vmem.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507
  1. // SPDX-License-Identifier: GPL-2.0
  2. #include <linux/sched/task.h>
  3. #include <linux/pgtable.h>
  4. #include <linux/kasan.h>
  5. #include <asm/page-states.h>
  6. #include <asm/pgalloc.h>
  7. #include <asm/facility.h>
  8. #include <asm/sections.h>
  9. #include <asm/ctlreg.h>
  10. #include <asm/physmem_info.h>
  11. #include <asm/maccess.h>
  12. #include <asm/abs_lowcore.h>
  13. #include "decompressor.h"
  14. #include "boot.h"
  15. struct ctlreg __bootdata_preserved(s390_invalid_asce);
  16. #ifdef CONFIG_PROC_FS
  17. atomic_long_t __bootdata_preserved(direct_pages_count[PG_DIRECT_MAP_MAX]);
  18. #endif
  19. #define init_mm (*(struct mm_struct *)vmlinux.init_mm_off)
  20. #define swapper_pg_dir vmlinux.swapper_pg_dir_off
  21. #define invalid_pg_dir vmlinux.invalid_pg_dir_off
  22. enum populate_mode {
  23. POPULATE_NONE,
  24. POPULATE_DIRECT,
  25. POPULATE_LOWCORE,
  26. POPULATE_ABS_LOWCORE,
  27. POPULATE_IDENTITY,
  28. POPULATE_KERNEL,
  29. #ifdef CONFIG_KASAN
  30. POPULATE_KASAN_MAP_SHADOW,
  31. POPULATE_KASAN_ZERO_SHADOW,
  32. POPULATE_KASAN_SHALLOW
  33. #endif
  34. };
  35. static void pgtable_populate(unsigned long addr, unsigned long end, enum populate_mode mode);
  36. #ifdef CONFIG_KASAN
  37. #define kasan_early_shadow_page vmlinux.kasan_early_shadow_page_off
  38. #define kasan_early_shadow_pte ((pte_t *)vmlinux.kasan_early_shadow_pte_off)
  39. #define kasan_early_shadow_pmd ((pmd_t *)vmlinux.kasan_early_shadow_pmd_off)
  40. #define kasan_early_shadow_pud ((pud_t *)vmlinux.kasan_early_shadow_pud_off)
  41. #define kasan_early_shadow_p4d ((p4d_t *)vmlinux.kasan_early_shadow_p4d_off)
  42. #define __sha(x) ((unsigned long)kasan_mem_to_shadow((void *)x))
  43. static pte_t pte_z;
  44. static inline void kasan_populate(unsigned long start, unsigned long end, enum populate_mode mode)
  45. {
  46. start = PAGE_ALIGN_DOWN(__sha(start));
  47. end = PAGE_ALIGN(__sha(end));
  48. pgtable_populate(start, end, mode);
  49. }
  50. static void kasan_populate_shadow(unsigned long kernel_start, unsigned long kernel_end)
  51. {
  52. pmd_t pmd_z = __pmd(__pa(kasan_early_shadow_pte) | _SEGMENT_ENTRY);
  53. pud_t pud_z = __pud(__pa(kasan_early_shadow_pmd) | _REGION3_ENTRY);
  54. p4d_t p4d_z = __p4d(__pa(kasan_early_shadow_pud) | _REGION2_ENTRY);
  55. unsigned long memgap_start = 0;
  56. unsigned long untracked_end;
  57. unsigned long start, end;
  58. int i;
  59. pte_z = __pte(__pa(kasan_early_shadow_page) | pgprot_val(PAGE_KERNEL_RO));
  60. if (!machine.has_nx)
  61. pte_z = clear_pte_bit(pte_z, __pgprot(_PAGE_NOEXEC));
  62. crst_table_init((unsigned long *)kasan_early_shadow_p4d, p4d_val(p4d_z));
  63. crst_table_init((unsigned long *)kasan_early_shadow_pud, pud_val(pud_z));
  64. crst_table_init((unsigned long *)kasan_early_shadow_pmd, pmd_val(pmd_z));
  65. memset64((u64 *)kasan_early_shadow_pte, pte_val(pte_z), PTRS_PER_PTE);
  66. __arch_set_page_dat(kasan_early_shadow_p4d, 1UL << CRST_ALLOC_ORDER);
  67. __arch_set_page_dat(kasan_early_shadow_pud, 1UL << CRST_ALLOC_ORDER);
  68. __arch_set_page_dat(kasan_early_shadow_pmd, 1UL << CRST_ALLOC_ORDER);
  69. __arch_set_page_dat(kasan_early_shadow_pte, 1);
  70. for_each_physmem_usable_range(i, &start, &end) {
  71. kasan_populate((unsigned long)__identity_va(start),
  72. (unsigned long)__identity_va(end),
  73. POPULATE_KASAN_MAP_SHADOW);
  74. if (memgap_start && physmem_info.info_source == MEM_DETECT_DIAG260) {
  75. kasan_populate((unsigned long)__identity_va(memgap_start),
  76. (unsigned long)__identity_va(start),
  77. POPULATE_KASAN_ZERO_SHADOW);
  78. }
  79. memgap_start = end;
  80. }
  81. kasan_populate(kernel_start + TEXT_OFFSET, kernel_end, POPULATE_KASAN_MAP_SHADOW);
  82. kasan_populate(0, (unsigned long)__identity_va(0), POPULATE_KASAN_ZERO_SHADOW);
  83. kasan_populate(AMODE31_START, AMODE31_END, POPULATE_KASAN_ZERO_SHADOW);
  84. if (IS_ENABLED(CONFIG_KASAN_VMALLOC)) {
  85. untracked_end = VMALLOC_START;
  86. /* shallowly populate kasan shadow for vmalloc and modules */
  87. kasan_populate(VMALLOC_START, MODULES_END, POPULATE_KASAN_SHALLOW);
  88. } else {
  89. untracked_end = MODULES_VADDR;
  90. }
  91. /* populate kasan shadow for untracked memory */
  92. kasan_populate((unsigned long)__identity_va(ident_map_size), untracked_end,
  93. POPULATE_KASAN_ZERO_SHADOW);
  94. kasan_populate(kernel_end, _REGION1_SIZE, POPULATE_KASAN_ZERO_SHADOW);
  95. }
  96. static bool kasan_pgd_populate_zero_shadow(pgd_t *pgd, unsigned long addr,
  97. unsigned long end, enum populate_mode mode)
  98. {
  99. if (mode == POPULATE_KASAN_ZERO_SHADOW &&
  100. IS_ALIGNED(addr, PGDIR_SIZE) && end - addr >= PGDIR_SIZE) {
  101. pgd_populate(&init_mm, pgd, kasan_early_shadow_p4d);
  102. return true;
  103. }
  104. return false;
  105. }
  106. static bool kasan_p4d_populate_zero_shadow(p4d_t *p4d, unsigned long addr,
  107. unsigned long end, enum populate_mode mode)
  108. {
  109. if (mode == POPULATE_KASAN_ZERO_SHADOW &&
  110. IS_ALIGNED(addr, P4D_SIZE) && end - addr >= P4D_SIZE) {
  111. p4d_populate(&init_mm, p4d, kasan_early_shadow_pud);
  112. return true;
  113. }
  114. return false;
  115. }
  116. static bool kasan_pud_populate_zero_shadow(pud_t *pud, unsigned long addr,
  117. unsigned long end, enum populate_mode mode)
  118. {
  119. if (mode == POPULATE_KASAN_ZERO_SHADOW &&
  120. IS_ALIGNED(addr, PUD_SIZE) && end - addr >= PUD_SIZE) {
  121. pud_populate(&init_mm, pud, kasan_early_shadow_pmd);
  122. return true;
  123. }
  124. return false;
  125. }
  126. static bool kasan_pmd_populate_zero_shadow(pmd_t *pmd, unsigned long addr,
  127. unsigned long end, enum populate_mode mode)
  128. {
  129. if (mode == POPULATE_KASAN_ZERO_SHADOW &&
  130. IS_ALIGNED(addr, PMD_SIZE) && end - addr >= PMD_SIZE) {
  131. pmd_populate(&init_mm, pmd, kasan_early_shadow_pte);
  132. return true;
  133. }
  134. return false;
  135. }
  136. static bool kasan_pte_populate_zero_shadow(pte_t *pte, enum populate_mode mode)
  137. {
  138. if (mode == POPULATE_KASAN_ZERO_SHADOW) {
  139. set_pte(pte, pte_z);
  140. return true;
  141. }
  142. return false;
  143. }
  144. #else
  145. static inline void kasan_populate_shadow(unsigned long kernel_start, unsigned long kernel_end)
  146. {
  147. }
  148. static inline bool kasan_pgd_populate_zero_shadow(pgd_t *pgd, unsigned long addr,
  149. unsigned long end, enum populate_mode mode)
  150. {
  151. return false;
  152. }
  153. static inline bool kasan_p4d_populate_zero_shadow(p4d_t *p4d, unsigned long addr,
  154. unsigned long end, enum populate_mode mode)
  155. {
  156. return false;
  157. }
  158. static inline bool kasan_pud_populate_zero_shadow(pud_t *pud, unsigned long addr,
  159. unsigned long end, enum populate_mode mode)
  160. {
  161. return false;
  162. }
  163. static inline bool kasan_pmd_populate_zero_shadow(pmd_t *pmd, unsigned long addr,
  164. unsigned long end, enum populate_mode mode)
  165. {
  166. return false;
  167. }
  168. static bool kasan_pte_populate_zero_shadow(pte_t *pte, enum populate_mode mode)
  169. {
  170. return false;
  171. }
  172. #endif
  173. /*
  174. * Mimic virt_to_kpte() in lack of init_mm symbol. Skip pmd NULL check though.
  175. */
  176. static inline pte_t *__virt_to_kpte(unsigned long va)
  177. {
  178. return pte_offset_kernel(pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va), va);
  179. }
  180. static void *boot_crst_alloc(unsigned long val)
  181. {
  182. unsigned long size = PAGE_SIZE << CRST_ALLOC_ORDER;
  183. unsigned long *table;
  184. table = (unsigned long *)physmem_alloc_top_down(RR_VMEM, size, size);
  185. crst_table_init(table, val);
  186. __arch_set_page_dat(table, 1UL << CRST_ALLOC_ORDER);
  187. return table;
  188. }
  189. static pte_t *boot_pte_alloc(void)
  190. {
  191. static void *pte_leftover;
  192. pte_t *pte;
  193. /*
  194. * handling pte_leftovers this way helps to avoid memory fragmentation
  195. * during POPULATE_KASAN_MAP_SHADOW when EDAT is off
  196. */
  197. if (!pte_leftover) {
  198. pte_leftover = (void *)physmem_alloc_top_down(RR_VMEM, PAGE_SIZE, PAGE_SIZE);
  199. pte = pte_leftover + _PAGE_TABLE_SIZE;
  200. __arch_set_page_dat(pte, 1);
  201. } else {
  202. pte = pte_leftover;
  203. pte_leftover = NULL;
  204. }
  205. memset64((u64 *)pte, _PAGE_INVALID, PTRS_PER_PTE);
  206. return pte;
  207. }
  208. static unsigned long _pa(unsigned long addr, unsigned long size, enum populate_mode mode)
  209. {
  210. switch (mode) {
  211. case POPULATE_NONE:
  212. return -1;
  213. case POPULATE_DIRECT:
  214. return addr;
  215. case POPULATE_LOWCORE:
  216. return __lowcore_pa(addr);
  217. case POPULATE_ABS_LOWCORE:
  218. return __abs_lowcore_pa(addr);
  219. case POPULATE_KERNEL:
  220. return __kernel_pa(addr);
  221. case POPULATE_IDENTITY:
  222. return __identity_pa(addr);
  223. #ifdef CONFIG_KASAN
  224. case POPULATE_KASAN_MAP_SHADOW:
  225. addr = physmem_alloc_top_down(RR_VMEM, size, size);
  226. memset((void *)addr, 0, size);
  227. return addr;
  228. #endif
  229. default:
  230. return -1;
  231. }
  232. }
  233. static bool large_allowed(enum populate_mode mode)
  234. {
  235. return (mode == POPULATE_DIRECT) || (mode == POPULATE_IDENTITY) || (mode == POPULATE_KERNEL);
  236. }
  237. static bool can_large_pud(pud_t *pu_dir, unsigned long addr, unsigned long end,
  238. enum populate_mode mode)
  239. {
  240. unsigned long size = end - addr;
  241. return machine.has_edat2 && large_allowed(mode) &&
  242. IS_ALIGNED(addr, PUD_SIZE) && (size >= PUD_SIZE) &&
  243. IS_ALIGNED(_pa(addr, size, mode), PUD_SIZE);
  244. }
  245. static bool can_large_pmd(pmd_t *pm_dir, unsigned long addr, unsigned long end,
  246. enum populate_mode mode)
  247. {
  248. unsigned long size = end - addr;
  249. return machine.has_edat1 && large_allowed(mode) &&
  250. IS_ALIGNED(addr, PMD_SIZE) && (size >= PMD_SIZE) &&
  251. IS_ALIGNED(_pa(addr, size, mode), PMD_SIZE);
  252. }
  253. static void pgtable_pte_populate(pmd_t *pmd, unsigned long addr, unsigned long end,
  254. enum populate_mode mode)
  255. {
  256. unsigned long pages = 0;
  257. pte_t *pte, entry;
  258. pte = pte_offset_kernel(pmd, addr);
  259. for (; addr < end; addr += PAGE_SIZE, pte++) {
  260. if (pte_none(*pte)) {
  261. if (kasan_pte_populate_zero_shadow(pte, mode))
  262. continue;
  263. entry = __pte(_pa(addr, PAGE_SIZE, mode));
  264. entry = set_pte_bit(entry, PAGE_KERNEL);
  265. if (!machine.has_nx)
  266. entry = clear_pte_bit(entry, __pgprot(_PAGE_NOEXEC));
  267. set_pte(pte, entry);
  268. pages++;
  269. }
  270. }
  271. if (mode == POPULATE_IDENTITY)
  272. update_page_count(PG_DIRECT_MAP_4K, pages);
  273. }
  274. static void pgtable_pmd_populate(pud_t *pud, unsigned long addr, unsigned long end,
  275. enum populate_mode mode)
  276. {
  277. unsigned long next, pages = 0;
  278. pmd_t *pmd, entry;
  279. pte_t *pte;
  280. pmd = pmd_offset(pud, addr);
  281. for (; addr < end; addr = next, pmd++) {
  282. next = pmd_addr_end(addr, end);
  283. if (pmd_none(*pmd)) {
  284. if (kasan_pmd_populate_zero_shadow(pmd, addr, next, mode))
  285. continue;
  286. if (can_large_pmd(pmd, addr, next, mode)) {
  287. entry = __pmd(_pa(addr, _SEGMENT_SIZE, mode));
  288. entry = set_pmd_bit(entry, SEGMENT_KERNEL);
  289. if (!machine.has_nx)
  290. entry = clear_pmd_bit(entry, __pgprot(_SEGMENT_ENTRY_NOEXEC));
  291. set_pmd(pmd, entry);
  292. pages++;
  293. continue;
  294. }
  295. pte = boot_pte_alloc();
  296. pmd_populate(&init_mm, pmd, pte);
  297. } else if (pmd_leaf(*pmd)) {
  298. continue;
  299. }
  300. pgtable_pte_populate(pmd, addr, next, mode);
  301. }
  302. if (mode == POPULATE_IDENTITY)
  303. update_page_count(PG_DIRECT_MAP_1M, pages);
  304. }
  305. static void pgtable_pud_populate(p4d_t *p4d, unsigned long addr, unsigned long end,
  306. enum populate_mode mode)
  307. {
  308. unsigned long next, pages = 0;
  309. pud_t *pud, entry;
  310. pmd_t *pmd;
  311. pud = pud_offset(p4d, addr);
  312. for (; addr < end; addr = next, pud++) {
  313. next = pud_addr_end(addr, end);
  314. if (pud_none(*pud)) {
  315. if (kasan_pud_populate_zero_shadow(pud, addr, next, mode))
  316. continue;
  317. if (can_large_pud(pud, addr, next, mode)) {
  318. entry = __pud(_pa(addr, _REGION3_SIZE, mode));
  319. entry = set_pud_bit(entry, REGION3_KERNEL);
  320. if (!machine.has_nx)
  321. entry = clear_pud_bit(entry, __pgprot(_REGION_ENTRY_NOEXEC));
  322. set_pud(pud, entry);
  323. pages++;
  324. continue;
  325. }
  326. pmd = boot_crst_alloc(_SEGMENT_ENTRY_EMPTY);
  327. pud_populate(&init_mm, pud, pmd);
  328. } else if (pud_leaf(*pud)) {
  329. continue;
  330. }
  331. pgtable_pmd_populate(pud, addr, next, mode);
  332. }
  333. if (mode == POPULATE_IDENTITY)
  334. update_page_count(PG_DIRECT_MAP_2G, pages);
  335. }
  336. static void pgtable_p4d_populate(pgd_t *pgd, unsigned long addr, unsigned long end,
  337. enum populate_mode mode)
  338. {
  339. unsigned long next;
  340. p4d_t *p4d;
  341. pud_t *pud;
  342. p4d = p4d_offset(pgd, addr);
  343. for (; addr < end; addr = next, p4d++) {
  344. next = p4d_addr_end(addr, end);
  345. if (p4d_none(*p4d)) {
  346. if (kasan_p4d_populate_zero_shadow(p4d, addr, next, mode))
  347. continue;
  348. pud = boot_crst_alloc(_REGION3_ENTRY_EMPTY);
  349. p4d_populate(&init_mm, p4d, pud);
  350. }
  351. pgtable_pud_populate(p4d, addr, next, mode);
  352. }
  353. }
  354. static void pgtable_populate(unsigned long addr, unsigned long end, enum populate_mode mode)
  355. {
  356. unsigned long next;
  357. pgd_t *pgd;
  358. p4d_t *p4d;
  359. pgd = pgd_offset(&init_mm, addr);
  360. for (; addr < end; addr = next, pgd++) {
  361. next = pgd_addr_end(addr, end);
  362. if (pgd_none(*pgd)) {
  363. if (kasan_pgd_populate_zero_shadow(pgd, addr, next, mode))
  364. continue;
  365. p4d = boot_crst_alloc(_REGION2_ENTRY_EMPTY);
  366. pgd_populate(&init_mm, pgd, p4d);
  367. }
  368. #ifdef CONFIG_KASAN
  369. if (mode == POPULATE_KASAN_SHALLOW)
  370. continue;
  371. #endif
  372. pgtable_p4d_populate(pgd, addr, next, mode);
  373. }
  374. }
  375. void setup_vmem(unsigned long kernel_start, unsigned long kernel_end, unsigned long asce_limit)
  376. {
  377. unsigned long lowcore_address = 0;
  378. unsigned long start, end;
  379. unsigned long asce_type;
  380. unsigned long asce_bits;
  381. pgd_t *init_mm_pgd;
  382. int i;
  383. /*
  384. * Mark whole memory as no-dat. This must be done before any
  385. * page tables are allocated, or kernel image builtin pages
  386. * are marked as dat tables.
  387. */
  388. for_each_physmem_online_range(i, &start, &end)
  389. __arch_set_page_nodat((void *)start, (end - start) >> PAGE_SHIFT);
  390. /*
  391. * init_mm->pgd contains virtual address of swapper_pg_dir.
  392. * It is unusable at this stage since DAT is yet off. Swap
  393. * it for physical address of swapper_pg_dir and restore
  394. * the virtual address after all page tables are created.
  395. */
  396. init_mm_pgd = init_mm.pgd;
  397. init_mm.pgd = (pgd_t *)swapper_pg_dir;
  398. if (asce_limit == _REGION1_SIZE) {
  399. asce_type = _REGION2_ENTRY_EMPTY;
  400. asce_bits = _ASCE_TYPE_REGION2 | _ASCE_TABLE_LENGTH;
  401. } else {
  402. asce_type = _REGION3_ENTRY_EMPTY;
  403. asce_bits = _ASCE_TYPE_REGION3 | _ASCE_TABLE_LENGTH;
  404. }
  405. s390_invalid_asce.val = invalid_pg_dir | _ASCE_TYPE_REGION3 | _ASCE_TABLE_LENGTH;
  406. crst_table_init((unsigned long *)swapper_pg_dir, asce_type);
  407. crst_table_init((unsigned long *)invalid_pg_dir, _REGION3_ENTRY_EMPTY);
  408. __arch_set_page_dat((void *)swapper_pg_dir, 1UL << CRST_ALLOC_ORDER);
  409. __arch_set_page_dat((void *)invalid_pg_dir, 1UL << CRST_ALLOC_ORDER);
  410. if (relocate_lowcore)
  411. lowcore_address = LOWCORE_ALT_ADDRESS;
  412. /*
  413. * To allow prefixing the lowcore must be mapped with 4KB pages.
  414. * To prevent creation of a large page at address 0 first map
  415. * the lowcore and create the identity mapping only afterwards.
  416. */
  417. pgtable_populate(lowcore_address,
  418. lowcore_address + sizeof(struct lowcore),
  419. POPULATE_LOWCORE);
  420. for_each_physmem_usable_range(i, &start, &end) {
  421. pgtable_populate((unsigned long)__identity_va(start),
  422. (unsigned long)__identity_va(end),
  423. POPULATE_IDENTITY);
  424. }
  425. /*
  426. * [kernel_start..kernel_start + TEXT_OFFSET] region is never
  427. * accessed as per the linker script:
  428. *
  429. * . = TEXT_OFFSET;
  430. *
  431. * Therefore, skip mapping TEXT_OFFSET bytes to prevent access to
  432. * [__kaslr_offset_phys..__kaslr_offset_phys + TEXT_OFFSET] region.
  433. */
  434. pgtable_populate(kernel_start + TEXT_OFFSET, kernel_end, POPULATE_KERNEL);
  435. pgtable_populate(AMODE31_START, AMODE31_END, POPULATE_DIRECT);
  436. pgtable_populate(__abs_lowcore, __abs_lowcore + sizeof(struct lowcore),
  437. POPULATE_ABS_LOWCORE);
  438. pgtable_populate(__memcpy_real_area, __memcpy_real_area + PAGE_SIZE,
  439. POPULATE_NONE);
  440. memcpy_real_ptep = __identity_va(__virt_to_kpte(__memcpy_real_area));
  441. kasan_populate_shadow(kernel_start, kernel_end);
  442. get_lowcore()->kernel_asce.val = swapper_pg_dir | asce_bits;
  443. get_lowcore()->user_asce = s390_invalid_asce;
  444. local_ctl_load(1, &get_lowcore()->kernel_asce);
  445. local_ctl_load(7, &get_lowcore()->user_asce);
  446. local_ctl_load(13, &get_lowcore()->kernel_asce);
  447. init_mm.context.asce = get_lowcore()->kernel_asce.val;
  448. init_mm.pgd = init_mm_pgd;
  449. }