Skip to contentSkip to conversation

mtest01: correct the ALLOC_THRESHOLD definition on s390x

PATCH New Li Wang 3 participants 4 messages Nov 28, 2019 – Dec 2, 2019

lore
Share & tools
lore

Li Wang

author
#0 lore raw

PATCH 1/1 mtest01: correct the ALLOC_THRESHOLD definition on s390x

mtest01 hits a problem on s390x platform. The situation is that if children's memory allocating is ongoing and the test remaining time is in an emergency, the parent will break from the while loop and try to revoke children, obviously, it doesn't have enough time to wait for the children's status change to 'T'. Then it occurs timeout issue as below:

mtest01.c:134: INFO: Filling up 80% of free ram which is 5868864 kbytes mtest01.c:149: INFO: ... child 38289 starting
mtest01.c:149: INFO: ... child 38288 starting
mtest01.c:208: WARN: the remaininig time is not enough for testing mtest01.c:218: FAIL: kbytes allocated (and written to) less than expected 5868864 Test timeouted, sending SIGKILL!

The immediate cause is the memory allocating is too slow to finish fleetly on such a small virtual s390x machine, because ALLOC_THRESHOLD does not take real effort. Here let's correct the allocating threshold definition to make sure each child allocates less memory.

And, another fix for the fail handling. In case children are still on memory allocating, the remaining 15 seconds is not enough to wait for process status change. We kill them directly to avoid test timeout occurring.

Cc Rachel Sibley <[email protected]>
---
 testcases/kernel/mem/mtest01/mtest01.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
testcases/kernel/mem/mtest01/mtest01.c +9 -4
index 960a2cef8..446d26897 100644
@@ -33,7 +33,7 @@
 
 #define FIVE_HUNDRED_MB         (500ULL*1024*1024)
 
-#if defined(_s390_)
+#if defined(__s390__) || defined(__s390x__)
 #define ALLOC_THRESHOLD		FIVE_HUNDRED_MB
 #elif defined(TST_ABI32)
 #define ALLOC_THRESHOLD		(2*FIVE_HUNDRED_MB)
@@ -216,11 +216,16 @@ static void mem_test(void)
 	if (children_done < pid_cntr) {
 		tst_res(TFAIL, "kbytes allocated %sless than expected %llu",
 				write_msg, alloc_maxbytes / 1024);
-	} else {
-		tst_res(TPASS, "%llu kbytes allocated %s",
-				alloc_maxbytes / 1024, write_msg);
+
+		for (i = 0; i < pid_cntr; i++)
+			kill(pid_list[i], SIGKILL);
+
+		return;
 	}
 
+	tst_res(TPASS, "%llu kbytes allocated %s",
+			alloc_maxbytes / 1024, write_msg);
+
 	for (i = 0; i < pid_cntr; i++) {
 		TST_PROCESS_STATE_WAIT(pid_list[i], 'T');
 		kill(pid_list[i], SIGCONT);
Signature · 1 line
2.20.1
#1 lore raw

PATCH 1/1 mtest01: correct the ALLOC_THRESHOLD definition on s390x

Hi Li,

Quoted 18 lines

mtest01 hits a problem on s390x platform. The situation is that if children's memory allocating is ongoing and the test remaining time is in an emergency, the parent will break from the while loop and try to revoke children, obviously, it doesn't have enough time to wait for the children's status change to 'T'. Then it occurs timeout issue as below:

mtest01.c:134: INFO: Filling up 80% of free ram which is 5868864 kbytes mtest01.c:149: INFO: ... child 38289 starting
mtest01.c:149: INFO: ... child 38288 starting
mtest01.c:208: WARN: the remaininig time is not enough for testing mtest01.c:218: FAIL: kbytes allocated (and written to) less than expected 5868864 Test timeouted, sending SIGKILL!

The immediate cause is the memory allocating is too slow to finish fleetly on such a small virtual s390x machine, because ALLOC_THRESHOLD does not take real effort. Here let's correct the allocating threshold definition to make sure each child allocates less memory.

Quoted 6 lines

And, another fix for the fail handling. In case children are still on memory allocating, the remaining 15 seconds is not enough to wait for process status change. We kill them directly to avoid test timeout occurring.

Cc Rachel Sibley <[email protected]>

Kind regards,
Petr

#2 lore raw

PATCH 1/1 mtest01: correct the ALLOC_THRESHOLD definition on s390x

In reply to Li Wang

----- Original Message -----

Quoted 9 lines

mtest01 hits a problem on s390x platform. The situation is that if children's memory allocating is ongoing and the test remaining time is in an emergency, the
parent will break from the while loop and try to revoke children, obviously, it
doesn't have enough time to wait for the children's status change to 'T'. Then it
occurs timeout issue as below:

Quoted 7 lines

mtest01.c:134: INFO: Filling up 80% of free ram which is 5868864 kbytes mtest01.c:149: INFO: ... child 38289 starting
mtest01.c:149: INFO: ... child 38288 starting
mtest01.c:208: WARN: the remaininig time is not enough for testing mtest01.c:218: FAIL: kbytes allocated (and written to) less than expected 5868864
Test timeouted, sending SIGKILL!

I/O on that system looks painfully slow, if it can't fill ~5G of swap in 5 minutes.

author
#3 lore raw

PATCH 1/1 mtest01: correct the ALLOC_THRESHOLD definition on s390x

Pushed, thanks for the review.

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20191202/78ce45e4/attachment.htm>