Re: [PATCH v2 06/13] x86/PMU: Add public xenpmu.h

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 09/30/2013 09:30 AM, Jan Beulich wrote:
>>>> On 30.09.13 at 15:25, Boris Ostrovsky <boris.ostrovsky@oracle.com> wrote:
>> On 09/25/2013 10:04 AM, Jan Beulich wrote:
>>>>>> On 20.09.13 at 11:42, Boris Ostrovsky <boris.ostrovsky@oracle.com> wrote:
>>>> +/* Intel PMU registers and structures */
>>>> +#define XENPMU_CORE2_MAX_ARCH_PMCS     16
>>>> +#define XENPMU_CORE2_MAX_FIXED_PMCS    4
>>>> +struct core2_vpmu_context {
>>>> +    uint64_t global_ctrl;
>>>> +    uint64_t global_ovf_ctrl;
>>>> +    uint64_t global_status;
>>>> +    uint64_t global_ovf_status;
>>>> +    uint64_t fixed_ctrl;
>>>> +    uint64_t ds_area;
>>>> +    uint64_t pebs_enable;
>>>> +    uint64_t debugctl;
>>>> +    uint64_t fix_counters[XENPMU_CORE2_MAX_FIXED_PMCS];
>>>> +    struct {
>>>> +        uint64_t counter;
>>>> +        uint64_t control;
>>>> +    } arch_msr_pair[XENPMU_CORE2_MAX_ARCH_PMCS];
>>>> +};
>>> I realize that using embedded arrays in both AMD and Intel
>>> structures makes things easier to implement, but it reduces
>>> forward compatibility. I'd therefore prefer those to be made
>>> handles.
>> (I missed this comment earlier).
>>
>> This is not done because it's easier but because I need to keep the
>> structure in a shared page.
> Then a more dynamic layout (with just the array base offset in the
> page specified in the structure) would still be preferable as being
> more extensible.
>
>

I'll see how I can do this, I am afraid this may end up being a bit too 
convoluted (the
arrays will still need to be part of some structure with decribes a 
context).

-boris