Re: [Xen-devel] XEN tools for ARM64 build issue

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 25.11.16 at 18:59, <julien.grall@xxxxxxx> wrote:
> On 25/11/16 17:00, Julien Grall wrote:
>> On 25/11/16 15:52, Jan Beulich wrote:
>>>>>> On 25.11.16 at 16:30, <julien.grall@xxxxxxx> wrote:
>>>> On 23/11/16 10:47, Jan Beulich wrote:
>>>>>>>> On 23.11.16 at 11:29, <andrii.anisov@xxxxxxxxx> wrote:
>>>>>> Building latest XEN master branch
>>>>>> (58bd0c7985890e0292212f94a56235228a3445c3) for salvator-x platform
>>>>>> using
>>>>>> the same yocto as here
>>>>>> https://wiki.xen.org/wiki/Xen_ARM_with_Virtualization_Extensions/Salvator-X
>>>>>>  
>>>>>>
>>>>>> I've
>>>>>> faced following issue:
>>>>>>
>>>>>> Maximum error count (200) exceeded
>>>>>>
>>>>>> /home/aanisov/DEV/salvatorx-new/build/tmp/work/aarch64-poky-linux/xen/4.8.0+
>>>>>>
>>>>>> gitAUTOINC+58bd0c7985-r0/git/tools/libxl/dsdt_anycpu_arm.asl
>>>>>>      3:         OperationRegion ( MSUM, SystemMemory, \_SB.MSUA, 1 )
>>>>>> Error    4084 -
>>>>>>
>>>>>>                          Object does not exist ^  (\_SB.MSUA)
>>>>>
>>>>> Looking at the source I see that this statement gets included for
>>>>> x86 only. Are you perhaps doing a cross build of the ARM tools
>>>>> on an x86 host? If so, it looks like some work would be needed to
>>>>> make that work (again? not sure if it ever worked), as there are
>>>>> various build host architecture conditionals in mk_dsdt.c.
>>>>
>>>> I gave a look at this error, I had in mind to replace all the #if
>>>> defined(__*__) by the CONFIG_* equivalent.
>>>>
>>>> However, the CONFIG_* are not available for the tools. I guess we need
>>>> to add them in the tools config.h. Any opinions?
>>>
>>> The CONFIG_* values are available in the Makefile (from
>>> config/arm*.mk). I think it would be better if the tool was told
>>> via command line option what architecture to build tables for.
>>> That would imply converting the problematic #if-s to if()s. Of
>>> course the per-arch #include-s at the top may be a problem
>>> here, but then again I'm not convinced the full xen/arch-arm.h
>>> is really needed here (I'm pretty sure including
>>> xen/hvm/hvm_info_table.h would cause no issue for ARM, the
>>> more that nominally this is an arch-independent header anyway).
>>
>> FYI, I am only planning to fix the compilation issue as this is a
>> problem for some user.
>>
>> Regarding hvm/hvm_info_table.h, It contains few things that are x86
>> specific (such as the PFN).
>>
>> Furthermore, I don't think this header should be included on ARM because
>> it defines HVM_MAX_VCPUS and may cause confusion with GUEST_MAX_VCPUS
>> the correct value for ARM guests.
> 
> I forgot to mention that xen/arch-arm.h is necessary in order to get 
> GUEST_MAX_VCPUS.

I think we should strive to solve this in a more elegant way, without
(almost) colliding definitions, and without the need to include stuff
we don't really need here. But yes, as a first step just fixing the
build issue without converting to runtime controlled behavior is going
to be fine. We just shouldn't put the issue fully to rest afterwards.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel