1 /******************************************************************************
2 * Xen balloon driver - enables returning/claiming memory to/from Xen.
3 *
4 * Copyright (c) 2003, B Dragovic
5 * Copyright (c) 2003-2004, M Williamson, K Fraser
6 * Copyright (c) 2005 Dan M. Smith, IBM Corporation
7 * Copyright (c) 2010 Daniel Kiper
8 *
9 * Memory hotplug support was written by Daniel Kiper. Work on
10 * it was sponsored by Google under Google Summer of Code 2010
11 * program. Jeremy Fitzhardinge from Citrix was the mentor for
12 * this project.
13 *
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License version 2
16 * as published by the Free Software Foundation; or, when distributed
17 * separately from the Linux kernel or incorporated into other
18 * software packages, subject to the following license:
19 *
20 * Permission is hereby granted, free of charge, to any person obtaining a copy
21 * of this source file (the "Software"), to deal in the Software without
22 * restriction, including without limitation the rights to use, copy, modify,
23 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
24 * and to permit persons to whom the Software is furnished to do so, subject to
25 * the following conditions:
26 *
27 * The above copyright notice and this permission notice shall be included in
28 * all copies or substantial portions of the Software.
29 *
30 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
31 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
32 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
33 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
34 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
35 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
36 * IN THE SOFTWARE.
37 */
38
39 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
40
41 #include <linux/cpu.h>
42 #include <linux/kernel.h>
43 #include <linux/sched.h>
44 #include <linux/cred.h>
45 #include <linux/errno.h>
46 #include <linux/mm.h>
47 #include <linux/bootmem.h>
48 #include <linux/pagemap.h>
49 #include <linux/highmem.h>
50 #include <linux/mutex.h>
51 #include <linux/list.h>
52 #include <linux/gfp.h>
53 #include <linux/notifier.h>
54 #include <linux/memory.h>
55 #include <linux/memory_hotplug.h>
56 #include <linux/percpu-defs.h>
57 #include <linux/slab.h>
58 #include <linux/sysctl.h>
59
60 #include <asm/page.h>
61 #include <asm/pgalloc.h>
62 #include <asm/pgtable.h>
63 #include <asm/tlb.h>
64
65 #include <asm/xen/hypervisor.h>
66 #include <asm/xen/hypercall.h>
67
68 #include <xen/xen.h>
69 #include <xen/interface/xen.h>
70 #include <xen/interface/memory.h>
71 #include <xen/balloon.h>
72 #include <xen/features.h>
73 #include <xen/page.h>
74
75 static int xen_hotplug_unpopulated;
76
77 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
78
79 static int zero;
80 static int one = 1;
81
82 static struct ctl_table balloon_table[] = {
83 {
84 .procname = "hotplug_unpopulated",
85 .data = &xen_hotplug_unpopulated,
86 .maxlen = sizeof(int),
87 .mode = 0644,
88 .proc_handler = proc_dointvec_minmax,
89 .extra1 = &zero,
90 .extra2 = &one,
91 },
92 { }
93 };
94
95 static struct ctl_table balloon_root[] = {
96 {
97 .procname = "balloon",
98 .mode = 0555,
99 .child = balloon_table,
100 },
101 { }
102 };
103
104 static struct ctl_table xen_root[] = {
105 {
106 .procname = "xen",
107 .mode = 0555,
108 .child = balloon_root,
109 },
110 { }
111 };
112
113 #endif
114
115 /*
116 * Use one extent per PAGE_SIZE to avoid to break down the page into
117 * multiple frame.
118 */
119 #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
120
121 /*
122 * balloon_process() state:
123 *
124 * BP_DONE: done or nothing to do,
125 * BP_WAIT: wait to be rescheduled,
126 * BP_EAGAIN: error, go to sleep,
127 * BP_ECANCELED: error, balloon operation canceled.
128 */
129
130 enum bp_state {
131 BP_DONE,
132 BP_WAIT,
133 BP_EAGAIN,
134 BP_ECANCELED
135 };
136
137
138 static DEFINE_MUTEX(balloon_mutex);
139
140 struct balloon_stats balloon_stats;
141 EXPORT_SYMBOL_GPL(balloon_stats);
142
143 /* We increase/decrease in batches which fit in a page */
144 static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
145
146
147 /* List of ballooned pages, threaded through the mem_map array. */
148 static LIST_HEAD(ballooned_pages);
149 static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
150
151 /* Main work function, always executed in process context. */
152 static void balloon_process(struct work_struct *work);
153 static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
154
155 /* When ballooning out (allocating memory to return to Xen) we don't really
156 want the kernel to try too hard since that can trigger the oom killer. */
157 #define GFP_BALLOON \
158 (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
159
scrub_page(struct page * page)160 static void scrub_page(struct page *page)
161 {
162 #ifdef CONFIG_XEN_SCRUB_PAGES
163 clear_highpage(page);
164 #endif
165 }
166
167 /* balloon_append: add the given page to the balloon. */
__balloon_append(struct page * page)168 static void __balloon_append(struct page *page)
169 {
170 /* Lowmem is re-populated first, so highmem pages go at list tail. */
171 if (PageHighMem(page)) {
172 list_add_tail(&page->lru, &ballooned_pages);
173 balloon_stats.balloon_high++;
174 } else {
175 list_add(&page->lru, &ballooned_pages);
176 balloon_stats.balloon_low++;
177 }
178 wake_up(&balloon_wq);
179 }
180
balloon_append(struct page * page)181 static void balloon_append(struct page *page)
182 {
183 __balloon_append(page);
184 }
185
186 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
balloon_retrieve(bool require_lowmem)187 static struct page *balloon_retrieve(bool require_lowmem)
188 {
189 struct page *page;
190
191 if (list_empty(&ballooned_pages))
192 return NULL;
193
194 page = list_entry(ballooned_pages.next, struct page, lru);
195 if (require_lowmem && PageHighMem(page))
196 return NULL;
197 list_del(&page->lru);
198
199 if (PageHighMem(page))
200 balloon_stats.balloon_high--;
201 else
202 balloon_stats.balloon_low--;
203
204 return page;
205 }
206
balloon_next_page(struct page * page)207 static struct page *balloon_next_page(struct page *page)
208 {
209 struct list_head *next = page->lru.next;
210 if (next == &ballooned_pages)
211 return NULL;
212 return list_entry(next, struct page, lru);
213 }
214
update_schedule(enum bp_state state)215 static enum bp_state update_schedule(enum bp_state state)
216 {
217 if (state == BP_WAIT)
218 return BP_WAIT;
219
220 if (state == BP_ECANCELED)
221 return BP_ECANCELED;
222
223 if (state == BP_DONE) {
224 balloon_stats.schedule_delay = 1;
225 balloon_stats.retry_count = 1;
226 return BP_DONE;
227 }
228
229 ++balloon_stats.retry_count;
230
231 if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
232 balloon_stats.retry_count > balloon_stats.max_retry_count) {
233 balloon_stats.schedule_delay = 1;
234 balloon_stats.retry_count = 1;
235 return BP_ECANCELED;
236 }
237
238 balloon_stats.schedule_delay <<= 1;
239
240 if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
241 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
242
243 return BP_EAGAIN;
244 }
245
246 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
release_memory_resource(struct resource * resource)247 static void release_memory_resource(struct resource *resource)
248 {
249 if (!resource)
250 return;
251
252 /*
253 * No need to reset region to identity mapped since we now
254 * know that no I/O can be in this region
255 */
256 release_resource(resource);
257 kfree(resource);
258 }
259
additional_memory_resource(phys_addr_t size)260 static struct resource *additional_memory_resource(phys_addr_t size)
261 {
262 struct resource *res;
263 int ret;
264
265 res = kzalloc(sizeof(*res), GFP_KERNEL);
266 if (!res)
267 return NULL;
268
269 res->name = "System RAM";
270 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
271
272 ret = allocate_resource(&iomem_resource, res,
273 size, 0, -1,
274 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
275 if (ret < 0) {
276 pr_err("Cannot allocate new System RAM resource\n");
277 kfree(res);
278 return NULL;
279 }
280
281 #ifdef CONFIG_SPARSEMEM
282 {
283 unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
284 unsigned long pfn = res->start >> PAGE_SHIFT;
285
286 if (pfn > limit) {
287 pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
288 pfn, limit);
289 release_memory_resource(res);
290 return NULL;
291 }
292 }
293 #endif
294
295 return res;
296 }
297
reserve_additional_memory(void)298 static enum bp_state reserve_additional_memory(void)
299 {
300 long credit;
301 struct resource *resource;
302 int nid, rc;
303 unsigned long balloon_hotplug;
304
305 credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
306 - balloon_stats.total_pages;
307
308 /*
309 * Already hotplugged enough pages? Wait for them to be
310 * onlined.
311 */
312 if (credit <= 0)
313 return BP_WAIT;
314
315 balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
316
317 resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
318 if (!resource)
319 goto err;
320
321 nid = memory_add_physaddr_to_nid(resource->start);
322
323 #ifdef CONFIG_XEN_HAVE_PVMMU
324 /*
325 * We don't support PV MMU when Linux and Xen is using
326 * different page granularity.
327 */
328 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
329
330 /*
331 * add_memory() will build page tables for the new memory so
332 * the p2m must contain invalid entries so the correct
333 * non-present PTEs will be written.
334 *
335 * If a failure occurs, the original (identity) p2m entries
336 * are not restored since this region is now known not to
337 * conflict with any devices.
338 */
339 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
340 unsigned long pfn, i;
341
342 pfn = PFN_DOWN(resource->start);
343 for (i = 0; i < balloon_hotplug; i++) {
344 if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
345 pr_warn("set_phys_to_machine() failed, no memory added\n");
346 goto err;
347 }
348 }
349 }
350 #endif
351
352 /*
353 * add_memory_resource() will call online_pages() which in its turn
354 * will call xen_online_page() callback causing deadlock if we don't
355 * release balloon_mutex here. Unlocking here is safe because the
356 * callers drop the mutex before trying again.
357 */
358 mutex_unlock(&balloon_mutex);
359 /* add_memory_resource() requires the device_hotplug lock */
360 lock_device_hotplug();
361 rc = add_memory_resource(nid, resource, memhp_auto_online);
362 unlock_device_hotplug();
363 mutex_lock(&balloon_mutex);
364
365 if (rc) {
366 pr_warn("Cannot add additional memory (%i)\n", rc);
367 goto err;
368 }
369
370 balloon_stats.total_pages += balloon_hotplug;
371
372 return BP_WAIT;
373 err:
374 release_memory_resource(resource);
375 return BP_ECANCELED;
376 }
377
xen_online_page(struct page * page)378 static void xen_online_page(struct page *page)
379 {
380 __online_page_set_limits(page);
381
382 mutex_lock(&balloon_mutex);
383
384 __balloon_append(page);
385
386 mutex_unlock(&balloon_mutex);
387 }
388
xen_memory_notifier(struct notifier_block * nb,unsigned long val,void * v)389 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
390 {
391 if (val == MEM_ONLINE)
392 schedule_delayed_work(&balloon_worker, 0);
393
394 return NOTIFY_OK;
395 }
396
397 static struct notifier_block xen_memory_nb = {
398 .notifier_call = xen_memory_notifier,
399 .priority = 0
400 };
401 #else
reserve_additional_memory(void)402 static enum bp_state reserve_additional_memory(void)
403 {
404 balloon_stats.target_pages = balloon_stats.current_pages +
405 balloon_stats.target_unpopulated;
406 return BP_ECANCELED;
407 }
408 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
409
current_credit(void)410 static long current_credit(void)
411 {
412 return balloon_stats.target_pages - balloon_stats.current_pages;
413 }
414
balloon_is_inflated(void)415 static bool balloon_is_inflated(void)
416 {
417 return balloon_stats.balloon_low || balloon_stats.balloon_high;
418 }
419
increase_reservation(unsigned long nr_pages)420 static enum bp_state increase_reservation(unsigned long nr_pages)
421 {
422 int rc;
423 unsigned long i;
424 struct page *page;
425 struct xen_memory_reservation reservation = {
426 .address_bits = 0,
427 .extent_order = EXTENT_ORDER,
428 .domid = DOMID_SELF
429 };
430
431 if (nr_pages > ARRAY_SIZE(frame_list))
432 nr_pages = ARRAY_SIZE(frame_list);
433
434 page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
435 for (i = 0; i < nr_pages; i++) {
436 if (!page) {
437 nr_pages = i;
438 break;
439 }
440
441 /* XENMEM_populate_physmap requires a PFN based on Xen
442 * granularity.
443 */
444 frame_list[i] = page_to_xen_pfn(page);
445 page = balloon_next_page(page);
446 }
447
448 set_xen_guest_handle(reservation.extent_start, frame_list);
449 reservation.nr_extents = nr_pages;
450 rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
451 if (rc <= 0)
452 return BP_EAGAIN;
453
454 for (i = 0; i < rc; i++) {
455 page = balloon_retrieve(false);
456 BUG_ON(page == NULL);
457
458 #ifdef CONFIG_XEN_HAVE_PVMMU
459 /*
460 * We don't support PV MMU when Linux and Xen is using
461 * different page granularity.
462 */
463 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
464
465 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
466 unsigned long pfn = page_to_pfn(page);
467
468 set_phys_to_machine(pfn, frame_list[i]);
469
470 /* Link back into the page tables if not highmem. */
471 if (!PageHighMem(page)) {
472 int ret;
473 ret = HYPERVISOR_update_va_mapping(
474 (unsigned long)__va(pfn << PAGE_SHIFT),
475 mfn_pte(frame_list[i], PAGE_KERNEL),
476 0);
477 BUG_ON(ret);
478 }
479 }
480 #endif
481
482 /* Relinquish the page back to the allocator. */
483 free_reserved_page(page);
484 }
485
486 balloon_stats.current_pages += rc;
487
488 return BP_DONE;
489 }
490
decrease_reservation(unsigned long nr_pages,gfp_t gfp)491 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
492 {
493 enum bp_state state = BP_DONE;
494 unsigned long i;
495 struct page *page, *tmp;
496 int ret;
497 struct xen_memory_reservation reservation = {
498 .address_bits = 0,
499 .extent_order = EXTENT_ORDER,
500 .domid = DOMID_SELF
501 };
502 LIST_HEAD(pages);
503
504 if (nr_pages > ARRAY_SIZE(frame_list))
505 nr_pages = ARRAY_SIZE(frame_list);
506
507 for (i = 0; i < nr_pages; i++) {
508 page = alloc_page(gfp);
509 if (page == NULL) {
510 nr_pages = i;
511 state = BP_EAGAIN;
512 break;
513 }
514 adjust_managed_page_count(page, -1);
515 scrub_page(page);
516 list_add(&page->lru, &pages);
517 }
518
519 /*
520 * Ensure that ballooned highmem pages don't have kmaps.
521 *
522 * Do this before changing the p2m as kmap_flush_unused()
523 * reads PTEs to obtain pages (and hence needs the original
524 * p2m entry).
525 */
526 kmap_flush_unused();
527
528 /*
529 * Setup the frame, update direct mapping, invalidate P2M,
530 * and add to balloon.
531 */
532 i = 0;
533 list_for_each_entry_safe(page, tmp, &pages, lru) {
534 /* XENMEM_decrease_reservation requires a GFN */
535 frame_list[i++] = xen_page_to_gfn(page);
536
537 #ifdef CONFIG_XEN_HAVE_PVMMU
538 /*
539 * We don't support PV MMU when Linux and Xen is using
540 * different page granularity.
541 */
542 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
543
544 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
545 unsigned long pfn = page_to_pfn(page);
546
547 if (!PageHighMem(page)) {
548 ret = HYPERVISOR_update_va_mapping(
549 (unsigned long)__va(pfn << PAGE_SHIFT),
550 __pte_ma(0), 0);
551 BUG_ON(ret);
552 }
553 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
554 }
555 #endif
556 list_del(&page->lru);
557
558 balloon_append(page);
559 }
560
561 flush_tlb_all();
562
563 set_xen_guest_handle(reservation.extent_start, frame_list);
564 reservation.nr_extents = nr_pages;
565 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
566 BUG_ON(ret != nr_pages);
567
568 balloon_stats.current_pages -= nr_pages;
569
570 return state;
571 }
572
573 /*
574 * As this is a work item it is guaranteed to run as a single instance only.
575 * We may of course race updates of the target counts (which are protected
576 * by the balloon lock), or with changes to the Xen hard limit, but we will
577 * recover from these in time.
578 */
balloon_process(struct work_struct * work)579 static void balloon_process(struct work_struct *work)
580 {
581 enum bp_state state = BP_DONE;
582 long credit;
583
584
585 do {
586 mutex_lock(&balloon_mutex);
587
588 credit = current_credit();
589
590 if (credit > 0) {
591 if (balloon_is_inflated())
592 state = increase_reservation(credit);
593 else
594 state = reserve_additional_memory();
595 }
596
597 if (credit < 0) {
598 long n_pages;
599
600 n_pages = min(-credit, si_mem_available());
601 state = decrease_reservation(n_pages, GFP_BALLOON);
602 if (state == BP_DONE && n_pages != -credit &&
603 n_pages < totalreserve_pages)
604 state = BP_EAGAIN;
605 }
606
607 state = update_schedule(state);
608
609 mutex_unlock(&balloon_mutex);
610
611 cond_resched();
612
613 } while (credit && state == BP_DONE);
614
615 /* Schedule more work if there is some still to be done. */
616 if (state == BP_EAGAIN)
617 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
618 }
619
620 /* Resets the Xen limit, sets new target, and kicks off processing. */
balloon_set_new_target(unsigned long target)621 void balloon_set_new_target(unsigned long target)
622 {
623 /* No need for lock. Not read-modify-write updates. */
624 balloon_stats.target_pages = target;
625 schedule_delayed_work(&balloon_worker, 0);
626 }
627 EXPORT_SYMBOL_GPL(balloon_set_new_target);
628
add_ballooned_pages(int nr_pages)629 static int add_ballooned_pages(int nr_pages)
630 {
631 enum bp_state st;
632
633 if (xen_hotplug_unpopulated) {
634 st = reserve_additional_memory();
635 if (st != BP_ECANCELED) {
636 mutex_unlock(&balloon_mutex);
637 wait_event(balloon_wq,
638 !list_empty(&ballooned_pages));
639 mutex_lock(&balloon_mutex);
640 return 0;
641 }
642 }
643
644 if (si_mem_available() < nr_pages)
645 return -ENOMEM;
646
647 st = decrease_reservation(nr_pages, GFP_USER);
648 if (st != BP_DONE)
649 return -ENOMEM;
650
651 return 0;
652 }
653
654 /**
655 * alloc_xenballooned_pages - get pages that have been ballooned out
656 * @nr_pages: Number of pages to get
657 * @pages: pages returned
658 * @return 0 on success, error otherwise
659 */
alloc_xenballooned_pages(int nr_pages,struct page ** pages)660 int alloc_xenballooned_pages(int nr_pages, struct page **pages)
661 {
662 int pgno = 0;
663 struct page *page;
664 int ret;
665
666 mutex_lock(&balloon_mutex);
667
668 balloon_stats.target_unpopulated += nr_pages;
669
670 while (pgno < nr_pages) {
671 page = balloon_retrieve(true);
672 if (page) {
673 pages[pgno++] = page;
674 #ifdef CONFIG_XEN_HAVE_PVMMU
675 /*
676 * We don't support PV MMU when Linux and Xen is using
677 * different page granularity.
678 */
679 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
680
681 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
682 ret = xen_alloc_p2m_entry(page_to_pfn(page));
683 if (ret < 0)
684 goto out_undo;
685 }
686 #endif
687 } else {
688 ret = add_ballooned_pages(nr_pages - pgno);
689 if (ret < 0)
690 goto out_undo;
691 }
692 }
693 mutex_unlock(&balloon_mutex);
694 return 0;
695 out_undo:
696 mutex_unlock(&balloon_mutex);
697 free_xenballooned_pages(pgno, pages);
698 return ret;
699 }
700 EXPORT_SYMBOL(alloc_xenballooned_pages);
701
702 /**
703 * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
704 * @nr_pages: Number of pages
705 * @pages: pages to return
706 */
free_xenballooned_pages(int nr_pages,struct page ** pages)707 void free_xenballooned_pages(int nr_pages, struct page **pages)
708 {
709 int i;
710
711 mutex_lock(&balloon_mutex);
712
713 for (i = 0; i < nr_pages; i++) {
714 if (pages[i])
715 balloon_append(pages[i]);
716 }
717
718 balloon_stats.target_unpopulated -= nr_pages;
719
720 /* The balloon may be too large now. Shrink it if needed. */
721 if (current_credit())
722 schedule_delayed_work(&balloon_worker, 0);
723
724 mutex_unlock(&balloon_mutex);
725 }
726 EXPORT_SYMBOL(free_xenballooned_pages);
727
728 #ifdef CONFIG_XEN_PV
balloon_add_region(unsigned long start_pfn,unsigned long pages)729 static void __init balloon_add_region(unsigned long start_pfn,
730 unsigned long pages)
731 {
732 unsigned long pfn, extra_pfn_end;
733 struct page *page;
734
735 /*
736 * If the amount of usable memory has been limited (e.g., with
737 * the 'mem' command line parameter), don't add pages beyond
738 * this limit.
739 */
740 extra_pfn_end = min(max_pfn, start_pfn + pages);
741
742 for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
743 page = pfn_to_page(pfn);
744 /* totalram_pages and totalhigh_pages do not
745 include the boot-time balloon extension, so
746 don't subtract from it. */
747 __balloon_append(page);
748 }
749
750 balloon_stats.total_pages += extra_pfn_end - start_pfn;
751 }
752 #endif
753
balloon_init(void)754 static int __init balloon_init(void)
755 {
756 if (!xen_domain())
757 return -ENODEV;
758
759 pr_info("Initialising balloon driver\n");
760
761 #ifdef CONFIG_XEN_PV
762 balloon_stats.current_pages = xen_pv_domain()
763 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
764 : get_num_physpages();
765 #else
766 balloon_stats.current_pages = get_num_physpages();
767 #endif
768 balloon_stats.target_pages = balloon_stats.current_pages;
769 balloon_stats.balloon_low = 0;
770 balloon_stats.balloon_high = 0;
771 balloon_stats.total_pages = balloon_stats.current_pages;
772
773 balloon_stats.schedule_delay = 1;
774 balloon_stats.max_schedule_delay = 32;
775 balloon_stats.retry_count = 1;
776 balloon_stats.max_retry_count = 4;
777
778 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
779 set_online_page_callback(&xen_online_page);
780 register_memory_notifier(&xen_memory_nb);
781 register_sysctl_table(xen_root);
782 #endif
783
784 #ifdef CONFIG_XEN_PV
785 {
786 int i;
787
788 /*
789 * Initialize the balloon with pages from the extra memory
790 * regions (see arch/x86/xen/setup.c).
791 */
792 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
793 if (xen_extra_mem[i].n_pfns)
794 balloon_add_region(xen_extra_mem[i].start_pfn,
795 xen_extra_mem[i].n_pfns);
796 }
797 #endif
798
799 /* Init the xen-balloon driver. */
800 xen_balloon_init();
801
802 return 0;
803 }
804 subsys_initcall(balloon_init);
805