discontig.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811
  1. /*
  2. * Copyright (c) 2000, 2003 Silicon Graphics, Inc. All rights reserved.
  3. * Copyright (c) 2001 Intel Corp.
  4. * Copyright (c) 2001 Tony Luck <tony.luck@intel.com>
  5. * Copyright (c) 2002 NEC Corp.
  6. * Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com>
  7. * Copyright (c) 2004 Silicon Graphics, Inc
  8. * Russ Anderson <rja@sgi.com>
  9. * Jesse Barnes <jbarnes@sgi.com>
  10. * Jack Steiner <steiner@sgi.com>
  11. */
  12. /*
  13. * Platform initialization for Discontig Memory
  14. */
  15. #include <linux/kernel.h>
  16. #include <linux/mm.h>
  17. #include <linux/swap.h>
  18. #include <linux/bootmem.h>
  19. #include <linux/acpi.h>
  20. #include <linux/efi.h>
  21. #include <linux/nodemask.h>
  22. #include <asm/pgalloc.h>
  23. #include <asm/tlb.h>
  24. #include <asm/meminit.h>
  25. #include <asm/numa.h>
  26. #include <asm/sections.h>
  27. /*
  28. * Track per-node information needed to setup the boot memory allocator, the
  29. * per-node areas, and the real VM.
  30. */
  31. struct early_node_data {
  32. struct ia64_node_data *node_data;
  33. unsigned long pernode_addr;
  34. unsigned long pernode_size;
  35. struct bootmem_data bootmem_data;
  36. unsigned long num_physpages;
  37. unsigned long num_dma_physpages;
  38. unsigned long min_pfn;
  39. unsigned long max_pfn;
  40. };
  41. static struct early_node_data mem_data[MAX_NUMNODES] __initdata;
  42. static nodemask_t memory_less_mask __initdata;
  43. static pg_data_t *pgdat_list[MAX_NUMNODES];
  44. /*
  45. * To prevent cache aliasing effects, align per-node structures so that they
  46. * start at addresses that are strided by node number.
  47. */
  48. #define MAX_NODE_ALIGN_OFFSET (32 * 1024 * 1024)
  49. #define NODEDATA_ALIGN(addr, node) \
  50. ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + \
  51. (((node)*PERCPU_PAGE_SIZE) & (MAX_NODE_ALIGN_OFFSET - 1)))
  52. /**
  53. * build_node_maps - callback to setup bootmem structs for each node
  54. * @start: physical start of range
  55. * @len: length of range
  56. * @node: node where this range resides
  57. *
  58. * We allocate a struct bootmem_data for each piece of memory that we wish to
  59. * treat as a virtually contiguous block (i.e. each node). Each such block
  60. * must start on an %IA64_GRANULE_SIZE boundary, so we round the address down
  61. * if necessary. Any non-existent pages will simply be part of the virtual
  62. * memmap. We also update min_low_pfn and max_low_pfn here as we receive
  63. * memory ranges from the caller.
  64. */
  65. static int __init build_node_maps(unsigned long start, unsigned long len,
  66. int node)
  67. {
  68. unsigned long cstart, epfn, end = start + len;
  69. struct bootmem_data *bdp = &mem_data[node].bootmem_data;
  70. epfn = GRANULEROUNDUP(end) >> PAGE_SHIFT;
  71. cstart = GRANULEROUNDDOWN(start);
  72. if (!bdp->node_low_pfn) {
  73. bdp->node_boot_start = cstart;
  74. bdp->node_low_pfn = epfn;
  75. } else {
  76. bdp->node_boot_start = min(cstart, bdp->node_boot_start);
  77. bdp->node_low_pfn = max(epfn, bdp->node_low_pfn);
  78. }
  79. min_low_pfn = min(min_low_pfn, bdp->node_boot_start>>PAGE_SHIFT);
  80. max_low_pfn = max(max_low_pfn, bdp->node_low_pfn);
  81. return 0;
  82. }
  83. /**
  84. * early_nr_cpus_node - return number of cpus on a given node
  85. * @node: node to check
  86. *
  87. * Count the number of cpus on @node. We can't use nr_cpus_node() yet because
  88. * acpi_boot_init() (which builds the node_to_cpu_mask array) hasn't been
  89. * called yet. Note that node 0 will also count all non-existent cpus.
  90. */
  91. static int __meminit early_nr_cpus_node(int node)
  92. {
  93. int cpu, n = 0;
  94. for (cpu = 0; cpu < NR_CPUS; cpu++)
  95. if (node == node_cpuid[cpu].nid)
  96. n++;
  97. return n;
  98. }
  99. /**
  100. * compute_pernodesize - compute size of pernode data
  101. * @node: the node id.
  102. */
  103. static unsigned long __meminit compute_pernodesize(int node)
  104. {
  105. unsigned long pernodesize = 0, cpus;
  106. cpus = early_nr_cpus_node(node);
  107. pernodesize += PERCPU_PAGE_SIZE * cpus;
  108. pernodesize += node * L1_CACHE_BYTES;
  109. pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
  110. pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
  111. pernodesize = PAGE_ALIGN(pernodesize);
  112. return pernodesize;
  113. }
  114. /**
  115. * per_cpu_node_setup - setup per-cpu areas on each node
  116. * @cpu_data: per-cpu area on this node
  117. * @node: node to setup
  118. *
  119. * Copy the static per-cpu data into the region we just set aside and then
  120. * setup __per_cpu_offset for each CPU on this node. Return a pointer to
  121. * the end of the area.
  122. */
  123. static void *per_cpu_node_setup(void *cpu_data, int node)
  124. {
  125. #ifdef CONFIG_SMP
  126. int cpu;
  127. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  128. if (node == node_cpuid[cpu].nid) {
  129. memcpy(__va(cpu_data), __phys_per_cpu_start,
  130. __per_cpu_end - __per_cpu_start);
  131. __per_cpu_offset[cpu] = (char*)__va(cpu_data) -
  132. __per_cpu_start;
  133. cpu_data += PERCPU_PAGE_SIZE;
  134. }
  135. }
  136. #endif
  137. return cpu_data;
  138. }
  139. /**
  140. * fill_pernode - initialize pernode data.
  141. * @node: the node id.
  142. * @pernode: physical address of pernode data
  143. * @pernodesize: size of the pernode data
  144. */
  145. static void __init fill_pernode(int node, unsigned long pernode,
  146. unsigned long pernodesize)
  147. {
  148. void *cpu_data;
  149. int cpus = early_nr_cpus_node(node);
  150. struct bootmem_data *bdp = &mem_data[node].bootmem_data;
  151. mem_data[node].pernode_addr = pernode;
  152. mem_data[node].pernode_size = pernodesize;
  153. memset(__va(pernode), 0, pernodesize);
  154. cpu_data = (void *)pernode;
  155. pernode += PERCPU_PAGE_SIZE * cpus;
  156. pernode += node * L1_CACHE_BYTES;
  157. pgdat_list[node] = __va(pernode);
  158. pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
  159. mem_data[node].node_data = __va(pernode);
  160. pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
  161. pgdat_list[node]->bdata = bdp;
  162. pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
  163. cpu_data = per_cpu_node_setup(cpu_data, node);
  164. return;
  165. }
  166. /**
  167. * find_pernode_space - allocate memory for memory map and per-node structures
  168. * @start: physical start of range
  169. * @len: length of range
  170. * @node: node where this range resides
  171. *
  172. * This routine reserves space for the per-cpu data struct, the list of
  173. * pg_data_ts and the per-node data struct. Each node will have something like
  174. * the following in the first chunk of addr. space large enough to hold it.
  175. *
  176. * ________________________
  177. * | |
  178. * |~~~~~~~~~~~~~~~~~~~~~~~~| <-- NODEDATA_ALIGN(start, node) for the first
  179. * | PERCPU_PAGE_SIZE * | start and length big enough
  180. * | cpus_on_this_node | Node 0 will also have entries for all non-existent cpus.
  181. * |------------------------|
  182. * | local pg_data_t * |
  183. * |------------------------|
  184. * | local ia64_node_data |
  185. * |------------------------|
  186. * | ??? |
  187. * |________________________|
  188. *
  189. * Once this space has been set aside, the bootmem maps are initialized. We
  190. * could probably move the allocation of the per-cpu and ia64_node_data space
  191. * outside of this function and use alloc_bootmem_node(), but doing it here
  192. * is straightforward and we get the alignments we want so...
  193. */
  194. static int __init find_pernode_space(unsigned long start, unsigned long len,
  195. int node)
  196. {
  197. unsigned long epfn;
  198. unsigned long pernodesize = 0, pernode, pages, mapsize;
  199. struct bootmem_data *bdp = &mem_data[node].bootmem_data;
  200. epfn = (start + len) >> PAGE_SHIFT;
  201. pages = bdp->node_low_pfn - (bdp->node_boot_start >> PAGE_SHIFT);
  202. mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
  203. /*
  204. * Make sure this memory falls within this node's usable memory
  205. * since we may have thrown some away in build_maps().
  206. */
  207. if (start < bdp->node_boot_start || epfn > bdp->node_low_pfn)
  208. return 0;
  209. /* Don't setup this node's local space twice... */
  210. if (mem_data[node].pernode_addr)
  211. return 0;
  212. /*
  213. * Calculate total size needed, incl. what's necessary
  214. * for good alignment and alias prevention.
  215. */
  216. pernodesize = compute_pernodesize(node);
  217. pernode = NODEDATA_ALIGN(start, node);
  218. /* Is this range big enough for what we want to store here? */
  219. if (start + len > (pernode + pernodesize + mapsize))
  220. fill_pernode(node, pernode, pernodesize);
  221. return 0;
  222. }
  223. /**
  224. * free_node_bootmem - free bootmem allocator memory for use
  225. * @start: physical start of range
  226. * @len: length of range
  227. * @node: node where this range resides
  228. *
  229. * Simply calls the bootmem allocator to free the specified ranged from
  230. * the given pg_data_t's bdata struct. After this function has been called
  231. * for all the entries in the EFI memory map, the bootmem allocator will
  232. * be ready to service allocation requests.
  233. */
  234. static int __init free_node_bootmem(unsigned long start, unsigned long len,
  235. int node)
  236. {
  237. free_bootmem_node(pgdat_list[node], start, len);
  238. return 0;
  239. }
  240. /**
  241. * reserve_pernode_space - reserve memory for per-node space
  242. *
  243. * Reserve the space used by the bootmem maps & per-node space in the boot
  244. * allocator so that when we actually create the real mem maps we don't
  245. * use their memory.
  246. */
  247. static void __init reserve_pernode_space(void)
  248. {
  249. unsigned long base, size, pages;
  250. struct bootmem_data *bdp;
  251. int node;
  252. for_each_online_node(node) {
  253. pg_data_t *pdp = pgdat_list[node];
  254. if (node_isset(node, memory_less_mask))
  255. continue;
  256. bdp = pdp->bdata;
  257. /* First the bootmem_map itself */
  258. pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
  259. size = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
  260. base = __pa(bdp->node_bootmem_map);
  261. reserve_bootmem_node(pdp, base, size);
  262. /* Now the per-node space */
  263. size = mem_data[node].pernode_size;
  264. base = __pa(mem_data[node].pernode_addr);
  265. reserve_bootmem_node(pdp, base, size);
  266. }
  267. }
  268. static void __meminit scatter_node_data(void)
  269. {
  270. pg_data_t **dst;
  271. int node;
  272. for_each_online_node(node) {
  273. dst = LOCAL_DATA_ADDR(pgdat_list[node])->pg_data_ptrs;
  274. memcpy(dst, pgdat_list, sizeof(pgdat_list));
  275. }
  276. }
  277. /**
  278. * initialize_pernode_data - fixup per-cpu & per-node pointers
  279. *
  280. * Each node's per-node area has a copy of the global pg_data_t list, so
  281. * we copy that to each node here, as well as setting the per-cpu pointer
  282. * to the local node data structure. The active_cpus field of the per-node
  283. * structure gets setup by the platform_cpu_init() function later.
  284. */
  285. static void __init initialize_pernode_data(void)
  286. {
  287. int cpu, node;
  288. scatter_node_data();
  289. #ifdef CONFIG_SMP
  290. /* Set the node_data pointer for each per-cpu struct */
  291. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  292. node = node_cpuid[cpu].nid;
  293. per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data;
  294. }
  295. #else
  296. {
  297. struct cpuinfo_ia64 *cpu0_cpu_info;
  298. cpu = 0;
  299. node = node_cpuid[cpu].nid;
  300. cpu0_cpu_info = (struct cpuinfo_ia64 *)(__phys_per_cpu_start +
  301. ((char *)&per_cpu__cpu_info - __per_cpu_start));
  302. cpu0_cpu_info->node_data = mem_data[node].node_data;
  303. }
  304. #endif /* CONFIG_SMP */
  305. }
  306. /**
  307. * memory_less_node_alloc - * attempt to allocate memory on the best NUMA slit
  308. * node but fall back to any other node when __alloc_bootmem_node fails
  309. * for best.
  310. * @nid: node id
  311. * @pernodesize: size of this node's pernode data
  312. */
  313. static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize)
  314. {
  315. void *ptr = NULL;
  316. u8 best = 0xff;
  317. int bestnode = -1, node, anynode = 0;
  318. for_each_online_node(node) {
  319. if (node_isset(node, memory_less_mask))
  320. continue;
  321. else if (node_distance(nid, node) < best) {
  322. best = node_distance(nid, node);
  323. bestnode = node;
  324. }
  325. anynode = node;
  326. }
  327. if (bestnode == -1)
  328. bestnode = anynode;
  329. ptr = __alloc_bootmem_node(pgdat_list[bestnode], pernodesize,
  330. PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
  331. return ptr;
  332. }
  333. /**
  334. * memory_less_nodes - allocate and initialize CPU only nodes pernode
  335. * information.
  336. */
  337. static void __init memory_less_nodes(void)
  338. {
  339. unsigned long pernodesize;
  340. void *pernode;
  341. int node;
  342. for_each_node_mask(node, memory_less_mask) {
  343. pernodesize = compute_pernodesize(node);
  344. pernode = memory_less_node_alloc(node, pernodesize);
  345. fill_pernode(node, __pa(pernode), pernodesize);
  346. }
  347. return;
  348. }
  349. #ifdef CONFIG_SPARSEMEM
  350. /**
  351. * register_sparse_mem - notify SPARSEMEM that this memory range exists.
  352. * @start: physical start of range
  353. * @end: physical end of range
  354. * @arg: unused
  355. *
  356. * Simply calls SPARSEMEM to register memory section(s).
  357. */
  358. static int __init register_sparse_mem(unsigned long start, unsigned long end,
  359. void *arg)
  360. {
  361. int nid;
  362. start = __pa(start) >> PAGE_SHIFT;
  363. end = __pa(end) >> PAGE_SHIFT;
  364. nid = early_pfn_to_nid(start);
  365. memory_present(nid, start, end);
  366. return 0;
  367. }
  368. static void __init arch_sparse_init(void)
  369. {
  370. efi_memmap_walk(register_sparse_mem, NULL);
  371. sparse_init();
  372. }
  373. #else
  374. #define arch_sparse_init() do {} while (0)
  375. #endif
  376. /**
  377. * find_memory - walk the EFI memory map and setup the bootmem allocator
  378. *
  379. * Called early in boot to setup the bootmem allocator, and to
  380. * allocate the per-cpu and per-node structures.
  381. */
  382. void __init find_memory(void)
  383. {
  384. int node;
  385. reserve_memory();
  386. if (num_online_nodes() == 0) {
  387. printk(KERN_ERR "node info missing!\n");
  388. node_set_online(0);
  389. }
  390. nodes_or(memory_less_mask, memory_less_mask, node_online_map);
  391. min_low_pfn = -1;
  392. max_low_pfn = 0;
  393. /* These actually end up getting called by call_pernode_memory() */
  394. efi_memmap_walk(filter_rsvd_memory, build_node_maps);
  395. efi_memmap_walk(filter_rsvd_memory, find_pernode_space);
  396. for_each_online_node(node)
  397. if (mem_data[node].bootmem_data.node_low_pfn) {
  398. node_clear(node, memory_less_mask);
  399. mem_data[node].min_pfn = ~0UL;
  400. }
  401. /*
  402. * Initialize the boot memory maps in reverse order since that's
  403. * what the bootmem allocator expects
  404. */
  405. for (node = MAX_NUMNODES - 1; node >= 0; node--) {
  406. unsigned long pernode, pernodesize, map;
  407. struct bootmem_data *bdp;
  408. if (!node_online(node))
  409. continue;
  410. else if (node_isset(node, memory_less_mask))
  411. continue;
  412. bdp = &mem_data[node].bootmem_data;
  413. pernode = mem_data[node].pernode_addr;
  414. pernodesize = mem_data[node].pernode_size;
  415. map = pernode + pernodesize;
  416. init_bootmem_node(pgdat_list[node],
  417. map>>PAGE_SHIFT,
  418. bdp->node_boot_start>>PAGE_SHIFT,
  419. bdp->node_low_pfn);
  420. }
  421. efi_memmap_walk(filter_rsvd_memory, free_node_bootmem);
  422. reserve_pernode_space();
  423. memory_less_nodes();
  424. initialize_pernode_data();
  425. max_pfn = max_low_pfn;
  426. find_initrd();
  427. }
  428. #ifdef CONFIG_SMP
  429. /**
  430. * per_cpu_init - setup per-cpu variables
  431. *
  432. * find_pernode_space() does most of this already, we just need to set
  433. * local_per_cpu_offset
  434. */
  435. void __cpuinit *per_cpu_init(void)
  436. {
  437. int cpu;
  438. static int first_time = 1;
  439. if (smp_processor_id() != 0)
  440. return __per_cpu_start + __per_cpu_offset[smp_processor_id()];
  441. if (first_time) {
  442. first_time = 0;
  443. for (cpu = 0; cpu < NR_CPUS; cpu++)
  444. per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
  445. }
  446. return __per_cpu_start + __per_cpu_offset[smp_processor_id()];
  447. }
  448. #endif /* CONFIG_SMP */
  449. #ifdef CONFIG_VIRTUAL_MEM_MAP
  450. static inline int find_next_valid_pfn_for_pgdat(pg_data_t *pgdat, int i)
  451. {
  452. unsigned long end_address, hole_next_pfn;
  453. unsigned long stop_address;
  454. end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
  455. end_address = PAGE_ALIGN(end_address);
  456. stop_address = (unsigned long) &vmem_map[
  457. pgdat->node_start_pfn + pgdat->node_spanned_pages];
  458. do {
  459. pgd_t *pgd;
  460. pud_t *pud;
  461. pmd_t *pmd;
  462. pte_t *pte;
  463. pgd = pgd_offset_k(end_address);
  464. if (pgd_none(*pgd)) {
  465. end_address += PGDIR_SIZE;
  466. continue;
  467. }
  468. pud = pud_offset(pgd, end_address);
  469. if (pud_none(*pud)) {
  470. end_address += PUD_SIZE;
  471. continue;
  472. }
  473. pmd = pmd_offset(pud, end_address);
  474. if (pmd_none(*pmd)) {
  475. end_address += PMD_SIZE;
  476. continue;
  477. }
  478. pte = pte_offset_kernel(pmd, end_address);
  479. retry_pte:
  480. if (pte_none(*pte)) {
  481. end_address += PAGE_SIZE;
  482. pte++;
  483. if ((end_address < stop_address) &&
  484. (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
  485. goto retry_pte;
  486. continue;
  487. }
  488. /* Found next valid vmem_map page */
  489. break;
  490. } while (end_address < stop_address);
  491. end_address = min(end_address, stop_address);
  492. end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
  493. hole_next_pfn = end_address / sizeof(struct page);
  494. return hole_next_pfn - pgdat->node_start_pfn;
  495. }
  496. #else
  497. static inline int find_next_valid_pfn_for_pgdat(pg_data_t *pgdat, int i)
  498. {
  499. return i + 1;
  500. }
  501. #endif
  502. /**
  503. * show_mem - give short summary of memory stats
  504. *
  505. * Shows a simple page count of reserved and used pages in the system.
  506. * For discontig machines, it does this on a per-pgdat basis.
  507. */
  508. void show_mem(void)
  509. {
  510. int i, total_reserved = 0;
  511. int total_shared = 0, total_cached = 0;
  512. unsigned long total_present = 0;
  513. pg_data_t *pgdat;
  514. printk("Mem-info:\n");
  515. show_free_areas();
  516. printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
  517. for_each_online_pgdat(pgdat) {
  518. unsigned long present;
  519. unsigned long flags;
  520. int shared = 0, cached = 0, reserved = 0;
  521. printk("Node ID: %d\n", pgdat->node_id);
  522. pgdat_resize_lock(pgdat, &flags);
  523. present = pgdat->node_present_pages;
  524. for(i = 0; i < pgdat->node_spanned_pages; i++) {
  525. struct page *page;
  526. if (pfn_valid(pgdat->node_start_pfn + i))
  527. page = pfn_to_page(pgdat->node_start_pfn + i);
  528. else {
  529. i = find_next_valid_pfn_for_pgdat(pgdat, i) - 1;
  530. continue;
  531. }
  532. if (PageReserved(page))
  533. reserved++;
  534. else if (PageSwapCache(page))
  535. cached++;
  536. else if (page_count(page))
  537. shared += page_count(page)-1;
  538. }
  539. pgdat_resize_unlock(pgdat, &flags);
  540. total_present += present;
  541. total_reserved += reserved;
  542. total_cached += cached;
  543. total_shared += shared;
  544. printk("\t%ld pages of RAM\n", present);
  545. printk("\t%d reserved pages\n", reserved);
  546. printk("\t%d pages shared\n", shared);
  547. printk("\t%d pages swap cached\n", cached);
  548. }
  549. printk("%ld pages of RAM\n", total_present);
  550. printk("%d reserved pages\n", total_reserved);
  551. printk("%d pages shared\n", total_shared);
  552. printk("%d pages swap cached\n", total_cached);
  553. printk("Total of %ld pages in page table cache\n",
  554. pgtable_quicklist_total_size());
  555. printk("%d free buffer pages\n", nr_free_buffer_pages());
  556. }
  557. /**
  558. * call_pernode_memory - use SRAT to call callback functions with node info
  559. * @start: physical start of range
  560. * @len: length of range
  561. * @arg: function to call for each range
  562. *
  563. * efi_memmap_walk() knows nothing about layout of memory across nodes. Find
  564. * out to which node a block of memory belongs. Ignore memory that we cannot
  565. * identify, and split blocks that run across multiple nodes.
  566. *
  567. * Take this opportunity to round the start address up and the end address
  568. * down to page boundaries.
  569. */
  570. void call_pernode_memory(unsigned long start, unsigned long len, void *arg)
  571. {
  572. unsigned long rs, re, end = start + len;
  573. void (*func)(unsigned long, unsigned long, int);
  574. int i;
  575. start = PAGE_ALIGN(start);
  576. end &= PAGE_MASK;
  577. if (start >= end)
  578. return;
  579. func = arg;
  580. if (!num_node_memblks) {
  581. /* No SRAT table, so assume one node (node 0) */
  582. if (start < end)
  583. (*func)(start, end - start, 0);
  584. return;
  585. }
  586. for (i = 0; i < num_node_memblks; i++) {
  587. rs = max(start, node_memblk[i].start_paddr);
  588. re = min(end, node_memblk[i].start_paddr +
  589. node_memblk[i].size);
  590. if (rs < re)
  591. (*func)(rs, re - rs, node_memblk[i].nid);
  592. if (re == end)
  593. break;
  594. }
  595. }
  596. /**
  597. * count_node_pages - callback to build per-node memory info structures
  598. * @start: physical start of range
  599. * @len: length of range
  600. * @node: node where this range resides
  601. *
  602. * Each node has it's own number of physical pages, DMAable pages, start, and
  603. * end page frame number. This routine will be called by call_pernode_memory()
  604. * for each piece of usable memory and will setup these values for each node.
  605. * Very similar to build_maps().
  606. */
  607. static __init int count_node_pages(unsigned long start, unsigned long len, int node)
  608. {
  609. unsigned long end = start + len;
  610. mem_data[node].num_physpages += len >> PAGE_SHIFT;
  611. if (start <= __pa(MAX_DMA_ADDRESS))
  612. mem_data[node].num_dma_physpages +=
  613. (min(end, __pa(MAX_DMA_ADDRESS)) - start) >>PAGE_SHIFT;
  614. start = GRANULEROUNDDOWN(start);
  615. start = ORDERROUNDDOWN(start);
  616. end = GRANULEROUNDUP(end);
  617. mem_data[node].max_pfn = max(mem_data[node].max_pfn,
  618. end >> PAGE_SHIFT);
  619. mem_data[node].min_pfn = min(mem_data[node].min_pfn,
  620. start >> PAGE_SHIFT);
  621. return 0;
  622. }
  623. /**
  624. * paging_init - setup page tables
  625. *
  626. * paging_init() sets up the page tables for each node of the system and frees
  627. * the bootmem allocator memory for general use.
  628. */
  629. void __init paging_init(void)
  630. {
  631. unsigned long max_dma;
  632. unsigned long zones_size[MAX_NR_ZONES];
  633. unsigned long zholes_size[MAX_NR_ZONES];
  634. unsigned long pfn_offset = 0;
  635. int node;
  636. max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
  637. arch_sparse_init();
  638. efi_memmap_walk(filter_rsvd_memory, count_node_pages);
  639. #ifdef CONFIG_VIRTUAL_MEM_MAP
  640. vmalloc_end -= PAGE_ALIGN(max_low_pfn * sizeof(struct page));
  641. vmem_map = (struct page *) vmalloc_end;
  642. efi_memmap_walk(create_mem_map_page_table, NULL);
  643. printk("Virtual mem_map starts at 0x%p\n", vmem_map);
  644. #endif
  645. for_each_online_node(node) {
  646. memset(zones_size, 0, sizeof(zones_size));
  647. memset(zholes_size, 0, sizeof(zholes_size));
  648. num_physpages += mem_data[node].num_physpages;
  649. if (mem_data[node].min_pfn >= max_dma) {
  650. /* All of this node's memory is above ZONE_DMA */
  651. zones_size[ZONE_NORMAL] = mem_data[node].max_pfn -
  652. mem_data[node].min_pfn;
  653. zholes_size[ZONE_NORMAL] = mem_data[node].max_pfn -
  654. mem_data[node].min_pfn -
  655. mem_data[node].num_physpages;
  656. } else if (mem_data[node].max_pfn < max_dma) {
  657. /* All of this node's memory is in ZONE_DMA */
  658. zones_size[ZONE_DMA] = mem_data[node].max_pfn -
  659. mem_data[node].min_pfn;
  660. zholes_size[ZONE_DMA] = mem_data[node].max_pfn -
  661. mem_data[node].min_pfn -
  662. mem_data[node].num_dma_physpages;
  663. } else {
  664. /* This node has memory in both zones */
  665. zones_size[ZONE_DMA] = max_dma -
  666. mem_data[node].min_pfn;
  667. zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] -
  668. mem_data[node].num_dma_physpages;
  669. zones_size[ZONE_NORMAL] = mem_data[node].max_pfn -
  670. max_dma;
  671. zholes_size[ZONE_NORMAL] = zones_size[ZONE_NORMAL] -
  672. (mem_data[node].num_physpages -
  673. mem_data[node].num_dma_physpages);
  674. }
  675. pfn_offset = mem_data[node].min_pfn;
  676. #ifdef CONFIG_VIRTUAL_MEM_MAP
  677. NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset;
  678. #endif
  679. free_area_init_node(node, NODE_DATA(node), zones_size,
  680. pfn_offset, zholes_size);
  681. }
  682. zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
  683. }
  684. pg_data_t *arch_alloc_nodedata(int nid)
  685. {
  686. unsigned long size = compute_pernodesize(nid);
  687. return kzalloc(size, GFP_KERNEL);
  688. }
  689. void arch_free_nodedata(pg_data_t *pgdat)
  690. {
  691. kfree(pgdat);
  692. }
  693. void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat)
  694. {
  695. pgdat_list[update_node] = update_pgdat;
  696. scatter_node_data();
  697. }