Re: [v6][PATCH 2/2] xen:vtd: missing RMRR mapping while share EPT

["Chen, Tiejun" <tiejun.chen@intel.com>]

On 2014/8/1 6:44, Tian, Kevin wrote:
>> From: Chen, Tiejun
>> Sent: Thursday, July 31, 2014 2:46 AM
>>
>> On 2014/7/30 18:47, Jan Beulich wrote:
>>>>>> On 30.07.14 at 12:40, <tiejun.chen@intel.com> wrote:
>>>> On 2014/7/30 18:25, Jan Beulich wrote:
>>>>>>>> On 30.07.14 at 11:40, <tiejun.chen@intel.com> wrote:
>>>>>>     From what those codes mean, it just return regardless whether they
>>>>>> really conflict. And this is just a good assumption, so if I'm
>>>>>> understanding this properly, actually our patches do this thing
>>>>>> precisely because we further check if this assumption is true, then take
>>>>>> necessary actions.
>>>>>
>>>>> Except that the pointed out check prevents the code you modify
>>>>> from being reached at all, i.e. as long as that check is there it
>>>>> doesn't matter (for any passed through USB device) what action
>>>>> rmrr_identity_mapping() takes.
>>>>>
>>>>
>>>> Sorry, what do you mean?
>>>>
>>>>    From my point of view these two patches should be better than drop
>>>> simply RMRR for any PT USB device no matter if its really necessary.
>>>
>>> I mean that for USB devices your patches change nothing without
>>> said check also getting removed.
>>>
>>
>> Jan,
>>
>> For USB instance, I think currently we still keep that until we have a
>> complete solution since this is always safe.
>>
>> Additionally, I'm trying to figure out that solution. As I mentioned
>> previously, I think we can reserve all RMRR once when a guest call
>> XENMEM_machine_memory_map to create its own memory. What about this
>> idea? Or other better suggestions?
>>
>
> It's not guest to create its own memory. It's control panel to create guest
> memory layout. We need reserve RMRRs there, and if doing that maybe we

I mean here we can reserve those RMRR memory in e820 as long as we 
create one VM.

> can leave user space to setup identity mapping for RMRR instead of doing
> that in Xen?
>

 From my point of view that may be complex. And I'm not sure if this can 
work out hotplug case.  And especially, the memory range covered by RMRR 
should not be big, right? So I think its acceptable to reserve them even 
guest always don't use them.

Thanks
Tiejun