Re: [PATCH bpf-next] perf/bpf_counter: use bpf_map_create instead of bpf_create_map

[Song Liu <songliubraving@fb.com>]

> On Dec 6, 2021, at 6:37 PM, Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
> 
> On Mon, Dec 6, 2021 at 3:08 PM Song Liu <song@kernel.org> wrote:
>> 
>> bpf_create_map is deprecated. Replace it with bpf_map_create.
>> 
>> Fixes: 992c4225419a ("libbpf: Unify low-level map creation APIs w/ new bpf_map_create()")
> 
> This is not a bug fix, it's an improvement. So I don't think "Fixes: "
> is warranted here, tbh.

I got compilation errors before this change, like

util/bpf_counter.c: In function ‘bperf_lock_attr_map’:
util/bpf_counter.c:323:3: error: ‘bpf_create_map’ is deprecated: libbpf v0.7+: use bpf_map_create() instead [-Werror=deprecated-declarations]
   map_fd = bpf_create_map(BPF_MAP_TYPE_HASH,
   ^~~~~~
In file included from util/bpf_counter.h:7,
                 from util/bpf_counter.c:15:
/data/users/songliubraving/kernel/linux-git/tools/lib/bpf/bpf.h:91:16: note: declared here
 LIBBPF_API int bpf_create_map(enum bpf_map_type map_type, int key_size,
                ^~~~~~~~~~~~~~
cc1: all warnings being treated as errors
make[4]: *** [/data/users/songliubraving/kernel/linux-git/tools/build/Makefile.build:96: util/bpf_counter.o] Error 1
make[4]: *** Waiting for unfinished jobs....
make[3]: *** [/data/users/songliubraving/kernel/linux-git/tools/build/Makefile.build:139: util] Error 2
make[2]: *** [Makefile.perf:665: perf-in.o] Error 2
make[1]: *** [Makefile.perf:240: sub-make] Error 2
make: *** [Makefile:70: all] Error 2

Do we plan to remove bpf_create_map in the future? If not, we can probably just
add '#pragma GCC diagnostic ignored "-Wdeprecated-declarations"' can call it done? 

> 
>> Signed-off-by: Song Liu <song@kernel.org>
>> ---
>> tools/perf/util/bpf_counter.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>> 
>> diff --git a/tools/perf/util/bpf_counter.c b/tools/perf/util/bpf_counter.c
>> index c17d4a43ce065..ed150a9b3a0c0 100644
>> --- a/tools/perf/util/bpf_counter.c
>> +++ b/tools/perf/util/bpf_counter.c
>> @@ -320,10 +320,10 @@ static int bperf_lock_attr_map(struct target *target)
>>        }
>> 
>>        if (access(path, F_OK)) {
>> -               map_fd = bpf_create_map(BPF_MAP_TYPE_HASH,
>> +               map_fd = bpf_map_create(BPF_MAP_TYPE_HASH, NULL,
> 
> I think perf is trying to be linkable with libbpf as a shared library,
> so on some older versions of libbpf bpf_map_create() won't be (yet)
> available. So to make this work, I think you'll need to define your
> own weak bpf_map_create function that will use bpf_create_map().

Hmm... I didn't know the plan to link libbpf as shared library. In this case, 
maybe the #pragma solution is preferred? 

Thanks,
Song