Re: [PATCH V2 6/7] perf, x86: Use LBR call stack to get user callchain

[Stephane Eranian]

But still, if I do:
$  perf record -g -e cpu/cycles/k ........

Looks like your code will activate LBR cstack for user stack even
though I don't use it.
I know it won't generate any user samples (theoretically) but you are
still commandeering
the LBR resource which other events may want to use.

On Wed, Oct 24, 2012 at 2:11 PM, Stephane Eranian <eranian@xxxxxxxxxx> wrote:
> On Wed, Oct 24, 2012 at 1:52 PM, Yan, Zheng <zheng.z.yan@xxxxxxxxx> wrote:
>> On 10/24/2012 07:47 PM, Stephane Eranian wrote:
>>> On Wed, Oct 24, 2012 at 1:23 PM, Yan, Zheng <zheng.z.yan@xxxxxxxxx> wrote:
>>>> On 10/24/2012 04:57 PM, Stephane Eranian wrote:
>>>>> On Wed, Oct 24, 2012 at 7:59 AM, Yan, Zheng <zheng.z.yan@xxxxxxxxx> wrote:
>>>>>> From: "Yan, Zheng" <zheng.z.yan@xxxxxxxxx>
>>>>>>
>>>>>> Try enabling the LBR call stack feature if event request recording
>>>>>> callchain. Try utilizing the LBR call stack to get user callchain
>>>>>> in case of there is no frame pointer.
>>>>>>
>>>>>> Signed-off-by: Yan, Zheng <zheng.z.yan@xxxxxxxxx>
>>>>>> ---
>>>>>>  arch/x86/kernel/cpu/perf_event.c           | 126 +++++++++++++++++++++--------
>>>>>>  arch/x86/kernel/cpu/perf_event.h           |   7 ++
>>>>>>  arch/x86/kernel/cpu/perf_event_intel.c     |  20 ++---
>>>>>>  arch/x86/kernel/cpu/perf_event_intel_lbr.c |   3 +
>>>>>>  include/linux/perf_event.h                 |   6 ++
>>>>>>  kernel/events/core.c                       |  11 ++-
>>>>>>  6 files changed, 124 insertions(+), 49 deletions(-)
>>>>>>
>>>>>> diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
>>>>>> index 8ae8044..3bf2100 100644
>>>>>> --- a/arch/x86/kernel/cpu/perf_event.c
>>>>>> +++ b/arch/x86/kernel/cpu/perf_event.c
>>>>>> @@ -398,35 +398,46 @@ int x86_pmu_hw_config(struct perf_event *event)
>>>>>>
>>>>>>                 if (event->attr.precise_ip > precise)
>>>>>>                         return -EOPNOTSUPP;
>>>>>> -               /*
>>>>>> -                * check that PEBS LBR correction does not conflict with
>>>>>> -                * whatever the user is asking with attr->branch_sample_type
>>>>>> -                */
>>>>>> -               if (event->attr.precise_ip > 1 && x86_pmu.intel_cap.pebs_format < 2) {
>>>>>> -                       u64 *br_type = &event->attr.branch_sample_type;
>>>>>> -
>>>>>> -                       if (has_branch_stack(event)) {
>>>>>> -                               if (!precise_br_compat(event))
>>>>>> -                                       return -EOPNOTSUPP;
>>>>>> -
>>>>>> -                               /* branch_sample_type is compatible */
>>>>>> -
>>>>>> -                       } else {
>>>>>> -                               /*
>>>>>> -                                * user did not specify  branch_sample_type
>>>>>> -                                *
>>>>>> -                                * For PEBS fixups, we capture all
>>>>>> -                                * the branches at the priv level of the
>>>>>> -                                * event.
>>>>>> -                                */
>>>>>> -                               *br_type = PERF_SAMPLE_BRANCH_ANY;
>>>>>> -
>>>>>> -                               if (!event->attr.exclude_user)
>>>>>> -                                       *br_type |= PERF_SAMPLE_BRANCH_USER;
>>>>>> -
>>>>>> -                               if (!event->attr.exclude_kernel)
>>>>>> -                                       *br_type |= PERF_SAMPLE_BRANCH_KERNEL;
>>>>>> -                       }
>>>>>> +       }
>>>>>> +       /*
>>>>>> +        * check that PEBS LBR correction does not conflict with
>>>>>> +        * whatever the user is asking with attr->branch_sample_type
>>>>>> +        */
>>>>>> +       if (event->attr.precise_ip > 1 && x86_pmu.intel_cap.pebs_format < 2) {
>>>>>> +               u64 *br_type = &event->attr.branch_sample_type;
>>>>>> +
>>>>>> +               if (has_branch_stack(event)) {
>>>>>> +                       if (!precise_br_compat(event))
>>>>>> +                               return -EOPNOTSUPP;
>>>>>> +
>>>>>> +                       /* branch_sample_type is compatible */
>>>>>> +
>>>>>> +               } else {
>>>>>> +                       /*
>>>>>> +                        * user did not specify  branch_sample_type
>>>>>> +                        *
>>>>>> +                        * For PEBS fixups, we capture all
>>>>>> +                        * the branches at the priv level of the
>>>>>> +                        * event.
>>>>>> +                        */
>>>>>> +                       *br_type = PERF_SAMPLE_BRANCH_ANY;
>>>>>> +
>>>>>> +                       if (!event->attr.exclude_user)
>>>>>> +                               *br_type |= PERF_SAMPLE_BRANCH_USER;
>>>>>> +
>>>>>> +                       if (!event->attr.exclude_kernel)
>>>>>> +                               *br_type |= PERF_SAMPLE_BRANCH_KERNEL;
>>>>>> +               }
>>>>>> +       } else if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) {
>>>>>> +               if (!has_branch_stack(event) && x86_pmu.attr_lbr_callstack) {
>>>>>> +                       /*
>>>>>> +                        * user did not specify branch_sample_type,
>>>>>> +                        * try using the LBR call stack facility to
>>>>>> +                        * record call chains in the user space.
>>>>>> +                        */
>>>>>> +                       event->attr.branch_sample_type =
>>>>>> +                               PERF_SAMPLE_BRANCH_USER |
>>>>>> +                               PERF_SAMPLE_BRANCH_CALL_STACK;
>>>>>
>>>>> You are forcing user level here, but how do you know the user wanted
>>>>> ONLY user level
>>>>> callchains?
>>>>>
>>>>>
>>>>
>>>> The LBR call stack is used only when the frame pointer approach doesn't work.
>>>
>>> And where is that determination made?
>>
>> check code that is added to perf_callchain_user and perf_callchain_user32
>>
> Are you saying you can run in a mode where you get kernel call stack via
> frame-pointer and the user call stack via LBR cstack for a single event?
>
>
>>>
>>>> I think the kernel has frame pointer for the most cases. The second reason is
>>>> that the LBR call stack only has 16 entries. I think it's too small to record
>>>> both kernel and user call chains.
>>>>
>>> It's even too small for many object oriented user programs as well.
>>>
>>>> Regards
>>>> Yan, Zheng
>>>>
>>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/