lkml.org 
[lkml]   [2012]   [Oct]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH V2 1/7] perf, x86: Reduce lbr_sel_map size
From
On Wed, Oct 24, 2012 at 10:37 AM, Yan, Zheng <zheng.z.yan@intel.com> wrote:
> On 10/24/2012 04:23 PM, Yan, Zheng wrote:
>> On 10/24/2012 04:15 PM, Stephane Eranian wrote:
>>> On Wed, Oct 24, 2012 at 9:49 AM, Yan, Zheng <zheng.z.yan@intel.com> wrote:
>>>> On 10/24/2012 03:28 PM, Stephane Eranian wrote:
>>>>> On Wed, Oct 24, 2012 at 7:59 AM, Yan, Zheng <zheng.z.yan@intel.com> wrote:
>>>>>> From: "Yan, Zheng" <zheng.z.yan@intel.com>
>>>>>>
>>>>>> The index of lbr_sel_map is bit value of perf branch_sample_type.
>>>>>> By using bit shift as index, we can reduce lbr_sel_map size.
>>>>>>
>>>>>> Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
>>>>>> ---
>>>>>> arch/x86/kernel/cpu/perf_event.h | 4 +++
>>>>>> arch/x86/kernel/cpu/perf_event_intel_lbr.c | 50 ++++++++++++++----------------
>>>>>> include/uapi/linux/perf_event.h | 42 +++++++++++++++++--------
>>>>>> 3 files changed, 56 insertions(+), 40 deletions(-)
>>>>>>
>>>>>> diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h
>>>>>> index d3b3bb7..ea6749a 100644
>>>>>> --- a/arch/x86/kernel/cpu/perf_event.h
>>>>>> +++ b/arch/x86/kernel/cpu/perf_event.h
>>>>>> @@ -412,6 +412,10 @@ struct x86_pmu {
>>>>>> struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
>>>>>> };
>>>>>>
>>>>>> +enum {
>>>>>> + PERF_SAMPLE_BRANCH_SELECT_MAP_SIZE = PERF_SAMPLE_BRANCH_MAX_SHIFT,
>>>>>> +};
>>>>>> +
>>>>> What's the point on the extraneous definition?
>>>>
>>>> because later patches will add map PERF_SAMPLE_BRANCH_CALL_STACK, it will make
>>>> "PERF_SAMPLE_BRANCH_SELECT_MAP_SIZE != PERF_SAMPLE_BRANCH_MAX_SHIFT"
>>>>
>>> And you are not going to do:
>>>
>>> enum perf_branch_sample_type_shift {
>>> ...
>>> PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT = 10
>>> PERF_SAMPLE_BRANCH_MAX_SHIFT
>>> };
>>>
>>> PERF_SAMPLE_BRANCH_CALL_STACK = 1 << PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT
>>>
>>> Unless you're telling you are not going to add a mapping for
>>> PERF_SAMPLE_CALL_STACK to the
>>> lbr_sel_map[]?
>>>
>>
>> I think include/uapi/linux/perf_event.h should only contain definition for user API.
>> So I added PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT and PERF_SAMPLE_BRANCH_CALL_STACK to
>> arch/x86/kernel/cpu/perf_event.h. Please check patch 1.
>>
>
> Sorry, I mean patch 2.
>
Yeah, I figured that one. The part I was missing was that you're
trying to fit this
under PERF_SAMPLE_CALLCHAIN.

So now, looks like we have 3 different ways of getting user call stacks:
- PERF_SAMPLE_CALLCHAIN via frame pointer
- PERF_SAMPLE_CALLCHAIN via LBR cstack on HSW
- PERF_SAMPLE_USER_CSTACK via stack copying + dwarf

And presumably all are available under perf record -g.

The difference I see with LBR cstack is that the callstack is much
smaller, max 16 deep
and it has the limitations you mentioned in the cover message.


\
 
 \ /
  Last update: 2012-10-24 11:21    [W:0.076 / U:0.020 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site