Re: [PATCH v3 5/5] tools, libxl: handle the iomem parameter with the memory_mapping hcall

[Julien Grall <julien.grall@linaro.org>]

Hello Arianna,

On 03/18/2014 04:15 PM, Arianna Avanzini wrote:
> On 03/17/2014 01:35 PM, Julien Grall wrote:
>> On 03/15/2014 08:11 PM, Arianna Avanzini wrote:
>>> diff --git a/tools/libxl/libxl_create.c b/tools/libxl/libxl_create.c
>>> index d015cf4..dda4df3 100644
>>> --- a/tools/libxl/libxl_create.c
>>> +++ b/tools/libxl/libxl_create.c
>>> @@ -1107,7 +1107,24 @@ static void domcreate_launch_dm(libxl__egc *egc, libxl__multidev *multidev,
>>>                   "failed give dom%d access to iomem range %"PRIx64"-%"PRIx64,
>>>                   domid, io->start, io->start + io->number - 1);
>>>              ret = ERROR_FAIL;
>>> +            continue;
>>>          }
>>> +#if defined(__x86_32__) || defined(__x86_64__)
>>> +        if (b_info->type != LIBXL_DOMAIN_TYPE_HVM)
>>
>> Calling xc_domain_memory_memory_mapping might be also useful for PVH.
>> AFAIK, on x86 PVH is consider as LIBXL_DOMAIN_TYPE_PV.
>>
>> The best solution is to check if the domain use auto-translated physmap
>> (e.g translation between PFN and MFN is done by Xen). But, after talking
>> with Ian C. it's not possible to know it on libxl.
>>
>> You might need to introduce an architecture helper (libxl_arm.c and
>> libxl_x86.c are respectively only compiled on ARM and x86), which check
>> if you can call xc_domain_memory_mapping or not.
>>
> 
> Sorry if I bother you with a question to see if I understood correctly, the
> helper should determine whether the guest domain is auto-translated, right?

Right, and we will have to call xc_domain_memory_mapping only if the
domain is auto-translated.

> 
>>> +            continue;
>>> +#endif /* __x86_32__ || __x86_64__ */
>>> +#if defined(__arm__) || defined(__x86_32__) || defined(__x86_64__)
>>
>> You don't need to ifdef this part. Libxl doesn't support other arch for now.
>>
> 
> I think it was Jan Beulich who suggested using a whitelist approach here during
> the review of the v2 patchset. Is it / is he OK to drop this by now?

I think we can remove it as long as we have an arch-function to check
whether this is needed to be called.

If Xen will be ported to a new architecture, the developper will have to
implement the new helper for its architecture.

Regards,

-- 
Julien Grall