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 #include <linux/kernel.h>
  40 #include <linux/sched.h>
  41 #include <linux/errno.h>
  42 #include <linux/mm.h>
  43 #include <linux/bootmem.h>
  44 #include <linux/pagemap.h>
  45 #include <linux/highmem.h>
  46 #include <linux/mutex.h>
  47 #include <linux/list.h>
  48 #include <linux/gfp.h>
  49 #include <linux/notifier.h>
  50 #include <linux/memory.h>
  51 #include <linux/memory_hotplug.h>
  52
  53 #include <asm/page.h>
  54 #include <asm/pgalloc.h>
  55 #include <asm/pgtable.h>
  56 #include <asm/tlb.h>
  57 #include <asm/e820.h>
  58
  59 #include <asm/xen/hypervisor.h>
  60 #include <asm/xen/hypercall.h>
  61
  62 #include <xen/xen.h>
  63 #include <xen/interface/xen.h>
  64 #include <xen/interface/memory.h>
  65 #include <xen/balloon.h>
  66 #include <xen/features.h>
  67 #include <xen/page.h>
  68
  69 /*
  70  * balloon_process() state:
  71  *
  72  * BP_DONE: done or nothing to do,
  73  * BP_EAGAIN: error, go to sleep,
  74  * BP_ECANCELED: error, balloon operation canceled.
  75  */
  76
  77 enum bp_state {
  78         BP_DONE,
  79         BP_EAGAIN,
  80         BP_ECANCELED
  81 };
  82
  83
  84 static DEFINE_MUTEX(balloon_mutex);
  85
  86 struct balloon_stats balloon_stats;
  87 EXPORT_SYMBOL_GPL(balloon_stats);
  88
  89 /* We increase/decrease in batches which fit in a page */
  90 static unsigned long frame_list[PAGE_SIZE / sizeof(unsigned long)];
  91
  92 #ifdef CONFIG_HIGHMEM
  93 #define inc_totalhigh_pages() (totalhigh_pages++)
  94 #define dec_totalhigh_pages() (totalhigh_pages--)
  95 #else
  96 #define inc_totalhigh_pages() do {} while(0)
  97 #define dec_totalhigh_pages() do {} while(0)
  98 #endif
  99
 100 /* List of ballooned pages, threaded through the mem_map array. */
 101 static LIST_HEAD(ballooned_pages);
 102
 103 /* Main work function, always executed in process context. */
 104 static void balloon_process(struct work_struct *work);
 105 static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
 106
 107 /* When ballooning out (allocating memory to return to Xen) we don't really
 108    want the kernel to try too hard since that can trigger the oom killer. */
 109 #define GFP_BALLOON \
 110         (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
 111
 112 static void scrub_page(struct page *page)
 113 {
 114 #ifdef CONFIG_XEN_SCRUB_PAGES
 115         clear_highpage(page);
 116 #endif
 117 }
 118
 119 /* balloon_append: add the given page to the balloon. */
 120 static void __balloon_append(struct page *page)
 121 {
 122         /* Lowmem is re-populated first, so highmem pages go at list tail. */
 123         if (PageHighMem(page)) {
 124                 list_add_tail(&page->lru, &ballooned_pages);
 125                 balloon_stats.balloon_high++;
 126         } else {
 127                 list_add(&page->lru, &ballooned_pages);
 128                 balloon_stats.balloon_low++;
 129         }
 130 }
 131
 132 static void balloon_append(struct page *page)
 133 {
 134         __balloon_append(page);
 135         if (PageHighMem(page))
 136                 dec_totalhigh_pages();
 137         totalram_pages--;
 138 }
 139
 140 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
 141 static struct page *balloon_retrieve(bool prefer_highmem)
 142 {
 143         struct page *page;
 144
 145         if (list_empty(&ballooned_pages))
 146                 return NULL;
 147
 148         if (prefer_highmem)
 149                 page = list_entry(ballooned_pages.prev, struct page, lru);
 150         else
 151                 page = list_entry(ballooned_pages.next, struct page, lru);
 152         list_del(&page->lru);
 153
 154         if (PageHighMem(page)) {
 155                 balloon_stats.balloon_high--;
 156                 inc_totalhigh_pages();
 157         }
 158         else
 159                 balloon_stats.balloon_low--;
 160
 161         totalram_pages++;
 162
 163         return page;
 164 }
 165
 166 static struct page *balloon_first_page(void)
 167 {
 168         if (list_empty(&ballooned_pages))
 169                 return NULL;
 170         return list_entry(ballooned_pages.next, struct page, lru);
 171 }
 172
 173 static struct page *balloon_next_page(struct page *page)
 174 {
 175         struct list_head *next = page->lru.next;
 176         if (next == &ballooned_pages)
 177                 return NULL;
 178         return list_entry(next, struct page, lru);
 179 }
 180
 181 static enum bp_state update_schedule(enum bp_state state)
 182 {
 183         if (state == BP_DONE) {
 184                 balloon_stats.schedule_delay = 1;
 185                 balloon_stats.retry_count = 1;
 186                 return BP_DONE;
 187         }
 188
 189         ++balloon_stats.retry_count;
 190
 191         if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
 192                         balloon_stats.retry_count > balloon_stats.max_retry_count) {
 193                 balloon_stats.schedule_delay = 1;
 194                 balloon_stats.retry_count = 1;
 195                 return BP_ECANCELED;
 196         }
 197
 198         balloon_stats.schedule_delay <<= 1;
 199
 200         if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
 201                 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
 202
 203         return BP_EAGAIN;
 204 }
 205
 206 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 207 static long current_credit(void)
 208 {
 209         return balloon_stats.target_pages - balloon_stats.current_pages -
 210                 balloon_stats.hotplug_pages;
 211 }
 212
 213 static bool balloon_is_inflated(void)
 214 {
 215         if (balloon_stats.balloon_low || balloon_stats.balloon_high ||
 216                         balloon_stats.balloon_hotplug)
 217                 return true;
 218         else
 219                 return false;
 220 }
 221
 222 /*
 223  * reserve_additional_memory() adds memory region of size >= credit above
 224  * max_pfn. New region is section aligned and size is modified to be multiple
 225  * of section size. Those features allow optimal use of address space and
 226  * establish proper alignment when this function is called first time after
 227  * boot (last section not fully populated at boot time contains unused memory
 228  * pages with PG_reserved bit not set; online_pages_range() does not allow page
 229  * onlining in whole range if first onlined page does not have PG_reserved
 230  * bit set). Real size of added memory is established at page onlining stage.
 231  */
 232
 233 static enum bp_state reserve_additional_memory(long credit)
 234 {
 235         int nid, rc;
 236         u64 hotplug_start_paddr;
 237         unsigned long balloon_hotplug = credit;
 238
 239         hotplug_start_paddr = PFN_PHYS(SECTION_ALIGN_UP(max_pfn));
 240         balloon_hotplug = round_up(balloon_hotplug, PAGES_PER_SECTION);
 241         nid = memory_add_physaddr_to_nid(hotplug_start_paddr);
 242
 243         rc = add_memory(nid, hotplug_start_paddr, balloon_hotplug << PAGE_SHIFT);
 244
 245         if (rc) {
 246                 pr_info("xen_balloon: %s: add_memory() failed: %i\n", __func__, rc);
 247                 return BP_EAGAIN;
 248         }
 249
 250         balloon_hotplug -= credit;
 251
 252         balloon_stats.hotplug_pages += credit;
 253         balloon_stats.balloon_hotplug = balloon_hotplug;
 254
 255         return BP_DONE;
 256 }
 257
 258 static void xen_online_page(struct page *page)
 259 {
 260         __online_page_set_limits(page);
 261
 262         mutex_lock(&balloon_mutex);
 263
 264         __balloon_append(page);
 265
 266         if (balloon_stats.hotplug_pages)
 267                 --balloon_stats.hotplug_pages;
 268         else
 269                 --balloon_stats.balloon_hotplug;
 270
 271         mutex_unlock(&balloon_mutex);
 272 }
 273
 274 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
 275 {
 276         if (val == MEM_ONLINE)
 277                 schedule_delayed_work(&balloon_worker, 0);
 278
 279         return NOTIFY_OK;
 280 }
 281
 282 static struct notifier_block xen_memory_nb = {
 283         .notifier_call = xen_memory_notifier,
 284         .priority = 0
 285 };
 286 #else
 287 static long current_credit(void)
 288 {
 289         unsigned long target = balloon_stats.target_pages;
 290
 291         target = min(target,
 292                      balloon_stats.current_pages +
 293                      balloon_stats.balloon_low +
 294                      balloon_stats.balloon_high);
 295
 296         return target - balloon_stats.current_pages;
 297 }
 298
 299 static bool balloon_is_inflated(void)
 300 {
 301         if (balloon_stats.balloon_low || balloon_stats.balloon_high)
 302                 return true;
 303         else
 304                 return false;
 305 }
 306
 307 static enum bp_state reserve_additional_memory(long credit)
 308 {
 309         balloon_stats.target_pages = balloon_stats.current_pages;
 310         return BP_DONE;
 311 }
 312 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
 313
 314 static enum bp_state increase_reservation(unsigned long nr_pages)
 315 {
 316         int rc;
 317         unsigned long  pfn, i;
 318         struct page   *page;
 319         struct xen_memory_reservation reservation = {
 320                 .address_bits = 0,
 321                 .extent_order = 0,
 322                 .domid        = DOMID_SELF
 323         };
 324
 325 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 326         if (!balloon_stats.balloon_low && !balloon_stats.balloon_high) {
 327                 nr_pages = min(nr_pages, balloon_stats.balloon_hotplug);
 328                 balloon_stats.hotplug_pages += nr_pages;
 329                 balloon_stats.balloon_hotplug -= nr_pages;
 330                 return BP_DONE;
 331         }
 332 #endif
 333
 334         if (nr_pages > ARRAY_SIZE(frame_list))
 335                 nr_pages = ARRAY_SIZE(frame_list);
 336
 337         page = balloon_first_page();
 338         for (i = 0; i < nr_pages; i++) {
 339                 if (!page) {
 340                         nr_pages = i;
 341                         break;
 342                 }
 343                 frame_list[i] = page_to_pfn(page);
 344                 page = balloon_next_page(page);
 345         }
 346
 347         set_xen_guest_handle(reservation.extent_start, frame_list);
 348         reservation.nr_extents = nr_pages;
 349         rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
 350         if (rc <= 0)
 351                 return BP_EAGAIN;
 352
 353         for (i = 0; i < rc; i++) {
 354                 page = balloon_retrieve(false);
 355                 BUG_ON(page == NULL);
 356
 357                 pfn = page_to_pfn(page);
 358                 BUG_ON(!xen_feature(XENFEAT_auto_translated_physmap) &&
 359                        phys_to_machine_mapping_valid(pfn));
 360
 361                 set_phys_to_machine(pfn, frame_list[i]);
 362
 363                 /* Link back into the page tables if not highmem. */
 364                 if (xen_pv_domain() && !PageHighMem(page)) {
 365                         int ret;
 366                         ret = HYPERVISOR_update_va_mapping(
 367                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 368                                 mfn_pte(frame_list[i], PAGE_KERNEL),
 369                                 0);
 370                         BUG_ON(ret);
 371                 }
 372
 373                 /* Relinquish the page back to the allocator. */
 374                 ClearPageReserved(page);
 375                 init_page_count(page);
 376                 __free_page(page);
 377         }
 378
 379         balloon_stats.current_pages += rc;
 380
 381         return BP_DONE;
 382 }
 383
 384 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
 385 {
 386         enum bp_state state = BP_DONE;
 387         unsigned long  pfn, i;
 388         struct page   *page;
 389         int ret;
 390         struct xen_memory_reservation reservation = {
 391                 .address_bits = 0,
 392                 .extent_order = 0,
 393                 .domid        = DOMID_SELF
 394         };
 395
 396 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 397         if (balloon_stats.hotplug_pages) {
 398                 nr_pages = min(nr_pages, balloon_stats.hotplug_pages);
 399                 balloon_stats.hotplug_pages -= nr_pages;
 400                 balloon_stats.balloon_hotplug += nr_pages;
 401                 return BP_DONE;
 402         }
 403 #endif
 404
 405         if (nr_pages > ARRAY_SIZE(frame_list))
 406                 nr_pages = ARRAY_SIZE(frame_list);
 407
 408         for (i = 0; i < nr_pages; i++) {
 409                 if ((page = alloc_page(gfp)) == NULL) {
 410                         nr_pages = i;
 411                         state = BP_EAGAIN;
 412                         break;
 413                 }
 414
 415                 pfn = page_to_pfn(page);
 416                 frame_list[i] = pfn_to_mfn(pfn);
 417
 418                 scrub_page(page);
 419
 420                 if (xen_pv_domain() && !PageHighMem(page)) {
 421                         ret = HYPERVISOR_update_va_mapping(
 422                                 (unsigned long)__va(pfn << PAGE_SHIFT),
 423                                 __pte_ma(0), 0);
 424                         BUG_ON(ret);
 425                 }
 426
 427         }
 428
 429         /* Ensure that ballooned highmem pages don't have kmaps. */
 430         kmap_flush_unused();
 431         flush_tlb_all();
 432
 433         /* No more mappings: invalidate P2M and add to balloon. */
 434         for (i = 0; i < nr_pages; i++) {
 435                 pfn = mfn_to_pfn(frame_list[i]);
 436                 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 437                 balloon_append(pfn_to_page(pfn));
 438         }
 439
 440         set_xen_guest_handle(reservation.extent_start, frame_list);
 441         reservation.nr_extents   = nr_pages;
 442         ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
 443         BUG_ON(ret != nr_pages);
 444
 445         balloon_stats.current_pages -= nr_pages;
 446
 447         return state;
 448 }
 449
 450 /*
 451  * We avoid multiple worker processes conflicting via the balloon mutex.
 452  * We may of course race updates of the target counts (which are protected
 453  * by the balloon lock), or with changes to the Xen hard limit, but we will
 454  * recover from these in time.
 455  */
 456 static void balloon_process(struct work_struct *work)
 457 {
 458         enum bp_state state = BP_DONE;
 459         long credit;
 460
 461         mutex_lock(&balloon_mutex);
 462
 463         do {
 464                 credit = current_credit();
 465
 466                 if (credit > 0) {
 467                         if (balloon_is_inflated())
 468                                 state = increase_reservation(credit);
 469                         else
 470                                 state = reserve_additional_memory(credit);
 471                 }
 472
 473                 if (credit < 0)
 474                         state = decrease_reservation(-credit, GFP_BALLOON);
 475
 476                 state = update_schedule(state);
 477
 478 #ifndef CONFIG_PREEMPT
 479                 if (need_resched())
 480                         schedule();
 481 #endif
 482         } while (credit && state == BP_DONE);
 483
 484         /* Schedule more work if there is some still to be done. */
 485         if (state == BP_EAGAIN)
 486                 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
 487
 488         mutex_unlock(&balloon_mutex);
 489 }
 490
 491 /* Resets the Xen limit, sets new target, and kicks off processing. */
 492 void balloon_set_new_target(unsigned long target)
 493 {
 494         /* No need for lock. Not read-modify-write updates. */
 495         balloon_stats.target_pages = target;
 496         schedule_delayed_work(&balloon_worker, 0);
 497 }
 498 EXPORT_SYMBOL_GPL(balloon_set_new_target);
 499
 500 /**
 501  * alloc_xenballooned_pages - get pages that have been ballooned out
 502  * @nr_pages: Number of pages to get
 503  * @pages: pages returned
 504  * @return 0 on success, error otherwise
 505  */
 506 int alloc_xenballooned_pages(int nr_pages, struct page** pages)
 507 {
 508         int pgno = 0;
 509         struct page* page;
 510         mutex_lock(&balloon_mutex);
 511         while (pgno < nr_pages) {
 512                 page = balloon_retrieve(true);
 513                 if (page) {
 514                         pages[pgno++] = page;
 515                 } else {
 516                         enum bp_state st;
 517                         st = decrease_reservation(nr_pages - pgno, GFP_HIGHUSER);
 518                         if (st != BP_DONE)
 519                                 goto out_undo;
 520                 }
 521         }
 522         mutex_unlock(&balloon_mutex);
 523         return 0;
 524  out_undo:
 525         while (pgno)
 526                 balloon_append(pages[--pgno]);
 527         /* Free the memory back to the kernel soon */
 528         schedule_delayed_work(&balloon_worker, 0);
 529         mutex_unlock(&balloon_mutex);
 530         return -ENOMEM;
 531 }
 532 EXPORT_SYMBOL(alloc_xenballooned_pages);
 533
 534 /**
 535  * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
 536  * @nr_pages: Number of pages
 537  * @pages: pages to return
 538  */
 539 void free_xenballooned_pages(int nr_pages, struct page** pages)
 540 {
 541         int i;
 542
 543         mutex_lock(&balloon_mutex);
 544
 545         for (i = 0; i < nr_pages; i++) {
 546                 if (pages[i])
 547                         balloon_append(pages[i]);
 548         }
 549
 550         /* The balloon may be too large now. Shrink it if needed. */
 551         if (current_credit())
 552                 schedule_delayed_work(&balloon_worker, 0);
 553
 554         mutex_unlock(&balloon_mutex);
 555 }
 556 EXPORT_SYMBOL(free_xenballooned_pages);
 557
 558 static int __init balloon_init(void)
 559 {
 560         unsigned long pfn, extra_pfn_end;
 561         struct page *page;
 562
 563         if (!xen_domain())
 564                 return -ENODEV;
 565
 566         pr_info("xen/balloon: Initialising balloon driver.\n");
 567
 568         balloon_stats.current_pages = xen_pv_domain()
 569                 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
 570                 : max_pfn;
 571         balloon_stats.target_pages  = balloon_stats.current_pages;
 572         balloon_stats.balloon_low   = 0;
 573         balloon_stats.balloon_high  = 0;
 574
 575         balloon_stats.schedule_delay = 1;
 576         balloon_stats.max_schedule_delay = 32;
 577         balloon_stats.retry_count = 1;
 578         balloon_stats.max_retry_count = RETRY_UNLIMITED;
 579
 580 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 581         balloon_stats.hotplug_pages = 0;
 582         balloon_stats.balloon_hotplug = 0;
 583
 584         set_online_page_callback(&xen_online_page);
 585         register_memory_notifier(&xen_memory_nb);
 586 #endif
 587
 588         /*
 589          * Initialize the balloon with pages from the extra memory
 590          * region (see arch/x86/xen/setup.c).
 591          *
 592          * If the amount of usable memory has been limited (e.g., with
 593          * the 'mem' command line parameter), don't add pages beyond
 594          * this limit.
 595          */
 596         extra_pfn_end = min(max_pfn,
 597                             (unsigned long)PFN_DOWN(xen_extra_mem_start
 598                                                     + xen_extra_mem_size));
 599         for (pfn = PFN_UP(xen_extra_mem_start);
 600              pfn < extra_pfn_end;
 601              pfn++) {
 602                 page = pfn_to_page(pfn);
 603                 /* totalram_pages and totalhigh_pages do not include the boot-time
 604                    balloon extension, so don't subtract from it. */
 605                 __balloon_append(page);
 606         }
 607
 608         return 0;
 609 }
 610
 611 subsys_initcall(balloon_init);
 612
 613 MODULE_LICENSE("GPL");