Re: [PATCH] sched: fix infinity loop in update_blocked_averages

From: Vincent Guittot
Date: Thu Dec 27 2018 - 12:03:49 EST

On Thu, 27 Dec 2018 at 17:40, Sargun Dhillon <sargun@xxxxxxxxx> wrote:
>
> On Thu, Dec 27, 2018 at 5:23 AM Vincent Guittot
> <vincent.guittot@xxxxxxxxxx> wrote:
> >
> > Adding Sargun and Dimitry who faced similar problem
> > Adding Tejun
> >
> > On Thu, 27 Dec 2018 at 11:21, Vincent Guittot
> > <vincent.guittot@xxxxxxxxxx> wrote:
> > >
> > > Le Thursday 27 Dec 2018 Ã 10:21:53 (+0100), Vincent Guittot a Ãcrit :
> > > > Hi Xie,
> > > >
> > > > On Thu, 27 Dec 2018 at 03:57, Xie XiuQi <xiexiuqi@xxxxxxxxxx> wrote:
> > > > >
> > > > > Zhepeng Xie report a bug, there is a infinity loop in
> > > > > update_blocked_averages().
> > > > >
> > > > > PID: 14233 TASK: ffff800b2de08fc0 CPU: 1 COMMAND: "docker"
> > > > > #0 [ffff00002213b9d0] update_blocked_averages at ffff00000811e4a8
> > > > > #1 [ffff00002213ba60] pick_next_task_fair at ffff00000812a3b4
> > > > > #2 [ffff00002213baf0] __schedule at ffff000008deaa88
> > > > > #3 [ffff00002213bb70] schedule at ffff000008deb1b8
> > > > > #4 [ffff00002213bb80] futex_wait_queue_me at ffff000008180754
> > > > > #5 [ffff00002213bbd0] futex_wait at ffff00000818192c
> > > > > #6 [ffff00002213bd00] do_futex at ffff000008183ee4
> > > > > #7 [ffff00002213bde0] __arm64_sys_futex at ffff000008184398
> > > > > #8 [ffff00002213be60] el0_svc_common at ffff0000080979ac
> > > > > #9 [ffff00002213bea0] el0_svc_handler at ffff000008097a6c
> > > > > #10 [ffff00002213bff0] el0_svc at ffff000008084044
> > > > >
> > > > > rq->tmp_alone_branch introduced in 4.10, used to point to
> > > > > the new beg of the list. If this cfs_rq is deleted somewhere
> > > > > else, then the tmp_alone_branch will be illegal and cause
> > > > > a list_add corruption.
> > > >
> > > > shouldn't all the sequence be protected by rq_lock ?
> > > >
> > > >
> > > > >
> > > > > (When enabled DEBUG_LIST, we fould this list_add corruption)
> > > > >
> > > > > [ 2546.741103] list_add corruption. next->prev should be prev
> > > > > (ffff800b4d61ad40), but was ffff800ba434fa38. (next=ffff800b6a95e740).
> > > > > [ 2546.741130] ------------[ cut here ]------------
> > > > > [ 2546.741132] kernel BUG at lib/list_debug.c:25!
> > > > > [ 2546.741136] Internal error: Oops - BUG: 0 [#1] SMP
> > > > > [ 2546.742870] CPU: 1 PID: 29428 Comm: docker-runc Kdump: loaded Tainted: G E 4.19.5-1.aarch64 #1
> > > > > [ 2546.745415] Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
> > > > > [ 2546.747402] pstate: 40000085 (nZcv daIf -PAN -UAO)
> > > > > [ 2546.749015] pc : __list_add_valid+0x50/0x90
> > > > > [ 2546.750485] lr : __list_add_valid+0x50/0x90
> > > > > [ 2546.751975] sp : ffff00001b5eb910
> > > > > [ 2546.753286] x29: ffff00001b5eb910 x28: ffff800abacf0000
> > > > > [ 2546.754976] x27: ffff00001b5ebbb0 x26: ffff000009570000
> > > > > [ 2546.756665] x25: ffff00000960d000 x24: 00000250f41ca8f8
> > > > > [ 2546.758366] x23: ffff800b6a95e740 x22: ffff800b4d61ad40
> > > > > [ 2546.760066] x21: ffff800b4d61ad40 x20: ffff800ba434f080
> > > > > [ 2546.761742] x19: ffff800b4d61ac00 x18: ffffffffffffffff
> > > > > [ 2546.763425] x17: 0000000000000000 x16: 0000000000000000
> > > > > [ 2546.765089] x15: ffff000009570748 x14: 6666662073617720
> > > > > [ 2546.766755] x13: 747562202c293034 x12: 6461313664346230
> > > > > [ 2546.768429] x11: 3038666666662820 x10: 0000000000000000
> > > > > [ 2546.770124] x9 : 0000000000000001 x8 : ffff000009f34a0f
> > > > > [ 2546.771831] x7 : 0000000000000000 x6 : 000000000000250d
> > > > > [ 2546.773525] x5 : 0000000000000000 x4 : 0000000000000000
> > > > > [ 2546.775227] x3 : 0000000000000000 x2 : 70ef7f624013ca00
> > > > > [ 2546.776929] x1 : 0000000000000000 x0 : 0000000000000075
> > > > > [ 2546.778623] Process docker-runc (pid: 29428, stack limit = 0x00000000293494a2)
> > > > > [ 2546.780742] Call trace:
> > > > > [ 2546.781955] __list_add_valid+0x50/0x90
> > > > > [ 2546.783469] enqueue_entity+0x4a0/0x6e8
> > > > > [ 2546.784957] enqueue_task_fair+0xac/0x610
> > > > > [ 2546.786502] sched_move_task+0x134/0x178
> > > > > [ 2546.787993] cpu_cgroup_attach+0x40/0x78
> > > > > [ 2546.789540] cgroup_migrate_execute+0x378/0x3a8
> > > > > [ 2546.791169] cgroup_migrate+0x6c/0x90
> > > > > [ 2546.792663] cgroup_attach_task+0x148/0x238
> > > > > [ 2546.794211] __cgroup1_procs_write.isra.2+0xf8/0x160
> > > > > [ 2546.795935] cgroup1_procs_write+0x38/0x48
> > > > > [ 2546.797492] cgroup_file_write+0xa0/0x170
> > > > > [ 2546.799010] kernfs_fop_write+0x114/0x1e0
> > > > > [ 2546.800558] __vfs_write+0x60/0x190
> > > > > [ 2546.801977] vfs_write+0xac/0x1c0
> > > > > [ 2546.803341] ksys_write+0x6c/0xd8
> > > > > [ 2546.804674] __arm64_sys_write+0x24/0x30
> > > > > [ 2546.806146] el0_svc_common+0x78/0x100
> > > > > [ 2546.807584] el0_svc_handler+0x38/0x88
> > > > > [ 2546.809017] el0_svc+0x8/0xc
> > > > >
> > > >
> > > > Have you got more details about the sequence that generates this bug ?
> > > > Is it easily reproducible ?
> > > >
> > > > > In this patch, we move rq->tmp_alone_branch point to its prev before delete it
> > > > > from list.
> > > > >
> > > > > Reported-by: Zhipeng Xie <xiezhipeng1@xxxxxxxxxx>
> > > > > Cc: Bin Li <huawei.libin@xxxxxxxxxx>
> > > > > Cc: <stable@xxxxxxxxxxxxxxx> [4.10+]
> > > > > Fixes: 9c2791f936ef (sched/fair: Fix hierarchical order in rq->leaf_cfs_rq_list)
> > > >
> > > > If it only happens in update_blocked_averages(), the del leaf has been added by:
> > > > a9e7f6544b9c (sched/fair: Fix O(nr_cgroups) in load balance path)
> > > >
> > > > > Signed-off-by: Xie XiuQi <xiexiuqi@xxxxxxxxxx>
> > > > > Tested-by: Zhipeng Xie <xiezhipeng1@xxxxxxxxxx>
> > > > > ---
> > > > > kernel/sched/fair.c | 5 +++++
> > > > > 1 file changed, 5 insertions(+)
> > > > >
> > > > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > > > > index ac855b2..7a72702 100644
> > > > > --- a/kernel/sched/fair.c
> > > > > +++ b/kernel/sched/fair.c
> > > > > @@ -347,6 +347,11 @@ static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
> > > > > static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
> > > > > {
> > > > > if (cfs_rq->on_list) {
> > > > > + struct rq *rq = rq_of(cfs_rq);
> > > > > +
> > > > > + if (rq->tmp_alone_branch == &cfs_rq->leaf_cfs_rq_list)
> > > > > + rq->tmp_alone_branch = cfs_rq->leaf_cfs_rq_list.prev;
> > > > > +
> > >
> > > I'm afraid that your patch will break the ordering of leaf_cfs_rq_list
> > >
> > > Can you tried the patch below:
> > >
> > > ---
> > > kernel/sched/fair.c | 7 -------
> > > 1 file changed, 7 deletions(-)
> > >
> > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > > index ca46964..4d51b2d 100644
> > > --- a/kernel/sched/fair.c
> > > +++ b/kernel/sched/fair.c
> > > @@ -7694,13 +7694,6 @@ static void update_blocked_averages(int cpu)
> > > if (se && !skip_blocked_update(se))
> > > update_load_avg(cfs_rq_of(se), se, 0);
> > >
> > > - /*
> > > - * There can be a lot of idle CPU cgroups. Don't let fully
> > > - * decayed cfs_rqs linger on the list.
> > > - */
> > > - if (cfs_rq_is_decayed(cfs_rq))
> > > - list_del_leaf_cfs_rq(cfs_rq);
> > > -
> > > /* Don't need periodic decay once load/util_avg are null */
> > > if (cfs_rq_has_blocked(cfs_rq))
> > > done = false;
> > > --
> > > 2.7.4
> > >
> > >
> > >
> > >
> Tested-by: Sargun Dhillon <sargun@xxxxxxxxx>

Thanks

>
> This patch fixes things for me. I imagine this code that's being
> removed has a purpose?

In the original behavior, the cs_rq was removed from the list only
when the cgroup was removed.
patch a9e7f6544b9c (sched/fair: Fix O(nr_cgroups) in load balance
path) has added an optimization which remove the cfs_rq when there
were no blocked load to update in order to optimize the loop but it
has introduced a race condition that create this infinite loop. The
patch fixes the problem by removing the optimization.
I will look at re-adding the optimization once i will have afix for
the race condition


> > > > > list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
> > > > > cfs_rq->on_list = 0;
> > > > > }
> > > > > --
> > > > > 1.8.3.1
> > > > >