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 struct resource *additional_memory_resource(phys_addr_t size)
 250 {
 251         struct resource *res;
 252         int ret;
 253
 254         res = kzalloc(sizeof(*res), GFP_KERNEL);
 255         if (!res)
 256                 return NULL;
 257
 258         res->name = "System RAM";
 259         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 260
 261         ret = allocate_resource(&iomem_resource, res,
 262                                 size, 0, -1,
 263                                 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
 264         if (ret < 0) {
 265                 pr_err("Cannot allocate new System RAM resource\n");
 266                 kfree(res);
 267                 return NULL;
 268         }
 269
 270         return res;
 271 }
 272
 273 static void release_memory_resource(struct resource *resource)
 274 {
 275         if (!resource)
 276                 return;
 277
 278         /*
 279          * No need to reset region to identity mapped since we now
 280          * know that no I/O can be in this region
 281          */
 282         release_resource(resource);
 283         kfree(resource);
 284 }
 285
 286 static enum bp_state reserve_additional_memory(void)
 287 {
 288         long credit;
 289         struct resource *resource;
 290         int nid, rc;
 291         unsigned long balloon_hotplug;
 292
 293         credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
 294                 - balloon_stats.total_pages;
 295
 296         /*
 297          * Already hotplugged enough pages?  Wait for them to be
 298          * onlined.
 299          */
 300         if (credit <= 0)
 301                 return BP_WAIT;
 302
 303         balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
 304
 305         resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
 306         if (!resource)
 307                 goto err;
 308
 309         nid = memory_add_physaddr_to_nid(resource->start);
 310
 311 #ifdef CONFIG_XEN_HAVE_PVMMU
 312         /*
 313          * We don't support PV MMU when Linux and Xen is using
 314          * different page granularity.
 315          */
 316         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 317
 318         /*
 319          * add_memory() will build page tables for the new memory so
 320          * the p2m must contain invalid entries so the correct
 321          * non-present PTEs will be written.
 322          *
 323          * If a failure occurs, the original (identity) p2m entries
 324          * are not restored since this region is now known not to
 325          * conflict with any devices.
 326          */
 327         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 328                 unsigned long pfn, i;
 329
 330                 pfn = PFN_DOWN(resource->start);
 331                 for (i = 0; i < balloon_hotplug; i++) {
 332                         if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
 333                                 pr_warn("set_phys_to_machine() failed, no memory added\n");
 334                                 goto err;
 335                         }
 336                 }
 337         }
 338 #endif
 339
 340         /*
 341          * add_memory_resource() will call online_pages() which in its turn
 342          * will call xen_online_page() callback causing deadlock if we don't
 343          * release balloon_mutex here. Unlocking here is safe because the
 344          * callers drop the mutex before trying again.
 345          */
 346         mutex_unlock(&balloon_mutex);
 347         rc = add_memory_resource(nid, resource, memhp_auto_online);
 348         mutex_lock(&balloon_mutex);
 349
 350         if (rc) {
 351                 pr_warn("Cannot add additional memory (%i)\n", rc);
 352                 goto err;
 353         }
 354
 355         balloon_stats.total_pages += balloon_hotplug;
 356
 357         return BP_WAIT;
 358   err:
 359         release_memory_resource(resource);
 360         return BP_ECANCELED;
 361 }
 362
 363 static void xen_online_page(struct page *page)
 364 {
 365         __online_page_set_limits(page);
 366
 367         mutex_lock(&balloon_mutex);
 368
 369         __balloon_append(page);
 370
 371         mutex_unlock(&balloon_mutex);
 372 }
 373
 374 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
 375 {
 376         if (val == MEM_ONLINE)
 377                 schedule_delayed_work(&balloon_worker, 0);
 378
 379         return NOTIFY_OK;
 380 }
 381
 382 static struct notifier_block xen_memory_nb = {
 383         .notifier_call = xen_memory_notifier,
 384         .priority = 0
 385 };
 386 #else
 387 static enum bp_state reserve_additional_memory(void)
 388 {
 389         balloon_stats.target_pages = balloon_stats.current_pages;
 390         return BP_ECANCELED;
 391 }
 392 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
 393
 394 static long current_credit(void)
 395 {
 396         return balloon_stats.target_pages - balloon_stats.current_pages;
 397 }
 398
 399 static bool balloon_is_inflated(void)
 400 {
 401         return balloon_stats.balloon_low || balloon_stats.balloon_high;
 402 }
 403
 404 static enum bp_state increase_reservation(unsigned long nr_pages)
 405 {
 406         int rc;
 407         unsigned long i;
 408         struct page   *page;
 409         struct xen_memory_reservation reservation = {
 410                 .address_bits = 0,
 411                 .extent_order = EXTENT_ORDER,
 412                 .domid        = DOMID_SELF
 413         };
 414
 415         if (nr_pages > ARRAY_SIZE(frame_list))
 416                 nr_pages = ARRAY_SIZE(frame_list);
 417
 418         page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
 419         for (i = 0; i < nr_pages; i++) {
 420                 if (!page) {
 421                         nr_pages = i;
 422                         break;
 423                 }
 424
 425                 /* XENMEM_populate_physmap requires a PFN based on Xen
 426                  * granularity.
 427                  */
 428                 frame_list[i] = page_to_xen_pfn(page);
 429                 page = balloon_next_page(page);
 430         }
 431
 432         set_xen_guest_handle(reservation.extent_start, frame_list);
 433         reservation.nr_extents = nr_pages;
 434         rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
 435         if (rc <= 0)
 436                 return BP_EAGAIN;
 437
 438         for (i = 0; i < rc; i++) {
 439                 page = balloon_retrieve(false);
 440                 BUG_ON(page == NULL);
 441
 442 #ifdef CONFIG_XEN_HAVE_PVMMU
 443                 /*
 444                  * We don't support PV MMU when Linux and Xen is using
 445                  * different page granularity.
 446                  */
 447                 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 448
 449                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 450                         unsigned long pfn = page_to_pfn(page);
 451
 452                         set_phys_to_machine(pfn, frame_list[i]);
 453
 454                         /* Link back into the page tables if not highmem. */
 455                         if (!PageHighMem(page)) {
 456                                 int ret;
 457                                 ret = HYPERVISOR_update_va_mapping(
 458                                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 459                                                 mfn_pte(frame_list[i], PAGE_KERNEL),
 460                                                 0);
 461                                 BUG_ON(ret);
 462                         }
 463                 }
 464 #endif
 465
 466                 /* Relinquish the page back to the allocator. */
 467                 __free_reserved_page(page);
 468         }
 469
 470         balloon_stats.current_pages += rc;
 471
 472         return BP_DONE;
 473 }
 474
 475 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
 476 {
 477         enum bp_state state = BP_DONE;
 478         unsigned long i;
 479         struct page *page, *tmp;
 480         int ret;
 481         struct xen_memory_reservation reservation = {
 482                 .address_bits = 0,
 483                 .extent_order = EXTENT_ORDER,
 484                 .domid        = DOMID_SELF
 485         };
 486         LIST_HEAD(pages);
 487
 488         if (nr_pages > ARRAY_SIZE(frame_list))
 489                 nr_pages = ARRAY_SIZE(frame_list);
 490
 491         for (i = 0; i < nr_pages; i++) {
 492                 page = alloc_page(gfp);
 493                 if (page == NULL) {
 494                         nr_pages = i;
 495                         state = BP_EAGAIN;
 496                         break;
 497                 }
 498                 scrub_page(page);
 499                 list_add(&page->lru, &pages);
 500         }
 501
 502         /*
 503          * Ensure that ballooned highmem pages don't have kmaps.
 504          *
 505          * Do this before changing the p2m as kmap_flush_unused()
 506          * reads PTEs to obtain pages (and hence needs the original
 507          * p2m entry).
 508          */
 509         kmap_flush_unused();
 510
 511         /*
 512          * Setup the frame, update direct mapping, invalidate P2M,
 513          * and add to balloon.
 514          */
 515         i = 0;
 516         list_for_each_entry_safe(page, tmp, &pages, lru) {
 517                 /* XENMEM_decrease_reservation requires a GFN */
 518                 frame_list[i++] = xen_page_to_gfn(page);
 519
 520 #ifdef CONFIG_XEN_HAVE_PVMMU
 521                 /*
 522                  * We don't support PV MMU when Linux and Xen is using
 523                  * different page granularity.
 524                  */
 525                 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 526
 527                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 528                         unsigned long pfn = page_to_pfn(page);
 529
 530                         if (!PageHighMem(page)) {
 531                                 ret = HYPERVISOR_update_va_mapping(
 532                                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 533                                                 __pte_ma(0), 0);
 534                                 BUG_ON(ret);
 535                         }
 536                         __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 537                 }
 538 #endif
 539                 list_del(&page->lru);
 540
 541                 balloon_append(page);
 542         }
 543
 544         flush_tlb_all();
 545
 546         set_xen_guest_handle(reservation.extent_start, frame_list);
 547         reservation.nr_extents   = nr_pages;
 548         ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
 549         BUG_ON(ret != nr_pages);
 550
 551         balloon_stats.current_pages -= nr_pages;
 552
 553         return state;
 554 }
 555
 556 /*
 557  * As this is a work item it is guaranteed to run as a single instance only.
 558  * We may of course race updates of the target counts (which are protected
 559  * by the balloon lock), or with changes to the Xen hard limit, but we will
 560  * recover from these in time.
 561  */
 562 static void balloon_process(struct work_struct *work)
 563 {
 564         enum bp_state state = BP_DONE;
 565         long credit;
 566
 567
 568         do {
 569                 mutex_lock(&balloon_mutex);
 570
 571                 credit = current_credit();
 572
 573                 if (credit > 0) {
 574                         if (balloon_is_inflated())
 575                                 state = increase_reservation(credit);
 576                         else
 577                                 state = reserve_additional_memory();
 578                 }
 579
 580                 if (credit < 0)
 581                         state = decrease_reservation(-credit, GFP_BALLOON);
 582
 583                 state = update_schedule(state);
 584
 585                 mutex_unlock(&balloon_mutex);
 586
 587                 cond_resched();
 588
 589         } while (credit && state == BP_DONE);
 590
 591         /* Schedule more work if there is some still to be done. */
 592         if (state == BP_EAGAIN)
 593                 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
 594 }
 595
 596 /* Resets the Xen limit, sets new target, and kicks off processing. */
 597 void balloon_set_new_target(unsigned long target)
 598 {
 599         /* No need for lock. Not read-modify-write updates. */
 600         balloon_stats.target_pages = target;
 601         schedule_delayed_work(&balloon_worker, 0);
 602 }
 603 EXPORT_SYMBOL_GPL(balloon_set_new_target);
 604
 605 static int add_ballooned_pages(int nr_pages)
 606 {
 607         enum bp_state st;
 608
 609         if (xen_hotplug_unpopulated) {
 610                 st = reserve_additional_memory();
 611                 if (st != BP_ECANCELED) {
 612                         mutex_unlock(&balloon_mutex);
 613                         wait_event(balloon_wq,
 614                                    !list_empty(&ballooned_pages));
 615                         mutex_lock(&balloon_mutex);
 616                         return 0;
 617                 }
 618         }
 619
 620         st = decrease_reservation(nr_pages, GFP_USER);
 621         if (st != BP_DONE)
 622                 return -ENOMEM;
 623
 624         return 0;
 625 }
 626
 627 /**
 628  * alloc_xenballooned_pages - get pages that have been ballooned out
 629  * @nr_pages: Number of pages to get
 630  * @pages: pages returned
 631  * @return 0 on success, error otherwise
 632  */
 633 int alloc_xenballooned_pages(int nr_pages, struct page **pages)
 634 {
 635         int pgno = 0;
 636         struct page *page;
 637         int ret;
 638
 639         mutex_lock(&balloon_mutex);
 640
 641         balloon_stats.target_unpopulated += nr_pages;
 642
 643         while (pgno < nr_pages) {
 644                 page = balloon_retrieve(true);
 645                 if (page) {
 646                         pages[pgno++] = page;
 647 #ifdef CONFIG_XEN_HAVE_PVMMU
 648                         /*
 649                          * We don't support PV MMU when Linux and Xen is using
 650                          * different page granularity.
 651                          */
 652                         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
 653
 654                         ret = xen_alloc_p2m_entry(page_to_pfn(page));
 655                         if (ret < 0)
 656                                 goto out_undo;
 657 #endif
 658                 } else {
 659                         ret = add_ballooned_pages(nr_pages - pgno);
 660                         if (ret < 0)
 661                                 goto out_undo;
 662                 }
 663         }
 664         mutex_unlock(&balloon_mutex);
 665         return 0;
 666  out_undo:
 667         mutex_unlock(&balloon_mutex);
 668         free_xenballooned_pages(pgno, pages);
 669         return ret;
 670 }
 671 EXPORT_SYMBOL(alloc_xenballooned_pages);
 672
 673 /**
 674  * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
 675  * @nr_pages: Number of pages
 676  * @pages: pages to return
 677  */
 678 void free_xenballooned_pages(int nr_pages, struct page **pages)
 679 {
 680         int i;
 681
 682         mutex_lock(&balloon_mutex);
 683
 684         for (i = 0; i < nr_pages; i++) {
 685                 if (pages[i])
 686                         balloon_append(pages[i]);
 687         }
 688
 689         balloon_stats.target_unpopulated -= nr_pages;
 690
 691         /* The balloon may be too large now. Shrink it if needed. */
 692         if (current_credit())
 693                 schedule_delayed_work(&balloon_worker, 0);
 694
 695         mutex_unlock(&balloon_mutex);
 696 }
 697 EXPORT_SYMBOL(free_xenballooned_pages);
 698
 699 static void __init balloon_add_region(unsigned long start_pfn,
 700                                       unsigned long pages)
 701 {
 702         unsigned long pfn, extra_pfn_end;
 703         struct page *page;
 704
 705         /*
 706          * If the amount of usable memory has been limited (e.g., with
 707          * the 'mem' command line parameter), don't add pages beyond
 708          * this limit.
 709          */
 710         extra_pfn_end = min(max_pfn, start_pfn + pages);
 711
 712         for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
 713                 page = pfn_to_page(pfn);
 714                 /* totalram_pages and totalhigh_pages do not
 715                    include the boot-time balloon extension, so
 716                    don't subtract from it. */
 717                 __balloon_append(page);
 718         }
 719
 720         balloon_stats.total_pages += extra_pfn_end - start_pfn;
 721 }
 722
 723 static int __init balloon_init(void)
 724 {
 725         int i;
 726
 727         if (!xen_domain())
 728                 return -ENODEV;
 729
 730         pr_info("Initialising balloon driver\n");
 731
 732         balloon_stats.current_pages = xen_pv_domain()
 733                 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
 734                 : get_num_physpages();
 735         balloon_stats.target_pages  = balloon_stats.current_pages;
 736         balloon_stats.balloon_low   = 0;
 737         balloon_stats.balloon_high  = 0;
 738         balloon_stats.total_pages   = balloon_stats.current_pages;
 739
 740         balloon_stats.schedule_delay = 1;
 741         balloon_stats.max_schedule_delay = 32;
 742         balloon_stats.retry_count = 1;
 743         balloon_stats.max_retry_count = RETRY_UNLIMITED;
 744
 745 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 746         set_online_page_callback(&xen_online_page);
 747         register_memory_notifier(&xen_memory_nb);
 748         register_sysctl_table(xen_root);
 749 #endif
 750
 751         /*
 752          * Initialize the balloon with pages from the extra memory
 753          * regions (see arch/x86/xen/setup.c).
 754          */
 755         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
 756                 if (xen_extra_mem[i].n_pfns)
 757                         balloon_add_region(xen_extra_mem[i].start_pfn,
 758                                            xen_extra_mem[i].n_pfns);
 759
 760         return 0;
 761 }
 762 subsys_initcall(balloon_init);