Re: [PATCH bpf-next 1/3] perf/core: introduce perf_event_mmap_bpf_prog

[Alexei Starovoitov <ast@fb.com>]

On 9/19/18 4:30 PM, Song Liu wrote:
>
>
>> On Sep 19, 2018, at 3:39 PM, Alexei Starovoitov <ast@kernel.org> wrote:
>>
>> introduce perf_event_mmap_bpf_prog() helper to emit RECORD_MMAP events
>> into perf ring buffer.
>> It's used by bpf load/unload logic to notify user space of addresses
>> and names of JITed bpf programs.
>>
>> Note that event->mmap.pid == -1 is an existing indicator of kernel event.
>> In addition use event->mmap.tid == BPF_FS_MAGIC to indicate bpf related
>> RECORD_MMAP event.
>>
>> Alternatively it's possible to introduce new 'enum perf_event_type' command
>> specificially for bpf prog load/unload, but existing RECORD_MMAP
>> is very close, so the choice made by this patch is to extend it.
>>
>> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
>
> Acked-by: Song Liu <songliubraving@fb.com>
>
> I guess we should also use this for kernel modules load/unload?

yes. that's possible.
There is similar issue with modules today that get unloaded
before 'perf report'.
Synthetic record_mmap for modules that perf emits into perf.data
has filename==module_name. It solves typical use case though.
We can extend record_mmap further and let kernel emit them
for module load/unload.
Some new magic for 'tid' would be necessary.