init.c 27 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056
  1. /*
  2. * linux/arch/parisc/mm/init.c
  3. *
  4. * Copyright (C) 1995 Linus Torvalds
  5. * Copyright 1999 SuSE GmbH
  6. * changed by Philipp Rumpf
  7. * Copyright 1999 Philipp Rumpf (prumpf@tux.org)
  8. * Copyright 2004 Randolph Chung (tausq@debian.org)
  9. * Copyright 2006-2007 Helge Deller (deller@gmx.de)
  10. *
  11. */
  12. #include <linux/module.h>
  13. #include <linux/mm.h>
  14. #include <linux/bootmem.h>
  15. #include <linux/gfp.h>
  16. #include <linux/delay.h>
  17. #include <linux/init.h>
  18. #include <linux/pci.h> /* for hppa_dma_ops and pcxl_dma_ops */
  19. #include <linux/initrd.h>
  20. #include <linux/swap.h>
  21. #include <linux/unistd.h>
  22. #include <linux/nodemask.h> /* for node_online_map */
  23. #include <linux/pagemap.h> /* for release_pages and page_cache_release */
  24. #include <asm/pgalloc.h>
  25. #include <asm/pgtable.h>
  26. #include <asm/tlb.h>
  27. #include <asm/pdc_chassis.h>
  28. #include <asm/mmzone.h>
  29. #include <asm/sections.h>
  30. extern int data_start;
  31. #if PT_NLEVELS == 3
  32. /* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout
  33. * with the first pmd adjacent to the pgd and below it. gcc doesn't actually
  34. * guarantee that global objects will be laid out in memory in the same order
  35. * as the order of declaration, so put these in different sections and use
  36. * the linker script to order them. */
  37. pmd_t pmd0[PTRS_PER_PMD] __attribute__ ((__section__ (".data..vm0.pmd"), aligned(PAGE_SIZE)));
  38. #endif
  39. pgd_t swapper_pg_dir[PTRS_PER_PGD] __attribute__ ((__section__ (".data..vm0.pgd"), aligned(PAGE_SIZE)));
  40. pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __attribute__ ((__section__ (".data..vm0.pte"), aligned(PAGE_SIZE)));
  41. #ifdef CONFIG_DISCONTIGMEM
  42. struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
  43. signed char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
  44. #endif
  45. static struct resource data_resource = {
  46. .name = "Kernel data",
  47. .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
  48. };
  49. static struct resource code_resource = {
  50. .name = "Kernel code",
  51. .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
  52. };
  53. static struct resource pdcdata_resource = {
  54. .name = "PDC data (Page Zero)",
  55. .start = 0,
  56. .end = 0x9ff,
  57. .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
  58. };
  59. static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
  60. /* The following array is initialized from the firmware specific
  61. * information retrieved in kernel/inventory.c.
  62. */
  63. physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
  64. int npmem_ranges __read_mostly;
  65. #ifdef CONFIG_64BIT
  66. #define MAX_MEM (~0UL)
  67. #else /* !CONFIG_64BIT */
  68. #define MAX_MEM (3584U*1024U*1024U)
  69. #endif /* !CONFIG_64BIT */
  70. static unsigned long mem_limit __read_mostly = MAX_MEM;
  71. static void __init mem_limit_func(void)
  72. {
  73. char *cp, *end;
  74. unsigned long limit;
  75. /* We need this before __setup() functions are called */
  76. limit = MAX_MEM;
  77. for (cp = boot_command_line; *cp; ) {
  78. if (memcmp(cp, "mem=", 4) == 0) {
  79. cp += 4;
  80. limit = memparse(cp, &end);
  81. if (end != cp)
  82. break;
  83. cp = end;
  84. } else {
  85. while (*cp != ' ' && *cp)
  86. ++cp;
  87. while (*cp == ' ')
  88. ++cp;
  89. }
  90. }
  91. if (limit < mem_limit)
  92. mem_limit = limit;
  93. }
  94. #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
  95. static void __init setup_bootmem(void)
  96. {
  97. unsigned long bootmap_size;
  98. unsigned long mem_max;
  99. unsigned long bootmap_pages;
  100. unsigned long bootmap_start_pfn;
  101. unsigned long bootmap_pfn;
  102. #ifndef CONFIG_DISCONTIGMEM
  103. physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
  104. int npmem_holes;
  105. #endif
  106. int i, sysram_resource_count;
  107. disable_sr_hashing(); /* Turn off space register hashing */
  108. /*
  109. * Sort the ranges. Since the number of ranges is typically
  110. * small, and performance is not an issue here, just do
  111. * a simple insertion sort.
  112. */
  113. for (i = 1; i < npmem_ranges; i++) {
  114. int j;
  115. for (j = i; j > 0; j--) {
  116. unsigned long tmp;
  117. if (pmem_ranges[j-1].start_pfn <
  118. pmem_ranges[j].start_pfn) {
  119. break;
  120. }
  121. tmp = pmem_ranges[j-1].start_pfn;
  122. pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
  123. pmem_ranges[j].start_pfn = tmp;
  124. tmp = pmem_ranges[j-1].pages;
  125. pmem_ranges[j-1].pages = pmem_ranges[j].pages;
  126. pmem_ranges[j].pages = tmp;
  127. }
  128. }
  129. #ifndef CONFIG_DISCONTIGMEM
  130. /*
  131. * Throw out ranges that are too far apart (controlled by
  132. * MAX_GAP).
  133. */
  134. for (i = 1; i < npmem_ranges; i++) {
  135. if (pmem_ranges[i].start_pfn -
  136. (pmem_ranges[i-1].start_pfn +
  137. pmem_ranges[i-1].pages) > MAX_GAP) {
  138. npmem_ranges = i;
  139. printk("Large gap in memory detected (%ld pages). "
  140. "Consider turning on CONFIG_DISCONTIGMEM\n",
  141. pmem_ranges[i].start_pfn -
  142. (pmem_ranges[i-1].start_pfn +
  143. pmem_ranges[i-1].pages));
  144. break;
  145. }
  146. }
  147. #endif
  148. if (npmem_ranges > 1) {
  149. /* Print the memory ranges */
  150. printk(KERN_INFO "Memory Ranges:\n");
  151. for (i = 0; i < npmem_ranges; i++) {
  152. unsigned long start;
  153. unsigned long size;
  154. size = (pmem_ranges[i].pages << PAGE_SHIFT);
  155. start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
  156. printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
  157. i,start, start + (size - 1), size >> 20);
  158. }
  159. }
  160. sysram_resource_count = npmem_ranges;
  161. for (i = 0; i < sysram_resource_count; i++) {
  162. struct resource *res = &sysram_resources[i];
  163. res->name = "System RAM";
  164. res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
  165. res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
  166. res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  167. request_resource(&iomem_resource, res);
  168. }
  169. /*
  170. * For 32 bit kernels we limit the amount of memory we can
  171. * support, in order to preserve enough kernel address space
  172. * for other purposes. For 64 bit kernels we don't normally
  173. * limit the memory, but this mechanism can be used to
  174. * artificially limit the amount of memory (and it is written
  175. * to work with multiple memory ranges).
  176. */
  177. mem_limit_func(); /* check for "mem=" argument */
  178. mem_max = 0;
  179. for (i = 0; i < npmem_ranges; i++) {
  180. unsigned long rsize;
  181. rsize = pmem_ranges[i].pages << PAGE_SHIFT;
  182. if ((mem_max + rsize) > mem_limit) {
  183. printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
  184. if (mem_max == mem_limit)
  185. npmem_ranges = i;
  186. else {
  187. pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT)
  188. - (mem_max >> PAGE_SHIFT);
  189. npmem_ranges = i + 1;
  190. mem_max = mem_limit;
  191. }
  192. break;
  193. }
  194. mem_max += rsize;
  195. }
  196. printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
  197. #ifndef CONFIG_DISCONTIGMEM
  198. /* Merge the ranges, keeping track of the holes */
  199. {
  200. unsigned long end_pfn;
  201. unsigned long hole_pages;
  202. npmem_holes = 0;
  203. end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
  204. for (i = 1; i < npmem_ranges; i++) {
  205. hole_pages = pmem_ranges[i].start_pfn - end_pfn;
  206. if (hole_pages) {
  207. pmem_holes[npmem_holes].start_pfn = end_pfn;
  208. pmem_holes[npmem_holes++].pages = hole_pages;
  209. end_pfn += hole_pages;
  210. }
  211. end_pfn += pmem_ranges[i].pages;
  212. }
  213. pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
  214. npmem_ranges = 1;
  215. }
  216. #endif
  217. bootmap_pages = 0;
  218. for (i = 0; i < npmem_ranges; i++)
  219. bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
  220. bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
  221. #ifdef CONFIG_DISCONTIGMEM
  222. for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
  223. memset(NODE_DATA(i), 0, sizeof(pg_data_t));
  224. NODE_DATA(i)->bdata = &bootmem_node_data[i];
  225. }
  226. memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
  227. for (i = 0; i < npmem_ranges; i++) {
  228. node_set_state(i, N_NORMAL_MEMORY);
  229. node_set_online(i);
  230. }
  231. #endif
  232. /*
  233. * Initialize and free the full range of memory in each range.
  234. * Note that the only writing these routines do are to the bootmap,
  235. * and we've made sure to locate the bootmap properly so that they
  236. * won't be writing over anything important.
  237. */
  238. bootmap_pfn = bootmap_start_pfn;
  239. max_pfn = 0;
  240. for (i = 0; i < npmem_ranges; i++) {
  241. unsigned long start_pfn;
  242. unsigned long npages;
  243. start_pfn = pmem_ranges[i].start_pfn;
  244. npages = pmem_ranges[i].pages;
  245. bootmap_size = init_bootmem_node(NODE_DATA(i),
  246. bootmap_pfn,
  247. start_pfn,
  248. (start_pfn + npages) );
  249. free_bootmem_node(NODE_DATA(i),
  250. (start_pfn << PAGE_SHIFT),
  251. (npages << PAGE_SHIFT) );
  252. bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
  253. if ((start_pfn + npages) > max_pfn)
  254. max_pfn = start_pfn + npages;
  255. }
  256. /* IOMMU is always used to access "high mem" on those boxes
  257. * that can support enough mem that a PCI device couldn't
  258. * directly DMA to any physical addresses.
  259. * ISA DMA support will need to revisit this.
  260. */
  261. max_low_pfn = max_pfn;
  262. /* bootmap sizing messed up? */
  263. BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages);
  264. /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
  265. #define PDC_CONSOLE_IO_IODC_SIZE 32768
  266. reserve_bootmem_node(NODE_DATA(0), 0UL,
  267. (unsigned long)(PAGE0->mem_free +
  268. PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
  269. reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
  270. (unsigned long)(_end - _text), BOOTMEM_DEFAULT);
  271. reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
  272. ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
  273. BOOTMEM_DEFAULT);
  274. #ifndef CONFIG_DISCONTIGMEM
  275. /* reserve the holes */
  276. for (i = 0; i < npmem_holes; i++) {
  277. reserve_bootmem_node(NODE_DATA(0),
  278. (pmem_holes[i].start_pfn << PAGE_SHIFT),
  279. (pmem_holes[i].pages << PAGE_SHIFT),
  280. BOOTMEM_DEFAULT);
  281. }
  282. #endif
  283. #ifdef CONFIG_BLK_DEV_INITRD
  284. if (initrd_start) {
  285. printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
  286. if (__pa(initrd_start) < mem_max) {
  287. unsigned long initrd_reserve;
  288. if (__pa(initrd_end) > mem_max) {
  289. initrd_reserve = mem_max - __pa(initrd_start);
  290. } else {
  291. initrd_reserve = initrd_end - initrd_start;
  292. }
  293. initrd_below_start_ok = 1;
  294. printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
  295. reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
  296. initrd_reserve, BOOTMEM_DEFAULT);
  297. }
  298. }
  299. #endif
  300. data_resource.start = virt_to_phys(&data_start);
  301. data_resource.end = virt_to_phys(_end) - 1;
  302. code_resource.start = virt_to_phys(_text);
  303. code_resource.end = virt_to_phys(&data_start)-1;
  304. /* We don't know which region the kernel will be in, so try
  305. * all of them.
  306. */
  307. for (i = 0; i < sysram_resource_count; i++) {
  308. struct resource *res = &sysram_resources[i];
  309. request_resource(res, &code_resource);
  310. request_resource(res, &data_resource);
  311. }
  312. request_resource(&sysram_resources[0], &pdcdata_resource);
  313. }
  314. static void __init map_pages(unsigned long start_vaddr,
  315. unsigned long start_paddr, unsigned long size,
  316. pgprot_t pgprot, int force)
  317. {
  318. pgd_t *pg_dir;
  319. pmd_t *pmd;
  320. pte_t *pg_table;
  321. unsigned long end_paddr;
  322. unsigned long start_pmd;
  323. unsigned long start_pte;
  324. unsigned long tmp1;
  325. unsigned long tmp2;
  326. unsigned long address;
  327. unsigned long vaddr;
  328. unsigned long ro_start;
  329. unsigned long ro_end;
  330. unsigned long fv_addr;
  331. unsigned long gw_addr;
  332. extern const unsigned long fault_vector_20;
  333. extern void * const linux_gateway_page;
  334. ro_start = __pa((unsigned long)_text);
  335. ro_end = __pa((unsigned long)&data_start);
  336. fv_addr = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
  337. gw_addr = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
  338. end_paddr = start_paddr + size;
  339. pg_dir = pgd_offset_k(start_vaddr);
  340. #if PTRS_PER_PMD == 1
  341. start_pmd = 0;
  342. #else
  343. start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
  344. #endif
  345. start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
  346. address = start_paddr;
  347. vaddr = start_vaddr;
  348. while (address < end_paddr) {
  349. #if PTRS_PER_PMD == 1
  350. pmd = (pmd_t *)__pa(pg_dir);
  351. #else
  352. pmd = (pmd_t *)pgd_address(*pg_dir);
  353. /*
  354. * pmd is physical at this point
  355. */
  356. if (!pmd) {
  357. pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE << PMD_ORDER);
  358. pmd = (pmd_t *) __pa(pmd);
  359. }
  360. pgd_populate(NULL, pg_dir, __va(pmd));
  361. #endif
  362. pg_dir++;
  363. /* now change pmd to kernel virtual addresses */
  364. pmd = (pmd_t *)__va(pmd) + start_pmd;
  365. for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
  366. /*
  367. * pg_table is physical at this point
  368. */
  369. pg_table = (pte_t *)pmd_address(*pmd);
  370. if (!pg_table) {
  371. pg_table = (pte_t *)
  372. alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE);
  373. pg_table = (pte_t *) __pa(pg_table);
  374. }
  375. pmd_populate_kernel(NULL, pmd, __va(pg_table));
  376. /* now change pg_table to kernel virtual addresses */
  377. pg_table = (pte_t *) __va(pg_table) + start_pte;
  378. for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
  379. pte_t pte;
  380. /*
  381. * Map the fault vector writable so we can
  382. * write the HPMC checksum.
  383. */
  384. if (force)
  385. pte = __mk_pte(address, pgprot);
  386. else if (core_kernel_text(vaddr) &&
  387. address != fv_addr)
  388. pte = __mk_pte(address, PAGE_KERNEL_EXEC);
  389. else
  390. #if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
  391. if (address >= ro_start && address < ro_end
  392. && address != fv_addr
  393. && address != gw_addr)
  394. pte = __mk_pte(address, PAGE_KERNEL_RO);
  395. else
  396. #endif
  397. pte = __mk_pte(address, pgprot);
  398. if (address >= end_paddr) {
  399. if (force)
  400. break;
  401. else
  402. pte_val(pte) = 0;
  403. }
  404. set_pte(pg_table, pte);
  405. address += PAGE_SIZE;
  406. vaddr += PAGE_SIZE;
  407. }
  408. start_pte = 0;
  409. if (address >= end_paddr)
  410. break;
  411. }
  412. start_pmd = 0;
  413. }
  414. }
  415. void free_initmem(void)
  416. {
  417. unsigned long init_begin = (unsigned long)__init_begin;
  418. unsigned long init_end = (unsigned long)__init_end;
  419. /* The init text pages are marked R-X. We have to
  420. * flush the icache and mark them RW-
  421. *
  422. * This is tricky, because map_pages is in the init section.
  423. * Do a dummy remap of the data section first (the data
  424. * section is already PAGE_KERNEL) to pull in the TLB entries
  425. * for map_kernel */
  426. map_pages(init_begin, __pa(init_begin), init_end - init_begin,
  427. PAGE_KERNEL_RWX, 1);
  428. /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
  429. * map_pages */
  430. map_pages(init_begin, __pa(init_begin), init_end - init_begin,
  431. PAGE_KERNEL, 1);
  432. /* force the kernel to see the new TLB entries */
  433. __flush_tlb_range(0, init_begin, init_end);
  434. /* Attempt to catch anyone trying to execute code here
  435. * by filling the page with BRK insns.
  436. */
  437. memset((void *)init_begin, 0x00, init_end - init_begin);
  438. /* finally dump all the instructions which were cached, since the
  439. * pages are no-longer executable */
  440. flush_icache_range(init_begin, init_end);
  441. free_initmem_default(-1);
  442. /* set up a new led state on systems shipped LED State panel */
  443. pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
  444. }
  445. #ifdef CONFIG_DEBUG_RODATA
  446. void mark_rodata_ro(void)
  447. {
  448. /* rodata memory was already mapped with KERNEL_RO access rights by
  449. pagetable_init() and map_pages(). No need to do additional stuff here */
  450. printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
  451. (unsigned long)(__end_rodata - __start_rodata) >> 10);
  452. }
  453. #endif
  454. /*
  455. * Just an arbitrary offset to serve as a "hole" between mapping areas
  456. * (between top of physical memory and a potential pcxl dma mapping
  457. * area, and below the vmalloc mapping area).
  458. *
  459. * The current 32K value just means that there will be a 32K "hole"
  460. * between mapping areas. That means that any out-of-bounds memory
  461. * accesses will hopefully be caught. The vmalloc() routines leaves
  462. * a hole of 4kB between each vmalloced area for the same reason.
  463. */
  464. /* Leave room for gateway page expansion */
  465. #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
  466. #error KERNEL_MAP_START is in gateway reserved region
  467. #endif
  468. #define MAP_START (KERNEL_MAP_START)
  469. #define VM_MAP_OFFSET (32*1024)
  470. #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
  471. & ~(VM_MAP_OFFSET-1)))
  472. void *parisc_vmalloc_start __read_mostly;
  473. EXPORT_SYMBOL(parisc_vmalloc_start);
  474. #ifdef CONFIG_PA11
  475. unsigned long pcxl_dma_start __read_mostly;
  476. #endif
  477. void __init mem_init(void)
  478. {
  479. /* Do sanity checks on page table constants */
  480. BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
  481. BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
  482. BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
  483. BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
  484. > BITS_PER_LONG);
  485. high_memory = __va((max_pfn << PAGE_SHIFT));
  486. set_max_mapnr(page_to_pfn(virt_to_page(high_memory - 1)) + 1);
  487. free_all_bootmem();
  488. #ifdef CONFIG_PA11
  489. if (hppa_dma_ops == &pcxl_dma_ops) {
  490. pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
  491. parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
  492. + PCXL_DMA_MAP_SIZE);
  493. } else {
  494. pcxl_dma_start = 0;
  495. parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
  496. }
  497. #else
  498. parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
  499. #endif
  500. mem_init_print_info(NULL);
  501. #ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
  502. printk("virtual kernel memory layout:\n"
  503. " vmalloc : 0x%p - 0x%p (%4ld MB)\n"
  504. " memory : 0x%p - 0x%p (%4ld MB)\n"
  505. " .init : 0x%p - 0x%p (%4ld kB)\n"
  506. " .data : 0x%p - 0x%p (%4ld kB)\n"
  507. " .text : 0x%p - 0x%p (%4ld kB)\n",
  508. (void*)VMALLOC_START, (void*)VMALLOC_END,
  509. (VMALLOC_END - VMALLOC_START) >> 20,
  510. __va(0), high_memory,
  511. ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
  512. __init_begin, __init_end,
  513. ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
  514. _etext, _edata,
  515. ((unsigned long)_edata - (unsigned long)_etext) >> 10,
  516. _text, _etext,
  517. ((unsigned long)_etext - (unsigned long)_text) >> 10);
  518. #endif
  519. }
  520. unsigned long *empty_zero_page __read_mostly;
  521. EXPORT_SYMBOL(empty_zero_page);
  522. void show_mem(unsigned int filter)
  523. {
  524. int i,free = 0,total = 0,reserved = 0;
  525. int shared = 0, cached = 0;
  526. printk(KERN_INFO "Mem-info:\n");
  527. show_free_areas(filter);
  528. if (filter & SHOW_MEM_FILTER_PAGE_COUNT)
  529. return;
  530. #ifndef CONFIG_DISCONTIGMEM
  531. i = max_mapnr;
  532. while (i-- > 0) {
  533. total++;
  534. if (PageReserved(mem_map+i))
  535. reserved++;
  536. else if (PageSwapCache(mem_map+i))
  537. cached++;
  538. else if (!page_count(&mem_map[i]))
  539. free++;
  540. else
  541. shared += page_count(&mem_map[i]) - 1;
  542. }
  543. #else
  544. for (i = 0; i < npmem_ranges; i++) {
  545. int j;
  546. for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
  547. struct page *p;
  548. unsigned long flags;
  549. pgdat_resize_lock(NODE_DATA(i), &flags);
  550. p = nid_page_nr(i, j) - node_start_pfn(i);
  551. total++;
  552. if (PageReserved(p))
  553. reserved++;
  554. else if (PageSwapCache(p))
  555. cached++;
  556. else if (!page_count(p))
  557. free++;
  558. else
  559. shared += page_count(p) - 1;
  560. pgdat_resize_unlock(NODE_DATA(i), &flags);
  561. }
  562. }
  563. #endif
  564. printk(KERN_INFO "%d pages of RAM\n", total);
  565. printk(KERN_INFO "%d reserved pages\n", reserved);
  566. printk(KERN_INFO "%d pages shared\n", shared);
  567. printk(KERN_INFO "%d pages swap cached\n", cached);
  568. #ifdef CONFIG_DISCONTIGMEM
  569. {
  570. struct zonelist *zl;
  571. int i, j;
  572. for (i = 0; i < npmem_ranges; i++) {
  573. zl = node_zonelist(i, 0);
  574. for (j = 0; j < MAX_NR_ZONES; j++) {
  575. struct zoneref *z;
  576. struct zone *zone;
  577. printk("Zone list for zone %d on node %d: ", j, i);
  578. for_each_zone_zonelist(zone, z, zl, j)
  579. printk("[%d/%s] ", zone_to_nid(zone),
  580. zone->name);
  581. printk("\n");
  582. }
  583. }
  584. }
  585. #endif
  586. }
  587. /*
  588. * pagetable_init() sets up the page tables
  589. *
  590. * Note that gateway_init() places the Linux gateway page at page 0.
  591. * Since gateway pages cannot be dereferenced this has the desirable
  592. * side effect of trapping those pesky NULL-reference errors in the
  593. * kernel.
  594. */
  595. static void __init pagetable_init(void)
  596. {
  597. int range;
  598. /* Map each physical memory range to its kernel vaddr */
  599. for (range = 0; range < npmem_ranges; range++) {
  600. unsigned long start_paddr;
  601. unsigned long end_paddr;
  602. unsigned long size;
  603. start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
  604. end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
  605. size = pmem_ranges[range].pages << PAGE_SHIFT;
  606. map_pages((unsigned long)__va(start_paddr), start_paddr,
  607. size, PAGE_KERNEL, 0);
  608. }
  609. #ifdef CONFIG_BLK_DEV_INITRD
  610. if (initrd_end && initrd_end > mem_limit) {
  611. printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
  612. map_pages(initrd_start, __pa(initrd_start),
  613. initrd_end - initrd_start, PAGE_KERNEL, 0);
  614. }
  615. #endif
  616. empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
  617. memset(empty_zero_page, 0, PAGE_SIZE);
  618. }
  619. static void __init gateway_init(void)
  620. {
  621. unsigned long linux_gateway_page_addr;
  622. /* FIXME: This is 'const' in order to trick the compiler
  623. into not treating it as DP-relative data. */
  624. extern void * const linux_gateway_page;
  625. linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
  626. /*
  627. * Setup Linux Gateway page.
  628. *
  629. * The Linux gateway page will reside in kernel space (on virtual
  630. * page 0), so it doesn't need to be aliased into user space.
  631. */
  632. map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
  633. PAGE_SIZE, PAGE_GATEWAY, 1);
  634. }
  635. #ifdef CONFIG_HPUX
  636. void
  637. map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
  638. {
  639. pgd_t *pg_dir;
  640. pmd_t *pmd;
  641. pte_t *pg_table;
  642. unsigned long start_pmd;
  643. unsigned long start_pte;
  644. unsigned long address;
  645. unsigned long hpux_gw_page_addr;
  646. /* FIXME: This is 'const' in order to trick the compiler
  647. into not treating it as DP-relative data. */
  648. extern void * const hpux_gateway_page;
  649. hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
  650. /*
  651. * Setup HP-UX Gateway page.
  652. *
  653. * The HP-UX gateway page resides in the user address space,
  654. * so it needs to be aliased into each process.
  655. */
  656. pg_dir = pgd_offset(mm,hpux_gw_page_addr);
  657. #if PTRS_PER_PMD == 1
  658. start_pmd = 0;
  659. #else
  660. start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
  661. #endif
  662. start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
  663. address = __pa(&hpux_gateway_page);
  664. #if PTRS_PER_PMD == 1
  665. pmd = (pmd_t *)__pa(pg_dir);
  666. #else
  667. pmd = (pmd_t *) pgd_address(*pg_dir);
  668. /*
  669. * pmd is physical at this point
  670. */
  671. if (!pmd) {
  672. pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
  673. pmd = (pmd_t *) __pa(pmd);
  674. }
  675. __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
  676. #endif
  677. /* now change pmd to kernel virtual addresses */
  678. pmd = (pmd_t *)__va(pmd) + start_pmd;
  679. /*
  680. * pg_table is physical at this point
  681. */
  682. pg_table = (pte_t *) pmd_address(*pmd);
  683. if (!pg_table)
  684. pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
  685. __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
  686. /* now change pg_table to kernel virtual addresses */
  687. pg_table = (pte_t *) __va(pg_table) + start_pte;
  688. set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
  689. }
  690. EXPORT_SYMBOL(map_hpux_gateway_page);
  691. #endif
  692. void __init paging_init(void)
  693. {
  694. int i;
  695. setup_bootmem();
  696. pagetable_init();
  697. gateway_init();
  698. flush_cache_all_local(); /* start with known state */
  699. flush_tlb_all_local(NULL);
  700. for (i = 0; i < npmem_ranges; i++) {
  701. unsigned long zones_size[MAX_NR_ZONES] = { 0, };
  702. zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
  703. #ifdef CONFIG_DISCONTIGMEM
  704. /* Need to initialize the pfnnid_map before we can initialize
  705. the zone */
  706. {
  707. int j;
  708. for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
  709. j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
  710. j++) {
  711. pfnnid_map[j] = i;
  712. }
  713. }
  714. #endif
  715. free_area_init_node(i, zones_size,
  716. pmem_ranges[i].start_pfn, NULL);
  717. }
  718. }
  719. #ifdef CONFIG_PA20
  720. /*
  721. * Currently, all PA20 chips have 18 bit protection IDs, which is the
  722. * limiting factor (space ids are 32 bits).
  723. */
  724. #define NR_SPACE_IDS 262144
  725. #else
  726. /*
  727. * Currently we have a one-to-one relationship between space IDs and
  728. * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
  729. * support 15 bit protection IDs, so that is the limiting factor.
  730. * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
  731. * probably not worth the effort for a special case here.
  732. */
  733. #define NR_SPACE_IDS 32768
  734. #endif /* !CONFIG_PA20 */
  735. #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
  736. #define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
  737. static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
  738. static unsigned long dirty_space_id[SID_ARRAY_SIZE];
  739. static unsigned long space_id_index;
  740. static unsigned long free_space_ids = NR_SPACE_IDS - 1;
  741. static unsigned long dirty_space_ids = 0;
  742. static DEFINE_SPINLOCK(sid_lock);
  743. unsigned long alloc_sid(void)
  744. {
  745. unsigned long index;
  746. spin_lock(&sid_lock);
  747. if (free_space_ids == 0) {
  748. if (dirty_space_ids != 0) {
  749. spin_unlock(&sid_lock);
  750. flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
  751. spin_lock(&sid_lock);
  752. }
  753. BUG_ON(free_space_ids == 0);
  754. }
  755. free_space_ids--;
  756. index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
  757. space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
  758. space_id_index = index;
  759. spin_unlock(&sid_lock);
  760. return index << SPACEID_SHIFT;
  761. }
  762. void free_sid(unsigned long spaceid)
  763. {
  764. unsigned long index = spaceid >> SPACEID_SHIFT;
  765. unsigned long *dirty_space_offset;
  766. dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
  767. index &= (BITS_PER_LONG - 1);
  768. spin_lock(&sid_lock);
  769. BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
  770. *dirty_space_offset |= (1L << index);
  771. dirty_space_ids++;
  772. spin_unlock(&sid_lock);
  773. }
  774. #ifdef CONFIG_SMP
  775. static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
  776. {
  777. int i;
  778. /* NOTE: sid_lock must be held upon entry */
  779. *ndirtyptr = dirty_space_ids;
  780. if (dirty_space_ids != 0) {
  781. for (i = 0; i < SID_ARRAY_SIZE; i++) {
  782. dirty_array[i] = dirty_space_id[i];
  783. dirty_space_id[i] = 0;
  784. }
  785. dirty_space_ids = 0;
  786. }
  787. return;
  788. }
  789. static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
  790. {
  791. int i;
  792. /* NOTE: sid_lock must be held upon entry */
  793. if (ndirty != 0) {
  794. for (i = 0; i < SID_ARRAY_SIZE; i++) {
  795. space_id[i] ^= dirty_array[i];
  796. }
  797. free_space_ids += ndirty;
  798. space_id_index = 0;
  799. }
  800. }
  801. #else /* CONFIG_SMP */
  802. static void recycle_sids(void)
  803. {
  804. int i;
  805. /* NOTE: sid_lock must be held upon entry */
  806. if (dirty_space_ids != 0) {
  807. for (i = 0; i < SID_ARRAY_SIZE; i++) {
  808. space_id[i] ^= dirty_space_id[i];
  809. dirty_space_id[i] = 0;
  810. }
  811. free_space_ids += dirty_space_ids;
  812. dirty_space_ids = 0;
  813. space_id_index = 0;
  814. }
  815. }
  816. #endif
  817. /*
  818. * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
  819. * purged, we can safely reuse the space ids that were released but
  820. * not flushed from the tlb.
  821. */
  822. #ifdef CONFIG_SMP
  823. static unsigned long recycle_ndirty;
  824. static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
  825. static unsigned int recycle_inuse;
  826. void flush_tlb_all(void)
  827. {
  828. int do_recycle;
  829. __inc_irq_stat(irq_tlb_count);
  830. do_recycle = 0;
  831. spin_lock(&sid_lock);
  832. if (dirty_space_ids > RECYCLE_THRESHOLD) {
  833. BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
  834. get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
  835. recycle_inuse++;
  836. do_recycle++;
  837. }
  838. spin_unlock(&sid_lock);
  839. on_each_cpu(flush_tlb_all_local, NULL, 1);
  840. if (do_recycle) {
  841. spin_lock(&sid_lock);
  842. recycle_sids(recycle_ndirty,recycle_dirty_array);
  843. recycle_inuse = 0;
  844. spin_unlock(&sid_lock);
  845. }
  846. }
  847. #else
  848. void flush_tlb_all(void)
  849. {
  850. __inc_irq_stat(irq_tlb_count);
  851. spin_lock(&sid_lock);
  852. flush_tlb_all_local(NULL);
  853. recycle_sids();
  854. spin_unlock(&sid_lock);
  855. }
  856. #endif
  857. #ifdef CONFIG_BLK_DEV_INITRD
  858. void free_initrd_mem(unsigned long start, unsigned long end)
  859. {
  860. free_reserved_area((void *)start, (void *)end, -1, "initrd");
  861. }
  862. #endif