xref: /linux/mm/balloon.c (revision 6aacab308a5dfd222b2d23662bbae60c11007cfb)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Common interface for implementing a memory balloon, including support
4  * for migration of pages inflated in a memory balloon.
5  *
6  * Copyright (C) 2012, Red Hat, Inc.  Rafael Aquini <aquini@redhat.com>
7  */
8 #include <linux/mm.h>
9 #include <linux/slab.h>
10 #include <linux/export.h>
11 #include <linux/balloon.h>
12 
13 /*
14  * Lock protecting the balloon_dev_info of all devices. We don't really
15  * expect more than one device.
16  */
17 static DEFINE_SPINLOCK(balloon_pages_lock);
18 
19 /**
20  * balloon_page_insert - insert a page into the balloon's page list and make
21  *			 the page->private assignment accordingly.
22  * @balloon : pointer to balloon device
23  * @page    : page to be assigned as a 'balloon page'
24  *
25  * Caller must ensure the balloon_pages_lock is held.
26  */
27 static void balloon_page_insert(struct balloon_dev_info *balloon,
28 				       struct page *page)
29 {
30 	lockdep_assert_held(&balloon_pages_lock);
31 	__SetPageOffline(page);
32 	if (IS_ENABLED(CONFIG_BALLOON_MIGRATION)) {
33 		SetPageMovableOps(page);
34 		set_page_private(page, (unsigned long)balloon);
35 	}
36 	list_add(&page->lru, &balloon->pages);
37 }
38 
39 /**
40  * balloon_page_finalize - prepare a balloon page that was removed from the
41  *			   balloon list for release to the page allocator
42  * @page: page to be released to the page allocator
43  *
44  * Caller must ensure the balloon_pages_lock is held.
45  */
46 static void balloon_page_finalize(struct page *page)
47 {
48 	lockdep_assert_held(&balloon_pages_lock);
49 	if (IS_ENABLED(CONFIG_BALLOON_MIGRATION))
50 		set_page_private(page, 0);
51 	/* PageOffline is sticky until the page is freed to the buddy. */
52 }
53 
54 static void balloon_page_enqueue_one(struct balloon_dev_info *b_dev_info,
55 				     struct page *page)
56 {
57 	balloon_page_insert(b_dev_info, page);
58 	if (b_dev_info->adjust_managed_page_count)
59 		adjust_managed_page_count(page, -1);
60 	__count_vm_event(BALLOON_INFLATE);
61 	inc_node_page_state(page, NR_BALLOON_PAGES);
62 }
63 
64 /**
65  * balloon_page_list_enqueue() - inserts a list of pages into the balloon page
66  *				 list.
67  * @b_dev_info: balloon device descriptor where we will insert a new page to
68  * @pages: pages to enqueue - allocated using balloon_page_alloc.
69  *
70  * Driver must call this function to properly enqueue balloon pages before
71  * definitively removing them from the guest system.
72  *
73  * Return: number of pages that were enqueued.
74  */
75 size_t balloon_page_list_enqueue(struct balloon_dev_info *b_dev_info,
76 				 struct list_head *pages)
77 {
78 	struct page *page, *tmp;
79 	unsigned long flags;
80 	size_t n_pages = 0;
81 
82 	spin_lock_irqsave(&balloon_pages_lock, flags);
83 	list_for_each_entry_safe(page, tmp, pages, lru) {
84 		list_del(&page->lru);
85 		balloon_page_enqueue_one(b_dev_info, page);
86 		n_pages++;
87 	}
88 	spin_unlock_irqrestore(&balloon_pages_lock, flags);
89 	return n_pages;
90 }
91 EXPORT_SYMBOL_GPL(balloon_page_list_enqueue);
92 
93 /**
94  * balloon_page_list_dequeue() - removes pages from balloon's page list and
95  *				 returns a list of the pages.
96  * @b_dev_info: balloon device descriptor where we will grab a page from.
97  * @pages: pointer to the list of pages that would be returned to the caller.
98  * @n_req_pages: number of requested pages.
99  *
100  * Driver must call this function to properly de-allocate a previous enlisted
101  * balloon pages before definitively releasing it back to the guest system.
102  * This function tries to remove @n_req_pages from the ballooned pages and
103  * return them to the caller in the @pages list.
104  *
105  * Note that this function may fail to dequeue some pages even if the balloon
106  * isn't empty - since the page list can be temporarily empty due to compaction
107  * of isolated pages.
108  *
109  * Return: number of pages that were added to the @pages list.
110  */
111 size_t balloon_page_list_dequeue(struct balloon_dev_info *b_dev_info,
112 				 struct list_head *pages, size_t n_req_pages)
113 {
114 	struct page *page, *tmp;
115 	unsigned long flags;
116 	size_t n_pages = 0;
117 
118 	spin_lock_irqsave(&balloon_pages_lock, flags);
119 	list_for_each_entry_safe(page, tmp, &b_dev_info->pages, lru) {
120 		if (n_pages == n_req_pages)
121 			break;
122 		list_del(&page->lru);
123 		if (b_dev_info->adjust_managed_page_count)
124 			adjust_managed_page_count(page, 1);
125 		balloon_page_finalize(page);
126 		__count_vm_event(BALLOON_DEFLATE);
127 		list_add(&page->lru, pages);
128 		dec_node_page_state(page, NR_BALLOON_PAGES);
129 		n_pages++;
130 	}
131 	spin_unlock_irqrestore(&balloon_pages_lock, flags);
132 
133 	return n_pages;
134 }
135 EXPORT_SYMBOL_GPL(balloon_page_list_dequeue);
136 
137 /**
138  * balloon_page_alloc - allocates a new page for insertion into the balloon
139  *			page list.
140  *
141  * Driver must call this function to properly allocate a new balloon page.
142  * Driver must call balloon_page_enqueue before definitively removing the page
143  * from the guest system.
144  *
145  * Return: struct page for the allocated page or NULL on allocation failure.
146  */
147 struct page *balloon_page_alloc(void)
148 {
149 	gfp_t gfp_flags = __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
150 
151 	if (IS_ENABLED(CONFIG_BALLOON_MIGRATION))
152 		gfp_flags |= GFP_HIGHUSER_MOVABLE;
153 	else
154 		gfp_flags |= GFP_HIGHUSER;
155 
156 	return alloc_page(gfp_flags);
157 }
158 EXPORT_SYMBOL_GPL(balloon_page_alloc);
159 
160 /**
161  * balloon_page_enqueue - inserts a new page into the balloon page list.
162  *
163  * @b_dev_info: balloon device descriptor where we will insert a new page
164  * @page: new page to enqueue - allocated using balloon_page_alloc.
165  *
166  * Drivers must call this function to properly enqueue a new allocated balloon
167  * page before definitively removing the page from the guest system.
168  *
169  * Drivers must not enqueue pages while page->lru is still in
170  * use, and must not use page->lru until a page was unqueued again.
171  */
172 void balloon_page_enqueue(struct balloon_dev_info *b_dev_info,
173 			  struct page *page)
174 {
175 	unsigned long flags;
176 
177 	spin_lock_irqsave(&balloon_pages_lock, flags);
178 	balloon_page_enqueue_one(b_dev_info, page);
179 	spin_unlock_irqrestore(&balloon_pages_lock, flags);
180 }
181 EXPORT_SYMBOL_GPL(balloon_page_enqueue);
182 
183 /**
184  * balloon_page_dequeue - removes a page from balloon's page list and returns
185  *			  its address to allow the driver to release the page.
186  * @b_dev_info: balloon device descriptor where we will grab a page from.
187  *
188  * Driver must call this function to properly dequeue a previously enqueued page
189  * before definitively releasing it back to the guest system.
190  *
191  * Caller must perform its own accounting to ensure that this
192  * function is called only if some pages are actually enqueued.
193  *
194  * Note that this function may fail to dequeue some pages even if there are
195  * some enqueued pages - since the page list can be temporarily empty due to
196  * the compaction of isolated pages.
197  *
198  * TODO: remove the caller accounting requirements, and allow caller to wait
199  * until all pages can be dequeued.
200  *
201  * Return: struct page for the dequeued page, or NULL if no page was dequeued.
202  */
203 struct page *balloon_page_dequeue(struct balloon_dev_info *b_dev_info)
204 {
205 	unsigned long flags;
206 	LIST_HEAD(pages);
207 	int n_pages;
208 
209 	n_pages = balloon_page_list_dequeue(b_dev_info, &pages, 1);
210 
211 	if (n_pages != 1) {
212 		/*
213 		 * If we are unable to dequeue a balloon page because the page
214 		 * list is empty and there are no isolated pages, then something
215 		 * went out of track and some balloon pages are lost.
216 		 * BUG() here, otherwise the balloon driver may get stuck in
217 		 * an infinite loop while attempting to release all its pages.
218 		 */
219 		spin_lock_irqsave(&balloon_pages_lock, flags);
220 		if (unlikely(list_empty(&b_dev_info->pages) &&
221 			     !b_dev_info->isolated_pages))
222 			BUG();
223 		spin_unlock_irqrestore(&balloon_pages_lock, flags);
224 		return NULL;
225 	}
226 	return list_first_entry(&pages, struct page, lru);
227 }
228 EXPORT_SYMBOL_GPL(balloon_page_dequeue);
229 
230 #ifdef CONFIG_BALLOON_MIGRATION
231 static struct balloon_dev_info *balloon_page_device(struct page *page)
232 {
233 	return (struct balloon_dev_info *)page_private(page);
234 }
235 
236 static bool balloon_page_isolate(struct page *page, isolate_mode_t mode)
237 
238 {
239 	struct balloon_dev_info *b_dev_info;
240 	unsigned long flags;
241 
242 	spin_lock_irqsave(&balloon_pages_lock, flags);
243 	b_dev_info = balloon_page_device(page);
244 	if (!b_dev_info) {
245 		/*
246 		 * The page already got deflated and removed from the
247 		 * balloon list.
248 		 */
249 		spin_unlock_irqrestore(&balloon_pages_lock, flags);
250 		return false;
251 	}
252 	list_del(&page->lru);
253 	b_dev_info->isolated_pages++;
254 	spin_unlock_irqrestore(&balloon_pages_lock, flags);
255 
256 	return true;
257 }
258 
259 static void balloon_page_putback(struct page *page)
260 {
261 	struct balloon_dev_info *b_dev_info = balloon_page_device(page);
262 	unsigned long flags;
263 
264 	/*
265 	 * When we isolated the page, the page was still inflated in a balloon
266 	 * device. As isolated balloon pages cannot get deflated, we still have
267 	 * a balloon device here.
268 	 */
269 	if (WARN_ON_ONCE(!b_dev_info))
270 		return;
271 
272 	spin_lock_irqsave(&balloon_pages_lock, flags);
273 	list_add(&page->lru, &b_dev_info->pages);
274 	b_dev_info->isolated_pages--;
275 	spin_unlock_irqrestore(&balloon_pages_lock, flags);
276 }
277 
278 static int balloon_page_migrate(struct page *newpage, struct page *page,
279 		enum migrate_mode mode)
280 {
281 	struct balloon_dev_info *b_dev_info = balloon_page_device(page);
282 	unsigned long flags;
283 	int rc;
284 
285 	/*
286 	 * When we isolated the page, the page was still inflated in a balloon
287 	 * device. As isolated balloon pages cannot get deflated, we still have
288 	 * a balloon device here.
289 	 */
290 	if (WARN_ON_ONCE(!b_dev_info))
291 		return -EAGAIN;
292 
293 	rc = b_dev_info->migratepage(b_dev_info, newpage, page, mode);
294 	if (rc < 0 && rc != -ENOENT)
295 		return rc;
296 
297 	spin_lock_irqsave(&balloon_pages_lock, flags);
298 	if (!rc) {
299 		/* Insert the new page into the balloon list. */
300 		get_page(newpage);
301 		balloon_page_insert(b_dev_info, newpage);
302 		__count_vm_event(BALLOON_MIGRATE);
303 
304 		if (b_dev_info->adjust_managed_page_count &&
305 		    page_zone(page) != page_zone(newpage)) {
306 			/*
307 			 * When we migrate a page to a different zone we
308 			 * have to fixup the count of both involved zones.
309 			 */
310 			adjust_managed_page_count(page, 1);
311 			adjust_managed_page_count(newpage, -1);
312 		}
313 	} else {
314 		/* Old page was deflated but new page not inflated. */
315 		__count_vm_event(BALLOON_DEFLATE);
316 
317 		if (b_dev_info->adjust_managed_page_count)
318 			adjust_managed_page_count(page, 1);
319 	}
320 
321 	b_dev_info->isolated_pages--;
322 
323 	/* Free the now-deflated page we isolated in balloon_page_isolate(). */
324 	balloon_page_finalize(page);
325 	spin_unlock_irqrestore(&balloon_pages_lock, flags);
326 
327 	put_page(page);
328 
329 	return 0;
330 }
331 
332 static const struct movable_operations balloon_mops = {
333 	.migrate_page = balloon_page_migrate,
334 	.isolate_page = balloon_page_isolate,
335 	.putback_page = balloon_page_putback,
336 };
337 
338 static int __init balloon_init(void)
339 {
340 	return set_movable_ops(&balloon_mops, PGTY_offline);
341 }
342 core_initcall(balloon_init);
343 
344 #endif /* CONFIG_BALLOON_MIGRATION */
345