Re: [help] host kernel panic in kvm's wakeup_handler()

["Longpeng (Mike)" <longpeng2@huawei.com>]

On 2017/5/26 18:40, Paolo Bonzini wrote:

> 
> 
> On 24/05/2017 07:04, Longpeng (Mike) wrote:
>>>> it crashed at *1ec1* and %rax get a wrong value(0xdead000000100100) at *1e92*,
>>>> it seems the *blocked_vcpu_on_cpu* list is corrupted, but kvm only access this
>>>> list in pre_block/post_block/wakeup_handler, and these three functions seems good.
>>>>
>>>> kvm version is 4.4-stable.
>>>>
>>>> Do you have any ideas? Any suggestion would be greatly appreciated, thanks!
>>>>
>>> Is this only seen with posted interrupt support enabled?  Booting with
>>> intremap=nopost on the kernel commandline would disable it.  Thanks,
>>
>> We tested with PI support enabled, but we not sure if it only occurs with PI
>> enabled yet.
> 
> This code should not run at all with PI disabled, since the handler is
> only reachable through an IRTE.
> 
> As you said, the list manipulation in those function is fairly simple.
> If you have a reproducer, you can try running it with CONFIG_LIST_DEBUG
> and see what you get.
> 


OK.

We have already started test for a long time, but didn't panic yet.

Thanks :)

> Thanks,
> 
> Paolo
> 
> .
> 


-- 
Regards,
Longpeng(Mike)