Skip to contentSkip to conversation

selftests/mm: do not try to split below filesystem block size

PATCH In review Pankaj Raghav (Samsung) 2 participants 3 messages Aug 26, 2024

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 selftests/mm: do not try to split below filesystem block size

From: Pankaj Raghav <[email protected]>

There is no point trying to split pagecache thp below the blocksize of the filesystem as that is the minimum order that pagecache needs to maintain to support blocksizes greater than pagesize [1].

Set the lower limit for the splitting order to be the fs blocksize order.

As the number of tests will now depend on the minimum splitting order, move the file preparation before calling ksft_set_plan().

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/

---
 .../selftests/mm/split_huge_page_test.c       | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)
tools/testing/selftests/mm/split_huge_page_test.c +13 -6
index e5e8dafc9d94..187fe9107998 100644
@@ -9,11 +9,13 @@
 #include <stdlib.h>
 #include <stdarg.h>
 #include <unistd.h>
+#include <math.h>
 #include <inttypes.h>
 #include <string.h>
 #include <fcntl.h>
 #include <sys/mman.h>
 #include <sys/mount.h>
+#include <sys/stat.h>
 #include <malloc.h>
 #include <stdbool.h>
 #include <time.h>
@@ -404,9 +406,10 @@ void split_thp_in_pagecache_to_order(size_t fd_size, int order, const char *fs_l
 
 int main(int argc, char **argv)
 {
-	int i;
+	int i, min_split_order = 0;
 	size_t fd_size;
 	char *optional_xfs_path = NULL;
+	struct stat filestat;
 	char fs_loc_template[] = "/tmp/thp_fs_XXXXXX";
 	const char *fs_loc;
 	bool created_tmp;
@@ -421,8 +424,6 @@ int main(int argc, char **argv)
 	if (argc > 1)
 		optional_xfs_path = argv[1];
 
-	ksft_set_plan(3+9);
-
 	pagesize = getpagesize();
 	pageshift = ffs(pagesize) - 1;
 	pmd_pagesize = read_pmd_pagesize();
@@ -431,13 +432,19 @@ int main(int argc, char **argv)
 
 	fd_size = 2 * pmd_pagesize;
 
+	created_tmp = prepare_thp_fs(optional_xfs_path, fs_loc_template,
+			&fs_loc);
+
+	if (!stat(fs_loc, &filestat))
+		min_split_order = log2(filestat.st_blksize) - pageshift;
+
+	ksft_set_plan(3 + 9 - min_split_order);
+
 	split_pmd_thp();
 	split_pte_mapped_thp();
 	split_file_backed_thp();
 
-	created_tmp = prepare_thp_fs(optional_xfs_path, fs_loc_template,
-			&fs_loc);
-	for (i = 8; i >= 0; i--)
+	for (i = 8; i >= min_split_order; i--)
 		split_thp_in_pagecache_to_order(fd_size, i, fs_loc);
 	cleanup_thp_fs(fs_loc, created_tmp);
 

base-commit: 5771112c37523a2344b346d7fe613694a2566df9

Signature · 1 line
2.44.1
reviewer
#1 lore raw
Quoted 5 lines · Pankaj Raghav (Samsung)

From: Pankaj Raghav <[email protected]>

There is no point trying to split pagecache thp below the blocksize of the filesystem as that is the minimum order that pagecache needs to maintain to support blocksizes greater than pagesize [1].

But the purpose of the tests is to make sure all cases are properly handled, right? If we do not test splitting pagecache large folio below the block size, we will never know if a kernel change breaks the handling.

Just my two cents.

Quoted 7 lines

Set the lower limit for the splitting order to be the fs blocksize order.

As the number of tests will now depend on the minimum splitting order, move the file preparation before calling ksft_set_plan().

Quoted 6 lines

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/

---
 .../selftests/mm/split_huge_page_test.c       | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

--
Best Regards,
Yan, Zi

#2 lore raw
Quoted 8 lines · Zi Yan
Quoted 5 lines · Pankaj Raghav (Samsung)

From: Pankaj Raghav <[email protected]>

There is no point trying to split pagecache thp below the blocksize of the filesystem as that is the minimum order that pagecache needs to maintain to support blocksizes greater than pagesize [1].

Quoted 5 lines

But the purpose of the tests is to make sure all cases are properly handled, right? If we do not test splitting pagecache large folio below the block size, we will never know if a kernel change breaks the handling.

Just my two cents.

That is a fair point. Let's ignore this patch then :)

Quoted 19 lines
Quoted 8 lines

Set the lower limit for the splitting order to be the fs blocksize order.

As the number of tests will now depend on the minimum splitting order, move the file preparation before calling ksft_set_plan().

Quoted 6 lines

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/

---
 .../selftests/mm/split_huge_page_test.c       | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

--
Best Regards,
Yan, Zi

Signature · 1 line
Pankaj Raghav