forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-10-09 244b2c5ca8b14627e4a17755e5922221e121c771 
 kernel/mm/page-writeback.c
....@@ -1,3 +1,4 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /*
23  * mm/page-writeback.c
34  *
....@@ -40,6 +41,9 @@
4041 #include <trace/events/writeback.h>
4142 
4243 #include "internal.h"
44
+
45
+#undef CREATE_TRACE_POINT
46
+#include <trace/hooks/mm.h>
4347 
4448 /*
4549  * Sleep at most 200ms at a time in balance_dirty_pages().
....@@ -256,7 +260,7 @@
256260  * requiring writeback.
257261  *
258262  * This number of dirtyable pages is the base value of which the
259
- * user-configurable dirty ratio is the effictive number of pages that
263
+ * user-configurable dirty ratio is the effective number of pages that
260264  * are allowed to be actually dirtied.  Per individual zone, or
261265  * globally by using the sum of dirtyable pages over all zones.
262266  *
....@@ -270,7 +274,7 @@
270274  * node_dirtyable_memory - number of dirtyable pages in a node
271275  * @pgdat: the node
272276  *
273
- * Returns the node's number of pages potentially available for dirty
277
+ * Return: the node's number of pages potentially available for dirty
274278  * page cache.  This is the base value for the per-node dirty limits.
275279  */
276280 static unsigned long node_dirtyable_memory(struct pglist_data *pgdat)
....@@ -355,7 +359,7 @@
355359 /**
356360  * global_dirtyable_memory - number of globally dirtyable pages
357361  *
358
- * Returns the global number of pages potentially available for dirty
362
+ * Return: the global number of pages potentially available for dirty
359363  * page cache.  This is the base value for the global dirty limits.
360364  */
361365 static unsigned long global_dirtyable_memory(void)
....@@ -386,8 +390,7 @@
386390  * Calculate @dtc->thresh and ->bg_thresh considering
387391  * vm_dirty_{bytes|ratio} and dirty_background_{bytes|ratio}.  The caller
388392  * must ensure that @dtc->avail is set before calling this function.  The
389
- * dirty limits will be lifted by 1/4 for PF_LESS_THROTTLE (ie. nfsd) and
390
- * real-time tasks.
393
+ * dirty limits will be lifted by 1/4 for real-time tasks.
391394  */
392395 static void domain_dirty_limits(struct dirty_throttle_control *dtc)
393396 {
....@@ -435,7 +438,7 @@
435438 	if (bg_thresh >= thresh)
436439 		bg_thresh = thresh / 2;
437440 	tsk = current;
438
-	if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
441
+	if (rt_task(tsk)) {
439442 		bg_thresh += bg_thresh / 4 + global_wb_domain.dirty_limit / 32;
440443 		thresh += thresh / 4 + global_wb_domain.dirty_limit / 32;
441444 	}
....@@ -470,7 +473,7 @@
470473  * node_dirty_limit - maximum number of dirty pages allowed in a node
471474  * @pgdat: the node
472475  *
473
- * Returns the maximum number of dirty pages allowed in a node, based
476
+ * Return: the maximum number of dirty pages allowed in a node, based
474477  * on the node's dirtyable memory.
475478  */
476479 static unsigned long node_dirty_limit(struct pglist_data *pgdat)
....@@ -485,7 +488,7 @@
485488 	else
486489 		dirty = vm_dirty_ratio * node_memory / 100;
487490 
488
-	if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk))
491
+	if (rt_task(tsk))
489492 		dirty += dirty / 4;
490493 
491494 	return dirty;
....@@ -495,7 +498,7 @@
495498  * node_dirty_ok - tells whether a node is within its dirty limits
496499  * @pgdat: the node to check
497500  *
498
- * Returns %true when the dirty pages in @pgdat are within the node's
501
+ * Return: %true when the dirty pages in @pgdat are within the node's
499502  * dirty limit, %false if the limit is exceeded.
500503  */
501504 bool node_dirty_ok(struct pglist_data *pgdat)
....@@ -504,15 +507,13 @@
504507 	unsigned long nr_pages = 0;
505508 
506509 	nr_pages += node_page_state(pgdat, NR_FILE_DIRTY);
507
-	nr_pages += node_page_state(pgdat, NR_UNSTABLE_NFS);
508510 	nr_pages += node_page_state(pgdat, NR_WRITEBACK);
509511 
510512 	return nr_pages <= limit;
511513 }
512514 
513515 int dirty_background_ratio_handler(struct ctl_table *table, int write,
514
-		void __user *buffer, size_t *lenp,
515
-		loff_t *ppos)
516
+		void *buffer, size_t *lenp, loff_t *ppos)
516517 {
517518 	int ret;
518519 
....@@ -523,8 +524,7 @@
523524 }
524525 
525526 int dirty_background_bytes_handler(struct ctl_table *table, int write,
526
-		void __user *buffer, size_t *lenp,
527
-		loff_t *ppos)
527
+		void *buffer, size_t *lenp, loff_t *ppos)
528528 {
529529 	int ret;
530530 
....@@ -534,9 +534,8 @@
534534 	return ret;
535535 }
536536 
537
-int dirty_ratio_handler(struct ctl_table *table, int write,
538
-		void __user *buffer, size_t *lenp,
539
-		loff_t *ppos)
537
+int dirty_ratio_handler(struct ctl_table *table, int write, void *buffer,
538
+		size_t *lenp, loff_t *ppos)
540539 {
541540 	int old_ratio = vm_dirty_ratio;
542541 	int ret;
....@@ -550,8 +549,7 @@
550549 }
551550 
552551 int dirty_bytes_handler(struct ctl_table *table, int write,
553
-		void __user *buffer, size_t *lenp,
554
-		loff_t *ppos)
552
+		void *buffer, size_t *lenp, loff_t *ppos)
555553 {
556554 	unsigned long old_bytes = vm_dirty_bytes;
557555 	int ret;
....@@ -743,9 +741,6 @@
743741  * __wb_calc_thresh - @wb's share of dirty throttling threshold
744742  * @dtc: dirty_throttle_context of interest
745743  *
746
- * Returns @wb's dirty limit in pages. The term "dirty" in the context of
747
- * dirty balancing includes all PG_dirty, PG_writeback and NFS unstable pages.
748
- *
749744  * Note that balance_dirty_pages() will only seriously take it as a hard limit
750745  * when sleeping max_pause per page is not enough to keep the dirty pages under
751746  * control. For example, when the device is completely stalled due to some error
....@@ -759,13 +754,16 @@
759754  *
760755  * The wb's share of dirty limit will be adapting to its throughput and
761756  * bounded by the bdi->min_ratio and/or bdi->max_ratio parameters, if set.
757
+ *
758
+ * Return: @wb's dirty limit in pages. The term "dirty" in the context of
759
+ * dirty balancing includes all PG_dirty and PG_writeback pages.
762760  */
763761 static unsigned long __wb_calc_thresh(struct dirty_throttle_control *dtc)
764762 {
765763 	struct wb_domain *dom = dtc_dom(dtc);
766764 	unsigned long thresh = dtc->thresh;
767765 	u64 wb_thresh;
768
-	long numerator, denominator;
766
+	unsigned long numerator, denominator;
769767 	unsigned long wb_min_ratio, wb_max_ratio;
770768 
771769 	/*
....@@ -776,7 +774,7 @@
776774 
777775 	wb_thresh = (thresh * (100 - bdi_min_ratio)) / 100;
778776 	wb_thresh *= numerator;
779
-	do_div(wb_thresh, denominator);
777
+	wb_thresh = div64_ul(wb_thresh, denominator);
780778 
781779 	wb_min_max_ratio(dtc->wb, &wb_min_ratio, &wb_max_ratio);
782780 
....@@ -1101,7 +1099,7 @@
11011099 	bw = written - min(written, wb->written_stamp);
11021100 	bw *= HZ;
11031101 	if (unlikely(elapsed > period)) {
1104
-		do_div(bw, elapsed);
1102
+		bw = div64_ul(bw, elapsed);
11051103 		avg = bw;
11061104 		goto out;
11071105 	}
....@@ -1566,7 +1564,7 @@
15661564 	struct dirty_throttle_control * const mdtc = mdtc_valid(&mdtc_stor) ?
15671565 						     &mdtc_stor : NULL;
15681566 	struct dirty_throttle_control *sdtc;
1569
-	unsigned long nr_reclaimable;	/* = file_dirty + unstable_nfs */
1567
+	unsigned long nr_reclaimable;	/* = file_dirty */
15701568 	long period;
15711569 	long pause;
15721570 	long max_pause;
....@@ -1586,14 +1584,7 @@
15861584 		unsigned long m_thresh = 0;
15871585 		unsigned long m_bg_thresh = 0;
15881586 
1589
-		/*
1590
-		 * Unstable writes are a feature of certain networked
1591
-		 * filesystems (i.e. NFS) in which data may have been
1592
-		 * written to the server's write cache, but has not yet
1593
-		 * been flushed to permanent storage.
1594
-		 */
1595
-		nr_reclaimable = global_node_page_state(NR_FILE_DIRTY) +
1596
-					global_node_page_state(NR_UNSTABLE_NFS);
1587
+		nr_reclaimable = global_node_page_state(NR_FILE_DIRTY);
15971588 		gdtc->avail = global_dirtyable_memory();
15981589 		gdtc->dirty = nr_reclaimable + global_node_page_state(NR_WRITEBACK);
15991590 
....@@ -1637,6 +1628,9 @@
16371628 			}
16381629 		}
16391630 
1631
+		trace_android_vh_mm_dirty_limits(gdtc, strictlimit, dirty, bg_thresh,
1632
+				nr_reclaimable, pages_dirtied);
1633
+
16401634 		/*
16411635 		 * Throttle it only when the background writeback cannot
16421636 		 * catch-up. This avoids (excessively) small writeouts
....@@ -1652,8 +1646,12 @@
16521646 		if (dirty <= dirty_freerun_ceiling(thresh, bg_thresh) &&
16531647 		    (!mdtc ||
16541648 		     m_dirty <= dirty_freerun_ceiling(m_thresh, m_bg_thresh))) {
1655
-			unsigned long intv = dirty_poll_interval(dirty, thresh);
1656
-			unsigned long m_intv = ULONG_MAX;
1649
+			unsigned long intv;
1650
+			unsigned long m_intv;
1651
+
1652
+free_running:
1653
+			intv = dirty_poll_interval(dirty, thresh);
1654
+			m_intv = ULONG_MAX;
16571655 
16581656 			current->dirty_paused_when = now;
16591657 			current->nr_dirtied = 0;
....@@ -1666,12 +1664,25 @@
16661664 		if (unlikely(!writeback_in_progress(wb)))
16671665 			wb_start_background_writeback(wb);
16681666 
1667
+		mem_cgroup_flush_foreign(wb);
1668
+
16691669 		/*
16701670 		 * Calculate global domain's pos_ratio and select the
16711671 		 * global dtc by default.
16721672 		 */
1673
-		if (!strictlimit)
1673
+		if (!strictlimit) {
16741674 			wb_dirty_limits(gdtc);
1675
+
1676
+			if ((current->flags & PF_LOCAL_THROTTLE) &&
1677
+			    gdtc->wb_dirty <
1678
+			    dirty_freerun_ceiling(gdtc->wb_thresh,
1679
+						  gdtc->wb_bg_thresh))
1680
+				/*
1681
+				 * LOCAL_THROTTLE tasks must not be throttled
1682
+				 * when below the per-wb freerun ceiling.
1683
+				 */
1684
+				goto free_running;
1685
+		}
16751686 
16761687 		dirty_exceeded = (gdtc->wb_dirty > gdtc->wb_thresh) &&
16771688 			((gdtc->dirty > gdtc->thresh) || strictlimit);
....@@ -1686,9 +1697,20 @@
16861697 			 * both global and memcg domains.  Choose the one
16871698 			 * w/ lower pos_ratio.
16881699 			 */
1689
-			if (!strictlimit)
1700
+			if (!strictlimit) {
16901701 				wb_dirty_limits(mdtc);
16911702 
1703
+				if ((current->flags & PF_LOCAL_THROTTLE) &&
1704
+				    mdtc->wb_dirty <
1705
+				    dirty_freerun_ceiling(mdtc->wb_thresh,
1706
+							  mdtc->wb_bg_thresh))
1707
+					/*
1708
+					 * LOCAL_THROTTLE tasks must not be
1709
+					 * throttled when below the per-wb
1710
+					 * freerun ceiling.
1711
+					 */
1712
+					goto free_running;
1713
+			}
16921714 			dirty_exceeded |= (mdtc->wb_dirty > mdtc->wb_thresh) &&
16931715 				((mdtc->dirty > mdtc->thresh) || strictlimit);
16941716 
....@@ -1866,7 +1888,7 @@
18661888 	int ratelimit;
18671889 	int *p;
18681890 
1869
-	if (!bdi_cap_account_dirty(bdi))
1891
+	if (!(bdi->capabilities & BDI_CAP_WRITEBACK))
18701892 		return;
18711893 
18721894 	if (inode_cgwb_enabled(inode))
....@@ -1918,7 +1940,9 @@
19181940  * @wb: bdi_writeback of interest
19191941  *
19201942  * Determines whether background writeback should keep writing @wb or it's
1921
- * clean enough.  Returns %true if writeback should continue.
1943
+ * clean enough.
1944
+ *
1945
+ * Return: %true if writeback should continue.
19221946  */
19231947 bool wb_over_bg_thresh(struct bdi_writeback *wb)
19241948 {
....@@ -1927,21 +1951,27 @@
19271951 	struct dirty_throttle_control * const gdtc = &gdtc_stor;
19281952 	struct dirty_throttle_control * const mdtc = mdtc_valid(&mdtc_stor) ?
19291953 						     &mdtc_stor : NULL;
1954
+	unsigned long reclaimable;
1955
+	unsigned long thresh;
19301956 
19311957 	/*
19321958 	 * Similar to balance_dirty_pages() but ignores pages being written
19331959 	 * as we're trying to decide whether to put more under writeback.
19341960 	 */
19351961 	gdtc->avail = global_dirtyable_memory();
1936
-	gdtc->dirty = global_node_page_state(NR_FILE_DIRTY) +
1937
-		      global_node_page_state(NR_UNSTABLE_NFS);
1962
+	gdtc->dirty = global_node_page_state(NR_FILE_DIRTY);
19381963 	domain_dirty_limits(gdtc);
19391964 
19401965 	if (gdtc->dirty > gdtc->bg_thresh)
19411966 		return true;
19421967 
1943
-	if (wb_stat(wb, WB_RECLAIMABLE) >
1944
-	    wb_calc_thresh(gdtc->wb, gdtc->bg_thresh))
1968
+	thresh = wb_calc_thresh(gdtc->wb, gdtc->bg_thresh);
1969
+	if (thresh < 2 * wb_stat_error())
1970
+		reclaimable = wb_stat_sum(wb, WB_RECLAIMABLE);
1971
+	else
1972
+		reclaimable = wb_stat(wb, WB_RECLAIMABLE);
1973
+
1974
+	if (reclaimable > thresh)
19451975 		return true;
19461976 
19471977 	if (mdtc) {
....@@ -1955,8 +1985,13 @@
19551985 		if (mdtc->dirty > mdtc->bg_thresh)
19561986 			return true;
19571987 
1958
-		if (wb_stat(wb, WB_RECLAIMABLE) >
1959
-		    wb_calc_thresh(mdtc->wb, mdtc->bg_thresh))
1988
+		thresh = wb_calc_thresh(mdtc->wb, mdtc->bg_thresh);
1989
+		if (thresh < 2 * wb_stat_error())
1990
+			reclaimable = wb_stat_sum(wb, WB_RECLAIMABLE);
1991
+		else
1992
+			reclaimable = wb_stat(wb, WB_RECLAIMABLE);
1993
+
1994
+		if (reclaimable > thresh)
19601995 			return true;
19611996 	}
19621997 
....@@ -1967,7 +2002,7 @@
19672002  * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
19682003  */
19692004 int dirty_writeback_centisecs_handler(struct ctl_table *table, int write,
1970
-	void __user *buffer, size_t *length, loff_t *ppos)
2005
+		void *buffer, size_t *length, loff_t *ppos)
19712006 {
19722007 	unsigned int old_interval = dirty_writeback_interval;
19732008 	int ret;
....@@ -2059,13 +2094,11 @@
20592094  * Called early on to tune the page writeback dirty limits.
20602095  *
20612096  * We used to scale dirty pages according to how total memory
2062
- * related to pages that could be allocated for buffers (by
2063
- * comparing nr_free_buffer_pages() to vm_total_pages.
2097
+ * related to pages that could be allocated for buffers.
20642098  *
20652099  * However, that was when we used "dirty_ratio" to scale with
20662100  * all memory, and we don't do that any more. "dirty_ratio"
2067
- * is now applied to total non-HIGHPAGE memory (by subtracting
2068
- * totalhigh_pages from vm_total_pages), and as such we can't
2101
+ * is now applied to total non-HIGHPAGE memory, and as such we can't
20692102  * get into the old insane situation any more where we had
20702103  * large amounts of dirty pages compared to a small amount of
20712104  * non-HIGHMEM memory.
....@@ -2097,34 +2130,25 @@
20972130  * dirty pages in the file (thus it is important for this function to be quick
20982131  * so that it can tag pages faster than a dirtying process can create them).
20992132  */
2100
-/*
2101
- * We tag pages in batches of WRITEBACK_TAG_BATCH to reduce the i_pages lock
2102
- * latency.
2103
- */
21042133 void tag_pages_for_writeback(struct address_space *mapping,
21052134 			     pgoff_t start, pgoff_t end)
21062135 {
2107
-#define WRITEBACK_TAG_BATCH 4096
2108
-	unsigned long tagged = 0;
2109
-	struct radix_tree_iter iter;
2110
-	void **slot;
2136
+	XA_STATE(xas, &mapping->i_pages, start);
2137
+	unsigned int tagged = 0;
2138
+	void *page;
21112139 
2112
-	xa_lock_irq(&mapping->i_pages);
2113
-	radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, start,
2114
-							PAGECACHE_TAG_DIRTY) {
2115
-		if (iter.index > end)
2116
-			break;
2117
-		radix_tree_iter_tag_set(&mapping->i_pages, &iter,
2118
-							PAGECACHE_TAG_TOWRITE);
2119
-		tagged++;
2120
-		if ((tagged % WRITEBACK_TAG_BATCH) != 0)
2140
+	xas_lock_irq(&xas);
2141
+	xas_for_each_marked(&xas, page, end, PAGECACHE_TAG_DIRTY) {
2142
+		xas_set_mark(&xas, PAGECACHE_TAG_TOWRITE);
2143
+		if (++tagged % XA_CHECK_SCHED)
21212144 			continue;
2122
-		slot = radix_tree_iter_resume(slot, &iter);
2123
-		xa_unlock_irq(&mapping->i_pages);
2145
+
2146
+		xas_pause(&xas);
2147
+		xas_unlock_irq(&xas);
21242148 		cond_resched();
2125
-		xa_lock_irq(&mapping->i_pages);
2149
+		xas_lock_irq(&xas);
21262150 	}
2127
-	xa_unlock_irq(&mapping->i_pages);
2151
+	xas_unlock_irq(&xas);
21282152 }
21292153 EXPORT_SYMBOL(tag_pages_for_writeback);
21302154 
....@@ -2156,6 +2180,8 @@
21562180  * lock/page writeback access order inversion - we should only ever lock
21572181  * multiple pages in ascending page->index order, and looping back to the start
21582182  * of the file violates that rule and causes deadlocks.
2183
+ *
2184
+ * Return: %0 on success, negative error code otherwise
21592185  */
21602186 int write_cache_pages(struct address_space *mapping,
21612187 		      struct writeback_control *wbc, writepage_t writepage,
....@@ -2166,17 +2192,15 @@
21662192 	int error;
21672193 	struct pagevec pvec;
21682194 	int nr_pages;
2169
-	pgoff_t uninitialized_var(writeback_index);
21702195 	pgoff_t index;
21712196 	pgoff_t end;		/* Inclusive */
21722197 	pgoff_t done_index;
21732198 	int range_whole = 0;
2174
-	int tag;
2199
+	xa_mark_t tag;
21752200 
21762201 	pagevec_init(&pvec);
21772202 	if (wbc->range_cyclic) {
2178
-		writeback_index = mapping->writeback_index; /* prev offset */
2179
-		index = writeback_index;
2203
+		index = mapping->writeback_index; /* prev offset */
21802204 		end = -1;
21812205 	} else {
21822206 		index = wbc->range_start >> PAGE_SHIFT;
....@@ -2184,12 +2208,12 @@
21842208 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
21852209 			range_whole = 1;
21862210 	}
2187
-	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2188
-		tag = PAGECACHE_TAG_TOWRITE;
2189
-	else
2190
-		tag = PAGECACHE_TAG_DIRTY;
2191
-	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2211
+	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) {
21922212 		tag_pages_for_writeback(mapping, index, end);
2213
+		tag = PAGECACHE_TAG_TOWRITE;
2214
+	} else {
2215
+		tag = PAGECACHE_TAG_DIRTY;
2216
+	}
21932217 	done_index = index;
21942218 	while (!done && (index <= end)) {
21952219 		int i;
....@@ -2314,6 +2338,8 @@
23142338  *
23152339  * This is a library function, which implements the writepages()
23162340  * address_space_operation.
2341
+ *
2342
+ * Return: %0 on success, negative error code otherwise
23172343  */
23182344 int generic_writepages(struct address_space *mapping,
23192345 		       struct writeback_control *wbc)
....@@ -2360,6 +2386,8 @@
23602386  *
23612387  * Note that the mapping's AS_EIO/AS_ENOSPC flags will be cleared when this
23622388  * function returns.
2389
+ *
2390
+ * Return: %0 on success, negative error code otherwise
23632391  */
23642392 int write_one_page(struct page *page)
23652393 {
....@@ -2413,7 +2441,7 @@
24132441 
24142442 	trace_writeback_dirty_page(page, mapping);
24152443 
2416
-	if (mapping_cap_account_dirty(mapping)) {
2444
+	if (mapping_can_writeback(mapping)) {
24172445 		struct bdi_writeback *wb;
24182446 
24192447 		inode_attach_wb(inode, page);
....@@ -2427,9 +2455,10 @@
24272455 		task_io_account_write(PAGE_SIZE);
24282456 		current->nr_dirtied++;
24292457 		this_cpu_inc(bdp_ratelimits);
2458
+
2459
+		mem_cgroup_track_foreign_dirty(page, wb);
24302460 	}
24312461 }
2432
-EXPORT_SYMBOL(account_page_dirtied);
24332462 
24342463 /*
24352464  * Helper function for deaccounting dirty page without writeback.
....@@ -2439,7 +2468,7 @@
24392468 void account_page_cleaned(struct page *page, struct address_space *mapping,
24402469 			  struct bdi_writeback *wb)
24412470 {
2442
-	if (mapping_cap_account_dirty(mapping)) {
2471
+	if (mapping_can_writeback(mapping)) {
24432472 		dec_lruvec_page_state(page, NR_FILE_DIRTY);
24442473 		dec_zone_page_state(page, NR_ZONE_WRITE_PENDING);
24452474 		dec_wb_stat(wb, WB_RECLAIMABLE);
....@@ -2449,7 +2478,7 @@
24492478 
24502479 /*
24512480  * For address_spaces which do not use buffers.  Just tag the page as dirty in
2452
- * its radix tree.
2481
+ * the xarray.
24532482  *
24542483  * This is also used when a single buffer is being dirtied: we want to set the
24552484  * page dirty in that case, but not all the buffers.  This is a "bottom-up"
....@@ -2475,7 +2504,7 @@
24752504 		BUG_ON(page_mapping(page) != mapping);
24762505 		WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
24772506 		account_page_dirtied(page, mapping);
2478
-		radix_tree_tag_set(&mapping->i_pages, page_index(page),
2507
+		__xa_set_mark(&mapping->i_pages, page_index(page),
24792508 				   PAGECACHE_TAG_DIRTY);
24802509 		xa_unlock_irqrestore(&mapping->i_pages, flags);
24812510 		unlock_page_memcg(page);
....@@ -2502,7 +2531,7 @@
25022531 {
25032532 	struct address_space *mapping = page->mapping;
25042533 
2505
-	if (mapping && mapping_cap_account_dirty(mapping)) {
2534
+	if (mapping && mapping_can_writeback(mapping)) {
25062535 		struct inode *inode = mapping->host;
25072536 		struct bdi_writeback *wb;
25082537 		struct wb_lock_cookie cookie = {};
....@@ -2614,7 +2643,7 @@
26142643 {
26152644 	struct address_space *mapping = page_mapping(page);
26162645 
2617
-	if (mapping_cap_account_dirty(mapping)) {
2646
+	if (mapping_can_writeback(mapping)) {
26182647 		struct inode *inode = mapping->host;
26192648 		struct bdi_writeback *wb;
26202649 		struct wb_lock_cookie cookie = {};
....@@ -2638,13 +2667,13 @@
26382667  * Returns true if the page was previously dirty.
26392668  *
26402669  * This is for preparing to put the page under writeout.  We leave the page
2641
- * tagged as dirty in the radix tree so that a concurrent write-for-sync
2670
+ * tagged as dirty in the xarray so that a concurrent write-for-sync
26422671  * can discover it via a PAGECACHE_TAG_DIRTY walk.  The ->writepage
26432672  * implementation will run either set_page_writeback() or set_page_dirty(),
2644
- * at which stage we bring the page's dirty flag and radix-tree dirty tag
2673
+ * at which stage we bring the page's dirty flag and xarray dirty tag
26452674  * back into sync.
26462675  *
2647
- * This incoherency between the page's dirty flag and radix-tree tag is
2676
+ * This incoherency between the page's dirty flag and xarray tag is
26482677  * unfortunate, but it only exists while the page is locked.
26492678  */
26502679 int clear_page_dirty_for_io(struct page *page)
....@@ -2652,9 +2681,9 @@
26522681 	struct address_space *mapping = page_mapping(page);
26532682 	int ret = 0;
26542683 
2655
-	BUG_ON(!PageLocked(page));
2684
+	VM_BUG_ON_PAGE(!PageLocked(page), page);
26562685 
2657
-	if (mapping && mapping_cap_account_dirty(mapping)) {
2686
+	if (mapping && mapping_can_writeback(mapping)) {
26582687 		struct inode *inode = mapping->host;
26592688 		struct bdi_writeback *wb;
26602689 		struct wb_lock_cookie cookie = {};
....@@ -2725,9 +2754,9 @@
27252754 		xa_lock_irqsave(&mapping->i_pages, flags);
27262755 		ret = TestClearPageWriteback(page);
27272756 		if (ret) {
2728
-			radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2757
+			__xa_clear_mark(&mapping->i_pages, page_index(page),
27292758 						PAGECACHE_TAG_WRITEBACK);
2730
-			if (bdi_cap_account_writeback(bdi)) {
2759
+			if (bdi->capabilities & BDI_CAP_WRITEBACK_ACCT) {
27312760 				struct bdi_writeback *wb = inode_to_wb(inode);
27322761 
27332762 				dec_wb_stat(wb, WB_WRITEBACK);
....@@ -2743,12 +2772,6 @@
27432772 	} else {
27442773 		ret = TestClearPageWriteback(page);
27452774 	}
2746
-	/*
2747
-	 * NOTE: Page might be free now! Writeback doesn't hold a page
2748
-	 * reference on its own, it relies on truncation to wait for
2749
-	 * the clearing of PG_writeback. The below can only access
2750
-	 * page state that is static across allocation cycles.
2751
-	 */
27522775 	if (ret) {
27532776 		dec_lruvec_state(lruvec, NR_WRITEBACK);
27542777 		dec_zone_page_state(page, NR_ZONE_WRITE_PENDING);
....@@ -2761,15 +2784,17 @@
27612784 int __test_set_page_writeback(struct page *page, bool keep_write)
27622785 {
27632786 	struct address_space *mapping = page_mapping(page);
2764
-	int ret;
2787
+	int ret, access_ret;
27652788 
27662789 	lock_page_memcg(page);
27672790 	if (mapping && mapping_use_writeback_tags(mapping)) {
2791
+		XA_STATE(xas, &mapping->i_pages, page_index(page));
27682792 		struct inode *inode = mapping->host;
27692793 		struct backing_dev_info *bdi = inode_to_bdi(inode);
27702794 		unsigned long flags;
27712795 
2772
-		xa_lock_irqsave(&mapping->i_pages, flags);
2796
+		xas_lock_irqsave(&xas, flags);
2797
+		xas_load(&xas);
27732798 		ret = TestSetPageWriteback(page);
27742799 		if (!ret) {
27752800 			bool on_wblist;
....@@ -2777,9 +2802,8 @@
27772802 			on_wblist = mapping_tagged(mapping,
27782803 						   PAGECACHE_TAG_WRITEBACK);
27792804 
2780
-			radix_tree_tag_set(&mapping->i_pages, page_index(page),
2781
-						PAGECACHE_TAG_WRITEBACK);
2782
-			if (bdi_cap_account_writeback(bdi))
2805
+			xas_set_mark(&xas, PAGECACHE_TAG_WRITEBACK);
2806
+			if (bdi->capabilities & BDI_CAP_WRITEBACK_ACCT)
27832807 				inc_wb_stat(inode_to_wb(inode), WB_WRITEBACK);
27842808 
27852809 			/*
....@@ -2791,12 +2815,10 @@
27912815 				sb_mark_inode_writeback(mapping->host);
27922816 		}
27932817 		if (!PageDirty(page))
2794
-			radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2795
-						PAGECACHE_TAG_DIRTY);
2818
+			xas_clear_mark(&xas, PAGECACHE_TAG_DIRTY);
27962819 		if (!keep_write)
2797
-			radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2798
-						PAGECACHE_TAG_TOWRITE);
2799
-		xa_unlock_irqrestore(&mapping->i_pages, flags);
2820
+			xas_clear_mark(&xas, PAGECACHE_TAG_TOWRITE);
2821
+		xas_unlock_irqrestore(&xas, flags);
28002822 	} else {
28012823 		ret = TestSetPageWriteback(page);
28022824 	}
....@@ -2805,20 +2827,29 @@
28052827 		inc_zone_page_state(page, NR_ZONE_WRITE_PENDING);
28062828 	}
28072829 	unlock_page_memcg(page);
2830
+	access_ret = arch_make_page_accessible(page);
2831
+	/*
2832
+	 * If writeback has been triggered on a page that cannot be made
2833
+	 * accessible, it is too late to recover here.
2834
+	 */
2835
+	VM_BUG_ON_PAGE(access_ret != 0, page);
2836
+
28082837 	return ret;
28092838 
28102839 }
28112840 EXPORT_SYMBOL(__test_set_page_writeback);
28122841 
28132842 /*
2814
- * Return true if any of the pages in the mapping are marked with the
2815
- * passed tag.
2843
+ * Wait for a page to complete writeback
28162844  */
2817
-int mapping_tagged(struct address_space *mapping, int tag)
2845
+void wait_on_page_writeback(struct page *page)
28182846 {
2819
-	return radix_tree_tagged(&mapping->i_pages, tag);
2847
+	while (PageWriteback(page)) {
2848
+		trace_wait_on_page_writeback(page, page_mapping(page));
2849
+		wait_on_page_bit(page, PG_writeback);
2850
+	}
28202851 }
2821
-EXPORT_SYMBOL(mapping_tagged);
2852
+EXPORT_SYMBOL_GPL(wait_on_page_writeback);
28222853 
28232854 /**
28242855  * wait_for_stable_page() - wait for writeback to finish, if necessary.
....@@ -2830,7 +2861,8 @@
28302861  */
28312862 void wait_for_stable_page(struct page *page)
28322863 {
2833
-	if (bdi_cap_stable_pages_required(inode_to_bdi(page->mapping->host)))
2864
+	page = thp_head(page);
2865
+	if (page->mapping->host->i_sb->s_iflags & SB_I_STABLE_WRITES)
28342866 		wait_on_page_writeback(page);
28352867 }
28362868 EXPORT_SYMBOL_GPL(wait_for_stable_page);