Date:   Sun, 1 Mar 2020 20:26:03 -0800
From:   Yonghong Song <yhs@...com>
To:     Song Liu <songliubraving@...com>
CC:     Networking <netdev@...r.kernel.org>, bpf <bpf@...r.kernel.org>,
        Kernel Team <Kernel-team@...com>,
        "ast@...nel.org" <ast@...nel.org>,
        "daniel@...earbox.net" <daniel@...earbox.net>,
        "arnaldo.melo@...il.com" <arnaldo.melo@...il.com>,
        "jolsa@...nel.org" <jolsa@...nel.org>
Subject: Re: [PATCH v2 bpf-next 1/2] bpftool: introduce "prog profile" command



On 3/1/20 7:42 PM, Yonghong Song wrote:
> 
> 
> On 3/1/20 2:37 PM, Song Liu wrote:
>>
>>
>>> On Feb 29, 2020, at 7:52 PM, Yonghong Song <yhs@...com> wrote:
>>>
>>>
>>>
>>> On 2/28/20 3:40 PM, Song Liu wrote:
>>>> With fentry/fexit programs, it is possible to profile BPF program with
>>>> hardware counters. Introduce bpftool "prog profile", which measures key
>>>> metrics of a BPF program.
>>>> bpftool prog profile command creates per-cpu perf events. Then it 
>>>> attaches
>>>> fentry/fexit programs to the target BPF program. The fentry program 
>>>> saves
>>>> perf event value to a map. The fexit program reads the perf event 
>>>> again,
>>>> and calculates the difference, which is the instructions/cycles used by
>>>> the target program.
>>>> Example input and output:
>>>>    ./bpftool prog profile 3 id 337 cycles instructions llc_misses
>>>>          4228 run_cnt
>>>>       3403698 cycles                                              
>>>> (84.08%)
>>>>       3525294 instructions   #  1.04 insn per cycle               
>>>> (84.05%)
>>>>            13 llc_misses     #  3.69 LLC misses per million isns  
>>>> (83.50%)
>>>> This command measures cycles and instructions for BPF program with id
>>>> 337 for 3 seconds. The program has triggered 4228 times. The rest of 
>>>> the
>>>> output is similar to perf-stat. In this example, the counters were only
>>>> counting ~84% of the time because of time multiplexing of perf 
>>>> counters.
>>>> Note that, this approach measures cycles and instructions in very small
>>>> increments. So the fentry/fexit programs introduce noticeable errors to
>>>> the measurement results.
>>>> The fentry/fexit programs are generated with BPF skeletons. 
>>>> Therefore, we
>>>> build bpftool twice. The first time _bpftool is built without 
>>>> skeletons.
>>>> Then, _bpftool is used to generate the skeletons. The second time, 
>>>> bpftool
>>>> is built with skeletons.
>>>> Signed-off-by: Song Liu <songliubraving@...com>
>>>> ---
>>>>   tools/bpf/bpftool/Makefile                |  18 +
>>>>   tools/bpf/bpftool/prog.c                  | 428 
>>>> +++++++++++++++++++++-
>>>>   tools/bpf/bpftool/skeleton/profiler.bpf.c | 171 +++++++++
>>>>   tools/bpf/bpftool/skeleton/profiler.h     |  47 +++
>>>>   tools/scripts/Makefile.include            |   1 +
>>>>   5 files changed, 664 insertions(+), 1 deletion(-)
>>>>   create mode 100644 tools/bpf/bpftool/skeleton/profiler.bpf.c
>>>>   create mode 100644 tools/bpf/bpftool/skeleton/profiler.h
>>>> diff --git a/tools/bpf/bpftool/Makefile b/tools/bpf/bpftool/Makefile
>>>> index c4e810335810..c035fc107027 100644
>>>> --- a/tools/bpf/bpftool/Makefile
>>>> +++ b/tools/bpf/bpftool/Makefile
>>>> @@ -59,6 +59,7 @@ LIBS = $(LIBBPF) -lelf -lz
>>>>     INSTALL ?= install
>>>>   RM ?= rm -f
>>>> +CLANG ?= clang
>>>>     FEATURE_USER = .bpftool
>>>>   FEATURE_TESTS = libbfd disassembler-four-args reallocarray zlib
>>>> @@ -110,6 +111,22 @@ SRCS += $(BFD_SRCS)
>>>>   endif
>>>>     OBJS = $(patsubst %.c,$(OUTPUT)%.o,$(SRCS)) $(OUTPUT)disasm.o
>>>> +_OBJS = $(filter-out $(OUTPUT)prog.o,$(OBJS)) $(OUTPUT)_prog.o
>>>> +
>>>> +$(OUTPUT)_prog.o: prog.c
>>>> +    $(QUIET_CC)$(COMPILE.c) -MMD -DBPFTOOL_WITHOUT_SKELETONS -o $@ $<
>>>> +
>>>> +$(OUTPUT)_bpftool: $(_OBJS) $(LIBBPF)
>>>> +    $(QUIET_LINK)$(CC) $(CFLAGS) $(LDFLAGS) -o $@ $(_OBJS) $(LIBS)
>>>> +
>>>> +skeleton/profiler.bpf.o: skeleton/profiler.bpf.c
>>>> +    $(QUIET_CLANG)$(CLANG) -g -O2 -target bpf -c $< -o $@
>>>
>>> With a fresh checkout, applying this patch and just selftests like
>>>   make -C tools/testing/selftests/bpf
>>>
>>> I got the following build error:
>>>
>>> make[2]: Leaving directory `/data/users/yhs/work/net-next/tools/lib/bpf'
>>> clang -g -O2 -target bpf -c skeleton/profiler.bpf.c -o 
>>> skeleton/profiler.bpf.o
>>> skeleton/profiler.bpf.c:5:10: fatal error: 'bpf/bpf_helpers.h' file 
>>> not found
>>> #include <bpf/bpf_helpers.h>
>>>          ^~~~~~~~~~~~~~~~~~~
>>> 1 error generated.
>>> make[1]: *** [skeleton/profiler.bpf.o] Error 1
>>>
>>> I think Makefile should be tweaked to avoid selftest failure.
>>
>> Hmm... I am not seeing this error. The build succeeded in the test.
> 
> Just tried again with a *fresh* checkout of tools/ directory with the 
> patch.
>    -bash-4.4$ make -C tools/testing/selftests/bpf
>    ...
>    LINK 
> /data/users/yhs/work/net-next/tools/testing/selftests/bpf/tools/build/bpftool//libbpf/libbpf.a 
> 
>    LINK 
> /data/users/yhs/work/net-next/tools/testing/selftests/bpf/tools/build/bpftool/_bpftool 
> 
> make[1]: *** No rule to make target `skeleton/profiler.bpf.c', needed by 
> `skeleton/profiler.bpf.o'.  Stop.
> make: *** 
> [/data/users/yhs/work/net-next/tools/testing/selftests/bpf/tools/sbin/bpftool] 
> Error 2
> make: Leaving directory 
> `/data/users/yhs/work/net-next/tools/testing/selftests/bpf
> 
> The error is different from my previous try, but the build still fails.

FYI. The following change fixed build in my env.

diff --git a/tools/bpf/bpftool/Makefile b/tools/bpf/bpftool/Makefile
index c035fc107027..20a90d8450f8 100644
--- a/tools/bpf/bpftool/Makefile
+++ b/tools/bpf/bpftool/Makefile
@@ -120,7 +120,7 @@ $(OUTPUT)_bpftool: $(_OBJS) $(LIBBPF)
         $(QUIET_LINK)$(CC) $(CFLAGS) $(LDFLAGS) -o $@ $(_OBJS) $(LIBS)

  skeleton/profiler.bpf.o: skeleton/profiler.bpf.c
-       $(QUIET_CLANG)$(CLANG) -g -O2 -target bpf -c $< -o $@
+       $(QUIET_CLANG)$(CLANG) -I$(srctree)/tools/lib -g -O2 -target bpf 
-c $< -o $@

  profiler.skel.h: $(OUTPUT)_bpftool skeleton/profiler.bpf.o
         $(QUIET_GEN)$(OUTPUT)./_bpftool gen skeleton 
skeleton/profiler.bpf.o > $@

> 
>>
>> Thanks,
>> Song
>>

Powered by blists - more mailing lists