[B][SRU][PATCH 0/5] Fix for cfs_bandwidth01 test in LTP/sched

Tue Jun 22 07:49:10 UTC 2021

From: Po-Hsu Lin <po-hsu.lin at canonical.com>

[Impact]
Test case cfs_bandwidth01 in LTP sched test suite is a reproducer
of a CFS unthrottle_cfs_rq() issue (fe61468b2cbc2b sched/fair: Fix
enqueue_task_fair warning).

This test triggers a warning on our 4.15 kernel:
 LTP: starting cfs_bandwidth01 (cfs_bandwidth01 -i 5)
 ------------[ cut here ]------------
 rq->tmp_alone_branch != &rq->leaf_cfs_rq_list
 WARNING: CPU: 0 PID: 0 at /build/linux-fYK9kF/linux-4.15.0/kernel/sched/fair.c:393 unthrottle_cfs_rq+0x16f/0x200
 Modules linked in: input_leds joydev serio_raw mac_hid qemu_fw_cfg kvm_intel kvm irqbypass sch_fq_codel binfmt_misc ib_iser rdma_cm iw_cm ib_cm ib_core iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi nfsd auth_rpcgss nfs_acl lockd grace sunrpc ip_tables x_tables autofs4 btrfs zstd_compress raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c raid1 raid0 multipath linear cirrus ttm drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops drm psmouse virtio_blk pata_acpi floppy virtio_net i2c_piix4
 CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-144-generic #148-Ubuntu
 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
 RIP: 0010:unthrottle_cfs_rq+0x16f/0x200
 RSP: 0018:ffff989ebfc03e80 EFLAGS: 00010082
 RAX: 0000000000000000 RBX: ffff989eb4c6ac00 RCX: 0000000000000000
 RDX: 0000000000000005 RSI: ffffffffacb63c4d RDI: 0000000000000046
 RBP: ffff989ebfc03ea8 R08: 000000af39e61b33 R09: ffffffffacb63c20
 R10: 0000000000000000 R11: 0000000000000001 R12: ffff989eb57fe400
 R13: ffff989ebfc21900 R14: 0000000000000001 R15: 0000000000000001
 FS: 0000000000000000(0000) GS:ffff989ebfc00000(0000) knlGS:0000000000000000
 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 000055593258d618 CR3: 000000007a044000 CR4: 00000000000006f0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 Call Trace:
 <IRQ>
 distribute_cfs_runtime+0xc3/0x110
 sched_cfs_period_timer+0xff/0x220
 ? sched_cfs_slack_timer+0xd0/0xd0
 __hrtimer_run_queues+0xdf/0x230
 hrtimer_interrupt+0xa0/0x1d0
 smp_apic_timer_interrupt+0x6f/0x140
 apic_timer_interrupt+0x90/0xa0
 </IRQ>
 RIP: 0010:native_safe_halt+0x12/0x20
 RSP: 0018:ffffffffac603e28 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff11
 RAX: ffffffffabbc9280 RBX: 0000000000000000 RCX: 0000000000000000
 RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
 RBP: ffffffffac603e28 R08: 000000af39850067 R09: ffff989e73749d00
 R10: 0000000000000000 R11: 7fffffffffffffff R12: 0000000000000000
 R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
 ? __sched_text_end+0x1/0x1
 default_idle+0x20/0x100
 arch_cpu_idle+0x15/0x20
 default_idle_call+0x23/0x30
 do_idle+0x172/0x1f0
 cpu_startup_entry+0x73/0x80
 rest_init+0xae/0xb0
 start_kernel+0x4dc/0x500
 x86_64_start_reservations+0x24/0x26
 x86_64_start_kernel+0x74/0x77
 secondary_startup_64+0xa5/0xb0
 Code: 50 09 00 00 49 39 85 60 09 00 00 74 68 80 3d 3a 6e 54 01 00 75 5f 31 db 48 c7 c7 c0 3d 2d ac c6 05 28 6e 54 01 01 e8 11 36 fc ff <0f> 0b 48 85 db 74 43 49 8b 85 78 09 00 00 49 39 85 70 09 00 00
 ---[ end trace b6b9a70bc2945c0c ]---

[Fix]
Base on the test case description, we will need these fixes:
  * fe61468b2cbc2b sched/fair: Fix enqueue_task_fair warning
  * b34cb07dde7c23 sched/fair: Fix enqueue_task_fair() warning some more
  * 39f23ce07b9355 sched/fair: Fix unthrottle_cfs_rq() for leaf_cfs_rq list
  * 6d4d22468dae3d sched/fair: Reorder enqueue/dequeue_task_fair path
  * 5ab297bab98431 sched/fair: Fix reordering of enqueue/dequeue_task_fair()

Backport needed for Bionic since we're missing some new variables /
coding style changes introduced in the following commits (and their
corresponding fixes):
  * 97fb7a0a8944bd sched: Clean up and harmonize the coding style of the scheduler code base
  * 9f68395333ad7f sched/pelt: Add a new runnable average signal
  * 6212437f0f6043 sched/fair: Fix runnable_avg for throttled cfs
  * 43e9f7f231e40e sched/fair: Start tracking SCHED_IDLE tasks count in cfs_rq

I have also searched in the upstream tree to see if there is any other
commit claim to be a fix of these but didn't see any.

[Test]
Test kernel can be found here:
https://people.canonical.com/~phlin/kernel/lp-1931325-cfs_bandwidth01/

With these patches applied, the test can pass without triggering this
warning.

<<<test_start>>>
tag=cfs_bandwidth01 stime=1624260713
cmdline="cfs_bandwidth01 -i 5"
contacts=""
analysis=exit
<<<test_output>>>
incrementing stop
tst_test.c:1313: TINFO: Timeout per run is 0h 05m 00s
tst_buffers.c:55: TINFO: Test is using guarded buffers
cfs_bandwidth01.c:49: TINFO: Set 'worker1/cpu.max' = '3000 10000'
cfs_bandwidth01.c:49: TINFO: Set 'worker2/cpu.max' = '2000 10000'
cfs_bandwidth01.c:49: TINFO: Set 'worker3/cpu.max' = '3000 10000'
cfs_bandwidth01.c:113: TPASS: Scheduled bandwidth constrained workers
cfs_bandwidth01.c:49: TINFO: Set 'level2/cpu.max' = '5000 10000'
cfs_bandwidth01.c:125: TPASS: Workers exited
cfs_bandwidth01.c:113: TPASS: Scheduled bandwidth constrained workers
cfs_bandwidth01.c:49: TINFO: Set 'level2/cpu.max' = '5000 10000'
cfs_bandwidth01.c:125: TPASS: Workers exited
cfs_bandwidth01.c:113: TPASS: Scheduled bandwidth constrained workers
cfs_bandwidth01.c:49: TINFO: Set 'level2/cpu.max' = '5000 10000'
cfs_bandwidth01.c:125: TPASS: Workers exited
cfs_bandwidth01.c:113: TPASS: Scheduled bandwidth constrained workers
cfs_bandwidth01.c:49: TINFO: Set 'level2/cpu.max' = '5000 10000'
cfs_bandwidth01.c:125: TPASS: Workers exited
cfs_bandwidth01.c:113: TPASS: Scheduled bandwidth constrained workers
cfs_bandwidth01.c:49: TINFO: Set 'level2/cpu.max' = '5000 10000'
cfs_bandwidth01.c:125: TPASS: Workers exited

Summary:
passed 10
failed 0
broken 0
skipped 0
warnings 0

I have also run the whole sched test suite in LTP to make sure there
is no other issues caused by this patchset.

[Where problems could occur]
* CFS (Completely Fair Scheduler) is the process scheduling system in
the kernel, if the patch is incorrect it might affect the sched
functionality. Especially system with CONFIG_FAIR_GROUP_SCHED and
CONFIG_CFS_BANDWIDTH enabled.

[Other Info]
Test case description:
 * Creates a multi-level CGroup hierarchy with the cpu controller
 * enabled. The leaf groups are populated with "busy" processes which
 * simulate intermittent cpu load. They spin for some time then sleep
 * then repeat.
 *
 * Both the trunk and leaf groups are set cpu bandwidth limits. The
 * busy processes will intermittently exceed these limits. Causing
 * them to be throttled. When they begin sleeping this will then cause
 * them to be unthrottle.

Phil Auld (1):
  sched/fair: Fix enqueue_task_fair() warning some more

Vincent Guittot (4):
  sched/fair: Fix enqueue_task_fair warning
  sched/fair: Reorder enqueue/dequeue_task_fair path
  sched/fair: Fix reordering of enqueue/dequeue_task_fair()
  sched/fair: Fix unthrottle_cfs_rq() for leaf_cfs_rq list

 kernel/sched/fair.c | 109 +++++++++++++++++++++++++++++++++++++---------------
 1 file changed, 77 insertions(+), 32 deletions(-)

-- 
2.7.4