xref: /linux/drivers/xen/balloon.c (revision ba0ad6ed89fd5dada3b7b65ef2b08e95d449d4ab)
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/freezer.h>
47 #include <linux/kthread.h>
48 #include <linux/mm.h>
49 #include <linux/memblock.h>
50 #include <linux/pagemap.h>
51 #include <linux/highmem.h>
52 #include <linux/mutex.h>
53 #include <linux/list.h>
54 #include <linux/gfp.h>
55 #include <linux/notifier.h>
56 #include <linux/memory.h>
57 #include <linux/memory_hotplug.h>
58 #include <linux/percpu-defs.h>
59 #include <linux/slab.h>
60 #include <linux/sysctl.h>
61 #include <linux/moduleparam.h>
62 #include <linux/jiffies.h>
63 
64 #include <asm/page.h>
65 #include <asm/tlb.h>
66 
67 #include <asm/xen/hypervisor.h>
68 #include <asm/xen/hypercall.h>
69 
70 #include <xen/xen.h>
71 #include <xen/interface/xen.h>
72 #include <xen/interface/memory.h>
73 #include <xen/balloon.h>
74 #include <xen/features.h>
75 #include <xen/page.h>
76 #include <xen/mem-reservation.h>
77 
78 #undef MODULE_PARAM_PREFIX
79 #define MODULE_PARAM_PREFIX "xen."
80 
81 static uint __read_mostly balloon_boot_timeout = 180;
82 module_param(balloon_boot_timeout, uint, 0444);
83 
84 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
85 static int xen_hotplug_unpopulated;
86 
87 static struct ctl_table balloon_table[] = {
88 	{
89 		.procname	= "hotplug_unpopulated",
90 		.data		= &xen_hotplug_unpopulated,
91 		.maxlen		= sizeof(int),
92 		.mode		= 0644,
93 		.proc_handler	= proc_dointvec_minmax,
94 		.extra1         = SYSCTL_ZERO,
95 		.extra2         = SYSCTL_ONE,
96 	},
97 	{ }
98 };
99 
100 #else
101 #define xen_hotplug_unpopulated 0
102 #endif
103 
104 /*
105  * Use one extent per PAGE_SIZE to avoid to break down the page into
106  * multiple frame.
107  */
108 #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
109 
110 /*
111  * balloon_thread() state:
112  *
113  * BP_DONE: done or nothing to do,
114  * BP_WAIT: wait to be rescheduled,
115  * BP_EAGAIN: error, go to sleep,
116  * BP_ECANCELED: error, balloon operation canceled.
117  */
118 
119 static enum bp_state {
120 	BP_DONE,
121 	BP_WAIT,
122 	BP_EAGAIN,
123 	BP_ECANCELED
124 } balloon_state = BP_DONE;
125 
126 /* Main waiting point for xen-balloon thread. */
127 static DECLARE_WAIT_QUEUE_HEAD(balloon_thread_wq);
128 
129 static DEFINE_MUTEX(balloon_mutex);
130 
131 struct balloon_stats balloon_stats;
132 EXPORT_SYMBOL_GPL(balloon_stats);
133 
134 /* We increase/decrease in batches which fit in a page */
135 static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
136 
137 
138 /* List of ballooned pages, threaded through the mem_map array. */
139 static LIST_HEAD(ballooned_pages);
140 static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
141 
142 /* When ballooning out (allocating memory to return to Xen) we don't really
143    want the kernel to try too hard since that can trigger the oom killer. */
144 #define GFP_BALLOON \
145 	(GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
146 
147 /* balloon_append: add the given page to the balloon. */
148 static void balloon_append(struct page *page)
149 {
150 	__SetPageOffline(page);
151 
152 	/* Lowmem is re-populated first, so highmem pages go at list tail. */
153 	if (PageHighMem(page)) {
154 		list_add_tail(&page->lru, &ballooned_pages);
155 		balloon_stats.balloon_high++;
156 	} else {
157 		list_add(&page->lru, &ballooned_pages);
158 		balloon_stats.balloon_low++;
159 	}
160 	wake_up(&balloon_wq);
161 }
162 
163 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
164 static struct page *balloon_retrieve(bool require_lowmem)
165 {
166 	struct page *page;
167 
168 	if (list_empty(&ballooned_pages))
169 		return NULL;
170 
171 	page = list_entry(ballooned_pages.next, struct page, lru);
172 	if (require_lowmem && PageHighMem(page))
173 		return NULL;
174 	list_del(&page->lru);
175 
176 	if (PageHighMem(page))
177 		balloon_stats.balloon_high--;
178 	else
179 		balloon_stats.balloon_low--;
180 
181 	__ClearPageOffline(page);
182 	return page;
183 }
184 
185 static struct page *balloon_next_page(struct page *page)
186 {
187 	struct list_head *next = page->lru.next;
188 	if (next == &ballooned_pages)
189 		return NULL;
190 	return list_entry(next, struct page, lru);
191 }
192 
193 static void update_schedule(void)
194 {
195 	if (balloon_state == BP_WAIT || balloon_state == BP_ECANCELED)
196 		return;
197 
198 	if (balloon_state == BP_DONE) {
199 		balloon_stats.schedule_delay = 1;
200 		balloon_stats.retry_count = 1;
201 		return;
202 	}
203 
204 	++balloon_stats.retry_count;
205 
206 	if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
207 			balloon_stats.retry_count > balloon_stats.max_retry_count) {
208 		balloon_stats.schedule_delay = 1;
209 		balloon_stats.retry_count = 1;
210 		balloon_state = BP_ECANCELED;
211 		return;
212 	}
213 
214 	balloon_stats.schedule_delay <<= 1;
215 
216 	if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
217 		balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
218 
219 	balloon_state = BP_EAGAIN;
220 }
221 
222 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
223 static void release_memory_resource(struct resource *resource)
224 {
225 	if (!resource)
226 		return;
227 
228 	/*
229 	 * No need to reset region to identity mapped since we now
230 	 * know that no I/O can be in this region
231 	 */
232 	release_resource(resource);
233 	kfree(resource);
234 }
235 
236 static struct resource *additional_memory_resource(phys_addr_t size)
237 {
238 	struct resource *res;
239 	int ret;
240 
241 	res = kzalloc(sizeof(*res), GFP_KERNEL);
242 	if (!res)
243 		return NULL;
244 
245 	res->name = "System RAM";
246 	res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
247 
248 	ret = allocate_resource(&iomem_resource, res,
249 				size, 0, -1,
250 				PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
251 	if (ret < 0) {
252 		pr_err("Cannot allocate new System RAM resource\n");
253 		kfree(res);
254 		return NULL;
255 	}
256 
257 	return res;
258 }
259 
260 static enum bp_state reserve_additional_memory(void)
261 {
262 	long credit;
263 	struct resource *resource;
264 	int nid, rc;
265 	unsigned long balloon_hotplug;
266 
267 	credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
268 		- balloon_stats.total_pages;
269 
270 	/*
271 	 * Already hotplugged enough pages?  Wait for them to be
272 	 * onlined.
273 	 */
274 	if (credit <= 0)
275 		return BP_WAIT;
276 
277 	balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
278 
279 	resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
280 	if (!resource)
281 		goto err;
282 
283 	nid = memory_add_physaddr_to_nid(resource->start);
284 
285 #ifdef CONFIG_XEN_HAVE_PVMMU
286 	/*
287 	 * We don't support PV MMU when Linux and Xen is using
288 	 * different page granularity.
289 	 */
290 	BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
291 
292         /*
293          * add_memory() will build page tables for the new memory so
294          * the p2m must contain invalid entries so the correct
295          * non-present PTEs will be written.
296          *
297          * If a failure occurs, the original (identity) p2m entries
298          * are not restored since this region is now known not to
299          * conflict with any devices.
300          */
301 	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
302 		unsigned long pfn, i;
303 
304 		pfn = PFN_DOWN(resource->start);
305 		for (i = 0; i < balloon_hotplug; i++) {
306 			if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
307 				pr_warn("set_phys_to_machine() failed, no memory added\n");
308 				goto err;
309 			}
310                 }
311 	}
312 #endif
313 
314 	/*
315 	 * add_memory_resource() will call online_pages() which in its turn
316 	 * will call xen_online_page() callback causing deadlock if we don't
317 	 * release balloon_mutex here. Unlocking here is safe because the
318 	 * callers drop the mutex before trying again.
319 	 */
320 	mutex_unlock(&balloon_mutex);
321 	/* add_memory_resource() requires the device_hotplug lock */
322 	lock_device_hotplug();
323 	rc = add_memory_resource(nid, resource, MHP_MERGE_RESOURCE);
324 	unlock_device_hotplug();
325 	mutex_lock(&balloon_mutex);
326 
327 	if (rc) {
328 		pr_warn("Cannot add additional memory (%i)\n", rc);
329 		goto err;
330 	}
331 
332 	balloon_stats.total_pages += balloon_hotplug;
333 
334 	return BP_WAIT;
335   err:
336 	release_memory_resource(resource);
337 	return BP_ECANCELED;
338 }
339 
340 static void xen_online_page(struct page *page, unsigned int order)
341 {
342 	unsigned long i, size = (1 << order);
343 	unsigned long start_pfn = page_to_pfn(page);
344 	struct page *p;
345 
346 	pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn);
347 	mutex_lock(&balloon_mutex);
348 	for (i = 0; i < size; i++) {
349 		p = pfn_to_page(start_pfn + i);
350 		balloon_append(p);
351 	}
352 	mutex_unlock(&balloon_mutex);
353 }
354 
355 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
356 {
357 	if (val == MEM_ONLINE)
358 		wake_up(&balloon_thread_wq);
359 
360 	return NOTIFY_OK;
361 }
362 
363 static struct notifier_block xen_memory_nb = {
364 	.notifier_call = xen_memory_notifier,
365 	.priority = 0
366 };
367 #else
368 static enum bp_state reserve_additional_memory(void)
369 {
370 	balloon_stats.target_pages = balloon_stats.current_pages +
371 				     balloon_stats.target_unpopulated;
372 	return BP_ECANCELED;
373 }
374 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
375 
376 static long current_credit(void)
377 {
378 	return balloon_stats.target_pages - balloon_stats.current_pages;
379 }
380 
381 static bool balloon_is_inflated(void)
382 {
383 	return balloon_stats.balloon_low || balloon_stats.balloon_high;
384 }
385 
386 static enum bp_state increase_reservation(unsigned long nr_pages)
387 {
388 	int rc;
389 	unsigned long i;
390 	struct page   *page;
391 
392 	if (nr_pages > ARRAY_SIZE(frame_list))
393 		nr_pages = ARRAY_SIZE(frame_list);
394 
395 	page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
396 	for (i = 0; i < nr_pages; i++) {
397 		if (!page) {
398 			nr_pages = i;
399 			break;
400 		}
401 
402 		frame_list[i] = page_to_xen_pfn(page);
403 		page = balloon_next_page(page);
404 	}
405 
406 	rc = xenmem_reservation_increase(nr_pages, frame_list);
407 	if (rc <= 0)
408 		return BP_EAGAIN;
409 
410 	for (i = 0; i < rc; i++) {
411 		page = balloon_retrieve(false);
412 		BUG_ON(page == NULL);
413 
414 		xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]);
415 
416 		/* Relinquish the page back to the allocator. */
417 		free_reserved_page(page);
418 	}
419 
420 	balloon_stats.current_pages += rc;
421 
422 	return BP_DONE;
423 }
424 
425 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
426 {
427 	enum bp_state state = BP_DONE;
428 	unsigned long i;
429 	struct page *page, *tmp;
430 	int ret;
431 	LIST_HEAD(pages);
432 
433 	if (nr_pages > ARRAY_SIZE(frame_list))
434 		nr_pages = ARRAY_SIZE(frame_list);
435 
436 	for (i = 0; i < nr_pages; i++) {
437 		page = alloc_page(gfp);
438 		if (page == NULL) {
439 			nr_pages = i;
440 			state = BP_EAGAIN;
441 			break;
442 		}
443 		adjust_managed_page_count(page, -1);
444 		xenmem_reservation_scrub_page(page);
445 		list_add(&page->lru, &pages);
446 	}
447 
448 	/*
449 	 * Ensure that ballooned highmem pages don't have kmaps.
450 	 *
451 	 * Do this before changing the p2m as kmap_flush_unused()
452 	 * reads PTEs to obtain pages (and hence needs the original
453 	 * p2m entry).
454 	 */
455 	kmap_flush_unused();
456 
457 	/*
458 	 * Setup the frame, update direct mapping, invalidate P2M,
459 	 * and add to balloon.
460 	 */
461 	i = 0;
462 	list_for_each_entry_safe(page, tmp, &pages, lru) {
463 		frame_list[i++] = xen_page_to_gfn(page);
464 
465 		xenmem_reservation_va_mapping_reset(1, &page);
466 
467 		list_del(&page->lru);
468 
469 		balloon_append(page);
470 	}
471 
472 	flush_tlb_all();
473 
474 	ret = xenmem_reservation_decrease(nr_pages, frame_list);
475 	BUG_ON(ret != nr_pages);
476 
477 	balloon_stats.current_pages -= nr_pages;
478 
479 	return state;
480 }
481 
482 /*
483  * Stop waiting if either state is BP_DONE and ballooning action is
484  * needed, or if the credit has changed while state is not BP_DONE.
485  */
486 static bool balloon_thread_cond(long credit)
487 {
488 	if (balloon_state == BP_DONE)
489 		credit = 0;
490 
491 	return current_credit() != credit || kthread_should_stop();
492 }
493 
494 /*
495  * As this is a kthread it is guaranteed to run as a single instance only.
496  * We may of course race updates of the target counts (which are protected
497  * by the balloon lock), or with changes to the Xen hard limit, but we will
498  * recover from these in time.
499  */
500 static int balloon_thread(void *unused)
501 {
502 	long credit;
503 	unsigned long timeout;
504 
505 	set_freezable();
506 	for (;;) {
507 		switch (balloon_state) {
508 		case BP_DONE:
509 		case BP_ECANCELED:
510 			timeout = 3600 * HZ;
511 			break;
512 		case BP_EAGAIN:
513 			timeout = balloon_stats.schedule_delay * HZ;
514 			break;
515 		case BP_WAIT:
516 			timeout = HZ;
517 			break;
518 		}
519 
520 		credit = current_credit();
521 
522 		wait_event_freezable_timeout(balloon_thread_wq,
523 			balloon_thread_cond(credit), timeout);
524 
525 		if (kthread_should_stop())
526 			return 0;
527 
528 		mutex_lock(&balloon_mutex);
529 
530 		credit = current_credit();
531 
532 		if (credit > 0) {
533 			if (balloon_is_inflated())
534 				balloon_state = increase_reservation(credit);
535 			else
536 				balloon_state = reserve_additional_memory();
537 		}
538 
539 		if (credit < 0) {
540 			long n_pages;
541 
542 			n_pages = min(-credit, si_mem_available());
543 			balloon_state = decrease_reservation(n_pages,
544 							     GFP_BALLOON);
545 			if (balloon_state == BP_DONE && n_pages != -credit &&
546 			    n_pages < totalreserve_pages)
547 				balloon_state = BP_EAGAIN;
548 		}
549 
550 		update_schedule();
551 
552 		mutex_unlock(&balloon_mutex);
553 
554 		cond_resched();
555 	}
556 }
557 
558 /* Resets the Xen limit, sets new target, and kicks off processing. */
559 void balloon_set_new_target(unsigned long target)
560 {
561 	/* No need for lock. Not read-modify-write updates. */
562 	balloon_stats.target_pages = target;
563 	wake_up(&balloon_thread_wq);
564 }
565 EXPORT_SYMBOL_GPL(balloon_set_new_target);
566 
567 static int add_ballooned_pages(unsigned int nr_pages)
568 {
569 	enum bp_state st;
570 
571 	if (xen_hotplug_unpopulated) {
572 		st = reserve_additional_memory();
573 		if (st != BP_ECANCELED) {
574 			int rc;
575 
576 			mutex_unlock(&balloon_mutex);
577 			rc = wait_event_interruptible(balloon_wq,
578 				   !list_empty(&ballooned_pages));
579 			mutex_lock(&balloon_mutex);
580 			return rc ? -ENOMEM : 0;
581 		}
582 	}
583 
584 	if (si_mem_available() < nr_pages)
585 		return -ENOMEM;
586 
587 	st = decrease_reservation(nr_pages, GFP_USER);
588 	if (st != BP_DONE)
589 		return -ENOMEM;
590 
591 	return 0;
592 }
593 
594 /**
595  * xen_alloc_ballooned_pages - get pages that have been ballooned out
596  * @nr_pages: Number of pages to get
597  * @pages: pages returned
598  * @return 0 on success, error otherwise
599  */
600 int xen_alloc_ballooned_pages(unsigned int nr_pages, struct page **pages)
601 {
602 	unsigned int pgno = 0;
603 	struct page *page;
604 	int ret;
605 
606 	mutex_lock(&balloon_mutex);
607 
608 	balloon_stats.target_unpopulated += nr_pages;
609 
610 	while (pgno < nr_pages) {
611 		page = balloon_retrieve(true);
612 		if (page) {
613 			pages[pgno++] = page;
614 #ifdef CONFIG_XEN_HAVE_PVMMU
615 			/*
616 			 * We don't support PV MMU when Linux and Xen is using
617 			 * different page granularity.
618 			 */
619 			BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
620 
621 			if (!xen_feature(XENFEAT_auto_translated_physmap)) {
622 				ret = xen_alloc_p2m_entry(page_to_pfn(page));
623 				if (ret < 0)
624 					goto out_undo;
625 			}
626 #endif
627 		} else {
628 			ret = add_ballooned_pages(nr_pages - pgno);
629 			if (ret < 0)
630 				goto out_undo;
631 		}
632 	}
633 	mutex_unlock(&balloon_mutex);
634 	return 0;
635  out_undo:
636 	mutex_unlock(&balloon_mutex);
637 	xen_free_ballooned_pages(pgno, pages);
638 	/*
639 	 * NB: xen_free_ballooned_pages will only subtract pgno pages, but since
640 	 * target_unpopulated is incremented with nr_pages at the start we need
641 	 * to remove the remaining ones also, or accounting will be screwed.
642 	 */
643 	balloon_stats.target_unpopulated -= nr_pages - pgno;
644 	return ret;
645 }
646 EXPORT_SYMBOL(xen_alloc_ballooned_pages);
647 
648 /**
649  * xen_free_ballooned_pages - return pages retrieved with get_ballooned_pages
650  * @nr_pages: Number of pages
651  * @pages: pages to return
652  */
653 void xen_free_ballooned_pages(unsigned int nr_pages, struct page **pages)
654 {
655 	unsigned int i;
656 
657 	mutex_lock(&balloon_mutex);
658 
659 	for (i = 0; i < nr_pages; i++) {
660 		if (pages[i])
661 			balloon_append(pages[i]);
662 	}
663 
664 	balloon_stats.target_unpopulated -= nr_pages;
665 
666 	/* The balloon may be too large now. Shrink it if needed. */
667 	if (current_credit())
668 		wake_up(&balloon_thread_wq);
669 
670 	mutex_unlock(&balloon_mutex);
671 }
672 EXPORT_SYMBOL(xen_free_ballooned_pages);
673 
674 static void __init balloon_add_regions(void)
675 {
676 #if defined(CONFIG_XEN_PV)
677 	unsigned long start_pfn, pages;
678 	unsigned long pfn, extra_pfn_end;
679 	unsigned int i;
680 
681 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
682 		pages = xen_extra_mem[i].n_pfns;
683 		if (!pages)
684 			continue;
685 
686 		start_pfn = xen_extra_mem[i].start_pfn;
687 
688 		/*
689 		 * If the amount of usable memory has been limited (e.g., with
690 		 * the 'mem' command line parameter), don't add pages beyond
691 		 * this limit.
692 		 */
693 		extra_pfn_end = min(max_pfn, start_pfn + pages);
694 
695 		for (pfn = start_pfn; pfn < extra_pfn_end; pfn++)
696 			balloon_append(pfn_to_page(pfn));
697 
698 		balloon_stats.total_pages += extra_pfn_end - start_pfn;
699 	}
700 #endif
701 }
702 
703 static int __init balloon_init(void)
704 {
705 	struct task_struct *task;
706 
707 	if (!xen_domain())
708 		return -ENODEV;
709 
710 	pr_info("Initialising balloon driver\n");
711 
712 #ifdef CONFIG_XEN_PV
713 	balloon_stats.current_pages = xen_pv_domain()
714 		? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
715 		: get_num_physpages();
716 #else
717 	balloon_stats.current_pages = get_num_physpages();
718 #endif
719 	balloon_stats.target_pages  = balloon_stats.current_pages;
720 	balloon_stats.balloon_low   = 0;
721 	balloon_stats.balloon_high  = 0;
722 	balloon_stats.total_pages   = balloon_stats.current_pages;
723 
724 	balloon_stats.schedule_delay = 1;
725 	balloon_stats.max_schedule_delay = 32;
726 	balloon_stats.retry_count = 1;
727 	balloon_stats.max_retry_count = 4;
728 
729 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
730 	set_online_page_callback(&xen_online_page);
731 	register_memory_notifier(&xen_memory_nb);
732 	register_sysctl_init("xen/balloon", balloon_table);
733 #endif
734 
735 	balloon_add_regions();
736 
737 	task = kthread_run(balloon_thread, NULL, "xen-balloon");
738 	if (IS_ERR(task)) {
739 		pr_err("xen-balloon thread could not be started, ballooning will not work!\n");
740 		return PTR_ERR(task);
741 	}
742 
743 	/* Init the xen-balloon driver. */
744 	xen_balloon_init();
745 
746 	return 0;
747 }
748 subsys_initcall(balloon_init);
749 
750 static int __init balloon_wait_finish(void)
751 {
752 	long credit, last_credit = 0;
753 	unsigned long last_changed = 0;
754 
755 	if (!xen_domain())
756 		return -ENODEV;
757 
758 	/* PV guests don't need to wait. */
759 	if (xen_pv_domain() || !current_credit())
760 		return 0;
761 
762 	pr_notice("Waiting for initial ballooning down having finished.\n");
763 
764 	while ((credit = current_credit()) < 0) {
765 		if (credit != last_credit) {
766 			last_changed = jiffies;
767 			last_credit = credit;
768 		}
769 		if (balloon_state == BP_ECANCELED) {
770 			pr_warn_once("Initial ballooning failed, %ld pages need to be freed.\n",
771 				     -credit);
772 			if (time_is_before_eq_jiffies(last_changed + HZ * balloon_boot_timeout))
773 				panic("Initial ballooning failed!\n");
774 		}
775 
776 		schedule_timeout_interruptible(HZ / 10);
777 	}
778 
779 	pr_notice("Initial ballooning down finished.\n");
780 
781 	return 0;
782 }
783 late_initcall_sync(balloon_wait_finish);
784