Re: [Qemu-devel] [PATCH] x86 tlb_fill in memory_helper.c

[Paolo Bonzini <pbonzini@redhat.com>]

On 16/02/2017 14:27, Alexander Boettcher wrote:
> On 16.02.2017 13:57, Paolo Bonzini wrote:
>>
>>
>> On 16/02/2017 13:57, Alexander Boettcher wrote:
>>> Hello Paolo,
>>>
>>> On 16.02.2017 12:53, Paolo Bonzini wrote:
>>>> As you noticed, the problem is that now cpu_vmexit is called without a
>>>> cpu_restore_state before.
>>>>
>>>> The right fix is to pass the retaddr to cpu_vmexit (via
>>>> cpu_svm_check_intercept_param); something like this untested patch:
>>>
>>> ...
>>>
>>>>
>>>> If it works for you we can include it.
>>>
>>> thank you for the quick patch. It works like a charm.
>>>
>>> Do you take care of adding the patch, please ? (I can of course also do.)
>>
>> Great, I will look at it.  Have you debugged the exact failure mode to
>> include something in the commit message?
> 
> I'm not familiar with the SVM model, but what I did was to
> 
> compare mainly the output of the svm exceptions on Qemu 2.4.1 and
> 2.5.0++ as reported by the Nova hypervisor runing in Qemu.
> 
> There one gets mysterious SVM exception 0x4e (PF), where
> 
> virtaddr = cr2 = VMCB::exitinfo2
> 
> cr2 is around the 0 address - where actually is nothing in the guest VM.
> 
> Later on one gets a SVM exception of 0x7f, and then your VM is gone.
> 
> Hope it helps to make up a useful commit message,

Not really, but thanks for trying! :)  I'll just write down what I said
in my first message to you.

Paolo