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/init.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/sysdev.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
  43 #include "plpar_wrappers.h"
  44
  45 #define CMM_DRIVER_VERSION      "1.0.0"
  46 #define CMM_DEFAULT_DELAY       1
  47 #define CMM_HOTPLUG_DELAY       5
  48 #define CMM_DEBUG                       0
  49 #define CMM_DISABLE             0
  50 #define CMM_OOM_KB              1024
  51 #define CMM_MIN_MEM_MB          256
  52 #define KB2PAGES(_p)            ((_p)>>(PAGE_SHIFT-10))
  53 #define PAGES2KB(_p)            ((_p)<<(PAGE_SHIFT-10))
  54 /*
  55  * The priority level tries to ensure that this notifier is called as
  56  * late as possible to reduce thrashing in the shared memory pool.
  57  */
  58 #define CMM_MEM_HOTPLUG_PRI     1
  59 #define CMM_MEM_ISOLATE_PRI     15
  60
  61 static unsigned int delay = CMM_DEFAULT_DELAY;
  62 static unsigned int hotplug_delay = CMM_HOTPLUG_DELAY;
  63 static unsigned int oom_kb = CMM_OOM_KB;
  64 static unsigned int cmm_debug = CMM_DEBUG;
  65 static unsigned int cmm_disabled = CMM_DISABLE;
  66 static unsigned long min_mem_mb = CMM_MIN_MEM_MB;
  67 static struct sys_device cmm_sysdev;
  68
  69 MODULE_AUTHOR("Brian King <brking@linux.vnet.ibm.com>");
  70 MODULE_DESCRIPTION("IBM System p Collaborative Memory Manager");
  71 MODULE_LICENSE("GPL");
  72 MODULE_VERSION(CMM_DRIVER_VERSION);
  73
  74 module_param_named(delay, delay, uint, S_IRUGO | S_IWUSR);
  75 MODULE_PARM_DESC(delay, "Delay (in seconds) between polls to query hypervisor paging requests. "
  76                  "[Default=" __stringify(CMM_DEFAULT_DELAY) "]");
  77 module_param_named(hotplug_delay, hotplug_delay, uint, S_IRUGO | S_IWUSR);
  78 MODULE_PARM_DESC(delay, "Delay (in seconds) after memory hotplug remove "
  79                  "before loaning resumes. "
  80                  "[Default=" __stringify(CMM_HOTPLUG_DELAY) "]");
  81 module_param_named(oom_kb, oom_kb, uint, S_IRUGO | S_IWUSR);
  82 MODULE_PARM_DESC(oom_kb, "Amount of memory in kb to free on OOM. "
  83                  "[Default=" __stringify(CMM_OOM_KB) "]");
  84 module_param_named(min_mem_mb, min_mem_mb, ulong, S_IRUGO | S_IWUSR);
  85 MODULE_PARM_DESC(min_mem_mb, "Minimum amount of memory (in MB) to not balloon. "
  86                  "[Default=" __stringify(CMM_MIN_MEM_MB) "]");
  87 module_param_named(debug, cmm_debug, uint, S_IRUGO | S_IWUSR);
  88 MODULE_PARM_DESC(debug, "Enable module debugging logging. Set to 1 to enable. "
  89                  "[Default=" __stringify(CMM_DEBUG) "]");
  90
  91 #define CMM_NR_PAGES ((PAGE_SIZE - sizeof(void *) - sizeof(unsigned long)) / sizeof(unsigned long))
  92
  93 #define cmm_dbg(...) if (cmm_debug) { printk(KERN_INFO "cmm: "__VA_ARGS__); }
  94
  95 struct cmm_page_array {
  96         struct cmm_page_array *next;
  97         unsigned long index;
  98         unsigned long page[CMM_NR_PAGES];
  99 };
 100
 101 static unsigned long loaned_pages;
 102 static unsigned long loaned_pages_target;
 103 static unsigned long oom_freed_pages;
 104
 105 static struct cmm_page_array *cmm_page_list;
 106 static DEFINE_SPINLOCK(cmm_lock);
 107
 108 static DEFINE_MUTEX(hotplug_mutex);
 109 static int hotplug_occurred; /* protected by the hotplug mutex */
 110
 111 static struct task_struct *cmm_thread_ptr;
 112
 113 /**
 114  * cmm_alloc_pages - Allocate pages and mark them as loaned
 115  * @nr: number of pages to allocate
 116  *
 117  * Return value:
 118  *      number of pages requested to be allocated which were not
 119  **/
 120 static long cmm_alloc_pages(long nr)
 121 {
 122         struct cmm_page_array *pa, *npa;
 123         unsigned long addr;
 124         long rc;
 125
 126         cmm_dbg("Begin request for %ld pages\n", nr);
 127
 128         while (nr) {
 129                 /* Exit if a hotplug operation is in progress or occurred */
 130                 if (mutex_trylock(&hotplug_mutex)) {
 131                         if (hotplug_occurred) {
 132                                 mutex_unlock(&hotplug_mutex);
 133                                 break;
 134                         }
 135                         mutex_unlock(&hotplug_mutex);
 136                 } else {
 137                         break;
 138                 }
 139
 140                 addr = __get_free_page(GFP_NOIO | __GFP_NOWARN |
 141                                        __GFP_NORETRY | __GFP_NOMEMALLOC);
 142                 if (!addr)
 143                         break;
 144                 spin_lock(&cmm_lock);
 145                 pa = cmm_page_list;
 146                 if (!pa || pa->index >= CMM_NR_PAGES) {
 147                         /* Need a new page for the page list. */
 148                         spin_unlock(&cmm_lock);
 149                         npa = (struct cmm_page_array *)__get_free_page(
 150                                         GFP_NOIO | __GFP_NOWARN |
 151                                         __GFP_NORETRY | __GFP_NOMEMALLOC);
 152                         if (!npa) {
 153                                 pr_info("%s: Can not allocate new page list\n", __func__);
 154                                 free_page(addr);
 155                                 break;
 156                         }
 157                         spin_lock(&cmm_lock);
 158                         pa = cmm_page_list;
 159
 160                         if (!pa || pa->index >= CMM_NR_PAGES) {
 161                                 npa->next = pa;
 162                                 npa->index = 0;
 163                                 pa = npa;
 164                                 cmm_page_list = pa;
 165                         } else
 166                                 free_page((unsigned long) npa);
 167                 }
 168
 169                 if ((rc = plpar_page_set_loaned(__pa(addr)))) {
 170                         pr_err("%s: Can not set page to loaned. rc=%ld\n", __func__, rc);
 171                         spin_unlock(&cmm_lock);
 172                         free_page(addr);
 173                         break;
 174                 }
 175
 176                 pa->page[pa->index++] = addr;
 177                 loaned_pages++;
 178                 totalram_pages--;
 179                 spin_unlock(&cmm_lock);
 180                 nr--;
 181         }
 182
 183         cmm_dbg("End request with %ld pages unfulfilled\n", nr);
 184         return nr;
 185 }
 186
 187 /**
 188  * cmm_free_pages - Free pages and mark them as active
 189  * @nr: number of pages to free
 190  *
 191  * Return value:
 192  *      number of pages requested to be freed which were not
 193  **/
 194 static long cmm_free_pages(long nr)
 195 {
 196         struct cmm_page_array *pa;
 197         unsigned long addr;
 198
 199         cmm_dbg("Begin free of %ld pages.\n", nr);
 200         spin_lock(&cmm_lock);
 201         pa = cmm_page_list;
 202         while (nr) {
 203                 if (!pa || pa->index <= 0)
 204                         break;
 205                 addr = pa->page[--pa->index];
 206
 207                 if (pa->index == 0) {
 208                         pa = pa->next;
 209                         free_page((unsigned long) cmm_page_list);
 210                         cmm_page_list = pa;
 211                 }
 212
 213                 plpar_page_set_active(__pa(addr));
 214                 free_page(addr);
 215                 loaned_pages--;
 216                 nr--;
 217                 totalram_pages++;
 218         }
 219         spin_unlock(&cmm_lock);
 220         cmm_dbg("End request with %ld pages unfulfilled\n", nr);
 221         return nr;
 222 }
 223
 224 /**
 225  * cmm_oom_notify - OOM notifier
 226  * @self:       notifier block struct
 227  * @dummy:      not used
 228  * @parm:       returned - number of pages freed
 229  *
 230  * Return value:
 231  *      NOTIFY_OK
 232  **/
 233 static int cmm_oom_notify(struct notifier_block *self,
 234                           unsigned long dummy, void *parm)
 235 {
 236         unsigned long *freed = parm;
 237         long nr = KB2PAGES(oom_kb);
 238
 239         cmm_dbg("OOM processing started\n");
 240         nr = cmm_free_pages(nr);
 241         loaned_pages_target = loaned_pages;
 242         *freed += KB2PAGES(oom_kb) - nr;
 243         oom_freed_pages += KB2PAGES(oom_kb) - nr;
 244         cmm_dbg("OOM processing complete\n");
 245         return NOTIFY_OK;
 246 }
 247
 248 /**
 249  * cmm_get_mpp - Read memory performance parameters
 250  *
 251  * Makes hcall to query the current page loan request from the hypervisor.
 252  *
 253  * Return value:
 254  *      nothing
 255  **/
 256 static void cmm_get_mpp(void)
 257 {
 258         int rc;
 259         struct hvcall_mpp_data mpp_data;
 260         signed long active_pages_target, page_loan_request, target;
 261         signed long total_pages = totalram_pages + loaned_pages;
 262         signed long min_mem_pages = (min_mem_mb * 1024 * 1024) / PAGE_SIZE;
 263
 264         rc = h_get_mpp(&mpp_data);
 265
 266         if (rc != H_SUCCESS)
 267                 return;
 268
 269         page_loan_request = div_s64((s64)mpp_data.loan_request, PAGE_SIZE);
 270         target = page_loan_request + (signed long)loaned_pages;
 271
 272         if (target < 0 || total_pages < min_mem_pages)
 273                 target = 0;
 274
 275         if (target > oom_freed_pages)
 276                 target -= oom_freed_pages;
 277         else
 278                 target = 0;
 279
 280         active_pages_target = total_pages - target;
 281
 282         if (min_mem_pages > active_pages_target)
 283                 target = total_pages - min_mem_pages;
 284
 285         if (target < 0)
 286                 target = 0;
 287
 288         loaned_pages_target = target;
 289
 290         cmm_dbg("delta = %ld, loaned = %lu, target = %lu, oom = %lu, totalram = %lu\n",
 291                 page_loan_request, loaned_pages, loaned_pages_target,
 292                 oom_freed_pages, totalram_pages);
 293 }
 294
 295 static struct notifier_block cmm_oom_nb = {
 296         .notifier_call = cmm_oom_notify
 297 };
 298
 299 /**
 300  * cmm_thread - CMM task thread
 301  * @dummy:      not used
 302  *
 303  * Return value:
 304  *      0
 305  **/
 306 static int cmm_thread(void *dummy)
 307 {
 308         unsigned long timeleft;
 309
 310         while (1) {
 311                 timeleft = msleep_interruptible(delay * 1000);
 312
 313                 if (kthread_should_stop() || timeleft)
 314                         break;
 315
 316                 if (mutex_trylock(&hotplug_mutex)) {
 317                         if (hotplug_occurred) {
 318                                 hotplug_occurred = 0;
 319                                 mutex_unlock(&hotplug_mutex);
 320                                 cmm_dbg("Hotplug operation has occurred, "
 321                                                 "loaning activity suspended "
 322                                                 "for %d seconds.\n",
 323                                                 hotplug_delay);
 324                                 timeleft = msleep_interruptible(hotplug_delay *
 325                                                 1000);
 326                                 if (kthread_should_stop() || timeleft)
 327                                         break;
 328                                 continue;
 329                         }
 330                         mutex_unlock(&hotplug_mutex);
 331                 } else {
 332                         cmm_dbg("Hotplug operation in progress, activity "
 333                                         "suspended\n");
 334                         continue;
 335                 }
 336
 337                 cmm_get_mpp();
 338
 339                 if (loaned_pages_target > loaned_pages) {
 340                         if (cmm_alloc_pages(loaned_pages_target - loaned_pages))
 341                                 loaned_pages_target = loaned_pages;
 342                 } else if (loaned_pages_target < loaned_pages)
 343                         cmm_free_pages(loaned_pages - loaned_pages_target);
 344         }
 345         return 0;
 346 }
 347
 348 #define CMM_SHOW(name, format, args...)                 \
 349         static ssize_t show_##name(struct sys_device *dev,      \
 350                                    struct sysdev_attribute *attr,       \
 351                                    char *buf)                   \
 352         {                                                       \
 353                 return sprintf(buf, format, ##args);            \
 354         }                                                       \
 355         static SYSDEV_ATTR(name, S_IRUGO, show_##name, NULL)
 356
 357 CMM_SHOW(loaned_kb, "%lu\n", PAGES2KB(loaned_pages));
 358 CMM_SHOW(loaned_target_kb, "%lu\n", PAGES2KB(loaned_pages_target));
 359
 360 static ssize_t show_oom_pages(struct sys_device *dev,
 361                               struct sysdev_attribute *attr, char *buf)
 362 {
 363         return sprintf(buf, "%lu\n", PAGES2KB(oom_freed_pages));
 364 }
 365
 366 static ssize_t store_oom_pages(struct sys_device *dev,
 367                                struct sysdev_attribute *attr,
 368                                const char *buf, size_t count)
 369 {
 370         unsigned long val = simple_strtoul (buf, NULL, 10);
 371
 372         if (!capable(CAP_SYS_ADMIN))
 373                 return -EPERM;
 374         if (val != 0)
 375                 return -EBADMSG;
 376
 377         oom_freed_pages = 0;
 378         return count;
 379 }
 380
 381 static SYSDEV_ATTR(oom_freed_kb, S_IWUSR| S_IRUGO,
 382                    show_oom_pages, store_oom_pages);
 383
 384 static struct sysdev_attribute *cmm_attrs[] = {
 385         &attr_loaned_kb,
 386         &attr_loaned_target_kb,
 387         &attr_oom_freed_kb,
 388 };
 389
 390 static struct sysdev_class cmm_sysdev_class = {
 391         .name = "cmm",
 392 };
 393
 394 /**
 395  * cmm_sysfs_register - Register with sysfs
 396  *
 397  * Return value:
 398  *      0 on success / other on failure
 399  **/
 400 static int cmm_sysfs_register(struct sys_device *sysdev)
 401 {
 402         int i, rc;
 403
 404         if ((rc = sysdev_class_register(&cmm_sysdev_class)))
 405                 return rc;
 406
 407         sysdev->id = 0;
 408         sysdev->cls = &cmm_sysdev_class;
 409
 410         if ((rc = sysdev_register(sysdev)))
 411                 goto class_unregister;
 412
 413         for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++) {
 414                 if ((rc = sysdev_create_file(sysdev, cmm_attrs[i])))
 415                         goto fail;
 416         }
 417
 418         return 0;
 419
 420 fail:
 421         while (--i >= 0)
 422                 sysdev_remove_file(sysdev, cmm_attrs[i]);
 423         sysdev_unregister(sysdev);
 424 class_unregister:
 425         sysdev_class_unregister(&cmm_sysdev_class);
 426         return rc;
 427 }
 428
 429 /**
 430  * cmm_unregister_sysfs - Unregister from sysfs
 431  *
 432  **/
 433 static void cmm_unregister_sysfs(struct sys_device *sysdev)
 434 {
 435         int i;
 436
 437         for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++)
 438                 sysdev_remove_file(sysdev, cmm_attrs[i]);
 439         sysdev_unregister(sysdev);
 440         sysdev_class_unregister(&cmm_sysdev_class);
 441 }
 442
 443 /**
 444  * cmm_reboot_notifier - Make sure pages are not still marked as "loaned"
 445  *
 446  **/
 447 static int cmm_reboot_notifier(struct notifier_block *nb,
 448                                unsigned long action, void *unused)
 449 {
 450         if (action == SYS_RESTART) {
 451                 if (cmm_thread_ptr)
 452                         kthread_stop(cmm_thread_ptr);
 453                 cmm_thread_ptr = NULL;
 454                 cmm_free_pages(loaned_pages);
 455         }
 456         return NOTIFY_DONE;
 457 }
 458
 459 static struct notifier_block cmm_reboot_nb = {
 460         .notifier_call = cmm_reboot_notifier,
 461 };
 462
 463 /**
 464  * cmm_count_pages - Count the number of pages loaned in a particular range.
 465  *
 466  * @arg: memory_isolate_notify structure with address range and count
 467  *
 468  * Return value:
 469  *      0 on success
 470  **/
 471 static unsigned long cmm_count_pages(void *arg)
 472 {
 473         struct memory_isolate_notify *marg = arg;
 474         struct cmm_page_array *pa;
 475         unsigned long start = (unsigned long)pfn_to_kaddr(marg->start_pfn);
 476         unsigned long end = start + (marg->nr_pages << PAGE_SHIFT);
 477         unsigned long idx;
 478
 479         spin_lock(&cmm_lock);
 480         pa = cmm_page_list;
 481         while (pa) {
 482                 if ((unsigned long)pa >= start && (unsigned long)pa < end)
 483                         marg->pages_found++;
 484                 for (idx = 0; idx < pa->index; idx++)
 485                         if (pa->page[idx] >= start && pa->page[idx] < end)
 486                                 marg->pages_found++;
 487                 pa = pa->next;
 488         }
 489         spin_unlock(&cmm_lock);
 490         return 0;
 491 }
 492
 493 /**
 494  * cmm_memory_isolate_cb - Handle memory isolation notifier calls
 495  * @self:       notifier block struct
 496  * @action:     action to take
 497  * @arg:        struct memory_isolate_notify data for handler
 498  *
 499  * Return value:
 500  *      NOTIFY_OK or notifier error based on subfunction return value
 501  **/
 502 static int cmm_memory_isolate_cb(struct notifier_block *self,
 503                                  unsigned long action, void *arg)
 504 {
 505         int ret = 0;
 506
 507         if (action == MEM_ISOLATE_COUNT)
 508                 ret = cmm_count_pages(arg);
 509
 510         if (ret)
 511                 ret = notifier_from_errno(ret);
 512         else
 513                 ret = NOTIFY_OK;
 514
 515         return 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                                 continue;
 564                         }
 565                 }
 566                 pa_curr = pa_curr->next;
 567         }
 568
 569         /* Search for page list structures in the range to be offlined */
 570         pa_last = NULL;
 571         pa_curr = cmm_page_list;
 572         while (pa_curr) {
 573                 if (((unsigned long)pa_curr >= start_page) &&
 574                                 ((unsigned long)pa_curr < end_page)) {
 575                         npa = (struct cmm_page_array *)__get_free_page(
 576                                         GFP_NOIO | __GFP_NOWARN |
 577                                         __GFP_NORETRY | __GFP_NOMEMALLOC);
 578                         if (!npa) {
 579                                 spin_unlock(&cmm_lock);
 580                                 cmm_dbg("Failed to allocate memory for list "
 581                                                 "management. Memory hotplug "
 582                                                 "failed.\n");
 583                                 return ENOMEM;
 584                         }
 585                         memcpy(npa, pa_curr, PAGE_SIZE);
 586                         if (pa_curr == cmm_page_list)
 587                                 cmm_page_list = npa;
 588                         if (pa_last)
 589                                 pa_last->next = npa;
 590                         free_page((unsigned long) pa_curr);
 591                         freed++;
 592                         pa_curr = npa;
 593                 }
 594
 595                 pa_last = pa_curr;
 596                 pa_curr = pa_curr->next;
 597         }
 598
 599         spin_unlock(&cmm_lock);
 600         cmm_dbg("Released %ld pages in the search range.\n", freed);
 601
 602         return 0;
 603 }
 604
 605 /**
 606  * cmm_memory_cb - Handle memory hotplug notifier calls
 607  * @self:       notifier block struct
 608  * @action:     action to take
 609  * @arg:        struct memory_notify data for handler
 610  *
 611  * Return value:
 612  *      NOTIFY_OK or notifier error based on subfunction return value
 613  *
 614  **/
 615 static int cmm_memory_cb(struct notifier_block *self,
 616                         unsigned long action, void *arg)
 617 {
 618         int ret = 0;
 619
 620         switch (action) {
 621         case MEM_GOING_OFFLINE:
 622                 mutex_lock(&hotplug_mutex);
 623                 hotplug_occurred = 1;
 624                 ret = cmm_mem_going_offline(arg);
 625                 break;
 626         case MEM_OFFLINE:
 627         case MEM_CANCEL_OFFLINE:
 628                 mutex_unlock(&hotplug_mutex);
 629                 cmm_dbg("Memory offline operation complete.\n");
 630                 break;
 631         case MEM_GOING_ONLINE:
 632         case MEM_ONLINE:
 633         case MEM_CANCEL_ONLINE:
 634                 break;
 635         }
 636
 637         if (ret)
 638                 ret = notifier_from_errno(ret);
 639         else
 640                 ret = NOTIFY_OK;
 641
 642         return ret;
 643 }
 644
 645 static struct notifier_block cmm_mem_nb = {
 646         .notifier_call = cmm_memory_cb,
 647         .priority = CMM_MEM_HOTPLUG_PRI
 648 };
 649
 650 /**
 651  * cmm_init - Module initialization
 652  *
 653  * Return value:
 654  *      0 on success / other on failure
 655  **/
 656 static int cmm_init(void)
 657 {
 658         int rc = -ENOMEM;
 659
 660         if (!firmware_has_feature(FW_FEATURE_CMO))
 661                 return -EOPNOTSUPP;
 662
 663         if ((rc = register_oom_notifier(&cmm_oom_nb)) < 0)
 664                 return rc;
 665
 666         if ((rc = register_reboot_notifier(&cmm_reboot_nb)))
 667                 goto out_oom_notifier;
 668
 669         if ((rc = cmm_sysfs_register(&cmm_sysdev)))
 670                 goto out_reboot_notifier;
 671
 672         if (register_memory_notifier(&cmm_mem_nb) ||
 673             register_memory_isolate_notifier(&cmm_mem_isolate_nb))
 674                 goto out_unregister_notifier;
 675
 676         if (cmm_disabled)
 677                 return rc;
 678
 679         cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
 680         if (IS_ERR(cmm_thread_ptr)) {
 681                 rc = PTR_ERR(cmm_thread_ptr);
 682                 goto out_unregister_notifier;
 683         }
 684
 685         return rc;
 686
 687 out_unregister_notifier:
 688         unregister_memory_notifier(&cmm_mem_nb);
 689         unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
 690         cmm_unregister_sysfs(&cmm_sysdev);
 691 out_reboot_notifier:
 692         unregister_reboot_notifier(&cmm_reboot_nb);
 693 out_oom_notifier:
 694         unregister_oom_notifier(&cmm_oom_nb);
 695         return rc;
 696 }
 697
 698 /**
 699  * cmm_exit - Module exit
 700  *
 701  * Return value:
 702  *      nothing
 703  **/
 704 static void cmm_exit(void)
 705 {
 706         if (cmm_thread_ptr)
 707                 kthread_stop(cmm_thread_ptr);
 708         unregister_oom_notifier(&cmm_oom_nb);
 709         unregister_reboot_notifier(&cmm_reboot_nb);
 710         unregister_memory_notifier(&cmm_mem_nb);
 711         unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
 712         cmm_free_pages(loaned_pages);
 713         cmm_unregister_sysfs(&cmm_sysdev);
 714 }
 715
 716 /**
 717  * cmm_set_disable - Disable/Enable CMM
 718  *
 719  * Return value:
 720  *      0 on success / other on failure
 721  **/
 722 static int cmm_set_disable(const char *val, struct kernel_param *kp)
 723 {
 724         int disable = simple_strtoul(val, NULL, 10);
 725
 726         if (disable != 0 && disable != 1)
 727                 return -EINVAL;
 728
 729         if (disable && !cmm_disabled) {
 730                 if (cmm_thread_ptr)
 731                         kthread_stop(cmm_thread_ptr);
 732                 cmm_thread_ptr = NULL;
 733                 cmm_free_pages(loaned_pages);
 734         } else if (!disable && cmm_disabled) {
 735                 cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
 736                 if (IS_ERR(cmm_thread_ptr))
 737                         return PTR_ERR(cmm_thread_ptr);
 738         }
 739
 740         cmm_disabled = disable;
 741         return 0;
 742 }
 743
 744 module_param_call(disable, cmm_set_disable, param_get_uint,
 745                   &cmm_disabled, S_IRUGO | S_IWUSR);
 746 MODULE_PARM_DESC(disable, "Disable CMM. Set to 1 to disable. "
 747                  "[Default=" __stringify(CMM_DISABLE) "]");
 748
 749 module_init(cmm_init);
 750 module_exit(cmm_exit);