Re: [PATCH bpf-next 05/12] bpf: Add missed value to kprobe perf link info

From: Hou Tao <houtao@huaweicloud.com>
To: Jiri Olsa <jolsa@kernel.org>
Cc: bpf@vger.kernel.org, Martin KaFai Lau <kafai@fb.com>,
	Song Liu <songliubraving@fb.com>, Yonghong Song <yhs@fb.com>,
	John Fastabend <john.fastabend@gmail.com>,
	KP Singh <kpsingh@chromium.org>,
	Stanislav Fomichev <sdf@google.com>, Hao Luo <haoluo@google.com>,
	Daniel Xu <dxu@dxuuu.xyz>, Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Andrii Nakryiko <andrii@kernel.org>
Subject: Re: [PATCH bpf-next 05/12] bpf: Add missed value to kprobe perf link info
Date: Tue, 5 Sep 2023 10:23:27 +0800	[thread overview]
Message-ID: <4540a09e-43a4-c258-710c-af895138ba6b@huaweicloud.com> (raw)
In-Reply-To: <20230828075537.194192-6-jolsa@kernel.org>

Hi,

On 8/28/2023 3:55 PM, Jiri Olsa wrote:
> Add missed value to kprobe attached through perf link info to
> hold the stats of missed kprobe handler execution.
>
> The kprobe's missed counter gets incremented when kprobe handler
> is not executed due to another kprobe running on the same cpu.
>
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
>  include/linux/trace_events.h   |  6 ++++--
>  include/uapi/linux/bpf.h       |  1 +
>  kernel/bpf/syscall.c           | 14 ++++++++------
>  kernel/trace/bpf_trace.c       |  5 +++--
>  kernel/trace/trace_kprobe.c    |  5 ++++-
>  tools/include/uapi/linux/bpf.h |  1 +
>  6 files changed, 21 insertions(+), 11 deletions(-)
>

SNIP
> diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
> index 17c21c0b2dd1..998c88874507 100644
> --- a/kernel/trace/trace_kprobe.c
> +++ b/kernel/trace/trace_kprobe.c
> @@ -1546,7 +1546,8 @@ NOKPROBE_SYMBOL(kretprobe_perf_func);
>  
>  int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
>  			const char **symbol, u64 *probe_offset,
> -			u64 *probe_addr, bool perf_type_tracepoint)
> +			u64 *probe_addr, unsigned long *missed,
> +			bool perf_type_tracepoint)
>  {
>  	const char *pevent = trace_event_name(event->tp_event);
>  	const char *group = event->tp_event->class->system;
> @@ -1565,6 +1566,8 @@ int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
>  	*probe_addr = kallsyms_show_value(current_cred()) ?
>  		      (unsigned long)tk->rp.kp.addr : 0;
>  	*symbol = tk->symbol;
> +	if (missed)
> +		*missed = tk->rp.kp.nmissed;

According to the implement of probes_profile_seq_show(), the missed
counter for kretprobe should be tk->rp.kp.nmissed + tk->rp.nmissed. I
think it would be a good idea to factor out a common helper to get the
missed counter for kprobe or kretprobe.
>  	return 0;
>  }
>  #endif	/* CONFIG_PERF_EVENTS */
> diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
> index b754edfb0cd7..5a39c7a13499 100644
> --- a/tools/include/uapi/linux/bpf.h
> +++ b/tools/include/uapi/linux/bpf.h
> @@ -6539,6 +6539,7 @@ struct bpf_link_info {
>  					__u32 name_len;
>  					__u32 offset; /* offset from func_name */
>  					__u64 addr;
> +					__u64 missed;
>  				} kprobe; /* BPF_PERF_EVENT_KPROBE, BPF_PERF_EVENT_KRETPROBE */
>  				struct {
>  					__aligned_u64 tp_name;   /* in/out */