ubuntu_ltp sched:starvation fails with ARM64 on bos2 region

[Kevin Becker]

For the 2024.03.04 cycle and s2024.02.05 cycle, ubuntu_ltp sched:starvation failed on j/realtime on bos2 region. Switching it to bos1 region allowed it to pass during the s2024.02.05 cycle.

Examples of failure from 2024.03.04:

tst_test.c:1741: TINFO: LTP version: 20230929-407-g58aa31f55
tst_test.c:1625: TINFO: Timeout per run is 0h 00m 30s
starvation.c:57: TPASS: sched_setaffinity(0, sizeof(mask), &mask) returned 0
tst_test.c:1633: TINFO: Updating max runtime to 0h 04m 00s
tst_test.c:1625: TINFO: Timeout per run is 0h 04m 30s
Test timeouted, sending SIGKILL!
tst_test.c:1673: TINFO: Killed the leftover descendant processes
tst_test.c:1679: TINFO: If you are running on slow machine, try exporting LTP_TIMEOUT_MUL > 1
tst_test.c:1681: TBROK: Test killed! (timeout?)

[Kevin Becker (kevinbecker)]

Running the test on bos1 for the 2024.03.04 cycle made the test pass, similar to s2024.02.05.

[Po-Hsu Lin (cypressyew)]

This issue can be found on F-5.15 64k as well.

Experiments with F-5.15.0-102-generic-64k shows it will take:
 * 4m37.231s
 * 4m32.198s
 * 4m32.442s
 * 4m33.488s
 * 4m34.977s
 * 4m33.371s
 * 4m33.577s
 * 4m32.672s
 * 4m34.560s
 * 4m37.162s
For this test to run on bos02. Just slightly above the 4m30s threshold.

Setting LTP_TIMEOUT_MUL=2 can bump the timeout to 5m00s and this test will pass.

[Po-Hsu Lin (cypressyew)]

Experiments with J-5.15.0-1058-realtime on bos02 ARM64 shows it will take:
  * 5m4.261s
  * 5m8.267s
  * 5m9.418s
  * 5m11.923s
  * 5m4.715s
  * 5m17.411s
  * 5m11.225s
  * 5m7.947s
  * 5m16.900s
  * 5m28.553s

In this case we will need LTP_TIMEOUT_MUL=3 to bump the LTP timeout to 5m30s (or even set it to 4).
Thoughts?

BTW I have a hint added for F-5.15.0-102-generic-64k.

[Po-Hsu Lin (cypressyew)]

Test shows 4m30s is not enough for bare-metal ARM64 node starmie-kernel with N-realtime 6.8.1-1002.2.

[Po-Hsu Lin (cypressyew)]

Testing timeout multiplier on node starmie with N-realtime 6.8.1-1002.2.

[Po-Hsu Lin (cypressyew)]

This test took 5 minutes and 29 seconds to run on node starmie with N-realtime 6.8.1-1002.2.

INFO: Test start time: Wed Jun 12 19:24:57 UTC 2024
COMMAND: /opt/ltp/bin/ltp-pan -q -e -S -a 125277 -n 125277 -f /tmp/ltp-d3rpbGfBN7/alltests -l /dev/null -C /dev/null -T /dev/null
LOG File: /dev/null
FAILED COMMAND File: /dev/null
TCONF COMMAND File: /dev/null
Running tests.......
tst_test.c:1733: TINFO: LTP version: 20230929-607-g8f21ebba4
tst_test.c:1617: TINFO: Timeout per run is 0h 05m 00s
starvation.c:57: TPASS: sched_setaffinity(0, sizeof(mask), &mask) returned 0
tst_test.c:1625: TINFO: Updating max runtime to 0h 04m 00s
tst_test.c:1617: TINFO: Timeout per run is 0h 09m 00s
starvation.c:100: TPASS: wait_for_pid(child_pid) passed

Summary:
passed 2
failed 0
broken 0
skipped 0
warnings 0
INFO: ltp-pan reported all tests PASS
LTP Version: 20230929-607-g8f21ebba4
INFO: Test end time: Wed Jun 12 19:30:26 UTC 2024

[Po-Hsu Lin (cypressyew)]

This test took 4 minutes and 34 seconds to run on node starmie with N-realtime 6.8.1-1008
Combined with the test result in #6 I think it's better to set the multiplier to 4 (6 minutes)

[Po-Hsu Lin (cypressyew)]

Patch landed, hints removed.