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 <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
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 device cmm_dev;
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, 0644);
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, 0644);
79MODULE_PARM_DESC(hotplug_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, 0644);
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, 0644);
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, 0644);
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
114static 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
130static 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
148
149
150
151
152
153static 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
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
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
222
223
224
225
226
227static 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
259
260
261
262
263
264
265
266static 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
283
284
285
286
287
288
289static 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
328static struct notifier_block cmm_oom_nb = {
329 .notifier_call = cmm_oom_notify
330};
331
332
333
334
335
336
337
338
339static 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, 0444, show_##name, NULL)
389
390CMM_SHOW(loaned_kb, "%lu\n", PAGES2KB(loaned_pages));
391CMM_SHOW(loaned_target_kb, "%lu\n", PAGES2KB(loaned_pages_target));
392
393static 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
399static 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
414static DEVICE_ATTR(oom_freed_kb, 0644,
415 show_oom_pages, store_oom_pages);
416
417static struct device_attribute *cmm_attrs[] = {
418 &dev_attr_loaned_kb,
419 &dev_attr_loaned_target_kb,
420 &dev_attr_oom_freed_kb,
421};
422
423static struct bus_type cmm_subsys = {
424 .name = "cmm",
425 .dev_name = "cmm",
426};
427
428
429
430
431
432
433
434static 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
454fail:
455 while (--i >= 0)
456 device_remove_file(dev, cmm_attrs[i]);
457 device_unregister(dev);
458subsys_unregister:
459 bus_unregister(&cmm_subsys);
460 return rc;
461}
462
463
464
465
466
467static 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
479
480
481static 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
493static struct notifier_block cmm_reboot_nb = {
494 .notifier_call = cmm_reboot_notifier,
495};
496
497
498
499
500
501
502
503
504
505static 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
529
530
531
532
533
534
535
536static 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
547static struct notifier_block cmm_mem_isolate_nb = {
548 .notifier_call = cmm_memory_isolate_cb,
549 .priority = CMM_MEM_ISOLATE_PRI
550};
551
552
553
554
555
556
557
558
559static 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
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
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
635
636
637
638
639
640
641
642
643static 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
668static struct notifier_block cmm_mem_nb = {
669 .notifier_call = cmm_memory_cb,
670 .priority = CMM_MEM_HOTPLUG_PRI
671};
672
673
674
675
676
677
678
679static 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
710out_unregister_notifier:
711 unregister_memory_notifier(&cmm_mem_nb);
712 unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
713 cmm_unregister_sysfs(&cmm_dev);
714out_reboot_notifier:
715 unregister_reboot_notifier(&cmm_reboot_nb);
716out_oom_notifier:
717 unregister_oom_notifier(&cmm_oom_nb);
718 return rc;
719}
720
721
722
723
724
725
726
727static 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
741
742
743
744
745static 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
767module_param_call(disable, cmm_set_disable, param_get_uint,
768 &cmm_disabled, 0644);
769MODULE_PARM_DESC(disable, "Disable CMM. Set to 1 to disable. "
770 "[Default=" __stringify(CMM_DISABLE) "]");
771
772module_init(cmm_init);
773module_exit(cmm_exit);
774