hulk inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I8KZ3H CVE: NA
----------------------------------------
In the test case like below, there is change going to hang system in stop affinity timer function.
for i in `seq 1 100000` do echo $1 > /sys/fs/cgroup/cpu/cpu.dynamic_affinity_mode 2> &1 done
[ 259.061097] CPU: 0 PID: 10223 Comm: sh Kdump: loaded Not tainted 5.10.0OLK-5.10-SP3_SG+ #102 [ 259.061098] Hardware name: Huawei TaiShan 200 (Model 5280)/BC82AMDD, BIOS 1.08 12/14/2019 [ 259.061098] pstate: 60400089 (nZCv daIf +PAN -UAO -TCO BTYPE=--) [ 259.061099] pc : hrtimer_active+0x2c/0x7c [ 259.061099] lr : hrtimer_cancel+0x3c/0x60 [ 259.061100] sp : ffff800167f33c10 [ 259.061100] x29: ffff800167f33c10 x28: ffff003006bde040 [ 259.061101] x27: 0000000000000000 x26: 0000000000000000 [ 259.061102] x25: 0000000000000000 x24: ffff0020b2cdea20 [ 259.061103] x23: ffff800167f33d48 x22: ffff800011cc6100 [ 259.061105] x21: ffff80001170a120 x20: ffff800011728d88 [ 259.061106] x19: ffff00208e53c418 x18: 0000000000000000 [ 259.061107] x17: 0000000000000000 x16: 0000000000000000 [ 259.061108] x15: 0000aaab02369e80 x14: 0000000000000000 [ 259.061109] x13: 0000000000000000 x12: 0000000000000000 [ 259.061110] x11: 0000000000000000 x10: 0000000000000008 [ 259.061111] x9 : ffff80001017c6bc x8 : 00000000ffffffc9 [ 259.061112] x7 : ffff0020cb082581 x6 : 0000000000000000 [ 259.061114] x5 : ffff0020cb082582 x4 : 0000000000000000 [ 259.061115] x3 : ffff202fffb9adc0 x2 : ffff202fffb9ae00 [ 259.061116] x1 : 0000000000003b82 x0 : ffff00208e53c418 [ 259.061117] Call trace: [ 259.061117] hrtimer_active+0x2c/0x7c [ 259.061118] stop_auto_affinity+0xf0/0x170 [ 259.061118] cpu_affinity_mode_write_u64+0x2c/0x60 [ 259.061119] cgroup_file_write+0x128/0x1b0 [ 259.061119] kernfs_fop_write_iter+0x130/0x1c0 [ 259.061120] new_sync_write+0xec/0x18c [ 259.061121] vfs_write+0x214/0x2ac [ 259.061121] ksys_write+0x70/0xfc [ 259.061122] __arm64_sys_write+0x24/0x30 [ 259.061122] invoke_syscall+0x50/0x11c [ 259.061123] el0_svc_common.constprop.0+0x158/0x164 [ 259.061123] do_el0_svc+0x2c/0xac [ 259.061123] el0_svc+0x20/0x30 [ 259.061124] el0_sync_handler+0xb0/0xb4 [ 259.061124] el0_sync+0x160/0x180
The root cause was that hrtimer_cancel are holding spin_lock, which the timer are running.
Thread: hrtimer:
acqurie lock acqurie lock cancel hrtimer sync ... // deadlock ... release lock release lock
Thread will wait hrtimer exit forever, and the hrtimer are waiting spin_ lock forever.
Fixes: 16980bd872f6 ("sched: Introduce smart grid scheduling strategy for cfs") Signed-off-by: Yipeng Zou zouyipeng@huawei.com --- kernel/sched/fair.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index ccd2a060c2df..ce6bb4304949 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -5403,6 +5403,12 @@ static enum hrtimer_restart sched_auto_affi_period_timer(struct hrtimer *timer) }
raw_spin_lock_irqsave(&auto_affi->lock, flags); + /* May be re-entrant by stop_auto_affinity, So check again. */ + if (auto_affi->period_active == 0) { + raw_spin_unlock_irqrestore(&auto_affi->lock, flags); + return HRTIMER_NORESTART; + } + if (util_avg_sum * 100 >= tg_capacity * sysctl_sched_util_low_pct) { affinity_domain_up(tg); } else if (util_avg_sum * 100 < tg_capacity * @@ -5496,8 +5502,6 @@ void stop_auto_affinity(struct auto_affinity *auto_affi) mutex_unlock(&smart_grid_used_mutex); return; } - - hrtimer_cancel(&auto_affi->period_timer); auto_affi->period_active = 0; auto_affi->mode = 0; ad->curr_level = ad->dcount > 0 ? ad->dcount - 1 : 0;
-
Yipeng Zou