*
100          * The following mapping attributes may be updated in live
101          * kernel mappings without the need for break-before-make.
102          */
103         static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE;
104 
105         return old  == 0 || new  == 0 || ((old ^ new) & ~mask) == 0;
106 }
107 
108 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
109                                   unsigned long end, unsigned long pfn,
110                                   pgprot_t prot,
111                                   phys_addr_t (*pgtable_alloc)(void))
112 {
113         pte_t *pte;
114 
115         BUG_ON(pmd_sect(*pmd));
116         if (pmd_none(*pmd)) {
117                 phys_addr_t pte_phys;
118                 BUG_ON(!pgtable_alloc);
119                 pte_phys = pgtable_alloc();
120                 pte = pte_set_fixmap(pte_phys);
121                 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
122                 pte_clear_fixmap();
123         }
124         BUG_ON(pmd_bad(*pmd));
125 
126         pte = pte_set_fixmap_offset(pmd, addr);
127         do {
128                 pte_t old_pte = *pte;
129 
130                 set_pte(pte, pfn_pte(pfn, prot));
131                 pfn++;
132 
133                 /*
134                  * After the PTE entry has been populated once, we
135                  * only allow updates to the permission attributes.
136                  */
137                 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte)));
138 
139         } while (pte++, addr += PAGE_SIZE, addr != end);
140 
141         pte_clear_fixmap();
142 }
143 
144 static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
145                                   phys_addr_t phys, pgprot_t prot,
146                                   phys_addr_t (*pgtable_alloc)(void),
147                                   bool page_mappings_only)
148 {
149         pmd_t *pmd;
150         unsigned long next;
151 
152         /*
153          * Check for initial section mappings in the pgd/pud and remove them.
154          */
155         BUG_ON(pud_sect(*pud));
156         if (pud_none(*pud)) {
157                 phys_addr_t pmd_phys;
158                 BUG_ON(!pgtable_alloc);
159                 pmd_phys = pgtable_alloc();
160                 pmd = pmd_set_fixmap(pmd_phys);
161                 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
162                 pmd_clear_fixmap();
163         }
164         BUG_ON(pud_bad(*pud));
165 
166         pmd = pmd_set_fixmap_offset(pud, addr);
167         do {
168                 pmd_t old_pmd = *pmd;
169 
170                 next = pmd_addr_end(addr, end);
171 
172                 /* try section mapping first */
173                 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
174                       !page_mappings_only) {
175                         pmd_set_huge(pmd, phys, prot);
176 
177                         /*
178                          * After the PMD entry has been populated once, we
179                          * only allow updates to the permission attributes.
180                          */
181                         BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
182                                                       pmd_val(*pmd)));
183                 } else {
184                         alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
185                                        prot, pgtable_alloc);
186 
187                         BUG_ON(pmd_val(old_pmd) != 0 &&
188                                pmd_val(old_pmd) != pmd_val(*pmd));
189                 }
190                 phys += next - addr;
191         } while (pmd++, addr = next, addr != end);
192 
193         pmd_clear_fixmap();
194 }
195 
196 static inline bool use_1G_block(unsigned long addr, unsigned long next,
197                         unsigned long phys)
198 {
199         if (PAGE_SHIFT != 12)
200                 return false;
201 
202         if (((addr | next | phys) & ~PUD_MASK) != 0)
203                 return false;
204 
205         return true;
206 }
207 
208 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
209                                   phys_addr_t phys, pgprot_t prot,
210                                   phys_addr_t (*pgtable_alloc)(void),
211                                   bool page_mappings_only)
212 {
213         pud_t *pud;
214         unsigned long next;
215 
216         if (pgd_none(*pgd)) {
217                 phys_addr_t pud_phys;
218                 BUG_ON(!pgtable_alloc);
219                 pud_phys = pgtable_alloc();
220                 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
221         }
222         BUG_ON(pgd_bad(*pgd));
223 
224         pud = pud_set_fixmap_offset(pgd, addr);
225         do {
226                 pud_t old_pud = *pud;
227 
228                 next = pud_addr_end(addr, end);
229 
230                 /*
231                  * For 4K granule only, attempt to put down a 1GB block
232                  */
233                 if (use_1G_block(addr, next, phys) && !page_mappings_only) {
234                         pud_set_huge(pud, phys, prot);
235 
236                         /*
237                          * After the PUD entry has been populated once, we
238                          * only allow updates to the permission attributes.
239                          */
240                         BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
241                                                       pud_val(*pud)));
242                 } else {
243                         alloc_init_pmd(pud, addr, next, phys, prot,
244                                        pgtable_alloc, page_mappings_only);
245 
246                         BUG_ON(pud_val(old_pud) != 0 &&
247                                pud_val(old_pud) != pud_val(*pud));
248                 }
249                 phys += next - addr;
250         } while (pud++, addr = next, addr != end);
251 
252         pud_clear_fixmap();
253 }
254 
255 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
256                                  unsigned long virt, phys_addr_t size,
257                                  pgprot_t prot,
258                                  phys_addr_t (*pgtable_alloc)(void),
259                                  bool page_mappings_only)
260 {
261         unsigned long addr, length, end, next;
262         pgd_t *pgd = pgd_offset_raw(pgdir, virt);
263 
264         /*
265          * If the virtual and physical address don't have the same offset
266          * within a page, we cannot map the region as the caller expects.
267          */
268         if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
269                 return;
270 
271         phys &= PAGE_MASK;
272         addr = virt & PAGE_MASK;
273         length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
274 
275         end = addr + length;
276         do {
277                 next = pgd_addr_end(addr, end);
278                 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
279                                page_mappings_only);
280                 phys += next - addr;
281         } while (pgd++, addr = next, addr != end);
282 }
283 
284 static phys_addr_t pgd_pgtable_alloc(void)
285 {
286         void *ptr = (void *)__get_free_page(PGALLOC_GFP);
287         if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
288                 BUG();
289 
290         /* Ensure the zeroed page is visible to the page table walker */
291         dsb(ishst);
292         return __pa(ptr);
293 }
294 
295 /*
296  * This function can only be used to modify existing table entries,
297  * without allocating new levels of table. Note that this permits the
298  * creation of new section or page entries.
299  */
300 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
301                                   phys_addr_t size, pgprot_t prot)
302 {
303         if (virt < VMALLOC_START) {
304                 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
305                         &phys, virt);
306                 return;
307         }
308         __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, false);
309 }
310 
311 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
312                                unsigned long virt, phys_addr_t size,
313                                pgprot_t prot, bool page_mappings_only)
314 {
315         BUG_ON(mm == &init_mm);
316 
317         __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
318                              pgd_pgtable_alloc, page_mappings_only);
319 }
320 
321 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
322                                   phys_addr_t size, pgprot_t prot)
323 {
324         if (virt < VMALLOC_START) {
325                 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
326                         &phys, virt);
327                 return;
328         }
329 
330         __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
331                              NULL, debug_pagealloc_enabled());
332 }
333 
334 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
335 {
336         unsigned long kernel_start = __pa(_text);
337         unsigned long kernel_end = __pa(__init_begin);
338 
339         /*
340          * Take care not to create a writable alias for the
341          * read-only text and rodata sections of the kernel image.
342          */
343 
344         /* No overlap with the kernel text/rodata */
345         if (end < kernel_start || start >= kernel_end) {
346                 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
347                                      end - start, PAGE_KERNEL,
348                                      early_pgtable_alloc,
349                                      debug_pagealloc_enabled());
350                 return;
351         }
352 
353         /*
354          * This block overlaps the kernel text/rodata mappings.
355          * Map the portion(s) which don't overlap.
356          */
357         if (start < kernel_start)
358                 __create_pgd_mapping(pgd, start,
359                                      __phys_to_virt(start),
360                                      kernel_start - start, PAGE_KERNEL,
361                                      early_pgtable_alloc,
362                                      debug_pagealloc_enabled());
363         if (kernel_end < end)
364                 __create_pgd_mapping(pgd, kernel_end,
365                                      __phys_to_virt(kernel_end),
366                                      end - kernel_end, PAGE_KERNEL,
367                                      early_pgtable_alloc,
368                                      debug_pagealloc_enabled());
369 
370         /*
371          * Map the linear alias of the [_text, __init_begin) interval as
372          * read-only/non-executable. This makes the contents of the
373          * region accessible to subsystems such as hibernate, but
374          * protects it from inadvertent modification or execution.
375          */
376         __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
377                              kernel_end - kernel_start, PAGE_KERNEL_RO,
378                              early_pgtable_alloc, debug_pagealloc_enabled());
379 }
380 
381 static void __init map_mem(pgd_t *pgd)
382 {
383         struct memblock_region *reg;
384 
385         /* map all the memory banks */
386         for_each_memblock(memory, reg) {
387                 phys_addr_t start = reg->base;
388                 phys_addr_t end = start + reg->size;
389 
390                 if (start >= end)
391                         break;
392                 if (memblock_is_nomap(reg))
393                         continue;
394 
395                 __map_memblock(pgd, start, end);
396         }
397 }
398 
399 void mark_rodata_ro(void)
400 {
401         unsigned long section_size;
402 
403         section_size = (unsigned long)_etext - (unsigned long)_text;
404         create_mapping_late(__pa(_text), (unsigned long)_text,
405                             section_size, PAGE_KERNEL_ROX);
406         /*
407          * mark .rodata as read only. Use __init_begin rather than __end_rodata
408          * to cover NOTES and EXCEPTION_TABLE.
409          */
410         section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
411         create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
412                             section_size, PAGE_KERNEL_RO);
413 
414         /* flush the TLBs after updating live kernel mappings */
415         flush_tlb_all();
416 
417         debug_checkwx();
418 }
419 
420 static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
421                                       pgprot_t prot, struct vm_struct *vma)
422 {
423         phys_addr_t pa_start = __pa(va_start);
424         unsigned long size = va_end - va_start;
425 
426         BUG_ON(!PAGE_ALIGNED(pa_start));
427         BUG_ON(!PAGE_ALIGNED(size));
428 
429         __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
430                              early_pgtable_alloc, debug_pagealloc_enabled());
431 
432         vma->addr       = va_start;
433         vma->phys_addr  = pa_start;
434         vma->size       = size;
435         vma->flags      = VM_MAP;
436         vma->caller     = __builtin_return_address(0);
437 
438         vm_area_add_early(vma);
439 }
440 
441 /*
442  * Create fine-grained mappings for the kernel.
443  */
444 static void __init map_kernel(pgd_t *pgd)
445 {
446         static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
447 
448         map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
449         map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
450         map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
451                            &vmlinux_init);
452         map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
453 
454         if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
455                 /*
456                  * The fixmap falls in a separate pgd to the kernel, and doesn't
457                  * live in the carveout for the swapper_pg_dir. We can simply
458                  * re-use the existing dir for the fixmap.
459                  */
460                 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
461                         *pgd_offset_k(FIXADDR_START));
462         } else if (CONFIG_PGTABLE_LEVELS > 3) {
463                 /*
464                  * The fixmap shares its top level pgd entry with the kernel
465                  * mapping. This can really only occur when we are running
466                  * with 16k/4 levels, so we can simply reuse the pud level
467                  * entry instead.
468                  */
469                 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
470                 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
471                         __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
472                 pud_clear_fixmap();
473         } else {
474                 BUG();
475         }
476 
477         kasan_copy_shadow(pgd);
478 }
479 
480 /*
481  * paging_init() sets up the page tables, initialises the zone memory
482  * maps and sets up the zero page.
483  */
484 void __init paging_init(void)
485 {
486         phys_addr_t pgd_phys = early_pgtable_alloc();
487         pgd_t *pgd = pgd_set_fixmap(pgd_phys);
488 
489         map_kernel(pgd);
490         map_mem(pgd);
491 
492         /*
493          * We want to reuse the original swapper_pg_dir so we don't have to
494          * communicate the new address to non-coherent secondaries in
495          * secondary_entry, and so cpu_switch_mm can generate the address with
496          * adrp+add rather than a load from some global variable.
497          *
498          * To do this we need to go via a temporary pgd.
499          */
500         cpu_replace_ttbr1(__va(pgd_phys));
501         memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
502         cpu_replace_ttbr1(swapper_pg_dir);
503 
504         pgd_clear_fixmap();
505         memblock_free(pgd_phys, PAGE_SIZE);
506 
507         /*
508          * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
509          * allocated with it.
510          */
511         memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
512                       SWAPPER_DIR_SIZE - PAGE_SIZE);
513 }
514 
515 /*
516  * Check whether a kernel address is valid (derived from arch/x86/).
517  */
518 int kern_addr_valid(unsigned long addr)
519 {
520         pgd_t *pgd;
521         pud_t *pud;
522         pmd_t *pmd;
523         pte_t *pte;
524 
525         if ((((long)addr) >> VA_BITS) != -1UL)
526                 return 0;
527 
528         pgd = pgd_offset_k(addr);
529         if (pgd_none(*pgd))
530                 return 0;
531 
532         pud = pud_offset(pgd, addr);
533         if (pud_none(*pud))
534                 return 0;
535 
536         if (pud_sect(*pud))
537                 return pfn_valid(pud_pfn(*pud));
538 
539         pmd = pmd_offset(pud, addr);
540         if (pmd_none(*pmd))
541                 return 0;
542 
543         if (pmd_sect(*pmd))
544                 return pfn_valid(pmd_pfn(*pmd));
545 
546         pte = pte_offset_kernel(pmd, addr);
547         if (pte_none(*pte))
548                 return 0;
549 
550         return pfn_valid(pte_pfn(*pte));
551 }
552 #ifdef CONFIG_SPARSEMEM_VMEMMAP
553 #if !ARM64_SWAPPER_USES_SECTION_MAPS
554 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
555 {
556         return vmemmap_populate_basepages(start, end, node);
557 }
558 #else   /* !ARM64_SWAPPER_USES_SECTION_MAPS */
559 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
560 {
561         unsigned long addr = start;
562         unsigned long next;
563         pgd_t *pgd;
564         pud_t *pud;
565         pmd_t *pmd;
566 
567         do {
568                 next = pmd_addr_end(addr, end);
569 
570                 pgd = vmemmap_pgd_populate(addr, node);
571                 if (!pgd)
572                         return -ENOMEM;
573 
574                 pud = vmemmap_pud_populate(pgd, addr, node);
575                 if (!pud)
576                         return -ENOMEM;
577 
578                 pmd = pmd_offset(pud, addr);
579                 if (pmd_none(*pmd)) {
580                         void *p = NULL;
581 
582                         p = vmemmap_alloc_block_buf(PMD_SIZE, node);
583                         if (!p)
584                                 return -ENOMEM;
585 
586                         set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
587                 } else
588                         vmemmap_verify((pte_t *)pmd, node, addr, next);
589         } while (addr = next, addr != end);
590 
591         return 0;
592 }
593 #endif  /* CONFIG_ARM64_64K_PAGES */
594 void vmemmap_free(unsigned long start, unsigned long end)
595 {
596 }
597 #endif  /* CONFIG_SPARSEMEM_VMEMMAP */
598 
599 static inline pud_t * fixmap_pud(unsigned long addr)
600 {
601         pgd_t *pgd = pgd_offset_k(addr);
602 
603         BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
604 
605         return pud_offset_kimg(pgd, addr);
606 }
607 
608 static inline pmd_t * fixmap_pmd(unsigned long addr)
609 {
610         pud_t *pud = fixmap_pud(addr);
611 
612         BUG_ON(pud_none(*pud) || pud_bad(*pud));
613 
614         return pmd_offset_kimg(pud, addr);
615 }
616 
617 static inline pte_t * fixmap_pte(unsigned long addr)
618 {
619         return &bm_pte[pte_index(addr)];
620 }
621 
622 void __init early_fixmap_init(void)
623 {
624         pgd_t *pgd;
625         pud_t *pud;
626         pmd_t *pmd;
627         unsigned long addr = FIXADDR_START;
628 
629         pgd = pgd_offset_k(addr);
630         if (CONFIG_PGTABLE_LEVELS > 3 &&
631             !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
632                 /*
633                  * We only end up here if the kernel mapping and the fixmap
634                  * share the top level pgd entry, which should only happen on
635                  * 16k/4 levels configurations.
636                  */
637                 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
638                 pud = pud_offset_kimg(pgd, addr);
639         } else {
640                 pgd_populate(&init_mm, pgd, bm_pud);
641                 pud = fixmap_pud(addr);
642         }
643         pud_populate(&init_mm, pud, bm_pmd);
644         pmd = fixmap_pmd(addr);
645         pmd_populate_kernel(&init_mm, pmd, bm_pte);
646 
647         /*
648          * The boot-ioremap range spans multiple pmds, for which
649          * we are not prepared:
650          */
651         BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
652                      != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
653 
654         if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
655              || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
656                 WARN_ON(1);
657                 pr_warn("pmd %p != %p, %p\n",
658                         pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
659                         fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
660                 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
661                         fix_to_virt(FIX_BTMAP_BEGIN));
662                 pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
663                         fix_to_virt(FIX_BTMAP_END));
664 
665                 pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
666                 pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
667         }
668 }
669 
670 void __set_fixmap(enum fixed_addresses idx,
671                                phys_addr_t phys, pgprot_t flags)
672 {
673         unsigned long addr = __fix_to_virt(idx);
674         pte_t *pte;
675 
676         BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
677 
678         pte = fixmap_pte(addr);
679 
680         if (pgprot_val(flags)) {
681                 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
682         } else {
683                 pte_clear(&init_mm, addr, pte);
684                 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
685         }
686 }
687 
688 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
689 {
690         const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
691         int offset;
692         void *dt_virt;
693 
694         /*
695          * Check whether the physical FDT address is set and meets the minimum
696          * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
697          * at least 8 bytes so that we can always access the magic and size
698          * fields of the FDT header after mapping the first chunk, double check
699          * here if that is indeed the case.
700          */
701         BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
702         if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
703                 return NULL;
704 
705         /*
706          * Make sure that the FDT region can be mapped without the need to
707          * allocate additional translation table pages, so that it is safe
708          * to call create_mapping_noalloc() this early.
709          *
710          * On 64k pages, the FDT will be mapped using PTEs, so we need to
711          * be in the same PMD as the rest of the fixmap.
712          * On 4k pages, we'll use section mappings for the FDT so we only
713          * have to be in the same PUD.
714          */
715         BUILD_BUG_ON(dt_virt_base % SZ_2M);
716 
717         BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
718                      __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
719 
720         offset = dt_phys % SWAPPER_BLOCK_SIZE;
721         dt_virt = (void *)dt_virt_base + offset;
722 
723         /* map the first chunk so we can read the size from the header */
724         create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
725                         dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
726 
727         if (fdt_magic(dt_virt) != FDT_MAGIC)
728                 return NULL;
729 
730         *size = fdt_totalsize(dt_virt);
731         if (*size > MAX_FDT_SIZE)
732                 return NULL;
733 
734         if (offset + *size > SWAPPER_BLOCK_SIZE)
735                 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
736                                round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
737 
738         return dt_virt;
739 }
740 
741 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
742 {
743         void *dt_virt;
744         int size;
745 
746         dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
747         if (!dt_virt)
748                 return NULL;
749 
750         memblock_reserve(dt_phys, size);
751         return dt_virt;
752 }
753 
754 int __init arch_ioremap_pud_supported(void)
755 {
756         /* only 4k granule supports level 1 block mappings */
757         return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
758 }
759 
760 int __init arch_ioremap_pmd_supported(void)
761 {
762         return 1;
763 }
764 
765 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
766 {
767         BUG_ON(phys & ~PUD_MASK);
768         set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
769         return 1;
770 }
771 
772 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
773 {
774         BUG_ON(phys & ~PMD_MASK);
775         set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
776         return 1;
777 }
778 
779 int pud_clear_huge(pud_t *pud)
780 {
781         if (!pud_sect(*pud))
782                 return 0;
783         pud_clear(pud);
784         return 1;
785 }
786 
787 int pmd_clear_huge(pmd_t *pmd)
788 {
789         if (!pmd_sect(*pmd))
790                 return 0;
791         pmd_clear(pmd);
792         return 1;
793 }
794 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp