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/errno.h>
  45 #include <linux/mm.h>
  46 #include <linux/bootmem.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/highmem.h>
  49 #include <linux/mutex.h>
  50 #include <linux/list.h>
  51 #include <linux/gfp.h>
  52 #include <linux/notifier.h>
  53 #include <linux/memory.h>
  54 #include <linux/memory_hotplug.h>
  55 #include <linux/percpu-defs.h>
  56 #include <linux/slab.h>
  57 #include <linux/sysctl.h>
  58
  59 #include <asm/page.h>
  60 #include <asm/pgalloc.h>
  61 #include <asm/pgtable.h>
  62 #include <asm/tlb.h>
  63
  64 #include <asm/xen/hypervisor.h>
  65 #include <asm/xen/hypercall.h>
  66
  67 #include <xen/xen.h>
  68 #include <xen/interface/xen.h>
  69 #include <xen/interface/memory.h>
  70 #include <xen/balloon.h>
  71 #include <xen/features.h>
  72 #include <xen/page.h>
  73
  74 static int xen_hotplug_unpopulated;
  75
  76 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
  77
  78 static int zero;
  79 static int one = 1;
  80
  81 static struct ctl_table balloon_table[] = {
  82         {
  83                 .procname       = "hotplug_unpopulated",
  84                 .data           = &xen_hotplug_unpopulated,
  85                 .maxlen         = sizeof(int),
  86                 .mode           = 0644,
  87                 .proc_handler   = proc_dointvec_minmax,
  88                 .extra1         = &zero,
  89                 .extra2         = &one,
  90         },
  91         { }
  92 };
  93
  94 static struct ctl_table balloon_root[] = {
  95         {
  96                 .procname       = "balloon",
  97                 .mode           = 0555,
  98                 .child          = balloon_table,
  99         },
 100         { }
 101 };
 102
 103 static struct ctl_table xen_root[] = {
 104         {
 105                 .procname       = "xen",
 106                 .mode           = 0555,
 107                 .child          = balloon_root,
 108         },
 109         { }
 110 };
 111
 112 #endif
 113
 114 /*
 115  * Use one extent per PAGE_SIZE to avoid to break down the page into
 116  * multiple frame.
 117  */
 118 #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
 119
 120 /*
 121  * balloon_process() state:
 122  *
 123  * BP_DONE: done or nothing to do,
 124  * BP_WAIT: wait to be rescheduled,
 125  * BP_EAGAIN: error, go to sleep,
 126  * BP_ECANCELED: error, balloon operation canceled.
 127  */
 128
 129 enum bp_state {
 130         BP_DONE,
 131         BP_WAIT,
 132         BP_EAGAIN,
 133         BP_ECANCELED
 134 };
 135
 136
 137 static DEFINE_MUTEX(balloon_mutex);
 138
 139 struct balloon_stats balloon_stats;
 140 EXPORT_SYMBOL_GPL(balloon_stats);
 141
 142 /* We increase/decrease in batches which fit in a page */
 143 static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
 144
 145
 146 /* List of ballooned pages, threaded through the mem_map array. */
 147 static LIST_HEAD(ballooned_pages);
 148 static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
 149
 150 /* Main work function, always executed in process context. */
 151 static void balloon_process(struct work_struct *work);
 152 static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
 153
 154 /* When ballooning out (allocating memory to return to Xen) we don't really
 155    want the kernel to try too hard since that can trigger the oom killer. */
 156 #define GFP_BALLOON \
 157         (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
 158
 159 static void scrub_page(struct page *page)
 160 {
 161 #ifdef CONFIG_XEN_SCRUB_PAGES
 162         clear_highpage(page);
 163 #endif
 164 }
 165
 166 /* balloon_append: add the given page to the balloon. */
 167 static void __balloon_append(struct page *page)
 168 {
 169         /* Lowmem is re-populated first, so highmem pages go at list tail. */
 170         if (PageHighMem(page)) {
 171                 list_add_tail(&page->lru, &ballooned_pages);
 172                 balloon_stats.balloon_high++;
 173         } else {
 174                 list_add(&page->lru, &ballooned_pages);
 175                 balloon_stats.balloon_low++;
 176         }
 177         wake_up(&balloon_wq);
 178 }
 179
 180 static void balloon_append(struct page *page)
 181 {
 182         __balloon_append(page);
 183         adjust_managed_page_count(page, -1);
 184 }
 185
 186 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
 187 static struct page *balloon_retrieve(bool require_lowmem)
 188 {
 189         struct page *page;
 190
 191         if (list_empty(&ballooned_pages))
 192                 return NULL;
 193
 194         page = list_entry(ballooned_pages.next, struct page, lru);
 195         if (require_lowmem && PageHighMem(page))
 196                 return NULL;
 197         list_del(&page->lru);
 198
 199         if (PageHighMem(page))
 200                 balloon_stats.balloon_high--;
 201         else
 202                 balloon_stats.balloon_low--;
 203
 204         adjust_managed_page_count(page, 1);
 205
 206         return page;
 207 }
 208
 209 static struct page *balloon_next_page(struct page *page)
 210 {
 211         struct list_head *next = page->lru.next;
 212         if (next == &ballooned_pages)
 213                 return NULL;
 214         return list_entry(next, struct page, lru);
 215 }
 216
 217 static enum bp_state update_schedule(enum bp_state state)
 218 {
 219         if (state == BP_WAIT)
 220                 return BP_WAIT;
 221
 222         if (state == BP_ECANCELED)
 223                 return BP_ECANCELED;
 224
 225         if (state == BP_DONE) {
 226                 balloon_stats.schedule_delay = 1;
 227                 balloon_stats.retry_count = 1;
 228                 return BP_DONE;
 229         }
 230
 231         ++balloon_stats.retry_count;
 232
 233         if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
 234                         balloon_stats.retry_count > balloon_stats.max_retry_count) {
 235                 balloon_stats.schedule_delay = 1;
 236                 balloon_stats.retry_count = 1;
 237                 return BP_ECANCELED;
 238         }
 239
 240         balloon_stats.schedule_delay <<= 1;
 241
 242         if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
 243                 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
 244
 245         return BP_EAGAIN;
 246 }
 247
 248 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 249 static void release_memory_resource(struct resource *resource)
 250 {
 251         if (!resource)
 252                 return;
 253
 254         /*
 255          * No need to reset region to identity mapped since we now
 256          * know that no I/O can be in this region
 257          */
 258         release_resource(resource);
 259         kfree(resource);
 260 }
 261
 262 static struct resource *additional_memory_resource(phys_addr_t size)
 263 {
 264         struct resource *res;
 265         int ret;
 266
 267         res = kzalloc(sizeof(*res), GFP_KERNEL);
 268         if (!res)
 269                 return NULL;
 270
 271         res->name = "System RAM";
 272         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 273
 274         ret = allocate_resource(&iomem_resource, res,
 275                                 size, 0, -1,
 276                                 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
 277         if (ret < 0) {
 278                 pr_err("Cannot allocate new System RAM resource\n");
 279                 kfree(res);
 280                 return NULL;
 281         }
 282
 283 #ifdef CONFIG_SPARSEMEM
 284         {
 285                 unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
 286                 unsigned long pfn = res->start >> PAGE_SHIFT;
 287
 288                 if (pfn > limit) {
 289                         pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
 290                                pfn, limit);
 291                         release_memory_resource(res);
 292                         return NULL;
 293                 }
 294         }
 295 #endif
 296
 297         return res;
 298 }
 299
 300 static enum bp_state reserve_additional_memory(void)
 301 {
 302         long credit;
 303         struct resource *resource;
 304         int nid, rc;
 305         unsigned long balloon_hotplug;
 306
 307         credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
 308                 - balloon_stats.total_pages;
 309
 310         /*
 311          * Already hotplugged enough pages?  Wait for them to be
 312          * onlined.
 313          */
 314         if (credit <= 0)
 315                 return BP_WAIT;
 316
 317         balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
 318
 319         resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
 320         if (!resource)
 321                 goto err;
 322
 323         nid = memory_add_physaddr_to_nid(resource->start);
 324
 325 #ifdef CONFIG_XEN_HAVE_PVMMU
 326         /*
 327          * We don't support PV MMU when Linux and Xen is using
 328          * different page granularity.
 329          */
 330         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 331
 332         /*
 333          * add_memory() will build page tables for the new memory so
 334          * the p2m must contain invalid entries so the correct
 335          * non-present PTEs will be written.
 336          *
 337          * If a failure occurs, the original (identity) p2m entries
 338          * are not restored since this region is now known not to
 339          * conflict with any devices.
 340          */
 341         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 342                 unsigned long pfn, i;
 343
 344                 pfn = PFN_DOWN(resource->start);
 345                 for (i = 0; i < balloon_hotplug; i++) {
 346                         if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
 347                                 pr_warn("set_phys_to_machine() failed, no memory added\n");
 348                                 goto err;
 349                         }
 350                 }
 351         }
 352 #endif
 353
 354         /*
 355          * add_memory_resource() will call online_pages() which in its turn
 356          * will call xen_online_page() callback causing deadlock if we don't
 357          * release balloon_mutex here. Unlocking here is safe because the
 358          * callers drop the mutex before trying again.
 359          */
 360         mutex_unlock(&balloon_mutex);
 361         rc = add_memory_resource(nid, resource, memhp_auto_online);
 362         mutex_lock(&balloon_mutex);
 363
 364         if (rc) {
 365                 pr_warn("Cannot add additional memory (%i)\n", rc);
 366                 goto err;
 367         }
 368
 369         balloon_stats.total_pages += balloon_hotplug;
 370
 371         return BP_WAIT;
 372   err:
 373         release_memory_resource(resource);
 374         return BP_ECANCELED;
 375 }
 376
 377 static void xen_online_page(struct page *page)
 378 {
 379         __online_page_set_limits(page);
 380
 381         mutex_lock(&balloon_mutex);
 382
 383         __balloon_append(page);
 384
 385         mutex_unlock(&balloon_mutex);
 386 }
 387
 388 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
 389 {
 390         if (val == MEM_ONLINE)
 391                 schedule_delayed_work(&balloon_worker, 0);
 392
 393         return NOTIFY_OK;
 394 }
 395
 396 static struct notifier_block xen_memory_nb = {
 397         .notifier_call = xen_memory_notifier,
 398         .priority = 0
 399 };
 400 #else
 401 static enum bp_state reserve_additional_memory(void)
 402 {
 403         balloon_stats.target_pages = balloon_stats.current_pages;
 404         return BP_ECANCELED;
 405 }
 406 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
 407
 408 static long current_credit(void)
 409 {
 410         return balloon_stats.target_pages - balloon_stats.current_pages;
 411 }
 412
 413 static bool balloon_is_inflated(void)
 414 {
 415         return balloon_stats.balloon_low || balloon_stats.balloon_high;
 416 }
 417
 418 static enum bp_state increase_reservation(unsigned long nr_pages)
 419 {
 420         int rc;
 421         unsigned long i;
 422         struct page   *page;
 423         struct xen_memory_reservation reservation = {
 424                 .address_bits = 0,
 425                 .extent_order = EXTENT_ORDER,
 426                 .domid        = DOMID_SELF
 427         };
 428
 429         if (nr_pages > ARRAY_SIZE(frame_list))
 430                 nr_pages = ARRAY_SIZE(frame_list);
 431
 432         page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
 433         for (i = 0; i < nr_pages; i++) {
 434                 if (!page) {
 435                         nr_pages = i;
 436                         break;
 437                 }
 438
 439                 /* XENMEM_populate_physmap requires a PFN based on Xen
 440                  * granularity.
 441                  */
 442                 frame_list[i] = page_to_xen_pfn(page);
 443                 page = balloon_next_page(page);
 444         }
 445
 446         set_xen_guest_handle(reservation.extent_start, frame_list);
 447         reservation.nr_extents = nr_pages;
 448         rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
 449         if (rc <= 0)
 450                 return BP_EAGAIN;
 451
 452         for (i = 0; i < rc; i++) {
 453                 page = balloon_retrieve(false);
 454                 BUG_ON(page == NULL);
 455
 456 #ifdef CONFIG_XEN_HAVE_PVMMU
 457                 /*
 458                  * We don't support PV MMU when Linux and Xen is using
 459                  * different page granularity.
 460                  */
 461                 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 462
 463                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 464                         unsigned long pfn = page_to_pfn(page);
 465
 466                         set_phys_to_machine(pfn, frame_list[i]);
 467
 468                         /* Link back into the page tables if not highmem. */
 469                         if (!PageHighMem(page)) {
 470                                 int ret;
 471                                 ret = HYPERVISOR_update_va_mapping(
 472                                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 473                                                 mfn_pte(frame_list[i], PAGE_KERNEL),
 474                                                 0);
 475                                 BUG_ON(ret);
 476                         }
 477                 }
 478 #endif
 479
 480                 /* Relinquish the page back to the allocator. */
 481                 __free_reserved_page(page);
 482         }
 483
 484         balloon_stats.current_pages += rc;
 485
 486         return BP_DONE;
 487 }
 488
 489 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
 490 {
 491         enum bp_state state = BP_DONE;
 492         unsigned long i;
 493         struct page *page, *tmp;
 494         int ret;
 495         struct xen_memory_reservation reservation = {
 496                 .address_bits = 0,
 497                 .extent_order = EXTENT_ORDER,
 498                 .domid        = DOMID_SELF
 499         };
 500         LIST_HEAD(pages);
 501
 502         if (nr_pages > ARRAY_SIZE(frame_list))
 503                 nr_pages = ARRAY_SIZE(frame_list);
 504
 505         for (i = 0; i < nr_pages; i++) {
 506                 page = alloc_page(gfp);
 507                 if (page == NULL) {
 508                         nr_pages = i;
 509                         state = BP_EAGAIN;
 510                         break;
 511                 }
 512                 scrub_page(page);
 513                 list_add(&page->lru, &pages);
 514         }
 515
 516         /*
 517          * Ensure that ballooned highmem pages don't have kmaps.
 518          *
 519          * Do this before changing the p2m as kmap_flush_unused()
 520          * reads PTEs to obtain pages (and hence needs the original
 521          * p2m entry).
 522          */
 523         kmap_flush_unused();
 524
 525         /*
 526          * Setup the frame, update direct mapping, invalidate P2M,
 527          * and add to balloon.
 528          */
 529         i = 0;
 530         list_for_each_entry_safe(page, tmp, &pages, lru) {
 531                 /* XENMEM_decrease_reservation requires a GFN */
 532                 frame_list[i++] = xen_page_to_gfn(page);
 533
 534 #ifdef CONFIG_XEN_HAVE_PVMMU
 535                 /*
 536                  * We don't support PV MMU when Linux and Xen is using
 537                  * different page granularity.
 538                  */
 539                 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 540
 541                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 542                         unsigned long pfn = page_to_pfn(page);
 543
 544                         if (!PageHighMem(page)) {
 545                                 ret = HYPERVISOR_update_va_mapping(
 546                                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 547                                                 __pte_ma(0), 0);
 548                                 BUG_ON(ret);
 549                         }
 550                         __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 551                 }
 552 #endif
 553                 list_del(&page->lru);
 554
 555                 balloon_append(page);
 556         }
 557
 558         flush_tlb_all();
 559
 560         set_xen_guest_handle(reservation.extent_start, frame_list);
 561         reservation.nr_extents   = nr_pages;
 562         ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
 563         BUG_ON(ret != nr_pages);
 564
 565         balloon_stats.current_pages -= nr_pages;
 566
 567         return state;
 568 }
 569
 570 /*
 571  * As this is a work item it is guaranteed to run as a single instance only.
 572  * We may of course race updates of the target counts (which are protected
 573  * by the balloon lock), or with changes to the Xen hard limit, but we will
 574  * recover from these in time.
 575  */
 576 static void balloon_process(struct work_struct *work)
 577 {
 578         enum bp_state state = BP_DONE;
 579         long credit;
 580
 581
 582         do {
 583                 mutex_lock(&balloon_mutex);
 584
 585                 credit = current_credit();
 586
 587                 if (credit > 0) {
 588                         if (balloon_is_inflated())
 589                                 state = increase_reservation(credit);
 590                         else
 591                                 state = reserve_additional_memory();
 592                 }
 593
 594                 if (credit < 0)
 595                         state = decrease_reservation(-credit, GFP_BALLOON);
 596
 597                 state = update_schedule(state);
 598
 599                 mutex_unlock(&balloon_mutex);
 600
 601                 cond_resched();
 602
 603         } while (credit && state == BP_DONE);
 604
 605         /* Schedule more work if there is some still to be done. */
 606         if (state == BP_EAGAIN)
 607                 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
 608 }
 609
 610 /* Resets the Xen limit, sets new target, and kicks off processing. */
 611 void balloon_set_new_target(unsigned long target)
 612 {
 613         /* No need for lock. Not read-modify-write updates. */
 614         balloon_stats.target_pages = target;
 615         schedule_delayed_work(&balloon_worker, 0);
 616 }
 617 EXPORT_SYMBOL_GPL(balloon_set_new_target);
 618
 619 static int add_ballooned_pages(int nr_pages)
 620 {
 621         enum bp_state st;
 622
 623         if (xen_hotplug_unpopulated) {
 624                 st = reserve_additional_memory();
 625                 if (st != BP_ECANCELED) {
 626                         mutex_unlock(&balloon_mutex);
 627                         wait_event(balloon_wq,
 628                                    !list_empty(&ballooned_pages));
 629                         mutex_lock(&balloon_mutex);
 630                         return 0;
 631                 }
 632         }
 633
 634         st = decrease_reservation(nr_pages, GFP_USER);
 635         if (st != BP_DONE)
 636                 return -ENOMEM;
 637
 638         return 0;
 639 }
 640
 641 /**
 642  * alloc_xenballooned_pages - get pages that have been ballooned out
 643  * @nr_pages: Number of pages to get
 644  * @pages: pages returned
 645  * @return 0 on success, error otherwise
 646  */
 647 int alloc_xenballooned_pages(int nr_pages, struct page **pages)
 648 {
 649         int pgno = 0;
 650         struct page *page;
 651         int ret;
 652
 653         mutex_lock(&balloon_mutex);
 654
 655         balloon_stats.target_unpopulated += nr_pages;
 656
 657         while (pgno < nr_pages) {
 658                 page = balloon_retrieve(true);
 659                 if (page) {
 660                         pages[pgno++] = page;
 661 #ifdef CONFIG_XEN_HAVE_PVMMU
 662                         /*
 663                          * We don't support PV MMU when Linux and Xen is using
 664                          * different page granularity.
 665                          */
 666                         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 667
 668                         ret = xen_alloc_p2m_entry(page_to_pfn(page));
 669                         if (ret < 0)
 670                                 goto out_undo;
 671 #endif
 672                 } else {
 673                         ret = add_ballooned_pages(nr_pages - pgno);
 674                         if (ret < 0)
 675                                 goto out_undo;
 676                 }
 677         }
 678         mutex_unlock(&balloon_mutex);
 679         return 0;
 680  out_undo:
 681         mutex_unlock(&balloon_mutex);
 682         free_xenballooned_pages(pgno, pages);
 683         return ret;
 684 }
 685 EXPORT_SYMBOL(alloc_xenballooned_pages);
 686
 687 /**
 688  * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
 689  * @nr_pages: Number of pages
 690  * @pages: pages to return
 691  */
 692 void free_xenballooned_pages(int nr_pages, struct page **pages)
 693 {
 694         int i;
 695
 696         mutex_lock(&balloon_mutex);
 697
 698         for (i = 0; i < nr_pages; i++) {
 699                 if (pages[i])
 700                         balloon_append(pages[i]);
 701         }
 702
 703         balloon_stats.target_unpopulated -= nr_pages;
 704
 705         /* The balloon may be too large now. Shrink it if needed. */
 706         if (current_credit())
 707                 schedule_delayed_work(&balloon_worker, 0);
 708
 709         mutex_unlock(&balloon_mutex);
 710 }
 711 EXPORT_SYMBOL(free_xenballooned_pages);
 712
 713 static void __init balloon_add_region(unsigned long start_pfn,
 714                                       unsigned long pages)
 715 {
 716         unsigned long pfn, extra_pfn_end;
 717         struct page *page;
 718
 719         /*
 720          * If the amount of usable memory has been limited (e.g., with
 721          * the 'mem' command line parameter), don't add pages beyond
 722          * this limit.
 723          */
 724         extra_pfn_end = min(max_pfn, start_pfn + pages);
 725
 726         for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
 727                 page = pfn_to_page(pfn);
 728                 /* totalram_pages and totalhigh_pages do not
 729                    include the boot-time balloon extension, so
 730                    don't subtract from it. */
 731                 __balloon_append(page);
 732         }
 733
 734         balloon_stats.total_pages += extra_pfn_end - start_pfn;
 735 }
 736
 737 static int __init balloon_init(void)
 738 {
 739         int i;
 740
 741         if (!xen_domain())
 742                 return -ENODEV;
 743
 744         pr_info("Initialising balloon driver\n");
 745
 746         balloon_stats.current_pages = xen_pv_domain()
 747                 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
 748                 : get_num_physpages();
 749         balloon_stats.target_pages  = balloon_stats.current_pages;
 750         balloon_stats.balloon_low   = 0;
 751         balloon_stats.balloon_high  = 0;
 752         balloon_stats.total_pages   = balloon_stats.current_pages;
 753
 754         balloon_stats.schedule_delay = 1;
 755         balloon_stats.max_schedule_delay = 32;
 756         balloon_stats.retry_count = 1;
 757         balloon_stats.max_retry_count = RETRY_UNLIMITED;
 758
 759 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 760         set_online_page_callback(&xen_online_page);
 761         register_memory_notifier(&xen_memory_nb);
 762         register_sysctl_table(xen_root);
 763 #endif
 764
 765         /*
 766          * Initialize the balloon with pages from the extra memory
 767          * regions (see arch/x86/xen/setup.c).
 768          */
 769         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
 770                 if (xen_extra_mem[i].n_pfns)
 771                         balloon_add_region(xen_extra_mem[i].start_pfn,
 772                                            xen_extra_mem[i].n_pfns);
 773
 774         return 0;
 775 }
 776 subsys_initcall(balloon_init);