Re: [RFC v2 1/1] memory: Delete assertion in memory_region_unregister_iommu_notifier

[Jason Wang <jasowang@redhat.com>]

On 2020/7/1 下午8:16, Peter Xu wrote:
> On Wed, Jul 01, 2020 at 04:11:49PM +0800, Jason Wang wrote:
>> On 2020/6/30 下午11:20, Peter Xu wrote:
>>> On Tue, Jun 30, 2020 at 05:23:31PM +0800, Jason Wang wrote:
>>>>>> Ok, we had a dedicated mr for interrupt:
>>>>>>
>>>>>> memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
>>>>>> VTD_INTERRUPT_ADDR_FIRST,
>>>>>> &vtd_dev_as->iommu_ir, 1);
>>>>>>
>>>>>> So it should be fine. I guess the reason that I'm asking is that I thought
>>>>>> "IR" means "Interrupt remapping" but in fact it means "Interrupt Region"?
>>> I was meaning "interrupt remapping", and of course it's the interrupt region
>>> too when IR enabled...
>>
>> Right.
>>
>>
>>>>>> But I'm still not clear about the invalidation part for interrupt region,
>>>>>> maybe you can elaborate a little more on this.
>>>>>>
>>>>>> Btw, I think guest can trigger the assert in vtd_do_iommu_translate() if we
>>>>>> teach vhost to DMA to that region:
>>>>> Why would we want to?
>>>> I meant a buggy(malicious) guest driver.
>>> Yes seems possible.  Do you want to post a patch?  Let me know if you want me
>>> to...  Thanks,
>>
>> Yes please.
> Oh wait...  Actually the comment above explains...
>
>      /*
>       * We have standalone memory region for interrupt addresses, we
>       * should never receive translation requests in this region.
>       */
>      assert(!vtd_is_interrupt_addr(addr));
>
> I overlooked myself that the IR region will be there even if ir=off.


Yes, but the point stands still but the issue is still if ir=off.


>    So I
> think the assert should stand.


Do you mean vhost can't trigger the assert()? If yes, I don't get how it 
can't.

Thanks


>