forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-10-09 244b2c5ca8b14627e4a17755e5922221e121c771 
 kernel/mm/page_isolation.c
....@@ -15,62 +15,30 @@
1515 #define CREATE_TRACE_POINTS
1616 #include <trace/events/page_isolation.h>
1717 
18
-static int set_migratetype_isolate(struct page *page, int migratetype,
19
-				bool skip_hwpoisoned_pages)
18
+static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags)
2019 {
21
-	struct zone *zone;
22
-	unsigned long flags, pfn;
23
-	struct memory_isolate_notify arg;
24
-	int notifier_ret;
25
-	int ret = -EBUSY;
26
-
27
-	zone = page_zone(page);
20
+	struct zone *zone = page_zone(page);
21
+	struct page *unmovable;
22
+	unsigned long flags;
2823 
2924 	spin_lock_irqsave(&zone->lock, flags);
3025 
3126 	/*
3227 	 * We assume the caller intended to SET migrate type to isolate.
3328 	 * If it is already set, then someone else must have raced and
34
-	 * set it before us.  Return -EBUSY
29
+	 * set it before us.
3530 	 */
36
-	if (is_migrate_isolate_page(page))
37
-		goto out;
31
+	if (is_migrate_isolate_page(page)) {
32
+		spin_unlock_irqrestore(&zone->lock, flags);
33
+		return -EBUSY;
34
+	}
3835 
39
-	pfn = page_to_pfn(page);
40
-	arg.start_pfn = pfn;
41
-	arg.nr_pages = pageblock_nr_pages;
42
-	arg.pages_found = 0;
43
-
44
-	/*
45
-	 * It may be possible to isolate a pageblock even if the
46
-	 * migratetype is not MIGRATE_MOVABLE. The memory isolation
47
-	 * notifier chain is used by balloon drivers to return the
48
-	 * number of pages in a range that are held by the balloon
49
-	 * driver to shrink memory. If all the pages are accounted for
50
-	 * by balloons, are free, or on the LRU, isolation can continue.
51
-	 * Later, for example, when memory hotplug notifier runs, these
52
-	 * pages reported as "can be isolated" should be isolated(freed)
53
-	 * by the balloon driver through the memory notifier chain.
54
-	 */
55
-	notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
56
-	notifier_ret = notifier_to_errno(notifier_ret);
57
-	if (notifier_ret)
58
-		goto out;
5936 	/*
6037 	 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
6138 	 * We just check MOVABLE pages.
6239 	 */
63
-	if (!has_unmovable_pages(zone, page, arg.pages_found, migratetype,
64
-				 skip_hwpoisoned_pages))
65
-		ret = 0;
66
-
67
-	/*
68
-	 * immobile means "not-on-lru" pages. If immobile is larger than
69
-	 * removable-by-driver pages reported by notifier, we'll fail.
70
-	 */
71
-
72
-out:
73
-	if (!ret) {
40
+	unmovable = has_unmovable_pages(zone, page, migratetype, isol_flags);
41
+	if (!unmovable) {
7442 		unsigned long nr_pages;
7543 		int mt = get_pageblock_migratetype(page);
7644 
....@@ -80,12 +48,20 @@
8048 									NULL);
8149 
8250 		__mod_zone_freepage_state(zone, -nr_pages, mt);
51
+		spin_unlock_irqrestore(&zone->lock, flags);
52
+		return 0;
8353 	}
8454 
8555 	spin_unlock_irqrestore(&zone->lock, flags);
86
-	if (!ret)
87
-		drain_all_pages(zone);
88
-	return ret;
56
+	if (isol_flags & REPORT_FAILURE) {
57
+		/*
58
+		 * printk() with zone->lock held will likely trigger a
59
+		 * lockdep splat, so defer it here.
60
+		 */
61
+		dump_page(unmovable, "unmovable page");
62
+	}
63
+
64
+	return -EBUSY;
8965 }
9066 
9167 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
....@@ -111,7 +87,7 @@
11187 	 * these pages to be merged.
11288 	 */
11389 	if (PageBuddy(page)) {
114
-		order = page_order(page);
90
+		order = buddy_order(page);
11591 		if (order >= pageblock_order) {
11692 			pfn = page_to_pfn(page);
11793 			buddy_pfn = __find_buddy_pfn(pfn, order);
....@@ -129,19 +105,22 @@
129105 	 * If we isolate freepage with more than pageblock_order, there
130106 	 * should be no freepage in the range, so we could avoid costly
131107 	 * pageblock scanning for freepage moving.
108
+	 *
109
+	 * We didn't actually touch any of the isolated pages, so place them
110
+	 * to the tail of the freelist. This is an optimization for memory
111
+	 * onlining - just onlined memory won't immediately be considered for
112
+	 * allocation.
132113 	 */
133114 	if (!isolated_page) {
134115 		nr_pages = move_freepages_block(zone, page, migratetype, NULL);
135116 		__mod_zone_freepage_state(zone, nr_pages, migratetype);
136117 	}
137118 	set_pageblock_migratetype(page, migratetype);
119
+	if (isolated_page)
120
+		__putback_isolated_page(page, order, migratetype);
138121 	zone->nr_isolate_pageblock--;
139122 out:
140123 	spin_unlock_irqrestore(&zone->lock, flags);
141
-	if (isolated_page) {
142
-		post_alloc_hook(page, order, __GFP_MOVABLE);
143
-		__free_pages(page, order);
144
-	}
145124 }
146125 
147126 static inline struct page *
....@@ -152,8 +131,6 @@
152131 	for (i = 0; i < nr_pages; i++) {
153132 		struct page *page;
154133 
155
-		if (!pfn_valid_within(pfn + i))
156
-			continue;
157134 		page = pfn_to_online_page(pfn + i);
158135 		if (!page)
159136 			continue;
....@@ -162,30 +139,50 @@
162139 	return NULL;
163140 }
164141 
165
-/*
166
- * start_isolate_page_range() -- make page-allocation-type of range of pages
167
- * to be MIGRATE_ISOLATE.
168
- * @start_pfn: The lower PFN of the range to be isolated.
169
- * @end_pfn: The upper PFN of the range to be isolated.
170
- * @migratetype: migrate type to set in error recovery.
142
+/**
143
+ * start_isolate_page_range() - make page-allocation-type of range of pages to
144
+ * be MIGRATE_ISOLATE.
145
+ * @start_pfn:		The lower PFN of the range to be isolated.
146
+ * @end_pfn:		The upper PFN of the range to be isolated.
147
+ *			start_pfn/end_pfn must be aligned to pageblock_order.
148
+ * @migratetype:	Migrate type to set in error recovery.
149
+ * @flags:		The following flags are allowed (they can be combined in
150
+ *			a bit mask)
151
+ *			MEMORY_OFFLINE - isolate to offline (!allocate) memory
152
+ *					 e.g., skip over PageHWPoison() pages
153
+ *					 and PageOffline() pages.
154
+ *			REPORT_FAILURE - report details about the failure to
155
+ *			isolate the range
171156  *
172157  * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
173158  * the range will never be allocated. Any free pages and pages freed in the
174
- * future will not be allocated again.
175
- *
176
- * start_pfn/end_pfn must be aligned to pageblock_order.
177
- * Return 0 on success and -EBUSY if any part of range cannot be isolated.
159
+ * future will not be allocated again. If specified range includes migrate types
160
+ * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
161
+ * pages in the range finally, the caller have to free all pages in the range.
162
+ * test_page_isolated() can be used for test it.
178163  *
179164  * There is no high level synchronization mechanism that prevents two threads
180
- * from trying to isolate overlapping ranges.  If this happens, one thread
165
+ * from trying to isolate overlapping ranges. If this happens, one thread
181166  * will notice pageblocks in the overlapping range already set to isolate.
182167  * This happens in set_migratetype_isolate, and set_migratetype_isolate
183
- * returns an error.  We then clean up by restoring the migration type on
184
- * pageblocks we may have modified and return -EBUSY to caller.  This
168
+ * returns an error. We then clean up by restoring the migration type on
169
+ * pageblocks we may have modified and return -EBUSY to caller. This
185170  * prevents two threads from simultaneously working on overlapping ranges.
171
+ *
172
+ * Please note that there is no strong synchronization with the page allocator
173
+ * either. Pages might be freed while their page blocks are marked ISOLATED.
174
+ * A call to drain_all_pages() after isolation can flush most of them. However
175
+ * in some cases pages might still end up on pcp lists and that would allow
176
+ * for their allocation even when they are in fact isolated already. Depending
177
+ * on how strong of a guarantee the caller needs, further drain_all_pages()
178
+ * might be needed (e.g. __offline_pages will need to call it after check for
179
+ * isolated range for a next retry).
180
+ *
181
+ * Return: 0 on success and -EBUSY if any part of range cannot be isolated.
186182  */
187183 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
188
-			     unsigned migratetype, bool skip_hwpoisoned_pages)
184
+			     unsigned migratetype, int flags,
185
+			     unsigned long *failed_pfn)
189186 {
190187 	unsigned long pfn;
191188 	unsigned long undo_pfn;
....@@ -198,10 +195,13 @@
198195 	     pfn < end_pfn;
199196 	     pfn += pageblock_nr_pages) {
200197 		page = __first_valid_page(pfn, pageblock_nr_pages);
201
-		if (page &&
202
-		    set_migratetype_isolate(page, migratetype, skip_hwpoisoned_pages)) {
203
-			undo_pfn = pfn;
204
-			goto undo;
198
+		if (page) {
199
+			if (set_migratetype_isolate(page, migratetype, flags)) {
200
+				undo_pfn = pfn;
201
+				if (failed_pfn)
202
+					*failed_pfn = page_to_pfn(page);
203
+				goto undo;
204
+			}
205205 		}
206206 	}
207207 	return 0;
....@@ -221,7 +221,7 @@
221221 /*
222222  * Make isolated pages available again.
223223  */
224
-int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
224
+void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
225225 			    unsigned migratetype)
226226 {
227227 	unsigned long pfn;
....@@ -238,7 +238,6 @@
238238 			continue;
239239 		unset_migratetype_isolate(page, migratetype);
240240 	}
241
-	return 0;
242241 }
243242 /*
244243  * Test all pages in the range is free(means isolated) or not.
....@@ -249,7 +248,7 @@
249248  */
250249 static unsigned long
251250 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
252
-				  bool skip_hwpoisoned_pages)
251
+				  int flags)
253252 {
254253 	struct page *page;
255254 
....@@ -265,14 +264,22 @@
265264 			 * the correct MIGRATE_ISOLATE freelist. There is no
266265 			 * simple way to verify that as VM_BUG_ON(), though.
267266 			 */
268
-			pfn += 1 << page_order(page);
269
-		else if (skip_hwpoisoned_pages && PageHWPoison(page))
267
+			pfn += 1 << buddy_order(page);
268
+		else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
270269 			/* A HWPoisoned page cannot be also PageBuddy */
270
+			pfn++;
271
+		else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
272
+			 !page_count(page))
273
+			/*
274
+			 * The responsible driver agreed to skip PageOffline()
275
+			 * pages when offlining memory by dropping its
276
+			 * reference in MEM_GOING_OFFLINE.
277
+			 */
271278 			pfn++;
272279 		else
273280 			break;
274281 	}
275
-#ifdef CONFIG_ARCH_ROCKCHIP
282
+#ifdef CONFIG_NO_GKI
276283 	if (pfn < end_pfn)
277284 		dump_page_owner(page);
278285 #endif
....@@ -282,7 +289,7 @@
282289 
283290 /* Caller should ensure that requested range is in a single zone */
284291 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
285
-			bool skip_hwpoisoned_pages)
292
+			int isol_flags, unsigned long *failed_pfn)
286293 {
287294 	unsigned long pfn, flags;
288295 	struct page *page;
....@@ -304,16 +311,16 @@
304311 	/* Check all pages are free or marked as ISOLATED */
305312 	zone = page_zone(page);
306313 	spin_lock_irqsave(&zone->lock, flags);
307
-	pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn,
308
-						skip_hwpoisoned_pages);
314
+	pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
309315 	spin_unlock_irqrestore(&zone->lock, flags);
310316 
311317 	trace_test_pages_isolated(start_pfn, end_pfn, pfn);
318
+	if (pfn < end_pfn) {
319
+		page_pinner_failure_detect(pfn_to_page(pfn));
320
+		if (failed_pfn)
321
+			*failed_pfn = pfn;
322
+		return -EBUSY;
323
+	}
312324 
313
-	return pfn < end_pfn ? -EBUSY : 0;
314
-}
315
-
316
-struct page *alloc_migrate_target(struct page *page, unsigned long private)
317
-{
318
-	return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]);
325
+	return 0;
319326 }