Linux Audio

Check our new training course

Loading...
v6.8
  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
 81static uint __read_mostly balloon_boot_timeout = 180;
 82module_param(balloon_boot_timeout, uint, 0444);
 83
 84#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 85static int xen_hotplug_unpopulated;
 86
 87static 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#else
100#define xen_hotplug_unpopulated 0
101#endif
102
103/*
104 * Use one extent per PAGE_SIZE to avoid to break down the page into
105 * multiple frame.
106 */
107#define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
108
109/*
110 * balloon_thread() state:
111 *
112 * BP_DONE: done or nothing to do,
113 * BP_WAIT: wait to be rescheduled,
114 * BP_EAGAIN: error, go to sleep,
115 * BP_ECANCELED: error, balloon operation canceled.
116 */
117
118static enum bp_state {
119	BP_DONE,
120	BP_WAIT,
121	BP_EAGAIN,
122	BP_ECANCELED
123} balloon_state = BP_DONE;
124
125/* Main waiting point for xen-balloon thread. */
126static DECLARE_WAIT_QUEUE_HEAD(balloon_thread_wq);
127
128static DEFINE_MUTEX(balloon_mutex);
129
130struct balloon_stats balloon_stats;
131EXPORT_SYMBOL_GPL(balloon_stats);
132
133/* We increase/decrease in batches which fit in a page */
134static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
 
135
136
137/* List of ballooned pages, threaded through the mem_map array. */
138static LIST_HEAD(ballooned_pages);
139static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
 
 
 
140
141/* When ballooning out (allocating memory to return to Xen) we don't really
142   want the kernel to try too hard since that can trigger the oom killer. */
143#define GFP_BALLOON \
144	(GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
145
146/* balloon_append: add the given page to the balloon. */
147static void balloon_append(struct page *page)
148{
149	__SetPageOffline(page);
 
 
 
150
 
 
 
151	/* Lowmem is re-populated first, so highmem pages go at list tail. */
152	if (PageHighMem(page)) {
153		list_add_tail(&page->lru, &ballooned_pages);
154		balloon_stats.balloon_high++;
155	} else {
156		list_add(&page->lru, &ballooned_pages);
157		balloon_stats.balloon_low++;
158	}
159	wake_up(&balloon_wq);
 
 
 
 
 
160}
161
162/* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
163static struct page *balloon_retrieve(bool require_lowmem)
164{
165	struct page *page;
166
167	if (list_empty(&ballooned_pages))
168		return NULL;
169
170	page = list_entry(ballooned_pages.next, struct page, lru);
171	if (require_lowmem && PageHighMem(page))
172		return NULL;
 
173	list_del(&page->lru);
174
175	if (PageHighMem(page))
176		balloon_stats.balloon_high--;
177	else
178		balloon_stats.balloon_low--;
179
180	__ClearPageOffline(page);
 
181	return page;
182}
183
184static struct page *balloon_next_page(struct page *page)
185{
186	struct list_head *next = page->lru.next;
187	if (next == &ballooned_pages)
188		return NULL;
189	return list_entry(next, struct page, lru);
190}
191
192static void update_schedule(void)
193{
194	if (balloon_state == BP_WAIT || balloon_state == BP_ECANCELED)
195		return;
196
197	if (balloon_state == BP_DONE) {
198		balloon_stats.schedule_delay = 1;
199		balloon_stats.retry_count = 1;
200		return;
201	}
202
203	++balloon_stats.retry_count;
204
205	if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
206			balloon_stats.retry_count > balloon_stats.max_retry_count) {
207		balloon_stats.schedule_delay = 1;
208		balloon_stats.retry_count = 1;
209		balloon_state = BP_ECANCELED;
210		return;
211	}
212
213	balloon_stats.schedule_delay <<= 1;
214
215	if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
216		balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
217
218	balloon_state = BP_EAGAIN;
219}
220
221#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
222static void release_memory_resource(struct resource *resource)
223{
224	if (!resource)
225		return;
226
227	/*
228	 * No need to reset region to identity mapped since we now
229	 * know that no I/O can be in this region
230	 */
231	release_resource(resource);
232	kfree(resource);
233}
234
235static struct resource *additional_memory_resource(phys_addr_t size)
236{
237	struct resource *res;
238	int ret;
239
240	res = kzalloc(sizeof(*res), GFP_KERNEL);
241	if (!res)
242		return NULL;
243
244	res->name = "System RAM";
245	res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
246
247	ret = allocate_resource(&iomem_resource, res,
248				size, 0, -1,
249				PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
250	if (ret < 0) {
251		pr_err("Cannot allocate new System RAM resource\n");
252		kfree(res);
253		return NULL;
254	}
255
256	return res;
257}
258
259static enum bp_state reserve_additional_memory(void)
 
 
 
 
 
 
 
 
 
 
 
260{
261	long credit;
262	struct resource *resource;
263	int nid, rc;
264	unsigned long balloon_hotplug;
265
266	credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
267		- balloon_stats.total_pages;
268
269	/*
270	 * Already hotplugged enough pages?  Wait for them to be
271	 * onlined.
272	 */
273	if (credit <= 0)
274		return BP_WAIT;
275
276	balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
277
278	resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
279	if (!resource)
280		goto err;
281
282	nid = memory_add_physaddr_to_nid(resource->start);
 
 
283
284#ifdef CONFIG_XEN_HAVE_PVMMU
285	/*
286	 * We don't support PV MMU when Linux and Xen is using
287	 * different page granularity.
288	 */
289	BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
290
291        /*
292         * add_memory() will build page tables for the new memory so
293         * the p2m must contain invalid entries so the correct
294         * non-present PTEs will be written.
295         *
296         * If a failure occurs, the original (identity) p2m entries
297         * are not restored since this region is now known not to
298         * conflict with any devices.
299         */ 
300	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
301		unsigned long pfn, i;
302
303		pfn = PFN_DOWN(resource->start);
304		for (i = 0; i < balloon_hotplug; i++) {
305			if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
306				pr_warn("set_phys_to_machine() failed, no memory added\n");
307				goto err;
308			}
309                }
310	}
311#endif
312
313	/*
314	 * add_memory_resource() will call online_pages() which in its turn
315	 * will call xen_online_page() callback causing deadlock if we don't
316	 * release balloon_mutex here. Unlocking here is safe because the
317	 * callers drop the mutex before trying again.
318	 */
319	mutex_unlock(&balloon_mutex);
320	/* add_memory_resource() requires the device_hotplug lock */
321	lock_device_hotplug();
322	rc = add_memory_resource(nid, resource, MHP_MERGE_RESOURCE);
323	unlock_device_hotplug();
324	mutex_lock(&balloon_mutex);
325
326	if (rc) {
327		pr_warn("Cannot add additional memory (%i)\n", rc);
328		goto err;
329	}
330
331	balloon_stats.total_pages += balloon_hotplug;
332
333	return BP_WAIT;
334  err:
335	release_memory_resource(resource);
336	return BP_ECANCELED;
337}
338
339static void xen_online_page(struct page *page, unsigned int order)
340{
341	unsigned long i, size = (1 << order);
342	unsigned long start_pfn = page_to_pfn(page);
343	struct page *p;
344
345	pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn);
346	mutex_lock(&balloon_mutex);
347	for (i = 0; i < size; i++) {
348		p = pfn_to_page(start_pfn + i);
349		balloon_append(p);
350	}
 
 
 
 
351	mutex_unlock(&balloon_mutex);
352}
353
354static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
355{
356	if (val == MEM_ONLINE)
357		wake_up(&balloon_thread_wq);
358
359	return NOTIFY_OK;
360}
361
362static struct notifier_block xen_memory_nb = {
363	.notifier_call = xen_memory_notifier,
364	.priority = 0
365};
366#else
367static enum bp_state reserve_additional_memory(void)
368{
369	balloon_stats.target_pages = balloon_stats.current_pages +
370				     balloon_stats.target_unpopulated;
371	return BP_ECANCELED;
 
 
 
 
 
372}
373#endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
374
375static long current_credit(void)
376{
377	return balloon_stats.target_pages - balloon_stats.current_pages;
 
 
 
378}
379
380static bool balloon_is_inflated(void)
381{
382	return balloon_stats.balloon_low || balloon_stats.balloon_high;
 
383}
 
384
385static enum bp_state increase_reservation(unsigned long nr_pages)
386{
387	int rc;
388	unsigned long i;
389	struct page   *page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
390
391	if (nr_pages > ARRAY_SIZE(frame_list))
392		nr_pages = ARRAY_SIZE(frame_list);
393
394	page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
395	for (i = 0; i < nr_pages; i++) {
396		if (!page) {
397			nr_pages = i;
398			break;
399		}
400
401		frame_list[i] = page_to_xen_pfn(page);
402		page = balloon_next_page(page);
403	}
404
405	rc = xenmem_reservation_increase(nr_pages, frame_list);
 
 
406	if (rc <= 0)
407		return BP_EAGAIN;
408
409	for (i = 0; i < rc; i++) {
410		page = balloon_retrieve(false);
411		BUG_ON(page == NULL);
412
413		xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
414
415		/* Relinquish the page back to the allocator. */
416		free_reserved_page(page);
417	}
418
419	balloon_stats.current_pages += rc;
420
421	return BP_DONE;
422}
423
424static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
425{
426	enum bp_state state = BP_DONE;
427	unsigned long i;
428	struct page *page, *tmp;
429	int ret;
430	LIST_HEAD(pages);
 
 
 
 
 
 
 
 
 
 
 
 
 
431
432	if (nr_pages > ARRAY_SIZE(frame_list))
433		nr_pages = ARRAY_SIZE(frame_list);
434
435	for (i = 0; i < nr_pages; i++) {
436		page = alloc_page(gfp);
437		if (page == NULL) {
438			nr_pages = i;
439			state = BP_EAGAIN;
440			break;
441		}
442		adjust_managed_page_count(page, -1);
443		xenmem_reservation_scrub_page(page);
444		list_add(&page->lru, &pages);
445	}
446
447	/*
448	 * Ensure that ballooned highmem pages don't have kmaps.
449	 *
450	 * Do this before changing the p2m as kmap_flush_unused()
451	 * reads PTEs to obtain pages (and hence needs the original
452	 * p2m entry).
453	 */
454	kmap_flush_unused();
455
456	/*
457	 * Setup the frame, update direct mapping, invalidate P2M,
458	 * and add to balloon.
459	 */
460	i = 0;
461	list_for_each_entry_safe(page, tmp, &pages, lru) {
462		frame_list[i++] = xen_page_to_gfn(page);
463
464		xenmem_reservation_va_mapping_reset(1, &page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
465
466		list_del(&page->lru);
 
 
467
468		balloon_append(page);
469	}
470
471	flush_tlb_all();
472
473	ret = xenmem_reservation_decrease(nr_pages, frame_list);
 
 
474	BUG_ON(ret != nr_pages);
475
476	balloon_stats.current_pages -= nr_pages;
477
478	return state;
479}
480
481/*
482 * Stop waiting if either state is BP_DONE and ballooning action is
483 * needed, or if the credit has changed while state is not BP_DONE.
484 */
485static bool balloon_thread_cond(long credit)
486{
487	if (balloon_state == BP_DONE)
488		credit = 0;
489
490	return current_credit() != credit || kthread_should_stop();
491}
492
493/*
494 * As this is a kthread it is guaranteed to run as a single instance only.
495 * We may of course race updates of the target counts (which are protected
496 * by the balloon lock), or with changes to the Xen hard limit, but we will
497 * recover from these in time.
498 */
499static int balloon_thread(void *unused)
500{
 
501	long credit;
502	unsigned long timeout;
503
504	set_freezable();
505	for (;;) {
506		switch (balloon_state) {
507		case BP_DONE:
508		case BP_ECANCELED:
509			timeout = 3600 * HZ;
510			break;
511		case BP_EAGAIN:
512			timeout = balloon_stats.schedule_delay * HZ;
513			break;
514		case BP_WAIT:
515			timeout = HZ;
516			break;
517		}
518
519		credit = current_credit();
520
521		wait_event_freezable_timeout(balloon_thread_wq,
522			balloon_thread_cond(credit), timeout);
523
524		if (kthread_should_stop())
525			return 0;
526
527		mutex_lock(&balloon_mutex);
528
 
529		credit = current_credit();
530
531		if (credit > 0) {
532			if (balloon_is_inflated())
533				balloon_state = increase_reservation(credit);
534			else
535				balloon_state = reserve_additional_memory();
536		}
537
538		if (credit < 0) {
539			long n_pages;
540
541			n_pages = min(-credit, si_mem_available());
542			balloon_state = decrease_reservation(n_pages,
543							     GFP_BALLOON);
544			if (balloon_state == BP_DONE && n_pages != -credit &&
545			    n_pages < totalreserve_pages)
546				balloon_state = BP_EAGAIN;
547		}
548
549		update_schedule();
 
 
 
 
550
551		mutex_unlock(&balloon_mutex);
 
 
552
553		cond_resched();
554	}
 
 
 
 
 
 
 
 
 
 
 
555}
556
557/* Resets the Xen limit, sets new target, and kicks off processing. */
558void balloon_set_new_target(unsigned long target)
559{
560	/* No need for lock. Not read-modify-write updates. */
561	balloon_stats.target_pages = target;
562	wake_up(&balloon_thread_wq);
563}
564EXPORT_SYMBOL_GPL(balloon_set_new_target);
565
566static int add_ballooned_pages(unsigned int nr_pages)
567{
568	enum bp_state st;
569
570	if (xen_hotplug_unpopulated) {
571		st = reserve_additional_memory();
572		if (st != BP_ECANCELED) {
573			int rc;
574
575			mutex_unlock(&balloon_mutex);
576			rc = wait_event_interruptible(balloon_wq,
577				   !list_empty(&ballooned_pages));
578			mutex_lock(&balloon_mutex);
579			return rc ? -ENOMEM : 0;
580		}
581	}
582
583	if (si_mem_available() < nr_pages)
584		return -ENOMEM;
585
586	st = decrease_reservation(nr_pages, GFP_USER);
587	if (st != BP_DONE)
588		return -ENOMEM;
589
590	return 0;
591}
592
593/**
594 * xen_alloc_ballooned_pages - get pages that have been ballooned out
595 * @nr_pages: Number of pages to get
596 * @pages: pages returned
 
597 * @return 0 on success, error otherwise
598 */
599int xen_alloc_ballooned_pages(unsigned int nr_pages, struct page **pages)
600{
601	unsigned int pgno = 0;
602	struct page *page;
603	int ret;
604
605	mutex_lock(&balloon_mutex);
606
607	balloon_stats.target_unpopulated += nr_pages;
608
609	while (pgno < nr_pages) {
610		page = balloon_retrieve(true);
611		if (page) {
612			pages[pgno++] = page;
613#ifdef CONFIG_XEN_HAVE_PVMMU
614			/*
615			 * We don't support PV MMU when Linux and Xen is using
616			 * different page granularity.
617			 */
618			BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
619
620			if (!xen_feature(XENFEAT_auto_translated_physmap)) {
621				ret = xen_alloc_p2m_entry(page_to_pfn(page));
622				if (ret < 0)
623					goto out_undo;
624			}
625#endif
626		} else {
627			ret = add_ballooned_pages(nr_pages - pgno);
628			if (ret < 0)
 
 
 
 
629				goto out_undo;
630		}
631	}
632	mutex_unlock(&balloon_mutex);
633	return 0;
634 out_undo:
 
 
 
 
635	mutex_unlock(&balloon_mutex);
636	xen_free_ballooned_pages(pgno, pages);
637	/*
638	 * NB: xen_free_ballooned_pages will only subtract pgno pages, but since
639	 * target_unpopulated is incremented with nr_pages at the start we need
640	 * to remove the remaining ones also, or accounting will be screwed.
641	 */
642	balloon_stats.target_unpopulated -= nr_pages - pgno;
643	return ret;
644}
645EXPORT_SYMBOL(xen_alloc_ballooned_pages);
646
647/**
648 * xen_free_ballooned_pages - return pages retrieved with get_ballooned_pages
649 * @nr_pages: Number of pages
650 * @pages: pages to return
651 */
652void xen_free_ballooned_pages(unsigned int nr_pages, struct page **pages)
653{
654	unsigned int i;
655
656	mutex_lock(&balloon_mutex);
657
658	for (i = 0; i < nr_pages; i++) {
659		if (pages[i])
660			balloon_append(pages[i]);
661	}
662
663	balloon_stats.target_unpopulated -= nr_pages;
664
665	/* The balloon may be too large now. Shrink it if needed. */
666	if (current_credit())
667		wake_up(&balloon_thread_wq);
668
669	mutex_unlock(&balloon_mutex);
670}
671EXPORT_SYMBOL(xen_free_ballooned_pages);
672
673static void __init balloon_add_regions(void)
 
674{
675#if defined(CONFIG_XEN_PV)
676	unsigned long start_pfn, pages;
677	unsigned long pfn, extra_pfn_end;
678	unsigned int i;
679
680	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
681		pages = xen_extra_mem[i].n_pfns;
682		if (!pages)
683			continue;
 
 
684
685		start_pfn = xen_extra_mem[i].start_pfn;
 
 
 
 
 
 
 
686
687		/*
688		 * If the amount of usable memory has been limited (e.g., with
689		 * the 'mem' command line parameter), don't add pages beyond
690		 * this limit.
691		 */
692		extra_pfn_end = min(max_pfn, start_pfn + pages);
 
 
 
 
 
 
 
693
694		for (pfn = start_pfn; pfn < extra_pfn_end; pfn++)
695			balloon_append(pfn_to_page(pfn));
696
697		balloon_stats.total_pages += extra_pfn_end - start_pfn;
 
 
 
 
 
 
 
 
 
 
698	}
699#endif
700}
701
 
 
 
 
702static int __init balloon_init(void)
703{
704	struct task_struct *task;
705
706	if (!xen_domain())
707		return -ENODEV;
708
 
 
 
 
 
 
 
 
 
 
 
 
 
 
709	pr_info("Initialising balloon driver\n");
710
711#ifdef CONFIG_XEN_PV
712	balloon_stats.current_pages = xen_pv_domain()
713		? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
714		: get_num_physpages();
715#else
716	balloon_stats.current_pages = get_num_physpages();
717#endif
718	balloon_stats.target_pages  = balloon_stats.current_pages;
719	balloon_stats.balloon_low   = 0;
720	balloon_stats.balloon_high  = 0;
721	balloon_stats.total_pages   = balloon_stats.current_pages;
722
723	balloon_stats.schedule_delay = 1;
724	balloon_stats.max_schedule_delay = 32;
725	balloon_stats.retry_count = 1;
726	balloon_stats.max_retry_count = 4;
727
728#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 
 
 
729	set_online_page_callback(&xen_online_page);
730	register_memory_notifier(&xen_memory_nb);
731	register_sysctl_init("xen/balloon", balloon_table);
732#endif
733
734	balloon_add_regions();
735
736	task = kthread_run(balloon_thread, NULL, "xen-balloon");
737	if (IS_ERR(task)) {
738		pr_err("xen-balloon thread could not be started, ballooning will not work!\n");
739		return PTR_ERR(task);
740	}
741
742	/* Init the xen-balloon driver. */
743	xen_balloon_init();
744
745	return 0;
746}
 
747subsys_initcall(balloon_init);
748
749static int __init balloon_wait_finish(void)
750{
751	long credit, last_credit = 0;
752	unsigned long last_changed = 0;
753
754	if (!xen_domain())
755		return -ENODEV;
756
757	/* PV guests don't need to wait. */
758	if (xen_pv_domain() || !current_credit())
759		return 0;
760
761	pr_notice("Waiting for initial ballooning down having finished.\n");
762
763	while ((credit = current_credit()) < 0) {
764		if (credit != last_credit) {
765			last_changed = jiffies;
766			last_credit = credit;
767		}
768		if (balloon_state == BP_ECANCELED) {
769			pr_warn_once("Initial ballooning failed, %ld pages need to be freed.\n",
770				     -credit);
771			if (time_is_before_eq_jiffies(last_changed + HZ * balloon_boot_timeout))
772				panic("Initial ballooning failed!\n");
773		}
774
775		schedule_timeout_interruptible(HZ / 10);
776	}
777
778	pr_notice("Initial ballooning down finished.\n");
779
780	return 0;
781}
782late_initcall_sync(balloon_wait_finish);
 
 
v3.15
  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/errno.h>
 45#include <linux/module.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
 58#include <asm/page.h>
 59#include <asm/pgalloc.h>
 60#include <asm/pgtable.h>
 61#include <asm/tlb.h>
 62
 63#include <asm/xen/hypervisor.h>
 64#include <asm/xen/hypercall.h>
 65
 66#include <xen/xen.h>
 67#include <xen/interface/xen.h>
 68#include <xen/interface/memory.h>
 69#include <xen/balloon.h>
 70#include <xen/features.h>
 71#include <xen/page.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 72
 73/*
 74 * balloon_process() state:
 75 *
 76 * BP_DONE: done or nothing to do,
 
 77 * BP_EAGAIN: error, go to sleep,
 78 * BP_ECANCELED: error, balloon operation canceled.
 79 */
 80
 81enum bp_state {
 82	BP_DONE,
 
 83	BP_EAGAIN,
 84	BP_ECANCELED
 85};
 86
 
 
 87
 88static DEFINE_MUTEX(balloon_mutex);
 89
 90struct balloon_stats balloon_stats;
 91EXPORT_SYMBOL_GPL(balloon_stats);
 92
 93/* We increase/decrease in batches which fit in a page */
 94static xen_pfn_t frame_list[PAGE_SIZE / sizeof(unsigned long)];
 95static DEFINE_PER_CPU(struct page *, balloon_scratch_page);
 96
 97
 98/* List of ballooned pages, threaded through the mem_map array. */
 99static LIST_HEAD(ballooned_pages);
100
101/* Main work function, always executed in process context. */
102static void balloon_process(struct work_struct *work);
103static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
104
105/* When ballooning out (allocating memory to return to Xen) we don't really
106   want the kernel to try too hard since that can trigger the oom killer. */
107#define GFP_BALLOON \
108	(GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
109
110static void scrub_page(struct page *page)
 
111{
112#ifdef CONFIG_XEN_SCRUB_PAGES
113	clear_highpage(page);
114#endif
115}
116
117/* balloon_append: add the given page to the balloon. */
118static void __balloon_append(struct page *page)
119{
120	/* Lowmem is re-populated first, so highmem pages go at list tail. */
121	if (PageHighMem(page)) {
122		list_add_tail(&page->lru, &ballooned_pages);
123		balloon_stats.balloon_high++;
124	} else {
125		list_add(&page->lru, &ballooned_pages);
126		balloon_stats.balloon_low++;
127	}
128}
129
130static void balloon_append(struct page *page)
131{
132	__balloon_append(page);
133	adjust_managed_page_count(page, -1);
134}
135
136/* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
137static struct page *balloon_retrieve(bool prefer_highmem)
138{
139	struct page *page;
140
141	if (list_empty(&ballooned_pages))
142		return NULL;
143
144	if (prefer_highmem)
145		page = list_entry(ballooned_pages.prev, struct page, lru);
146	else
147		page = list_entry(ballooned_pages.next, struct page, lru);
148	list_del(&page->lru);
149
150	if (PageHighMem(page))
151		balloon_stats.balloon_high--;
152	else
153		balloon_stats.balloon_low--;
154
155	adjust_managed_page_count(page, 1);
156
157	return page;
158}
159
160static struct page *balloon_next_page(struct page *page)
161{
162	struct list_head *next = page->lru.next;
163	if (next == &ballooned_pages)
164		return NULL;
165	return list_entry(next, struct page, lru);
166}
167
168static enum bp_state update_schedule(enum bp_state state)
169{
170	if (state == BP_DONE) {
 
 
 
171		balloon_stats.schedule_delay = 1;
172		balloon_stats.retry_count = 1;
173		return BP_DONE;
174	}
175
176	++balloon_stats.retry_count;
177
178	if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
179			balloon_stats.retry_count > balloon_stats.max_retry_count) {
180		balloon_stats.schedule_delay = 1;
181		balloon_stats.retry_count = 1;
182		return BP_ECANCELED;
 
183	}
184
185	balloon_stats.schedule_delay <<= 1;
186
187	if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
188		balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
189
190	return BP_EAGAIN;
191}
192
193#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
194static long current_credit(void)
195{
196	return balloon_stats.target_pages - balloon_stats.current_pages -
197		balloon_stats.hotplug_pages;
 
 
 
 
 
 
 
198}
199
200static bool balloon_is_inflated(void)
201{
202	if (balloon_stats.balloon_low || balloon_stats.balloon_high ||
203			balloon_stats.balloon_hotplug)
204		return true;
205	else
206		return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
207}
208
209/*
210 * reserve_additional_memory() adds memory region of size >= credit above
211 * max_pfn. New region is section aligned and size is modified to be multiple
212 * of section size. Those features allow optimal use of address space and
213 * establish proper alignment when this function is called first time after
214 * boot (last section not fully populated at boot time contains unused memory
215 * pages with PG_reserved bit not set; online_pages_range() does not allow page
216 * onlining in whole range if first onlined page does not have PG_reserved
217 * bit set). Real size of added memory is established at page onlining stage.
218 */
219
220static enum bp_state reserve_additional_memory(long credit)
221{
 
 
222	int nid, rc;
223	u64 hotplug_start_paddr;
224	unsigned long balloon_hotplug = credit;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
225
226	hotplug_start_paddr = PFN_PHYS(SECTION_ALIGN_UP(max_pfn));
227	balloon_hotplug = round_up(balloon_hotplug, PAGES_PER_SECTION);
228	nid = memory_add_physaddr_to_nid(hotplug_start_paddr);
229
230	rc = add_memory(nid, hotplug_start_paddr, balloon_hotplug << PAGE_SHIFT);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
231
232	if (rc) {
233		pr_info("%s: add_memory() failed: %i\n", __func__, rc);
234		return BP_EAGAIN;
235	}
236
237	balloon_hotplug -= credit;
238
239	balloon_stats.hotplug_pages += credit;
240	balloon_stats.balloon_hotplug = balloon_hotplug;
241
242	return BP_DONE;
243}
244
245static void xen_online_page(struct page *page)
246{
247	__online_page_set_limits(page);
 
 
248
 
249	mutex_lock(&balloon_mutex);
250
251	__balloon_append(page);
252
253	if (balloon_stats.hotplug_pages)
254		--balloon_stats.hotplug_pages;
255	else
256		--balloon_stats.balloon_hotplug;
257
258	mutex_unlock(&balloon_mutex);
259}
260
261static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
262{
263	if (val == MEM_ONLINE)
264		schedule_delayed_work(&balloon_worker, 0);
265
266	return NOTIFY_OK;
267}
268
269static struct notifier_block xen_memory_nb = {
270	.notifier_call = xen_memory_notifier,
271	.priority = 0
272};
273#else
274static long current_credit(void)
275{
276	unsigned long target = balloon_stats.target_pages;
277
278	target = min(target,
279		     balloon_stats.current_pages +
280		     balloon_stats.balloon_low +
281		     balloon_stats.balloon_high);
282
283	return target - balloon_stats.current_pages;
284}
 
285
286static bool balloon_is_inflated(void)
287{
288	if (balloon_stats.balloon_low || balloon_stats.balloon_high)
289		return true;
290	else
291		return false;
292}
293
294static enum bp_state reserve_additional_memory(long credit)
295{
296	balloon_stats.target_pages = balloon_stats.current_pages;
297	return BP_DONE;
298}
299#endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
300
301static enum bp_state increase_reservation(unsigned long nr_pages)
302{
303	int rc;
304	unsigned long  pfn, i;
305	struct page   *page;
306	struct xen_memory_reservation reservation = {
307		.address_bits = 0,
308		.extent_order = 0,
309		.domid        = DOMID_SELF
310	};
311
312#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
313	if (!balloon_stats.balloon_low && !balloon_stats.balloon_high) {
314		nr_pages = min(nr_pages, balloon_stats.balloon_hotplug);
315		balloon_stats.hotplug_pages += nr_pages;
316		balloon_stats.balloon_hotplug -= nr_pages;
317		return BP_DONE;
318	}
319#endif
320
321	if (nr_pages > ARRAY_SIZE(frame_list))
322		nr_pages = ARRAY_SIZE(frame_list);
323
324	page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
325	for (i = 0; i < nr_pages; i++) {
326		if (!page) {
327			nr_pages = i;
328			break;
329		}
330		frame_list[i] = page_to_pfn(page);
 
331		page = balloon_next_page(page);
332	}
333
334	set_xen_guest_handle(reservation.extent_start, frame_list);
335	reservation.nr_extents = nr_pages;
336	rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
337	if (rc <= 0)
338		return BP_EAGAIN;
339
340	for (i = 0; i < rc; i++) {
341		page = balloon_retrieve(false);
342		BUG_ON(page == NULL);
343
344		pfn = page_to_pfn(page);
345
346#ifdef CONFIG_XEN_HAVE_PVMMU
347		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
348			set_phys_to_machine(pfn, frame_list[i]);
349
350			/* Link back into the page tables if not highmem. */
351			if (!PageHighMem(page)) {
352				int ret;
353				ret = HYPERVISOR_update_va_mapping(
354						(unsigned long)__va(pfn << PAGE_SHIFT),
355						mfn_pte(frame_list[i], PAGE_KERNEL),
356						0);
357				BUG_ON(ret);
358			}
359		}
360#endif
361
362		/* Relinquish the page back to the allocator. */
363		__free_reserved_page(page);
364	}
365
366	balloon_stats.current_pages += rc;
367
368	return BP_DONE;
369}
370
371static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
372{
373	enum bp_state state = BP_DONE;
374	unsigned long  pfn, i;
375	struct page   *page;
376	int ret;
377	struct xen_memory_reservation reservation = {
378		.address_bits = 0,
379		.extent_order = 0,
380		.domid        = DOMID_SELF
381	};
382
383#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
384	if (balloon_stats.hotplug_pages) {
385		nr_pages = min(nr_pages, balloon_stats.hotplug_pages);
386		balloon_stats.hotplug_pages -= nr_pages;
387		balloon_stats.balloon_hotplug += nr_pages;
388		return BP_DONE;
389	}
390#endif
391
392	if (nr_pages > ARRAY_SIZE(frame_list))
393		nr_pages = ARRAY_SIZE(frame_list);
394
395	for (i = 0; i < nr_pages; i++) {
396		page = alloc_page(gfp);
397		if (page == NULL) {
398			nr_pages = i;
399			state = BP_EAGAIN;
400			break;
401		}
402		scrub_page(page);
403
404		frame_list[i] = page_to_pfn(page);
405	}
406
407	/*
408	 * Ensure that ballooned highmem pages don't have kmaps.
409	 *
410	 * Do this before changing the p2m as kmap_flush_unused()
411	 * reads PTEs to obtain pages (and hence needs the original
412	 * p2m entry).
413	 */
414	kmap_flush_unused();
415
416	/* Update direct mapping, invalidate P2M, and add to balloon. */
417	for (i = 0; i < nr_pages; i++) {
418		pfn = frame_list[i];
419		frame_list[i] = pfn_to_mfn(pfn);
420		page = pfn_to_page(pfn);
 
 
421
422#ifdef CONFIG_XEN_HAVE_PVMMU
423		/*
424		 * Ballooned out frames are effectively replaced with
425		 * a scratch frame.  Ensure direct mappings and the
426		 * p2m are consistent.
427		 */
428		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
429			unsigned long p;
430			struct page   *scratch_page = get_balloon_scratch_page();
431
432			if (!PageHighMem(page)) {
433				ret = HYPERVISOR_update_va_mapping(
434						(unsigned long)__va(pfn << PAGE_SHIFT),
435						pfn_pte(page_to_pfn(scratch_page),
436							PAGE_KERNEL_RO), 0);
437				BUG_ON(ret);
438			}
439			p = page_to_pfn(scratch_page);
440			__set_phys_to_machine(pfn, pfn_to_mfn(p));
441
442			put_balloon_scratch_page();
443		}
444#endif
445
446		balloon_append(page);
447	}
448
449	flush_tlb_all();
450
451	set_xen_guest_handle(reservation.extent_start, frame_list);
452	reservation.nr_extents   = nr_pages;
453	ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
454	BUG_ON(ret != nr_pages);
455
456	balloon_stats.current_pages -= nr_pages;
457
458	return state;
459}
460
461/*
462 * We avoid multiple worker processes conflicting via the balloon mutex.
 
 
 
 
 
 
 
 
 
 
 
 
463 * We may of course race updates of the target counts (which are protected
464 * by the balloon lock), or with changes to the Xen hard limit, but we will
465 * recover from these in time.
466 */
467static void balloon_process(struct work_struct *work)
468{
469	enum bp_state state = BP_DONE;
470	long credit;
 
471
472	mutex_lock(&balloon_mutex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
473
474	do {
475		credit = current_credit();
476
477		if (credit > 0) {
478			if (balloon_is_inflated())
479				state = increase_reservation(credit);
480			else
481				state = reserve_additional_memory(credit);
482		}
483
484		if (credit < 0)
485			state = decrease_reservation(-credit, GFP_BALLOON);
486
487		state = update_schedule(state);
 
 
 
 
 
 
488
489#ifndef CONFIG_PREEMPT
490		if (need_resched())
491			schedule();
492#endif
493	} while (credit && state == BP_DONE);
494
495	/* Schedule more work if there is some still to be done. */
496	if (state == BP_EAGAIN)
497		schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
498
499	mutex_unlock(&balloon_mutex);
500}
501
502struct page *get_balloon_scratch_page(void)
503{
504	struct page *ret = get_cpu_var(balloon_scratch_page);
505	BUG_ON(ret == NULL);
506	return ret;
507}
508
509void put_balloon_scratch_page(void)
510{
511	put_cpu_var(balloon_scratch_page);
512}
513
514/* Resets the Xen limit, sets new target, and kicks off processing. */
515void balloon_set_new_target(unsigned long target)
516{
517	/* No need for lock. Not read-modify-write updates. */
518	balloon_stats.target_pages = target;
519	schedule_delayed_work(&balloon_worker, 0);
520}
521EXPORT_SYMBOL_GPL(balloon_set_new_target);
522
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
523/**
524 * alloc_xenballooned_pages - get pages that have been ballooned out
525 * @nr_pages: Number of pages to get
526 * @pages: pages returned
527 * @highmem: allow highmem pages
528 * @return 0 on success, error otherwise
529 */
530int alloc_xenballooned_pages(int nr_pages, struct page **pages, bool highmem)
531{
532	int pgno = 0;
533	struct page *page;
 
 
534	mutex_lock(&balloon_mutex);
 
 
 
535	while (pgno < nr_pages) {
536		page = balloon_retrieve(highmem);
537		if (page && (highmem || !PageHighMem(page))) {
538			pages[pgno++] = page;
 
 
 
 
 
 
 
 
 
 
 
 
 
539		} else {
540			enum bp_state st;
541			if (page)
542				balloon_append(page);
543			st = decrease_reservation(nr_pages - pgno,
544					highmem ? GFP_HIGHUSER : GFP_USER);
545			if (st != BP_DONE)
546				goto out_undo;
547		}
548	}
549	mutex_unlock(&balloon_mutex);
550	return 0;
551 out_undo:
552	while (pgno)
553		balloon_append(pages[--pgno]);
554	/* Free the memory back to the kernel soon */
555	schedule_delayed_work(&balloon_worker, 0);
556	mutex_unlock(&balloon_mutex);
557	return -ENOMEM;
 
 
 
 
 
 
 
558}
559EXPORT_SYMBOL(alloc_xenballooned_pages);
560
561/**
562 * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
563 * @nr_pages: Number of pages
564 * @pages: pages to return
565 */
566void free_xenballooned_pages(int nr_pages, struct page **pages)
567{
568	int i;
569
570	mutex_lock(&balloon_mutex);
571
572	for (i = 0; i < nr_pages; i++) {
573		if (pages[i])
574			balloon_append(pages[i]);
575	}
576
 
 
577	/* The balloon may be too large now. Shrink it if needed. */
578	if (current_credit())
579		schedule_delayed_work(&balloon_worker, 0);
580
581	mutex_unlock(&balloon_mutex);
582}
583EXPORT_SYMBOL(free_xenballooned_pages);
584
585static void __init balloon_add_region(unsigned long start_pfn,
586				      unsigned long pages)
587{
 
 
588	unsigned long pfn, extra_pfn_end;
589	struct page *page;
590
591	/*
592	 * If the amount of usable memory has been limited (e.g., with
593	 * the 'mem' command line parameter), don't add pages beyond
594	 * this limit.
595	 */
596	extra_pfn_end = min(max_pfn, start_pfn + pages);
597
598	for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
599		page = pfn_to_page(pfn);
600		/* totalram_pages and totalhigh_pages do not
601		   include the boot-time balloon extension, so
602		   don't subtract from it. */
603		__balloon_append(page);
604	}
605}
606
607static int alloc_balloon_scratch_page(int cpu)
608{
609	if (per_cpu(balloon_scratch_page, cpu) != NULL)
610		return 0;
611
612	per_cpu(balloon_scratch_page, cpu) = alloc_page(GFP_KERNEL);
613	if (per_cpu(balloon_scratch_page, cpu) == NULL) {
614		pr_warn("Failed to allocate balloon_scratch_page for cpu %d\n", cpu);
615		return -ENOMEM;
616	}
617
618	return 0;
619}
620
 
 
621
622static int balloon_cpu_notify(struct notifier_block *self,
623				    unsigned long action, void *hcpu)
624{
625	int cpu = (long)hcpu;
626	switch (action) {
627	case CPU_UP_PREPARE:
628		if (alloc_balloon_scratch_page(cpu))
629			return NOTIFY_BAD;
630		break;
631	default:
632		break;
633	}
634	return NOTIFY_OK;
635}
636
637static struct notifier_block balloon_cpu_notifier = {
638	.notifier_call	= balloon_cpu_notify,
639};
640
641static int __init balloon_init(void)
642{
643	int i, cpu;
644
645	if (!xen_domain())
646		return -ENODEV;
647
648	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
649		register_cpu_notifier(&balloon_cpu_notifier);
650
651		get_online_cpus();
652		for_each_online_cpu(cpu) {
653			if (alloc_balloon_scratch_page(cpu)) {
654				put_online_cpus();
655				unregister_cpu_notifier(&balloon_cpu_notifier);
656				return -ENOMEM;
657			}
658		}
659		put_online_cpus();
660	}
661
662	pr_info("Initialising balloon driver\n");
663
 
664	balloon_stats.current_pages = xen_pv_domain()
665		? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
666		: get_num_physpages();
 
 
 
667	balloon_stats.target_pages  = balloon_stats.current_pages;
668	balloon_stats.balloon_low   = 0;
669	balloon_stats.balloon_high  = 0;
 
670
671	balloon_stats.schedule_delay = 1;
672	balloon_stats.max_schedule_delay = 32;
673	balloon_stats.retry_count = 1;
674	balloon_stats.max_retry_count = RETRY_UNLIMITED;
675
676#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
677	balloon_stats.hotplug_pages = 0;
678	balloon_stats.balloon_hotplug = 0;
679
680	set_online_page_callback(&xen_online_page);
681	register_memory_notifier(&xen_memory_nb);
 
682#endif
683
684	/*
685	 * Initialize the balloon with pages from the extra memory
686	 * regions (see arch/x86/xen/setup.c).
687	 */
688	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
689		if (xen_extra_mem[i].size)
690			balloon_add_region(PFN_UP(xen_extra_mem[i].start),
691					   PFN_DOWN(xen_extra_mem[i].size));
 
 
692
693	return 0;
694}
695
696subsys_initcall(balloon_init);
697
698static int __init balloon_clear(void)
699{
700	int cpu;
 
701
702	for_each_possible_cpu(cpu)
703		per_cpu(balloon_scratch_page, cpu) = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
704
705	return 0;
706}
707early_initcall(balloon_clear);
708
709MODULE_LICENSE("GPL");