mmu.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802
  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/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/mm.h>
  31. #include <asm/barrier.h>
  32. #include <asm/cputype.h>
  33. #include <asm/fixmap.h>
  34. #include <asm/kasan.h>
  35. #include <asm/kernel-pgtable.h>
  36. #include <asm/sections.h>
  37. #include <asm/setup.h>
  38. #include <asm/sizes.h>
  39. #include <asm/tlb.h>
  40. #include <asm/memblock.h>
  41. #include <asm/mmu_context.h>
  42. #include <asm/ptdump.h>
  43. u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
  44. u64 kimage_voffset __ro_after_init;
  45. EXPORT_SYMBOL(kimage_voffset);
  46. /*
  47. * Empty_zero_page is a special page that is used for zero-initialized data
  48. * and COW.
  49. */
  50. unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
  51. EXPORT_SYMBOL(empty_zero_page);
  52. static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
  53. static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
  54. static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
  55. pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  56. unsigned long size, pgprot_t vma_prot)
  57. {
  58. if (!pfn_valid(pfn))
  59. return pgprot_noncached(vma_prot);
  60. else if (file->f_flags & O_SYNC)
  61. return pgprot_writecombine(vma_prot);
  62. return vma_prot;
  63. }
  64. EXPORT_SYMBOL(phys_mem_access_prot);
  65. static phys_addr_t __init early_pgtable_alloc(void)
  66. {
  67. phys_addr_t phys;
  68. void *ptr;
  69. phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
  70. /*
  71. * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
  72. * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
  73. * any level of table.
  74. */
  75. ptr = pte_set_fixmap(phys);
  76. memset(ptr, 0, PAGE_SIZE);
  77. /*
  78. * Implicit barriers also ensure the zeroed page is visible to the page
  79. * table walker
  80. */
  81. pte_clear_fixmap();
  82. return phys;
  83. }
  84. static bool pgattr_change_is_safe(u64 old, u64 new)
  85. {
  86. /*
  87. * The following mapping attributes may be updated in live
  88. * kernel mappings without the need for break-before-make.
  89. */
  90. static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE;
  91. return old == 0 || new == 0 || ((old ^ new) & ~mask) == 0;
  92. }
  93. static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
  94. unsigned long end, unsigned long pfn,
  95. pgprot_t prot,
  96. phys_addr_t (*pgtable_alloc)(void))
  97. {
  98. pte_t *pte;
  99. BUG_ON(pmd_sect(*pmd));
  100. if (pmd_none(*pmd)) {
  101. phys_addr_t pte_phys;
  102. BUG_ON(!pgtable_alloc);
  103. pte_phys = pgtable_alloc();
  104. pte = pte_set_fixmap(pte_phys);
  105. __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
  106. pte_clear_fixmap();
  107. }
  108. BUG_ON(pmd_bad(*pmd));
  109. pte = pte_set_fixmap_offset(pmd, addr);
  110. do {
  111. pte_t old_pte = *pte;
  112. set_pte(pte, pfn_pte(pfn, prot));
  113. pfn++;
  114. /*
  115. * After the PTE entry has been populated once, we
  116. * only allow updates to the permission attributes.
  117. */
  118. BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte)));
  119. } while (pte++, addr += PAGE_SIZE, addr != end);
  120. pte_clear_fixmap();
  121. }
  122. static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
  123. phys_addr_t phys, pgprot_t prot,
  124. phys_addr_t (*pgtable_alloc)(void),
  125. bool page_mappings_only)
  126. {
  127. pmd_t *pmd;
  128. unsigned long next;
  129. /*
  130. * Check for initial section mappings in the pgd/pud and remove them.
  131. */
  132. BUG_ON(pud_sect(*pud));
  133. if (pud_none(*pud)) {
  134. phys_addr_t pmd_phys;
  135. BUG_ON(!pgtable_alloc);
  136. pmd_phys = pgtable_alloc();
  137. pmd = pmd_set_fixmap(pmd_phys);
  138. __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
  139. pmd_clear_fixmap();
  140. }
  141. BUG_ON(pud_bad(*pud));
  142. pmd = pmd_set_fixmap_offset(pud, addr);
  143. do {
  144. pmd_t old_pmd = *pmd;
  145. next = pmd_addr_end(addr, end);
  146. /* try section mapping first */
  147. if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
  148. !page_mappings_only) {
  149. pmd_set_huge(pmd, phys, prot);
  150. /*
  151. * After the PMD entry has been populated once, we
  152. * only allow updates to the permission attributes.
  153. */
  154. BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
  155. pmd_val(*pmd)));
  156. } else {
  157. alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
  158. prot, pgtable_alloc);
  159. BUG_ON(pmd_val(old_pmd) != 0 &&
  160. pmd_val(old_pmd) != pmd_val(*pmd));
  161. }
  162. phys += next - addr;
  163. } while (pmd++, addr = next, addr != end);
  164. pmd_clear_fixmap();
  165. }
  166. static inline bool use_1G_block(unsigned long addr, unsigned long next,
  167. unsigned long phys)
  168. {
  169. if (PAGE_SHIFT != 12)
  170. return false;
  171. if (((addr | next | phys) & ~PUD_MASK) != 0)
  172. return false;
  173. return true;
  174. }
  175. static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
  176. phys_addr_t phys, pgprot_t prot,
  177. phys_addr_t (*pgtable_alloc)(void),
  178. bool page_mappings_only)
  179. {
  180. pud_t *pud;
  181. unsigned long next;
  182. if (pgd_none(*pgd)) {
  183. phys_addr_t pud_phys;
  184. BUG_ON(!pgtable_alloc);
  185. pud_phys = pgtable_alloc();
  186. __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
  187. }
  188. BUG_ON(pgd_bad(*pgd));
  189. pud = pud_set_fixmap_offset(pgd, addr);
  190. do {
  191. pud_t old_pud = *pud;
  192. next = pud_addr_end(addr, end);
  193. /*
  194. * For 4K granule only, attempt to put down a 1GB block
  195. */
  196. if (use_1G_block(addr, next, phys) && !page_mappings_only) {
  197. pud_set_huge(pud, phys, prot);
  198. /*
  199. * After the PUD entry has been populated once, we
  200. * only allow updates to the permission attributes.
  201. */
  202. BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
  203. pud_val(*pud)));
  204. } else {
  205. alloc_init_pmd(pud, addr, next, phys, prot,
  206. pgtable_alloc, page_mappings_only);
  207. BUG_ON(pud_val(old_pud) != 0 &&
  208. pud_val(old_pud) != pud_val(*pud));
  209. }
  210. phys += next - addr;
  211. } while (pud++, addr = next, addr != end);
  212. pud_clear_fixmap();
  213. }
  214. static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
  215. unsigned long virt, phys_addr_t size,
  216. pgprot_t prot,
  217. phys_addr_t (*pgtable_alloc)(void),
  218. bool page_mappings_only)
  219. {
  220. unsigned long addr, length, end, next;
  221. pgd_t *pgd = pgd_offset_raw(pgdir, virt);
  222. /*
  223. * If the virtual and physical address don't have the same offset
  224. * within a page, we cannot map the region as the caller expects.
  225. */
  226. if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
  227. return;
  228. phys &= PAGE_MASK;
  229. addr = virt & PAGE_MASK;
  230. length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
  231. end = addr + length;
  232. do {
  233. next = pgd_addr_end(addr, end);
  234. alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
  235. page_mappings_only);
  236. phys += next - addr;
  237. } while (pgd++, addr = next, addr != end);
  238. }
  239. static phys_addr_t pgd_pgtable_alloc(void)
  240. {
  241. void *ptr = (void *)__get_free_page(PGALLOC_GFP);
  242. if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
  243. BUG();
  244. /* Ensure the zeroed page is visible to the page table walker */
  245. dsb(ishst);
  246. return __pa(ptr);
  247. }
  248. /*
  249. * This function can only be used to modify existing table entries,
  250. * without allocating new levels of table. Note that this permits the
  251. * creation of new section or page entries.
  252. */
  253. static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
  254. phys_addr_t size, pgprot_t prot)
  255. {
  256. if (virt < VMALLOC_START) {
  257. pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
  258. &phys, virt);
  259. return;
  260. }
  261. __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, false);
  262. }
  263. void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
  264. unsigned long virt, phys_addr_t size,
  265. pgprot_t prot, bool page_mappings_only)
  266. {
  267. BUG_ON(mm == &init_mm);
  268. __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
  269. pgd_pgtable_alloc, page_mappings_only);
  270. }
  271. static void create_mapping_late(phys_addr_t phys, unsigned long virt,
  272. phys_addr_t size, pgprot_t prot)
  273. {
  274. if (virt < VMALLOC_START) {
  275. pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
  276. &phys, virt);
  277. return;
  278. }
  279. __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
  280. NULL, debug_pagealloc_enabled());
  281. }
  282. static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
  283. {
  284. phys_addr_t kernel_start = __pa_symbol(_text);
  285. phys_addr_t kernel_end = __pa_symbol(__init_begin);
  286. /*
  287. * Take care not to create a writable alias for the
  288. * read-only text and rodata sections of the kernel image.
  289. */
  290. /* No overlap with the kernel text/rodata */
  291. if (end < kernel_start || start >= kernel_end) {
  292. __create_pgd_mapping(pgd, start, __phys_to_virt(start),
  293. end - start, PAGE_KERNEL,
  294. early_pgtable_alloc,
  295. debug_pagealloc_enabled());
  296. return;
  297. }
  298. /*
  299. * This block overlaps the kernel text/rodata mappings.
  300. * Map the portion(s) which don't overlap.
  301. */
  302. if (start < kernel_start)
  303. __create_pgd_mapping(pgd, start,
  304. __phys_to_virt(start),
  305. kernel_start - start, PAGE_KERNEL,
  306. early_pgtable_alloc,
  307. debug_pagealloc_enabled());
  308. if (kernel_end < end)
  309. __create_pgd_mapping(pgd, kernel_end,
  310. __phys_to_virt(kernel_end),
  311. end - kernel_end, PAGE_KERNEL,
  312. early_pgtable_alloc,
  313. debug_pagealloc_enabled());
  314. /*
  315. * Map the linear alias of the [_text, __init_begin) interval as
  316. * read-only/non-executable. This makes the contents of the
  317. * region accessible to subsystems such as hibernate, but
  318. * protects it from inadvertent modification or execution.
  319. */
  320. __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
  321. kernel_end - kernel_start, PAGE_KERNEL_RO,
  322. early_pgtable_alloc, debug_pagealloc_enabled());
  323. }
  324. static void __init map_mem(pgd_t *pgd)
  325. {
  326. struct memblock_region *reg;
  327. /* map all the memory banks */
  328. for_each_memblock(memory, reg) {
  329. phys_addr_t start = reg->base;
  330. phys_addr_t end = start + reg->size;
  331. if (start >= end)
  332. break;
  333. if (memblock_is_nomap(reg))
  334. continue;
  335. __map_memblock(pgd, start, end);
  336. }
  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_symbol(_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_symbol(__start_rodata), (unsigned long)__start_rodata,
  350. section_size, PAGE_KERNEL_RO);
  351. /* flush the TLBs after updating live kernel mappings */
  352. flush_tlb_all();
  353. debug_checkwx();
  354. }
  355. static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
  356. pgprot_t prot, struct vm_struct *vma)
  357. {
  358. phys_addr_t pa_start = __pa_symbol(va_start);
  359. unsigned long size = va_end - va_start;
  360. BUG_ON(!PAGE_ALIGNED(pa_start));
  361. BUG_ON(!PAGE_ALIGNED(size));
  362. __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
  363. early_pgtable_alloc, debug_pagealloc_enabled());
  364. vma->addr = va_start;
  365. vma->phys_addr = pa_start;
  366. vma->size = size;
  367. vma->flags = VM_MAP;
  368. vma->caller = __builtin_return_address(0);
  369. vm_area_add_early(vma);
  370. }
  371. /*
  372. * Create fine-grained mappings for the kernel.
  373. */
  374. static void __init map_kernel(pgd_t *pgd)
  375. {
  376. static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
  377. map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
  378. map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
  379. map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
  380. &vmlinux_init);
  381. map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
  382. if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
  383. /*
  384. * The fixmap falls in a separate pgd to the kernel, and doesn't
  385. * live in the carveout for the swapper_pg_dir. We can simply
  386. * re-use the existing dir for the fixmap.
  387. */
  388. set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
  389. *pgd_offset_k(FIXADDR_START));
  390. } else if (CONFIG_PGTABLE_LEVELS > 3) {
  391. /*
  392. * The fixmap shares its top level pgd entry with the kernel
  393. * mapping. This can really only occur when we are running
  394. * with 16k/4 levels, so we can simply reuse the pud level
  395. * entry instead.
  396. */
  397. BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
  398. set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
  399. __pud(__pa_symbol(bm_pmd) | PUD_TYPE_TABLE));
  400. pud_clear_fixmap();
  401. } else {
  402. BUG();
  403. }
  404. kasan_copy_shadow(pgd);
  405. }
  406. /*
  407. * paging_init() sets up the page tables, initialises the zone memory
  408. * maps and sets up the zero page.
  409. */
  410. void __init paging_init(void)
  411. {
  412. phys_addr_t pgd_phys = early_pgtable_alloc();
  413. pgd_t *pgd = pgd_set_fixmap(pgd_phys);
  414. map_kernel(pgd);
  415. map_mem(pgd);
  416. /*
  417. * We want to reuse the original swapper_pg_dir so we don't have to
  418. * communicate the new address to non-coherent secondaries in
  419. * secondary_entry, and so cpu_switch_mm can generate the address with
  420. * adrp+add rather than a load from some global variable.
  421. *
  422. * To do this we need to go via a temporary pgd.
  423. */
  424. cpu_replace_ttbr1(__va(pgd_phys));
  425. memcpy(swapper_pg_dir, pgd, PGD_SIZE);
  426. cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
  427. pgd_clear_fixmap();
  428. memblock_free(pgd_phys, PAGE_SIZE);
  429. /*
  430. * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
  431. * allocated with it.
  432. */
  433. memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE,
  434. SWAPPER_DIR_SIZE - PAGE_SIZE);
  435. }
  436. /*
  437. * Check whether a kernel address is valid (derived from arch/x86/).
  438. */
  439. int kern_addr_valid(unsigned long addr)
  440. {
  441. pgd_t *pgd;
  442. pud_t *pud;
  443. pmd_t *pmd;
  444. pte_t *pte;
  445. if ((((long)addr) >> VA_BITS) != -1UL)
  446. return 0;
  447. pgd = pgd_offset_k(addr);
  448. if (pgd_none(*pgd))
  449. return 0;
  450. pud = pud_offset(pgd, addr);
  451. if (pud_none(*pud))
  452. return 0;
  453. if (pud_sect(*pud))
  454. return pfn_valid(pud_pfn(*pud));
  455. pmd = pmd_offset(pud, addr);
  456. if (pmd_none(*pmd))
  457. return 0;
  458. if (pmd_sect(*pmd))
  459. return pfn_valid(pmd_pfn(*pmd));
  460. pte = pte_offset_kernel(pmd, addr);
  461. if (pte_none(*pte))
  462. return 0;
  463. return pfn_valid(pte_pfn(*pte));
  464. }
  465. #ifdef CONFIG_SPARSEMEM_VMEMMAP
  466. #if !ARM64_SWAPPER_USES_SECTION_MAPS
  467. int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
  468. {
  469. return vmemmap_populate_basepages(start, end, node);
  470. }
  471. #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
  472. int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
  473. {
  474. unsigned long addr = start;
  475. unsigned long next;
  476. pgd_t *pgd;
  477. pud_t *pud;
  478. pmd_t *pmd;
  479. do {
  480. next = pmd_addr_end(addr, end);
  481. pgd = vmemmap_pgd_populate(addr, node);
  482. if (!pgd)
  483. return -ENOMEM;
  484. pud = vmemmap_pud_populate(pgd, addr, node);
  485. if (!pud)
  486. return -ENOMEM;
  487. pmd = pmd_offset(pud, addr);
  488. if (pmd_none(*pmd)) {
  489. void *p = NULL;
  490. p = vmemmap_alloc_block_buf(PMD_SIZE, node);
  491. if (!p)
  492. return -ENOMEM;
  493. set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
  494. } else
  495. vmemmap_verify((pte_t *)pmd, node, addr, next);
  496. } while (addr = next, addr != end);
  497. return 0;
  498. }
  499. #endif /* CONFIG_ARM64_64K_PAGES */
  500. void vmemmap_free(unsigned long start, unsigned long end)
  501. {
  502. }
  503. #endif /* CONFIG_SPARSEMEM_VMEMMAP */
  504. static inline pud_t * fixmap_pud(unsigned long addr)
  505. {
  506. pgd_t *pgd = pgd_offset_k(addr);
  507. BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
  508. return pud_offset_kimg(pgd, addr);
  509. }
  510. static inline pmd_t * fixmap_pmd(unsigned long addr)
  511. {
  512. pud_t *pud = fixmap_pud(addr);
  513. BUG_ON(pud_none(*pud) || pud_bad(*pud));
  514. return pmd_offset_kimg(pud, addr);
  515. }
  516. static inline pte_t * fixmap_pte(unsigned long addr)
  517. {
  518. return &bm_pte[pte_index(addr)];
  519. }
  520. /*
  521. * The p*d_populate functions call virt_to_phys implicitly so they can't be used
  522. * directly on kernel symbols (bm_p*d). This function is called too early to use
  523. * lm_alias so __p*d_populate functions must be used to populate with the
  524. * physical address from __pa_symbol.
  525. */
  526. void __init early_fixmap_init(void)
  527. {
  528. pgd_t *pgd;
  529. pud_t *pud;
  530. pmd_t *pmd;
  531. unsigned long addr = FIXADDR_START;
  532. pgd = pgd_offset_k(addr);
  533. if (CONFIG_PGTABLE_LEVELS > 3 &&
  534. !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa_symbol(bm_pud))) {
  535. /*
  536. * We only end up here if the kernel mapping and the fixmap
  537. * share the top level pgd entry, which should only happen on
  538. * 16k/4 levels configurations.
  539. */
  540. BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
  541. pud = pud_offset_kimg(pgd, addr);
  542. } else {
  543. if (pgd_none(*pgd))
  544. __pgd_populate(pgd, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
  545. pud = fixmap_pud(addr);
  546. }
  547. if (pud_none(*pud))
  548. __pud_populate(pud, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
  549. pmd = fixmap_pmd(addr);
  550. __pmd_populate(pmd, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
  551. /*
  552. * The boot-ioremap range spans multiple pmds, for which
  553. * we are not prepared:
  554. */
  555. BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
  556. != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
  557. if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
  558. || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
  559. WARN_ON(1);
  560. pr_warn("pmd %p != %p, %p\n",
  561. pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
  562. fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
  563. pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
  564. fix_to_virt(FIX_BTMAP_BEGIN));
  565. pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
  566. fix_to_virt(FIX_BTMAP_END));
  567. pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
  568. pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
  569. }
  570. }
  571. void __set_fixmap(enum fixed_addresses idx,
  572. phys_addr_t phys, pgprot_t flags)
  573. {
  574. unsigned long addr = __fix_to_virt(idx);
  575. pte_t *pte;
  576. BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
  577. pte = fixmap_pte(addr);
  578. if (pgprot_val(flags)) {
  579. set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
  580. } else {
  581. pte_clear(&init_mm, addr, pte);
  582. flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
  583. }
  584. }
  585. void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
  586. {
  587. const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
  588. int offset;
  589. void *dt_virt;
  590. /*
  591. * Check whether the physical FDT address is set and meets the minimum
  592. * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
  593. * at least 8 bytes so that we can always access the magic and size
  594. * fields of the FDT header after mapping the first chunk, double check
  595. * here if that is indeed the case.
  596. */
  597. BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
  598. if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
  599. return NULL;
  600. /*
  601. * Make sure that the FDT region can be mapped without the need to
  602. * allocate additional translation table pages, so that it is safe
  603. * to call create_mapping_noalloc() this early.
  604. *
  605. * On 64k pages, the FDT will be mapped using PTEs, so we need to
  606. * be in the same PMD as the rest of the fixmap.
  607. * On 4k pages, we'll use section mappings for the FDT so we only
  608. * have to be in the same PUD.
  609. */
  610. BUILD_BUG_ON(dt_virt_base % SZ_2M);
  611. BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
  612. __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
  613. offset = dt_phys % SWAPPER_BLOCK_SIZE;
  614. dt_virt = (void *)dt_virt_base + offset;
  615. /* map the first chunk so we can read the size from the header */
  616. create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
  617. dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
  618. if (fdt_magic(dt_virt) != FDT_MAGIC)
  619. return NULL;
  620. *size = fdt_totalsize(dt_virt);
  621. if (*size > MAX_FDT_SIZE)
  622. return NULL;
  623. if (offset + *size > SWAPPER_BLOCK_SIZE)
  624. create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
  625. round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
  626. return dt_virt;
  627. }
  628. void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
  629. {
  630. void *dt_virt;
  631. int size;
  632. dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
  633. if (!dt_virt)
  634. return NULL;
  635. memblock_reserve(dt_phys, size);
  636. return dt_virt;
  637. }
  638. int __init arch_ioremap_pud_supported(void)
  639. {
  640. /* only 4k granule supports level 1 block mappings */
  641. return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
  642. }
  643. int __init arch_ioremap_pmd_supported(void)
  644. {
  645. return 1;
  646. }
  647. int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
  648. {
  649. BUG_ON(phys & ~PUD_MASK);
  650. set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
  651. return 1;
  652. }
  653. int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
  654. {
  655. BUG_ON(phys & ~PMD_MASK);
  656. set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
  657. return 1;
  658. }
  659. int pud_clear_huge(pud_t *pud)
  660. {
  661. if (!pud_sect(*pud))
  662. return 0;
  663. pud_clear(pud);
  664. return 1;
  665. }
  666. int pmd_clear_huge(pmd_t *pmd)
  667. {
  668. if (!pmd_sect(*pmd))
  669. return 0;
  670. pmd_clear(pmd);
  671. return 1;
  672. }