drivers/xen/balloon.c

   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
  39 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  40
  41 #include <linux/cpu.h>
  42 #include <linux/kernel.h>
  43 #include <linux/sched.h>
  44 #include <linux/cred.h>
  45 #include <linux/errno.h>
  46 #include <linux/freezer.h>
  47 #include <linux/kthread.h>
  48 #include <linux/mm.h>
  49 #include <linux/memblock.h>
  50 #include <linux/pagemap.h>
  51 #include <linux/highmem.h>
  52 #include <linux/mutex.h>
  53 #include <linux/list.h>
  54 #include <linux/gfp.h>
  55 #include <linux/notifier.h>
  56 #include <linux/memory.h>
  57 #include <linux/memory_hotplug.h>
  58 #include <linux/percpu-defs.h>
  59 #include <linux/slab.h>
  60 #include <linux/sysctl.h>
  61 #include <linux/moduleparam.h>
  62 #include <linux/jiffies.h>
  63
  64 #include <asm/page.h>
  65 #include <asm/tlb.h>
  66
  67 #include <asm/xen/hypervisor.h>
  68 #include <asm/xen/hypercall.h>
  69
  70 #include <xen/xen.h>
  71 #include <xen/interface/xen.h>
  72 #include <xen/interface/memory.h>
  73 #include <xen/balloon.h>
  74 #include <xen/features.h>
  75 #include <xen/page.h>
  76 #include <xen/mem-reservation.h>
  77
  78 #undef MODULE_PARAM_PREFIX
  79 #define MODULE_PARAM_PREFIX "xen."
  80
  81 static uint __read_mostly balloon_boot_timeout = 180;
  82 module_param(balloon_boot_timeout, uint, 0444);
  83
  84 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
  85 static int xen_hotplug_unpopulated;
  86
  87 static struct ctl_table balloon_table[] = {
  88         {
  89                 .procname       = "hotplug_unpopulated",
  90                 .data           = &xen_hotplug_unpopulated,
  91                 .maxlen         = sizeof(int),
  92                 .mode           = 0644,
  93                 .proc_handler   = proc_dointvec_minmax,
  94                 .extra1         = SYSCTL_ZERO,
  95                 .extra2         = SYSCTL_ONE,
  96         },
  97         { }
  98 };
  99
 100 #else
 101 #define xen_hotplug_unpopulated 0
 102 #endif
 103
 104 /*
 105  * Use one extent per PAGE_SIZE to avoid to break down the page into
 106  * multiple frame.
 107  */
 108 #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
 109
 110 /*
 111  * balloon_thread() state:
 112  *
 113  * BP_DONE: done or nothing to do,
 114  * BP_WAIT: wait to be rescheduled,
 115  * BP_EAGAIN: error, go to sleep,
 116  * BP_ECANCELED: error, balloon operation canceled.
 117  */
 118
 119 static enum bp_state {
 120         BP_DONE,
 121         BP_WAIT,
 122         BP_EAGAIN,
 123         BP_ECANCELED
 124 } balloon_state = BP_DONE;
 125
 126 /* Main waiting point for xen-balloon thread. */
 127 static DECLARE_WAIT_QUEUE_HEAD(balloon_thread_wq);
 128
 129 static DEFINE_MUTEX(balloon_mutex);
 130
 131 struct balloon_stats balloon_stats;
 132 EXPORT_SYMBOL_GPL(balloon_stats);
 133
 134 /* We increase/decrease in batches which fit in a page */
 135 static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
 136
 137
 138 /* List of ballooned pages, threaded through the mem_map array. */
 139 static LIST_HEAD(ballooned_pages);
 140 static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
 141
 142 /* When ballooning out (allocating memory to return to Xen) we don't really
 143    want the kernel to try too hard since that can trigger the oom killer. */
 144 #define GFP_BALLOON \
 145         (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
 146
 147 /* balloon_append: add the given page to the balloon. */
 148 static void balloon_append(struct page *page)
 149 {
 150         __SetPageOffline(page);
 151
 152         /* Lowmem is re-populated first, so highmem pages go at list tail. */
 153         if (PageHighMem(page)) {
 154                 list_add_tail(&page->lru, &ballooned_pages);
 155                 balloon_stats.balloon_high++;
 156         } else {
 157                 list_add(&page->lru, &ballooned_pages);
 158                 balloon_stats.balloon_low++;
 159         }
 160         wake_up(&balloon_wq);
 161 }
 162
 163 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
 164 static struct page *balloon_retrieve(bool require_lowmem)
 165 {
 166         struct page *page;
 167
 168         if (list_empty(&ballooned_pages))
 169                 return NULL;
 170
 171         page = list_entry(ballooned_pages.next, struct page, lru);
 172         if (require_lowmem && PageHighMem(page))
 173                 return NULL;
 174         list_del(&page->lru);
 175
 176         if (PageHighMem(page))
 177                 balloon_stats.balloon_high--;
 178         else
 179                 balloon_stats.balloon_low--;
 180
 181         __ClearPageOffline(page);
 182         return page;
 183 }
 184
 185 static struct page *balloon_next_page(struct page *page)
 186 {
 187         struct list_head *next = page->lru.next;
 188         if (next == &ballooned_pages)
 189                 return NULL;
 190         return list_entry(next, struct page, lru);
 191 }
 192
 193 static void update_schedule(void)
 194 {
 195         if (balloon_state == BP_WAIT || balloon_state == BP_ECANCELED)
 196                 return;
 197
 198         if (balloon_state == BP_DONE) {
 199                 balloon_stats.schedule_delay = 1;
 200                 balloon_stats.retry_count = 1;
 201                 return;
 202         }
 203
 204         ++balloon_stats.retry_count;
 205
 206         if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
 207                         balloon_stats.retry_count > balloon_stats.max_retry_count) {
 208                 balloon_stats.schedule_delay = 1;
 209                 balloon_stats.retry_count = 1;
 210                 balloon_state = BP_ECANCELED;
 211                 return;
 212         }
 213
 214         balloon_stats.schedule_delay <<= 1;
 215
 216         if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
 217                 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
 218
 219         balloon_state = BP_EAGAIN;
 220 }
 221
 222 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 223 static void release_memory_resource(struct resource *resource)
 224 {
 225         if (!resource)
 226                 return;
 227
 228         /*
 229          * No need to reset region to identity mapped since we now
 230          * know that no I/O can be in this region
 231          */
 232         release_resource(resource);
 233         kfree(resource);
 234 }
 235
 236 static struct resource *additional_memory_resource(phys_addr_t size)
 237 {
 238         struct resource *res;
 239         int ret;
 240
 241         res = kzalloc(sizeof(*res), GFP_KERNEL);
 242         if (!res)
 243                 return NULL;
 244
 245         res->name = "System RAM";
 246         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 247
 248         ret = allocate_resource(&iomem_resource, res,
 249                                 size, 0, -1,
 250                                 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
 251         if (ret < 0) {
 252                 pr_err("Cannot allocate new System RAM resource\n");
 253                 kfree(res);
 254                 return NULL;
 255         }
 256
 257         return res;
 258 }
 259
 260 static enum bp_state reserve_additional_memory(void)
 261 {
 262         long credit;
 263         struct resource *resource;
 264         int nid, rc;
 265         unsigned long balloon_hotplug;
 266
 267         credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
 268                 - balloon_stats.total_pages;
 269
 270         /*
 271          * Already hotplugged enough pages?  Wait for them to be
 272          * onlined.
 273          */
 274         if (credit <= 0)
 275                 return BP_WAIT;
 276
 277         balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
 278
 279         resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
 280         if (!resource)
 281                 goto err;
 282
 283         nid = memory_add_physaddr_to_nid(resource->start);
 284
 285 #ifdef CONFIG_XEN_HAVE_PVMMU
 286         /*
 287          * We don't support PV MMU when Linux and Xen is using
 288          * different page granularity.
 289          */
 290         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 291
 292         /*
 293          * add_memory() will build page tables for the new memory so
 294          * the p2m must contain invalid entries so the correct
 295          * non-present PTEs will be written.
 296          *
 297          * If a failure occurs, the original (identity) p2m entries
 298          * are not restored since this region is now known not to
 299          * conflict with any devices.
 300          */
 301         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 302                 unsigned long pfn, i;
 303
 304                 pfn = PFN_DOWN(resource->start);
 305                 for (i = 0; i < balloon_hotplug; i++) {
 306                         if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
 307                                 pr_warn("set_phys_to_machine() failed, no memory added\n");
 308                                 goto err;
 309                         }
 310                 }
 311         }
 312 #endif
 313
 314         /*
 315          * add_memory_resource() will call online_pages() which in its turn
 316          * will call xen_online_page() callback causing deadlock if we don't
 317          * release balloon_mutex here. Unlocking here is safe because the
 318          * callers drop the mutex before trying again.
 319          */
 320         mutex_unlock(&balloon_mutex);
 321         /* add_memory_resource() requires the device_hotplug lock */
 322         lock_device_hotplug();
 323         rc = add_memory_resource(nid, resource, MHP_MERGE_RESOURCE);
 324         unlock_device_hotplug();
 325         mutex_lock(&balloon_mutex);
 326
 327         if (rc) {
 328                 pr_warn("Cannot add additional memory (%i)\n", rc);
 329                 goto err;
 330         }
 331
 332         balloon_stats.total_pages += balloon_hotplug;
 333
 334         return BP_WAIT;
 335   err:
 336         release_memory_resource(resource);
 337         return BP_ECANCELED;
 338 }
 339
 340 static void xen_online_page(struct page *page, unsigned int order)
 341 {
 342         unsigned long i, size = (1 << order);
 343         unsigned long start_pfn = page_to_pfn(page);
 344         struct page *p;
 345
 346         pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn);
 347         mutex_lock(&balloon_mutex);
 348         for (i = 0; i < size; i++) {
 349                 p = pfn_to_page(start_pfn + i);
 350                 balloon_append(p);
 351         }
 352         mutex_unlock(&balloon_mutex);
 353 }
 354
 355 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
 356 {
 357         if (val == MEM_ONLINE)
 358                 wake_up(&balloon_thread_wq);
 359
 360         return NOTIFY_OK;
 361 }
 362
 363 static struct notifier_block xen_memory_nb = {
 364         .notifier_call = xen_memory_notifier,
 365         .priority = 0
 366 };
 367 #else
 368 static enum bp_state reserve_additional_memory(void)
 369 {
 370         balloon_stats.target_pages = balloon_stats.current_pages +
 371                                      balloon_stats.target_unpopulated;
 372         return BP_ECANCELED;
 373 }
 374 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
 375
 376 static long current_credit(void)
 377 {
 378         return balloon_stats.target_pages - balloon_stats.current_pages;
 379 }
 380
 381 static bool balloon_is_inflated(void)
 382 {
 383         return balloon_stats.balloon_low || balloon_stats.balloon_high;
 384 }
 385
 386 static enum bp_state increase_reservation(unsigned long nr_pages)
 387 {
 388         int rc;
 389         unsigned long i;
 390         struct page   *page;
 391
 392         if (nr_pages > ARRAY_SIZE(frame_list))
 393                 nr_pages = ARRAY_SIZE(frame_list);
 394
 395         page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
 396         for (i = 0; i < nr_pages; i++) {
 397                 if (!page) {
 398                         nr_pages = i;
 399                         break;
 400                 }
 401
 402                 frame_list[i] = page_to_xen_pfn(page);
 403                 page = balloon_next_page(page);
 404         }
 405
 406         rc = xenmem_reservation_increase(nr_pages, frame_list);
 407         if (rc <= 0)
 408                 return BP_EAGAIN;
 409
 410         for (i = 0; i < rc; i++) {
 411                 page = balloon_retrieve(false);
 412                 BUG_ON(page == NULL);
 413
 414                 xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]);
 415
 416                 /* Relinquish the page back to the allocator. */
 417                 free_reserved_page(page);
 418         }
 419
 420         balloon_stats.current_pages += rc;
 421
 422         return BP_DONE;
 423 }
 424
 425 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
 426 {
 427         enum bp_state state = BP_DONE;
 428         unsigned long i;
 429         struct page *page, *tmp;
 430         int ret;
 431         LIST_HEAD(pages);
 432
 433         if (nr_pages > ARRAY_SIZE(frame_list))
 434                 nr_pages = ARRAY_SIZE(frame_list);
 435
 436         for (i = 0; i < nr_pages; i++) {
 437                 page = alloc_page(gfp);
 438                 if (page == NULL) {
 439                         nr_pages = i;
 440                         state = BP_EAGAIN;
 441                         break;
 442                 }
 443                 adjust_managed_page_count(page, -1);
 444                 xenmem_reservation_scrub_page(page);
 445                 list_add(&page->lru, &pages);
 446         }
 447
 448         /*
 449          * Ensure that ballooned highmem pages don't have kmaps.
 450          *
 451          * Do this before changing the p2m as kmap_flush_unused()
 452          * reads PTEs to obtain pages (and hence needs the original
 453          * p2m entry).
 454          */
 455         kmap_flush_unused();
 456
 457         /*
 458          * Setup the frame, update direct mapping, invalidate P2M,
 459          * and add to balloon.
 460          */
 461         i = 0;
 462         list_for_each_entry_safe(page, tmp, &pages, lru) {
 463                 frame_list[i++] = xen_page_to_gfn(page);
 464
 465                 xenmem_reservation_va_mapping_reset(1, &page);
 466
 467                 list_del(&page->lru);
 468
 469                 balloon_append(page);
 470         }
 471
 472         flush_tlb_all();
 473
 474         ret = xenmem_reservation_decrease(nr_pages, frame_list);
 475         BUG_ON(ret != nr_pages);
 476
 477         balloon_stats.current_pages -= nr_pages;
 478
 479         return state;
 480 }
 481
 482 /*
 483  * Stop waiting if either state is BP_DONE and ballooning action is
 484  * needed, or if the credit has changed while state is not BP_DONE.
 485  */
 486 static bool balloon_thread_cond(long credit)
 487 {
 488         if (balloon_state == BP_DONE)
 489                 credit = 0;
 490
 491         return current_credit() != credit || kthread_should_stop();
 492 }
 493
 494 /*
 495  * As this is a kthread it is guaranteed to run as a single instance only.
 496  * We may of course race updates of the target counts (which are protected
 497  * by the balloon lock), or with changes to the Xen hard limit, but we will
 498  * recover from these in time.
 499  */
 500 static int balloon_thread(void *unused)
 501 {
 502         long credit;
 503         unsigned long timeout;
 504
 505         set_freezable();
 506         for (;;) {
 507                 switch (balloon_state) {
 508                 case BP_DONE:
 509                 case BP_ECANCELED:
 510                         timeout = 3600 * HZ;
 511                         break;
 512                 case BP_EAGAIN:
 513                         timeout = balloon_stats.schedule_delay * HZ;
 514                         break;
 515                 case BP_WAIT:
 516                         timeout = HZ;
 517                         break;
 518                 }
 519
 520                 credit = current_credit();
 521
 522                 wait_event_freezable_timeout(balloon_thread_wq,
 523                         balloon_thread_cond(credit), timeout);
 524
 525                 if (kthread_should_stop())
 526                         return 0;
 527
 528                 mutex_lock(&balloon_mutex);
 529
 530                 credit = current_credit();
 531
 532                 if (credit > 0) {
 533                         if (balloon_is_inflated())
 534                                 balloon_state = increase_reservation(credit);
 535                         else
 536                                 balloon_state = reserve_additional_memory();
 537                 }
 538
 539                 if (credit < 0) {
 540                         long n_pages;
 541
 542                         n_pages = min(-credit, si_mem_available());
 543                         balloon_state = decrease_reservation(n_pages,
 544                                                              GFP_BALLOON);
 545                         if (balloon_state == BP_DONE && n_pages != -credit &&
 546                             n_pages < totalreserve_pages)
 547                                 balloon_state = BP_EAGAIN;
 548                 }
 549
 550                 update_schedule();
 551
 552                 mutex_unlock(&balloon_mutex);
 553
 554                 cond_resched();
 555         }
 556 }
 557
 558 /* Resets the Xen limit, sets new target, and kicks off processing. */
 559 void balloon_set_new_target(unsigned long target)
 560 {
 561         /* No need for lock. Not read-modify-write updates. */
 562         balloon_stats.target_pages = target;
 563         wake_up(&balloon_thread_wq);
 564 }
 565 EXPORT_SYMBOL_GPL(balloon_set_new_target);
 566
 567 static int add_ballooned_pages(unsigned int nr_pages)
 568 {
 569         enum bp_state st;
 570
 571         if (xen_hotplug_unpopulated) {
 572                 st = reserve_additional_memory();
 573                 if (st != BP_ECANCELED) {
 574                         int rc;
 575
 576                         mutex_unlock(&balloon_mutex);
 577                         rc = wait_event_interruptible(balloon_wq,
 578                                    !list_empty(&ballooned_pages));
 579                         mutex_lock(&balloon_mutex);
 580                         return rc ? -ENOMEM : 0;
 581                 }
 582         }
 583
 584         if (si_mem_available() < nr_pages)
 585                 return -ENOMEM;
 586
 587         st = decrease_reservation(nr_pages, GFP_USER);
 588         if (st != BP_DONE)
 589                 return -ENOMEM;
 590
 591         return 0;
 592 }
 593
 594 /**
 595  * xen_alloc_ballooned_pages - get pages that have been ballooned out
 596  * @nr_pages: Number of pages to get
 597  * @pages: pages returned
 598  * @return 0 on success, error otherwise
 599  */
 600 int xen_alloc_ballooned_pages(unsigned int nr_pages, struct page **pages)
 601 {
 602         unsigned int pgno = 0;
 603         struct page *page;
 604         int ret;
 605
 606         mutex_lock(&balloon_mutex);
 607
 608         balloon_stats.target_unpopulated += nr_pages;
 609
 610         while (pgno < nr_pages) {
 611                 page = balloon_retrieve(true);
 612                 if (page) {
 613                         pages[pgno++] = page;
 614 #ifdef CONFIG_XEN_HAVE_PVMMU
 615                         /*
 616                          * We don't support PV MMU when Linux and Xen is using
 617                          * different page granularity.
 618                          */
 619                         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 620
 621                         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 622                                 ret = xen_alloc_p2m_entry(page_to_pfn(page));
 623                                 if (ret < 0)
 624                                         goto out_undo;
 625                         }
 626 #endif
 627                 } else {
 628                         ret = add_ballooned_pages(nr_pages - pgno);
 629                         if (ret < 0)
 630                                 goto out_undo;
 631                 }
 632         }
 633         mutex_unlock(&balloon_mutex);
 634         return 0;
 635  out_undo:
 636         mutex_unlock(&balloon_mutex);
 637         xen_free_ballooned_pages(pgno, pages);
 638         /*
 639          * NB: xen_free_ballooned_pages will only subtract pgno pages, but since
 640          * target_unpopulated is incremented with nr_pages at the start we need
 641          * to remove the remaining ones also, or accounting will be screwed.
 642          */
 643         balloon_stats.target_unpopulated -= nr_pages - pgno;
 644         return ret;
 645 }
 646 EXPORT_SYMBOL(xen_alloc_ballooned_pages);
 647
 648 /**
 649  * xen_free_ballooned_pages - return pages retrieved with get_ballooned_pages
 650  * @nr_pages: Number of pages
 651  * @pages: pages to return
 652  */
 653 void xen_free_ballooned_pages(unsigned int nr_pages, struct page **pages)
 654 {
 655         unsigned int i;
 656
 657         mutex_lock(&balloon_mutex);
 658
 659         for (i = 0; i < nr_pages; i++) {
 660                 if (pages[i])
 661                         balloon_append(pages[i]);
 662         }
 663
 664         balloon_stats.target_unpopulated -= nr_pages;
 665
 666         /* The balloon may be too large now. Shrink it if needed. */
 667         if (current_credit())
 668                 wake_up(&balloon_thread_wq);
 669
 670         mutex_unlock(&balloon_mutex);
 671 }
 672 EXPORT_SYMBOL(xen_free_ballooned_pages);
 673
 674 static void __init balloon_add_regions(void)
 675 {
 676 #if defined(CONFIG_XEN_PV)
 677         unsigned long start_pfn, pages;
 678         unsigned long pfn, extra_pfn_end;
 679         unsigned int i;
 680
 681         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 682                 pages = xen_extra_mem[i].n_pfns;
 683                 if (!pages)
 684                         continue;
 685
 686                 start_pfn = xen_extra_mem[i].start_pfn;
 687
 688                 /*
 689                  * If the amount of usable memory has been limited (e.g., with
 690                  * the 'mem' command line parameter), don't add pages beyond
 691                  * this limit.
 692                  */
 693                 extra_pfn_end = min(max_pfn, start_pfn + pages);
 694
 695                 for (pfn = start_pfn; pfn < extra_pfn_end; pfn++)
 696                         balloon_append(pfn_to_page(pfn));
 697
 698                 balloon_stats.total_pages += extra_pfn_end - start_pfn;
 699         }
 700 #endif
 701 }
 702
 703 static int __init balloon_init(void)
 704 {
 705         struct task_struct *task;
 706
 707         if (!xen_domain())
 708                 return -ENODEV;
 709
 710         pr_info("Initialising balloon driver\n");
 711
 712 #ifdef CONFIG_XEN_PV
 713         balloon_stats.current_pages = xen_pv_domain()
 714                 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
 715                 : get_num_physpages();
 716 #else
 717         balloon_stats.current_pages = get_num_physpages();
 718 #endif
 719         balloon_stats.target_pages  = balloon_stats.current_pages;
 720         balloon_stats.balloon_low   = 0;
 721         balloon_stats.balloon_high  = 0;
 722         balloon_stats.total_pages   = balloon_stats.current_pages;
 723
 724         balloon_stats.schedule_delay = 1;
 725         balloon_stats.max_schedule_delay = 32;
 726         balloon_stats.retry_count = 1;
 727         balloon_stats.max_retry_count = 4;
 728
 729 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 730         set_online_page_callback(&xen_online_page);
 731         register_memory_notifier(&xen_memory_nb);
 732         register_sysctl_init("xen/balloon", balloon_table);
 733 #endif
 734
 735         balloon_add_regions();
 736
 737         task = kthread_run(balloon_thread, NULL, "xen-balloon");
 738         if (IS_ERR(task)) {
 739                 pr_err("xen-balloon thread could not be started, ballooning will not work!\n");
 740                 return PTR_ERR(task);
 741         }
 742
 743         /* Init the xen-balloon driver. */
 744         xen_balloon_init();
 745
 746         return 0;
 747 }
 748 subsys_initcall(balloon_init);
 749
 750 static int __init balloon_wait_finish(void)
 751 {
 752         long credit, last_credit = 0;
 753         unsigned long last_changed = 0;
 754
 755         if (!xen_domain())
 756                 return -ENODEV;
 757
 758         /* PV guests don't need to wait. */
 759         if (xen_pv_domain() || !current_credit())
 760                 return 0;
 761
 762         pr_notice("Waiting for initial ballooning down having finished.\n");
 763
 764         while ((credit = current_credit()) < 0) {
 765                 if (credit != last_credit) {
 766                         last_changed = jiffies;
 767                         last_credit = credit;
 768                 }
 769                 if (balloon_state == BP_ECANCELED) {
 770                         pr_warn_once("Initial ballooning failed, %ld pages need to be freed.\n",
 771                                      -credit);
 772                         if (time_is_before_eq_jiffies(last_changed + HZ * balloon_boot_timeout))
 773                                 panic("Initial ballooning failed!\n");
 774                 }
 775
 776                 schedule_timeout_interruptible(HZ / 10);
 777         }
 778
 779         pr_notice("Initial ballooning down finished.\n");
 780
 781         return 0;
 782 }
 783 late_initcall_sync(balloon_wait_finish);