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

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

On 2014/8/1 21:47, Jan Beulich wrote:
>>>> On 01.08.14 at 11:50, <tiejun.chen@intel.com> wrote:
>> On 2014/8/1 15:21, Jan Beulich wrote:
>>>>>> On 01.08.14 at 09:10, <tiejun.chen@intel.com> wrote:
>>>> On 2014/8/1 14:51, Jan Beulich wrote:
>>>>>>>> On 31.07.14 at 11:45, <tiejun.chen@intel.com> wrote:
>>>>>> 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?
>>>>>
>>>>> I don't think any HVM guest would ever call this, even more so that
>>>>> the call is restricted to Dom0. The reservation needs to be done
>>>>
>>>> Thanks for your correction. Actually I'm also afraid I may not find a
>>>> correct place where I want to go indeed since I'm not familiar this
>>>> process.
>>>>
>>>>> when guest memory gets populated (and its E820 constructed).
>>>>
>>>> Could you hint me where this action is covered?
>>>
>>> Memory population happens in tools/libxc/xc_hvm_build.c:setup_guest(),
>>> the E820 for the guest gets constructed in hvmloader (just grep for
>>> [eE]820).
>>>
>>
>> Thanks for your information.
>>
>> With further looking into this, instead of XENMEM_machine_memory_map, I
>> think we can go XENMEM_set_memory_map path, right?
>
> I'm not sure - this may be an additional piece to be done for
> consistency (if the domain builder doesn't already call this), but
> since hvmloader doesn't appear to call XENMEM_memory_map it
> won't do on its own I'm afraid.
>

Yes, current hvmloader can't do this on its own.

But in PV case, e820_host, seems be a refereed way to our goal. Even we 
may reuse some codes here so its a convenient approach.

Additionally, I want to know if patch v6 is fine to be acked.

Thanks
Tiejun