2 * Collaborative memory management interface.
4 * Copyright (C) 2008 IBM Corporation
5 * Author(s): Brian King (brking@linux.vnet.ibm.com),
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.
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.
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
23 #include <linux/ctype.h>
24 #include <linux/delay.h>
25 #include <linux/errno.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>
40 #include <asm/pgalloc.h>
41 #include <asm/uaccess.h>
42 #include <linux/memory.h>
44 #include "plpar_wrappers.h"
46 #define CMM_DRIVER_VERSION "1.0.0"
47 #define CMM_DEFAULT_DELAY 1
48 #define CMM_HOTPLUG_DELAY 5
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))
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.
59 #define CMM_MEM_HOTPLUG_PRI 1
60 #define CMM_MEM_ISOLATE_PRI 15
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 sys_device cmm_sysdev
;
70 MODULE_AUTHOR("Brian King <brking@linux.vnet.ibm.com>");
71 MODULE_DESCRIPTION("IBM System p Collaborative Memory Manager");
72 MODULE_LICENSE("GPL");
73 MODULE_VERSION(CMM_DRIVER_VERSION
);
75 module_param_named(delay
, delay
, uint
, S_IRUGO
| S_IWUSR
);
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
) "]");
88 module_param_named(debug
, cmm_debug
, uint
, S_IRUGO
| S_IWUSR
);
89 MODULE_PARM_DESC(debug
, "Enable module debugging logging. Set to 1 to enable. "
90 "[Default=" __stringify(CMM_DEBUG
) "]");
92 #define CMM_NR_PAGES ((PAGE_SIZE - sizeof(void *) - sizeof(unsigned long)) / sizeof(unsigned long))
94 #define cmm_dbg(...) if (cmm_debug) { printk(KERN_INFO "cmm: "__VA_ARGS__); }
96 struct cmm_page_array
{
97 struct cmm_page_array
*next
;
99 unsigned long page
[CMM_NR_PAGES
];
102 static unsigned long loaned_pages
;
103 static unsigned long loaned_pages_target
;
104 static unsigned long oom_freed_pages
;
106 static struct cmm_page_array
*cmm_page_list
;
107 static DEFINE_SPINLOCK(cmm_lock
);
109 static DEFINE_MUTEX(hotplug_mutex
);
110 static int hotplug_occurred
; /* protected by the hotplug mutex */
112 static struct task_struct
*cmm_thread_ptr
;
115 * cmm_alloc_pages - Allocate pages and mark them as loaned
116 * @nr: number of pages to allocate
119 * number of pages requested to be allocated which were not
121 static long cmm_alloc_pages(long nr
)
123 struct cmm_page_array
*pa
, *npa
;
127 cmm_dbg("Begin request for %ld pages\n", 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
);
136 mutex_unlock(&hotplug_mutex
);
141 addr
= __get_free_page(GFP_NOIO
| __GFP_NOWARN
|
142 __GFP_NORETRY
| __GFP_NOMEMALLOC
);
145 spin_lock(&cmm_lock
);
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(
151 GFP_NOIO
| __GFP_NOWARN
|
152 __GFP_NORETRY
| __GFP_NOMEMALLOC
);
154 pr_info("%s: Can not allocate new page list\n", __func__
);
158 spin_lock(&cmm_lock
);
161 if (!pa
|| pa
->index
>= CMM_NR_PAGES
) {
167 free_page((unsigned long) npa
);
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
);
177 pa
->page
[pa
->index
++] = addr
;
180 spin_unlock(&cmm_lock
);
184 cmm_dbg("End request with %ld pages unfulfilled\n", nr
);
189 * cmm_free_pages - Free pages and mark them as active
190 * @nr: number of pages to free
193 * number of pages requested to be freed which were not
195 static long cmm_free_pages(long nr
)
197 struct cmm_page_array
*pa
;
200 cmm_dbg("Begin free of %ld pages.\n", nr
);
201 spin_lock(&cmm_lock
);
204 if (!pa
|| pa
->index
<= 0)
206 addr
= pa
->page
[--pa
->index
];
208 if (pa
->index
== 0) {
210 free_page((unsigned long) cmm_page_list
);
214 plpar_page_set_active(__pa(addr
));
220 spin_unlock(&cmm_lock
);
221 cmm_dbg("End request with %ld pages unfulfilled\n", nr
);
226 * cmm_oom_notify - OOM notifier
227 * @self: notifier block struct
229 * @parm: returned - number of pages freed
234 static int cmm_oom_notify(struct notifier_block
*self
,
235 unsigned long dummy
, void *parm
)
237 unsigned long *freed
= parm
;
238 long nr
= KB2PAGES(oom_kb
);
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");
250 * cmm_get_mpp - Read memory performance parameters
252 * Makes hcall to query the current page loan request from the hypervisor.
257 static void cmm_get_mpp(void)
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
;
265 rc
= h_get_mpp(&mpp_data
);
270 page_loan_request
= div_s64((s64
)mpp_data
.loan_request
, PAGE_SIZE
);
271 target
= page_loan_request
+ (signed long)loaned_pages
;
273 if (target
< 0 || total_pages
< min_mem_pages
)
276 if (target
> oom_freed_pages
)
277 target
-= oom_freed_pages
;
281 active_pages_target
= total_pages
- target
;
283 if (min_mem_pages
> active_pages_target
)
284 target
= total_pages
- min_mem_pages
;
289 loaned_pages_target
= target
;
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
);
296 static struct notifier_block cmm_oom_nb
= {
297 .notifier_call
= cmm_oom_notify
301 * cmm_thread - CMM task thread
307 static int cmm_thread(void *dummy
)
309 unsigned long timeleft
;
312 timeleft
= msleep_interruptible(delay
* 1000);
314 if (kthread_should_stop() || timeleft
)
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 "
325 timeleft
= msleep_interruptible(hotplug_delay
*
327 if (kthread_should_stop() || timeleft
)
331 mutex_unlock(&hotplug_mutex
);
333 cmm_dbg("Hotplug operation in progress, activity "
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
);
349 #define CMM_SHOW(name, format, args...) \
350 static ssize_t show_##name(struct sys_device *dev, \
351 struct sysdev_attribute *attr, \
354 return sprintf(buf, format, ##args); \
356 static SYSDEV_ATTR(name, S_IRUGO, show_##name, NULL)
358 CMM_SHOW(loaned_kb
, "%lu\n", PAGES2KB(loaned_pages
));
359 CMM_SHOW(loaned_target_kb
, "%lu\n", PAGES2KB(loaned_pages_target
));
361 static ssize_t
show_oom_pages(struct sys_device
*dev
,
362 struct sysdev_attribute
*attr
, char *buf
)
364 return sprintf(buf
, "%lu\n", PAGES2KB(oom_freed_pages
));
367 static ssize_t
store_oom_pages(struct sys_device
*dev
,
368 struct sysdev_attribute
*attr
,
369 const char *buf
, size_t count
)
371 unsigned long val
= simple_strtoul (buf
, NULL
, 10);
373 if (!capable(CAP_SYS_ADMIN
))
382 static SYSDEV_ATTR(oom_freed_kb
, S_IWUSR
| S_IRUGO
,
383 show_oom_pages
, store_oom_pages
);
385 static struct sysdev_attribute
*cmm_attrs
[] = {
387 &attr_loaned_target_kb
,
391 static struct sysdev_class cmm_sysdev_class
= {
396 * cmm_sysfs_register - Register with sysfs
399 * 0 on success / other on failure
401 static int cmm_sysfs_register(struct sys_device
*sysdev
)
405 if ((rc
= sysdev_class_register(&cmm_sysdev_class
)))
409 sysdev
->cls
= &cmm_sysdev_class
;
411 if ((rc
= sysdev_register(sysdev
)))
412 goto class_unregister
;
414 for (i
= 0; i
< ARRAY_SIZE(cmm_attrs
); i
++) {
415 if ((rc
= sysdev_create_file(sysdev
, cmm_attrs
[i
])))
423 sysdev_remove_file(sysdev
, cmm_attrs
[i
]);
424 sysdev_unregister(sysdev
);
426 sysdev_class_unregister(&cmm_sysdev_class
);
431 * cmm_unregister_sysfs - Unregister from sysfs
434 static void cmm_unregister_sysfs(struct sys_device
*sysdev
)
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
);
445 * cmm_reboot_notifier - Make sure pages are not still marked as "loaned"
448 static int cmm_reboot_notifier(struct notifier_block
*nb
,
449 unsigned long action
, void *unused
)
451 if (action
== SYS_RESTART
) {
453 kthread_stop(cmm_thread_ptr
);
454 cmm_thread_ptr
= NULL
;
455 cmm_free_pages(loaned_pages
);
460 static struct notifier_block cmm_reboot_nb
= {
461 .notifier_call
= cmm_reboot_notifier
,
465 * cmm_count_pages - Count the number of pages loaned in a particular range.
467 * @arg: memory_isolate_notify structure with address range and count
472 static unsigned long cmm_count_pages(void *arg
)
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
);
480 spin_lock(&cmm_lock
);
483 if ((unsigned long)pa
>= start
&& (unsigned long)pa
< end
)
485 for (idx
= 0; idx
< pa
->index
; idx
++)
486 if (pa
->page
[idx
] >= start
&& pa
->page
[idx
] < end
)
490 spin_unlock(&cmm_lock
);
495 * cmm_memory_isolate_cb - Handle memory isolation notifier calls
496 * @self: notifier block struct
497 * @action: action to take
498 * @arg: struct memory_isolate_notify data for handler
501 * NOTIFY_OK or notifier error based on subfunction return value
503 static int cmm_memory_isolate_cb(struct notifier_block
*self
,
504 unsigned long action
, void *arg
)
508 if (action
== MEM_ISOLATE_COUNT
)
509 ret
= cmm_count_pages(arg
);
512 ret
= notifier_from_errno(ret
);
519 static struct notifier_block cmm_mem_isolate_nb
= {
520 .notifier_call
= cmm_memory_isolate_cb
,
521 .priority
= CMM_MEM_ISOLATE_PRI
525 * cmm_mem_going_offline - Unloan pages where memory is to be removed
526 * @arg: memory_notify structure with page range to be offlined
531 static int cmm_mem_going_offline(void *arg
)
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
;
538 unsigned long freed
= 0;
540 cmm_dbg("Memory going offline, searching 0x%lx (%ld pages).\n",
541 start_page
, marg
->nr_pages
);
542 spin_lock(&cmm_lock
);
544 /* Search the page list for pages in the range to be offlined */
545 pa_last
= pa_curr
= cmm_page_list
;
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
))
552 plpar_page_set_active(__pa(pa_curr
->page
[idx
]));
553 free_page(pa_curr
->page
[idx
]);
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
;
567 pa_curr
= pa_curr
->next
;
570 /* Search for page list structures in the range to be offlined */
572 pa_curr
= cmm_page_list
;
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
);
580 spin_unlock(&cmm_lock
);
581 cmm_dbg("Failed to allocate memory for list "
582 "management. Memory hotplug "
586 memcpy(npa
, pa_curr
, PAGE_SIZE
);
587 if (pa_curr
== cmm_page_list
)
591 free_page((unsigned long) pa_curr
);
597 pa_curr
= pa_curr
->next
;
600 spin_unlock(&cmm_lock
);
601 cmm_dbg("Released %ld pages in the search range.\n", freed
);
607 * cmm_memory_cb - Handle memory hotplug notifier calls
608 * @self: notifier block struct
609 * @action: action to take
610 * @arg: struct memory_notify data for handler
613 * NOTIFY_OK or notifier error based on subfunction return value
616 static int cmm_memory_cb(struct notifier_block
*self
,
617 unsigned long action
, void *arg
)
622 case MEM_GOING_OFFLINE
:
623 mutex_lock(&hotplug_mutex
);
624 hotplug_occurred
= 1;
625 ret
= cmm_mem_going_offline(arg
);
628 case MEM_CANCEL_OFFLINE
:
629 mutex_unlock(&hotplug_mutex
);
630 cmm_dbg("Memory offline operation complete.\n");
632 case MEM_GOING_ONLINE
:
634 case MEM_CANCEL_ONLINE
:
639 ret
= notifier_from_errno(ret
);
646 static struct notifier_block cmm_mem_nb
= {
647 .notifier_call
= cmm_memory_cb
,
648 .priority
= CMM_MEM_HOTPLUG_PRI
652 * cmm_init - Module initialization
655 * 0 on success / other on failure
657 static int cmm_init(void)
661 if (!firmware_has_feature(FW_FEATURE_CMO
))
664 if ((rc
= register_oom_notifier(&cmm_oom_nb
)) < 0)
667 if ((rc
= register_reboot_notifier(&cmm_reboot_nb
)))
668 goto out_oom_notifier
;
670 if ((rc
= cmm_sysfs_register(&cmm_sysdev
)))
671 goto out_reboot_notifier
;
673 if (register_memory_notifier(&cmm_mem_nb
) ||
674 register_memory_isolate_notifier(&cmm_mem_isolate_nb
))
675 goto out_unregister_notifier
;
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
;
688 out_unregister_notifier
:
689 unregister_memory_notifier(&cmm_mem_nb
);
690 unregister_memory_isolate_notifier(&cmm_mem_isolate_nb
);
691 cmm_unregister_sysfs(&cmm_sysdev
);
693 unregister_reboot_notifier(&cmm_reboot_nb
);
695 unregister_oom_notifier(&cmm_oom_nb
);
700 * cmm_exit - Module exit
705 static void cmm_exit(void)
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
);
718 * cmm_set_disable - Disable/Enable CMM
721 * 0 on success / other on failure
723 static int cmm_set_disable(const char *val
, struct kernel_param
*kp
)
725 int disable
= simple_strtoul(val
, NULL
, 10);
727 if (disable
!= 0 && disable
!= 1)
730 if (disable
&& !cmm_disabled
) {
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
);
741 cmm_disabled
= disable
;
745 module_param_call(disable
, cmm_set_disable
, param_get_uint
,
746 &cmm_disabled
, S_IRUGO
| S_IWUSR
);
747 MODULE_PARM_DESC(disable
, "Disable CMM. Set to 1 to disable. "
748 "[Default=" __stringify(CMM_DISABLE
) "]");
750 module_init(cmm_init
);
751 module_exit(cmm_exit
);