Re: [PATCH v3 11/14] libxl: get and set soft affinity

[George Dunlap <george.dunlap@eu.citrix.com>]

On 20/11/13 11:27, Ian Campbell wrote:
> On Tue, 2013-11-19 at 18:51 +0100, Dario Faggioli wrote:
>>>> +{
>>>> +    libxl_cputopology *topology;
>>>> +    libxl_bitmap ecpumap;
>>>> +    int nr_cpus = 0, rc;
>>>> +
>>>> +    topology = libxl_get_cpu_topology(ctx, &nr_cpus);
>>>> +    if (!topology) {
>>>> +        LIBXL__LOG(ctx, LIBXL__LOG_ERROR, "failed to retrieve CPU topology");
>>> It's not consistent within the file but I think for new functions we
>>> should use the LOG macro variants.
>>>
>> Right, but don't I need a gc to use it? Should I "make up" one just for
>> the purpose of using LOG/LOGE?
> I think a call to GC_INIT/GC_FREE should be cheap enough.
>
>>>> +        return ERROR_FAIL;
>>>> +    }
>>>> +    libxl_cputopology_list_free(topology, nr_cpus);
>>> Why are you retrieving this only to immediately throw it away?
>>>
>> Because I need nr_cpus. :-)
> Surely this is not the recommended way to get nr_cpus!
>
> libxl_get_cpu_topology() itself calls libxl_get_max_cpus() which seems
> like the obvious candidate.
>
>
>>>> diff --git a/tools/libxl/libxl.h b/tools/libxl/libxl.h
>>>> index c7dceda..504c57b 100644
>>>> --- a/tools/libxl/libxl.h
>>>> +++ b/tools/libxl/libxl.h
>>>> @@ -82,6 +82,20 @@
>>>>   #define LIBXL_HAVE_DOMAIN_NODEAFFINITY 1
>>>>   
>>>>   /*
>>>> + * LIBXL_HAVE_VCPUINFO_SOFTAFFINITY indicates that a 'cpumap_soft'
>>>> + * field (of libxl_bitmap type) is present in libxl_vcpuinfo,
>>>> + * containing the soft affinity for the vcpu.
>>>> + */
>>>> +#define LIBXL_HAVE_VCPUINFO_SOFTAFFINITY 1
>>>> +
>>>> +/*
>>>> + * LIBXL_HAVE_BUILDINFO_SOFTAFFINITY indicates that a 'cpumap_soft'
>>>> + * field (of libxl_bitmap type) is present in libxl_domain_build_info,
>>>> + * containing the soft affinity for the vcpu.
>>>> + */
>>>> +#define LIBXL_HAVE_BUILDINFO_SOFTAFFINITY 1
>>> Given that they arrive can we just use HAVE_SOFTRAFFINITY?
>>>
>> You mean just introducing one #define? Sure... For some reason I assumed
>> that every new field should come with it's own symbol. But if it's fine
>> to have one, I'm all for it. :-)
> I think it's ok.
>
>>>> +/* Flags, consistent with domctl.h */
>>>> +#define LIBXL_VCPUAFFINITY_HARD 1
>>>> +#define LIBXL_VCPUAFFINITY_SOFT 2
>>> Can these be an enum in the idl?
>>>
>> I think yes.
>>
>> I did actually check and, of all the enum-s in the IDL, none are used as
>> flags, they're rather used as "single values". OTOH, the only actual
>> flags I found (I think it was LIBXL_SUSPEND_DEBUG, LIBXL_SUSPEND_LIVE)
>> were defined like I did myself above... That's why I went for it.
> I have a feeling they predate the IDL, or at least the Enumeration
> support. It's true that we don't have any other bit fields in enums
> though. I can't see the harm, it's probably not worth introducing a new
> IDL type for them.

Since these are bits, not numbers, I don't think an enum is the right 
construct.  Or, the enum values should be the *bit numbers*, and the 
flags should be (1<<[bit_humber]).

  -George