drivers/misc/vmware_balloon.c

   1 /*
   2  * VMware Balloon driver.
   3  *
   4  * Copyright (C) 2000-2010, VMware, Inc. All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; version 2 of the License and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  13  * NON INFRINGEMENT.  See the GNU General Public License for more
  14  * details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Maintained by: Dmitry Torokhov <dtor@vmware.com>
  21  */
  22
  23 /*
  24  * This is VMware physical memory management driver for Linux. The driver
  25  * acts like a "balloon" that can be inflated to reclaim physical pages by
  26  * reserving them in the guest and invalidating them in the monitor,
  27  * freeing up the underlying machine pages so they can be allocated to
  28  * other guests.  The balloon can also be deflated to allow the guest to
  29  * use more physical memory. Higher level policies can control the sizes
  30  * of balloons in VMs in order to manage physical memory resources.
  31  */
  32
  33 //#define DEBUG
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/types.h>
  37 #include <linux/kernel.h>
  38 #include <linux/mm.h>
  39 #include <linux/sched.h>
  40 #include <linux/module.h>
  41 #include <linux/workqueue.h>
  42 #include <linux/debugfs.h>
  43 #include <linux/seq_file.h>
  44 #include <asm/hypervisor.h>
  45
  46 MODULE_AUTHOR("VMware, Inc.");
  47 MODULE_DESCRIPTION("VMware Memory Control (Balloon) Driver");
  48 MODULE_VERSION("1.2.1.0-K");
  49 MODULE_ALIAS("dmi:*:svnVMware*:*");
  50 MODULE_ALIAS("vmware_vmmemctl");
  51 MODULE_LICENSE("GPL");
  52
  53 /*
  54  * Various constants controlling rate of inflaint/deflating balloon,
  55  * measured in pages.
  56  */
  57
  58 /*
  59  * Rate of allocating memory when there is no memory pressure
  60  * (driver performs non-sleeping allocations).
  61  */
  62 #define VMW_BALLOON_NOSLEEP_ALLOC_MAX   16384U
  63
  64 /*
  65  * Rates of memory allocaton when guest experiences memory pressure
  66  * (driver performs sleeping allocations).
  67  */
  68 #define VMW_BALLOON_RATE_ALLOC_MIN      512U
  69 #define VMW_BALLOON_RATE_ALLOC_MAX      2048U
  70 #define VMW_BALLOON_RATE_ALLOC_INC      16U
  71
  72 /*
  73  * Rates for releasing pages while deflating balloon.
  74  */
  75 #define VMW_BALLOON_RATE_FREE_MIN       512U
  76 #define VMW_BALLOON_RATE_FREE_MAX       16384U
  77 #define VMW_BALLOON_RATE_FREE_INC       16U
  78
  79 /*
  80  * When guest is under memory pressure, use a reduced page allocation
  81  * rate for next several cycles.
  82  */
  83 #define VMW_BALLOON_SLOW_CYCLES         4
  84
  85 /*
  86  * Use __GFP_HIGHMEM to allow pages from HIGHMEM zone. We don't
  87  * allow wait (__GFP_WAIT) for NOSLEEP page allocations. Use
  88  * __GFP_NOWARN, to suppress page allocation failure warnings.
  89  */
  90 #define VMW_PAGE_ALLOC_NOSLEEP          (__GFP_HIGHMEM|__GFP_NOWARN)
  91
  92 /*
  93  * Use GFP_HIGHUSER when executing in a separate kernel thread
  94  * context and allocation can sleep.  This is less stressful to
  95  * the guest memory system, since it allows the thread to block
  96  * while memory is reclaimed, and won't take pages from emergency
  97  * low-memory pools.
  98  */
  99 #define VMW_PAGE_ALLOC_CANSLEEP         (GFP_HIGHUSER)
 100
 101 /* Maximum number of page allocations without yielding processor */
 102 #define VMW_BALLOON_YIELD_THRESHOLD     1024
 103
 104
 105 /*
 106  * Hypervisor communication port definitions.
 107  */
 108 #define VMW_BALLOON_HV_PORT             0x5670
 109 #define VMW_BALLOON_HV_MAGIC            0x456c6d6f
 110 #define VMW_BALLOON_PROTOCOL_VERSION    2
 111 #define VMW_BALLOON_GUEST_ID            1       /* Linux */
 112
 113 #define VMW_BALLOON_CMD_START           0
 114 #define VMW_BALLOON_CMD_GET_TARGET      1
 115 #define VMW_BALLOON_CMD_LOCK            2
 116 #define VMW_BALLOON_CMD_UNLOCK          3
 117 #define VMW_BALLOON_CMD_GUEST_ID        4
 118
 119 /* error codes */
 120 #define VMW_BALLOON_SUCCESS             0
 121 #define VMW_BALLOON_FAILURE             -1
 122 #define VMW_BALLOON_ERROR_CMD_INVALID   1
 123 #define VMW_BALLOON_ERROR_PPN_INVALID   2
 124 #define VMW_BALLOON_ERROR_PPN_LOCKED    3
 125 #define VMW_BALLOON_ERROR_PPN_UNLOCKED  4
 126 #define VMW_BALLOON_ERROR_PPN_PINNED    5
 127 #define VMW_BALLOON_ERROR_PPN_NOTNEEDED 6
 128 #define VMW_BALLOON_ERROR_RESET         7
 129 #define VMW_BALLOON_ERROR_BUSY          8
 130
 131 #define VMWARE_BALLOON_CMD(cmd, data, result)           \
 132 ({                                                      \
 133         unsigned long __stat, __dummy1, __dummy2;       \
 134         __asm__ __volatile__ ("inl (%%dx)" :            \
 135                 "=a"(__stat),                           \
 136                 "=c"(__dummy1),                         \
 137                 "=d"(__dummy2),                         \
 138                 "=b"(result) :                          \
 139                 "0"(VMW_BALLOON_HV_MAGIC),              \
 140                 "1"(VMW_BALLOON_CMD_##cmd),             \
 141                 "2"(VMW_BALLOON_HV_PORT),               \
 142                 "3"(data) :                             \
 143                 "memory");                              \
 144         result &= -1UL;                                 \
 145         __stat & -1UL;                                  \
 146 })
 147
 148 #ifdef CONFIG_DEBUG_FS
 149 struct vmballoon_stats {
 150         unsigned int timer;
 151
 152         /* allocation statustics */
 153         unsigned int alloc;
 154         unsigned int alloc_fail;
 155         unsigned int sleep_alloc;
 156         unsigned int sleep_alloc_fail;
 157         unsigned int refused_alloc;
 158         unsigned int refused_free;
 159         unsigned int free;
 160
 161         /* monitor operations */
 162         unsigned int lock;
 163         unsigned int lock_fail;
 164         unsigned int unlock;
 165         unsigned int unlock_fail;
 166         unsigned int target;
 167         unsigned int target_fail;
 168         unsigned int start;
 169         unsigned int start_fail;
 170         unsigned int guest_type;
 171         unsigned int guest_type_fail;
 172 };
 173
 174 #define STATS_INC(stat) (stat)++
 175 #else
 176 #define STATS_INC(stat)
 177 #endif
 178
 179 struct vmballoon {
 180
 181         /* list of reserved physical pages */
 182         struct list_head pages;
 183
 184         /* transient list of non-balloonable pages */
 185         struct list_head refused_pages;
 186
 187         /* balloon size in pages */
 188         unsigned int size;
 189         unsigned int target;
 190
 191         /* reset flag */
 192         bool reset_required;
 193
 194         /* adjustment rates (pages per second) */
 195         unsigned int rate_alloc;
 196         unsigned int rate_free;
 197
 198         /* slowdown page allocations for next few cycles */
 199         unsigned int slow_allocation_cycles;
 200
 201 #ifdef CONFIG_DEBUG_FS
 202         /* statistics */
 203         struct vmballoon_stats stats;
 204
 205         /* debugfs file exporting statistics */
 206         struct dentry *dbg_entry;
 207 #endif
 208
 209         struct sysinfo sysinfo;
 210
 211         struct delayed_work dwork;
 212 };
 213
 214 static struct vmballoon balloon;
 215 static struct workqueue_struct *vmballoon_wq;
 216
 217 /*
 218  * Send "start" command to the host, communicating supported version
 219  * of the protocol.
 220  */
 221 static bool vmballoon_send_start(struct vmballoon *b)
 222 {
 223         unsigned long status, dummy;
 224
 225         STATS_INC(b->stats.start);
 226
 227         status = VMWARE_BALLOON_CMD(START, VMW_BALLOON_PROTOCOL_VERSION, dummy);
 228         if (status == VMW_BALLOON_SUCCESS)
 229                 return true;
 230
 231         pr_debug("%s - failed, hv returns %ld\n", __func__, status);
 232         STATS_INC(b->stats.start_fail);
 233         return false;
 234 }
 235
 236 static bool vmballoon_check_status(struct vmballoon *b, unsigned long status)
 237 {
 238         switch (status) {
 239         case VMW_BALLOON_SUCCESS:
 240                 return true;
 241
 242         case VMW_BALLOON_ERROR_RESET:
 243                 b->reset_required = true;
 244                 /* fall through */
 245
 246         default:
 247                 return false;
 248         }
 249 }
 250
 251 /*
 252  * Communicate guest type to the host so that it can adjust ballooning
 253  * algorithm to the one most appropriate for the guest. This command
 254  * is normally issued after sending "start" command and is part of
 255  * standard reset sequence.
 256  */
 257 static bool vmballoon_send_guest_id(struct vmballoon *b)
 258 {
 259         unsigned long status, dummy;
 260
 261         status = VMWARE_BALLOON_CMD(GUEST_ID, VMW_BALLOON_GUEST_ID, dummy);
 262
 263         STATS_INC(b->stats.guest_type);
 264
 265         if (vmballoon_check_status(b, status))
 266                 return true;
 267
 268         pr_debug("%s - failed, hv returns %ld\n", __func__, status);
 269         STATS_INC(b->stats.guest_type_fail);
 270         return false;
 271 }
 272
 273 /*
 274  * Retrieve desired balloon size from the host.
 275  */
 276 static bool vmballoon_send_get_target(struct vmballoon *b, u32 *new_target)
 277 {
 278         unsigned long status;
 279         unsigned long target;
 280         unsigned long limit;
 281         u32 limit32;
 282
 283         /*
 284          * si_meminfo() is cheap. Moreover, we want to provide dynamic
 285          * max balloon size later. So let us call si_meminfo() every
 286          * iteration.
 287          */
 288         si_meminfo(&b->sysinfo);
 289         limit = b->sysinfo.totalram;
 290
 291         /* Ensure limit fits in 32-bits */
 292         limit32 = (u32)limit;
 293         if (limit != limit32)
 294                 return false;
 295
 296         /* update stats */
 297         STATS_INC(b->stats.target);
 298
 299         status = VMWARE_BALLOON_CMD(GET_TARGET, limit, target);
 300         if (vmballoon_check_status(b, status)) {
 301                 *new_target = target;
 302                 return true;
 303         }
 304
 305         pr_debug("%s - failed, hv returns %ld\n", __func__, status);
 306         STATS_INC(b->stats.target_fail);
 307         return false;
 308 }
 309
 310 /*
 311  * Notify the host about allocated page so that host can use it without
 312  * fear that guest will need it. Host may reject some pages, we need to
 313  * check the return value and maybe submit a different page.
 314  */
 315 static bool vmballoon_send_lock_page(struct vmballoon *b, unsigned long pfn)
 316 {
 317         unsigned long status, dummy;
 318         u32 pfn32;
 319
 320         pfn32 = (u32)pfn;
 321         if (pfn32 != pfn)
 322                 return false;
 323
 324         STATS_INC(b->stats.lock);
 325
 326         status = VMWARE_BALLOON_CMD(LOCK, pfn, dummy);
 327         if (vmballoon_check_status(b, status))
 328                 return true;
 329
 330         pr_debug("%s - ppn %lx, hv returns %ld\n", __func__, pfn, status);
 331         STATS_INC(b->stats.lock_fail);
 332         return false;
 333 }
 334
 335 /*
 336  * Notify the host that guest intends to release given page back into
 337  * the pool of available (to the guest) pages.
 338  */
 339 static bool vmballoon_send_unlock_page(struct vmballoon *b, unsigned long pfn)
 340 {
 341         unsigned long status, dummy;
 342         u32 pfn32;
 343
 344         pfn32 = (u32)pfn;
 345         if (pfn32 != pfn)
 346                 return false;
 347
 348         STATS_INC(b->stats.unlock);
 349
 350         status = VMWARE_BALLOON_CMD(UNLOCK, pfn, dummy);
 351         if (vmballoon_check_status(b, status))
 352                 return true;
 353
 354         pr_debug("%s - ppn %lx, hv returns %ld\n", __func__, pfn, status);
 355         STATS_INC(b->stats.unlock_fail);
 356         return false;
 357 }
 358
 359 /*
 360  * Quickly release all pages allocated for the balloon. This function is
 361  * called when host decides to "reset" balloon for one reason or another.
 362  * Unlike normal "deflate" we do not (shall not) notify host of the pages
 363  * being released.
 364  */
 365 static void vmballoon_pop(struct vmballoon *b)
 366 {
 367         struct page *page, *next;
 368         unsigned int count = 0;
 369
 370         list_for_each_entry_safe(page, next, &b->pages, lru) {
 371                 list_del(&page->lru);
 372                 __free_page(page);
 373                 STATS_INC(b->stats.free);
 374                 b->size--;
 375
 376                 if (++count >= b->rate_free) {
 377                         count = 0;
 378                         cond_resched();
 379                 }
 380         }
 381 }
 382
 383 /*
 384  * Perform standard reset sequence by popping the balloon (in case it
 385  * is not  empty) and then restarting protocol. This operation normally
 386  * happens when host responds with VMW_BALLOON_ERROR_RESET to a command.
 387  */
 388 static void vmballoon_reset(struct vmballoon *b)
 389 {
 390         /* free all pages, skipping monitor unlock */
 391         vmballoon_pop(b);
 392
 393         if (vmballoon_send_start(b)) {
 394                 b->reset_required = false;
 395                 if (!vmballoon_send_guest_id(b))
 396                         pr_err("failed to send guest ID to the host\n");
 397         }
 398 }
 399
 400 /*
 401  * Allocate (or reserve) a page for the balloon and notify the host.  If host
 402  * refuses the page put it on "refuse" list and allocate another one until host
 403  * is satisfied. "Refused" pages are released at the end of inflation cycle
 404  * (when we allocate b->rate_alloc pages).
 405  */
 406 static int vmballoon_reserve_page(struct vmballoon *b, bool can_sleep)
 407 {
 408         struct page *page;
 409         gfp_t flags;
 410         bool locked = false;
 411
 412         do {
 413                 if (!can_sleep)
 414                         STATS_INC(b->stats.alloc);
 415                 else
 416                         STATS_INC(b->stats.sleep_alloc);
 417
 418                 flags = can_sleep ? VMW_PAGE_ALLOC_CANSLEEP : VMW_PAGE_ALLOC_NOSLEEP;
 419                 page = alloc_page(flags);
 420                 if (!page) {
 421                         if (!can_sleep)
 422                                 STATS_INC(b->stats.alloc_fail);
 423                         else
 424                                 STATS_INC(b->stats.sleep_alloc_fail);
 425                         return -ENOMEM;
 426                 }
 427
 428                 /* inform monitor */
 429                 locked = vmballoon_send_lock_page(b, page_to_pfn(page));
 430                 if (!locked) {
 431                         if (b->reset_required) {
 432                                 __free_page(page);
 433                                 return -EIO;
 434                         }
 435
 436                         /* place on list of non-balloonable pages, retry allocation */
 437                         list_add(&page->lru, &b->refused_pages);
 438                         STATS_INC(b->stats.refused_alloc);
 439                 }
 440         } while (!locked);
 441
 442         /* track allocated page */
 443         list_add(&page->lru, &b->pages);
 444
 445         /* update balloon size */
 446         b->size++;
 447
 448         return 0;
 449 }
 450
 451 /*
 452  * Release the page allocated for the balloon. Note that we first notify
 453  * the host so it can make sure the page will be available for the guest
 454  * to use, if needed.
 455  */
 456 static int vmballoon_release_page(struct vmballoon *b, struct page *page)
 457 {
 458         if (!vmballoon_send_unlock_page(b, page_to_pfn(page)))
 459                 return -EIO;
 460
 461         list_del(&page->lru);
 462
 463         /* deallocate page */
 464         __free_page(page);
 465         STATS_INC(b->stats.free);
 466
 467         /* update balloon size */
 468         b->size--;
 469
 470         return 0;
 471 }
 472
 473 /*
 474  * Release pages that were allocated while attempting to inflate the
 475  * balloon but were refused by the host for one reason or another.
 476  */
 477 static void vmballoon_release_refused_pages(struct vmballoon *b)
 478 {
 479         struct page *page, *next;
 480
 481         list_for_each_entry_safe(page, next, &b->refused_pages, lru) {
 482                 list_del(&page->lru);
 483                 __free_page(page);
 484                 STATS_INC(b->stats.refused_free);
 485         }
 486 }
 487
 488 /*
 489  * Inflate the balloon towards its target size. Note that we try to limit
 490  * the rate of allocation to make sure we are not choking the rest of the
 491  * system.
 492  */
 493 static void vmballoon_inflate(struct vmballoon *b)
 494 {
 495         unsigned int goal;
 496         unsigned int rate;
 497         unsigned int i;
 498         unsigned int allocations = 0;
 499         int error = 0;
 500         bool alloc_can_sleep = false;
 501
 502         pr_debug("%s - size: %d, target %d\n", __func__, b->size, b->target);
 503
 504         /*
 505          * First try NOSLEEP page allocations to inflate balloon.
 506          *
 507          * If we do not throttle nosleep allocations, we can drain all
 508          * free pages in the guest quickly (if the balloon target is high).
 509          * As a side-effect, draining free pages helps to inform (force)
 510          * the guest to start swapping if balloon target is not met yet,
 511          * which is a desired behavior. However, balloon driver can consume
 512          * all available CPU cycles if too many pages are allocated in a
 513          * second. Therefore, we throttle nosleep allocations even when
 514          * the guest is not under memory pressure. OTOH, if we have already
 515          * predicted that the guest is under memory pressure, then we
 516          * slowdown page allocations considerably.
 517          */
 518
 519         goal = b->target - b->size;
 520         /*
 521          * Start with no sleep allocation rate which may be higher
 522          * than sleeping allocation rate.
 523          */
 524         rate = b->slow_allocation_cycles ?
 525                         b->rate_alloc : VMW_BALLOON_NOSLEEP_ALLOC_MAX;
 526
 527         pr_debug("%s - goal: %d, no-sleep rate: %d, sleep rate: %d\n",
 528                  __func__, goal, rate, b->rate_alloc);
 529
 530         for (i = 0; i < goal; i++) {
 531
 532                 error = vmballoon_reserve_page(b, alloc_can_sleep);
 533                 if (error) {
 534                         if (error != -ENOMEM) {
 535                                 /*
 536                                  * Not a page allocation failure, stop this
 537                                  * cycle. Maybe we'll get new target from
 538                                  * the host soon.
 539                                  */
 540                                 break;
 541                         }
 542
 543                         if (alloc_can_sleep) {
 544                                 /*
 545                                  * CANSLEEP page allocation failed, so guest
 546                                  * is under severe memory pressure. Quickly
 547                                  * decrease allocation rate.
 548                                  */
 549                                 b->rate_alloc = max(b->rate_alloc / 2,
 550                                                     VMW_BALLOON_RATE_ALLOC_MIN);
 551                                 break;
 552                         }
 553
 554                         /*
 555                          * NOSLEEP page allocation failed, so the guest is
 556                          * under memory pressure. Let us slow down page
 557                          * allocations for next few cycles so that the guest
 558                          * gets out of memory pressure. Also, if we already
 559                          * allocated b->rate_alloc pages, let's pause,
 560                          * otherwise switch to sleeping allocations.
 561                          */
 562                         b->slow_allocation_cycles = VMW_BALLOON_SLOW_CYCLES;
 563
 564                         if (i >= b->rate_alloc)
 565                                 break;
 566
 567                         alloc_can_sleep = true;
 568                         /* Lower rate for sleeping allocations. */
 569                         rate = b->rate_alloc;
 570                 }
 571
 572                 if (++allocations > VMW_BALLOON_YIELD_THRESHOLD) {
 573                         cond_resched();
 574                         allocations = 0;
 575                 }
 576
 577                 if (i >= rate) {
 578                         /* We allocated enough pages, let's take a break. */
 579                         break;
 580                 }
 581         }
 582
 583         /*
 584          * We reached our goal without failures so try increasing
 585          * allocation rate.
 586          */
 587         if (error == 0 && i >= b->rate_alloc) {
 588                 unsigned int mult = i / b->rate_alloc;
 589
 590                 b->rate_alloc =
 591                         min(b->rate_alloc + mult * VMW_BALLOON_RATE_ALLOC_INC,
 592                             VMW_BALLOON_RATE_ALLOC_MAX);
 593         }
 594
 595         vmballoon_release_refused_pages(b);
 596 }
 597
 598 /*
 599  * Decrease the size of the balloon allowing guest to use more memory.
 600  */
 601 static void vmballoon_deflate(struct vmballoon *b)
 602 {
 603         struct page *page, *next;
 604         unsigned int i = 0;
 605         unsigned int goal;
 606         int error;
 607
 608         pr_debug("%s - size: %d, target %d\n", __func__, b->size, b->target);
 609
 610         /* limit deallocation rate */
 611         goal = min(b->size - b->target, b->rate_free);
 612
 613         pr_debug("%s - goal: %d, rate: %d\n", __func__, goal, b->rate_free);
 614
 615         /* free pages to reach target */
 616         list_for_each_entry_safe(page, next, &b->pages, lru) {
 617                 error = vmballoon_release_page(b, page);
 618                 if (error) {
 619                         /* quickly decrease rate in case of error */
 620                         b->rate_free = max(b->rate_free / 2,
 621                                            VMW_BALLOON_RATE_FREE_MIN);
 622                         return;
 623                 }
 624
 625                 if (++i >= goal)
 626                         break;
 627         }
 628
 629         /* slowly increase rate if there were no errors */
 630         b->rate_free = min(b->rate_free + VMW_BALLOON_RATE_FREE_INC,
 631                            VMW_BALLOON_RATE_FREE_MAX);
 632 }
 633
 634 /*
 635  * Balloon work function: reset protocol, if needed, get the new size and
 636  * adjust balloon as needed. Repeat in 1 sec.
 637  */
 638 static void vmballoon_work(struct work_struct *work)
 639 {
 640         struct delayed_work *dwork = to_delayed_work(work);
 641         struct vmballoon *b = container_of(dwork, struct vmballoon, dwork);
 642         unsigned int target;
 643
 644         STATS_INC(b->stats.timer);
 645
 646         if (b->reset_required)
 647                 vmballoon_reset(b);
 648
 649         if (b->slow_allocation_cycles > 0)
 650                 b->slow_allocation_cycles--;
 651
 652         if (vmballoon_send_get_target(b, &target)) {
 653                 /* update target, adjust size */
 654                 b->target = target;
 655
 656                 if (b->size < target)
 657                         vmballoon_inflate(b);
 658                 else if (b->size > target)
 659                         vmballoon_deflate(b);
 660         }
 661
 662         queue_delayed_work(vmballoon_wq, dwork, round_jiffies_relative(HZ));
 663 }
 664
 665 /*
 666  * DEBUGFS Interface
 667  */
 668 #ifdef CONFIG_DEBUG_FS
 669
 670 static int vmballoon_debug_show(struct seq_file *f, void *offset)
 671 {
 672         struct vmballoon *b = f->private;
 673         struct vmballoon_stats *stats = &b->stats;
 674
 675         /* format size info */
 676         seq_printf(f,
 677                    "target:             %8d pages\n"
 678                    "current:            %8d pages\n",
 679                    b->target, b->size);
 680
 681         /* format rate info */
 682         seq_printf(f,
 683                    "rateNoSleepAlloc:   %8d pages/sec\n"
 684                    "rateSleepAlloc:     %8d pages/sec\n"
 685                    "rateFree:           %8d pages/sec\n",
 686                    VMW_BALLOON_NOSLEEP_ALLOC_MAX,
 687                    b->rate_alloc, b->rate_free);
 688
 689         seq_printf(f,
 690                    "\n"
 691                    "timer:              %8u\n"
 692                    "start:              %8u (%4u failed)\n"
 693                    "guestType:          %8u (%4u failed)\n"
 694                    "lock:               %8u (%4u failed)\n"
 695                    "unlock:             %8u (%4u failed)\n"
 696                    "target:             %8u (%4u failed)\n"
 697                    "primNoSleepAlloc:   %8u (%4u failed)\n"
 698                    "primCanSleepAlloc:  %8u (%4u failed)\n"
 699                    "primFree:           %8u\n"
 700                    "errAlloc:           %8u\n"
 701                    "errFree:            %8u\n",
 702                    stats->timer,
 703                    stats->start, stats->start_fail,
 704                    stats->guest_type, stats->guest_type_fail,
 705                    stats->lock,  stats->lock_fail,
 706                    stats->unlock, stats->unlock_fail,
 707                    stats->target, stats->target_fail,
 708                    stats->alloc, stats->alloc_fail,
 709                    stats->sleep_alloc, stats->sleep_alloc_fail,
 710                    stats->free,
 711                    stats->refused_alloc, stats->refused_free);
 712
 713         return 0;
 714 }
 715
 716 static int vmballoon_debug_open(struct inode *inode, struct file *file)
 717 {
 718         return single_open(file, vmballoon_debug_show, inode->i_private);
 719 }
 720
 721 static const struct file_operations vmballoon_debug_fops = {
 722         .owner          = THIS_MODULE,
 723         .open           = vmballoon_debug_open,
 724         .read           = seq_read,
 725         .llseek         = seq_lseek,
 726         .release        = single_release,
 727 };
 728
 729 static int __init vmballoon_debugfs_init(struct vmballoon *b)
 730 {
 731         int error;
 732
 733         b->dbg_entry = debugfs_create_file("vmmemctl", S_IRUGO, NULL, b,
 734                                            &vmballoon_debug_fops);
 735         if (IS_ERR(b->dbg_entry)) {
 736                 error = PTR_ERR(b->dbg_entry);
 737                 pr_err("failed to create debugfs entry, error: %d\n", error);
 738                 return error;
 739         }
 740
 741         return 0;
 742 }
 743
 744 static void __exit vmballoon_debugfs_exit(struct vmballoon *b)
 745 {
 746         debugfs_remove(b->dbg_entry);
 747 }
 748
 749 #else
 750
 751 static inline int vmballoon_debugfs_init(struct vmballoon *b)
 752 {
 753         return 0;
 754 }
 755
 756 static inline void vmballoon_debugfs_exit(struct vmballoon *b)
 757 {
 758 }
 759
 760 #endif  /* CONFIG_DEBUG_FS */
 761
 762 static int __init vmballoon_init(void)
 763 {
 764         int error;
 765
 766         /*
 767          * Check if we are running on VMware's hypervisor and bail out
 768          * if we are not.
 769          */
 770         if (x86_hyper != &x86_hyper_vmware)
 771                 return -ENODEV;
 772
 773         vmballoon_wq = create_freezeable_workqueue("vmmemctl");
 774         if (!vmballoon_wq) {
 775                 pr_err("failed to create workqueue\n");
 776                 return -ENOMEM;
 777         }
 778
 779         INIT_LIST_HEAD(&balloon.pages);
 780         INIT_LIST_HEAD(&balloon.refused_pages);
 781
 782         /* initialize rates */
 783         balloon.rate_alloc = VMW_BALLOON_RATE_ALLOC_MAX;
 784         balloon.rate_free = VMW_BALLOON_RATE_FREE_MAX;
 785
 786         INIT_DELAYED_WORK(&balloon.dwork, vmballoon_work);
 787
 788         /*
 789          * Start balloon.
 790          */
 791         if (!vmballoon_send_start(&balloon)) {
 792                 pr_err("failed to send start command to the host\n");
 793                 error = -EIO;
 794                 goto fail;
 795         }
 796
 797         if (!vmballoon_send_guest_id(&balloon)) {
 798                 pr_err("failed to send guest ID to the host\n");
 799                 error = -EIO;
 800                 goto fail;
 801         }
 802
 803         error = vmballoon_debugfs_init(&balloon);
 804         if (error)
 805                 goto fail;
 806
 807         queue_delayed_work(vmballoon_wq, &balloon.dwork, 0);
 808
 809         return 0;
 810
 811 fail:
 812         destroy_workqueue(vmballoon_wq);
 813         return error;
 814 }
 815 module_init(vmballoon_init);
 816
 817 static void __exit vmballoon_exit(void)
 818 {
 819         cancel_delayed_work_sync(&balloon.dwork);
 820         destroy_workqueue(vmballoon_wq);
 821
 822         vmballoon_debugfs_exit(&balloon);
 823
 824         /*
 825          * Deallocate all reserved memory, and reset connection with monitor.
 826          * Reset connection before deallocating memory to avoid potential for
 827          * additional spurious resets from guest touching deallocated pages.
 828          */
 829         vmballoon_send_start(&balloon);
 830         vmballoon_pop(&balloon);
 831 }
 832 module_exit(vmballoon_exit);