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 <linux/uaccess.h>
  41 #include <linux/memory.h>
  42 #include <asm/plpar_wrappers.h>
  43
  44 #include "pseries.h"
  45
  46 #define CMM_DRIVER_VERSION      "1.0.0"
  47 #define CMM_DEFAULT_DELAY       1
  48 #define CMM_HOTPLUG_DELAY       5
  49 #define CMM_DEBUG                       0
  50 #define CMM_DISABLE             0
  51 #define CMM_OOM_KB              1024
  52 #define CMM_MIN_MEM_MB          256
  53 #define KB2PAGES(_p)            ((_p)>>(PAGE_SHIFT-10))
  54 #define PAGES2KB(_p)            ((_p)<<(PAGE_SHIFT-10))
  55 /*
  56  * The priority level tries to ensure that this notifier is called as
  57  * late as possible to reduce thrashing in the shared memory pool.
  58  */
  59 #define CMM_MEM_HOTPLUG_PRI     1
  60 #define CMM_MEM_ISOLATE_PRI     15
  61
  62 static unsigned int delay = CMM_DEFAULT_DELAY;
  63 static unsigned int hotplug_delay = CMM_HOTPLUG_DELAY;
  64 static unsigned int oom_kb = CMM_OOM_KB;
  65 static unsigned int cmm_debug = CMM_DEBUG;
  66 static unsigned int cmm_disabled = CMM_DISABLE;
  67 static unsigned long min_mem_mb = CMM_MIN_MEM_MB;
  68 static struct device cmm_dev;
  69
  70 MODULE_AUTHOR("Brian King <brking@linux.vnet.ibm.com>");
  71 MODULE_DESCRIPTION("IBM System p Collaborative Memory Manager");
  72 MODULE_LICENSE("GPL");
  73 MODULE_VERSION(CMM_DRIVER_VERSION);
  74
  75 module_param_named(delay, delay, uint, S_IRUGO | S_IWUSR);
  76 MODULE_PARM_DESC(delay, "Delay (in seconds) between polls to query hypervisor paging requests. "
  77                  "[Default=" __stringify(CMM_DEFAULT_DELAY) "]");
  78 module_param_named(hotplug_delay, hotplug_delay, uint, S_IRUGO | S_IWUSR);
  79 MODULE_PARM_DESC(hotplug_delay, "Delay (in seconds) after memory hotplug remove "
  80                  "before loaning resumes. "
  81                  "[Default=" __stringify(CMM_HOTPLUG_DELAY) "]");
  82 module_param_named(oom_kb, oom_kb, uint, S_IRUGO | S_IWUSR);
  83 MODULE_PARM_DESC(oom_kb, "Amount of memory in kb to free on OOM. "
  84                  "[Default=" __stringify(CMM_OOM_KB) "]");
  85 module_param_named(min_mem_mb, min_mem_mb, ulong, S_IRUGO | S_IWUSR);
  86 MODULE_PARM_DESC(min_mem_mb, "Minimum amount of memory (in MB) to not balloon. "
  87                  "[Default=" __stringify(CMM_MIN_MEM_MB) "]");
  88 module_param_named(debug, cmm_debug, uint, S_IRUGO | S_IWUSR);
  89 MODULE_PARM_DESC(debug, "Enable module debugging logging. Set to 1 to enable. "
  90                  "[Default=" __stringify(CMM_DEBUG) "]");
  91
  92 #define CMM_NR_PAGES ((PAGE_SIZE - sizeof(void *) - sizeof(unsigned long)) / sizeof(unsigned long))
  93
  94 #define cmm_dbg(...) if (cmm_debug) { printk(KERN_INFO "cmm: "__VA_ARGS__); }
  95
  96 struct cmm_page_array {
  97         struct cmm_page_array *next;
  98         unsigned long index;
  99         unsigned long page[CMM_NR_PAGES];
 100 };
 101
 102 static unsigned long loaned_pages;
 103 static unsigned long loaned_pages_target;
 104 static unsigned long oom_freed_pages;
 105
 106 static struct cmm_page_array *cmm_page_list;
 107 static DEFINE_SPINLOCK(cmm_lock);
 108
 109 static DEFINE_MUTEX(hotplug_mutex);
 110 static int hotplug_occurred; /* protected by the hotplug mutex */
 111
 112 static struct task_struct *cmm_thread_ptr;
 113
 114 static long plpar_page_set_loaned(unsigned long vpa)
 115 {
 116         unsigned long cmo_page_sz = cmo_get_page_size();
 117         long rc = 0;
 118         int i;
 119
 120         for (i = 0; !rc && i < PAGE_SIZE; i += cmo_page_sz)
 121                 rc = plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_LOANED, vpa + i, 0);
 122
 123         for (i -= cmo_page_sz; rc && i != 0; i -= cmo_page_sz)
 124                 plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_ACTIVE,
 125                                    vpa + i - cmo_page_sz, 0);
 126
 127         return rc;
 128 }
 129
 130 static long plpar_page_set_active(unsigned long vpa)
 131 {
 132         unsigned long cmo_page_sz = cmo_get_page_size();
 133         long rc = 0;
 134         int i;
 135
 136         for (i = 0; !rc && i < PAGE_SIZE; i += cmo_page_sz)
 137                 rc = plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_ACTIVE, vpa + i, 0);
 138
 139         for (i -= cmo_page_sz; rc && i != 0; i -= cmo_page_sz)
 140                 plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_LOANED,
 141                                    vpa + i - cmo_page_sz, 0);
 142
 143         return rc;
 144 }
 145
 146 /**
 147  * cmm_alloc_pages - Allocate pages and mark them as loaned
 148  * @nr: number of pages to allocate
 149  *
 150  * Return value:
 151  *      number of pages requested to be allocated which were not
 152  **/
 153 static long cmm_alloc_pages(long nr)
 154 {
 155         struct cmm_page_array *pa, *npa;
 156         unsigned long addr;
 157         long rc;
 158
 159         cmm_dbg("Begin request for %ld pages\n", nr);
 160
 161         while (nr) {
 162                 /* Exit if a hotplug operation is in progress or occurred */
 163                 if (mutex_trylock(&hotplug_mutex)) {
 164                         if (hotplug_occurred) {
 165                                 mutex_unlock(&hotplug_mutex);
 166                                 break;
 167                         }
 168                         mutex_unlock(&hotplug_mutex);
 169                 } else {
 170                         break;
 171                 }
 172
 173                 addr = __get_free_page(GFP_NOIO | __GFP_NOWARN |
 174                                        __GFP_NORETRY | __GFP_NOMEMALLOC);
 175                 if (!addr)
 176                         break;
 177                 spin_lock(&cmm_lock);
 178                 pa = cmm_page_list;
 179                 if (!pa || pa->index >= CMM_NR_PAGES) {
 180                         /* Need a new page for the page list. */
 181                         spin_unlock(&cmm_lock);
 182                         npa = (struct cmm_page_array *)__get_free_page(
 183                                         GFP_NOIO | __GFP_NOWARN |
 184                                         __GFP_NORETRY | __GFP_NOMEMALLOC);
 185                         if (!npa) {
 186                                 pr_info("%s: Can not allocate new page list\n", __func__);
 187                                 free_page(addr);
 188                                 break;
 189                         }
 190                         spin_lock(&cmm_lock);
 191                         pa = cmm_page_list;
 192
 193                         if (!pa || pa->index >= CMM_NR_PAGES) {
 194                                 npa->next = pa;
 195                                 npa->index = 0;
 196                                 pa = npa;
 197                                 cmm_page_list = pa;
 198                         } else
 199                                 free_page((unsigned long) npa);
 200                 }
 201
 202                 if ((rc = plpar_page_set_loaned(__pa(addr)))) {
 203                         pr_err("%s: Can not set page to loaned. rc=%ld\n", __func__, rc);
 204                         spin_unlock(&cmm_lock);
 205                         free_page(addr);
 206                         break;
 207                 }
 208
 209                 pa->page[pa->index++] = addr;
 210                 loaned_pages++;
 211                 totalram_pages--;
 212                 spin_unlock(&cmm_lock);
 213                 nr--;
 214         }
 215
 216         cmm_dbg("End request with %ld pages unfulfilled\n", nr);
 217         return nr;
 218 }
 219
 220 /**
 221  * cmm_free_pages - Free pages and mark them as active
 222  * @nr: number of pages to free
 223  *
 224  * Return value:
 225  *      number of pages requested to be freed which were not
 226  **/
 227 static long cmm_free_pages(long nr)
 228 {
 229         struct cmm_page_array *pa;
 230         unsigned long addr;
 231
 232         cmm_dbg("Begin free of %ld pages.\n", nr);
 233         spin_lock(&cmm_lock);
 234         pa = cmm_page_list;
 235         while (nr) {
 236                 if (!pa || pa->index <= 0)
 237                         break;
 238                 addr = pa->page[--pa->index];
 239
 240                 if (pa->index == 0) {
 241                         pa = pa->next;
 242                         free_page((unsigned long) cmm_page_list);
 243                         cmm_page_list = pa;
 244                 }
 245
 246                 plpar_page_set_active(__pa(addr));
 247                 free_page(addr);
 248                 loaned_pages--;
 249                 nr--;
 250                 totalram_pages++;
 251         }
 252         spin_unlock(&cmm_lock);
 253         cmm_dbg("End request with %ld pages unfulfilled\n", nr);
 254         return nr;
 255 }
 256
 257 /**
 258  * cmm_oom_notify - OOM notifier
 259  * @self:       notifier block struct
 260  * @dummy:      not used
 261  * @parm:       returned - number of pages freed
 262  *
 263  * Return value:
 264  *      NOTIFY_OK
 265  **/
 266 static int cmm_oom_notify(struct notifier_block *self,
 267                           unsigned long dummy, void *parm)
 268 {
 269         unsigned long *freed = parm;
 270         long nr = KB2PAGES(oom_kb);
 271
 272         cmm_dbg("OOM processing started\n");
 273         nr = cmm_free_pages(nr);
 274         loaned_pages_target = loaned_pages;
 275         *freed += KB2PAGES(oom_kb) - nr;
 276         oom_freed_pages += KB2PAGES(oom_kb) - nr;
 277         cmm_dbg("OOM processing complete\n");
 278         return NOTIFY_OK;
 279 }
 280
 281 /**
 282  * cmm_get_mpp - Read memory performance parameters
 283  *
 284  * Makes hcall to query the current page loan request from the hypervisor.
 285  *
 286  * Return value:
 287  *      nothing
 288  **/
 289 static void cmm_get_mpp(void)
 290 {
 291         int rc;
 292         struct hvcall_mpp_data mpp_data;
 293         signed long active_pages_target, page_loan_request, target;
 294         signed long total_pages = totalram_pages + loaned_pages;
 295         signed long min_mem_pages = (min_mem_mb * 1024 * 1024) / PAGE_SIZE;
 296
 297         rc = h_get_mpp(&mpp_data);
 298
 299         if (rc != H_SUCCESS)
 300                 return;
 301
 302         page_loan_request = div_s64((s64)mpp_data.loan_request, PAGE_SIZE);
 303         target = page_loan_request + (signed long)loaned_pages;
 304
 305         if (target < 0 || total_pages < min_mem_pages)
 306                 target = 0;
 307
 308         if (target > oom_freed_pages)
 309                 target -= oom_freed_pages;
 310         else
 311                 target = 0;
 312
 313         active_pages_target = total_pages - target;
 314
 315         if (min_mem_pages > active_pages_target)
 316                 target = total_pages - min_mem_pages;
 317
 318         if (target < 0)
 319                 target = 0;
 320
 321         loaned_pages_target = target;
 322
 323         cmm_dbg("delta = %ld, loaned = %lu, target = %lu, oom = %lu, totalram = %lu\n",
 324                 page_loan_request, loaned_pages, loaned_pages_target,
 325                 oom_freed_pages, totalram_pages);
 326 }
 327
 328 static struct notifier_block cmm_oom_nb = {
 329         .notifier_call = cmm_oom_notify
 330 };
 331
 332 /**
 333  * cmm_thread - CMM task thread
 334  * @dummy:      not used
 335  *
 336  * Return value:
 337  *      0
 338  **/
 339 static int cmm_thread(void *dummy)
 340 {
 341         unsigned long timeleft;
 342
 343         while (1) {
 344                 timeleft = msleep_interruptible(delay * 1000);
 345
 346                 if (kthread_should_stop() || timeleft)
 347                         break;
 348
 349                 if (mutex_trylock(&hotplug_mutex)) {
 350                         if (hotplug_occurred) {
 351                                 hotplug_occurred = 0;
 352                                 mutex_unlock(&hotplug_mutex);
 353                                 cmm_dbg("Hotplug operation has occurred, "
 354                                                 "loaning activity suspended "
 355                                                 "for %d seconds.\n",
 356                                                 hotplug_delay);
 357                                 timeleft = msleep_interruptible(hotplug_delay *
 358                                                 1000);
 359                                 if (kthread_should_stop() || timeleft)
 360                                         break;
 361                                 continue;
 362                         }
 363                         mutex_unlock(&hotplug_mutex);
 364                 } else {
 365                         cmm_dbg("Hotplug operation in progress, activity "
 366                                         "suspended\n");
 367                         continue;
 368                 }
 369
 370                 cmm_get_mpp();
 371
 372                 if (loaned_pages_target > loaned_pages) {
 373                         if (cmm_alloc_pages(loaned_pages_target - loaned_pages))
 374                                 loaned_pages_target = loaned_pages;
 375                 } else if (loaned_pages_target < loaned_pages)
 376                         cmm_free_pages(loaned_pages - loaned_pages_target);
 377         }
 378         return 0;
 379 }
 380
 381 #define CMM_SHOW(name, format, args...)                 \
 382         static ssize_t show_##name(struct device *dev,  \
 383                                    struct device_attribute *attr,       \
 384                                    char *buf)                   \
 385         {                                                       \
 386                 return sprintf(buf, format, ##args);            \
 387         }                                                       \
 388         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 389
 390 CMM_SHOW(loaned_kb, "%lu\n", PAGES2KB(loaned_pages));
 391 CMM_SHOW(loaned_target_kb, "%lu\n", PAGES2KB(loaned_pages_target));
 392
 393 static ssize_t show_oom_pages(struct device *dev,
 394                               struct device_attribute *attr, char *buf)
 395 {
 396         return sprintf(buf, "%lu\n", PAGES2KB(oom_freed_pages));
 397 }
 398
 399 static ssize_t store_oom_pages(struct device *dev,
 400                                struct device_attribute *attr,
 401                                const char *buf, size_t count)
 402 {
 403         unsigned long val = simple_strtoul (buf, NULL, 10);
 404
 405         if (!capable(CAP_SYS_ADMIN))
 406                 return -EPERM;
 407         if (val != 0)
 408                 return -EBADMSG;
 409
 410         oom_freed_pages = 0;
 411         return count;
 412 }
 413
 414 static DEVICE_ATTR(oom_freed_kb, S_IWUSR | S_IRUGO,
 415                    show_oom_pages, store_oom_pages);
 416
 417 static struct device_attribute *cmm_attrs[] = {
 418         &dev_attr_loaned_kb,
 419         &dev_attr_loaned_target_kb,
 420         &dev_attr_oom_freed_kb,
 421 };
 422
 423 static struct bus_type cmm_subsys = {
 424         .name = "cmm",
 425         .dev_name = "cmm",
 426 };
 427
 428 /**
 429  * cmm_sysfs_register - Register with sysfs
 430  *
 431  * Return value:
 432  *      0 on success / other on failure
 433  **/
 434 static int cmm_sysfs_register(struct device *dev)
 435 {
 436         int i, rc;
 437
 438         if ((rc = subsys_system_register(&cmm_subsys, NULL)))
 439                 return rc;
 440
 441         dev->id = 0;
 442         dev->bus = &cmm_subsys;
 443
 444         if ((rc = device_register(dev)))
 445                 goto subsys_unregister;
 446
 447         for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++) {
 448                 if ((rc = device_create_file(dev, cmm_attrs[i])))
 449                         goto fail;
 450         }
 451
 452         return 0;
 453
 454 fail:
 455         while (--i >= 0)
 456                 device_remove_file(dev, cmm_attrs[i]);
 457         device_unregister(dev);
 458 subsys_unregister:
 459         bus_unregister(&cmm_subsys);
 460         return rc;
 461 }
 462
 463 /**
 464  * cmm_unregister_sysfs - Unregister from sysfs
 465  *
 466  **/
 467 static void cmm_unregister_sysfs(struct device *dev)
 468 {
 469         int i;
 470
 471         for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++)
 472                 device_remove_file(dev, cmm_attrs[i]);
 473         device_unregister(dev);
 474         bus_unregister(&cmm_subsys);
 475 }
 476
 477 /**
 478  * cmm_reboot_notifier - Make sure pages are not still marked as "loaned"
 479  *
 480  **/
 481 static int cmm_reboot_notifier(struct notifier_block *nb,
 482                                unsigned long action, void *unused)
 483 {
 484         if (action == SYS_RESTART) {
 485                 if (cmm_thread_ptr)
 486                         kthread_stop(cmm_thread_ptr);
 487                 cmm_thread_ptr = NULL;
 488                 cmm_free_pages(loaned_pages);
 489         }
 490         return NOTIFY_DONE;
 491 }
 492
 493 static struct notifier_block cmm_reboot_nb = {
 494         .notifier_call = cmm_reboot_notifier,
 495 };
 496
 497 /**
 498  * cmm_count_pages - Count the number of pages loaned in a particular range.
 499  *
 500  * @arg: memory_isolate_notify structure with address range and count
 501  *
 502  * Return value:
 503  *      0 on success
 504  **/
 505 static unsigned long cmm_count_pages(void *arg)
 506 {
 507         struct memory_isolate_notify *marg = arg;
 508         struct cmm_page_array *pa;
 509         unsigned long start = (unsigned long)pfn_to_kaddr(marg->start_pfn);
 510         unsigned long end = start + (marg->nr_pages << PAGE_SHIFT);
 511         unsigned long idx;
 512
 513         spin_lock(&cmm_lock);
 514         pa = cmm_page_list;
 515         while (pa) {
 516                 if ((unsigned long)pa >= start && (unsigned long)pa < end)
 517                         marg->pages_found++;
 518                 for (idx = 0; idx < pa->index; idx++)
 519                         if (pa->page[idx] >= start && pa->page[idx] < end)
 520                                 marg->pages_found++;
 521                 pa = pa->next;
 522         }
 523         spin_unlock(&cmm_lock);
 524         return 0;
 525 }
 526
 527 /**
 528  * cmm_memory_isolate_cb - Handle memory isolation notifier calls
 529  * @self:       notifier block struct
 530  * @action:     action to take
 531  * @arg:        struct memory_isolate_notify data for handler
 532  *
 533  * Return value:
 534  *      NOTIFY_OK or notifier error based on subfunction return value
 535  **/
 536 static int cmm_memory_isolate_cb(struct notifier_block *self,
 537                                  unsigned long action, void *arg)
 538 {
 539         int ret = 0;
 540
 541         if (action == MEM_ISOLATE_COUNT)
 542                 ret = cmm_count_pages(arg);
 543
 544         return notifier_from_errno(ret);
 545 }
 546
 547 static struct notifier_block cmm_mem_isolate_nb = {
 548         .notifier_call = cmm_memory_isolate_cb,
 549         .priority = CMM_MEM_ISOLATE_PRI
 550 };
 551
 552 /**
 553  * cmm_mem_going_offline - Unloan pages where memory is to be removed
 554  * @arg: memory_notify structure with page range to be offlined
 555  *
 556  * Return value:
 557  *      0 on success
 558  **/
 559 static int cmm_mem_going_offline(void *arg)
 560 {
 561         struct memory_notify *marg = arg;
 562         unsigned long start_page = (unsigned long)pfn_to_kaddr(marg->start_pfn);
 563         unsigned long end_page = start_page + (marg->nr_pages << PAGE_SHIFT);
 564         struct cmm_page_array *pa_curr, *pa_last, *npa;
 565         unsigned long idx;
 566         unsigned long freed = 0;
 567
 568         cmm_dbg("Memory going offline, searching 0x%lx (%ld pages).\n",
 569                         start_page, marg->nr_pages);
 570         spin_lock(&cmm_lock);
 571
 572         /* Search the page list for pages in the range to be offlined */
 573         pa_last = pa_curr = cmm_page_list;
 574         while (pa_curr) {
 575                 for (idx = (pa_curr->index - 1); (idx + 1) > 0; idx--) {
 576                         if ((pa_curr->page[idx] < start_page) ||
 577                             (pa_curr->page[idx] >= end_page))
 578                                 continue;
 579
 580                         plpar_page_set_active(__pa(pa_curr->page[idx]));
 581                         free_page(pa_curr->page[idx]);
 582                         freed++;
 583                         loaned_pages--;
 584                         totalram_pages++;
 585                         pa_curr->page[idx] = pa_last->page[--pa_last->index];
 586                         if (pa_last->index == 0) {
 587                                 if (pa_curr == pa_last)
 588                                         pa_curr = pa_last->next;
 589                                 pa_last = pa_last->next;
 590                                 free_page((unsigned long)cmm_page_list);
 591                                 cmm_page_list = pa_last;
 592                         }
 593                 }
 594                 pa_curr = pa_curr->next;
 595         }
 596
 597         /* Search for page list structures in the range to be offlined */
 598         pa_last = NULL;
 599         pa_curr = cmm_page_list;
 600         while (pa_curr) {
 601                 if (((unsigned long)pa_curr >= start_page) &&
 602                                 ((unsigned long)pa_curr < end_page)) {
 603                         npa = (struct cmm_page_array *)__get_free_page(
 604                                         GFP_NOIO | __GFP_NOWARN |
 605                                         __GFP_NORETRY | __GFP_NOMEMALLOC);
 606                         if (!npa) {
 607                                 spin_unlock(&cmm_lock);
 608                                 cmm_dbg("Failed to allocate memory for list "
 609                                                 "management. Memory hotplug "
 610                                                 "failed.\n");
 611                                 return -ENOMEM;
 612                         }
 613                         memcpy(npa, pa_curr, PAGE_SIZE);
 614                         if (pa_curr == cmm_page_list)
 615                                 cmm_page_list = npa;
 616                         if (pa_last)
 617                                 pa_last->next = npa;
 618                         free_page((unsigned long) pa_curr);
 619                         freed++;
 620                         pa_curr = npa;
 621                 }
 622
 623                 pa_last = pa_curr;
 624                 pa_curr = pa_curr->next;
 625         }
 626
 627         spin_unlock(&cmm_lock);
 628         cmm_dbg("Released %ld pages in the search range.\n", freed);
 629
 630         return 0;
 631 }
 632
 633 /**
 634  * cmm_memory_cb - Handle memory hotplug notifier calls
 635  * @self:       notifier block struct
 636  * @action:     action to take
 637  * @arg:        struct memory_notify data for handler
 638  *
 639  * Return value:
 640  *      NOTIFY_OK or notifier error based on subfunction return value
 641  *
 642  **/
 643 static int cmm_memory_cb(struct notifier_block *self,
 644                         unsigned long action, void *arg)
 645 {
 646         int ret = 0;
 647
 648         switch (action) {
 649         case MEM_GOING_OFFLINE:
 650                 mutex_lock(&hotplug_mutex);
 651                 hotplug_occurred = 1;
 652                 ret = cmm_mem_going_offline(arg);
 653                 break;
 654         case MEM_OFFLINE:
 655         case MEM_CANCEL_OFFLINE:
 656                 mutex_unlock(&hotplug_mutex);
 657                 cmm_dbg("Memory offline operation complete.\n");
 658                 break;
 659         case MEM_GOING_ONLINE:
 660         case MEM_ONLINE:
 661         case MEM_CANCEL_ONLINE:
 662                 break;
 663         }
 664
 665         return notifier_from_errno(ret);
 666 }
 667
 668 static struct notifier_block cmm_mem_nb = {
 669         .notifier_call = cmm_memory_cb,
 670         .priority = CMM_MEM_HOTPLUG_PRI
 671 };
 672
 673 /**
 674  * cmm_init - Module initialization
 675  *
 676  * Return value:
 677  *      0 on success / other on failure
 678  **/
 679 static int cmm_init(void)
 680 {
 681         int rc = -ENOMEM;
 682
 683         if (!firmware_has_feature(FW_FEATURE_CMO))
 684                 return -EOPNOTSUPP;
 685
 686         if ((rc = register_oom_notifier(&cmm_oom_nb)) < 0)
 687                 return rc;
 688
 689         if ((rc = register_reboot_notifier(&cmm_reboot_nb)))
 690                 goto out_oom_notifier;
 691
 692         if ((rc = cmm_sysfs_register(&cmm_dev)))
 693                 goto out_reboot_notifier;
 694
 695         if (register_memory_notifier(&cmm_mem_nb) ||
 696             register_memory_isolate_notifier(&cmm_mem_isolate_nb))
 697                 goto out_unregister_notifier;
 698
 699         if (cmm_disabled)
 700                 return rc;
 701
 702         cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
 703         if (IS_ERR(cmm_thread_ptr)) {
 704                 rc = PTR_ERR(cmm_thread_ptr);
 705                 goto out_unregister_notifier;
 706         }
 707
 708         return rc;
 709
 710 out_unregister_notifier:
 711         unregister_memory_notifier(&cmm_mem_nb);
 712         unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
 713         cmm_unregister_sysfs(&cmm_dev);
 714 out_reboot_notifier:
 715         unregister_reboot_notifier(&cmm_reboot_nb);
 716 out_oom_notifier:
 717         unregister_oom_notifier(&cmm_oom_nb);
 718         return rc;
 719 }
 720
 721 /**
 722  * cmm_exit - Module exit
 723  *
 724  * Return value:
 725  *      nothing
 726  **/
 727 static void cmm_exit(void)
 728 {
 729         if (cmm_thread_ptr)
 730                 kthread_stop(cmm_thread_ptr);
 731         unregister_oom_notifier(&cmm_oom_nb);
 732         unregister_reboot_notifier(&cmm_reboot_nb);
 733         unregister_memory_notifier(&cmm_mem_nb);
 734         unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
 735         cmm_free_pages(loaned_pages);
 736         cmm_unregister_sysfs(&cmm_dev);
 737 }
 738
 739 /**
 740  * cmm_set_disable - Disable/Enable CMM
 741  *
 742  * Return value:
 743  *      0 on success / other on failure
 744  **/
 745 static int cmm_set_disable(const char *val, const struct kernel_param *kp)
 746 {
 747         int disable = simple_strtoul(val, NULL, 10);
 748
 749         if (disable != 0 && disable != 1)
 750                 return -EINVAL;
 751
 752         if (disable && !cmm_disabled) {
 753                 if (cmm_thread_ptr)
 754                         kthread_stop(cmm_thread_ptr);
 755                 cmm_thread_ptr = NULL;
 756                 cmm_free_pages(loaned_pages);
 757         } else if (!disable && cmm_disabled) {
 758                 cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
 759                 if (IS_ERR(cmm_thread_ptr))
 760                         return PTR_ERR(cmm_thread_ptr);
 761         }
 762
 763         cmm_disabled = disable;
 764         return 0;
 765 }
 766
 767 module_param_call(disable, cmm_set_disable, param_get_uint,
 768                   &cmm_disabled, S_IRUGO | S_IWUSR);
 769 MODULE_PARM_DESC(disable, "Disable CMM. Set to 1 to disable. "
 770                  "[Default=" __stringify(CMM_DISABLE) "]");
 771
 772 module_init(cmm_init);
 773 module_exit(cmm_exit);