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);
v4.10.11
  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/mm.h>
 46#include <linux/bootmem.h>
 47#include <linux/pagemap.h>
 48#include <linux/highmem.h>
 49#include <linux/mutex.h>
 50#include <linux/list.h>
 51#include <linux/gfp.h>
 52#include <linux/notifier.h>
 53#include <linux/memory.h>
 54#include <linux/memory_hotplug.h>
 55#include <linux/percpu-defs.h>
 56#include <linux/slab.h>
 57#include <linux/sysctl.h>
 
 
 58
 59#include <asm/page.h>
 60#include <asm/pgalloc.h>
 61#include <asm/pgtable.h>
 62#include <asm/tlb.h>
 63
 64#include <asm/xen/hypervisor.h>
 65#include <asm/xen/hypercall.h>
 66
 67#include <xen/xen.h>
 68#include <xen/interface/xen.h>
 69#include <xen/interface/memory.h>
 70#include <xen/balloon.h>
 71#include <xen/features.h>
 72#include <xen/page.h>
 
 73
 74static int xen_hotplug_unpopulated;
 
 
 
 
 75
 76#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
 77
 78static int zero;
 79static int one = 1;
 80
 81static struct ctl_table balloon_table[] = {
 82	{
 83		.procname	= "hotplug_unpopulated",
 84		.data		= &xen_hotplug_unpopulated,
 85		.maxlen		= sizeof(int),
 86		.mode		= 0644,
 87		.proc_handler	= proc_dointvec_minmax,
 88		.extra1         = &zero,
 89		.extra2         = &one,
 90	},
 91	{ }
 92};
 93
 94static struct ctl_table balloon_root[] = {
 95	{
 96		.procname	= "balloon",
 97		.mode		= 0555,
 98		.child		= balloon_table,
 99	},
100	{ }
101};
102
103static struct ctl_table xen_root[] = {
104	{
105		.procname	= "xen",
106		.mode		= 0555,
107		.child		= balloon_root,
108	},
109	{ }
110};
111
 
 
112#endif
113
114/*
115 * Use one extent per PAGE_SIZE to avoid to break down the page into
116 * multiple frame.
117 */
118#define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
119
120/*
121 * balloon_process() state:
122 *
123 * BP_DONE: done or nothing to do,
124 * BP_WAIT: wait to be rescheduled,
125 * BP_EAGAIN: error, go to sleep,
126 * BP_ECANCELED: error, balloon operation canceled.
127 */
128
129enum bp_state {
130	BP_DONE,
131	BP_WAIT,
132	BP_EAGAIN,
133	BP_ECANCELED
134};
135
 
 
136
137static DEFINE_MUTEX(balloon_mutex);
138
139struct balloon_stats balloon_stats;
140EXPORT_SYMBOL_GPL(balloon_stats);
141
142/* We increase/decrease in batches which fit in a page */
143static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
144
145
146/* List of ballooned pages, threaded through the mem_map array. */
147static LIST_HEAD(ballooned_pages);
148static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
149
150/* Main work function, always executed in process context. */
151static void balloon_process(struct work_struct *work);
152static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
153
154/* When ballooning out (allocating memory to return to Xen) we don't really
155   want the kernel to try too hard since that can trigger the oom killer. */
156#define GFP_BALLOON \
157	(GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
158
159static void scrub_page(struct page *page)
 
160{
161#ifdef CONFIG_XEN_SCRUB_PAGES
162	clear_highpage(page);
163#endif
164}
165
166/* balloon_append: add the given page to the balloon. */
167static void __balloon_append(struct page *page)
168{
169	/* Lowmem is re-populated first, so highmem pages go at list tail. */
170	if (PageHighMem(page)) {
171		list_add_tail(&page->lru, &ballooned_pages);
172		balloon_stats.balloon_high++;
173	} else {
174		list_add(&page->lru, &ballooned_pages);
175		balloon_stats.balloon_low++;
176	}
177	wake_up(&balloon_wq);
178}
179
180static void balloon_append(struct page *page)
181{
182	__balloon_append(page);
183}
184
185/* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
186static struct page *balloon_retrieve(bool require_lowmem)
187{
188	struct page *page;
189
190	if (list_empty(&ballooned_pages))
191		return NULL;
192
193	page = list_entry(ballooned_pages.next, struct page, lru);
194	if (require_lowmem && PageHighMem(page))
195		return NULL;
196	list_del(&page->lru);
197
198	if (PageHighMem(page))
199		balloon_stats.balloon_high--;
200	else
201		balloon_stats.balloon_low--;
202
 
203	return page;
204}
205
206static struct page *balloon_next_page(struct page *page)
207{
208	struct list_head *next = page->lru.next;
209	if (next == &ballooned_pages)
210		return NULL;
211	return list_entry(next, struct page, lru);
212}
213
214static enum bp_state update_schedule(enum bp_state state)
215{
216	if (state == BP_WAIT)
217		return BP_WAIT;
218
219	if (state == BP_ECANCELED)
220		return BP_ECANCELED;
221
222	if (state == BP_DONE) {
223		balloon_stats.schedule_delay = 1;
224		balloon_stats.retry_count = 1;
225		return BP_DONE;
226	}
227
228	++balloon_stats.retry_count;
229
230	if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
231			balloon_stats.retry_count > balloon_stats.max_retry_count) {
232		balloon_stats.schedule_delay = 1;
233		balloon_stats.retry_count = 1;
234		return BP_ECANCELED;
 
235	}
236
237	balloon_stats.schedule_delay <<= 1;
238
239	if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
240		balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
241
242	return BP_EAGAIN;
243}
244
245#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
246static void release_memory_resource(struct resource *resource)
247{
248	if (!resource)
249		return;
250
251	/*
252	 * No need to reset region to identity mapped since we now
253	 * know that no I/O can be in this region
254	 */
255	release_resource(resource);
256	kfree(resource);
257}
258
259static struct resource *additional_memory_resource(phys_addr_t size)
260{
261	struct resource *res;
262	int ret;
263
264	res = kzalloc(sizeof(*res), GFP_KERNEL);
265	if (!res)
266		return NULL;
267
268	res->name = "System RAM";
269	res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
270
271	ret = allocate_resource(&iomem_resource, res,
272				size, 0, -1,
273				PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
274	if (ret < 0) {
275		pr_err("Cannot allocate new System RAM resource\n");
276		kfree(res);
277		return NULL;
278	}
279
280#ifdef CONFIG_SPARSEMEM
281	{
282		unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
283		unsigned long pfn = res->start >> PAGE_SHIFT;
284
285		if (pfn > limit) {
286			pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
287			       pfn, limit);
288			release_memory_resource(res);
289			return NULL;
290		}
291	}
292#endif
293
294	return res;
295}
296
297static enum bp_state reserve_additional_memory(void)
298{
299	long credit;
300	struct resource *resource;
301	int nid, rc;
302	unsigned long balloon_hotplug;
303
304	credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
305		- balloon_stats.total_pages;
306
307	/*
308	 * Already hotplugged enough pages?  Wait for them to be
309	 * onlined.
310	 */
311	if (credit <= 0)
312		return BP_WAIT;
313
314	balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
315
316	resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
317	if (!resource)
318		goto err;
319
320	nid = memory_add_physaddr_to_nid(resource->start);
321
322#ifdef CONFIG_XEN_HAVE_PVMMU
323	/*
324	 * We don't support PV MMU when Linux and Xen is using
325	 * different page granularity.
326	 */
327	BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
328
329        /*
330         * add_memory() will build page tables for the new memory so
331         * the p2m must contain invalid entries so the correct
332         * non-present PTEs will be written.
333         *
334         * If a failure occurs, the original (identity) p2m entries
335         * are not restored since this region is now known not to
336         * conflict with any devices.
337         */ 
338	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
339		unsigned long pfn, i;
340
341		pfn = PFN_DOWN(resource->start);
342		for (i = 0; i < balloon_hotplug; i++) {
343			if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
344				pr_warn("set_phys_to_machine() failed, no memory added\n");
345				goto err;
346			}
347                }
348	}
349#endif
350
351	/*
352	 * add_memory_resource() will call online_pages() which in its turn
353	 * will call xen_online_page() callback causing deadlock if we don't
354	 * release balloon_mutex here. Unlocking here is safe because the
355	 * callers drop the mutex before trying again.
356	 */
357	mutex_unlock(&balloon_mutex);
358	rc = add_memory_resource(nid, resource, memhp_auto_online);
 
 
 
359	mutex_lock(&balloon_mutex);
360
361	if (rc) {
362		pr_warn("Cannot add additional memory (%i)\n", rc);
363		goto err;
364	}
365
366	balloon_stats.total_pages += balloon_hotplug;
367
368	return BP_WAIT;
369  err:
370	release_memory_resource(resource);
371	return BP_ECANCELED;
372}
373
374static void xen_online_page(struct page *page)
375{
376	__online_page_set_limits(page);
 
 
377
 
378	mutex_lock(&balloon_mutex);
379
380	__balloon_append(page);
381
 
382	mutex_unlock(&balloon_mutex);
383}
384
385static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
386{
387	if (val == MEM_ONLINE)
388		schedule_delayed_work(&balloon_worker, 0);
389
390	return NOTIFY_OK;
391}
392
393static struct notifier_block xen_memory_nb = {
394	.notifier_call = xen_memory_notifier,
395	.priority = 0
396};
397#else
398static enum bp_state reserve_additional_memory(void)
399{
400	balloon_stats.target_pages = balloon_stats.current_pages;
 
401	return BP_ECANCELED;
402}
403#endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
404
405static long current_credit(void)
406{
407	return balloon_stats.target_pages - balloon_stats.current_pages;
408}
409
410static bool balloon_is_inflated(void)
411{
412	return balloon_stats.balloon_low || balloon_stats.balloon_high;
413}
414
415static enum bp_state increase_reservation(unsigned long nr_pages)
416{
417	int rc;
418	unsigned long i;
419	struct page   *page;
420	struct xen_memory_reservation reservation = {
421		.address_bits = 0,
422		.extent_order = EXTENT_ORDER,
423		.domid        = DOMID_SELF
424	};
425
426	if (nr_pages > ARRAY_SIZE(frame_list))
427		nr_pages = ARRAY_SIZE(frame_list);
428
429	page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
430	for (i = 0; i < nr_pages; i++) {
431		if (!page) {
432			nr_pages = i;
433			break;
434		}
435
436		/* XENMEM_populate_physmap requires a PFN based on Xen
437		 * granularity.
438		 */
439		frame_list[i] = page_to_xen_pfn(page);
440		page = balloon_next_page(page);
441	}
442
443	set_xen_guest_handle(reservation.extent_start, frame_list);
444	reservation.nr_extents = nr_pages;
445	rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
446	if (rc <= 0)
447		return BP_EAGAIN;
448
449	for (i = 0; i < rc; i++) {
450		page = balloon_retrieve(false);
451		BUG_ON(page == NULL);
452
453#ifdef CONFIG_XEN_HAVE_PVMMU
454		/*
455		 * We don't support PV MMU when Linux and Xen is using
456		 * different page granularity.
457		 */
458		BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
459
460		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
461			unsigned long pfn = page_to_pfn(page);
462
463			set_phys_to_machine(pfn, frame_list[i]);
464
465			/* Link back into the page tables if not highmem. */
466			if (!PageHighMem(page)) {
467				int ret;
468				ret = HYPERVISOR_update_va_mapping(
469						(unsigned long)__va(pfn << PAGE_SHIFT),
470						mfn_pte(frame_list[i], PAGE_KERNEL),
471						0);
472				BUG_ON(ret);
473			}
474		}
475#endif
476
477		/* Relinquish the page back to the allocator. */
478		free_reserved_page(page);
479	}
480
481	balloon_stats.current_pages += rc;
482
483	return BP_DONE;
484}
485
486static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
487{
488	enum bp_state state = BP_DONE;
489	unsigned long i;
490	struct page *page, *tmp;
491	int ret;
492	struct xen_memory_reservation reservation = {
493		.address_bits = 0,
494		.extent_order = EXTENT_ORDER,
495		.domid        = DOMID_SELF
496	};
497	LIST_HEAD(pages);
498
499	if (nr_pages > ARRAY_SIZE(frame_list))
500		nr_pages = ARRAY_SIZE(frame_list);
501
502	for (i = 0; i < nr_pages; i++) {
503		page = alloc_page(gfp);
504		if (page == NULL) {
505			nr_pages = i;
506			state = BP_EAGAIN;
507			break;
508		}
509		adjust_managed_page_count(page, -1);
510		scrub_page(page);
511		list_add(&page->lru, &pages);
512	}
513
514	/*
515	 * Ensure that ballooned highmem pages don't have kmaps.
516	 *
517	 * Do this before changing the p2m as kmap_flush_unused()
518	 * reads PTEs to obtain pages (and hence needs the original
519	 * p2m entry).
520	 */
521	kmap_flush_unused();
522
523	/*
524	 * Setup the frame, update direct mapping, invalidate P2M,
525	 * and add to balloon.
526	 */
527	i = 0;
528	list_for_each_entry_safe(page, tmp, &pages, lru) {
529		/* XENMEM_decrease_reservation requires a GFN */
530		frame_list[i++] = xen_page_to_gfn(page);
531
532#ifdef CONFIG_XEN_HAVE_PVMMU
533		/*
534		 * We don't support PV MMU when Linux and Xen is using
535		 * different page granularity.
536		 */
537		BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
538
539		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
540			unsigned long pfn = page_to_pfn(page);
541
542			if (!PageHighMem(page)) {
543				ret = HYPERVISOR_update_va_mapping(
544						(unsigned long)__va(pfn << PAGE_SHIFT),
545						__pte_ma(0), 0);
546				BUG_ON(ret);
547			}
548			__set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
549		}
550#endif
551		list_del(&page->lru);
552
553		balloon_append(page);
554	}
555
556	flush_tlb_all();
557
558	set_xen_guest_handle(reservation.extent_start, frame_list);
559	reservation.nr_extents   = nr_pages;
560	ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
561	BUG_ON(ret != nr_pages);
562
563	balloon_stats.current_pages -= nr_pages;
564
565	return state;
566}
567
568/*
569 * As this is a work item it is guaranteed to run as a single instance only.
 
 
 
 
 
 
 
 
 
 
 
 
570 * We may of course race updates of the target counts (which are protected
571 * by the balloon lock), or with changes to the Xen hard limit, but we will
572 * recover from these in time.
573 */
574static void balloon_process(struct work_struct *work)
575{
576	enum bp_state state = BP_DONE;
577	long credit;
 
578
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
579
580	do {
581		mutex_lock(&balloon_mutex);
582
583		credit = current_credit();
584
585		if (credit > 0) {
586			if (balloon_is_inflated())
587				state = increase_reservation(credit);
588			else
589				state = reserve_additional_memory();
590		}
591
592		if (credit < 0)
593			state = decrease_reservation(-credit, GFP_BALLOON);
 
 
 
 
 
 
 
 
594
595		state = update_schedule(state);
596
597		mutex_unlock(&balloon_mutex);
598
599		cond_resched();
600
601	} while (credit && state == BP_DONE);
602
603	/* Schedule more work if there is some still to be done. */
604	if (state == BP_EAGAIN)
605		schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
606}
607
608/* Resets the Xen limit, sets new target, and kicks off processing. */
609void balloon_set_new_target(unsigned long target)
610{
611	/* No need for lock. Not read-modify-write updates. */
612	balloon_stats.target_pages = target;
613	schedule_delayed_work(&balloon_worker, 0);
614}
615EXPORT_SYMBOL_GPL(balloon_set_new_target);
616
617static int add_ballooned_pages(int nr_pages)
618{
619	enum bp_state st;
620
621	if (xen_hotplug_unpopulated) {
622		st = reserve_additional_memory();
623		if (st != BP_ECANCELED) {
 
 
624			mutex_unlock(&balloon_mutex);
625			wait_event(balloon_wq,
626				   !list_empty(&ballooned_pages));
627			mutex_lock(&balloon_mutex);
628			return 0;
629		}
630	}
631
 
 
 
632	st = decrease_reservation(nr_pages, GFP_USER);
633	if (st != BP_DONE)
634		return -ENOMEM;
635
636	return 0;
637}
638
639/**
640 * alloc_xenballooned_pages - get pages that have been ballooned out
641 * @nr_pages: Number of pages to get
642 * @pages: pages returned
643 * @return 0 on success, error otherwise
644 */
645int alloc_xenballooned_pages(int nr_pages, struct page **pages)
646{
647	int pgno = 0;
648	struct page *page;
649	int ret;
650
651	mutex_lock(&balloon_mutex);
652
653	balloon_stats.target_unpopulated += nr_pages;
654
655	while (pgno < nr_pages) {
656		page = balloon_retrieve(true);
657		if (page) {
658			pages[pgno++] = page;
659#ifdef CONFIG_XEN_HAVE_PVMMU
660			/*
661			 * We don't support PV MMU when Linux and Xen is using
662			 * different page granularity.
663			 */
664			BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
665
666			ret = xen_alloc_p2m_entry(page_to_pfn(page));
667			if (ret < 0)
668				goto out_undo;
 
 
669#endif
670		} else {
671			ret = add_ballooned_pages(nr_pages - pgno);
672			if (ret < 0)
673				goto out_undo;
674		}
675	}
676	mutex_unlock(&balloon_mutex);
677	return 0;
678 out_undo:
679	mutex_unlock(&balloon_mutex);
680	free_xenballooned_pages(pgno, pages);
 
 
 
 
 
 
681	return ret;
682}
683EXPORT_SYMBOL(alloc_xenballooned_pages);
684
685/**
686 * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
687 * @nr_pages: Number of pages
688 * @pages: pages to return
689 */
690void free_xenballooned_pages(int nr_pages, struct page **pages)
691{
692	int i;
693
694	mutex_lock(&balloon_mutex);
695
696	for (i = 0; i < nr_pages; i++) {
697		if (pages[i])
698			balloon_append(pages[i]);
699	}
700
701	balloon_stats.target_unpopulated -= nr_pages;
702
703	/* The balloon may be too large now. Shrink it if needed. */
704	if (current_credit())
705		schedule_delayed_work(&balloon_worker, 0);
706
707	mutex_unlock(&balloon_mutex);
708}
709EXPORT_SYMBOL(free_xenballooned_pages);
710
711static void __init balloon_add_region(unsigned long start_pfn,
712				      unsigned long pages)
713{
 
 
714	unsigned long pfn, extra_pfn_end;
715	struct page *page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
716
717	/*
718	 * If the amount of usable memory has been limited (e.g., with
719	 * the 'mem' command line parameter), don't add pages beyond
720	 * this limit.
721	 */
722	extra_pfn_end = min(max_pfn, start_pfn + pages);
723
724	for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
725		page = pfn_to_page(pfn);
726		/* totalram_pages and totalhigh_pages do not
727		   include the boot-time balloon extension, so
728		   don't subtract from it. */
729		__balloon_append(page);
730	}
731
732	balloon_stats.total_pages += extra_pfn_end - start_pfn;
733}
734
735static int __init balloon_init(void)
736{
737	int i;
738
739	if (!xen_domain())
740		return -ENODEV;
741
742	pr_info("Initialising balloon driver\n");
743
 
744	balloon_stats.current_pages = xen_pv_domain()
745		? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
746		: get_num_physpages();
 
 
 
747	balloon_stats.target_pages  = balloon_stats.current_pages;
748	balloon_stats.balloon_low   = 0;
749	balloon_stats.balloon_high  = 0;
750	balloon_stats.total_pages   = balloon_stats.current_pages;
751
752	balloon_stats.schedule_delay = 1;
753	balloon_stats.max_schedule_delay = 32;
754	balloon_stats.retry_count = 1;
755	balloon_stats.max_retry_count = RETRY_UNLIMITED;
756
757#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
758	set_online_page_callback(&xen_online_page);
759	register_memory_notifier(&xen_memory_nb);
760	register_sysctl_table(xen_root);
761#endif
762
763	/*
764	 * Initialize the balloon with pages from the extra memory
765	 * regions (see arch/x86/xen/setup.c).
766	 */
767	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
768		if (xen_extra_mem[i].n_pfns)
769			balloon_add_region(xen_extra_mem[i].start_pfn,
770					   xen_extra_mem[i].n_pfns);
 
 
771
772	return 0;
773}
774subsys_initcall(balloon_init);