Re: [PATCH 0/5] perf core: Read from overwrite ring buffer

["Wangnan (F)" <wangnan0@huawei.com>]

Hi Peter,

Do you have any further comments about this patchset? What should be
the correct route for them to be merged into mainline?

Thank you.

On 2016/1/25 16:33, Wang Nan wrote:
> This is the v3 of this series. Compare with v2, tailsize method is
> removed, ioctl command PERF_EVENT_IOC_PAUSE_OUTPUT is changed to
> _IOW('$', 9, __u32) since it has an input value, commit message
> is slightly adjusted.
>
> New test result on ARM64 is below (x86_64 result is copied from previous
> email, test method is described in previous version [1]):
>
>                         x86_64        |       ARM64
>                -----------------------+---------------------
>                  MEAN         STDVAR  |    MEAN       STDVAR
>     BASE     :  879870.81   11913.13  |  808302.67   6951.47
>     RAWPERF  : 2603854.70  706658.40  | 3461675.12  54075.69
>     WRTBKWRD : 2313301.22    6727.96  | 3350177.98  23125.48
>     TAILSIZE : 2383051.86    5248.06  | 3556496.56  24802.17
>     RAWOVWRT : 2315273.18    5221.03  | 3458767.49  41560.32
>     RAWOVWRT*: 2323970.45    5103.39  | 3438967.06  34095.83
>
> ARM64 platform is a smartphone with 8 Hisilicon arm64 cores
> (big/little).
>
> Benckmarking result on both architechure support removing tailsize
> method because its performance penalty is higher than WRTBKWRD. Also, it
> seems we don't need to consider tailheader method (putting whole header
> at the end of records, kernel write to the ring buffer from beginning to
> end), because the above result shows WRTBKWRD and RAWOVWRT are similar,
> and tailheader is impossible to outperform RAWOVWRT in theory.
>
> [1] http://lkml.kernel.org/g/1453464834-233200-1-git-send-email-wangnan0@huawei.com
>
> Wang Nan (5):
>    perf core: Introduce new ioctl options to pause and resume ring buffer
>    perf core: Set event's default overflow_handler
>    perf core: Prepare writing into ring buffer from end
>    perf core: Add backward attribute to perf event
>    perf core: Reduce perf event output overhead by new overflow handler
>
>   include/linux/perf_event.h      | 22 +++++++++++--
>   include/uapi/linux/perf_event.h |  4 ++-
>   kernel/events/core.c            | 73 +++++++++++++++++++++++++++++++++++------
>   kernel/events/internal.h        | 11 +++++++
>   kernel/events/ring_buffer.c     | 63 +++++++++++++++++++++++++++++++----
>   5 files changed, 153 insertions(+), 20 deletions(-)
>
> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> Cc: He Kuang <hekuang@huawei.com>
> Cc: Alexei Starovoitov <ast@kernel.org>
> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
> Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
> Cc: Jiri Olsa <jolsa@kernel.org>
> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> Cc: Namhyung Kim <namhyung@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Zefan Li <lizefan@huawei.com>
> Cc: pi3orama@163.com
>