Re: [PATCH] perf_events: improve Intel event scheduling

[Stephane Eranian <eranian@google.com>]

On Fri, Dec 11, 2009 at 12:00 PM, stephane eranian
<eranian@googlemail.com> wrote:
>>> --- a/arch/x86/kernel/cpu/perf_event.c
>>> +++ b/arch/x86/kernel/cpu/perf_event.c
>>
>>> @@ -68,6 +69,15 @@ struct debug_store {
>>>       u64     pebs_event_reset[MAX_PEBS_EVENTS];
>>>  };
>>>
>>> +#define BITS_TO_U64(nr) DIV_ROUND_UP(nr, BITS_PER_BYTE * sizeof(u64))
>>
>> Do we need this, is it realistic to expect X86_PMC_IDX_MAX to be
>> anything else than 64?
>>
> The issue had to do with i386 mode where long are 32 bits  < 64. And in
> particular with the initializer .idxmsk[0] = V. In the future we may exceed
> 32 registers. That means the initializer would have to change. But I guess
> we have quite some ways before this case is reached. So I will revert all
> of this to unsigned long.
>
Well, in fact we have to use u64 because you are already using register
indexes > 32, e.g., for the Intel fixed counters which start at position 32.
Using unsigned long would make the static initializer uglier in this case.