Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
cmm.c
Go to the documentation of this file.
1 /*
2  * Collaborative memory management interface.
3  *
4  * Copyright (C) 2008 IBM Corporation
5  * Author(s): Brian King ([email protected]),
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/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/device.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  * 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 <[email protected]>");
71 MODULE_DESCRIPTION("IBM System p Collaborative Memory Manager");
72 MODULE_LICENSE("GPL");
74 
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(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) "]");
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 
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 
121 static 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  /* Exit if a hotplug operation is in progress or occurred */
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 
143  if (!addr)
144  break;
145  spin_lock(&cmm_lock);
146  pa = cmm_page_list;
147  if (!pa || pa->index >= CMM_NR_PAGES) {
148  /* Need a new page for the page list. */
149  spin_unlock(&cmm_lock);
150  npa = (struct cmm_page_array *)__get_free_page(
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 
195 static 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 
234 static 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 
257 static 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 
296 static struct notifier_block cmm_oom_nb = {
297  .notifier_call = cmm_oom_notify
298 };
299 
307 static 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 device *dev, \
351  struct device_attribute *attr, \
352  char *buf) \
353  { \
354  return sprintf(buf, format, ##args); \
355  } \
356  static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
357 
358 CMM_SHOW(loaned_kb, "%lu\n", PAGES2KB(loaned_pages));
359 CMM_SHOW(loaned_target_kb, "%lu\n", PAGES2KB(loaned_pages_target));
360 
361 static ssize_t show_oom_pages(struct device *dev,
362  struct device_attribute *attr, char *buf)
363 {
364  return sprintf(buf, "%lu\n", PAGES2KB(oom_freed_pages));
365 }
366 
367 static ssize_t store_oom_pages(struct device *dev,
368  struct device_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 
382 static DEVICE_ATTR(oom_freed_kb, S_IWUSR | S_IRUGO,
383  show_oom_pages, store_oom_pages);
384 
385 static struct device_attribute *cmm_attrs[] = {
386  &dev_attr_loaned_kb,
387  &dev_attr_loaned_target_kb,
388  &dev_attr_oom_freed_kb,
389 };
390 
391 static struct bus_type cmm_subsys = {
392  .name = "cmm",
393  .dev_name = "cmm",
394 };
395 
402 static int cmm_sysfs_register(struct device *dev)
403 {
404  int i, rc;
405 
406  if ((rc = subsys_system_register(&cmm_subsys, NULL)))
407  return rc;
408 
409  dev->id = 0;
410  dev->bus = &cmm_subsys;
411 
412  if ((rc = device_register(dev)))
413  goto subsys_unregister;
414 
415  for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++) {
416  if ((rc = device_create_file(dev, cmm_attrs[i])))
417  goto fail;
418  }
419 
420  return 0;
421 
422 fail:
423  while (--i >= 0)
424  device_remove_file(dev, cmm_attrs[i]);
425  device_unregister(dev);
426 subsys_unregister:
427  bus_unregister(&cmm_subsys);
428  return rc;
429 }
430 
435 static void cmm_unregister_sysfs(struct device *dev)
436 {
437  int i;
438 
439  for (i = 0; i < ARRAY_SIZE(cmm_attrs); i++)
440  device_remove_file(dev, cmm_attrs[i]);
441  device_unregister(dev);
442  bus_unregister(&cmm_subsys);
443 }
444 
449 static int cmm_reboot_notifier(struct notifier_block *nb,
450  unsigned long action, void *unused)
451 {
452  if (action == SYS_RESTART) {
453  if (cmm_thread_ptr)
454  kthread_stop(cmm_thread_ptr);
455  cmm_thread_ptr = NULL;
456  cmm_free_pages(loaned_pages);
457  }
458  return NOTIFY_DONE;
459 }
460 
461 static struct notifier_block cmm_reboot_nb = {
462  .notifier_call = cmm_reboot_notifier,
463 };
464 
473 static unsigned long cmm_count_pages(void *arg)
474 {
475  struct memory_isolate_notify *marg = arg;
476  struct cmm_page_array *pa;
477  unsigned long start = (unsigned long)pfn_to_kaddr(marg->start_pfn);
478  unsigned long end = start + (marg->nr_pages << PAGE_SHIFT);
479  unsigned long idx;
480 
481  spin_lock(&cmm_lock);
482  pa = cmm_page_list;
483  while (pa) {
484  if ((unsigned long)pa >= start && (unsigned long)pa < end)
485  marg->pages_found++;
486  for (idx = 0; idx < pa->index; idx++)
487  if (pa->page[idx] >= start && pa->page[idx] < end)
488  marg->pages_found++;
489  pa = pa->next;
490  }
491  spin_unlock(&cmm_lock);
492  return 0;
493 }
494 
504 static int cmm_memory_isolate_cb(struct notifier_block *self,
505  unsigned long action, void *arg)
506 {
507  int ret = 0;
508 
509  if (action == MEM_ISOLATE_COUNT)
510  ret = cmm_count_pages(arg);
511 
512  return notifier_from_errno(ret);
513 }
514 
515 static struct notifier_block cmm_mem_isolate_nb = {
516  .notifier_call = cmm_memory_isolate_cb,
517  .priority = CMM_MEM_ISOLATE_PRI
518 };
519 
527 static int cmm_mem_going_offline(void *arg)
528 {
529  struct memory_notify *marg = arg;
530  unsigned long start_page = (unsigned long)pfn_to_kaddr(marg->start_pfn);
531  unsigned long end_page = start_page + (marg->nr_pages << PAGE_SHIFT);
532  struct cmm_page_array *pa_curr, *pa_last, *npa;
533  unsigned long idx;
534  unsigned long freed = 0;
535 
536  cmm_dbg("Memory going offline, searching 0x%lx (%ld pages).\n",
537  start_page, marg->nr_pages);
538  spin_lock(&cmm_lock);
539 
540  /* Search the page list for pages in the range to be offlined */
541  pa_last = pa_curr = cmm_page_list;
542  while (pa_curr) {
543  for (idx = (pa_curr->index - 1); (idx + 1) > 0; idx--) {
544  if ((pa_curr->page[idx] < start_page) ||
545  (pa_curr->page[idx] >= end_page))
546  continue;
547 
548  plpar_page_set_active(__pa(pa_curr->page[idx]));
549  free_page(pa_curr->page[idx]);
550  freed++;
551  loaned_pages--;
552  totalram_pages++;
553  pa_curr->page[idx] = pa_last->page[--pa_last->index];
554  if (pa_last->index == 0) {
555  if (pa_curr == pa_last)
556  pa_curr = pa_last->next;
557  pa_last = pa_last->next;
558  free_page((unsigned long)cmm_page_list);
559  cmm_page_list = pa_last;
560  continue;
561  }
562  }
563  pa_curr = pa_curr->next;
564  }
565 
566  /* Search for page list structures in the range to be offlined */
567  pa_last = NULL;
568  pa_curr = cmm_page_list;
569  while (pa_curr) {
570  if (((unsigned long)pa_curr >= start_page) &&
571  ((unsigned long)pa_curr < end_page)) {
572  npa = (struct cmm_page_array *)__get_free_page(
575  if (!npa) {
576  spin_unlock(&cmm_lock);
577  cmm_dbg("Failed to allocate memory for list "
578  "management. Memory hotplug "
579  "failed.\n");
580  return ENOMEM;
581  }
582  memcpy(npa, pa_curr, PAGE_SIZE);
583  if (pa_curr == cmm_page_list)
584  cmm_page_list = npa;
585  if (pa_last)
586  pa_last->next = npa;
587  free_page((unsigned long) pa_curr);
588  freed++;
589  pa_curr = npa;
590  }
591 
592  pa_last = pa_curr;
593  pa_curr = pa_curr->next;
594  }
595 
596  spin_unlock(&cmm_lock);
597  cmm_dbg("Released %ld pages in the search range.\n", freed);
598 
599  return 0;
600 }
601 
612 static int cmm_memory_cb(struct notifier_block *self,
613  unsigned long action, void *arg)
614 {
615  int ret = 0;
616 
617  switch (action) {
618  case MEM_GOING_OFFLINE:
619  mutex_lock(&hotplug_mutex);
620  hotplug_occurred = 1;
621  ret = cmm_mem_going_offline(arg);
622  break;
623  case MEM_OFFLINE:
624  case MEM_CANCEL_OFFLINE:
625  mutex_unlock(&hotplug_mutex);
626  cmm_dbg("Memory offline operation complete.\n");
627  break;
628  case MEM_GOING_ONLINE:
629  case MEM_ONLINE:
630  case MEM_CANCEL_ONLINE:
631  break;
632  }
633 
634  return notifier_from_errno(ret);
635 }
636 
637 static struct notifier_block cmm_mem_nb = {
638  .notifier_call = cmm_memory_cb,
639  .priority = CMM_MEM_HOTPLUG_PRI
640 };
641 
648 static int cmm_init(void)
649 {
650  int rc = -ENOMEM;
651 
652  if (!firmware_has_feature(FW_FEATURE_CMO))
653  return -EOPNOTSUPP;
654 
655  if ((rc = register_oom_notifier(&cmm_oom_nb)) < 0)
656  return rc;
657 
658  if ((rc = register_reboot_notifier(&cmm_reboot_nb)))
659  goto out_oom_notifier;
660 
661  if ((rc = cmm_sysfs_register(&cmm_dev)))
662  goto out_reboot_notifier;
663 
664  if (register_memory_notifier(&cmm_mem_nb) ||
665  register_memory_isolate_notifier(&cmm_mem_isolate_nb))
666  goto out_unregister_notifier;
667 
668  if (cmm_disabled)
669  return rc;
670 
671  cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
672  if (IS_ERR(cmm_thread_ptr)) {
673  rc = PTR_ERR(cmm_thread_ptr);
674  goto out_unregister_notifier;
675  }
676 
677  return rc;
678 
679 out_unregister_notifier:
680  unregister_memory_notifier(&cmm_mem_nb);
681  unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
682  cmm_unregister_sysfs(&cmm_dev);
683 out_reboot_notifier:
684  unregister_reboot_notifier(&cmm_reboot_nb);
685 out_oom_notifier:
686  unregister_oom_notifier(&cmm_oom_nb);
687  return rc;
688 }
689 
696 static void cmm_exit(void)
697 {
698  if (cmm_thread_ptr)
699  kthread_stop(cmm_thread_ptr);
700  unregister_oom_notifier(&cmm_oom_nb);
701  unregister_reboot_notifier(&cmm_reboot_nb);
702  unregister_memory_notifier(&cmm_mem_nb);
703  unregister_memory_isolate_notifier(&cmm_mem_isolate_nb);
704  cmm_free_pages(loaned_pages);
705  cmm_unregister_sysfs(&cmm_dev);
706 }
707 
714 static int cmm_set_disable(const char *val, struct kernel_param *kp)
715 {
716  int disable = simple_strtoul(val, NULL, 10);
717 
718  if (disable != 0 && disable != 1)
719  return -EINVAL;
720 
721  if (disable && !cmm_disabled) {
722  if (cmm_thread_ptr)
723  kthread_stop(cmm_thread_ptr);
724  cmm_thread_ptr = NULL;
725  cmm_free_pages(loaned_pages);
726  } else if (!disable && cmm_disabled) {
727  cmm_thread_ptr = kthread_run(cmm_thread, NULL, "cmmthread");
728  if (IS_ERR(cmm_thread_ptr))
729  return PTR_ERR(cmm_thread_ptr);
730  }
731 
732  cmm_disabled = disable;
733  return 0;
734 }
735 
736 module_param_call(disable, cmm_set_disable, param_get_uint,
737  &cmm_disabled, S_IRUGO | S_IWUSR);
738 MODULE_PARM_DESC(disable, "Disable CMM. Set to 1 to disable. "
739  "[Default=" __stringify(CMM_DISABLE) "]");
740 
741 module_init(cmm_init);
742 module_exit(cmm_exit);