mmu.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777
  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. #include <linux/cache.h>
  20. #include <linux/export.h>
  21. #include <linux/kernel.h>
  22. #include <linux/errno.h>
  23. #include <linux/init.h>
  24. #include <linux/ioport.h>
  25. #include <linux/kexec.h>
  26. #include <linux/libfdt.h>
  27. #include <linux/mman.h>
  28. #include <linux/nodemask.h>
  29. #include <linux/memblock.h>
  30. #include <linux/fs.h>
  31. #include <linux/io.h>
  32. #include <linux/slab.h>
  33. #include <linux/stop_machine.h>
  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. u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
  46. u64 kimage_voffset __ro_after_init;
  47. EXPORT_SYMBOL(kimage_voffset);
  48. /*
  49. * Empty_zero_page is a special page that is used for zero-initialized data
  50. * and COW.
  51. */
  52. unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
  53. EXPORT_SYMBOL(empty_zero_page);
  54. static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
  55. static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
  56. static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
  57. pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  58. unsigned long size, pgprot_t vma_prot)
  59. {
  60. if (!pfn_valid(pfn))
  61. return pgprot_noncached(vma_prot);
  62. else if (file->f_flags & O_SYNC)
  63. return pgprot_writecombine(vma_prot);
  64. return vma_prot;
  65. }
  66. EXPORT_SYMBOL(phys_mem_access_prot);
  67. static phys_addr_t __init early_pgtable_alloc(void)
  68. {
  69. phys_addr_t phys;
  70. void *ptr;
  71. phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
  72. /*
  73. * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
  74. * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
  75. * any level of table.
  76. */
  77. ptr = pte_set_fixmap(phys);
  78. memset(ptr, 0, PAGE_SIZE);
  79. /*
  80. * Implicit barriers also ensure the zeroed page is visible to the page
  81. * table walker
  82. */
  83. pte_clear_fixmap();
  84. return phys;
  85. }
  86. static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
  87. unsigned long end, unsigned long pfn,
  88. pgprot_t prot,
  89. phys_addr_t (*pgtable_alloc)(void))
  90. {
  91. pte_t *pte;
  92. BUG_ON(pmd_sect(*pmd));
  93. if (pmd_none(*pmd)) {
  94. phys_addr_t pte_phys;
  95. BUG_ON(!pgtable_alloc);
  96. pte_phys = pgtable_alloc();
  97. pte = pte_set_fixmap(pte_phys);
  98. __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
  99. pte_clear_fixmap();
  100. }
  101. BUG_ON(pmd_bad(*pmd));
  102. pte = pte_set_fixmap_offset(pmd, addr);
  103. do {
  104. set_pte(pte, pfn_pte(pfn, prot));
  105. pfn++;
  106. } while (pte++, addr += PAGE_SIZE, addr != end);
  107. pte_clear_fixmap();
  108. }
  109. static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
  110. phys_addr_t phys, pgprot_t prot,
  111. phys_addr_t (*pgtable_alloc)(void),
  112. bool allow_block_mappings)
  113. {
  114. pmd_t *pmd;
  115. unsigned long next;
  116. /*
  117. * Check for initial section mappings in the pgd/pud and remove them.
  118. */
  119. BUG_ON(pud_sect(*pud));
  120. if (pud_none(*pud)) {
  121. phys_addr_t pmd_phys;
  122. BUG_ON(!pgtable_alloc);
  123. pmd_phys = pgtable_alloc();
  124. pmd = pmd_set_fixmap(pmd_phys);
  125. __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
  126. pmd_clear_fixmap();
  127. }
  128. BUG_ON(pud_bad(*pud));
  129. pmd = pmd_set_fixmap_offset(pud, addr);
  130. do {
  131. next = pmd_addr_end(addr, end);
  132. /* try section mapping first */
  133. if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
  134. allow_block_mappings) {
  135. pmd_t old_pmd =*pmd;
  136. pmd_set_huge(pmd, phys, prot);
  137. /*
  138. * Check for previous table entries created during
  139. * boot (__create_page_tables) and flush them.
  140. */
  141. if (!pmd_none(old_pmd)) {
  142. flush_tlb_all();
  143. if (pmd_table(old_pmd)) {
  144. phys_addr_t table = pmd_page_paddr(old_pmd);
  145. if (!WARN_ON_ONCE(slab_is_available()))
  146. memblock_free(table, PAGE_SIZE);
  147. }
  148. }
  149. } else {
  150. alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
  151. prot, pgtable_alloc);
  152. }
  153. phys += next - addr;
  154. } while (pmd++, addr = next, addr != end);
  155. pmd_clear_fixmap();
  156. }
  157. static inline bool use_1G_block(unsigned long addr, unsigned long next,
  158. unsigned long phys)
  159. {
  160. if (PAGE_SHIFT != 12)
  161. return false;
  162. if (((addr | next | phys) & ~PUD_MASK) != 0)
  163. return false;
  164. return true;
  165. }
  166. static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
  167. phys_addr_t phys, pgprot_t prot,
  168. phys_addr_t (*pgtable_alloc)(void),
  169. bool allow_block_mappings)
  170. {
  171. pud_t *pud;
  172. unsigned long next;
  173. if (pgd_none(*pgd)) {
  174. phys_addr_t pud_phys;
  175. BUG_ON(!pgtable_alloc);
  176. pud_phys = pgtable_alloc();
  177. __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
  178. }
  179. BUG_ON(pgd_bad(*pgd));
  180. pud = pud_set_fixmap_offset(pgd, addr);
  181. do {
  182. next = pud_addr_end(addr, end);
  183. /*
  184. * For 4K granule only, attempt to put down a 1GB block
  185. */
  186. if (use_1G_block(addr, next, phys) && allow_block_mappings) {
  187. pud_t old_pud = *pud;
  188. pud_set_huge(pud, phys, prot);
  189. /*
  190. * If we have an old value for a pud, it will
  191. * be pointing to a pmd table that we no longer
  192. * need (from swapper_pg_dir).
  193. *
  194. * Look up the old pmd table and free it.
  195. */
  196. if (!pud_none(old_pud)) {
  197. flush_tlb_all();
  198. if (pud_table(old_pud)) {
  199. phys_addr_t table = pud_page_paddr(old_pud);
  200. if (!WARN_ON_ONCE(slab_is_available()))
  201. memblock_free(table, PAGE_SIZE);
  202. }
  203. }
  204. } else {
  205. alloc_init_pmd(pud, addr, next, phys, prot,
  206. pgtable_alloc, allow_block_mappings);
  207. }
  208. phys += next - addr;
  209. } while (pud++, addr = next, addr != end);
  210. pud_clear_fixmap();
  211. }
  212. static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
  213. unsigned long virt, phys_addr_t size,
  214. pgprot_t prot,
  215. phys_addr_t (*pgtable_alloc)(void),
  216. bool allow_block_mappings)
  217. {
  218. unsigned long addr, length, end, next;
  219. pgd_t *pgd = pgd_offset_raw(pgdir, virt);
  220. /*
  221. * If the virtual and physical address don't have the same offset
  222. * within a page, we cannot map the region as the caller expects.
  223. */
  224. if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
  225. return;
  226. phys &= PAGE_MASK;
  227. addr = virt & PAGE_MASK;
  228. length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
  229. end = addr + length;
  230. do {
  231. next = pgd_addr_end(addr, end);
  232. alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
  233. allow_block_mappings);
  234. phys += next - addr;
  235. } while (pgd++, addr = next, addr != end);
  236. }
  237. static phys_addr_t pgd_pgtable_alloc(void)
  238. {
  239. void *ptr = (void *)__get_free_page(PGALLOC_GFP);
  240. if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
  241. BUG();
  242. /* Ensure the zeroed page is visible to the page table walker */
  243. dsb(ishst);
  244. return __pa(ptr);
  245. }
  246. /*
  247. * This function can only be used to modify existing table entries,
  248. * without allocating new levels of table. Note that this permits the
  249. * creation of new section or page entries.
  250. */
  251. static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
  252. phys_addr_t size, pgprot_t prot)
  253. {
  254. if (virt < VMALLOC_START) {
  255. pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
  256. &phys, virt);
  257. return;
  258. }
  259. __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, true);
  260. }
  261. void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
  262. unsigned long virt, phys_addr_t size,
  263. pgprot_t prot, bool allow_block_mappings)
  264. {
  265. BUG_ON(mm == &init_mm);
  266. __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
  267. pgd_pgtable_alloc, allow_block_mappings);
  268. }
  269. static void create_mapping_late(phys_addr_t phys, unsigned long virt,
  270. phys_addr_t size, pgprot_t prot)
  271. {
  272. if (virt < VMALLOC_START) {
  273. pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
  274. &phys, virt);
  275. return;
  276. }
  277. __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
  278. NULL, !debug_pagealloc_enabled());
  279. }
  280. static void __init __map_memblock(pgd_t *pgd, phys_addr_t start,
  281. phys_addr_t end, pgprot_t prot,
  282. bool allow_block_mappings)
  283. {
  284. __create_pgd_mapping(pgd, start, __phys_to_virt(start), end - start,
  285. prot, early_pgtable_alloc, allow_block_mappings);
  286. }
  287. static void __init map_mem(pgd_t *pgd)
  288. {
  289. unsigned long kernel_start = __pa(_text);
  290. unsigned long kernel_end = __pa(__init_begin);
  291. struct memblock_region *reg;
  292. /*
  293. * Take care not to create a writable alias for the
  294. * read-only text and rodata sections of the kernel image.
  295. * So temporarily mark them as NOMAP to skip mappings in
  296. * the following for-loop
  297. */
  298. memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
  299. #ifdef CONFIG_KEXEC_CORE
  300. if (crashk_res.end)
  301. memblock_mark_nomap(crashk_res.start,
  302. resource_size(&crashk_res));
  303. #endif
  304. /* map all the memory banks */
  305. for_each_memblock(memory, reg) {
  306. phys_addr_t start = reg->base;
  307. phys_addr_t end = start + reg->size;
  308. if (start >= end)
  309. break;
  310. if (memblock_is_nomap(reg))
  311. continue;
  312. __map_memblock(pgd, start, end,
  313. PAGE_KERNEL, !debug_pagealloc_enabled());
  314. }
  315. /*
  316. * Map the linear alias of the [_text, __init_begin) interval as
  317. * read-only/non-executable. This makes the contents of the
  318. * region accessible to subsystems such as hibernate, but
  319. * protects it from inadvertent modification or execution.
  320. */
  321. __map_memblock(pgd, kernel_start, kernel_end,
  322. PAGE_KERNEL_RO, !debug_pagealloc_enabled());
  323. memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
  324. #ifdef CONFIG_KEXEC_CORE
  325. /*
  326. * Use page-level mappings here so that we can shrink the region
  327. * in page granularity and put back unused memory to buddy system
  328. * through /sys/kernel/kexec_crash_size interface.
  329. */
  330. if (crashk_res.end) {
  331. __map_memblock(pgd, crashk_res.start, crashk_res.end + 1,
  332. PAGE_KERNEL, false);
  333. memblock_clear_nomap(crashk_res.start,
  334. resource_size(&crashk_res));
  335. }
  336. #endif
  337. }
  338. void mark_rodata_ro(void)
  339. {
  340. unsigned long section_size;
  341. section_size = (unsigned long)_etext - (unsigned long)_text;
  342. create_mapping_late(__pa(_text), (unsigned long)_text,
  343. section_size, PAGE_KERNEL_ROX);
  344. /*
  345. * mark .rodata as read only. Use __init_begin rather than __end_rodata
  346. * to cover NOTES and EXCEPTION_TABLE.
  347. */
  348. section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
  349. create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
  350. section_size, PAGE_KERNEL_RO);
  351. }
  352. static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
  353. pgprot_t prot, struct vm_struct *vma)
  354. {
  355. phys_addr_t pa_start = __pa(va_start);
  356. unsigned long size = va_end - va_start;
  357. BUG_ON(!PAGE_ALIGNED(pa_start));
  358. BUG_ON(!PAGE_ALIGNED(size));
  359. __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
  360. early_pgtable_alloc, !debug_pagealloc_enabled());
  361. vma->addr = va_start;
  362. vma->phys_addr = pa_start;
  363. vma->size = size;
  364. vma->flags = VM_MAP;
  365. vma->caller = __builtin_return_address(0);
  366. vm_area_add_early(vma);
  367. }
  368. /*
  369. * Create fine-grained mappings for the kernel.
  370. */
  371. static void __init map_kernel(pgd_t *pgd)
  372. {
  373. static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
  374. map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
  375. map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
  376. map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
  377. &vmlinux_init);
  378. map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
  379. if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
  380. /*
  381. * The fixmap falls in a separate pgd to the kernel, and doesn't
  382. * live in the carveout for the swapper_pg_dir. We can simply
  383. * re-use the existing dir for the fixmap.
  384. */
  385. set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
  386. *pgd_offset_k(FIXADDR_START));
  387. } else if (CONFIG_PGTABLE_LEVELS > 3) {
  388. /*
  389. * The fixmap shares its top level pgd entry with the kernel
  390. * mapping. This can really only occur when we are running
  391. * with 16k/4 levels, so we can simply reuse the pud level
  392. * entry instead.
  393. */
  394. BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
  395. set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
  396. __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
  397. pud_clear_fixmap();
  398. } else {
  399. BUG();
  400. }
  401. kasan_copy_shadow(pgd);
  402. }
  403. /*
  404. * paging_init() sets up the page tables, initialises the zone memory
  405. * maps and sets up the zero page.
  406. */
  407. void __init paging_init(void)
  408. {
  409. phys_addr_t pgd_phys = early_pgtable_alloc();
  410. pgd_t *pgd = pgd_set_fixmap(pgd_phys);
  411. map_kernel(pgd);
  412. map_mem(pgd);
  413. /*
  414. * We want to reuse the original swapper_pg_dir so we don't have to
  415. * communicate the new address to non-coherent secondaries in
  416. * secondary_entry, and so cpu_switch_mm can generate the address with
  417. * adrp+add rather than a load from some global variable.
  418. *
  419. * To do this we need to go via a temporary pgd.
  420. */
  421. cpu_replace_ttbr1(__va(pgd_phys));
  422. memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
  423. cpu_replace_ttbr1(swapper_pg_dir);
  424. pgd_clear_fixmap();
  425. memblock_free(pgd_phys, PAGE_SIZE);
  426. /*
  427. * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
  428. * allocated with it.
  429. */
  430. memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
  431. SWAPPER_DIR_SIZE - PAGE_SIZE);
  432. }
  433. /*
  434. * Check whether a kernel address is valid (derived from arch/x86/).
  435. */
  436. int kern_addr_valid(unsigned long addr)
  437. {
  438. pgd_t *pgd;
  439. pud_t *pud;
  440. pmd_t *pmd;
  441. pte_t *pte;
  442. if ((((long)addr) >> VA_BITS) != -1UL)
  443. return 0;
  444. pgd = pgd_offset_k(addr);
  445. if (pgd_none(*pgd))
  446. return 0;
  447. pud = pud_offset(pgd, addr);
  448. if (pud_none(*pud))
  449. return 0;
  450. if (pud_sect(*pud))
  451. return pfn_valid(pud_pfn(*pud));
  452. pmd = pmd_offset(pud, addr);
  453. if (pmd_none(*pmd))
  454. return 0;
  455. if (pmd_sect(*pmd))
  456. return pfn_valid(pmd_pfn(*pmd));
  457. pte = pte_offset_kernel(pmd, addr);
  458. if (pte_none(*pte))
  459. return 0;
  460. return pfn_valid(pte_pfn(*pte));
  461. }
  462. #ifdef CONFIG_SPARSEMEM_VMEMMAP
  463. #if !ARM64_SWAPPER_USES_SECTION_MAPS
  464. int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
  465. {
  466. return vmemmap_populate_basepages(start, end, node);
  467. }
  468. #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
  469. int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
  470. {
  471. unsigned long addr = start;
  472. unsigned long next;
  473. pgd_t *pgd;
  474. pud_t *pud;
  475. pmd_t *pmd;
  476. do {
  477. next = pmd_addr_end(addr, end);
  478. pgd = vmemmap_pgd_populate(addr, node);
  479. if (!pgd)
  480. return -ENOMEM;
  481. pud = vmemmap_pud_populate(pgd, addr, node);
  482. if (!pud)
  483. return -ENOMEM;
  484. pmd = pmd_offset(pud, addr);
  485. if (pmd_none(*pmd)) {
  486. void *p = NULL;
  487. p = vmemmap_alloc_block_buf(PMD_SIZE, node);
  488. if (!p)
  489. return -ENOMEM;
  490. set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
  491. } else
  492. vmemmap_verify((pte_t *)pmd, node, addr, next);
  493. } while (addr = next, addr != end);
  494. return 0;
  495. }
  496. #endif /* CONFIG_ARM64_64K_PAGES */
  497. void vmemmap_free(unsigned long start, unsigned long end)
  498. {
  499. }
  500. #endif /* CONFIG_SPARSEMEM_VMEMMAP */
  501. static inline pud_t * fixmap_pud(unsigned long addr)
  502. {
  503. pgd_t *pgd = pgd_offset_k(addr);
  504. BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
  505. return pud_offset_kimg(pgd, addr);
  506. }
  507. static inline pmd_t * fixmap_pmd(unsigned long addr)
  508. {
  509. pud_t *pud = fixmap_pud(addr);
  510. BUG_ON(pud_none(*pud) || pud_bad(*pud));
  511. return pmd_offset_kimg(pud, addr);
  512. }
  513. static inline pte_t * fixmap_pte(unsigned long addr)
  514. {
  515. return &bm_pte[pte_index(addr)];
  516. }
  517. void __init early_fixmap_init(void)
  518. {
  519. pgd_t *pgd;
  520. pud_t *pud;
  521. pmd_t *pmd;
  522. unsigned long addr = FIXADDR_START;
  523. pgd = pgd_offset_k(addr);
  524. if (CONFIG_PGTABLE_LEVELS > 3 &&
  525. !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
  526. /*
  527. * We only end up here if the kernel mapping and the fixmap
  528. * share the top level pgd entry, which should only happen on
  529. * 16k/4 levels configurations.
  530. */
  531. BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
  532. pud = pud_offset_kimg(pgd, addr);
  533. } else {
  534. pgd_populate(&init_mm, pgd, bm_pud);
  535. pud = fixmap_pud(addr);
  536. }
  537. pud_populate(&init_mm, pud, bm_pmd);
  538. pmd = fixmap_pmd(addr);
  539. pmd_populate_kernel(&init_mm, pmd, bm_pte);
  540. /*
  541. * The boot-ioremap range spans multiple pmds, for which
  542. * we are not prepared:
  543. */
  544. BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
  545. != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
  546. if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
  547. || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
  548. WARN_ON(1);
  549. pr_warn("pmd %p != %p, %p\n",
  550. pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
  551. fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
  552. pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
  553. fix_to_virt(FIX_BTMAP_BEGIN));
  554. pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
  555. fix_to_virt(FIX_BTMAP_END));
  556. pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
  557. pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
  558. }
  559. }
  560. void __set_fixmap(enum fixed_addresses idx,
  561. phys_addr_t phys, pgprot_t flags)
  562. {
  563. unsigned long addr = __fix_to_virt(idx);
  564. pte_t *pte;
  565. BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
  566. pte = fixmap_pte(addr);
  567. if (pgprot_val(flags)) {
  568. set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
  569. } else {
  570. pte_clear(&init_mm, addr, pte);
  571. flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
  572. }
  573. }
  574. void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
  575. {
  576. const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
  577. int offset;
  578. void *dt_virt;
  579. /*
  580. * Check whether the physical FDT address is set and meets the minimum
  581. * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
  582. * at least 8 bytes so that we can always access the magic and size
  583. * fields of the FDT header after mapping the first chunk, double check
  584. * here if that is indeed the case.
  585. */
  586. BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
  587. if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
  588. return NULL;
  589. /*
  590. * Make sure that the FDT region can be mapped without the need to
  591. * allocate additional translation table pages, so that it is safe
  592. * to call create_mapping_noalloc() this early.
  593. *
  594. * On 64k pages, the FDT will be mapped using PTEs, so we need to
  595. * be in the same PMD as the rest of the fixmap.
  596. * On 4k pages, we'll use section mappings for the FDT so we only
  597. * have to be in the same PUD.
  598. */
  599. BUILD_BUG_ON(dt_virt_base % SZ_2M);
  600. BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
  601. __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
  602. offset = dt_phys % SWAPPER_BLOCK_SIZE;
  603. dt_virt = (void *)dt_virt_base + offset;
  604. /* map the first chunk so we can read the size from the header */
  605. create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
  606. dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
  607. if (fdt_magic(dt_virt) != FDT_MAGIC)
  608. return NULL;
  609. *size = fdt_totalsize(dt_virt);
  610. if (*size > MAX_FDT_SIZE)
  611. return NULL;
  612. if (offset + *size > SWAPPER_BLOCK_SIZE)
  613. create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
  614. round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
  615. return dt_virt;
  616. }
  617. void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
  618. {
  619. void *dt_virt;
  620. int size;
  621. dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
  622. if (!dt_virt)
  623. return NULL;
  624. memblock_reserve(dt_phys, size);
  625. return dt_virt;
  626. }
  627. int __init arch_ioremap_pud_supported(void)
  628. {
  629. /* only 4k granule supports level 1 block mappings */
  630. return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
  631. }
  632. int __init arch_ioremap_pmd_supported(void)
  633. {
  634. return 1;
  635. }
  636. int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
  637. {
  638. BUG_ON(phys & ~PUD_MASK);
  639. set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
  640. return 1;
  641. }
  642. int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
  643. {
  644. BUG_ON(phys & ~PMD_MASK);
  645. set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
  646. return 1;
  647. }
  648. int pud_clear_huge(pud_t *pud)
  649. {
  650. if (!pud_sect(*pud))
  651. return 0;
  652. pud_clear(pud);
  653. return 1;
  654. }
  655. int pmd_clear_huge(pmd_t *pmd)
  656. {
  657. if (!pmd_sect(*pmd))
  658. return 0;
  659. pmd_clear(pmd);
  660. return 1;
  661. }