1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Collaborative memory management interface.
4  *
5  * Copyright (C) 2008 IBM Corporation
6  * Author(s): Brian King (brking@linux.vnet.ibm.com),
7  */
8 
9 #include <linux/ctype.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/fs.h>
13 #include <linux/gfp.h>
14 #include <linux/kthread.h>
15 #include <linux/module.h>
16 #include <linux/oom.h>
17 #include <linux/reboot.h>
18 #include <linux/sched.h>
19 #include <linux/stringify.h>
20 #include <linux/swap.h>
21 #include <linux/device.h>
22 #include <asm/firmware.h>
23 #include <asm/hvcall.h>
24 #include <asm/mmu.h>
25 #include <asm/pgalloc.h>
26 #include <linux/uaccess.h>
27 #include <linux/memory.h>
28 #include <asm/plpar_wrappers.h>
29 
30 #include "pseries.h"
31 
32 #define CMM_DRIVER_VERSION	"1.0.0"
33 #define CMM_DEFAULT_DELAY	1
34 #define CMM_HOTPLUG_DELAY	5
35 #define CMM_DEBUG			0
36 #define CMM_DISABLE		0
37 #define CMM_OOM_KB		1024
38 #define CMM_MIN_MEM_MB		256
39 #define KB2PAGES(_p)		((_p)>>(PAGE_SHIFT-10))
40 #define PAGES2KB(_p)		((_p)<<(PAGE_SHIFT-10))
41 /*
42  * The priority level tries to ensure that this notifier is called as
43  * late as possible to reduce thrashing in the shared memory pool.
44  */
45 #define CMM_MEM_HOTPLUG_PRI	1
46 #define CMM_MEM_ISOLATE_PRI	15
47 
48 static unsigned int delay = CMM_DEFAULT_DELAY;
49 static unsigned int hotplug_delay = CMM_HOTPLUG_DELAY;
50 static unsigned int oom_kb = CMM_OOM_KB;
51 static unsigned int cmm_debug = CMM_DEBUG;
52 static unsigned int cmm_disabled = CMM_DISABLE;
53 static unsigned long min_mem_mb = CMM_MIN_MEM_MB;
54 static struct device cmm_dev;
55 
56 MODULE_AUTHOR("Brian King <brking@linux.vnet.ibm.com>");
57 MODULE_DESCRIPTION("IBM System p Collaborative Memory Manager");
58 MODULE_LICENSE("GPL");
59 MODULE_VERSION(CMM_DRIVER_VERSION);
60 
61 module_param_named(delay, delay, uint, 0644);
62 MODULE_PARM_DESC(delay, "Delay (in seconds) between polls to query hypervisor paging requests. "
63 		 "[Default=" __stringify(CMM_DEFAULT_DELAY) "]");
64 module_param_named(hotplug_delay, hotplug_delay, uint, 0644);
65 MODULE_PARM_DESC(hotplug_delay, "Delay (in seconds) after memory hotplug remove "
66 		 "before loaning resumes. "
67 		 "[Default=" __stringify(CMM_HOTPLUG_DELAY) "]");
68 module_param_named(oom_kb, oom_kb, uint, 0644);
69 MODULE_PARM_DESC(oom_kb, "Amount of memory in kb to free on OOM. "
70 		 "[Default=" __stringify(CMM_OOM_KB) "]");
71 module_param_named(min_mem_mb, min_mem_mb, ulong, 0644);
72 MODULE_PARM_DESC(min_mem_mb, "Minimum amount of memory (in MB) to not balloon. "
73 		 "[Default=" __stringify(CMM_MIN_MEM_MB) "]");
74 module_param_named(debug, cmm_debug, uint, 0644);
75 MODULE_PARM_DESC(debug, "Enable module debugging logging. Set to 1 to enable. "
76 		 "[Default=" __stringify(CMM_DEBUG) "]");
77 
78 #define CMM_NR_PAGES ((PAGE_SIZE - sizeof(void *) - sizeof(unsigned long)) / sizeof(unsigned long))
79 
80 #define cmm_dbg(...) if (cmm_debug) { printk(KERN_INFO "cmm: "__VA_ARGS__); }
81 
82 struct cmm_page_array {
83 	struct cmm_page_array *next;
84 	unsigned long index;
85 	unsigned long page[CMM_NR_PAGES];
86 };
87 
88 static unsigned long loaned_pages;
89 static unsigned long loaned_pages_target;
90 static unsigned long oom_freed_pages;
91 
92 static struct cmm_page_array *cmm_page_list;
93 static DEFINE_SPINLOCK(cmm_lock);
94 
95 static DEFINE_MUTEX(hotplug_mutex);
96 static int hotplug_occurred; /* protected by the hotplug mutex */
97 
98 static struct task_struct *cmm_thread_ptr;
99 
100 static long plpar_page_set_loaned(unsigned long vpa)
101 {
102 	unsigned long cmo_page_sz = cmo_get_page_size();
103 	long rc = 0;
104 	int i;
105 
106 	for (i = 0; !rc && i < PAGE_SIZE; i += cmo_page_sz)
107 		rc = plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_LOANED, vpa + i, 0);
108 
109 	for (i -= cmo_page_sz; rc && i != 0; i -= cmo_page_sz)
110 		plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_ACTIVE,
111 				   vpa + i - cmo_page_sz, 0);
112 
113 	return rc;
114 }
115 
116 static long plpar_page_set_active(unsigned long vpa)
117 {
118 	unsigned long cmo_page_sz = cmo_get_page_size();
119 	long rc = 0;
120 	int i;
121 
122 	for (i = 0; !rc && i < PAGE_SIZE; i += cmo_page_sz)
123 		rc = plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_ACTIVE, vpa + i, 0);
124 
125 	for (i -= cmo_page_sz; rc && i != 0; i -= cmo_page_sz)
126 		plpar_hcall_norets(H_PAGE_INIT, H_PAGE_SET_LOANED,
127 				   vpa + i - cmo_page_sz, 0);
128 
129 	return rc;
130 }
131 
132 /**
133  * cmm_alloc_pages - Allocate pages and mark them as loaned
134  * @nr:	number of pages to allocate
135  *
136  * Return value:
137  * 	number of pages requested to be allocated which were not
138  **/
139 static long cmm_alloc_pages(long nr)
140 {
141 	struct cmm_page_array *pa, *npa;
142 	unsigned long addr;
143 	long rc;
144 
145 	cmm_dbg("Begin request for %ld pages\n", nr);
146 
147 	while (nr) {
148 		/* Exit if a hotplug operation is in progress or occurred */
149 		if (mutex_trylock(&hotplug_mutex)) {
150 			if (hotplug_occurred) {
151 				mutex_unlock(&hotplug_mutex);
152 				break;
153 			}
154 			mutex_unlock(&hotplug_mutex);
155 		} else {
156 			break;
157 		}
158 
159 		addr = __get_free_page(GFP_NOIO | __GFP_NOWARN |
160 				       __GFP_NORETRY | __GFP_NOMEMALLOC);
161 		if (!addr)
162 			break;
163 		spin_lock(&cmm_lock);
164 		pa = cmm_page_list;
165 		if (!pa || pa->index >= CMM_NR_PAGES) {
166 			/* Need a new page for the page list. */
167 			spin_unlock(&cmm_lock);
168 			npa = (struct cmm_page_array *)__get_free_page(
169 					GFP_NOIO | __GFP_NOWARN |
170 					__GFP_NORETRY | __GFP_NOMEMALLOC);
171 			if (!npa) {
172 				pr_info("%s: Can not allocate new page list\n", __func__);
173 				free_page(addr);
174 				break;
175 			}
176 			spin_lock(&cmm_lock);
177 			pa = cmm_page_list;
178 
179 			if (!pa || pa->index >= CMM_NR_PAGES) {
180 				npa->next = pa;
181 				npa->index = 0;
182 				pa = npa;
183 				cmm_page_list = pa;
184 			} else
185 				free_page((unsigned long) npa);
186 		}
187 
188 		if ((rc = plpar_page_set_loaned(__pa(addr)))) {
189 			pr_err("%s: Can not set page to loaned. rc=%ld\n", __func__, rc);
190 			spin_unlock(&cmm_lock);
191 			free_page(addr);
192 			break;
193 		}
194 
195 		pa->page[pa->index++] = addr;
196 		loaned_pages++;
197 		totalram_pages_dec();
198 		spin_unlock(&cmm_lock);
199 		nr--;
200 	}
201 
202 	cmm_dbg("End request with %ld pages unfulfilled\n", nr);
203 	return nr;
204 }
205 
206 /**
207  * cmm_free_pages - Free pages and mark them as active
208  * @nr:	number of pages to free
209  *
210  * Return value:
211  * 	number of pages requested to be freed which were not
212  **/
213 static long cmm_free_pages(long nr)
214 {
215 	struct cmm_page_array *pa;
216 	unsigned long addr;
217 
218 	cmm_dbg("Begin free of %ld pages.\n", nr);
219 	spin_lock(&cmm_lock);
220 	pa = cmm_page_list;
221 	while (nr) {
222 		if (!pa || pa->index <= 0)
223 			break;
224 		addr = pa->page[--pa->index];
225 
226 		if (pa->index == 0) {
227 			pa = pa->next;
228 			free_page((unsigned long) cmm_page_list);
229 			cmm_page_list = pa;
230 		}
231 
232 		plpar_page_set_active(__pa(addr));
233 		free_page(addr);
234 		loaned_pages--;
235 		nr--;
236 		totalram_pages_inc();
237 	}
238 	spin_unlock(&cmm_lock);
239 	cmm_dbg("End request with %ld pages unfulfilled\n", nr);
240 	return nr;
241 }
242 
243 /**
244  * cmm_oom_notify - OOM notifier
245  * @self:	notifier block struct
246  * @dummy:	not used
247  * @parm:	returned - number of pages freed
248  *
249  * Return value:
250  * 	NOTIFY_OK
251  **/
252 static int cmm_oom_notify(struct notifier_block *self,
253 			  unsigned long dummy, void *parm)
254 {
255 	unsigned long *freed = parm;
256 	long nr = KB2PAGES(oom_kb);
257 
258 	cmm_dbg("OOM processing started\n");
259 	nr = cmm_free_pages(nr);
260 	loaned_pages_target = loaned_pages;
261 	*freed += KB2PAGES(oom_kb) - nr;
262 	oom_freed_pages += KB2PAGES(oom_kb) - nr;
263 	cmm_dbg("OOM processing complete\n");
264 	return NOTIFY_OK;
265 }
266 
267 /**
268  * cmm_get_mpp - Read memory performance parameters
269  *
270  * Makes hcall to query the current page loan request from the hypervisor.
271  *
272  * Return value:
273  * 	nothing
274  **/
275 static void cmm_get_mpp(void)
276 {
277 	int rc;
278 	struct hvcall_mpp_data mpp_data;
279 	signed long active_pages_target, page_loan_request, target;
280 	signed long total_pages = totalram_pages() + loaned_pages;
281 	signed long min_mem_pages = (min_mem_mb * 1024 * 1024) / PAGE_SIZE;
282 
283 	rc = h_get_mpp(&mpp_data);
284 
285 	if (rc != H_SUCCESS)
286 		return;
287 
288 	page_loan_request = div_s64((s64)mpp_data.loan_request, PAGE_SIZE);
289 	target = page_loan_request + (signed long)loaned_pages;
290 
291 	if (target < 0 || total_pages < min_mem_pages)
292 		target = 0;
293 
294 	if (target > oom_freed_pages)
295 		target -= oom_freed_pages;
296 	else
297 		target = 0;
298 
299 	active_pages_target = total_pages - target;
300 
301 	if (min_mem_pages > active_pages_target)
302 		target = total_pages - min_mem_pages;
303 
304 	if (target < 0)
305 		target = 0;
306 
307 	loaned_pages_target = target;
308 
309 	cmm_dbg("delta = %ld, loaned = %lu, target = %lu, oom = %lu, totalram = %lu\n",
310 		page_loan_request, loaned_pages, loaned_pages_target,
311 		oom_freed_pages, totalram_pages());
312 }
313 
314 static struct notifier_block cmm_oom_nb = {
315 	.notifier_call = cmm_oom_notify
316 };
317 
318 /**
319  * cmm_thread - CMM task thread
320  * @dummy:	not used
321  *
322  * Return value:
323  * 	0
324  **/
325 static int cmm_thread(void *dummy)
326 {
327 	unsigned long timeleft;
328 
329 	while (1) {
330 		timeleft = msleep_interruptible(delay * 1000);
331 
332 		if (kthread_should_stop() || timeleft)
333 			break;
334 
335 		if (mutex_trylock(&hotplug_mutex)) {
336 			if (hotplug_occurred) {
337 				hotplug_occurred = 0;
338 				mutex_unlock(&hotplug_mutex);
339 				cmm_dbg("Hotplug operation has occurred, "
340 						"loaning activity suspended "
341 						"for %d seconds.\n",
342 						hotplug_delay);
343 				timeleft = msleep_interruptible(hotplug_delay *
344 						1000);
345 				if (kthread_should_stop() || timeleft)
346 					break;
347 				continue;
348 			}
349 			mutex_unlock(&hotplug_mutex);
350 		} else {
351 			cmm_dbg("Hotplug operation in progress, activity "
352 					"suspended\n");
353 			continue;
354 		}
355 
356 		cmm_get_mpp();
357 
358 		if (loaned_pages_target > loaned_pages) {
359 			if (cmm_alloc_pages(loaned_pages_target - loaned_pages))
360 				loaned_pages_target = loaned_pages;
361 		} else if (loaned_pages_target < loaned_pages)
362 			cmm_free_pages(loaned_pages - loaned_pages_target);
363 	}
364 	return 0;
365 }
366 
367 #define CMM_SHOW(name, format, args...)			\
368 	static ssize_t show_##name(struct device *dev,	\
369 				   struct device_attribute *attr,	\
370 				   char *buf)			\
371 	{							\
372 		return sprintf(buf, format, ##args);		\
373 	}							\
374 	static DEVICE_ATTR(name, 0444, show_##name, NULL)
375 
376 CMM_SHOW(loaned_kb, "%lu\n", PAGES2KB(loaned_pages));
377 CMM_SHOW(loaned_target_kb, "%lu\n", PAGES2KB(loaned_pages_target));
378 
379 static ssize_t show_oom_pages(struct device *dev,
380 			      struct device_attribute *attr, char *buf)
381 {
382 	return sprintf(buf, "%lu\n", PAGES2KB(oom_freed_pages));
383 }
384 
385 static ssize_t store_oom_pages(struct device *dev,
386 			       struct device_attribute *attr,
387 			       const char *buf, size_t count)
388 {
389 	unsigned long val = simple_strtoul (buf, NULL, 10);
390 
391 	if (!capable(CAP_SYS_ADMIN))
392 		return -EPERM;
393 	if (val != 0)
394 		return -EBADMSG;
395 
396 	oom_freed_pages = 0;
397 	return count;
398 }
399 
400 static DEVICE_ATTR(oom_freed_kb, 0644,
401 		   show_oom_pages, store_oom_pages);
402 
403 static struct device_attribute *cmm_attrs[] = {
404 	&dev_attr_loaned_kb,
405 	&dev_attr_loaned_target_kb,
406 	&dev_attr_oom_freed_kb,
407 };
408 
409 static struct bus_type cmm_subsys = {
410 	.name = "cmm",
411 	.dev_name = "cmm",
412 };
413 
414 /**
415  * cmm_sysfs_register - Register with sysfs
416  *
417  * Return value:
418  * 	0 on success / other on failure
419  **/
420 static int cmm_sysfs_register(struct device *dev)
421 {
422 	int i, rc;
423 
424 	if ((rc = subsys_system_register(&cmm_subsys, NULL)))
425 		return rc;
426 
427 	dev->id = 0;
428 	dev->bus = &cmm_subsys;
429 
430 	if ((rc = device_register(dev)))
431 		goto subsys_unregister;
432 
433 	for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++) {
434 		if ((rc = device_create_file(dev, cmm_attrs[i])))
435 			goto fail;
436 	}
437 
438 	return 0;
439 
440 fail:
441 	while (--i >= 0)
442 		device_remove_file(dev, cmm_attrs[i]);
443 	device_unregister(dev);
444 subsys_unregister:
445 	bus_unregister(&cmm_subsys);
446 	return rc;
447 }
448 
449 /**
450  * cmm_unregister_sysfs - Unregister from sysfs
451  *
452  **/
453 static void cmm_unregister_sysfs(struct device *dev)
454 {
455 	int i;
456 
457 	for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++)
458 		device_remove_file(dev, cmm_attrs[i]);
459 	device_unregister(dev);
460 	bus_unregister(&cmm_subsys);
461 }
462 
463 /**
464  * cmm_reboot_notifier - Make sure pages are not still marked as "loaned"
465  *
466  **/
467 static int cmm_reboot_notifier(struct notifier_block *nb,
468 			       unsigned long action, void *unused)
469 {
470 	if (action == SYS_RESTART) {
471 		if (cmm_thread_ptr)
472 			kthread_stop(cmm_thread_ptr);
473 		cmm_thread_ptr = NULL;
474 		cmm_free_pages(loaned_pages);
475 	}
476 	return NOTIFY_DONE;
477 }
478 
479 static struct notifier_block cmm_reboot_nb = {
480 	.notifier_call = cmm_reboot_notifier,
481 };
482 
483 /**
484  * cmm_count_pages - Count the number of pages loaned in a particular range.
485  *
486  * @arg: memory_isolate_notify structure with address range and count
487  *
488  * Return value:
489  *      0 on success
490  **/
491 static unsigned long cmm_count_pages(void *arg)
492 {
493 	struct memory_isolate_notify *marg = arg;
494 	struct cmm_page_array *pa;
495 	unsigned long start = (unsigned long)pfn_to_kaddr(marg->start_pfn);
496 	unsigned long end = start + (marg->nr_pages << PAGE_SHIFT);
497 	unsigned long idx;
498 
499 	spin_lock(&cmm_lock);
500 	pa = cmm_page_list;
501 	while (pa) {
502 		if ((unsigned long)pa >= start && (unsigned long)pa < end)
503 			marg->pages_found++;
504 		for (idx = 0; idx < pa->index; idx++)
505 			if (pa->page[idx] >= start && pa->page[idx] < end)
506 				marg->pages_found++;
507 		pa = pa->next;
508 	}
509 	spin_unlock(&cmm_lock);
510 	return 0;
511 }
512 
513 /**
514  * cmm_memory_isolate_cb - Handle memory isolation notifier calls
515  * @self:	notifier block struct
516  * @action:	action to take
517  * @arg:	struct memory_isolate_notify data for handler
518  *
519  * Return value:
520  *	NOTIFY_OK or notifier error based on subfunction return value
521  **/
522 static int cmm_memory_isolate_cb(struct notifier_block *self,
523 				 unsigned long action, void *arg)
524 {
525 	int ret = 0;
526 
527 	if (action == MEM_ISOLATE_COUNT)
528 		ret = cmm_count_pages(arg);
529 
530 	return notifier_from_errno(ret);
531 }
532 
533 static struct notifier_block cmm_mem_isolate_nb = {
534 	.notifier_call = cmm_memory_isolate_cb,
535 	.priority = CMM_MEM_ISOLATE_PRI
536 };
537 
538 /**
539  * cmm_mem_going_offline - Unloan pages where memory is to be removed
540  * @arg: memory_notify structure with page range to be offlined
541  *
542  * Return value:
543  *	0 on success
544  **/
545 static int cmm_mem_going_offline(void *arg)
546 {
547 	struct memory_notify *marg = arg;
548 	unsigned long start_page = (unsigned long)pfn_to_kaddr(marg->start_pfn);
549 	unsigned long end_page = start_page + (marg->nr_pages << PAGE_SHIFT);
550 	struct cmm_page_array *pa_curr, *pa_last, *npa;
551 	unsigned long idx;
552 	unsigned long freed = 0;
553 
554 	cmm_dbg("Memory going offline, searching 0x%lx (%ld pages).\n",
555 			start_page, marg->nr_pages);
556 	spin_lock(&cmm_lock);
557 
558 	/* Search the page list for pages in the range to be offlined */
559 	pa_last = pa_curr = cmm_page_list;
560 	while (pa_curr) {
561 		for (idx = (pa_curr->index - 1); (idx + 1) > 0; idx--) {
562 			if ((pa_curr->page[idx] < start_page) ||
563 			    (pa_curr->page[idx] >= end_page))
564 				continue;
565 
566 			plpar_page_set_active(__pa(pa_curr->page[idx]));
567 			free_page(pa_curr->page[idx]);
568 			freed++;
569 			loaned_pages--;
570 			totalram_pages_inc();
571 			pa_curr->page[idx] = pa_last->page[--pa_last->index];
572 			if (pa_last->index == 0) {
573 				if (pa_curr == pa_last)
574 					pa_curr = pa_last->next;
575 				pa_last = pa_last->next;
576 				free_page((unsigned long)cmm_page_list);
577 				cmm_page_list = pa_last;
578 			}
579 		}
580 		pa_curr = pa_curr->next;
581 	}
582 
583 	/* Search for page list structures in the range to be offlined */
584 	pa_last = NULL;
585 	pa_curr = cmm_page_list;
586 	while (pa_curr) {
587 		if (((unsigned long)pa_curr >= start_page) &&
588 				((unsigned long)pa_curr < end_page)) {
589 			npa = (struct cmm_page_array *)__get_free_page(
590 					GFP_NOIO | __GFP_NOWARN |
591 					__GFP_NORETRY | __GFP_NOMEMALLOC);
592 			if (!npa) {
593 				spin_unlock(&cmm_lock);
594 				cmm_dbg("Failed to allocate memory for list "
595 						"management. Memory hotplug "
596 						"failed.\n");
597 				return -ENOMEM;
598 			}
599 			memcpy(npa, pa_curr, PAGE_SIZE);
600 			if (pa_curr == cmm_page_list)
601 				cmm_page_list = npa;
602 			if (pa_last)
603 				pa_last->next = npa;
604 			free_page((unsigned long) pa_curr);
605 			freed++;
606 			pa_curr = npa;
607 		}
608 
609 		pa_last = pa_curr;
610 		pa_curr = pa_curr->next;
611 	}
612 
613 	spin_unlock(&cmm_lock);
614 	cmm_dbg("Released %ld pages in the search range.\n", freed);
615 
616 	return 0;
617 }
618 
619 /**
620  * cmm_memory_cb - Handle memory hotplug notifier calls
621  * @self:	notifier block struct
622  * @action:	action to take
623  * @arg:	struct memory_notify data for handler
624  *
625  * Return value:
626  *	NOTIFY_OK or notifier error based on subfunction return value
627  *
628  **/
629 static int cmm_memory_cb(struct notifier_block *self,
630 			unsigned long action, void *arg)
631 {
632 	int ret = 0;
633 
634 	switch (action) {
635 	case MEM_GOING_OFFLINE:
636 		mutex_lock(&hotplug_mutex);
637 		hotplug_occurred = 1;
638 		ret = cmm_mem_going_offline(arg);
639 		break;
640 	case MEM_OFFLINE:
641 	case MEM_CANCEL_OFFLINE:
642 		mutex_unlock(&hotplug_mutex);
643 		cmm_dbg("Memory offline operation complete.\n");
644 		break;
645 	case MEM_GOING_ONLINE:
646 	case MEM_ONLINE:
647 	case MEM_CANCEL_ONLINE:
648 		break;
649 	}
650 
651 	return notifier_from_errno(ret);
652 }
653 
654 static struct notifier_block cmm_mem_nb = {
655 	.notifier_call = cmm_memory_cb,
656 	.priority = CMM_MEM_HOTPLUG_PRI
657 };
658 
659 /**
660  * cmm_init - Module initialization
661  *
662  * Return value:
663  * 	0 on success / other on failure
664  **/
665 static int cmm_init(void)
666 {
667 	int rc = -ENOMEM;
668 
669 	if (!firmware_has_feature(FW_FEATURE_CMO))
670 		return -EOPNOTSUPP;
671 
672 	if ((rc = register_oom_notifier(&cmm_oom_nb)) < 0)
673 		return rc;
674 
675 	if ((rc = register_reboot_notifier(&cmm_reboot_nb)))
676 		goto out_oom_notifier;
677 
678 	if ((rc = cmm_sysfs_register(&cmm_dev)))
679 		goto out_reboot_notifier;
680 
681 	if (register_memory_notifier(&cmm_mem_nb) ||
682 	    register_memory_isolate_notifier(&cmm_mem_isolate_nb))
683 		goto out_unregister_notifier;
684 
685 	if (cmm_disabled)
686 		return rc;
687 
688 	cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
689 	if (IS_ERR(cmm_thread_ptr)) {
690 		rc = PTR_ERR(cmm_thread_ptr);
691 		goto out_unregister_notifier;
692 	}
693 
694 	return rc;
695 
696 out_unregister_notifier:
697 	unregister_memory_notifier(&cmm_mem_nb);
698 	unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
699 	cmm_unregister_sysfs(&cmm_dev);
700 out_reboot_notifier:
701 	unregister_reboot_notifier(&cmm_reboot_nb);
702 out_oom_notifier:
703 	unregister_oom_notifier(&cmm_oom_nb);
704 	return rc;
705 }
706 
707 /**
708  * cmm_exit - Module exit
709  *
710  * Return value:
711  * 	nothing
712  **/
713 static void cmm_exit(void)
714 {
715 	if (cmm_thread_ptr)
716 		kthread_stop(cmm_thread_ptr);
717 	unregister_oom_notifier(&cmm_oom_nb);
718 	unregister_reboot_notifier(&cmm_reboot_nb);
719 	unregister_memory_notifier(&cmm_mem_nb);
720 	unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
721 	cmm_free_pages(loaned_pages);
722 	cmm_unregister_sysfs(&cmm_dev);
723 }
724 
725 /**
726  * cmm_set_disable - Disable/Enable CMM
727  *
728  * Return value:
729  * 	0 on success / other on failure
730  **/
731 static int cmm_set_disable(const char *val, const struct kernel_param *kp)
732 {
733 	int disable = simple_strtoul(val, NULL, 10);
734 
735 	if (disable != 0 && disable != 1)
736 		return -EINVAL;
737 
738 	if (disable && !cmm_disabled) {
739 		if (cmm_thread_ptr)
740 			kthread_stop(cmm_thread_ptr);
741 		cmm_thread_ptr = NULL;
742 		cmm_free_pages(loaned_pages);
743 	} else if (!disable && cmm_disabled) {
744 		cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
745 		if (IS_ERR(cmm_thread_ptr))
746 			return PTR_ERR(cmm_thread_ptr);
747 	}
748 
749 	cmm_disabled = disable;
750 	return 0;
751 }
752 
753 module_param_call(disable, cmm_set_disable, param_get_uint,
754 		  &cmm_disabled, 0644);
755 MODULE_PARM_DESC(disable, "Disable CMM. Set to 1 to disable. "
756 		 "[Default=" __stringify(CMM_DISABLE) "]");
757 
758 module_init(cmm_init);
759 module_exit(cmm_exit);
760