1/*
  2 * mm/balloon_compaction.c
  3 *
  4 * Common interface for making balloon pages movable by compaction.
  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_compaction.h>
 12
 13/*
 14 * balloon_devinfo_alloc - allocates a balloon device information descriptor.
 15 * @balloon_dev_descriptor: pointer to reference the balloon device which
 16 *                          this struct balloon_dev_info will be servicing.
 17 *
 18 * Driver must call it to properly allocate and initialize an instance of
 19 * struct balloon_dev_info which will be used to reference a balloon device
 20 * as well as to keep track of the balloon device page list.
 21 */
 22struct balloon_dev_info *balloon_devinfo_alloc(void *balloon_dev_descriptor)
 23{
 24	struct balloon_dev_info *b_dev_info;
 25	b_dev_info = kmalloc(sizeof(*b_dev_info), GFP_KERNEL);
 26	if (!b_dev_info)
 27		return ERR_PTR(-ENOMEM);
 28
 29	b_dev_info->balloon_device = balloon_dev_descriptor;
 30	b_dev_info->mapping = NULL;
 31	b_dev_info->isolated_pages = 0;
 32	spin_lock_init(&b_dev_info->pages_lock);
 33	INIT_LIST_HEAD(&b_dev_info->pages);
 34
 35	return b_dev_info;
 36}
 37EXPORT_SYMBOL_GPL(balloon_devinfo_alloc);
 38
 39/*
 40 * balloon_page_enqueue - allocates a new page and inserts it into the balloon
 41 *			  page list.
 42 * @b_dev_info: balloon device decriptor where we will insert a new page to
 43 *
 44 * Driver must call it to properly allocate a new enlisted balloon page
 45 * before definetively removing it from the guest system.
 46 * This function returns the page address for the recently enqueued page or
 47 * NULL in the case we fail to allocate a new page this turn.
 48 */
 49struct page *balloon_page_enqueue(struct balloon_dev_info *b_dev_info)
 50{
 51	unsigned long flags;
 52	struct page *page = alloc_page(balloon_mapping_gfp_mask() |
 53					__GFP_NOMEMALLOC | __GFP_NORETRY);
 54	if (!page)
 55		return NULL;
 56
 57	/*
 58	 * Block others from accessing the 'page' when we get around to
 59	 * establishing additional references. We should be the only one
 60	 * holding a reference to the 'page' at this point.
 61	 */
 62	BUG_ON(!trylock_page(page));
 63	spin_lock_irqsave(&b_dev_info->pages_lock, flags);
 64	balloon_page_insert(page, b_dev_info->mapping, &b_dev_info->pages);
 65	spin_unlock_irqrestore(&b_dev_info->pages_lock, flags);
 66	unlock_page(page);
 67	return page;
 68}
 69EXPORT_SYMBOL_GPL(balloon_page_enqueue);
 70
 71/*
 72 * balloon_page_dequeue - removes a page from balloon's page list and returns
 73 *			  the its address to allow the driver release the page.
 74 * @b_dev_info: balloon device decriptor where we will grab a page from.
 75 *
 76 * Driver must call it to properly de-allocate a previous enlisted balloon page
 77 * before definetively releasing it back to the guest system.
 78 * This function returns the page address for the recently dequeued page or
 79 * NULL in the case we find balloon's page list temporarily empty due to
 80 * compaction isolated pages.
 81 */
 82struct page *balloon_page_dequeue(struct balloon_dev_info *b_dev_info)
 83{
 84	struct page *page, *tmp;
 85	unsigned long flags;
 86	bool dequeued_page;
 87
 88	dequeued_page = false;
 89	list_for_each_entry_safe(page, tmp, &b_dev_info->pages, lru) {
 90		/*
 91		 * Block others from accessing the 'page' while we get around
 92		 * establishing additional references and preparing the 'page'
 93		 * to be released by the balloon driver.
 94		 */
 95		if (trylock_page(page)) {
 96			spin_lock_irqsave(&b_dev_info->pages_lock, flags);
 97			/*
 98			 * Raise the page refcount here to prevent any wrong
 99			 * attempt to isolate this page, in case of coliding
100			 * with balloon_page_isolate() just after we release
101			 * the page lock.
102			 *
103			 * balloon_page_free() will take care of dropping
104			 * this extra refcount later.
105			 */
106			get_page(page);
107			balloon_page_delete(page);
108			spin_unlock_irqrestore(&b_dev_info->pages_lock, flags);
109			unlock_page(page);
110			dequeued_page = true;
111			break;
112		}
113	}
114
115	if (!dequeued_page) {
116		/*
117		 * If we are unable to dequeue a balloon page because the page
118		 * list is empty and there is no isolated pages, then something
119		 * went out of track and some balloon pages are lost.
120		 * BUG() here, otherwise the balloon driver may get stuck into
121		 * an infinite loop while attempting to release all its pages.
122		 */
123		spin_lock_irqsave(&b_dev_info->pages_lock, flags);
124		if (unlikely(list_empty(&b_dev_info->pages) &&
125			     !b_dev_info->isolated_pages))
126			BUG();
127		spin_unlock_irqrestore(&b_dev_info->pages_lock, flags);
128		page = NULL;
129	}
130	return page;
131}
132EXPORT_SYMBOL_GPL(balloon_page_dequeue);
133
134#ifdef CONFIG_BALLOON_COMPACTION
135/*
136 * balloon_mapping_alloc - allocates a special ->mapping for ballooned pages.
137 * @b_dev_info: holds the balloon device information descriptor.
138 * @a_ops: balloon_mapping address_space_operations descriptor.
139 *
140 * Driver must call it to properly allocate and initialize an instance of
141 * struct address_space which will be used as the special page->mapping for
142 * balloon device enlisted page instances.
143 */
144struct address_space *balloon_mapping_alloc(struct balloon_dev_info *b_dev_info,
145				const struct address_space_operations *a_ops)
146{
147	struct address_space *mapping;
148
149	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
150	if (!mapping)
151		return ERR_PTR(-ENOMEM);
152
153	/*
154	 * Give a clean 'zeroed' status to all elements of this special
155	 * balloon page->mapping struct address_space instance.
156	 */
157	address_space_init_once(mapping);
158
159	/*
160	 * Set mapping->flags appropriately, to allow balloon pages
161	 * ->mapping identification.
162	 */
163	mapping_set_balloon(mapping);
164	mapping_set_gfp_mask(mapping, balloon_mapping_gfp_mask());
165
166	/* balloon's page->mapping->a_ops callback descriptor */
167	mapping->a_ops = a_ops;
168
169	/*
170	 * Establish a pointer reference back to the balloon device descriptor
171	 * this particular page->mapping will be servicing.
172	 * This is used by compaction / migration procedures to identify and
173	 * access the balloon device pageset while isolating / migrating pages.
174	 *
175	 * As some balloon drivers can register multiple balloon devices
176	 * for a single guest, this also helps compaction / migration to
177	 * properly deal with multiple balloon pagesets, when required.
178	 */
179	mapping->private_data = b_dev_info;
180	b_dev_info->mapping = mapping;
181
182	return mapping;
183}
184EXPORT_SYMBOL_GPL(balloon_mapping_alloc);
185
186static inline void __isolate_balloon_page(struct page *page)
187{
188	struct balloon_dev_info *b_dev_info = page->mapping->private_data;
189	unsigned long flags;
190	spin_lock_irqsave(&b_dev_info->pages_lock, flags);
191	list_del(&page->lru);
192	b_dev_info->isolated_pages++;
193	spin_unlock_irqrestore(&b_dev_info->pages_lock, flags);
194}
195
196static inline void __putback_balloon_page(struct page *page)
197{
198	struct balloon_dev_info *b_dev_info = page->mapping->private_data;
199	unsigned long flags;
200	spin_lock_irqsave(&b_dev_info->pages_lock, flags);
201	list_add(&page->lru, &b_dev_info->pages);
202	b_dev_info->isolated_pages--;
203	spin_unlock_irqrestore(&b_dev_info->pages_lock, flags);
204}
205
206static inline int __migrate_balloon_page(struct address_space *mapping,
207		struct page *newpage, struct page *page, enum migrate_mode mode)
208{
209	return page->mapping->a_ops->migratepage(mapping, newpage, page, mode);
210}
211
212/* __isolate_lru_page() counterpart for a ballooned page */
213bool balloon_page_isolate(struct page *page)
214{
215	/*
216	 * Avoid burning cycles with pages that are yet under __free_pages(),
217	 * or just got freed under us.
218	 *
219	 * In case we 'win' a race for a balloon page being freed under us and
220	 * raise its refcount preventing __free_pages() from doing its job
221	 * the put_page() at the end of this block will take care of
222	 * release this page, thus avoiding a nasty leakage.
223	 */
224	if (likely(get_page_unless_zero(page))) {
225		/*
226		 * As balloon pages are not isolated from LRU lists, concurrent
227		 * compaction threads can race against page migration functions
228		 * as well as race against the balloon driver releasing a page.
229		 *
230		 * In order to avoid having an already isolated balloon page
231		 * being (wrongly) re-isolated while it is under migration,
232		 * or to avoid attempting to isolate pages being released by
233		 * the balloon driver, lets be sure we have the page lock
234		 * before proceeding with the balloon page isolation steps.
235		 */
236		if (likely(trylock_page(page))) {
237			/*
238			 * A ballooned page, by default, has just one refcount.
239			 * Prevent concurrent compaction threads from isolating
240			 * an already isolated balloon page by refcount check.
241			 */
242			if (__is_movable_balloon_page(page) &&
243			    page_count(page) == 2) {
244				__isolate_balloon_page(page);
245				unlock_page(page);
246				return true;
247			}
248			unlock_page(page);
249		}
250		put_page(page);
251	}
252	return false;
253}
254
255/* putback_lru_page() counterpart for a ballooned page */
256void balloon_page_putback(struct page *page)
257{
258	/*
259	 * 'lock_page()' stabilizes the page and prevents races against
260	 * concurrent isolation threads attempting to re-isolate it.
261	 */
262	lock_page(page);
263
264	if (__is_movable_balloon_page(page)) {
265		__putback_balloon_page(page);
266		/* drop the extra ref count taken for page isolation */
267		put_page(page);
268	} else {
269		WARN_ON(1);
270		dump_page(page, "not movable balloon page");
271	}
272	unlock_page(page);
273}
274
275/* move_to_new_page() counterpart for a ballooned page */
276int balloon_page_migrate(struct page *newpage,
277			 struct page *page, enum migrate_mode mode)
278{
279	struct address_space *mapping;
280	int rc = -EAGAIN;
281
282	/*
283	 * Block others from accessing the 'newpage' when we get around to
284	 * establishing additional references. We should be the only one
285	 * holding a reference to the 'newpage' at this point.
286	 */
287	BUG_ON(!trylock_page(newpage));
288
289	if (WARN_ON(!__is_movable_balloon_page(page))) {
290		dump_page(page, "not movable balloon page");
291		unlock_page(newpage);
292		return rc;
293	}
294
295	mapping = page->mapping;
296	if (mapping)
297		rc = __migrate_balloon_page(mapping, newpage, page, mode);
298
299	unlock_page(newpage);
300	return rc;
301}
302#endif /* CONFIG_BALLOON_COMPACTION */