Re: [PATCH v1 1/4] perf parse-events: Remove BPF event support

[Manu Bretelle <chantr4@gmail.com>]

On Thu, Oct 19, 2023 at 06:08:33PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Thu, Oct 19, 2023 at 12:45:08PM -0700, Manu Bretelle escreveu:
> > cc @quentin
> > 
> > On Fri, Aug 11, 2023 at 12:41:36PM -0300, Arnaldo Carvalho de Melo wrote:
> > > Em Fri, Aug 11, 2023 at 11:43:22AM -0300, Arnaldo Carvalho de Melo escreveu:
> > > > Right now it is not applying due to some clash with other changes and
> > > > when I tried to apply it manually there were some formatting issues:
> > > > 
> > > > ⬢[acme@toolbox perf-tools-next]$ head ~/wb/1.patch
> > > > From SRS0=EALy=D3=flex--irogers.bounces.google.com=3IDHVZAcKBAUnwtljwxlttlqj.htrfhrjpjwsjq.twl@kernel.org Thu Aug 10 17:53:46 2023
> > > > Delivered-To: arnaldo.melo@gmail.com
> > > > Received: from imap.gmail.com [64.233.186.109]
> > > > 	by quaco with IMAP (fetchmail-6.4.37)
> > > > 	for <acme@localhost> (single-drop); Thu, 10 Aug 2023 17:53:46 -0300 (-03)
> > > > Received: by 2002:a0c:ab03:0:b0:63d:780e:9480 with SMTP id h3csp908198qvb;
> > > >  Thu, 10 Aug 2023 11:49:52 -0700 (PDT)
> > > > X-Google-Smtp-Source: AGHT+IH9N/knUCyQ0tQ2Q0XBH0gqf8A8DB8/37YHWAJDKBmz7AGSV9CvCKYDuE3EwxriZFBwtZMs
> > > > X-Received: by 2002:a4a:6b4f:0:b0:56c:b2ab:9820 with SMTP id
> > > >  h15-20020a4a6b4f000000b0056cb2ab9820mr2695332oof.8.1691693392493; Thu, 10 Aug
> > > > ⬢[acme@toolbox perf-tools-next]$ patch -p1 < ~/wb/1.patch
> > > > patching file tools/perf/Documentation/perf-config.txt
> > > > patch: **** malformed patch at line 234: ith
> > > > 
> > > > ⬢[acme@toolbox perf-tools-next]$
> > > > 
> > > > I'm trying to apply it manually.
> > > 
> > > I have this extracted from this patch as the first patch in the series:
> > > 
> > > >From adc61b5774a9de62f34d593f164ca02daa6fb44c Mon Sep 17 00:00:00 2001
> > > From: Ian Rogers <irogers@google.com>
> > > Date: Fri, 11 Aug 2023 12:19:48 -0300
> > > Subject: [PATCH 1/1] perf bpf: Remove support for embedding clang for
> > >  compiling BPF events (-e foo.c)
> > > 
> > > This never was in the default build for perf, is difficult to maintain
> > > as it uses clang/llvm internals so ditch it, keeping, for now, the
> > > external compilation of .c BPF into .o bytecode and its subsequent
> > > loading, that is also going to be removed, do it separately to help
> > > bisection and to properly document what is being removed and why.
> > > 
> > > Committer notes:
> > > 
> > > Extracted from a larger patch and removed some leftovers, namely
> > > deleting these now unused feature tests:
> > > 
> > >     tools/build/feature/test-clang.cpp
> > >     tools/build/feature/test-cxx.cpp
> > >     tools/build/feature/test-llvm-version.cpp
> > >     tools/build/feature/test-llvm.cpp
> > > 
> > 
> > This seem to have broken `llvm` feature detection for `bpftool`.
> > 
> > The feature detections are still available in `tools/build/Makefile.feature` [0]
> > but the .cpp files are gone.
> > 
> > `bpftool` still rely on the `llvm` feature:
> > 
> >     $ git --no-pager grep 'feature-llvm'
> >     tools/bpf/bpftool/Makefile:ifeq ($(feature-llvm),1)
> > 
> > The result of testing llvm feature is:
> > 
> >     $ cat tools/build/feature/test-llvm.make.output
> >     cc1plus: fatal error: test-llvm.cpp: No such file or directory
> >     compilation terminated.
> > 
> > With current head:
> > 
> >     make -j $((4*$(nproc))) -C tools/bpf/bpftool && ./tools/bpf/bpftool/bpftool --version
> >     ...
> >     Auto-detecting system features:
> >     ...                         clang-bpf-co-re: [ on  ]
> >     ...                                    llvm: [ OFF ]
> >     ...                                  libcap: [ on  ]
> >     ...                                  libbfd: [ on  ]
> >     ...
> >     ...
> >     ...
> >     bpftool v7.3.0
> >     using libbpf v1.3
> >     features: libbfd, skeletons
> > 
> > After applying
> > 
> >     git show 56b11a2126bf2f422831ecf6112b87a4485b221b  tools/build/feature | \
> >         patch -p1 -R
> 
> 
> Ouch, so probably we need just to reintroduce that one
> tools/build/feature/test-llvm.cpp file.

Yes, I believe so. From git-grepping `feature-<feature>`, only `feature-llvm` came up.

I suppose as part of the cleanup the features should have been
removed from tools/build/Makefile.feature too.
 
> 
> Building perf these days ends up using bpftool, and the end result as
> noticed with me testing perf, perf trace with bpf, etc didn't change, so
> I didn't notice :-\
> 
> And:
> 
> ifeq ($(feature-llvm),1)
>   # If LLVM is available, use it for JIT disassembly
>   CFLAGS  += -DHAVE_LLVM_SUPPORT
>   LLVM_CONFIG_LIB_COMPONENTS := mcdisassembler all-targets
>   CFLAGS  += $(shell $(LLVM_CONFIG) --cflags --libs $(LLVM_CONFIG_LIB_COMPONENTS))
>   LIBS    += $(shell $(LLVM_CONFIG) --libs $(LLVM_CONFIG_LIB_COMPONENTS))
>   ifeq ($(shell $(LLVM_CONFIG) --shared-mode),static)
>     LIBS += $(shell $(LLVM_CONFIG) --system-libs $(LLVM_CONFIG_LIB_COMPONENTS))
>     LIBS += -lstdc++
>   endif
>   LDFLAGS += $(shell $(LLVM_CONFIG) --ldflags)
> else
>   # Fall back on libbfd
>   ifeq ($(feature-libbfd),1)
>     LIBS += -lbfd -ldl -lopcodes
>   else ifeq ($(feature-libbfd-liberty),1)
>     LIBS += -lbfd -ldl -lopcodes -liberty
>   else ifeq ($(feature-libbfd-liberty-z),1)
>     LIBS += -lbfd -ldl -lopcodes -liberty -lz
>   endif
> 
>   # If one of the above feature combinations is set, we support libbfd
>   ifneq ($(filter -lbfd,$(LIBS)),)
>     CFLAGS += -DHAVE_LIBBFD_SUPPORT
> 
>     # Libbfd interface changed over time, figure out what we need
>     ifeq ($(feature-disassembler-four-args), 1)
>       CFLAGS += -DDISASM_FOUR_ARGS_SIGNATURE
>     endif
>     ifeq ($(feature-disassembler-init-styled), 1)
>       CFLAGS += -DDISASM_INIT_STYLED
>     endif
>   endif
> endif
> 
> And there is a fallback to using binutils, so most people ended up not
> noticing.

Yes, definitely, the fallback make it harder to easily detect.
I detected it because I was manually building bpftool and it was odd that llvm
was not detected.

> 
> I wonder how to improve the current situation to detect these kinds of
> problems in the future, i.e. how to notice that some file needed by some
> Makefile, etc got removed or that some feature test fails because some
> change in the test .c files makes them fail and thus activates fallbacks
> like the one above :-\

I think it is tricky. Specifically to this situation, some CI could try to build
the different combinaison of bpftool and check the features through the build
`bpftool --version`.
This is actually a test that I run internally to make sure our build has some
feature enabled.
This is actually tested by bpftool in the GH CI:
https://github.com/libbpf/bpftool/blob/main/.github/workflows/build.yaml#L62

As a matter of fact, it would not have been detected because that CI uses a
different Makefile.feature.

Quentin and I were talking offline how we could improve bpftool CI at diff time.
This is an example where it would have helped :)

> 
> 
> So if I just get this back:
> 
> ⬢[acme@toolbox perf-tools-next]$ cat tools/build/feature/test-llvm.cpp
> // SPDX-License-Identifier: GPL-2.0
> #include "llvm/Support/ManagedStatic.h"
> #include "llvm/Support/raw_ostream.h"
> #define NUM_VERSION (((LLVM_VERSION_MAJOR) << 16) + (LLVM_VERSION_MINOR << 8) + LLVM_VERSION_PATCH)
> 
> #if NUM_VERSION < 0x030900
> # error "LLVM version too low"
> #endif
> int main()
> {
> 	llvm::errs() << "Hello World!\n";
> 	llvm::llvm_shutdown();
> 	return 0;
> }
> ⬢[acme@toolbox perf-tools-next]$
> 
> And install the llvm-devel package then it back working:
> 
> ⬢[acme@toolbox perf-tools-next]$ make -C tools/bpf/bpftool
> make: Entering directory '/home/acme/git/perf-tools-next/tools/bpf/bpftool'
> 
> Auto-detecting system features:
> ...                         clang-bpf-co-re: [ on  ]
> ...                                    llvm: [ on  ]
> ...                                  libcap: [ on  ]
> ...                                  libbfd: [ on  ]
> <SNIP>
> ⬢[acme@toolbox perf-tools-next]$ cat tools/build/feature/test-llvm.make.output
> ⬢[acme@toolbox perf-tools-next]$ ls -la tools/build/feature/test-llvm.
> test-llvm.bin          test-llvm.cpp          test-llvm.d            test-llvm.make.output
> ⬢[acme@toolbox perf-tools-next]$ ls -la tools/build/feature/test-llvm.bin
> -rwxr-xr-x. 1 acme acme 17712 Oct 19 18:04 tools/build/feature/test-llvm.bin
> ⬢[acme@toolbox perf-tools-next]$ ldd tools/build/feature/test-llvm.bin
> 	linux-vdso.so.1 (0x00007ffcaf5d9000)
> 	libLLVM-16.so => /lib64/libLLVM-16.so (0x00007fc4faefa000)
> 	libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007fc4faca6000)
> 	libm.so.6 => /lib64/libm.so.6 (0x00007fc4fabc5000)
> 	libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007fc4faba1000)
> 	libc.so.6 => /lib64/libc.so.6 (0x00007fc4fa9c3000)
> 	libffi.so.8 => /lib64/libffi.so.8 (0x00007fc4fa9b7000)
> 	libedit.so.0 => /lib64/libedit.so.0 (0x00007fc4fa978000)
> 	libz.so.1 => /lib64/libz.so.1 (0x00007fc4fa95e000)
> 	libtinfo.so.6 => /lib64/libtinfo.so.6 (0x00007fc4fa92b000)
> 	/lib64/ld-linux-x86-64.so.2 (0x00007fc502404000)
> ⬢[acme@toolbox perf-tools-next]$ sudo dnf install llvm-devel
> 
> I'll get this merged in my perf-tools-fixes-for-v6.6 that I'll submit
> tomorrow to Linus, thanks for reporting!
> 
> I'll add your:
> 
> Reported-by: Manu Bretelle <chantr4@gmail.com>
> 
> And:
> 
> Fixes: 56b11a2126bf2f42 ("perf bpf: Remove support for embedding clang for compiling BPF events (-e foo.c)")
> 
> Ok?

SGTM. Thanks for the quick turnaround.

Reviewed-by: Manu Bretelle <chantr4@gmail.com>

> 
> - Arnaldo