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