mmu.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792
  1. /*
  2. * linux/arch/arm/mm/mmu.c
  3. *
  4. * Copyright (C) 1995-2005 Russell King
  5. *
  6. * This program is free software; you can redistribute it and/or modify
  7. * it under the terms of the GNU General Public License version 2 as
  8. * published by the Free Software Foundation.
  9. */
  10. #include <linux/module.h>
  11. #include <linux/kernel.h>
  12. #include <linux/errno.h>
  13. #include <linux/init.h>
  14. #include <linux/bootmem.h>
  15. #include <linux/mman.h>
  16. #include <linux/nodemask.h>
  17. #include <asm/mach-types.h>
  18. #include <asm/setup.h>
  19. #include <asm/sizes.h>
  20. #include <asm/tlb.h>
  21. #include <asm/mach/arch.h>
  22. #include <asm/mach/map.h>
  23. #include "mm.h"
  24. DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  25. extern void _stext, _etext, __data_start, _end;
  26. extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
  27. /*
  28. * empty_zero_page is a special page that is used for
  29. * zero-initialized data and COW.
  30. */
  31. struct page *empty_zero_page;
  32. /*
  33. * The pmd table for the upper-most set of pages.
  34. */
  35. pmd_t *top_pmd;
  36. #define CPOLICY_UNCACHED 0
  37. #define CPOLICY_BUFFERED 1
  38. #define CPOLICY_WRITETHROUGH 2
  39. #define CPOLICY_WRITEBACK 3
  40. #define CPOLICY_WRITEALLOC 4
  41. static unsigned int cachepolicy __initdata = CPOLICY_WRITEBACK;
  42. static unsigned int ecc_mask __initdata = 0;
  43. pgprot_t pgprot_kernel;
  44. EXPORT_SYMBOL(pgprot_kernel);
  45. struct cachepolicy {
  46. const char policy[16];
  47. unsigned int cr_mask;
  48. unsigned int pmd;
  49. unsigned int pte;
  50. };
  51. static struct cachepolicy cache_policies[] __initdata = {
  52. {
  53. .policy = "uncached",
  54. .cr_mask = CR_W|CR_C,
  55. .pmd = PMD_SECT_UNCACHED,
  56. .pte = 0,
  57. }, {
  58. .policy = "buffered",
  59. .cr_mask = CR_C,
  60. .pmd = PMD_SECT_BUFFERED,
  61. .pte = PTE_BUFFERABLE,
  62. }, {
  63. .policy = "writethrough",
  64. .cr_mask = 0,
  65. .pmd = PMD_SECT_WT,
  66. .pte = PTE_CACHEABLE,
  67. }, {
  68. .policy = "writeback",
  69. .cr_mask = 0,
  70. .pmd = PMD_SECT_WB,
  71. .pte = PTE_BUFFERABLE|PTE_CACHEABLE,
  72. }, {
  73. .policy = "writealloc",
  74. .cr_mask = 0,
  75. .pmd = PMD_SECT_WBWA,
  76. .pte = PTE_BUFFERABLE|PTE_CACHEABLE,
  77. }
  78. };
  79. /*
  80. * These are useful for identifing cache coherency
  81. * problems by allowing the cache or the cache and
  82. * writebuffer to be turned off. (Note: the write
  83. * buffer should not be on and the cache off).
  84. */
  85. static void __init early_cachepolicy(char **p)
  86. {
  87. int i;
  88. for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
  89. int len = strlen(cache_policies[i].policy);
  90. if (memcmp(*p, cache_policies[i].policy, len) == 0) {
  91. cachepolicy = i;
  92. cr_alignment &= ~cache_policies[i].cr_mask;
  93. cr_no_alignment &= ~cache_policies[i].cr_mask;
  94. *p += len;
  95. break;
  96. }
  97. }
  98. if (i == ARRAY_SIZE(cache_policies))
  99. printk(KERN_ERR "ERROR: unknown or unsupported cache policy\n");
  100. flush_cache_all();
  101. set_cr(cr_alignment);
  102. }
  103. __early_param("cachepolicy=", early_cachepolicy);
  104. static void __init early_nocache(char **__unused)
  105. {
  106. char *p = "buffered";
  107. printk(KERN_WARNING "nocache is deprecated; use cachepolicy=%s\n", p);
  108. early_cachepolicy(&p);
  109. }
  110. __early_param("nocache", early_nocache);
  111. static void __init early_nowrite(char **__unused)
  112. {
  113. char *p = "uncached";
  114. printk(KERN_WARNING "nowb is deprecated; use cachepolicy=%s\n", p);
  115. early_cachepolicy(&p);
  116. }
  117. __early_param("nowb", early_nowrite);
  118. static void __init early_ecc(char **p)
  119. {
  120. if (memcmp(*p, "on", 2) == 0) {
  121. ecc_mask = PMD_PROTECTION;
  122. *p += 2;
  123. } else if (memcmp(*p, "off", 3) == 0) {
  124. ecc_mask = 0;
  125. *p += 3;
  126. }
  127. }
  128. __early_param("ecc=", early_ecc);
  129. static int __init noalign_setup(char *__unused)
  130. {
  131. cr_alignment &= ~CR_A;
  132. cr_no_alignment &= ~CR_A;
  133. set_cr(cr_alignment);
  134. return 1;
  135. }
  136. __setup("noalign", noalign_setup);
  137. #ifndef CONFIG_SMP
  138. void adjust_cr(unsigned long mask, unsigned long set)
  139. {
  140. unsigned long flags;
  141. mask &= ~CR_A;
  142. set &= mask;
  143. local_irq_save(flags);
  144. cr_no_alignment = (cr_no_alignment & ~mask) | set;
  145. cr_alignment = (cr_alignment & ~mask) | set;
  146. set_cr((get_cr() & ~mask) | set);
  147. local_irq_restore(flags);
  148. }
  149. #endif
  150. struct mem_types {
  151. unsigned int prot_pte;
  152. unsigned int prot_l1;
  153. unsigned int prot_sect;
  154. unsigned int domain;
  155. };
  156. static struct mem_types mem_types[] __initdata = {
  157. [MT_DEVICE] = {
  158. .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
  159. L_PTE_WRITE,
  160. .prot_l1 = PMD_TYPE_TABLE,
  161. .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_UNCACHED |
  162. PMD_SECT_AP_WRITE,
  163. .domain = DOMAIN_IO,
  164. },
  165. [MT_CACHECLEAN] = {
  166. .prot_sect = PMD_TYPE_SECT | PMD_BIT4,
  167. .domain = DOMAIN_KERNEL,
  168. },
  169. [MT_MINICLEAN] = {
  170. .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_MINICACHE,
  171. .domain = DOMAIN_KERNEL,
  172. },
  173. [MT_LOW_VECTORS] = {
  174. .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
  175. L_PTE_EXEC,
  176. .prot_l1 = PMD_TYPE_TABLE,
  177. .domain = DOMAIN_USER,
  178. },
  179. [MT_HIGH_VECTORS] = {
  180. .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
  181. L_PTE_USER | L_PTE_EXEC,
  182. .prot_l1 = PMD_TYPE_TABLE,
  183. .domain = DOMAIN_USER,
  184. },
  185. [MT_MEMORY] = {
  186. .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_AP_WRITE,
  187. .domain = DOMAIN_KERNEL,
  188. },
  189. [MT_ROM] = {
  190. .prot_sect = PMD_TYPE_SECT | PMD_BIT4,
  191. .domain = DOMAIN_KERNEL,
  192. },
  193. [MT_IXP2000_DEVICE] = { /* IXP2400 requires XCB=101 for on-chip I/O */
  194. .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
  195. L_PTE_WRITE,
  196. .prot_l1 = PMD_TYPE_TABLE,
  197. .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_UNCACHED |
  198. PMD_SECT_AP_WRITE | PMD_SECT_BUFFERABLE |
  199. PMD_SECT_TEX(1),
  200. .domain = DOMAIN_IO,
  201. },
  202. [MT_NONSHARED_DEVICE] = {
  203. .prot_l1 = PMD_TYPE_TABLE,
  204. .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_NONSHARED_DEV |
  205. PMD_SECT_AP_WRITE,
  206. .domain = DOMAIN_IO,
  207. }
  208. };
  209. /*
  210. * Adjust the PMD section entries according to the CPU in use.
  211. */
  212. static void __init build_mem_type_table(void)
  213. {
  214. struct cachepolicy *cp;
  215. unsigned int cr = get_cr();
  216. unsigned int user_pgprot, kern_pgprot;
  217. int cpu_arch = cpu_architecture();
  218. int i;
  219. #if defined(CONFIG_CPU_DCACHE_DISABLE)
  220. if (cachepolicy > CPOLICY_BUFFERED)
  221. cachepolicy = CPOLICY_BUFFERED;
  222. #elif defined(CONFIG_CPU_DCACHE_WRITETHROUGH)
  223. if (cachepolicy > CPOLICY_WRITETHROUGH)
  224. cachepolicy = CPOLICY_WRITETHROUGH;
  225. #endif
  226. if (cpu_arch < CPU_ARCH_ARMv5) {
  227. if (cachepolicy >= CPOLICY_WRITEALLOC)
  228. cachepolicy = CPOLICY_WRITEBACK;
  229. ecc_mask = 0;
  230. }
  231. /*
  232. * Xscale must not have PMD bit 4 set for section mappings.
  233. */
  234. if (cpu_is_xscale())
  235. for (i = 0; i < ARRAY_SIZE(mem_types); i++)
  236. mem_types[i].prot_sect &= ~PMD_BIT4;
  237. /*
  238. * ARMv5 and lower, excluding Xscale, bit 4 must be set for
  239. * page tables.
  240. */
  241. if (cpu_arch < CPU_ARCH_ARMv6 && !cpu_is_xscale())
  242. for (i = 0; i < ARRAY_SIZE(mem_types); i++)
  243. if (mem_types[i].prot_l1)
  244. mem_types[i].prot_l1 |= PMD_BIT4;
  245. cp = &cache_policies[cachepolicy];
  246. kern_pgprot = user_pgprot = cp->pte;
  247. /*
  248. * Enable CPU-specific coherency if supported.
  249. * (Only available on XSC3 at the moment.)
  250. */
  251. if (arch_is_coherent()) {
  252. if (cpu_is_xsc3()) {
  253. mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
  254. mem_types[MT_MEMORY].prot_pte |= L_PTE_SHARED;
  255. }
  256. }
  257. /*
  258. * ARMv6 and above have extended page tables.
  259. */
  260. if (cpu_arch >= CPU_ARCH_ARMv6 && (cr & CR_XP)) {
  261. /*
  262. * bit 4 becomes XN which we must clear for the
  263. * kernel memory mapping.
  264. */
  265. mem_types[MT_MEMORY].prot_sect &= ~PMD_SECT_XN;
  266. mem_types[MT_ROM].prot_sect &= ~PMD_SECT_XN;
  267. /*
  268. * Mark cache clean areas and XIP ROM read only
  269. * from SVC mode and no access from userspace.
  270. */
  271. mem_types[MT_ROM].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
  272. mem_types[MT_MINICLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
  273. mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
  274. /*
  275. * Mark the device area as "shared device"
  276. */
  277. mem_types[MT_DEVICE].prot_pte |= L_PTE_BUFFERABLE;
  278. mem_types[MT_DEVICE].prot_sect |= PMD_SECT_BUFFERED;
  279. #ifdef CONFIG_SMP
  280. /*
  281. * Mark memory with the "shared" attribute for SMP systems
  282. */
  283. user_pgprot |= L_PTE_SHARED;
  284. kern_pgprot |= L_PTE_SHARED;
  285. mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
  286. #endif
  287. }
  288. for (i = 0; i < 16; i++) {
  289. unsigned long v = pgprot_val(protection_map[i]);
  290. v = (v & ~(L_PTE_BUFFERABLE|L_PTE_CACHEABLE)) | user_pgprot;
  291. protection_map[i] = __pgprot(v);
  292. }
  293. mem_types[MT_LOW_VECTORS].prot_pte |= kern_pgprot;
  294. mem_types[MT_HIGH_VECTORS].prot_pte |= kern_pgprot;
  295. if (cpu_arch >= CPU_ARCH_ARMv5) {
  296. #ifndef CONFIG_SMP
  297. /*
  298. * Only use write-through for non-SMP systems
  299. */
  300. mem_types[MT_LOW_VECTORS].prot_pte &= ~L_PTE_BUFFERABLE;
  301. mem_types[MT_HIGH_VECTORS].prot_pte &= ~L_PTE_BUFFERABLE;
  302. #endif
  303. } else {
  304. mem_types[MT_MINICLEAN].prot_sect &= ~PMD_SECT_TEX(1);
  305. }
  306. pgprot_kernel = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG |
  307. L_PTE_DIRTY | L_PTE_WRITE |
  308. L_PTE_EXEC | kern_pgprot);
  309. mem_types[MT_LOW_VECTORS].prot_l1 |= ecc_mask;
  310. mem_types[MT_HIGH_VECTORS].prot_l1 |= ecc_mask;
  311. mem_types[MT_MEMORY].prot_sect |= ecc_mask | cp->pmd;
  312. mem_types[MT_ROM].prot_sect |= cp->pmd;
  313. switch (cp->pmd) {
  314. case PMD_SECT_WT:
  315. mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WT;
  316. break;
  317. case PMD_SECT_WB:
  318. case PMD_SECT_WBWA:
  319. mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WB;
  320. break;
  321. }
  322. printk("Memory policy: ECC %sabled, Data cache %s\n",
  323. ecc_mask ? "en" : "dis", cp->policy);
  324. }
  325. #define vectors_base() (vectors_high() ? 0xffff0000 : 0)
  326. /*
  327. * Create a SECTION PGD between VIRT and PHYS in domain
  328. * DOMAIN with protection PROT. This operates on half-
  329. * pgdir entry increments.
  330. */
  331. static inline void
  332. alloc_init_section(unsigned long virt, unsigned long phys, int prot)
  333. {
  334. pmd_t *pmdp = pmd_off_k(virt);
  335. if (virt & (1 << 20))
  336. pmdp++;
  337. *pmdp = __pmd(phys | prot);
  338. flush_pmd_entry(pmdp);
  339. }
  340. /*
  341. * Create a SUPER SECTION PGD between VIRT and PHYS with protection PROT
  342. */
  343. static inline void
  344. alloc_init_supersection(unsigned long virt, unsigned long phys, int prot)
  345. {
  346. int i;
  347. for (i = 0; i < 16; i += 1) {
  348. alloc_init_section(virt, phys, prot | PMD_SECT_SUPER);
  349. virt += (PGDIR_SIZE / 2);
  350. }
  351. }
  352. /*
  353. * Add a PAGE mapping between VIRT and PHYS in domain
  354. * DOMAIN with protection PROT. Note that due to the
  355. * way we map the PTEs, we must allocate two PTE_SIZE'd
  356. * blocks - one for the Linux pte table, and one for
  357. * the hardware pte table.
  358. */
  359. static inline void
  360. alloc_init_page(unsigned long virt, unsigned long phys, unsigned int prot_l1, pgprot_t prot)
  361. {
  362. pmd_t *pmdp = pmd_off_k(virt);
  363. pte_t *ptep;
  364. if (pmd_none(*pmdp)) {
  365. ptep = alloc_bootmem_low_pages(2 * PTRS_PER_PTE *
  366. sizeof(pte_t));
  367. __pmd_populate(pmdp, __pa(ptep) | prot_l1);
  368. }
  369. ptep = pte_offset_kernel(pmdp, virt);
  370. set_pte_ext(ptep, pfn_pte(phys >> PAGE_SHIFT, prot), 0);
  371. }
  372. /*
  373. * Create the page directory entries and any necessary
  374. * page tables for the mapping specified by `md'. We
  375. * are able to cope here with varying sizes and address
  376. * offsets, and we take full advantage of sections and
  377. * supersections.
  378. */
  379. void __init create_mapping(struct map_desc *md)
  380. {
  381. unsigned long virt, length;
  382. int prot_sect, prot_l1, domain;
  383. pgprot_t prot_pte;
  384. unsigned long off = (u32)__pfn_to_phys(md->pfn);
  385. if (md->virtual != vectors_base() && md->virtual < TASK_SIZE) {
  386. printk(KERN_WARNING "BUG: not creating mapping for "
  387. "0x%08llx at 0x%08lx in user region\n",
  388. __pfn_to_phys((u64)md->pfn), md->virtual);
  389. return;
  390. }
  391. if ((md->type == MT_DEVICE || md->type == MT_ROM) &&
  392. md->virtual >= PAGE_OFFSET && md->virtual < VMALLOC_END) {
  393. printk(KERN_WARNING "BUG: mapping for 0x%08llx at 0x%08lx "
  394. "overlaps vmalloc space\n",
  395. __pfn_to_phys((u64)md->pfn), md->virtual);
  396. }
  397. domain = mem_types[md->type].domain;
  398. prot_pte = __pgprot(mem_types[md->type].prot_pte);
  399. prot_l1 = mem_types[md->type].prot_l1 | PMD_DOMAIN(domain);
  400. prot_sect = mem_types[md->type].prot_sect | PMD_DOMAIN(domain);
  401. /*
  402. * Catch 36-bit addresses
  403. */
  404. if(md->pfn >= 0x100000) {
  405. if(domain) {
  406. printk(KERN_ERR "MM: invalid domain in supersection "
  407. "mapping for 0x%08llx at 0x%08lx\n",
  408. __pfn_to_phys((u64)md->pfn), md->virtual);
  409. return;
  410. }
  411. if((md->virtual | md->length | __pfn_to_phys(md->pfn))
  412. & ~SUPERSECTION_MASK) {
  413. printk(KERN_ERR "MM: cannot create mapping for "
  414. "0x%08llx at 0x%08lx invalid alignment\n",
  415. __pfn_to_phys((u64)md->pfn), md->virtual);
  416. return;
  417. }
  418. /*
  419. * Shift bits [35:32] of address into bits [23:20] of PMD
  420. * (See ARMv6 spec).
  421. */
  422. off |= (((md->pfn >> (32 - PAGE_SHIFT)) & 0xF) << 20);
  423. }
  424. virt = md->virtual;
  425. off -= virt;
  426. length = md->length;
  427. if (mem_types[md->type].prot_l1 == 0 &&
  428. (virt & 0xfffff || (virt + off) & 0xfffff || (virt + length) & 0xfffff)) {
  429. printk(KERN_WARNING "BUG: map for 0x%08lx at 0x%08lx can not "
  430. "be mapped using pages, ignoring.\n",
  431. __pfn_to_phys(md->pfn), md->virtual);
  432. return;
  433. }
  434. while ((virt & 0xfffff || (virt + off) & 0xfffff) && length >= PAGE_SIZE) {
  435. alloc_init_page(virt, virt + off, prot_l1, prot_pte);
  436. virt += PAGE_SIZE;
  437. length -= PAGE_SIZE;
  438. }
  439. /* N.B. ARMv6 supersections are only defined to work with domain 0.
  440. * Since domain assignments can in fact be arbitrary, the
  441. * 'domain == 0' check below is required to insure that ARMv6
  442. * supersections are only allocated for domain 0 regardless
  443. * of the actual domain assignments in use.
  444. */
  445. if ((cpu_architecture() >= CPU_ARCH_ARMv6 || cpu_is_xsc3())
  446. && domain == 0) {
  447. /*
  448. * Align to supersection boundary if !high pages.
  449. * High pages have already been checked for proper
  450. * alignment above and they will fail the SUPSERSECTION_MASK
  451. * check because of the way the address is encoded into
  452. * offset.
  453. */
  454. if (md->pfn <= 0x100000) {
  455. while ((virt & ~SUPERSECTION_MASK ||
  456. (virt + off) & ~SUPERSECTION_MASK) &&
  457. length >= (PGDIR_SIZE / 2)) {
  458. alloc_init_section(virt, virt + off, prot_sect);
  459. virt += (PGDIR_SIZE / 2);
  460. length -= (PGDIR_SIZE / 2);
  461. }
  462. }
  463. while (length >= SUPERSECTION_SIZE) {
  464. alloc_init_supersection(virt, virt + off, prot_sect);
  465. virt += SUPERSECTION_SIZE;
  466. length -= SUPERSECTION_SIZE;
  467. }
  468. }
  469. /*
  470. * A section mapping covers half a "pgdir" entry.
  471. */
  472. while (length >= (PGDIR_SIZE / 2)) {
  473. alloc_init_section(virt, virt + off, prot_sect);
  474. virt += (PGDIR_SIZE / 2);
  475. length -= (PGDIR_SIZE / 2);
  476. }
  477. while (length >= PAGE_SIZE) {
  478. alloc_init_page(virt, virt + off, prot_l1, prot_pte);
  479. virt += PAGE_SIZE;
  480. length -= PAGE_SIZE;
  481. }
  482. }
  483. /*
  484. * Create the architecture specific mappings
  485. */
  486. void __init iotable_init(struct map_desc *io_desc, int nr)
  487. {
  488. int i;
  489. for (i = 0; i < nr; i++)
  490. create_mapping(io_desc + i);
  491. }
  492. static inline void prepare_page_table(struct meminfo *mi)
  493. {
  494. unsigned long addr;
  495. /*
  496. * Clear out all the mappings below the kernel image.
  497. */
  498. for (addr = 0; addr < MODULE_START; addr += PGDIR_SIZE)
  499. pmd_clear(pmd_off_k(addr));
  500. #ifdef CONFIG_XIP_KERNEL
  501. /* The XIP kernel is mapped in the module area -- skip over it */
  502. addr = ((unsigned long)&_etext + PGDIR_SIZE - 1) & PGDIR_MASK;
  503. #endif
  504. for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
  505. pmd_clear(pmd_off_k(addr));
  506. /*
  507. * Clear out all the kernel space mappings, except for the first
  508. * memory bank, up to the end of the vmalloc region.
  509. */
  510. for (addr = __phys_to_virt(mi->bank[0].start + mi->bank[0].size);
  511. addr < VMALLOC_END; addr += PGDIR_SIZE)
  512. pmd_clear(pmd_off_k(addr));
  513. }
  514. /*
  515. * Reserve the various regions of node 0
  516. */
  517. void __init reserve_node_zero(pg_data_t *pgdat)
  518. {
  519. unsigned long res_size = 0;
  520. /*
  521. * Register the kernel text and data with bootmem.
  522. * Note that this can only be in node 0.
  523. */
  524. #ifdef CONFIG_XIP_KERNEL
  525. reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start);
  526. #else
  527. reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
  528. #endif
  529. /*
  530. * Reserve the page tables. These are already in use,
  531. * and can only be in node 0.
  532. */
  533. reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
  534. PTRS_PER_PGD * sizeof(pgd_t));
  535. /*
  536. * Hmm... This should go elsewhere, but we really really need to
  537. * stop things allocating the low memory; ideally we need a better
  538. * implementation of GFP_DMA which does not assume that DMA-able
  539. * memory starts at zero.
  540. */
  541. if (machine_is_integrator() || machine_is_cintegrator())
  542. res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
  543. /*
  544. * These should likewise go elsewhere. They pre-reserve the
  545. * screen memory region at the start of main system memory.
  546. */
  547. if (machine_is_edb7211())
  548. res_size = 0x00020000;
  549. if (machine_is_p720t())
  550. res_size = 0x00014000;
  551. /* H1940 and RX3715 need to reserve this for suspend */
  552. if (machine_is_h1940() || machine_is_rx3715()) {
  553. reserve_bootmem_node(pgdat, 0x30003000, 0x1000);
  554. reserve_bootmem_node(pgdat, 0x30081000, 0x1000);
  555. }
  556. #ifdef CONFIG_SA1111
  557. /*
  558. * Because of the SA1111 DMA bug, we want to preserve our
  559. * precious DMA-able memory...
  560. */
  561. res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
  562. #endif
  563. if (res_size)
  564. reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size);
  565. }
  566. /*
  567. * Set up device the mappings. Since we clear out the page tables for all
  568. * mappings above VMALLOC_END, we will remove any debug device mappings.
  569. * This means you have to be careful how you debug this function, or any
  570. * called function. This means you can't use any function or debugging
  571. * method which may touch any device, otherwise the kernel _will_ crash.
  572. */
  573. static void __init devicemaps_init(struct machine_desc *mdesc)
  574. {
  575. struct map_desc map;
  576. unsigned long addr;
  577. void *vectors;
  578. /*
  579. * Allocate the vector page early.
  580. */
  581. vectors = alloc_bootmem_low_pages(PAGE_SIZE);
  582. BUG_ON(!vectors);
  583. for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
  584. pmd_clear(pmd_off_k(addr));
  585. /*
  586. * Map the kernel if it is XIP.
  587. * It is always first in the modulearea.
  588. */
  589. #ifdef CONFIG_XIP_KERNEL
  590. map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
  591. map.virtual = MODULE_START;
  592. map.length = ((unsigned long)&_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
  593. map.type = MT_ROM;
  594. create_mapping(&map);
  595. #endif
  596. /*
  597. * Map the cache flushing regions.
  598. */
  599. #ifdef FLUSH_BASE
  600. map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS);
  601. map.virtual = FLUSH_BASE;
  602. map.length = SZ_1M;
  603. map.type = MT_CACHECLEAN;
  604. create_mapping(&map);
  605. #endif
  606. #ifdef FLUSH_BASE_MINICACHE
  607. map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS + SZ_1M);
  608. map.virtual = FLUSH_BASE_MINICACHE;
  609. map.length = SZ_1M;
  610. map.type = MT_MINICLEAN;
  611. create_mapping(&map);
  612. #endif
  613. /*
  614. * Create a mapping for the machine vectors at the high-vectors
  615. * location (0xffff0000). If we aren't using high-vectors, also
  616. * create a mapping at the low-vectors virtual address.
  617. */
  618. map.pfn = __phys_to_pfn(virt_to_phys(vectors));
  619. map.virtual = 0xffff0000;
  620. map.length = PAGE_SIZE;
  621. map.type = MT_HIGH_VECTORS;
  622. create_mapping(&map);
  623. if (!vectors_high()) {
  624. map.virtual = 0;
  625. map.type = MT_LOW_VECTORS;
  626. create_mapping(&map);
  627. }
  628. /*
  629. * Ask the machine support to map in the statically mapped devices.
  630. */
  631. if (mdesc->map_io)
  632. mdesc->map_io();
  633. /*
  634. * Finally flush the caches and tlb to ensure that we're in a
  635. * consistent state wrt the writebuffer. This also ensures that
  636. * any write-allocated cache lines in the vector page are written
  637. * back. After this point, we can start to touch devices again.
  638. */
  639. local_flush_tlb_all();
  640. flush_cache_all();
  641. }
  642. /*
  643. * paging_init() sets up the page tables, initialises the zone memory
  644. * maps, and sets up the zero page, bad page and bad page tables.
  645. */
  646. void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
  647. {
  648. void *zero_page;
  649. build_mem_type_table();
  650. prepare_page_table(mi);
  651. bootmem_init(mi);
  652. devicemaps_init(mdesc);
  653. top_pmd = pmd_off_k(0xffff0000);
  654. /*
  655. * allocate the zero page. Note that we count on this going ok.
  656. */
  657. zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
  658. memzero(zero_page, PAGE_SIZE);
  659. empty_zero_page = virt_to_page(zero_page);
  660. flush_dcache_page(empty_zero_page);
  661. }
  662. /*
  663. * In order to soft-boot, we need to insert a 1:1 mapping in place of
  664. * the user-mode pages. This will then ensure that we have predictable
  665. * results when turning the mmu off
  666. */
  667. void setup_mm_for_reboot(char mode)
  668. {
  669. unsigned long base_pmdval;
  670. pgd_t *pgd;
  671. int i;
  672. if (current->mm && current->mm->pgd)
  673. pgd = current->mm->pgd;
  674. else
  675. pgd = init_mm.pgd;
  676. base_pmdval = PMD_SECT_AP_WRITE | PMD_SECT_AP_READ | PMD_TYPE_SECT;
  677. if (cpu_architecture() <= CPU_ARCH_ARMv5TEJ && !cpu_is_xscale())
  678. base_pmdval |= PMD_BIT4;
  679. for (i = 0; i < FIRST_USER_PGD_NR + USER_PTRS_PER_PGD; i++, pgd++) {
  680. unsigned long pmdval = (i << PGDIR_SHIFT) | base_pmdval;
  681. pmd_t *pmd;
  682. pmd = pmd_off(pgd, i << PGDIR_SHIFT);
  683. pmd[0] = __pmd(pmdval);
  684. pmd[1] = __pmd(pmdval + (1 << (PGDIR_SHIFT - 1)));
  685. flush_pmd_entry(pmd);
  686. }
  687. }