KVM: BUG during OOM on host

KVM: BUG during OOM on host

[Wanpeng Li]

Hi all,

When the OOM-Killer kills the guest which is running memory hog
processes, there is tons of splatting as below: the page count reaches
0 before drop_spte. The splatting will disappear when disables THP on
host.

------------[ cut here ]------------
 WARNING: CPU: 4 PID: 28317 at arch/x86/kvm/mmu.c:682
mmu_spte_clear_track_bits+0x84/0x120 [kvm]
 CPU: 4 PID: 28317 Comm: kworker/4:2 Tainted: G    B D W       4.12.0-rc7+ #29
 Workqueue: events mmput_async_fn
 task: ffff9465283c3280 task.stack: ffffacb0083b4000
 RIP: 0010:mmu_spte_clear_track_bits+0x84/0x120 [kvm]
 Call Trace:
  drop_spte+0x1a/0xb0 [kvm]
  mmu_page_zap_pte+0xcc/0xe0 [kvm]
  kvm_mmu_prepare_zap_page+0x81/0x4c0 [kvm]
  ? _raw_spin_lock+0x65/0x70
  kvm_mmu_invalidate_zap_all_pages+0x171/0x240 [kvm]
  kvm_arch_flush_shadow_all+0xe/0x10 [kvm]
  kvm_mmu_notifier_release+0x54/0x90 [kvm]
  ? kvm_mmu_notifier_release+0x5/0x90 [kvm]
  __mmu_notifier_release+0x65/0x100
  exit_mmap+0x168/0x180
  ? __khugepaged_exit+0xdb/0x140
  ? rcu_read_lock_sched_held+0x79/0x80
  ? kmem_cache_free+0x2f2/0x350
  mmput_async_fn+0x4c/0x140
  process_one_work+0x203/0x690
  worker_thread+0x4e/0x3b0
  kthread+0x117/0x150

Regards,
Wanpeng Li

Re: KVM: BUG during OOM on host

[Paolo Bonzini]

On 29/06/2017 08:21, Wanpeng Li wrote:
> Hi all,
> 
> When the OOM-Killer kills the guest which is running memory hog
> processes, there is tons of splatting as below: the page count reaches
> 0 before drop_spte. The splatting will disappear when disables THP on
> host.

Can you reproduce it in 4.10?

Thanks,

Paolo

> ------------[ cut here ]------------
>  WARNING: CPU: 4 PID: 28317 at arch/x86/kvm/mmu.c:682
> mmu_spte_clear_track_bits+0x84/0x120 [kvm]
>  CPU: 4 PID: 28317 Comm: kworker/4:2 Tainted: G    B D W       4.12.0-rc7+ #29
>  Workqueue: events mmput_async_fn
>  task: ffff9465283c3280 task.stack: ffffacb0083b4000
>  RIP: 0010:mmu_spte_clear_track_bits+0x84/0x120 [kvm]
>  Call Trace:
>   drop_spte+0x1a/0xb0 [kvm]
>   mmu_page_zap_pte+0xcc/0xe0 [kvm]
>   kvm_mmu_prepare_zap_page+0x81/0x4c0 [kvm]
>   ? _raw_spin_lock+0x65/0x70
>   kvm_mmu_invalidate_zap_all_pages+0x171/0x240 [kvm]
>   kvm_arch_flush_shadow_all+0xe/0x10 [kvm]
>   kvm_mmu_notifier_release+0x54/0x90 [kvm]
>   ? kvm_mmu_notifier_release+0x5/0x90 [kvm]
>   __mmu_notifier_release+0x65/0x100
>   exit_mmap+0x168/0x180
>   ? __khugepaged_exit+0xdb/0x140
>   ? rcu_read_lock_sched_held+0x79/0x80
>   ? kmem_cache_free+0x2f2/0x350
>   mmput_async_fn+0x4c/0x140
>   process_one_work+0x203/0x690
>   worker_thread+0x4e/0x3b0
>   kthread+0x117/0x150
> 
> Regards,
> Wanpeng Li
> 

Re: KVM: BUG during OOM on host

[Wanpeng Li]

2017-06-30 1:14 GMT+08:00 Paolo Bonzini <pbonzini@redhat.com>:
>
>
> On 29/06/2017 08:21, Wanpeng Li wrote:
>> Hi all,
>>
>> When the OOM-Killer kills the guest which is running memory hog
>> processes, there is tons of splatting as below: the page count reaches
>> 0 before drop_spte. The splatting will disappear when disables THP on
>> host.
>
> Can you reproduce it in 4.10?

Sorry for the late response, it can't be reproduced in 4.10 Linus tree.

Regards,
Wanpeng Li

>
> Thanks,
>
> Paolo
>
>> ------------[ cut here ]------------
>>  WARNING: CPU: 4 PID: 28317 at arch/x86/kvm/mmu.c:682
>> mmu_spte_clear_track_bits+0x84/0x120 [kvm]
>>  CPU: 4 PID: 28317 Comm: kworker/4:2 Tainted: G    B D W       4.12.0-rc7+ #29
>>  Workqueue: events mmput_async_fn
>>  task: ffff9465283c3280 task.stack: ffffacb0083b4000
>>  RIP: 0010:mmu_spte_clear_track_bits+0x84/0x120 [kvm]
>>  Call Trace:
>>   drop_spte+0x1a/0xb0 [kvm]
>>   mmu_page_zap_pte+0xcc/0xe0 [kvm]
>>   kvm_mmu_prepare_zap_page+0x81/0x4c0 [kvm]
>>   ? _raw_spin_lock+0x65/0x70
>>   kvm_mmu_invalidate_zap_all_pages+0x171/0x240 [kvm]
>>   kvm_arch_flush_shadow_all+0xe/0x10 [kvm]
>>   kvm_mmu_notifier_release+0x54/0x90 [kvm]
>>   ? kvm_mmu_notifier_release+0x5/0x90 [kvm]
>>   __mmu_notifier_release+0x65/0x100
>>   exit_mmap+0x168/0x180
>>   ? __khugepaged_exit+0xdb/0x140
>>   ? rcu_read_lock_sched_held+0x79/0x80
>>   ? kmem_cache_free+0x2f2/0x350
>>   mmput_async_fn+0x4c/0x140
>>   process_one_work+0x203/0x690
>>   worker_thread+0x4e/0x3b0
>>   kthread+0x117/0x150
>>
>> Regards,
>> Wanpeng Li
>>