Skip to contentSkip to conversation

ksm: fix occasional page_volatile false positives

PATCH In review Andrea Arcangeli 3 participants 4 messages May 12, 2022 – May 13, 2022

lore
Share & tools
lore
#0 lore raw

cover ksm: fix occasional page_volatile false positives

Hello everyone,

Some LTP KSM tests (ksm0[2-4]) occasionally reported some page_volatile elevated failure that turned out to be a false positive from LTP.

This proposed fix has been verified by running the tests in a loop like this: "while ./ksm02; do : ; done".

Thanks,
Andrea

Andrea Arcangeli (1):
ksm: fix occasional page_volatile false positives

 testcases/kernel/mem/lib/mem.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw

PATCH 1/1 ksm: fix occasional page_volatile false positives

The KSM scan keeps running while checking the KSM status in sysfs, but during the KSM scan the the rmap_items in the stale unstable tree of the old pass are removed from it and are later reinserted in the new unstable tree of the current pass, leading to some page_volatile false positives.

The fix is stop the KSM scan temporarily while the KSM status is being read from sysfs.

For reference here's an instance of the fixed false positives:

mem.c:255: TFAIL: pages_volatile is not 0 but 1.
mem.c:255: TFAIL: pages_unshared is not 1 but 0.

---
 testcases/kernel/mem/lib/mem.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
testcases/kernel/mem/lib/mem.c +17 -0
index 090569ebb..8ddd7adf7 100644
@@ -261,14 +261,31 @@ static void final_group_check(int run, int pages_shared, int pages_sharing,
 			  int pages_volatile, int pages_unshared,
 			  int sleep_millisecs, int pages_to_scan)
 {
+	int ksm_run_orig;
+
 	tst_res(TINFO, "check!");
 	check("run", run);
+
+	/*
+	 * Temporarily stop the KSM scan during the checks: during the
+	 * KSM scan the rmap_items in the stale unstable tree of the
+	 * old pass are removed from it and are later reinserted in
+	 * the new unstable tree of the current pass. So if the checks
+	 * run in the race window between removal and re-insertion, it
+	 * can lead to unexpected false positives where page_volatile
+	 * is elevated and page_unshared is recessed.
+	 */
+	SAFE_FILE_SCANF(PATH_KSM "run", "%d", &ksm_run_orig);
+	SAFE_FILE_PRINTF(PATH_KSM "run", "0");
+
 	check("pages_shared", pages_shared);
 	check("pages_sharing", pages_sharing);
 	check("pages_volatile", pages_volatile);
 	check("pages_unshared", pages_unshared);
 	check("sleep_millisecs", sleep_millisecs);
 	check("pages_to_scan", pages_to_scan);
+
+	SAFE_FILE_PRINTF(PATH_KSM "run", "%d", ksm_run_orig);
 }
 
 void ksm_group_check(int run, int pages_shared, int pages_sharing,
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
Reviewed-by on v1 1/1
#2 lore raw

Andrea Arcangeli <[email protected]> wrote:

The KSM scan keeps running while checking the KSM status in sysfs, but

Quoted 8 lines

during the KSM scan the the rmap_items in the stale unstable tree of the old pass are removed from it and are later reinserted in the new unstable tree of the current pass, leading to some page_volatile false positives.

The fix is stop the KSM scan temporarily while the KSM status is being read from sysfs.

Quoted 9 lines

For reference here's an instance of the fixed false positives:

mem.c:255: TFAIL: pages_volatile is not 0 but 1.
mem.c:255: TFAIL: pages_unshared is not 1 but 0.

This email address is valid but I prefer to use the my name liwang@ for better remembering :). (Anyway this doesn't matter, we can ignore this.)

Quoted 3 lines

@Cryril, I vote for adding this patch in the new release. Plz take this into consideration, thanks.

Signature · 2 lines
Regards,
Li Wang
Signature · 2 lines
-- 
Mailing list info: https://lists.linux.it/listinfo/ltp