arch/powerpc/platforms/pseries/cmm.c

   1 /*
   2  * Collaborative memory management interface.
   3  *
   4  * Copyright (C) 2008 IBM Corporation
   5  * Author(s): Brian King (brking@linux.vnet.ibm.com),
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  */
  22
  23 #include <linux/ctype.h>
  24 #include <linux/delay.h>
  25 #include <linux/errno.h>
  26 #include <linux/fs.h>
  27 #include <linux/gfp.h>
  28 #include <linux/kthread.h>
  29 #include <linux/module.h>
  30 #include <linux/oom.h>
  31 #include <linux/reboot.h>
  32 #include <linux/sched.h>
  33 #include <linux/stringify.h>
  34 #include <linux/swap.h>
  35 #include <linux/device.h>
  36 #include <asm/firmware.h>
  37 #include <asm/hvcall.h>
  38 #include <asm/mmu.h>
  39 #include <asm/pgalloc.h>
  40 #include <asm/uaccess.h>
  41 #include <linux/memory.h>
  42 #include <asm/plpar_wrappers.h>
  43
  44 #define CMM_DRIVER_VERSION      "1.0.0"
  45 #define CMM_DEFAULT_DELAY       1
  46 #define CMM_HOTPLUG_DELAY       5
  47 #define CMM_DEBUG                       0
  48 #define CMM_DISABLE             0
  49 #define CMM_OOM_KB              1024
  50 #define CMM_MIN_MEM_MB          256
  51 #define KB2PAGES(_p)            ((_p)>>(PAGE_SHIFT-10))
  52 #define PAGES2KB(_p)            ((_p)<<(PAGE_SHIFT-10))
  53 /*
  54  * The priority level tries to ensure that this notifier is called as
  55  * late as possible to reduce thrashing in the shared memory pool.
  56  */
  57 #define CMM_MEM_HOTPLUG_PRI     1
  58 #define CMM_MEM_ISOLATE_PRI     15
  59
  60 static unsigned int delay = CMM_DEFAULT_DELAY;
  61 static unsigned int hotplug_delay = CMM_HOTPLUG_DELAY;
  62 static unsigned int oom_kb = CMM_OOM_KB;
  63 static unsigned int cmm_debug = CMM_DEBUG;
  64 static unsigned int cmm_disabled = CMM_DISABLE;
  65 static unsigned long min_mem_mb = CMM_MIN_MEM_MB;
  66 static struct device cmm_dev;
  67
  68 MODULE_AUTHOR("Brian King <brking@linux.vnet.ibm.com>");
  69 MODULE_DESCRIPTION("IBM System p Collaborative Memory Manager");
  70 MODULE_LICENSE("GPL");
  71 MODULE_VERSION(CMM_DRIVER_VERSION);
  72
  73 module_param_named(delay, delay, uint, S_IRUGO | S_IWUSR);
  74 MODULE_PARM_DESC(delay, "Delay (in seconds) between polls to query hypervisor paging requests. "
  75                  "[Default=" __stringify(CMM_DEFAULT_DELAY) "]");
  76 module_param_named(hotplug_delay, hotplug_delay, uint, S_IRUGO | S_IWUSR);
  77 MODULE_PARM_DESC(delay, "Delay (in seconds) after memory hotplug remove "
  78                  "before loaning resumes. "
  79                  "[Default=" __stringify(CMM_HOTPLUG_DELAY) "]");
  80 module_param_named(oom_kb, oom_kb, uint, S_IRUGO | S_IWUSR);
  81 MODULE_PARM_DESC(oom_kb, "Amount of memory in kb to free on OOM. "
  82                  "[Default=" __stringify(CMM_OOM_KB) "]");
  83 module_param_named(min_mem_mb, min_mem_mb, ulong, S_IRUGO | S_IWUSR);
  84 MODULE_PARM_DESC(min_mem_mb, "Minimum amount of memory (in MB) to not balloon. "
  85                  "[Default=" __stringify(CMM_MIN_MEM_MB) "]");
  86 module_param_named(debug, cmm_debug, uint, S_IRUGO | S_IWUSR);
  87 MODULE_PARM_DESC(debug, "Enable module debugging logging. Set to 1 to enable. "
  88                  "[Default=" __stringify(CMM_DEBUG) "]");
  89
  90 #define CMM_NR_PAGES ((PAGE_SIZE - sizeof(void *) - sizeof(unsigned long)) / sizeof(unsigned long))
  91
  92 #define cmm_dbg(...) if (cmm_debug) { printk(KERN_INFO "cmm: "__VA_ARGS__); }
  93
  94 struct cmm_page_array {
  95         struct cmm_page_array *next;
  96         unsigned long index;
  97         unsigned long page[CMM_NR_PAGES];
  98 };
  99
 100 static unsigned long loaned_pages;
 101 static unsigned long loaned_pages_target;
 102 static unsigned long oom_freed_pages;
 103
 104 static struct cmm_page_array *cmm_page_list;
 105 static DEFINE_SPINLOCK(cmm_lock);
 106
 107 static DEFINE_MUTEX(hotplug_mutex);
 108 static int hotplug_occurred; /* protected by the hotplug mutex */
 109
 110 static struct task_struct *cmm_thread_ptr;
 111
 112 /**
 113  * cmm_alloc_pages - Allocate pages and mark them as loaned
 114  * @nr: number of pages to allocate
 115  *
 116  * Return value:
 117  *      number of pages requested to be allocated which were not
 118  **/
 119 static long cmm_alloc_pages(long nr)
 120 {
 121         struct cmm_page_array *pa, *npa;
 122         unsigned long addr;
 123         long rc;
 124
 125         cmm_dbg("Begin request for %ld pages\n", nr);
 126
 127         while (nr) {
 128                 /* Exit if a hotplug operation is in progress or occurred */
 129                 if (mutex_trylock(&hotplug_mutex)) {
 130                         if (hotplug_occurred) {
 131                                 mutex_unlock(&hotplug_mutex);
 132                                 break;
 133                         }
 134                         mutex_unlock(&hotplug_mutex);
 135                 } else {
 136                         break;
 137                 }
 138
 139                 addr = __get_free_page(GFP_NOIO | __GFP_NOWARN |
 140                                        __GFP_NORETRY | __GFP_NOMEMALLOC);
 141                 if (!addr)
 142                         break;
 143                 spin_lock(&cmm_lock);
 144                 pa = cmm_page_list;
 145                 if (!pa || pa->index >= CMM_NR_PAGES) {
 146                         /* Need a new page for the page list. */
 147                         spin_unlock(&cmm_lock);
 148                         npa = (struct cmm_page_array *)__get_free_page(
 149                                         GFP_NOIO | __GFP_NOWARN |
 150                                         __GFP_NORETRY | __GFP_NOMEMALLOC);
 151                         if (!npa) {
 152                                 pr_info("%s: Can not allocate new page list\n", __func__);
 153                                 free_page(addr);
 154                                 break;
 155                         }
 156                         spin_lock(&cmm_lock);
 157                         pa = cmm_page_list;
 158
 159                         if (!pa || pa->index >= CMM_NR_PAGES) {
 160                                 npa->next = pa;
 161                                 npa->index = 0;
 162                                 pa = npa;
 163                                 cmm_page_list = pa;
 164                         } else
 165                                 free_page((unsigned long) npa);
 166                 }
 167
 168                 if ((rc = plpar_page_set_loaned(__pa(addr)))) {
 169                         pr_err("%s: Can not set page to loaned. rc=%ld\n", __func__, rc);
 170                         spin_unlock(&cmm_lock);
 171                         free_page(addr);
 172                         break;
 173                 }
 174
 175                 pa->page[pa->index++] = addr;
 176                 loaned_pages++;
 177                 totalram_pages--;
 178                 spin_unlock(&cmm_lock);
 179                 nr--;
 180         }
 181
 182         cmm_dbg("End request with %ld pages unfulfilled\n", nr);
 183         return nr;
 184 }
 185
 186 /**
 187  * cmm_free_pages - Free pages and mark them as active
 188  * @nr: number of pages to free
 189  *
 190  * Return value:
 191  *      number of pages requested to be freed which were not
 192  **/
 193 static long cmm_free_pages(long nr)
 194 {
 195         struct cmm_page_array *pa;
 196         unsigned long addr;
 197
 198         cmm_dbg("Begin free of %ld pages.\n", nr);
 199         spin_lock(&cmm_lock);
 200         pa = cmm_page_list;
 201         while (nr) {
 202                 if (!pa || pa->index <= 0)
 203                         break;
 204                 addr = pa->page[--pa->index];
 205
 206                 if (pa->index == 0) {
 207                         pa = pa->next;
 208                         free_page((unsigned long) cmm_page_list);
 209                         cmm_page_list = pa;
 210                 }
 211
 212                 plpar_page_set_active(__pa(addr));
 213                 free_page(addr);
 214                 loaned_pages--;
 215                 nr--;
 216                 totalram_pages++;
 217         }
 218         spin_unlock(&cmm_lock);
 219         cmm_dbg("End request with %ld pages unfulfilled\n", nr);
 220         return nr;
 221 }
 222
 223 /**
 224  * cmm_oom_notify - OOM notifier
 225  * @self:       notifier block struct
 226  * @dummy:      not used
 227  * @parm:       returned - number of pages freed
 228  *
 229  * Return value:
 230  *      NOTIFY_OK
 231  **/
 232 static int cmm_oom_notify(struct notifier_block *self,
 233                           unsigned long dummy, void *parm)
 234 {
 235         unsigned long *freed = parm;
 236         long nr = KB2PAGES(oom_kb);
 237
 238         cmm_dbg("OOM processing started\n");
 239         nr = cmm_free_pages(nr);
 240         loaned_pages_target = loaned_pages;
 241         *freed += KB2PAGES(oom_kb) - nr;
 242         oom_freed_pages += KB2PAGES(oom_kb) - nr;
 243         cmm_dbg("OOM processing complete\n");
 244         return NOTIFY_OK;
 245 }
 246
 247 /**
 248  * cmm_get_mpp - Read memory performance parameters
 249  *
 250  * Makes hcall to query the current page loan request from the hypervisor.
 251  *
 252  * Return value:
 253  *      nothing
 254  **/
 255 static void cmm_get_mpp(void)
 256 {
 257         int rc;
 258         struct hvcall_mpp_data mpp_data;
 259         signed long active_pages_target, page_loan_request, target;
 260         signed long total_pages = totalram_pages + loaned_pages;
 261         signed long min_mem_pages = (min_mem_mb * 1024 * 1024) / PAGE_SIZE;
 262
 263         rc = h_get_mpp(&mpp_data);
 264
 265         if (rc != H_SUCCESS)
 266                 return;
 267
 268         page_loan_request = div_s64((s64)mpp_data.loan_request, PAGE_SIZE);
 269         target = page_loan_request + (signed long)loaned_pages;
 270
 271         if (target < 0 || total_pages < min_mem_pages)
 272                 target = 0;
 273
 274         if (target > oom_freed_pages)
 275                 target -= oom_freed_pages;
 276         else
 277                 target = 0;
 278
 279         active_pages_target = total_pages - target;
 280
 281         if (min_mem_pages > active_pages_target)
 282                 target = total_pages - min_mem_pages;
 283
 284         if (target < 0)
 285                 target = 0;
 286
 287         loaned_pages_target = target;
 288
 289         cmm_dbg("delta = %ld, loaned = %lu, target = %lu, oom = %lu, totalram = %lu\n",
 290                 page_loan_request, loaned_pages, loaned_pages_target,
 291                 oom_freed_pages, totalram_pages);
 292 }
 293
 294 static struct notifier_block cmm_oom_nb = {
 295         .notifier_call = cmm_oom_notify
 296 };
 297
 298 /**
 299  * cmm_thread - CMM task thread
 300  * @dummy:      not used
 301  *
 302  * Return value:
 303  *      0
 304  **/
 305 static int cmm_thread(void *dummy)
 306 {
 307         unsigned long timeleft;
 308
 309         while (1) {
 310                 timeleft = msleep_interruptible(delay * 1000);
 311
 312                 if (kthread_should_stop() || timeleft)
 313                         break;
 314
 315                 if (mutex_trylock(&hotplug_mutex)) {
 316                         if (hotplug_occurred) {
 317                                 hotplug_occurred = 0;
 318                                 mutex_unlock(&hotplug_mutex);
 319                                 cmm_dbg("Hotplug operation has occurred, "
 320                                                 "loaning activity suspended "
 321                                                 "for %d seconds.\n",
 322                                                 hotplug_delay);
 323                                 timeleft = msleep_interruptible(hotplug_delay *
 324                                                 1000);
 325                                 if (kthread_should_stop() || timeleft)
 326                                         break;
 327                                 continue;
 328                         }
 329                         mutex_unlock(&hotplug_mutex);
 330                 } else {
 331                         cmm_dbg("Hotplug operation in progress, activity "
 332                                         "suspended\n");
 333                         continue;
 334                 }
 335
 336                 cmm_get_mpp();
 337
 338                 if (loaned_pages_target > loaned_pages) {
 339                         if (cmm_alloc_pages(loaned_pages_target - loaned_pages))
 340                                 loaned_pages_target = loaned_pages;
 341                 } else if (loaned_pages_target < loaned_pages)
 342                         cmm_free_pages(loaned_pages - loaned_pages_target);
 343         }
 344         return 0;
 345 }
 346
 347 #define CMM_SHOW(name, format, args...)                 \
 348         static ssize_t show_##name(struct device *dev,  \
 349                                    struct device_attribute *attr,       \
 350                                    char *buf)                   \
 351         {                                                       \
 352                 return sprintf(buf, format, ##args);            \
 353         }                                                       \
 354         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 355
 356 CMM_SHOW(loaned_kb, "%lu\n", PAGES2KB(loaned_pages));
 357 CMM_SHOW(loaned_target_kb, "%lu\n", PAGES2KB(loaned_pages_target));
 358
 359 static ssize_t show_oom_pages(struct device *dev,
 360                               struct device_attribute *attr, char *buf)
 361 {
 362         return sprintf(buf, "%lu\n", PAGES2KB(oom_freed_pages));
 363 }
 364
 365 static ssize_t store_oom_pages(struct device *dev,
 366                                struct device_attribute *attr,
 367                                const char *buf, size_t count)
 368 {
 369         unsigned long val = simple_strtoul (buf, NULL, 10);
 370
 371         if (!capable(CAP_SYS_ADMIN))
 372                 return -EPERM;
 373         if (val != 0)
 374                 return -EBADMSG;
 375
 376         oom_freed_pages = 0;
 377         return count;
 378 }
 379
 380 static DEVICE_ATTR(oom_freed_kb, S_IWUSR | S_IRUGO,
 381                    show_oom_pages, store_oom_pages);
 382
 383 static struct device_attribute *cmm_attrs[] = {
 384         &dev_attr_loaned_kb,
 385         &dev_attr_loaned_target_kb,
 386         &dev_attr_oom_freed_kb,
 387 };
 388
 389 static struct bus_type cmm_subsys = {
 390         .name = "cmm",
 391         .dev_name = "cmm",
 392 };
 393
 394 static void cmm_release_device(struct device *dev)
 395 {
 396 }
 397
 398 /**
 399  * cmm_sysfs_register - Register with sysfs
 400  *
 401  * Return value:
 402  *      0 on success / other on failure
 403  **/
 404 static int cmm_sysfs_register(struct device *dev)
 405 {
 406         int i, rc;
 407
 408         if ((rc = subsys_system_register(&cmm_subsys, NULL)))
 409                 return rc;
 410
 411         dev->id = 0;
 412         dev->bus = &cmm_subsys;
 413         dev->release = cmm_release_device;
 414
 415         if ((rc = device_register(dev)))
 416                 goto subsys_unregister;
 417
 418         for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++) {
 419                 if ((rc = device_create_file(dev, cmm_attrs[i])))
 420                         goto fail;
 421         }
 422
 423         return 0;
 424
 425 fail:
 426         while (--i >= 0)
 427                 device_remove_file(dev, cmm_attrs[i]);
 428         device_unregister(dev);
 429 subsys_unregister:
 430         bus_unregister(&cmm_subsys);
 431         return rc;
 432 }
 433
 434 /**
 435  * cmm_unregister_sysfs - Unregister from sysfs
 436  *
 437  **/
 438 static void cmm_unregister_sysfs(struct device *dev)
 439 {
 440         int i;
 441
 442         for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++)
 443                 device_remove_file(dev, cmm_attrs[i]);
 444         device_unregister(dev);
 445         bus_unregister(&cmm_subsys);
 446 }
 447
 448 /**
 449  * cmm_reboot_notifier - Make sure pages are not still marked as "loaned"
 450  *
 451  **/
 452 static int cmm_reboot_notifier(struct notifier_block *nb,
 453                                unsigned long action, void *unused)
 454 {
 455         if (action == SYS_RESTART) {
 456                 if (cmm_thread_ptr)
 457                         kthread_stop(cmm_thread_ptr);
 458                 cmm_thread_ptr = NULL;
 459                 cmm_free_pages(loaned_pages);
 460         }
 461         return NOTIFY_DONE;
 462 }
 463
 464 static struct notifier_block cmm_reboot_nb = {
 465         .notifier_call = cmm_reboot_notifier,
 466 };
 467
 468 /**
 469  * cmm_count_pages - Count the number of pages loaned in a particular range.
 470  *
 471  * @arg: memory_isolate_notify structure with address range and count
 472  *
 473  * Return value:
 474  *      0 on success
 475  **/
 476 static unsigned long cmm_count_pages(void *arg)
 477 {
 478         struct memory_isolate_notify *marg = arg;
 479         struct cmm_page_array *pa;
 480         unsigned long start = (unsigned long)pfn_to_kaddr(marg->start_pfn);
 481         unsigned long end = start + (marg->nr_pages << PAGE_SHIFT);
 482         unsigned long idx;
 483
 484         spin_lock(&cmm_lock);
 485         pa = cmm_page_list;
 486         while (pa) {
 487                 if ((unsigned long)pa >= start && (unsigned long)pa < end)
 488                         marg->pages_found++;
 489                 for (idx = 0; idx < pa->index; idx++)
 490                         if (pa->page[idx] >= start && pa->page[idx] < end)
 491                                 marg->pages_found++;
 492                 pa = pa->next;
 493         }
 494         spin_unlock(&cmm_lock);
 495         return 0;
 496 }
 497
 498 /**
 499  * cmm_memory_isolate_cb - Handle memory isolation notifier calls
 500  * @self:       notifier block struct
 501  * @action:     action to take
 502  * @arg:        struct memory_isolate_notify data for handler
 503  *
 504  * Return value:
 505  *      NOTIFY_OK or notifier error based on subfunction return value
 506  **/
 507 static int cmm_memory_isolate_cb(struct notifier_block *self,
 508                                  unsigned long action, void *arg)
 509 {
 510         int ret = 0;
 511
 512         if (action == MEM_ISOLATE_COUNT)
 513                 ret = cmm_count_pages(arg);
 514
 515         return notifier_from_errno(ret);
 516 }
 517
 518 static struct notifier_block cmm_mem_isolate_nb = {
 519         .notifier_call = cmm_memory_isolate_cb,
 520         .priority = CMM_MEM_ISOLATE_PRI
 521 };
 522
 523 /**
 524  * cmm_mem_going_offline - Unloan pages where memory is to be removed
 525  * @arg: memory_notify structure with page range to be offlined
 526  *
 527  * Return value:
 528  *      0 on success
 529  **/
 530 static int cmm_mem_going_offline(void *arg)
 531 {
 532         struct memory_notify *marg = arg;
 533         unsigned long start_page = (unsigned long)pfn_to_kaddr(marg->start_pfn);
 534         unsigned long end_page = start_page + (marg->nr_pages << PAGE_SHIFT);
 535         struct cmm_page_array *pa_curr, *pa_last, *npa;
 536         unsigned long idx;
 537         unsigned long freed = 0;
 538
 539         cmm_dbg("Memory going offline, searching 0x%lx (%ld pages).\n",
 540                         start_page, marg->nr_pages);
 541         spin_lock(&cmm_lock);
 542
 543         /* Search the page list for pages in the range to be offlined */
 544         pa_last = pa_curr = cmm_page_list;
 545         while (pa_curr) {
 546                 for (idx = (pa_curr->index - 1); (idx + 1) > 0; idx--) {
 547                         if ((pa_curr->page[idx] < start_page) ||
 548                             (pa_curr->page[idx] >= end_page))
 549                                 continue;
 550
 551                         plpar_page_set_active(__pa(pa_curr->page[idx]));
 552                         free_page(pa_curr->page[idx]);
 553                         freed++;
 554                         loaned_pages--;
 555                         totalram_pages++;
 556                         pa_curr->page[idx] = pa_last->page[--pa_last->index];
 557                         if (pa_last->index == 0) {
 558                                 if (pa_curr == pa_last)
 559                                         pa_curr = pa_last->next;
 560                                 pa_last = pa_last->next;
 561                                 free_page((unsigned long)cmm_page_list);
 562                                 cmm_page_list = pa_last;
 563                         }
 564                 }
 565                 pa_curr = pa_curr->next;
 566         }
 567
 568         /* Search for page list structures in the range to be offlined */
 569         pa_last = NULL;
 570         pa_curr = cmm_page_list;
 571         while (pa_curr) {
 572                 if (((unsigned long)pa_curr >= start_page) &&
 573                                 ((unsigned long)pa_curr < end_page)) {
 574                         npa = (struct cmm_page_array *)__get_free_page(
 575                                         GFP_NOIO | __GFP_NOWARN |
 576                                         __GFP_NORETRY | __GFP_NOMEMALLOC);
 577                         if (!npa) {
 578                                 spin_unlock(&cmm_lock);
 579                                 cmm_dbg("Failed to allocate memory for list "
 580                                                 "management. Memory hotplug "
 581                                                 "failed.\n");
 582                                 return ENOMEM;
 583                         }
 584                         memcpy(npa, pa_curr, PAGE_SIZE);
 585                         if (pa_curr == cmm_page_list)
 586                                 cmm_page_list = npa;
 587                         if (pa_last)
 588                                 pa_last->next = npa;
 589                         free_page((unsigned long) pa_curr);
 590                         freed++;
 591                         pa_curr = npa;
 592                 }
 593
 594                 pa_last = pa_curr;
 595                 pa_curr = pa_curr->next;
 596         }
 597
 598         spin_unlock(&cmm_lock);
 599         cmm_dbg("Released %ld pages in the search range.\n", freed);
 600
 601         return 0;
 602 }
 603
 604 /**
 605  * cmm_memory_cb - Handle memory hotplug notifier calls
 606  * @self:       notifier block struct
 607  * @action:     action to take
 608  * @arg:        struct memory_notify data for handler
 609  *
 610  * Return value:
 611  *      NOTIFY_OK or notifier error based on subfunction return value
 612  *
 613  **/
 614 static int cmm_memory_cb(struct notifier_block *self,
 615                         unsigned long action, void *arg)
 616 {
 617         int ret = 0;
 618
 619         switch (action) {
 620         case MEM_GOING_OFFLINE:
 621                 mutex_lock(&hotplug_mutex);
 622                 hotplug_occurred = 1;
 623                 ret = cmm_mem_going_offline(arg);
 624                 break;
 625         case MEM_OFFLINE:
 626         case MEM_CANCEL_OFFLINE:
 627                 mutex_unlock(&hotplug_mutex);
 628                 cmm_dbg("Memory offline operation complete.\n");
 629                 break;
 630         case MEM_GOING_ONLINE:
 631         case MEM_ONLINE:
 632         case MEM_CANCEL_ONLINE:
 633                 break;
 634         }
 635
 636         return notifier_from_errno(ret);
 637 }
 638
 639 static struct notifier_block cmm_mem_nb = {
 640         .notifier_call = cmm_memory_cb,
 641         .priority = CMM_MEM_HOTPLUG_PRI
 642 };
 643
 644 /**
 645  * cmm_init - Module initialization
 646  *
 647  * Return value:
 648  *      0 on success / other on failure
 649  **/
 650 static int cmm_init(void)
 651 {
 652         int rc = -ENOMEM;
 653
 654         if (!firmware_has_feature(FW_FEATURE_CMO))
 655                 return -EOPNOTSUPP;
 656
 657         if ((rc = register_oom_notifier(&cmm_oom_nb)) < 0)
 658                 return rc;
 659
 660         if ((rc = register_reboot_notifier(&cmm_reboot_nb)))
 661                 goto out_oom_notifier;
 662
 663         if ((rc = cmm_sysfs_register(&cmm_dev)))
 664                 goto out_reboot_notifier;
 665
 666         if (register_memory_notifier(&cmm_mem_nb) ||
 667             register_memory_isolate_notifier(&cmm_mem_isolate_nb))
 668                 goto out_unregister_notifier;
 669
 670         if (cmm_disabled)
 671                 return rc;
 672
 673         cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
 674         if (IS_ERR(cmm_thread_ptr)) {
 675                 rc = PTR_ERR(cmm_thread_ptr);
 676                 goto out_unregister_notifier;
 677         }
 678
 679         return rc;
 680
 681 out_unregister_notifier:
 682         unregister_memory_notifier(&cmm_mem_nb);
 683         unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
 684         cmm_unregister_sysfs(&cmm_dev);
 685 out_reboot_notifier:
 686         unregister_reboot_notifier(&cmm_reboot_nb);
 687 out_oom_notifier:
 688         unregister_oom_notifier(&cmm_oom_nb);
 689         return rc;
 690 }
 691
 692 /**
 693  * cmm_exit - Module exit
 694  *
 695  * Return value:
 696  *      nothing
 697  **/
 698 static void cmm_exit(void)
 699 {
 700         if (cmm_thread_ptr)
 701                 kthread_stop(cmm_thread_ptr);
 702         unregister_oom_notifier(&cmm_oom_nb);
 703         unregister_reboot_notifier(&cmm_reboot_nb);
 704         unregister_memory_notifier(&cmm_mem_nb);
 705         unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
 706         cmm_free_pages(loaned_pages);
 707         cmm_unregister_sysfs(&cmm_dev);
 708 }
 709
 710 /**
 711  * cmm_set_disable - Disable/Enable CMM
 712  *
 713  * Return value:
 714  *      0 on success / other on failure
 715  **/
 716 static int cmm_set_disable(const char *val, struct kernel_param *kp)
 717 {
 718         int disable = simple_strtoul(val, NULL, 10);
 719
 720         if (disable != 0 && disable != 1)
 721                 return -EINVAL;
 722
 723         if (disable && !cmm_disabled) {
 724                 if (cmm_thread_ptr)
 725                         kthread_stop(cmm_thread_ptr);
 726                 cmm_thread_ptr = NULL;
 727                 cmm_free_pages(loaned_pages);
 728         } else if (!disable && cmm_disabled) {
 729                 cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
 730                 if (IS_ERR(cmm_thread_ptr))
 731                         return PTR_ERR(cmm_thread_ptr);
 732         }
 733
 734         cmm_disabled = disable;
 735         return 0;
 736 }
 737
 738 module_param_call(disable, cmm_set_disable, param_get_uint,
 739                   &cmm_disabled, S_IRUGO | S_IWUSR);
 740 MODULE_PARM_DESC(disable, "Disable CMM. Set to 1 to disable. "
 741                  "[Default=" __stringify(CMM_DISABLE) "]");
 742
 743 module_init(cmm_init);
 744 module_exit(cmm_exit);