add new system file

hc

2024-02-20 102a0743326a03cd1a1202ceda21e175b7d3575c

 kernel/mm/page_io.c
..
..
@@ -25,25 +25,23 @@
25
25
 #include <linux/psi.h>
26
26
 #include <linux/uio.h>
27
27
 #include <linux/sched/task.h>
28
-#include <asm/pgtable.h>
28
+#include <trace/hooks/mm.h>
29
29
 
30
30
 static struct bio *get_swap_bio(gfp_t gfp_flags,
31
31
 				struct page *page, bio_end_io_t end_io)
32
32
 {
33
-	int i, nr = hpage_nr_pages(page);
34
33
 	struct bio *bio;
35
34
 
36
-	bio = bio_alloc(gfp_flags, nr);
35
+	bio = bio_alloc(gfp_flags, 1);
37
36
 	if (bio) {
38
37
 		struct block_device *bdev;
39
38
 
40
39
 		bio->bi_iter.bi_sector = map_swap_page(page, &bdev);
41
40
 		bio_set_dev(bio, bdev);
41
+		bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
42
42
 		bio->bi_end_io = end_io;
43
43
 
44
-		for (i = 0; i < nr; i++)
45
-			bio_add_page(bio, page + i, PAGE_SIZE, 0);
46
-		VM_BUG_ON(bio->bi_iter.bi_size != PAGE_SIZE * nr);
44
+		bio_add_page(bio, page, thp_size(page), 0);
47
45
 	}
48
46
 	return bio;
49
47
 }
..
..
@@ -63,62 +61,13 @@
63
61
 		 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
64
62
 		 */
65
63
 		set_page_dirty(page);
66
-		pr_alert("Write-error on swap-device (%u:%u:%llu)\n",
67
-			 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
68
-			 (unsigned long long)bio->bi_iter.bi_sector);
64
+		pr_alert_ratelimited("Write-error on swap-device (%u:%u:%llu)\n",
65
+				     MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
66
+				     (unsigned long long)bio->bi_iter.bi_sector);
69
67
 		ClearPageReclaim(page);
70
68
 	}
71
69
 	end_page_writeback(page);
72
70
 	bio_put(bio);
73
-}
74
-
75
-static void swap_slot_free_notify(struct page *page)
76
-{
77
-	struct swap_info_struct *sis;
78
-	struct gendisk *disk;
79
-	swp_entry_t entry;
80
-
81
-	/*
82
-	 * There is no guarantee that the page is in swap cache - the software
83
-	 * suspend code (at least) uses end_swap_bio_read() against a non-
84
-	 * swapcache page.  So we must check PG_swapcache before proceeding with
85
-	 * this optimization.
86
-	 */
87
-	if (unlikely(!PageSwapCache(page)))
88
-		return;
89
-
90
-	sis = page_swap_info(page);
91
-	if (!(sis->flags & SWP_BLKDEV))
92
-		return;
93
-
94
-	/*
95
-	 * The swap subsystem performs lazy swap slot freeing,
96
-	 * expecting that the page will be swapped out again.
97
-	 * So we can avoid an unnecessary write if the page
98
-	 * isn't redirtied.
99
-	 * This is good for real swap storage because we can
100
-	 * reduce unnecessary I/O and enhance wear-leveling
101
-	 * if an SSD is used as the as swap device.
102
-	 * But if in-memory swap device (eg zram) is used,
103
-	 * this causes a duplicated copy between uncompressed
104
-	 * data in VM-owned memory and compressed data in
105
-	 * zram-owned memory.  So let's free zram-owned memory
106
-	 * and make the VM-owned decompressed page *dirty*,
107
-	 * so the page should be swapped out somewhere again if
108
-	 * we again wish to reclaim it.
109
-	 */
110
-	disk = sis->bdev->bd_disk;
111
-	entry.val = page_private(page);
112
-	if (disk->fops->swap_slot_free_notify &&
113
-			__swap_count(sis, entry) == 1) {
114
-		unsigned long offset;
115
-
116
-		offset = swp_offset(entry);
117
-
118
-		SetPageDirty(page);
119
-		disk->fops->swap_slot_free_notify(sis->bdev,
120
-				offset);
121
-	}
122
71
 }
123
72
 
124
73
 static void end_swap_bio_read(struct bio *bio)
..
..
@@ -129,20 +78,21 @@
129
78
 	if (bio->bi_status) {
130
79
 		SetPageError(page);
131
80
 		ClearPageUptodate(page);
132
-		pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
133
-			 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
134
-			 (unsigned long long)bio->bi_iter.bi_sector);
81
+		pr_alert_ratelimited("Read-error on swap-device (%u:%u:%llu)\n",
82
+				     MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
83
+				     (unsigned long long)bio->bi_iter.bi_sector);
135
84
 		goto out;
136
85
 	}
137
86
 
138
87
 	SetPageUptodate(page);
139
-	swap_slot_free_notify(page);
140
88
 out:
141
89
 	unlock_page(page);
142
90
 	WRITE_ONCE(bio->bi_private, NULL);
143
91
 	bio_put(bio);
144
-	wake_up_process(waiter);
145
-	put_task_struct(waiter);
92
+	if (waiter) {
93
+		blk_wake_io_task(waiter);
94
+		put_task_struct(waiter);
95
+	}
146
96
 }
147
97
 
148
98
 int generic_swapfile_activate(struct swap_info_struct *sis,
..
..
@@ -165,7 +115,7 @@
165
115
 	blocks_per_page = PAGE_SIZE >> blkbits;
166
116
 
167
117
 	/*
168
-	 * Map all the blocks into the extent list.  This code doesn't try
118
+	 * Map all the blocks into the extent tree.  This code doesn't try
169
119
 	 * to be very smart.
170
120
 	 */
171
121
 	probe_block = 0;
..
..
@@ -178,8 +128,9 @@
178
128
 
179
129
 		cond_resched();
180
130
 
181
-		first_block = bmap(inode, probe_block);
182
-		if (first_block == 0)
131
+		first_block = probe_block;
132
+		ret = bmap(inode, &first_block);
133
+		if (ret || !first_block)
183
134
 			goto bad_bmap;
184
135
 
185
136
 		/*
..
..
@@ -194,9 +145,11 @@
194
145
 					block_in_page++) {
195
146
 			sector_t block;
196
147
 
197
-			block = bmap(inode, probe_block + block_in_page);
198
-			if (block == 0)
148
+			block = probe_block + block_in_page;
149
+			ret = bmap(inode, &block);
150
+			if (ret || !block)
199
151
 				goto bad_bmap;
152
+
200
153
 			if (block != first_block + block_in_page) {
201
154
 				/* Discontiguity */
202
155
 				probe_block++;
..
..
@@ -251,6 +204,16 @@
251
204
 		unlock_page(page);
252
205
 		goto out;
253
206
 	}
207
+	/*
208
+	 * Arch code may have to preserve more data than just the page
209
+	 * contents, e.g. memory tags.
210
+	 */
211
+	ret = arch_prepare_to_swap(page);
212
+	if (ret) {
213
+		set_page_dirty(page);
214
+		unlock_page(page);
215
+		goto out;
216
+	}
254
217
 	if (frontswap_store(page) == 0) {
255
218
 		set_page_writeback(page);
256
219
 		unlock_page(page);
..
..
@@ -268,8 +231,25 @@
268
231
 	if (unlikely(PageTransHuge(page)))
269
232
 		count_vm_event(THP_SWPOUT);
270
233
 #endif
271
-	count_vm_events(PSWPOUT, hpage_nr_pages(page));
234
+	count_vm_events(PSWPOUT, thp_nr_pages(page));
272
235
 }
236
+
237
+#if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
238
+static void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
239
+{
240
+	struct cgroup_subsys_state *css;
241
+
242
+	if (!page->mem_cgroup)
243
+		return;
244
+
245
+	rcu_read_lock();
246
+	css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
247
+	bio_associate_blkg_from_css(bio, css);
248
+	rcu_read_unlock();
249
+}
250
+#else
251
+#define bio_associate_blkg_from_page(bio, page)		do { } while (0)
252
+#endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
273
253
 
274
254
 int __swap_writepage(struct page *page, struct writeback_control *wbc,
275
255
 		bio_end_io_t end_write_func)
..
..
@@ -277,9 +257,10 @@
277
257
 	struct bio *bio;
278
258
 	int ret;
279
259
 	struct swap_info_struct *sis = page_swap_info(page);
260
+	bool skip = false;
280
261
 
281
262
 	VM_BUG_ON_PAGE(!PageSwapCache(page), page);
282
-	if (sis->flags & SWP_FILE) {
263
+	if (data_race(sis->flags & SWP_FS_OPS)) {
283
264
 		struct kiocb kiocb;
284
265
 		struct file *swap_file = sis->swap_file;
285
266
 		struct address_space *mapping = swap_file->f_mapping;
..
..
@@ -290,7 +271,7 @@
290
271
 		};
291
272
 		struct iov_iter from;
292
273
 
293
-		iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
274
+		iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
294
275
 		init_sync_kiocb(&kiocb, swap_file);
295
276
 		kiocb.ki_pos = page_file_offset(page);
296
277
 
..
..
@@ -298,6 +279,7 @@
298
279
 		unlock_page(page);
299
280
 		ret = mapping->a_ops->direct_IO(&kiocb, &from);
300
281
 		if (ret == PAGE_SIZE) {
282
+			trace_android_vh_count_pswpout(sis);
301
283
 			count_vm_event(PSWPOUT);
302
284
 			ret = 0;
303
285
 		} else {
..
..
@@ -320,29 +302,30 @@
320
302
 		return ret;
321
303
 	}
322
304
 
323
-	ret = bdev_write_page(sis->bdev, map_swap_page(page, &sis->bdev),
324
-			      page, wbc);
305
+	ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
325
306
 	if (!ret) {
326
-		count_swpout_vm_event(page);
307
+		trace_android_vh_count_swpout_vm_event(sis, page, &skip);
308
+		if (!skip)
309
+			count_swpout_vm_event(page);
327
310
 		return 0;
328
311
 	}
329
312
 
330
-	ret = 0;
331
313
 	bio = get_swap_bio(GFP_NOIO, page, end_write_func);
332
314
 	if (bio == NULL) {
333
315
 		set_page_dirty(page);
334
316
 		unlock_page(page);
335
-		ret = -ENOMEM;
336
-		goto out;
317
+		return -ENOMEM;
337
318
 	}
338
319
 	bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
339
-	bio_associate_blkcg_from_page(bio, page);
340
-	count_swpout_vm_event(page);
320
+	bio_associate_blkg_from_page(bio, page);
321
+	trace_android_vh_count_swpout_vm_event(sis, page, &skip);
322
+	if (!skip)
323
+		count_swpout_vm_event(page);
341
324
 	set_page_writeback(page);
342
325
 	unlock_page(page);
343
326
 	submit_bio(bio);
344
-out:
345
-	return ret;
327
+
328
+	return 0;
346
329
 }
347
330
 
348
331
 int swap_readpage(struct page *page, bool synchronous)
..
..
@@ -371,25 +354,25 @@
371
354
 		goto out;
372
355
 	}
373
356
 
374
-	if (sis->flags & SWP_FILE) {
357
+	if (data_race(sis->flags & SWP_FS_OPS)) {
375
358
 		struct file *swap_file = sis->swap_file;
376
359
 		struct address_space *mapping = swap_file->f_mapping;
377
360
 
378
361
 		ret = mapping->a_ops->readpage(swap_file, page);
379
-		if (!ret)
362
+		if (!ret) {
363
+			trace_android_vh_count_pswpin(sis);
380
364
 			count_vm_event(PSWPIN);
365
+		}
381
366
 		goto out;
382
367
 	}
383
368
 
384
-	ret = bdev_read_page(sis->bdev, map_swap_page(page, &sis->bdev), page);
385
-	if (!ret) {
386
-		if (trylock_page(page)) {
387
-			swap_slot_free_notify(page);
388
-			unlock_page(page);
369
+	if (sis->flags & SWP_SYNCHRONOUS_IO) {
370
+		ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
371
+		if (!ret) {
372
+			trace_android_vh_count_pswpin(sis);
373
+			count_vm_event(PSWPIN);
374
+			goto out;
389
375
 		}
390
-
391
-		count_vm_event(PSWPIN);
392
-		goto out;
393
376
 	}
394
377
 
395
378
 	ret = 0;
..
..
@@ -404,9 +387,13 @@
404
387
 	 * Keep this task valid during swap readpage because the oom killer may
405
388
 	 * attempt to access it in the page fault retry time check.
406
389
 	 */
407
-	get_task_struct(current);
408
-	bio->bi_private = current;
409
390
 	bio_set_op_attrs(bio, REQ_OP_READ, 0);
391
+	if (synchronous) {
392
+		bio->bi_opf |= REQ_HIPRI;
393
+		get_task_struct(current);
394
+		bio->bi_private = current;
395
+	}
396
+	trace_android_vh_count_pswpin(sis);
410
397
 	count_vm_event(PSWPIN);
411
398
 	bio_get(bio);
412
399
 	qc = submit_bio(bio);
..
..
@@ -415,8 +402,8 @@
415
402
 		if (!READ_ONCE(bio->bi_private))
416
403
 			break;
417
404
 
418
-		if (!blk_poll(disk->queue, qc))
419
-			break;
405
+		if (!blk_poll(disk->queue, qc, true))
406
+			blk_io_schedule();
420
407
 	}
421
408
 	__set_current_state(TASK_RUNNING);
422
409
 	bio_put(bio);
..
..
@@ -430,7 +417,7 @@
430
417
 {
431
418
 	struct swap_info_struct *sis = page_swap_info(page);
432
419
 
433
-	if (sis->flags & SWP_FILE) {
420
+	if (data_race(sis->flags & SWP_FS_OPS)) {
434
421
 		struct address_space *mapping = sis->swap_file->f_mapping;
435
422
 
436
423
 		VM_BUG_ON_PAGE(!PageSwapCache(page), page);