Re: [PATCH bpf-next v2] bpf: Add show_fdinfo for perf_event

[Andrii Nakryiko]

On Fri, Jun 6, 2025 at 8:03 AM Tao Chen <chen.dylane@xxxxxxxxx> wrote:
>
> After commit 1b715e1b0ec5 ("bpf: Support ->fill_link_info for perf_event") add
> perf_event info, we can also show the info with the method of cat /proc/[fd]/fdinfo.
>
> kprobe fdinfo:
> link_type:      perf
> link_id:        10
> prog_tag:       bcf7977d3b93787c
> prog_id:        20
> name:   bpf_fentry_test1
> offset: 0
> missed: 0
> addr:   ffffffffa28a2904
> event_type:     kprobe
> cookie: 3735928559
>
> uprobe fdinfo:
> link_type:      perf
> link_id:        13
> prog_tag:       bcf7977d3b93787c
> prog_id:        21
> name:   /proc/self/exe
> offset: 63dce4
> ref_ctr_offset: 33eee2a
> event_type:     uprobe
> cookie: 3735928559
>
> tracepoint fdinfo:
> link_type:      perf
> link_id:        11
> prog_tag:       bcf7977d3b93787c
> prog_id:        22
> tp_name:        sched_switch
> event_type:     tracepoint
> cookie: 3735928559
>
> perf_event fdinfo:
> link_type:      perf
> link_id:        12
> prog_tag:       bcf7977d3b93787c
> prog_id:        23
> type:   1
> config: 2
> event_type:     event
> cookie: 3735928559
>
> Signed-off-by: Tao Chen <chen.dylane@xxxxxxxxx>
> ---
>  kernel/bpf/syscall.c | 118 +++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 118 insertions(+)
>
> Change list:
> - v1 -> v2:
>    - Andrii suggested:
>      1. define event_type with string
>      2. print offset and addr with hex
>
>    - Jiri suggested:
>      1. add ref_ctr_offset for uprobe
> - v1:
>     https://lore.kernel.org/bpf/20250604163723.3175258-1-chen.dylane@xxxxxxxxx
>
> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index 89d027cd7ca..928ff129087 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -3795,6 +3795,31 @@ static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
>         info->perf_event.kprobe.cookie = event->bpf_cookie;
>         return 0;
>  }
> +
> +static void bpf_perf_link_fdinfo_kprobe(const struct perf_event *event,
> +                                       struct seq_file *seq)
> +{
> +       const char *name;
> +       int err;
> +       u32 prog_id, type;
> +       u64 offset, addr;
> +       unsigned long missed;
> +
> +       err = bpf_get_perf_event_info(event, &prog_id, &type, &name,
> +                                     &offset, &addr, &missed);
> +       if (err)
> +               return;
> +
> +       seq_printf(seq,
> +                  "name:\t%s\n"
> +                  "offset:\t%llx\n"

I used %#llx to have 0x prepended in front of hex values (here and in
a few other places). That seems to be what we do in other fdinfo
handlers. Applied to bpf-next, thanks.

> +                  "missed:\t%lu\n"
> +                  "addr:\t%llx\n"
> +                  "event_type:\t%s\n"
> +                  "cookie:\t%llu\n",
> +                  name, offset, missed, addr, type == BPF_FD_TYPE_KRETPROBE ?
> +                  "kretprobe" : "kprobe", event->bpf_cookie);
> +}
>  #endif
>
>  #ifdef CONFIG_UPROBE_EVENTS

[...]