Re: Subject: [BUG/RFC] write-open file THP cache purge can discard dirty page cache

[Pedro Falcato <pfalcato@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 586 bytes --]

On Tue, Jun 30, 2026 at 07:49:12PM +0100, Pedro Falcato wrote:
snip
> 
> Other idea: perhaps doing filemap_write_and_wait() after the nr_thps
> increment in collapse_file() will Just Work and result in a _much_
> simpler fix. And it avoids any weird forward-progress issues as no one can
> write to folios at that point.
> 

Gregg, if you could test this patch, it would be much appreciated. This patch
(hopefully) makes it so no dirty folio will ever coexist with a ro-THP, thus
hopefully sidestepping the entire issue in a simple way. Only compile-tested
and not reviewed.

-- 
Pedro

[-- Attachment #2: 0001-mm-khugepaged-write-all-dirty-folios-when-collapsing.patch --]
[-- Type: text/x-patch, Size: 2171 bytes --]

From 43d90a937f0b24656a8d0405035c3efcfdf0961e Mon Sep 17 00:00:00 2001
From: Pedro Falcato <pfalcato@suse.de>
Date: Tue, 30 Jun 2026 20:48:41 +0100
Subject: [PATCH] mm/khugepaged: write all dirty folios when collapsing

Signed-off-by: Pedro Falcato <pfalcato@suse.de>
---
 mm/khugepaged.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index a97b20617869..3f0f90ab16ba 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -60,6 +60,7 @@ enum scan_result {
 	SCAN_STORE_FAILED,
 	SCAN_COPY_MC,
 	SCAN_PAGE_FILLED,
+	SCAN_WRITEBACK_FAIL,
 };
 
 #define CREATE_TRACE_POINTS
@@ -1812,7 +1813,7 @@ static int collapse_file(struct mm_struct *mm, unsigned long addr,
 	pgoff_t index = 0, end = start + HPAGE_PMD_NR;
 	LIST_HEAD(pagelist);
 	XA_STATE_ORDER(xas, &mapping->i_pages, start, HPAGE_PMD_ORDER);
-	int nr_none = 0, result = SCAN_SUCCEED;
+	int nr_none = 0, result = SCAN_SUCCEED, err;
 	bool is_shmem = shmem_file(file);
 
 	VM_BUG_ON(!IS_ENABLED(CONFIG_READ_ONLY_THP_FOR_FS) && !is_shmem);
@@ -2043,6 +2044,17 @@ static int collapse_file(struct mm_struct *mm, unsigned long addr,
 	 */
 	try_to_unmap_flush();
 
+	/*
+	 * If collapse looks to be successful, flush any dirty pages
+	 * out the page cache. With the nr_thps incremented, there won't be
+	 * any new writers (nor new dirties).
+	 */
+	if (result == SCAN_SUCCEED && !is_shmem) {
+		err = filemap_write_and_wait(mapping);
+		if (err)
+			result = SCAN_WRITEBACK_FAIL;
+	}
+
 	if (result == SCAN_SUCCEED && nr_none &&
 	    !shmem_charge(mapping->host, nr_none))
 		result = SCAN_FAIL;
@@ -2210,9 +2222,10 @@ static int collapse_file(struct mm_struct *mm, unsigned long addr,
 	/*
 	 * Undo the updates of filemap_nr_thps_inc for non-SHMEM
 	 * file only. This undo is not needed unless failure is
-	 * due to SCAN_COPY_MC.
+	 * due to SCAN_COPY_MC or SCAN_WRITEBACK_FAIL.
 	 */
-	if (!is_shmem && result == SCAN_COPY_MC) {
+	if (!is_shmem && (result == SCAN_COPY_MC ||
+	    result == SCAN_WRITEBACK_FAIL)) {
 		filemap_nr_thps_dec(mapping);
 		/*
 		 * Paired with the fence in do_dentry_open() -> get_write_access()
-- 
2.55.0