1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
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
55
56
57#define CMM_MEM_HOTPLUG_PRI 1
58#define CMM_MEM_ISOLATE_PRI 15
59
60static unsigned int delay = CMM_DEFAULT_DELAY;
61static unsigned int hotplug_delay = CMM_HOTPLUG_DELAY;
62static unsigned int oom_kb = CMM_OOM_KB;
63static unsigned int cmm_debug = CMM_DEBUG;
64static unsigned int cmm_disabled = CMM_DISABLE;
65static unsigned long min_mem_mb = CMM_MIN_MEM_MB;
66static struct device cmm_dev;
67
68MODULE_AUTHOR("Brian King <brking@linux.vnet.ibm.com>");
69MODULE_DESCRIPTION("IBM System p Collaborative Memory Manager");
70MODULE_LICENSE("GPL");
71MODULE_VERSION(CMM_DRIVER_VERSION);
72
73module_param_named(delay, delay, uint, S_IRUGO | S_IWUSR);
74MODULE_PARM_DESC(delay, "Delay (in seconds) between polls to query hypervisor paging requests. "
75 "[Default=" __stringify(CMM_DEFAULT_DELAY) "]");
76module_param_named(hotplug_delay, hotplug_delay, uint, S_IRUGO | S_IWUSR);
77MODULE_PARM_DESC(delay, "Delay (in seconds) after memory hotplug remove "
78 "before loaning resumes. "
79 "[Default=" __stringify(CMM_HOTPLUG_DELAY) "]");
80module_param_named(oom_kb, oom_kb, uint, S_IRUGO | S_IWUSR);
81MODULE_PARM_DESC(oom_kb, "Amount of memory in kb to free on OOM. "
82 "[Default=" __stringify(CMM_OOM_KB) "]");
83module_param_named(min_mem_mb, min_mem_mb, ulong, S_IRUGO | S_IWUSR);
84MODULE_PARM_DESC(min_mem_mb, "Minimum amount of memory (in MB) to not balloon. "
85 "[Default=" __stringify(CMM_MIN_MEM_MB) "]");
86module_param_named(debug, cmm_debug, uint, S_IRUGO | S_IWUSR);
87MODULE_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
94struct cmm_page_array {
95 struct cmm_page_array *next;
96 unsigned long index;
97 unsigned long page[CMM_NR_PAGES];
98};
99
100static unsigned long loaned_pages;
101static unsigned long loaned_pages_target;
102static unsigned long oom_freed_pages;
103
104static struct cmm_page_array *cmm_page_list;
105static DEFINE_SPINLOCK(cmm_lock);
106
107static DEFINE_MUTEX(hotplug_mutex);
108static int hotplug_occurred;
109
110static struct task_struct *cmm_thread_ptr;
111
112
113
114
115
116
117
118
119static 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
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
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
188
189
190
191
192
193static 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
225
226
227
228
229
230
231
232static 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
249
250
251
252
253
254
255static 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
294static struct notifier_block cmm_oom_nb = {
295 .notifier_call = cmm_oom_notify
296};
297
298
299
300
301
302
303
304
305static 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
356CMM_SHOW(loaned_kb, "%lu\n", PAGES2KB(loaned_pages));
357CMM_SHOW(loaned_target_kb, "%lu\n", PAGES2KB(loaned_pages_target));
358
359static 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
365static 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
380static DEVICE_ATTR(oom_freed_kb, S_IWUSR | S_IRUGO,
381 show_oom_pages, store_oom_pages);
382
383static struct device_attribute *cmm_attrs[] = {
384 &dev_attr_loaned_kb,
385 &dev_attr_loaned_target_kb,
386 &dev_attr_oom_freed_kb,
387};
388
389static struct bus_type cmm_subsys = {
390 .name = "cmm",
391 .dev_name = "cmm",
392};
393
394
395
396
397
398
399
400static int cmm_sysfs_register(struct device *dev)
401{
402 int i, rc;
403
404 if ((rc = subsys_system_register(&cmm_subsys, NULL)))
405 return rc;
406
407 dev->id = 0;
408 dev->bus = &cmm_subsys;
409
410 if ((rc = device_register(dev)))
411 goto subsys_unregister;
412
413 for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++) {
414 if ((rc = device_create_file(dev, cmm_attrs[i])))
415 goto fail;
416 }
417
418 return 0;
419
420fail:
421 while (--i >= 0)
422 device_remove_file(dev, cmm_attrs[i]);
423 device_unregister(dev);
424subsys_unregister:
425 bus_unregister(&cmm_subsys);
426 return rc;
427}
428
429
430
431
432
433static void cmm_unregister_sysfs(struct device *dev)
434{
435 int i;
436
437 for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++)
438 device_remove_file(dev, cmm_attrs[i]);
439 device_unregister(dev);
440 bus_unregister(&cmm_subsys);
441}
442
443
444
445
446
447static 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
459static struct notifier_block cmm_reboot_nb = {
460 .notifier_call = cmm_reboot_notifier,
461};
462
463
464
465
466
467
468
469
470
471static 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
495
496
497
498
499
500
501
502static 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 return notifier_from_errno(ret);
511}
512
513static struct notifier_block cmm_mem_isolate_nb = {
514 .notifier_call = cmm_memory_isolate_cb,
515 .priority = CMM_MEM_ISOLATE_PRI
516};
517
518
519
520
521
522
523
524
525static int cmm_mem_going_offline(void *arg)
526{
527 struct memory_notify *marg = arg;
528 unsigned long start_page = (unsigned long)pfn_to_kaddr(marg->start_pfn);
529 unsigned long end_page = start_page + (marg->nr_pages << PAGE_SHIFT);
530 struct cmm_page_array *pa_curr, *pa_last, *npa;
531 unsigned long idx;
532 unsigned long freed = 0;
533
534 cmm_dbg("Memory going offline, searching 0x%lx (%ld pages).\n",
535 start_page, marg->nr_pages);
536 spin_lock(&cmm_lock);
537
538
539 pa_last = pa_curr = cmm_page_list;
540 while (pa_curr) {
541 for (idx = (pa_curr->index - 1); (idx + 1) > 0; idx--) {
542 if ((pa_curr->page[idx] < start_page) ||
543 (pa_curr->page[idx] >= end_page))
544 continue;
545
546 plpar_page_set_active(__pa(pa_curr->page[idx]));
547 free_page(pa_curr->page[idx]);
548 freed++;
549 loaned_pages--;
550 totalram_pages++;
551 pa_curr->page[idx] = pa_last->page[--pa_last->index];
552 if (pa_last->index == 0) {
553 if (pa_curr == pa_last)
554 pa_curr = pa_last->next;
555 pa_last = pa_last->next;
556 free_page((unsigned long)cmm_page_list);
557 cmm_page_list = pa_last;
558 }
559 }
560 pa_curr = pa_curr->next;
561 }
562
563
564 pa_last = NULL;
565 pa_curr = cmm_page_list;
566 while (pa_curr) {
567 if (((unsigned long)pa_curr >= start_page) &&
568 ((unsigned long)pa_curr < end_page)) {
569 npa = (struct cmm_page_array *)__get_free_page(
570 GFP_NOIO | __GFP_NOWARN |
571 __GFP_NORETRY | __GFP_NOMEMALLOC);
572 if (!npa) {
573 spin_unlock(&cmm_lock);
574 cmm_dbg("Failed to allocate memory for list "
575 "management. Memory hotplug "
576 "failed.\n");
577 return ENOMEM;
578 }
579 memcpy(npa, pa_curr, PAGE_SIZE);
580 if (pa_curr == cmm_page_list)
581 cmm_page_list = npa;
582 if (pa_last)
583 pa_last->next = npa;
584 free_page((unsigned long) pa_curr);
585 freed++;
586 pa_curr = npa;
587 }
588
589 pa_last = pa_curr;
590 pa_curr = pa_curr->next;
591 }
592
593 spin_unlock(&cmm_lock);
594 cmm_dbg("Released %ld pages in the search range.\n", freed);
595
596 return 0;
597}
598
599
600
601
602
603
604
605
606
607
608
609static int cmm_memory_cb(struct notifier_block *self,
610 unsigned long action, void *arg)
611{
612 int ret = 0;
613
614 switch (action) {
615 case MEM_GOING_OFFLINE:
616 mutex_lock(&hotplug_mutex);
617 hotplug_occurred = 1;
618 ret = cmm_mem_going_offline(arg);
619 break;
620 case MEM_OFFLINE:
621 case MEM_CANCEL_OFFLINE:
622 mutex_unlock(&hotplug_mutex);
623 cmm_dbg("Memory offline operation complete.\n");
624 break;
625 case MEM_GOING_ONLINE:
626 case MEM_ONLINE:
627 case MEM_CANCEL_ONLINE:
628 break;
629 }
630
631 return notifier_from_errno(ret);
632}
633
634static struct notifier_block cmm_mem_nb = {
635 .notifier_call = cmm_memory_cb,
636 .priority = CMM_MEM_HOTPLUG_PRI
637};
638
639
640
641
642
643
644
645static int cmm_init(void)
646{
647 int rc = -ENOMEM;
648
649 if (!firmware_has_feature(FW_FEATURE_CMO))
650 return -EOPNOTSUPP;
651
652 if ((rc = register_oom_notifier(&cmm_oom_nb)) < 0)
653 return rc;
654
655 if ((rc = register_reboot_notifier(&cmm_reboot_nb)))
656 goto out_oom_notifier;
657
658 if ((rc = cmm_sysfs_register(&cmm_dev)))
659 goto out_reboot_notifier;
660
661 if (register_memory_notifier(&cmm_mem_nb) ||
662 register_memory_isolate_notifier(&cmm_mem_isolate_nb))
663 goto out_unregister_notifier;
664
665 if (cmm_disabled)
666 return rc;
667
668 cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
669 if (IS_ERR(cmm_thread_ptr)) {
670 rc = PTR_ERR(cmm_thread_ptr);
671 goto out_unregister_notifier;
672 }
673
674 return rc;
675
676out_unregister_notifier:
677 unregister_memory_notifier(&cmm_mem_nb);
678 unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
679 cmm_unregister_sysfs(&cmm_dev);
680out_reboot_notifier:
681 unregister_reboot_notifier(&cmm_reboot_nb);
682out_oom_notifier:
683 unregister_oom_notifier(&cmm_oom_nb);
684 return rc;
685}
686
687
688
689
690
691
692
693static void cmm_exit(void)
694{
695 if (cmm_thread_ptr)
696 kthread_stop(cmm_thread_ptr);
697 unregister_oom_notifier(&cmm_oom_nb);
698 unregister_reboot_notifier(&cmm_reboot_nb);
699 unregister_memory_notifier(&cmm_mem_nb);
700 unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
701 cmm_free_pages(loaned_pages);
702 cmm_unregister_sysfs(&cmm_dev);
703}
704
705
706
707
708
709
710
711static int cmm_set_disable(const char *val, struct kernel_param *kp)
712{
713 int disable = simple_strtoul(val, NULL, 10);
714
715 if (disable != 0 && disable != 1)
716 return -EINVAL;
717
718 if (disable && !cmm_disabled) {
719 if (cmm_thread_ptr)
720 kthread_stop(cmm_thread_ptr);
721 cmm_thread_ptr = NULL;
722 cmm_free_pages(loaned_pages);
723 } else if (!disable && cmm_disabled) {
724 cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
725 if (IS_ERR(cmm_thread_ptr))
726 return PTR_ERR(cmm_thread_ptr);
727 }
728
729 cmm_disabled = disable;
730 return 0;
731}
732
733module_param_call(disable, cmm_set_disable, param_get_uint,
734 &cmm_disabled, S_IRUGO | S_IWUSR);
735MODULE_PARM_DESC(disable, "Disable CMM. Set to 1 to disable. "
736 "[Default=" __stringify(CMM_DISABLE) "]");
737
738module_init(cmm_init);
739module_exit(cmm_exit);
740