Skip to contentSkip to conversation

Unshare KSM pages before setting max_page_sharing

PATCH New Martin Doucha 3 participants 6 messages Nov 28, 2019 – Dec 4, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Unshare KSM pages before setting max_page_sharing

Setting max_page_sharing is possible only when there are no KSM shared pages in the system. Otherwise writing to max_page_sharing SysFS file will fail with EBUSY.

---
 testcases/kernel/mem/lib/mem.c | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
testcases/kernel/mem/lib/mem.c +10 -7
index a0c1b9b00..456259792 100644
@@ -482,8 +482,11 @@ void create_same_memory(int size, int num, int unit)
 	stop_ksm_children(child, num);
 
 	tst_res(TINFO, "KSM merging...");
-	if (access(PATH_KSM "max_page_sharing", F_OK) == 0)
+	if (access(PATH_KSM "max_page_sharing", F_OK) == 0) {
+		SAFE_FILE_PRINTF(PATH_KSM "run", "2");
 		SAFE_FILE_PRINTF(PATH_KSM "max_page_sharing", "%ld", size * pages * num);
+	}
+
 	SAFE_FILE_PRINTF(PATH_KSM "run", "1");
 	SAFE_FILE_PRINTF(PATH_KSM "pages_to_scan", "%ld", size * pages * num);
 	SAFE_FILE_PRINTF(PATH_KSM "sleep_millisecs", "0");
@@ -571,16 +574,16 @@ void test_ksm_merge_across_nodes(unsigned long nr_pages)
 	SAFE_FILE_PRINTF(PATH_KSM "sleep_millisecs", "0");
 	SAFE_FILE_PRINTF(PATH_KSM "pages_to_scan", "%ld",
 			 nr_pages * num_nodes);
-	if (access(PATH_KSM "max_page_sharing", F_OK) == 0)
-		SAFE_FILE_PRINTF(PATH_KSM "max_page_sharing",
-			"%ld", nr_pages * num_nodes);
 	/*
-	 * merge_across_nodes setting can be changed only when there
-	 * are no ksm shared pages in system, so set run 2 to unmerge
-	 * pages first, then to 1 after changing merge_across_nodes,
+	 * merge_across_nodes and max_page_sharing setting can be changed
+	 * only when there are no ksm shared pages in system, so set run 2
+	 * to unmerge pages first, then to 1 after changing merge_across_nodes,
 	 * to remerge according to the new setting.
 	 */
 	SAFE_FILE_PRINTF(PATH_KSM "run", "2");
+	if (access(PATH_KSM "max_page_sharing", F_OK) == 0)
+		SAFE_FILE_PRINTF(PATH_KSM "max_page_sharing",
+			"%ld", nr_pages * num_nodes);
 	tst_res(TINFO, "Start to test KSM with merge_across_nodes=1");
 	SAFE_FILE_PRINTF(PATH_KSM "merge_across_nodes", "1");
 	SAFE_FILE_PRINTF(PATH_KSM "run", "1");
Signature · 1 line
2.24.0
#1 lore raw

PATCH 1/1 Unshare KSM pages before setting max_page_sharing

Quoted 6 lines · Martin Doucha

Setting max_page_sharing is possible only when there are no KSM shared pages
in the system. Otherwise writing to max_page_sharing SysFS file will fail with EBUSY.

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20191129/5fd0585d/attachment.htm>
#2 lore raw

PATCH 1/1 Unshare KSM pages before setting max_page_sharing

In reply to Martin Doucha

Hi Martin,

Quoted 3 lines

Setting max_page_sharing is possible only when there are no KSM shared pages in the system. Otherwise writing to max_page_sharing SysFS file will fail with EBUSY.

Good catch!

BTW at least ksm01 will fail on VM host with running VMs, but that's another story not related to this patch.

mem.c:251: FAIL: pages_shared is not 1 but 3629. mem.c:251: FAIL: pages_sharing is not 98302 but 113789. mem.c:251: FAIL: pages_volatile is not 0 but 12. mem.c:251: FAIL: pages_unshared is not 1 but 48519.

Kind regards,
Petr

#3 lore raw

PATCH 1/1 Unshare KSM pages before setting max_page_sharing

Quoted 9 lines · Petr Vorel

Hi Martin,

Quoted 1 line

Setting max_page_sharing is possible only when there are no KSM shared

pages

Quoted 2 lines

in the system. Otherwise writing to max_page_sharing SysFS file will fail with EBUSY.

Pushed.

Quoted 6 lines

Good catch!

BTW at least ksm01 will fail on VM host with running VMs, but that's another story not related to this patch.

AFAIK, KSM is mainly designed for VMs pages merging. It seems the LTP ksm test can't run in parallel too, they are suggested to be the only test without any other ksm-test running in the background. Because ksmd does not distinguish which process the pages belong to, it just does merging and counting work in total for the whole system.

If we run ./ksm01 & ./ksm01, it will also fail like below I think.

Quoted 6 lines

mem.c:251: FAIL: pages_shared is not 1 but 3629. mem.c:251: FAIL: pages_sharing is not 98302 but 113789. mem.c:251: FAIL: pages_volatile is not 0 but 12. mem.c:251: FAIL: pages_unshared is not 1 but 48519.

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20191202/8c711629/attachment.htm>
#4 lore raw

PATCH 1/1 Unshare KSM pages before setting max_page_sharing

In reply to Petr Vorel
Quoted 7 lines · Petr Vorel

BTW at least ksm01 will fail on VM host with running VMs, but that's another story not related to this patch.

mem.c:251: FAIL: pages_shared is not 1 but 3629. mem.c:251: FAIL: pages_sharing is not 98302 but 113789. mem.c:251: FAIL: pages_volatile is not 0 but 12. mem.c:251: FAIL: pages_unshared is not 1 but 48519.

I'm aware of other KSM tests failing but the failures are rare and quite random so they'll be tricky to debug. The snippet you've sent looks like interference from another process running on the same VM but the failures I see in OpenQA look like some sort of race condition in kernel itself.

Signature · 7 lines
Martin Doucha   [email protected]
QA Engineer for Software Maintenance
SUSE LINUX, s.r.o.
CORSO IIa
Krizikova 148/34
186 00 Prague 8
Czech Republic
#5 lore raw

PATCH 1/1 Unshare KSM pages before setting max_page_sharing

Hi,

Quoted 8 lines
Quoted 7 lines

BTW at least ksm01 will fail on VM host with running VMs, but that's another story not related to this patch.

mem.c:251: FAIL: pages_shared is not 1 but 3629. mem.c:251: FAIL: pages_sharing is not 98302 but 113789. mem.c:251: FAIL: pages_volatile is not 0 but 12. mem.c:251: FAIL: pages_unshared is not 1 but 48519.

Quoted 5 lines

I'm aware of other KSM tests failing but the failures are rare and quite random so they'll be tricky to debug. The snippet you've sent looks like interference from another process running on the same VM but the failures I see in OpenQA look like some sort of race condition in kernel itself.

Yep, my failures are for sure different. I run the test on my workstation (KVM host), after start of about 15 VMs. Before test was ok. I haven't tested, but it should be ok when running on KVM guests.

Kind regards,
Petr