balloon.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807
  1. /******************************************************************************
  2. * Xen balloon driver - enables returning/claiming memory to/from Xen.
  3. *
  4. * Copyright (c) 2003, B Dragovic
  5. * Copyright (c) 2003-2004, M Williamson, K Fraser
  6. * Copyright (c) 2005 Dan M. Smith, IBM Corporation
  7. * Copyright (c) 2010 Daniel Kiper
  8. *
  9. * Memory hotplug support was written by Daniel Kiper. Work on
  10. * it was sponsored by Google under Google Summer of Code 2010
  11. * program. Jeremy Fitzhardinge from Citrix was the mentor for
  12. * this project.
  13. *
  14. * This program is free software; you can redistribute it and/or
  15. * modify it under the terms of the GNU General Public License version 2
  16. * as published by the Free Software Foundation; or, when distributed
  17. * separately from the Linux kernel or incorporated into other
  18. * software packages, subject to the following license:
  19. *
  20. * Permission is hereby granted, free of charge, to any person obtaining a copy
  21. * of this source file (the "Software"), to deal in the Software without
  22. * restriction, including without limitation the rights to use, copy, modify,
  23. * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  24. * and to permit persons to whom the Software is furnished to do so, subject to
  25. * the following conditions:
  26. *
  27. * The above copyright notice and this permission notice shall be included in
  28. * all copies or substantial portions of the Software.
  29. *
  30. * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  31. * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  32. * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  33. * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  34. * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  35. * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  36. * IN THE SOFTWARE.
  37. */
  38. #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  39. #include <linux/cpu.h>
  40. #include <linux/kernel.h>
  41. #include <linux/sched.h>
  42. #include <linux/cred.h>
  43. #include <linux/errno.h>
  44. #include <linux/freezer.h>
  45. #include <linux/kthread.h>
  46. #include <linux/mm.h>
  47. #include <linux/memblock.h>
  48. #include <linux/pagemap.h>
  49. #include <linux/highmem.h>
  50. #include <linux/mutex.h>
  51. #include <linux/list.h>
  52. #include <linux/gfp.h>
  53. #include <linux/notifier.h>
  54. #include <linux/memory.h>
  55. #include <linux/memory_hotplug.h>
  56. #include <linux/percpu-defs.h>
  57. #include <linux/slab.h>
  58. #include <linux/sysctl.h>
  59. #include <linux/moduleparam.h>
  60. #include <asm/page.h>
  61. #include <asm/tlb.h>
  62. #include <asm/xen/hypervisor.h>
  63. #include <asm/xen/hypercall.h>
  64. #include <xen/xen.h>
  65. #include <xen/interface/xen.h>
  66. #include <xen/interface/memory.h>
  67. #include <xen/balloon.h>
  68. #include <xen/features.h>
  69. #include <xen/page.h>
  70. #include <xen/mem-reservation.h>
  71. #undef MODULE_PARAM_PREFIX
  72. #define MODULE_PARAM_PREFIX "xen."
  73. static uint __read_mostly balloon_boot_timeout = 180;
  74. module_param(balloon_boot_timeout, uint, 0444);
  75. static int xen_hotplug_unpopulated;
  76. #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
  77. static struct ctl_table balloon_table[] = {
  78. {
  79. .procname = "hotplug_unpopulated",
  80. .data = &xen_hotplug_unpopulated,
  81. .maxlen = sizeof(int),
  82. .mode = 0644,
  83. .proc_handler = proc_dointvec_minmax,
  84. .extra1 = SYSCTL_ZERO,
  85. .extra2 = SYSCTL_ONE,
  86. },
  87. { }
  88. };
  89. static struct ctl_table balloon_root[] = {
  90. {
  91. .procname = "balloon",
  92. .mode = 0555,
  93. .child = balloon_table,
  94. },
  95. { }
  96. };
  97. static struct ctl_table xen_root[] = {
  98. {
  99. .procname = "xen",
  100. .mode = 0555,
  101. .child = balloon_root,
  102. },
  103. { }
  104. };
  105. #endif
  106. /*
  107. * Use one extent per PAGE_SIZE to avoid to break down the page into
  108. * multiple frame.
  109. */
  110. #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
  111. /*
  112. * balloon_thread() state:
  113. *
  114. * BP_DONE: done or nothing to do,
  115. * BP_WAIT: wait to be rescheduled,
  116. * BP_EAGAIN: error, go to sleep,
  117. * BP_ECANCELED: error, balloon operation canceled.
  118. */
  119. static enum bp_state {
  120. BP_DONE,
  121. BP_WAIT,
  122. BP_EAGAIN,
  123. BP_ECANCELED
  124. } balloon_state = BP_DONE;
  125. /* Main waiting point for xen-balloon thread. */
  126. static DECLARE_WAIT_QUEUE_HEAD(balloon_thread_wq);
  127. static DEFINE_MUTEX(balloon_mutex);
  128. struct balloon_stats balloon_stats;
  129. EXPORT_SYMBOL_GPL(balloon_stats);
  130. /* We increase/decrease in batches which fit in a page */
  131. static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
  132. /* List of ballooned pages, threaded through the mem_map array. */
  133. static LIST_HEAD(ballooned_pages);
  134. static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
  135. /* When ballooning out (allocating memory to return to Xen) we don't really
  136. want the kernel to try too hard since that can trigger the oom killer. */
  137. #define GFP_BALLOON \
  138. (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
  139. /* balloon_append: add the given page to the balloon. */
  140. static void balloon_append(struct page *page)
  141. {
  142. __SetPageOffline(page);
  143. /* Lowmem is re-populated first, so highmem pages go at list tail. */
  144. if (PageHighMem(page)) {
  145. list_add_tail(&page->lru, &ballooned_pages);
  146. balloon_stats.balloon_high++;
  147. } else {
  148. list_add(&page->lru, &ballooned_pages);
  149. balloon_stats.balloon_low++;
  150. }
  151. wake_up(&balloon_wq);
  152. }
  153. /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
  154. static struct page *balloon_retrieve(bool require_lowmem)
  155. {
  156. struct page *page;
  157. if (list_empty(&ballooned_pages))
  158. return NULL;
  159. page = list_entry(ballooned_pages.next, struct page, lru);
  160. if (require_lowmem && PageHighMem(page))
  161. return NULL;
  162. list_del(&page->lru);
  163. if (PageHighMem(page))
  164. balloon_stats.balloon_high--;
  165. else
  166. balloon_stats.balloon_low--;
  167. __ClearPageOffline(page);
  168. return page;
  169. }
  170. static struct page *balloon_next_page(struct page *page)
  171. {
  172. struct list_head *next = page->lru.next;
  173. if (next == &ballooned_pages)
  174. return NULL;
  175. return list_entry(next, struct page, lru);
  176. }
  177. static void update_schedule(void)
  178. {
  179. if (balloon_state == BP_WAIT || balloon_state == BP_ECANCELED)
  180. return;
  181. if (balloon_state == BP_DONE) {
  182. balloon_stats.schedule_delay = 1;
  183. balloon_stats.retry_count = 1;
  184. return;
  185. }
  186. ++balloon_stats.retry_count;
  187. if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
  188. balloon_stats.retry_count > balloon_stats.max_retry_count) {
  189. balloon_stats.schedule_delay = 1;
  190. balloon_stats.retry_count = 1;
  191. balloon_state = BP_ECANCELED;
  192. return;
  193. }
  194. balloon_stats.schedule_delay <<= 1;
  195. if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
  196. balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
  197. balloon_state = BP_EAGAIN;
  198. }
  199. #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
  200. static void release_memory_resource(struct resource *resource)
  201. {
  202. if (!resource)
  203. return;
  204. /*
  205. * No need to reset region to identity mapped since we now
  206. * know that no I/O can be in this region
  207. */
  208. release_resource(resource);
  209. kfree(resource);
  210. }
  211. static struct resource *additional_memory_resource(phys_addr_t size)
  212. {
  213. struct resource *res;
  214. int ret;
  215. res = kzalloc(sizeof(*res), GFP_KERNEL);
  216. if (!res)
  217. return NULL;
  218. res->name = "System RAM";
  219. res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
  220. ret = allocate_resource(&iomem_resource, res,
  221. size, 0, -1,
  222. PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
  223. if (ret < 0) {
  224. pr_err("Cannot allocate new System RAM resource\n");
  225. kfree(res);
  226. return NULL;
  227. }
  228. return res;
  229. }
  230. static enum bp_state reserve_additional_memory(void)
  231. {
  232. long credit;
  233. struct resource *resource;
  234. int nid, rc;
  235. unsigned long balloon_hotplug;
  236. credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
  237. - balloon_stats.total_pages;
  238. /*
  239. * Already hotplugged enough pages? Wait for them to be
  240. * onlined.
  241. */
  242. if (credit <= 0)
  243. return BP_WAIT;
  244. balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
  245. resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
  246. if (!resource)
  247. goto err;
  248. nid = memory_add_physaddr_to_nid(resource->start);
  249. #ifdef CONFIG_XEN_HAVE_PVMMU
  250. /*
  251. * We don't support PV MMU when Linux and Xen is using
  252. * different page granularity.
  253. */
  254. BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
  255. /*
  256. * add_memory() will build page tables for the new memory so
  257. * the p2m must contain invalid entries so the correct
  258. * non-present PTEs will be written.
  259. *
  260. * If a failure occurs, the original (identity) p2m entries
  261. * are not restored since this region is now known not to
  262. * conflict with any devices.
  263. */
  264. if (!xen_feature(XENFEAT_auto_translated_physmap)) {
  265. unsigned long pfn, i;
  266. pfn = PFN_DOWN(resource->start);
  267. for (i = 0; i < balloon_hotplug; i++) {
  268. if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
  269. pr_warn("set_phys_to_machine() failed, no memory added\n");
  270. goto err;
  271. }
  272. }
  273. }
  274. #endif
  275. /*
  276. * add_memory_resource() will call online_pages() which in its turn
  277. * will call xen_online_page() callback causing deadlock if we don't
  278. * release balloon_mutex here. Unlocking here is safe because the
  279. * callers drop the mutex before trying again.
  280. */
  281. mutex_unlock(&balloon_mutex);
  282. /* add_memory_resource() requires the device_hotplug lock */
  283. lock_device_hotplug();
  284. rc = add_memory_resource(nid, resource, MEMHP_MERGE_RESOURCE);
  285. unlock_device_hotplug();
  286. mutex_lock(&balloon_mutex);
  287. if (rc) {
  288. pr_warn("Cannot add additional memory (%i)\n", rc);
  289. goto err;
  290. }
  291. balloon_stats.total_pages += balloon_hotplug;
  292. return BP_WAIT;
  293. err:
  294. release_memory_resource(resource);
  295. return BP_ECANCELED;
  296. }
  297. static void xen_online_page(struct page *page, unsigned int order)
  298. {
  299. unsigned long i, size = (1 << order);
  300. unsigned long start_pfn = page_to_pfn(page);
  301. struct page *p;
  302. pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn);
  303. mutex_lock(&balloon_mutex);
  304. for (i = 0; i < size; i++) {
  305. p = pfn_to_page(start_pfn + i);
  306. balloon_append(p);
  307. }
  308. mutex_unlock(&balloon_mutex);
  309. }
  310. static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
  311. {
  312. if (val == MEM_ONLINE)
  313. wake_up(&balloon_thread_wq);
  314. return NOTIFY_OK;
  315. }
  316. static struct notifier_block xen_memory_nb = {
  317. .notifier_call = xen_memory_notifier,
  318. .priority = 0
  319. };
  320. #else
  321. static enum bp_state reserve_additional_memory(void)
  322. {
  323. balloon_stats.target_pages = balloon_stats.current_pages +
  324. balloon_stats.target_unpopulated;
  325. return BP_ECANCELED;
  326. }
  327. #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
  328. static long current_credit(void)
  329. {
  330. return balloon_stats.target_pages - balloon_stats.current_pages;
  331. }
  332. static bool balloon_is_inflated(void)
  333. {
  334. return balloon_stats.balloon_low || balloon_stats.balloon_high;
  335. }
  336. static enum bp_state increase_reservation(unsigned long nr_pages)
  337. {
  338. int rc;
  339. unsigned long i;
  340. struct page *page;
  341. if (nr_pages > ARRAY_SIZE(frame_list))
  342. nr_pages = ARRAY_SIZE(frame_list);
  343. page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
  344. for (i = 0; i < nr_pages; i++) {
  345. if (!page) {
  346. nr_pages = i;
  347. break;
  348. }
  349. frame_list[i] = page_to_xen_pfn(page);
  350. page = balloon_next_page(page);
  351. }
  352. rc = xenmem_reservation_increase(nr_pages, frame_list);
  353. if (rc <= 0)
  354. return BP_EAGAIN;
  355. for (i = 0; i < rc; i++) {
  356. page = balloon_retrieve(false);
  357. BUG_ON(page == NULL);
  358. xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]);
  359. /* Relinquish the page back to the allocator. */
  360. free_reserved_page(page);
  361. }
  362. balloon_stats.current_pages += rc;
  363. return BP_DONE;
  364. }
  365. static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
  366. {
  367. enum bp_state state = BP_DONE;
  368. unsigned long i;
  369. struct page *page, *tmp;
  370. int ret;
  371. LIST_HEAD(pages);
  372. if (nr_pages > ARRAY_SIZE(frame_list))
  373. nr_pages = ARRAY_SIZE(frame_list);
  374. for (i = 0; i < nr_pages; i++) {
  375. page = alloc_page(gfp);
  376. if (page == NULL) {
  377. nr_pages = i;
  378. state = BP_EAGAIN;
  379. break;
  380. }
  381. adjust_managed_page_count(page, -1);
  382. xenmem_reservation_scrub_page(page);
  383. list_add(&page->lru, &pages);
  384. }
  385. /*
  386. * Ensure that ballooned highmem pages don't have kmaps.
  387. *
  388. * Do this before changing the p2m as kmap_flush_unused()
  389. * reads PTEs to obtain pages (and hence needs the original
  390. * p2m entry).
  391. */
  392. kmap_flush_unused();
  393. /*
  394. * Setup the frame, update direct mapping, invalidate P2M,
  395. * and add to balloon.
  396. */
  397. i = 0;
  398. list_for_each_entry_safe(page, tmp, &pages, lru) {
  399. frame_list[i++] = xen_page_to_gfn(page);
  400. xenmem_reservation_va_mapping_reset(1, &page);
  401. list_del(&page->lru);
  402. balloon_append(page);
  403. }
  404. flush_tlb_all();
  405. ret = xenmem_reservation_decrease(nr_pages, frame_list);
  406. BUG_ON(ret != nr_pages);
  407. balloon_stats.current_pages -= nr_pages;
  408. return state;
  409. }
  410. /*
  411. * Stop waiting if either state is BP_DONE and ballooning action is
  412. * needed, or if the credit has changed while state is not BP_DONE.
  413. */
  414. static bool balloon_thread_cond(long credit)
  415. {
  416. if (balloon_state == BP_DONE)
  417. credit = 0;
  418. return current_credit() != credit || kthread_should_stop();
  419. }
  420. /*
  421. * As this is a kthread it is guaranteed to run as a single instance only.
  422. * We may of course race updates of the target counts (which are protected
  423. * by the balloon lock), or with changes to the Xen hard limit, but we will
  424. * recover from these in time.
  425. */
  426. static int balloon_thread(void *unused)
  427. {
  428. long credit;
  429. unsigned long timeout;
  430. set_freezable();
  431. for (;;) {
  432. switch (balloon_state) {
  433. case BP_DONE:
  434. case BP_ECANCELED:
  435. timeout = 3600 * HZ;
  436. break;
  437. case BP_EAGAIN:
  438. timeout = balloon_stats.schedule_delay * HZ;
  439. break;
  440. case BP_WAIT:
  441. timeout = HZ;
  442. break;
  443. }
  444. credit = current_credit();
  445. wait_event_freezable_timeout(balloon_thread_wq,
  446. balloon_thread_cond(credit), timeout);
  447. if (kthread_should_stop())
  448. return 0;
  449. mutex_lock(&balloon_mutex);
  450. credit = current_credit();
  451. if (credit > 0) {
  452. if (balloon_is_inflated())
  453. balloon_state = increase_reservation(credit);
  454. else
  455. balloon_state = reserve_additional_memory();
  456. }
  457. if (credit < 0) {
  458. long n_pages;
  459. n_pages = min(-credit, si_mem_available());
  460. balloon_state = decrease_reservation(n_pages,
  461. GFP_BALLOON);
  462. if (balloon_state == BP_DONE && n_pages != -credit &&
  463. n_pages < totalreserve_pages)
  464. balloon_state = BP_EAGAIN;
  465. }
  466. update_schedule();
  467. mutex_unlock(&balloon_mutex);
  468. cond_resched();
  469. }
  470. }
  471. /* Resets the Xen limit, sets new target, and kicks off processing. */
  472. void balloon_set_new_target(unsigned long target)
  473. {
  474. /* No need for lock. Not read-modify-write updates. */
  475. balloon_stats.target_pages = target;
  476. wake_up(&balloon_thread_wq);
  477. }
  478. EXPORT_SYMBOL_GPL(balloon_set_new_target);
  479. static int add_ballooned_pages(int nr_pages)
  480. {
  481. enum bp_state st;
  482. if (xen_hotplug_unpopulated) {
  483. st = reserve_additional_memory();
  484. if (st != BP_ECANCELED) {
  485. int rc;
  486. mutex_unlock(&balloon_mutex);
  487. rc = wait_event_interruptible(balloon_wq,
  488. !list_empty(&ballooned_pages));
  489. mutex_lock(&balloon_mutex);
  490. return rc ? -ENOMEM : 0;
  491. }
  492. }
  493. if (si_mem_available() < nr_pages)
  494. return -ENOMEM;
  495. st = decrease_reservation(nr_pages, GFP_USER);
  496. if (st != BP_DONE)
  497. return -ENOMEM;
  498. return 0;
  499. }
  500. /**
  501. * alloc_xenballooned_pages - get pages that have been ballooned out
  502. * @nr_pages: Number of pages to get
  503. * @pages: pages returned
  504. * @return 0 on success, error otherwise
  505. */
  506. int alloc_xenballooned_pages(int nr_pages, struct page **pages)
  507. {
  508. int pgno = 0;
  509. struct page *page;
  510. int ret;
  511. mutex_lock(&balloon_mutex);
  512. balloon_stats.target_unpopulated += nr_pages;
  513. while (pgno < nr_pages) {
  514. page = balloon_retrieve(true);
  515. if (page) {
  516. pages[pgno++] = page;
  517. #ifdef CONFIG_XEN_HAVE_PVMMU
  518. /*
  519. * We don't support PV MMU when Linux and Xen is using
  520. * different page granularity.
  521. */
  522. BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
  523. if (!xen_feature(XENFEAT_auto_translated_physmap)) {
  524. ret = xen_alloc_p2m_entry(page_to_pfn(page));
  525. if (ret < 0)
  526. goto out_undo;
  527. }
  528. #endif
  529. } else {
  530. ret = add_ballooned_pages(nr_pages - pgno);
  531. if (ret < 0)
  532. goto out_undo;
  533. }
  534. }
  535. mutex_unlock(&balloon_mutex);
  536. return 0;
  537. out_undo:
  538. mutex_unlock(&balloon_mutex);
  539. free_xenballooned_pages(pgno, pages);
  540. /*
  541. * NB: free_xenballooned_pages will only subtract pgno pages, but since
  542. * target_unpopulated is incremented with nr_pages at the start we need
  543. * to remove the remaining ones also, or accounting will be screwed.
  544. */
  545. balloon_stats.target_unpopulated -= nr_pages - pgno;
  546. return ret;
  547. }
  548. EXPORT_SYMBOL(alloc_xenballooned_pages);
  549. /**
  550. * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
  551. * @nr_pages: Number of pages
  552. * @pages: pages to return
  553. */
  554. void free_xenballooned_pages(int nr_pages, struct page **pages)
  555. {
  556. int i;
  557. mutex_lock(&balloon_mutex);
  558. for (i = 0; i < nr_pages; i++) {
  559. if (pages[i])
  560. balloon_append(pages[i]);
  561. }
  562. balloon_stats.target_unpopulated -= nr_pages;
  563. /* The balloon may be too large now. Shrink it if needed. */
  564. if (current_credit())
  565. wake_up(&balloon_thread_wq);
  566. mutex_unlock(&balloon_mutex);
  567. }
  568. EXPORT_SYMBOL(free_xenballooned_pages);
  569. #if defined(CONFIG_XEN_PV) && !defined(CONFIG_XEN_UNPOPULATED_ALLOC)
  570. static void __init balloon_add_region(unsigned long start_pfn,
  571. unsigned long pages)
  572. {
  573. unsigned long pfn, extra_pfn_end;
  574. /*
  575. * If the amount of usable memory has been limited (e.g., with
  576. * the 'mem' command line parameter), don't add pages beyond
  577. * this limit.
  578. */
  579. extra_pfn_end = min(max_pfn, start_pfn + pages);
  580. for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
  581. /* totalram_pages and totalhigh_pages do not
  582. include the boot-time balloon extension, so
  583. don't subtract from it. */
  584. balloon_append(pfn_to_page(pfn));
  585. }
  586. balloon_stats.total_pages += extra_pfn_end - start_pfn;
  587. }
  588. #endif
  589. static int __init balloon_init(void)
  590. {
  591. struct task_struct *task;
  592. if (!xen_domain())
  593. return -ENODEV;
  594. pr_info("Initialising balloon driver\n");
  595. #ifdef CONFIG_XEN_PV
  596. balloon_stats.current_pages = xen_pv_domain()
  597. ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
  598. : get_num_physpages();
  599. #else
  600. balloon_stats.current_pages = get_num_physpages();
  601. #endif
  602. balloon_stats.target_pages = balloon_stats.current_pages;
  603. balloon_stats.balloon_low = 0;
  604. balloon_stats.balloon_high = 0;
  605. balloon_stats.total_pages = balloon_stats.current_pages;
  606. balloon_stats.schedule_delay = 1;
  607. balloon_stats.max_schedule_delay = 32;
  608. balloon_stats.retry_count = 1;
  609. balloon_stats.max_retry_count = 4;
  610. #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
  611. set_online_page_callback(&xen_online_page);
  612. register_memory_notifier(&xen_memory_nb);
  613. register_sysctl_table(xen_root);
  614. #endif
  615. #if defined(CONFIG_XEN_PV) && !defined(CONFIG_XEN_UNPOPULATED_ALLOC)
  616. {
  617. int i;
  618. /*
  619. * Initialize the balloon with pages from the extra memory
  620. * regions (see arch/x86/xen/setup.c).
  621. */
  622. for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
  623. if (xen_extra_mem[i].n_pfns)
  624. balloon_add_region(xen_extra_mem[i].start_pfn,
  625. xen_extra_mem[i].n_pfns);
  626. }
  627. #endif
  628. task = kthread_run(balloon_thread, NULL, "xen-balloon");
  629. if (IS_ERR(task)) {
  630. pr_err("xen-balloon thread could not be started, ballooning will not work!\n");
  631. return PTR_ERR(task);
  632. }
  633. /* Init the xen-balloon driver. */
  634. xen_balloon_init();
  635. return 0;
  636. }
  637. subsys_initcall(balloon_init);
  638. static int __init balloon_wait_finish(void)
  639. {
  640. long credit, last_credit = 0;
  641. unsigned long last_changed = 0;
  642. if (!xen_domain())
  643. return -ENODEV;
  644. /* PV guests don't need to wait. */
  645. if (xen_pv_domain() || !current_credit())
  646. return 0;
  647. pr_notice("Waiting for initial ballooning down having finished.\n");
  648. while ((credit = current_credit()) < 0) {
  649. if (credit != last_credit) {
  650. last_changed = jiffies;
  651. last_credit = credit;
  652. }
  653. if (balloon_state == BP_ECANCELED) {
  654. pr_warn_once("Initial ballooning failed, %ld pages need to be freed.\n",
  655. -credit);
  656. if (jiffies - last_changed >= HZ * balloon_boot_timeout)
  657. panic("Initial ballooning failed!\n");
  658. }
  659. schedule_timeout_interruptible(HZ / 10);
  660. }
  661. pr_notice("Initial ballooning down finished.\n");
  662. return 0;
  663. }
  664. late_initcall_sync(balloon_wait_finish);