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/mm.h>
  47 #include <linux/bootmem.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
  60 #include <asm/page.h>
  61 #include <asm/pgalloc.h>
  62 #include <asm/pgtable.h>
  63 #include <asm/tlb.h>
  64
  65 #include <asm/xen/hypervisor.h>
  66 #include <asm/xen/hypercall.h>
  67
  68 #include <xen/xen.h>
  69 #include <xen/interface/xen.h>
  70 #include <xen/interface/memory.h>
  71 #include <xen/balloon.h>
  72 #include <xen/features.h>
  73 #include <xen/page.h>
  74
  75 static int xen_hotplug_unpopulated;
  76
  77 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
  78
  79 static int zero;
  80 static int one = 1;
  81
  82 static struct ctl_table balloon_table[] = {
  83         {
  84                 .procname       = "hotplug_unpopulated",
  85                 .data           = &xen_hotplug_unpopulated,
  86                 .maxlen         = sizeof(int),
  87                 .mode           = 0644,
  88                 .proc_handler   = proc_dointvec_minmax,
  89                 .extra1         = &zero,
  90                 .extra2         = &one,
  91         },
  92         { }
  93 };
  94
  95 static struct ctl_table balloon_root[] = {
  96         {
  97                 .procname       = "balloon",
  98                 .mode           = 0555,
  99                 .child          = balloon_table,
 100         },
 101         { }
 102 };
 103
 104 static struct ctl_table xen_root[] = {
 105         {
 106                 .procname       = "xen",
 107                 .mode           = 0555,
 108                 .child          = balloon_root,
 109         },
 110         { }
 111 };
 112
 113 #endif
 114
 115 /*
 116  * Use one extent per PAGE_SIZE to avoid to break down the page into
 117  * multiple frame.
 118  */
 119 #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
 120
 121 /*
 122  * balloon_process() state:
 123  *
 124  * BP_DONE: done or nothing to do,
 125  * BP_WAIT: wait to be rescheduled,
 126  * BP_EAGAIN: error, go to sleep,
 127  * BP_ECANCELED: error, balloon operation canceled.
 128  */
 129
 130 enum bp_state {
 131         BP_DONE,
 132         BP_WAIT,
 133         BP_EAGAIN,
 134         BP_ECANCELED
 135 };
 136
 137
 138 static DEFINE_MUTEX(balloon_mutex);
 139
 140 struct balloon_stats balloon_stats;
 141 EXPORT_SYMBOL_GPL(balloon_stats);
 142
 143 /* We increase/decrease in batches which fit in a page */
 144 static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
 145
 146
 147 /* List of ballooned pages, threaded through the mem_map array. */
 148 static LIST_HEAD(ballooned_pages);
 149 static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
 150
 151 /* Main work function, always executed in process context. */
 152 static void balloon_process(struct work_struct *work);
 153 static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
 154
 155 /* When ballooning out (allocating memory to return to Xen) we don't really
 156    want the kernel to try too hard since that can trigger the oom killer. */
 157 #define GFP_BALLOON \
 158         (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
 159
 160 static void scrub_page(struct page *page)
 161 {
 162 #ifdef CONFIG_XEN_SCRUB_PAGES
 163         clear_highpage(page);
 164 #endif
 165 }
 166
 167 /* balloon_append: add the given page to the balloon. */
 168 static void __balloon_append(struct page *page)
 169 {
 170         /* Lowmem is re-populated first, so highmem pages go at list tail. */
 171         if (PageHighMem(page)) {
 172                 list_add_tail(&page->lru, &ballooned_pages);
 173                 balloon_stats.balloon_high++;
 174         } else {
 175                 list_add(&page->lru, &ballooned_pages);
 176                 balloon_stats.balloon_low++;
 177         }
 178         wake_up(&balloon_wq);
 179 }
 180
 181 static void balloon_append(struct page *page)
 182 {
 183         __balloon_append(page);
 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         return page;
 205 }
 206
 207 static struct page *balloon_next_page(struct page *page)
 208 {
 209         struct list_head *next = page->lru.next;
 210         if (next == &ballooned_pages)
 211                 return NULL;
 212         return list_entry(next, struct page, lru);
 213 }
 214
 215 static enum bp_state update_schedule(enum bp_state state)
 216 {
 217         if (state == BP_WAIT)
 218                 return BP_WAIT;
 219
 220         if (state == BP_ECANCELED)
 221                 return BP_ECANCELED;
 222
 223         if (state == BP_DONE) {
 224                 balloon_stats.schedule_delay = 1;
 225                 balloon_stats.retry_count = 1;
 226                 return BP_DONE;
 227         }
 228
 229         ++balloon_stats.retry_count;
 230
 231         if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
 232                         balloon_stats.retry_count > balloon_stats.max_retry_count) {
 233                 balloon_stats.schedule_delay = 1;
 234                 balloon_stats.retry_count = 1;
 235                 return BP_ECANCELED;
 236         }
 237
 238         balloon_stats.schedule_delay <<= 1;
 239
 240         if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
 241                 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
 242
 243         return BP_EAGAIN;
 244 }
 245
 246 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 247 static void release_memory_resource(struct resource *resource)
 248 {
 249         if (!resource)
 250                 return;
 251
 252         /*
 253          * No need to reset region to identity mapped since we now
 254          * know that no I/O can be in this region
 255          */
 256         release_resource(resource);
 257         kfree(resource);
 258 }
 259
 260 static struct resource *additional_memory_resource(phys_addr_t size)
 261 {
 262         struct resource *res;
 263         int ret;
 264
 265         res = kzalloc(sizeof(*res), GFP_KERNEL);
 266         if (!res)
 267                 return NULL;
 268
 269         res->name = "System RAM";
 270         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 271
 272         ret = allocate_resource(&iomem_resource, res,
 273                                 size, 0, -1,
 274                                 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
 275         if (ret < 0) {
 276                 pr_err("Cannot allocate new System RAM resource\n");
 277                 kfree(res);
 278                 return NULL;
 279         }
 280
 281 #ifdef CONFIG_SPARSEMEM
 282         {
 283                 unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
 284                 unsigned long pfn = res->start >> PAGE_SHIFT;
 285
 286                 if (pfn > limit) {
 287                         pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
 288                                pfn, limit);
 289                         release_memory_resource(res);
 290                         return NULL;
 291                 }
 292         }
 293 #endif
 294
 295         return res;
 296 }
 297
 298 static enum bp_state reserve_additional_memory(void)
 299 {
 300         long credit;
 301         struct resource *resource;
 302         int nid, rc;
 303         unsigned long balloon_hotplug;
 304
 305         credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
 306                 - balloon_stats.total_pages;
 307
 308         /*
 309          * Already hotplugged enough pages?  Wait for them to be
 310          * onlined.
 311          */
 312         if (credit <= 0)
 313                 return BP_WAIT;
 314
 315         balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
 316
 317         resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
 318         if (!resource)
 319                 goto err;
 320
 321         nid = memory_add_physaddr_to_nid(resource->start);
 322
 323 #ifdef CONFIG_XEN_HAVE_PVMMU
 324         /*
 325          * We don't support PV MMU when Linux and Xen is using
 326          * different page granularity.
 327          */
 328         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 329
 330         /*
 331          * add_memory() will build page tables for the new memory so
 332          * the p2m must contain invalid entries so the correct
 333          * non-present PTEs will be written.
 334          *
 335          * If a failure occurs, the original (identity) p2m entries
 336          * are not restored since this region is now known not to
 337          * conflict with any devices.
 338          */
 339         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 340                 unsigned long pfn, i;
 341
 342                 pfn = PFN_DOWN(resource->start);
 343                 for (i = 0; i < balloon_hotplug; i++) {
 344                         if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
 345                                 pr_warn("set_phys_to_machine() failed, no memory added\n");
 346                                 goto err;
 347                         }
 348                 }
 349         }
 350 #endif
 351
 352         /*
 353          * add_memory_resource() will call online_pages() which in its turn
 354          * will call xen_online_page() callback causing deadlock if we don't
 355          * release balloon_mutex here. Unlocking here is safe because the
 356          * callers drop the mutex before trying again.
 357          */
 358         mutex_unlock(&balloon_mutex);
 359         rc = add_memory_resource(nid, resource, memhp_auto_online);
 360         mutex_lock(&balloon_mutex);
 361
 362         if (rc) {
 363                 pr_warn("Cannot add additional memory (%i)\n", rc);
 364                 goto err;
 365         }
 366
 367         balloon_stats.total_pages += balloon_hotplug;
 368
 369         return BP_WAIT;
 370   err:
 371         release_memory_resource(resource);
 372         return BP_ECANCELED;
 373 }
 374
 375 static void xen_online_page(struct page *page)
 376 {
 377         __online_page_set_limits(page);
 378
 379         mutex_lock(&balloon_mutex);
 380
 381         __balloon_append(page);
 382
 383         mutex_unlock(&balloon_mutex);
 384 }
 385
 386 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
 387 {
 388         if (val == MEM_ONLINE)
 389                 schedule_delayed_work(&balloon_worker, 0);
 390
 391         return NOTIFY_OK;
 392 }
 393
 394 static struct notifier_block xen_memory_nb = {
 395         .notifier_call = xen_memory_notifier,
 396         .priority = 0
 397 };
 398 #else
 399 static enum bp_state reserve_additional_memory(void)
 400 {
 401         balloon_stats.target_pages = balloon_stats.current_pages;
 402         return BP_ECANCELED;
 403 }
 404 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
 405
 406 static long current_credit(void)
 407 {
 408         return balloon_stats.target_pages - balloon_stats.current_pages;
 409 }
 410
 411 static bool balloon_is_inflated(void)
 412 {
 413         return balloon_stats.balloon_low || balloon_stats.balloon_high;
 414 }
 415
 416 static enum bp_state increase_reservation(unsigned long nr_pages)
 417 {
 418         int rc;
 419         unsigned long i;
 420         struct page   *page;
 421         struct xen_memory_reservation reservation = {
 422                 .address_bits = 0,
 423                 .extent_order = EXTENT_ORDER,
 424                 .domid        = DOMID_SELF
 425         };
 426
 427         if (nr_pages > ARRAY_SIZE(frame_list))
 428                 nr_pages = ARRAY_SIZE(frame_list);
 429
 430         page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
 431         for (i = 0; i < nr_pages; i++) {
 432                 if (!page) {
 433                         nr_pages = i;
 434                         break;
 435                 }
 436
 437                 /* XENMEM_populate_physmap requires a PFN based on Xen
 438                  * granularity.
 439                  */
 440                 frame_list[i] = page_to_xen_pfn(page);
 441                 page = balloon_next_page(page);
 442         }
 443
 444         set_xen_guest_handle(reservation.extent_start, frame_list);
 445         reservation.nr_extents = nr_pages;
 446         rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
 447         if (rc <= 0)
 448                 return BP_EAGAIN;
 449
 450         for (i = 0; i < rc; i++) {
 451                 page = balloon_retrieve(false);
 452                 BUG_ON(page == NULL);
 453
 454 #ifdef CONFIG_XEN_HAVE_PVMMU
 455                 /*
 456                  * We don't support PV MMU when Linux and Xen is using
 457                  * different page granularity.
 458                  */
 459                 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 460
 461                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 462                         unsigned long pfn = page_to_pfn(page);
 463
 464                         set_phys_to_machine(pfn, frame_list[i]);
 465
 466                         /* Link back into the page tables if not highmem. */
 467                         if (!PageHighMem(page)) {
 468                                 int ret;
 469                                 ret = HYPERVISOR_update_va_mapping(
 470                                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 471                                                 mfn_pte(frame_list[i], PAGE_KERNEL),
 472                                                 0);
 473                                 BUG_ON(ret);
 474                         }
 475                 }
 476 #endif
 477
 478                 /* Relinquish the page back to the allocator. */
 479                 free_reserved_page(page);
 480         }
 481
 482         balloon_stats.current_pages += rc;
 483
 484         return BP_DONE;
 485 }
 486
 487 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
 488 {
 489         enum bp_state state = BP_DONE;
 490         unsigned long i;
 491         struct page *page, *tmp;
 492         int ret;
 493         struct xen_memory_reservation reservation = {
 494                 .address_bits = 0,
 495                 .extent_order = EXTENT_ORDER,
 496                 .domid        = DOMID_SELF
 497         };
 498         LIST_HEAD(pages);
 499
 500         if (nr_pages > ARRAY_SIZE(frame_list))
 501                 nr_pages = ARRAY_SIZE(frame_list);
 502
 503         for (i = 0; i < nr_pages; i++) {
 504                 page = alloc_page(gfp);
 505                 if (page == NULL) {
 506                         nr_pages = i;
 507                         state = BP_EAGAIN;
 508                         break;
 509                 }
 510                 adjust_managed_page_count(page, -1);
 511                 scrub_page(page);
 512                 list_add(&page->lru, &pages);
 513         }
 514
 515         /*
 516          * Ensure that ballooned highmem pages don't have kmaps.
 517          *
 518          * Do this before changing the p2m as kmap_flush_unused()
 519          * reads PTEs to obtain pages (and hence needs the original
 520          * p2m entry).
 521          */
 522         kmap_flush_unused();
 523
 524         /*
 525          * Setup the frame, update direct mapping, invalidate P2M,
 526          * and add to balloon.
 527          */
 528         i = 0;
 529         list_for_each_entry_safe(page, tmp, &pages, lru) {
 530                 /* XENMEM_decrease_reservation requires a GFN */
 531                 frame_list[i++] = xen_page_to_gfn(page);
 532
 533 #ifdef CONFIG_XEN_HAVE_PVMMU
 534                 /*
 535                  * We don't support PV MMU when Linux and Xen is using
 536                  * different page granularity.
 537                  */
 538                 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 539
 540                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 541                         unsigned long pfn = page_to_pfn(page);
 542
 543                         if (!PageHighMem(page)) {
 544                                 ret = HYPERVISOR_update_va_mapping(
 545                                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 546                                                 __pte_ma(0), 0);
 547                                 BUG_ON(ret);
 548                         }
 549                         __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 550                 }
 551 #endif
 552                 list_del(&page->lru);
 553
 554                 balloon_append(page);
 555         }
 556
 557         flush_tlb_all();
 558
 559         set_xen_guest_handle(reservation.extent_start, frame_list);
 560         reservation.nr_extents   = nr_pages;
 561         ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
 562         BUG_ON(ret != nr_pages);
 563
 564         balloon_stats.current_pages -= nr_pages;
 565
 566         return state;
 567 }
 568
 569 /*
 570  * As this is a work item it is guaranteed to run as a single instance only.
 571  * We may of course race updates of the target counts (which are protected
 572  * by the balloon lock), or with changes to the Xen hard limit, but we will
 573  * recover from these in time.
 574  */
 575 static void balloon_process(struct work_struct *work)
 576 {
 577         enum bp_state state = BP_DONE;
 578         long credit;
 579
 580
 581         do {
 582                 mutex_lock(&balloon_mutex);
 583
 584                 credit = current_credit();
 585
 586                 if (credit > 0) {
 587                         if (balloon_is_inflated())
 588                                 state = increase_reservation(credit);
 589                         else
 590                                 state = reserve_additional_memory();
 591                 }
 592
 593                 if (credit < 0)
 594                         state = decrease_reservation(-credit, GFP_BALLOON);
 595
 596                 state = update_schedule(state);
 597
 598                 mutex_unlock(&balloon_mutex);
 599
 600                 cond_resched();
 601
 602         } while (credit && state == BP_DONE);
 603
 604         /* Schedule more work if there is some still to be done. */
 605         if (state == BP_EAGAIN)
 606                 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
 607 }
 608
 609 /* Resets the Xen limit, sets new target, and kicks off processing. */
 610 void balloon_set_new_target(unsigned long target)
 611 {
 612         /* No need for lock. Not read-modify-write updates. */
 613         balloon_stats.target_pages = target;
 614         schedule_delayed_work(&balloon_worker, 0);
 615 }
 616 EXPORT_SYMBOL_GPL(balloon_set_new_target);
 617
 618 static int add_ballooned_pages(int nr_pages)
 619 {
 620         enum bp_state st;
 621
 622         if (xen_hotplug_unpopulated) {
 623                 st = reserve_additional_memory();
 624                 if (st != BP_ECANCELED) {
 625                         mutex_unlock(&balloon_mutex);
 626                         wait_event(balloon_wq,
 627                                    !list_empty(&ballooned_pages));
 628                         mutex_lock(&balloon_mutex);
 629                         return 0;
 630                 }
 631         }
 632
 633         st = decrease_reservation(nr_pages, GFP_USER);
 634         if (st != BP_DONE)
 635                 return -ENOMEM;
 636
 637         return 0;
 638 }
 639
 640 /**
 641  * alloc_xenballooned_pages - get pages that have been ballooned out
 642  * @nr_pages: Number of pages to get
 643  * @pages: pages returned
 644  * @return 0 on success, error otherwise
 645  */
 646 int alloc_xenballooned_pages(int nr_pages, struct page **pages)
 647 {
 648         int pgno = 0;
 649         struct page *page;
 650         int ret;
 651
 652         mutex_lock(&balloon_mutex);
 653
 654         balloon_stats.target_unpopulated += nr_pages;
 655
 656         while (pgno < nr_pages) {
 657                 page = balloon_retrieve(true);
 658                 if (page) {
 659                         pages[pgno++] = page;
 660 #ifdef CONFIG_XEN_HAVE_PVMMU
 661                         /*
 662                          * We don't support PV MMU when Linux and Xen is using
 663                          * different page granularity.
 664                          */
 665                         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 666
 667                         ret = xen_alloc_p2m_entry(page_to_pfn(page));
 668                         if (ret < 0)
 669                                 goto out_undo;
 670 #endif
 671                 } else {
 672                         ret = add_ballooned_pages(nr_pages - pgno);
 673                         if (ret < 0)
 674                                 goto out_undo;
 675                 }
 676         }
 677         mutex_unlock(&balloon_mutex);
 678         return 0;
 679  out_undo:
 680         mutex_unlock(&balloon_mutex);
 681         free_xenballooned_pages(pgno, pages);
 682         return ret;
 683 }
 684 EXPORT_SYMBOL(alloc_xenballooned_pages);
 685
 686 /**
 687  * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
 688  * @nr_pages: Number of pages
 689  * @pages: pages to return
 690  */
 691 void free_xenballooned_pages(int nr_pages, struct page **pages)
 692 {
 693         int i;
 694
 695         mutex_lock(&balloon_mutex);
 696
 697         for (i = 0; i < nr_pages; i++) {
 698                 if (pages[i])
 699                         balloon_append(pages[i]);
 700         }
 701
 702         balloon_stats.target_unpopulated -= nr_pages;
 703
 704         /* The balloon may be too large now. Shrink it if needed. */
 705         if (current_credit())
 706                 schedule_delayed_work(&balloon_worker, 0);
 707
 708         mutex_unlock(&balloon_mutex);
 709 }
 710 EXPORT_SYMBOL(free_xenballooned_pages);
 711
 712 #ifdef CONFIG_XEN_PV
 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 #endif
 737
 738 static int __init balloon_init(void)
 739 {
 740         if (!xen_domain())
 741                 return -ENODEV;
 742
 743         pr_info("Initialising balloon driver\n");
 744
 745 #ifdef CONFIG_XEN_PV
 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 #else
 750         balloon_stats.current_pages = get_num_physpages();
 751 #endif
 752         balloon_stats.target_pages  = balloon_stats.current_pages;
 753         balloon_stats.balloon_low   = 0;
 754         balloon_stats.balloon_high  = 0;
 755         balloon_stats.total_pages   = balloon_stats.current_pages;
 756
 757         balloon_stats.schedule_delay = 1;
 758         balloon_stats.max_schedule_delay = 32;
 759         balloon_stats.retry_count = 1;
 760         balloon_stats.max_retry_count = RETRY_UNLIMITED;
 761
 762 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 763         set_online_page_callback(&xen_online_page);
 764         register_memory_notifier(&xen_memory_nb);
 765         register_sysctl_table(xen_root);
 766 #endif
 767
 768 #ifdef CONFIG_XEN_PV
 769         {
 770                 int i;
 771
 772                 /*
 773                  * Initialize the balloon with pages from the extra memory
 774                  * regions (see arch/x86/xen/setup.c).
 775                  */
 776                 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
 777                         if (xen_extra_mem[i].n_pfns)
 778                                 balloon_add_region(xen_extra_mem[i].start_pfn,
 779                                                    xen_extra_mem[i].n_pfns);
 780         }
 781 #endif
 782
 783         return 0;
 784 }
 785 subsys_initcall(balloon_init);