mmu.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785
  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/export.h>
  20. #include <linux/kernel.h>
  21. #include <linux/errno.h>
  22. #include <linux/init.h>
  23. #include <linux/libfdt.h>
  24. #include <linux/mman.h>
  25. #include <linux/nodemask.h>
  26. #include <linux/memblock.h>
  27. #include <linux/fs.h>
  28. #include <linux/io.h>
  29. #include <linux/slab.h>
  30. #include <linux/stop_machine.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 "mm.h"
  43. u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
  44. u64 kimage_voffset __read_mostly;
  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 void alloc_init_pte(pmd_t *pmd, unsigned long addr,
  85. unsigned long end, unsigned long pfn,
  86. pgprot_t prot,
  87. phys_addr_t (*pgtable_alloc)(void))
  88. {
  89. pte_t *pte;
  90. BUG_ON(pmd_sect(*pmd));
  91. if (pmd_none(*pmd)) {
  92. phys_addr_t pte_phys;
  93. BUG_ON(!pgtable_alloc);
  94. pte_phys = pgtable_alloc();
  95. pte = pte_set_fixmap(pte_phys);
  96. __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
  97. pte_clear_fixmap();
  98. }
  99. BUG_ON(pmd_bad(*pmd));
  100. pte = pte_set_fixmap_offset(pmd, addr);
  101. do {
  102. set_pte(pte, pfn_pte(pfn, prot));
  103. pfn++;
  104. } while (pte++, addr += PAGE_SIZE, addr != end);
  105. pte_clear_fixmap();
  106. }
  107. static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
  108. phys_addr_t phys, pgprot_t prot,
  109. phys_addr_t (*pgtable_alloc)(void),
  110. bool allow_block_mappings)
  111. {
  112. pmd_t *pmd;
  113. unsigned long next;
  114. /*
  115. * Check for initial section mappings in the pgd/pud and remove them.
  116. */
  117. BUG_ON(pud_sect(*pud));
  118. if (pud_none(*pud)) {
  119. phys_addr_t pmd_phys;
  120. BUG_ON(!pgtable_alloc);
  121. pmd_phys = pgtable_alloc();
  122. pmd = pmd_set_fixmap(pmd_phys);
  123. __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
  124. pmd_clear_fixmap();
  125. }
  126. BUG_ON(pud_bad(*pud));
  127. pmd = pmd_set_fixmap_offset(pud, addr);
  128. do {
  129. next = pmd_addr_end(addr, end);
  130. /* try section mapping first */
  131. if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
  132. allow_block_mappings) {
  133. pmd_t old_pmd =*pmd;
  134. pmd_set_huge(pmd, phys, prot);
  135. /*
  136. * Check for previous table entries created during
  137. * boot (__create_page_tables) and flush them.
  138. */
  139. if (!pmd_none(old_pmd)) {
  140. flush_tlb_all();
  141. if (pmd_table(old_pmd)) {
  142. phys_addr_t table = pmd_page_paddr(old_pmd);
  143. if (!WARN_ON_ONCE(slab_is_available()))
  144. memblock_free(table, PAGE_SIZE);
  145. }
  146. }
  147. } else {
  148. alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
  149. prot, pgtable_alloc);
  150. }
  151. phys += next - addr;
  152. } while (pmd++, addr = next, addr != end);
  153. pmd_clear_fixmap();
  154. }
  155. static inline bool use_1G_block(unsigned long addr, unsigned long next,
  156. unsigned long phys)
  157. {
  158. if (PAGE_SHIFT != 12)
  159. return false;
  160. if (((addr | next | phys) & ~PUD_MASK) != 0)
  161. return false;
  162. return true;
  163. }
  164. static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
  165. phys_addr_t phys, pgprot_t prot,
  166. phys_addr_t (*pgtable_alloc)(void),
  167. bool allow_block_mappings)
  168. {
  169. pud_t *pud;
  170. unsigned long next;
  171. if (pgd_none(*pgd)) {
  172. phys_addr_t pud_phys;
  173. BUG_ON(!pgtable_alloc);
  174. pud_phys = pgtable_alloc();
  175. __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
  176. }
  177. BUG_ON(pgd_bad(*pgd));
  178. pud = pud_set_fixmap_offset(pgd, addr);
  179. do {
  180. next = pud_addr_end(addr, end);
  181. /*
  182. * For 4K granule only, attempt to put down a 1GB block
  183. */
  184. if (use_1G_block(addr, next, phys) && allow_block_mappings) {
  185. pud_t old_pud = *pud;
  186. pud_set_huge(pud, phys, prot);
  187. /*
  188. * If we have an old value for a pud, it will
  189. * be pointing to a pmd table that we no longer
  190. * need (from swapper_pg_dir).
  191. *
  192. * Look up the old pmd table and free it.
  193. */
  194. if (!pud_none(old_pud)) {
  195. flush_tlb_all();
  196. if (pud_table(old_pud)) {
  197. phys_addr_t table = pud_page_paddr(old_pud);
  198. if (!WARN_ON_ONCE(slab_is_available()))
  199. memblock_free(table, PAGE_SIZE);
  200. }
  201. }
  202. } else {
  203. alloc_init_pmd(pud, addr, next, phys, prot,
  204. pgtable_alloc, allow_block_mappings);
  205. }
  206. phys += next - addr;
  207. } while (pud++, addr = next, addr != end);
  208. pud_clear_fixmap();
  209. }
  210. static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
  211. unsigned long virt, phys_addr_t size,
  212. pgprot_t prot,
  213. phys_addr_t (*pgtable_alloc)(void),
  214. bool allow_block_mappings)
  215. {
  216. unsigned long addr, length, end, next;
  217. pgd_t *pgd = pgd_offset_raw(pgdir, virt);
  218. /*
  219. * If the virtual and physical address don't have the same offset
  220. * within a page, we cannot map the region as the caller expects.
  221. */
  222. if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
  223. return;
  224. phys &= PAGE_MASK;
  225. addr = virt & PAGE_MASK;
  226. length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
  227. end = addr + length;
  228. do {
  229. next = pgd_addr_end(addr, end);
  230. alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
  231. allow_block_mappings);
  232. phys += next - addr;
  233. } while (pgd++, addr = next, addr != end);
  234. }
  235. static phys_addr_t pgd_pgtable_alloc(void)
  236. {
  237. void *ptr = (void *)__get_free_page(PGALLOC_GFP);
  238. if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
  239. BUG();
  240. /* Ensure the zeroed page is visible to the page table walker */
  241. dsb(ishst);
  242. return __pa(ptr);
  243. }
  244. /*
  245. * This function can only be used to modify existing table entries,
  246. * without allocating new levels of table. Note that this permits the
  247. * creation of new section or page entries.
  248. */
  249. static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
  250. phys_addr_t size, pgprot_t prot)
  251. {
  252. if (virt < VMALLOC_START) {
  253. pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
  254. &phys, virt);
  255. return;
  256. }
  257. __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, true);
  258. }
  259. void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
  260. unsigned long virt, phys_addr_t size,
  261. pgprot_t prot, bool allow_block_mappings)
  262. {
  263. BUG_ON(mm == &init_mm);
  264. __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
  265. pgd_pgtable_alloc, allow_block_mappings);
  266. }
  267. static void create_mapping_late(phys_addr_t phys, unsigned long virt,
  268. phys_addr_t size, pgprot_t prot)
  269. {
  270. if (virt < VMALLOC_START) {
  271. pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
  272. &phys, virt);
  273. return;
  274. }
  275. __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
  276. NULL, !debug_pagealloc_enabled());
  277. }
  278. static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
  279. {
  280. unsigned long kernel_start = __pa(_text);
  281. unsigned long kernel_end = __pa(__init_begin);
  282. /*
  283. * Take care not to create a writable alias for the
  284. * read-only text and rodata sections of the kernel image.
  285. */
  286. /* No overlap with the kernel text/rodata */
  287. if (end < kernel_start || start >= kernel_end) {
  288. __create_pgd_mapping(pgd, start, __phys_to_virt(start),
  289. end - start, PAGE_KERNEL,
  290. early_pgtable_alloc,
  291. !debug_pagealloc_enabled());
  292. return;
  293. }
  294. /*
  295. * This block overlaps the kernel text/rodata mappings.
  296. * Map the portion(s) which don't overlap.
  297. */
  298. if (start < kernel_start)
  299. __create_pgd_mapping(pgd, start,
  300. __phys_to_virt(start),
  301. kernel_start - start, PAGE_KERNEL,
  302. early_pgtable_alloc,
  303. !debug_pagealloc_enabled());
  304. if (kernel_end < end)
  305. __create_pgd_mapping(pgd, kernel_end,
  306. __phys_to_virt(kernel_end),
  307. end - kernel_end, PAGE_KERNEL,
  308. early_pgtable_alloc,
  309. !debug_pagealloc_enabled());
  310. /*
  311. * Map the linear alias of the [_text, __init_begin) interval as
  312. * read-only/non-executable. This makes the contents of the
  313. * region accessible to subsystems such as hibernate, but
  314. * protects it from inadvertent modification or execution.
  315. */
  316. __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
  317. kernel_end - kernel_start, PAGE_KERNEL_RO,
  318. early_pgtable_alloc, !debug_pagealloc_enabled());
  319. }
  320. static void __init map_mem(pgd_t *pgd)
  321. {
  322. struct memblock_region *reg;
  323. /* map all the memory banks */
  324. for_each_memblock(memory, reg) {
  325. phys_addr_t start = reg->base;
  326. phys_addr_t end = start + reg->size;
  327. if (start >= end)
  328. break;
  329. if (memblock_is_nomap(reg))
  330. continue;
  331. __map_memblock(pgd, start, end);
  332. }
  333. }
  334. void mark_rodata_ro(void)
  335. {
  336. unsigned long section_size;
  337. section_size = (unsigned long)_etext - (unsigned long)_text;
  338. create_mapping_late(__pa(_text), (unsigned long)_text,
  339. section_size, PAGE_KERNEL_ROX);
  340. /*
  341. * mark .rodata as read only. Use __init_begin rather than __end_rodata
  342. * to cover NOTES and EXCEPTION_TABLE.
  343. */
  344. section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
  345. create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
  346. section_size, PAGE_KERNEL_RO);
  347. }
  348. void fixup_init(void)
  349. {
  350. /*
  351. * Unmap the __init region but leave the VM area in place. This
  352. * prevents the region from being reused for kernel modules, which
  353. * is not supported by kallsyms.
  354. */
  355. unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
  356. }
  357. static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
  358. pgprot_t prot, struct vm_struct *vma)
  359. {
  360. phys_addr_t pa_start = __pa(va_start);
  361. unsigned long size = va_end - va_start;
  362. BUG_ON(!PAGE_ALIGNED(pa_start));
  363. BUG_ON(!PAGE_ALIGNED(size));
  364. __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
  365. early_pgtable_alloc, !debug_pagealloc_enabled());
  366. vma->addr = va_start;
  367. vma->phys_addr = pa_start;
  368. vma->size = size;
  369. vma->flags = VM_MAP;
  370. vma->caller = __builtin_return_address(0);
  371. vm_area_add_early(vma);
  372. }
  373. /*
  374. * Create fine-grained mappings for the kernel.
  375. */
  376. static void __init map_kernel(pgd_t *pgd)
  377. {
  378. static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
  379. map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
  380. map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
  381. map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
  382. &vmlinux_init);
  383. map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
  384. if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
  385. /*
  386. * The fixmap falls in a separate pgd to the kernel, and doesn't
  387. * live in the carveout for the swapper_pg_dir. We can simply
  388. * re-use the existing dir for the fixmap.
  389. */
  390. set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
  391. *pgd_offset_k(FIXADDR_START));
  392. } else if (CONFIG_PGTABLE_LEVELS > 3) {
  393. /*
  394. * The fixmap shares its top level pgd entry with the kernel
  395. * mapping. This can really only occur when we are running
  396. * with 16k/4 levels, so we can simply reuse the pud level
  397. * entry instead.
  398. */
  399. BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
  400. set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
  401. __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
  402. pud_clear_fixmap();
  403. } else {
  404. BUG();
  405. }
  406. kasan_copy_shadow(pgd);
  407. }
  408. /*
  409. * paging_init() sets up the page tables, initialises the zone memory
  410. * maps and sets up the zero page.
  411. */
  412. void __init paging_init(void)
  413. {
  414. phys_addr_t pgd_phys = early_pgtable_alloc();
  415. pgd_t *pgd = pgd_set_fixmap(pgd_phys);
  416. map_kernel(pgd);
  417. map_mem(pgd);
  418. /*
  419. * We want to reuse the original swapper_pg_dir so we don't have to
  420. * communicate the new address to non-coherent secondaries in
  421. * secondary_entry, and so cpu_switch_mm can generate the address with
  422. * adrp+add rather than a load from some global variable.
  423. *
  424. * To do this we need to go via a temporary pgd.
  425. */
  426. cpu_replace_ttbr1(__va(pgd_phys));
  427. memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
  428. cpu_replace_ttbr1(swapper_pg_dir);
  429. pgd_clear_fixmap();
  430. memblock_free(pgd_phys, PAGE_SIZE);
  431. /*
  432. * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
  433. * allocated with it.
  434. */
  435. memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
  436. SWAPPER_DIR_SIZE - PAGE_SIZE);
  437. }
  438. /*
  439. * Check whether a kernel address is valid (derived from arch/x86/).
  440. */
  441. int kern_addr_valid(unsigned long addr)
  442. {
  443. pgd_t *pgd;
  444. pud_t *pud;
  445. pmd_t *pmd;
  446. pte_t *pte;
  447. if ((((long)addr) >> VA_BITS) != -1UL)
  448. return 0;
  449. pgd = pgd_offset_k(addr);
  450. if (pgd_none(*pgd))
  451. return 0;
  452. pud = pud_offset(pgd, addr);
  453. if (pud_none(*pud))
  454. return 0;
  455. if (pud_sect(*pud))
  456. return pfn_valid(pud_pfn(*pud));
  457. pmd = pmd_offset(pud, addr);
  458. if (pmd_none(*pmd))
  459. return 0;
  460. if (pmd_sect(*pmd))
  461. return pfn_valid(pmd_pfn(*pmd));
  462. pte = pte_offset_kernel(pmd, addr);
  463. if (pte_none(*pte))
  464. return 0;
  465. return pfn_valid(pte_pfn(*pte));
  466. }
  467. #ifdef CONFIG_SPARSEMEM_VMEMMAP
  468. #if !ARM64_SWAPPER_USES_SECTION_MAPS
  469. int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
  470. {
  471. return vmemmap_populate_basepages(start, end, node);
  472. }
  473. #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
  474. int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
  475. {
  476. unsigned long addr = start;
  477. unsigned long next;
  478. pgd_t *pgd;
  479. pud_t *pud;
  480. pmd_t *pmd;
  481. do {
  482. next = pmd_addr_end(addr, end);
  483. pgd = vmemmap_pgd_populate(addr, node);
  484. if (!pgd)
  485. return -ENOMEM;
  486. pud = vmemmap_pud_populate(pgd, addr, node);
  487. if (!pud)
  488. return -ENOMEM;
  489. pmd = pmd_offset(pud, addr);
  490. if (pmd_none(*pmd)) {
  491. void *p = NULL;
  492. p = vmemmap_alloc_block_buf(PMD_SIZE, node);
  493. if (!p)
  494. return -ENOMEM;
  495. set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
  496. } else
  497. vmemmap_verify((pte_t *)pmd, node, addr, next);
  498. } while (addr = next, addr != end);
  499. return 0;
  500. }
  501. #endif /* CONFIG_ARM64_64K_PAGES */
  502. void vmemmap_free(unsigned long start, unsigned long end)
  503. {
  504. }
  505. #endif /* CONFIG_SPARSEMEM_VMEMMAP */
  506. static inline pud_t * fixmap_pud(unsigned long addr)
  507. {
  508. pgd_t *pgd = pgd_offset_k(addr);
  509. BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
  510. return pud_offset_kimg(pgd, addr);
  511. }
  512. static inline pmd_t * fixmap_pmd(unsigned long addr)
  513. {
  514. pud_t *pud = fixmap_pud(addr);
  515. BUG_ON(pud_none(*pud) || pud_bad(*pud));
  516. return pmd_offset_kimg(pud, addr);
  517. }
  518. static inline pte_t * fixmap_pte(unsigned long addr)
  519. {
  520. return &bm_pte[pte_index(addr)];
  521. }
  522. void __init early_fixmap_init(void)
  523. {
  524. pgd_t *pgd;
  525. pud_t *pud;
  526. pmd_t *pmd;
  527. unsigned long addr = FIXADDR_START;
  528. pgd = pgd_offset_k(addr);
  529. if (CONFIG_PGTABLE_LEVELS > 3 &&
  530. !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
  531. /*
  532. * We only end up here if the kernel mapping and the fixmap
  533. * share the top level pgd entry, which should only happen on
  534. * 16k/4 levels configurations.
  535. */
  536. BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
  537. pud = pud_offset_kimg(pgd, addr);
  538. } else {
  539. pgd_populate(&init_mm, pgd, bm_pud);
  540. pud = fixmap_pud(addr);
  541. }
  542. pud_populate(&init_mm, pud, bm_pmd);
  543. pmd = fixmap_pmd(addr);
  544. pmd_populate_kernel(&init_mm, pmd, bm_pte);
  545. /*
  546. * The boot-ioremap range spans multiple pmds, for which
  547. * we are not prepared:
  548. */
  549. BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
  550. != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
  551. if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
  552. || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
  553. WARN_ON(1);
  554. pr_warn("pmd %p != %p, %p\n",
  555. pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
  556. fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
  557. pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
  558. fix_to_virt(FIX_BTMAP_BEGIN));
  559. pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
  560. fix_to_virt(FIX_BTMAP_END));
  561. pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
  562. pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
  563. }
  564. }
  565. void __set_fixmap(enum fixed_addresses idx,
  566. phys_addr_t phys, pgprot_t flags)
  567. {
  568. unsigned long addr = __fix_to_virt(idx);
  569. pte_t *pte;
  570. BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
  571. pte = fixmap_pte(addr);
  572. if (pgprot_val(flags)) {
  573. set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
  574. } else {
  575. pte_clear(&init_mm, addr, pte);
  576. flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
  577. }
  578. }
  579. void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
  580. {
  581. const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
  582. int offset;
  583. void *dt_virt;
  584. /*
  585. * Check whether the physical FDT address is set and meets the minimum
  586. * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
  587. * at least 8 bytes so that we can always access the magic and size
  588. * fields of the FDT header after mapping the first chunk, double check
  589. * here if that is indeed the case.
  590. */
  591. BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
  592. if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
  593. return NULL;
  594. /*
  595. * Make sure that the FDT region can be mapped without the need to
  596. * allocate additional translation table pages, so that it is safe
  597. * to call create_mapping_noalloc() this early.
  598. *
  599. * On 64k pages, the FDT will be mapped using PTEs, so we need to
  600. * be in the same PMD as the rest of the fixmap.
  601. * On 4k pages, we'll use section mappings for the FDT so we only
  602. * have to be in the same PUD.
  603. */
  604. BUILD_BUG_ON(dt_virt_base % SZ_2M);
  605. BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
  606. __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
  607. offset = dt_phys % SWAPPER_BLOCK_SIZE;
  608. dt_virt = (void *)dt_virt_base + offset;
  609. /* map the first chunk so we can read the size from the header */
  610. create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
  611. dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
  612. if (fdt_magic(dt_virt) != FDT_MAGIC)
  613. return NULL;
  614. *size = fdt_totalsize(dt_virt);
  615. if (*size > MAX_FDT_SIZE)
  616. return NULL;
  617. if (offset + *size > SWAPPER_BLOCK_SIZE)
  618. create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
  619. round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
  620. return dt_virt;
  621. }
  622. void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
  623. {
  624. void *dt_virt;
  625. int size;
  626. dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
  627. if (!dt_virt)
  628. return NULL;
  629. memblock_reserve(dt_phys, size);
  630. return dt_virt;
  631. }
  632. int __init arch_ioremap_pud_supported(void)
  633. {
  634. /* only 4k granule supports level 1 block mappings */
  635. return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
  636. }
  637. int __init arch_ioremap_pmd_supported(void)
  638. {
  639. return 1;
  640. }
  641. int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
  642. {
  643. BUG_ON(phys & ~PUD_MASK);
  644. set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
  645. return 1;
  646. }
  647. int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
  648. {
  649. BUG_ON(phys & ~PMD_MASK);
  650. set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
  651. return 1;
  652. }
  653. int pud_clear_huge(pud_t *pud)
  654. {
  655. if (!pud_sect(*pud))
  656. return 0;
  657. pud_clear(pud);
  658. return 1;
  659. }
  660. int pmd_clear_huge(pmd_t *pmd)
  661. {
  662. if (!pmd_sect(*pmd))
  663. return 0;
  664. pmd_clear(pmd);
  665. return 1;
  666. }