Re: [RFC 09/13] perf kwork: Add workqueue report support

[Yang Jihong <yangjihong1@huawei.com>]

Hello,

On 2022/6/16 5:56, Namhyung Kim wrote:
> On Tue, Jun 14, 2022 at 8:22 PM Yang Jihong <yangjihong1@huawei.com> wrote:
>>
>> Hello,
>>
>> On 2022/6/15 5:54, Namhyung Kim wrote:
>>> On Mon, Jun 13, 2022 at 2:48 AM Yang Jihong <yangjihong1@huawei.com> wrote:
>>>>
>>>> Implements workqueue report function.
>>>>
>>>> test case:
>>>>
>>>>     # perf kwork -k workqueue rep
>>>>
>>>>       Kwork Name                | Cpu  | Total Runtime | Frequency | Max runtime   | Max runtime start   | Max runtime end     |
>>>>      ---------------------------------------------------------------------------------------------------------------------------
>>>>       (w)0xffffffff83e09fa0     | 0001 |   2152.678 ms |       194 |     12.376 ms |    2059361.546621 s |    2059361.558997 s |
>>>>       (w)0xffff888332fea180     | 0000 |     17.125 ms |       301 |      1.018 ms |    2059358.441070 s |    2059358.442089 s |
>>>>       (w)0xffff8881035a83d8     | 0007 |      7.556 ms |         3 |      3.212 ms |    2059362.614643 s |    2059362.617855 s |
>>>>       (w)0xffff888102fc14a0     | 0002 |      7.080 ms |         5 |      1.962 ms |    2059365.421753 s |    2059365.423714 s |
>>>>       (w)0xffffffff82f7da00     | 0000 |      4.277 ms |         7 |      3.778 ms |    2059360.851063 s |    2059360.854841 s |
>>>>       (w)0xffffffff8305d680     | 0006 |      1.796 ms |         1 |      1.796 ms |    2059360.046818 s |    2059360.048613 s |
>>>>       (w)0xffff8883339e9040     | 0005 |      1.659 ms |         2 |      1.619 ms |    2059361.266051 s |    2059361.267670 s |
>>>>       (w)0xffff888333de9040     | 0007 |      1.121 ms |         5 |      0.783 ms |    2059368.238059 s |    2059368.238842 s |
>>>>       (w)0xffff888332fe9040     | 0000 |      0.990 ms |         4 |      0.911 ms |    2059359.604075 s |    2059359.604986 s |
>>>>       (w)0xffff8883331e9040     | 0001 |      0.244 ms |         6 |      0.046 ms |    2059362.689277 s |    2059362.689323 s |
>>>>       (w)0xffff888102e44400     | 0007 |      0.239 ms |         2 |      0.137 ms |    2059363.117537 s |    2059363.117674 s |
>>>>       (w)0xffff8883333ea180     | 0002 |      0.141 ms |         5 |      0.049 ms |    2059365.423784 s |    2059365.423833 s |
>>>>       (w)0xffffffff83062f28     | 0006 |      0.084 ms |         2 |      0.047 ms |    2059358.208033 s |    2059358.208080 s |
>>>>       (w)0xffffffff8305ca48     | 0003 |      0.078 ms |         2 |      0.041 ms |    2059361.071371 s |    2059361.071412 s |
>>>>       (w)0xffff8883337e9040     | 0004 |      0.062 ms |         1 |      0.062 ms |    2059362.605723 s |    2059362.605785 s |
>>>>       (w)0xffff8881035a81e8     | 0001 |      0.056 ms |         1 |      0.056 ms |    2059363.118231 s |    2059363.118287 s |
>>>>       (w)0xffff8883335e9040     | 0003 |      0.026 ms |         1 |      0.026 ms |    2059358.573397 s |    2059358.573423 s |
>>>>       (w)0xffffffff83062e70     | 0006 |      0.023 ms |         1 |      0.023 ms |    2059368.398864 s |    2059368.398888 s |
>>>>       (w)0xffffffff83e06480     | 0002 |      0.000 ms |         1 |      0.000 ms |    2059359.986792 s |    2059359.986792 s |
>>>
>>> Using "function" in the tracepoint and symbolizing it would be
>>> far more intuitive.
>>>
>> OK，This is a simplified version that will be improved in the next
>> version, and I'd like to add the following features:
>> 1. Supports the kthread profile.
> 
> Could you elaborate more?
trace kthread tracepoints (sched:sched_kthread_work_queue_work, sched:
sched_kthread_work_execute_start and 
sched:sched_kthread_work_execute_end) can support kthread profile, 
because framework has been set up. we only need to add a new kthread 
class class.

> 
>> 2. Save runtime and latency in kernel using ebpf(similar to "perf
>> record: Implement off-cpu profiling with BPF") . This reduces the number
>> of interruptions caused by writing files to hard disks, which is closer
>> to the actual scenario.
> 
> Sounds great.
OK, I'll add it in next version.
> 
>>
>> This RFC is sent to discuss to see if this function is useful to the
>> community and can be accepted by the community. :)
> 
> Yeah I think it'd be useful.
Thanks for your affirmation.

Thanks,
Jihong