tools/perf/Documentation/perf-check.txt | 1 + tools/perf/Makefile.config | 13 + tools/perf/Makefile.perf | 24 +- tools/perf/builtin-check.c | 1 + tools/perf/tests/make | 2 + tools/perf/util/Build | 3 +- tools/perf/util/addr2line.c | 439 +++++++++++++++++++++ tools/perf/util/addr2line.h | 20 + tools/perf/util/annotate.c | 1 - tools/perf/util/capstone.c | 352 ++++++++++++----- tools/perf/util/config.c | 2 +- tools/perf/util/disasm.c | 18 +- tools/perf/util/disasm.h | 4 - tools/perf/util/dso.c | 112 ++++++ tools/perf/util/dso.h | 4 + tools/perf/util/libbfd.c | 4 +- tools/perf/util/libbfd.h | 6 +- tools/perf/util/llvm-c-helpers.cpp | 120 +++++- tools/perf/util/llvm-c-helpers.h | 24 +- tools/perf/util/llvm.c | 374 +++++++++++++----- tools/perf/util/llvm.h | 3 - tools/perf/util/srcline.c | 495 ++---------------------- tools/perf/util/srcline.h | 1 - 23 files changed, 1314 insertions(+), 709 deletions(-) create mode 100644 tools/perf/util/addr2line.c create mode 100644 tools/perf/util/addr2line.h
Linking against libcapstone and libLLVM can be a significant increase
in dependencies and file size if building statically. For something
like `perf record` the disassembler and addr2line functionality won't
be used. Support dynamically loading these libraries using dlopen and
then calling the appropriate functions found using dlsym.
The patch series:
1) feature check libLLVM support and avoid always reinitializing the
disassembler.
2) adds BPF JIT disassembly support to in memory disassemblers (LLVM
and capstone) by just directing them at the BPF info linear JIT
instructions (note this doesn't support source lines);
3) adds fallback to srcline's addr2line so that llvm_addr2line is
tried first, then the deprecated libbfd and then the forked command
tried next, moving the code for forking out of the main srcline.c
file in the process.
4) adds perf_ variants of the capstone/llvm functions that will either
directly call the function or use dlsym to discover it;
The addr2line LLVM functionality is written in C++. To avoid linking
against libLLVM for this, a new LIBLLVM_DYNAMIC option is added where
the C++ code with the libLLVM dependency will be built into a
libperf-llvm.so and that dlsym-ed and called against. Ideally LLVM
would extend their C API to avoid this.
v7: Refactor now the first 5 patches, that largely moved code around,
have landed. Move the dlopen code to the end of the series so that
the first 8 patches can be picked improving capstone/LLVM support
without adding the dlopen code. Rename the cover letter and
disassembler cleanup patches.
v6: Refactor the libbfd along with capstone and LLVM, previous patch
series had tried to avoid this by just removing the deprecated
BUILD_NONDISTRO code. Remove the libtracefs removal into its own
patch.
v5: Rebase and comment typo fix.
v4: Rebase and addition of a patch removing an unused struct variable.
v3: Add srcline addr2line fallback trying LLVM first then forking a
process. This came up in conversation with Steinar Gunderson
<sesse@google.com>.
Tweak the cover letter message to try to address Andi Kleen's
<ak@linux.intel.com> feedback that the series doesn't really
achieve anything.
v2: Add mangling of the function names in libperf-llvm.so to avoid
potential infinite recursion. Add BPF JIT disassembly support to
LLVM and capstone. Add/rebase the BUILD_NONDISTRO cleanup onto the
series from:
https://lore.kernel.org/lkml/20250111202851.1075338-1-irogers@google.com/
Some other minor additional clean up.
Ian Rogers (11):
perf check: Add libLLVM feature
perf llvm: Reduce LLVM initialization
perf dso: Move read_symbol from llvm/capstone to dso
perf dso: Support BPF programs in dso__read_symbol
perf dso: Clean up read_symbol error handling
perf disasm: Make ins__scnprintf and ins__is_nop static
perf srcline: Fallback between addr2line implementations
perf disasm: Remove unused evsel from annotate_args
perf capstone: Support for dlopen-ing libcapstone.so
perf llvm: Support for dlopen-ing libLLVM.so
perf llvm: Mangle libperf-llvm.so function names
tools/perf/Documentation/perf-check.txt | 1 +
tools/perf/Makefile.config | 13 +
tools/perf/Makefile.perf | 24 +-
tools/perf/builtin-check.c | 1 +
tools/perf/tests/make | 2 +
tools/perf/util/Build | 3 +-
tools/perf/util/addr2line.c | 439 +++++++++++++++++++++
tools/perf/util/addr2line.h | 20 +
tools/perf/util/annotate.c | 1 -
tools/perf/util/capstone.c | 352 ++++++++++++-----
tools/perf/util/config.c | 2 +-
tools/perf/util/disasm.c | 18 +-
tools/perf/util/disasm.h | 4 -
tools/perf/util/dso.c | 112 ++++++
tools/perf/util/dso.h | 4 +
tools/perf/util/libbfd.c | 4 +-
tools/perf/util/libbfd.h | 6 +-
tools/perf/util/llvm-c-helpers.cpp | 120 +++++-
tools/perf/util/llvm-c-helpers.h | 24 +-
tools/perf/util/llvm.c | 374 +++++++++++++-----
tools/perf/util/llvm.h | 3 -
tools/perf/util/srcline.c | 495 ++----------------------
tools/perf/util/srcline.h | 1 -
23 files changed, 1314 insertions(+), 709 deletions(-)
create mode 100644 tools/perf/util/addr2line.c
create mode 100644 tools/perf/util/addr2line.h
--
2.51.0.618.g983fd99d29-goog
On Sun, Oct 05, 2025 at 02:22:01PM -0700, Ian Rogers wrote:
> Linking against libcapstone and libLLVM can be a significant increase
> in dependencies and file size if building statically. For something
> like `perf record` the disassembler and addr2line functionality won't
> be used. Support dynamically loading these libraries using dlopen and
> then calling the appropriate functions found using dlsym.
>
> The patch series:
> 1) feature check libLLVM support and avoid always reinitializing the
> disassembler.
> 2) adds BPF JIT disassembly support to in memory disassemblers (LLVM
> and capstone) by just directing them at the BPF info linear JIT
> instructions (note this doesn't support source lines);
> 3) adds fallback to srcline's addr2line so that llvm_addr2line is
> tried first, then the deprecated libbfd and then the forked command
> tried next, moving the code for forking out of the main srcline.c
> file in the process.
> 4) adds perf_ variants of the capstone/llvm functions that will either
> directly call the function or use dlsym to discover it;
>
> The addr2line LLVM functionality is written in C++. To avoid linking
> against libLLVM for this, a new LIBLLVM_DYNAMIC option is added where
> the C++ code with the libLLVM dependency will be built into a
> libperf-llvm.so and that dlsym-ed and called against. Ideally LLVM
> would extend their C API to avoid this.
>
> v7: Refactor now the first 5 patches, that largely moved code around,
> have landed. Move the dlopen code to the end of the series so that
> the first 8 patches can be picked improving capstone/LLVM support
So I tentatively picked the first 8 patches, will test it now, hopefully
we can go with it to have BPF annotation...
Wait, will try to fix this one:
⬢ [acme@toolbx perf-tools-next]$ git log --oneline -1 ; time make -C tools/perf build-test
make_static: cd . && make LDFLAGS=-static NO_PERF_READ_VDSO32=1 NO_PERF_READ_VDSOX32=1 NO_JVMTI=1 NO_LIBTRACEEVENT=1 NO_LIBELF=1 -j32 DESTDIR=/tmp/tmp.w26bDGykTM
cd . && make LDFLAGS=-static NO_PERF_READ_VDSO32=1 NO_PERF_READ_VDSOX32=1 NO_JVMTI=1 NO_LIBTRACEEVENT=1 NO_LIBELF=1 -j32 DESTDIR=/tmp/tmp.w26bDGykTM
BUILD: Doing 'make -j32' parallel build
<SNIP>
Auto-detecting system features:
... libdw: [ OFF ]
... glibc: [ on ]
... libelf: [ OFF ]
... libnuma: [ OFF ]
... numa_num_possible_cpus: [ OFF ]
... libpython: [ OFF ]
... libcapstone: [ OFF ]
... llvm-perf: [ OFF ]
... zlib: [ OFF ]
... lzma: [ OFF ]
... get_cpuid: [ on ]
... bpf: [ on ]
... libaio: [ on ]
... libzstd: [ OFF ]
<SNIP>
CC tests/api-io.o
CC util/sha1.o
CC util/smt.o
LD util/intel-pt-decoder/perf-util-in.o
CC tests/demangle-java-test.o
CC util/strbuf.o
CC util/string.o
CC tests/demangle-ocaml-test.o
CC util/strlist.o
CC tests/demangle-rust-v0-test.o
CC tests/pfm.o
CC tests/parse-metric.o
CC util/strfilter.o
CC tests/pe-file-parsing.o
util/llvm.c: In function ‘init_llvm’:
util/llvm.c:78:17: error: implicit declaration of function ‘LLVMInitializeAllTargetInfos’ [-Wimplicit-function-declaration]
78 | LLVMInitializeAllTargetInfos();
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
util/llvm.c:79:17: error: implicit declaration of function ‘LLVMInitializeAllTargetMCs’ [-Wimplicit-function-declaration]
79 | LLVMInitializeAllTargetMCs();
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
util/llvm.c:80:17: error: implicit declaration of function ‘LLVMInitializeAllDisassemblers’ [-Wimplicit-function-declaration]
80 | LLVMInitializeAllDisassemblers();
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
util/llvm.c: At top level:
util/llvm.c:73:13: error: ‘init_llvm’ defined but not used [-Werror=unused-function]
73 | static void init_llvm(void)
| ^~~~~~~~~
cc1: all warnings being treated as errors
CC tests/expand-cgroup.o
CC util/top.o
CC tests/perf-time-to-tsc.o
CC util/usage.o
make[6]: *** [/home/acme/git/perf-tools-next/tools/build/Makefile.build:86: util/llvm.o] Error 1
make[6]: *** Waiting for unfinished jobs....
CC tests/dlfilter-test.o
CC tests/sigtrap.o
CC tests/event_groups.o
> without adding the dlopen code. Rename the cover letter and
> disassembler cleanup patches.
> v6: Refactor the libbfd along with capstone and LLVM, previous patch
> series had tried to avoid this by just removing the deprecated
> BUILD_NONDISTRO code. Remove the libtracefs removal into its own
> patch.
> v5: Rebase and comment typo fix.
> v4: Rebase and addition of a patch removing an unused struct variable.
> v3: Add srcline addr2line fallback trying LLVM first then forking a
> process. This came up in conversation with Steinar Gunderson
> <sesse@google.com>.
> Tweak the cover letter message to try to address Andi Kleen's
> <ak@linux.intel.com> feedback that the series doesn't really
> achieve anything.
> v2: Add mangling of the function names in libperf-llvm.so to avoid
> potential infinite recursion. Add BPF JIT disassembly support to
> LLVM and capstone. Add/rebase the BUILD_NONDISTRO cleanup onto the
> series from:
> https://lore.kernel.org/lkml/20250111202851.1075338-1-irogers@google.com/
> Some other minor additional clean up.
>
> Ian Rogers (11):
> perf check: Add libLLVM feature
> perf llvm: Reduce LLVM initialization
> perf dso: Move read_symbol from llvm/capstone to dso
> perf dso: Support BPF programs in dso__read_symbol
> perf dso: Clean up read_symbol error handling
> perf disasm: Make ins__scnprintf and ins__is_nop static
> perf srcline: Fallback between addr2line implementations
> perf disasm: Remove unused evsel from annotate_args
> perf capstone: Support for dlopen-ing libcapstone.so
> perf llvm: Support for dlopen-ing libLLVM.so
> perf llvm: Mangle libperf-llvm.so function names
>
> tools/perf/Documentation/perf-check.txt | 1 +
> tools/perf/Makefile.config | 13 +
> tools/perf/Makefile.perf | 24 +-
> tools/perf/builtin-check.c | 1 +
> tools/perf/tests/make | 2 +
> tools/perf/util/Build | 3 +-
> tools/perf/util/addr2line.c | 439 +++++++++++++++++++++
> tools/perf/util/addr2line.h | 20 +
> tools/perf/util/annotate.c | 1 -
> tools/perf/util/capstone.c | 352 ++++++++++++-----
> tools/perf/util/config.c | 2 +-
> tools/perf/util/disasm.c | 18 +-
> tools/perf/util/disasm.h | 4 -
> tools/perf/util/dso.c | 112 ++++++
> tools/perf/util/dso.h | 4 +
> tools/perf/util/libbfd.c | 4 +-
> tools/perf/util/libbfd.h | 6 +-
> tools/perf/util/llvm-c-helpers.cpp | 120 +++++-
> tools/perf/util/llvm-c-helpers.h | 24 +-
> tools/perf/util/llvm.c | 374 +++++++++++++-----
> tools/perf/util/llvm.h | 3 -
> tools/perf/util/srcline.c | 495 ++----------------------
> tools/perf/util/srcline.h | 1 -
> 23 files changed, 1314 insertions(+), 709 deletions(-)
> create mode 100644 tools/perf/util/addr2line.c
> create mode 100644 tools/perf/util/addr2line.h
>
> --
> 2.51.0.618.g983fd99d29-goog
>
On Mon, Oct 06, 2025 at 03:39:13PM -0300, Arnaldo Carvalho de Melo wrote:
> On Sun, Oct 05, 2025 at 02:22:01PM -0700, Ian Rogers wrote:
> > v7: Refactor now the first 5 patches, that largely moved code around,
> > have landed. Move the dlopen code to the end of the series so that
> > the first 8 patches can be picked improving capstone/LLVM support
> So I tentatively picked the first 8 patches, will test it now, hopefully
> we can go with it to have BPF annotation...
> Wait, will try to fix this one:
> ⬢ [acme@toolbx perf-tools-next]$ git log --oneline -1 ; time make -C tools/perf build-test
> make_static: cd . && make LDFLAGS=-static NO_PERF_READ_VDSO32=1 NO_PERF_READ_VDSOX32=1 NO_JVMTI=1 NO_LIBTRACEEVENT=1 NO_LIBELF=1 -j32 DESTDIR=/tmp/tmp.w26bDGykTM
> cd . && make LDFLAGS=-static NO_PERF_READ_VDSO32=1 NO_PERF_READ_VDSOX32=1 NO_JVMTI=1 NO_LIBTRACEEVENT=1 NO_LIBELF=1 -j32 DESTDIR=/tmp/tmp.w26bDGykTM
> BUILD: Doing 'make -j32' parallel build
> <SNIP>
> Auto-detecting system features:
> ... libdw: [ OFF ]
> ... glibc: [ on ]
> ... libelf: [ OFF ]
> ... libnuma: [ OFF ]
> ... numa_num_possible_cpus: [ OFF ]
> ... libpython: [ OFF ]
> ... libcapstone: [ OFF ]
> ... llvm-perf: [ OFF ]
> ... zlib: [ OFF ]
> ... lzma: [ OFF ]
> ... get_cpuid: [ on ]
> ... bpf: [ on ]
> ... libaio: [ on ]
> ... libzstd: [ OFF ]
> <SNIP>
> CC tests/api-io.o
> CC util/sha1.o
> CC util/smt.o
> LD util/intel-pt-decoder/perf-util-in.o
> CC tests/demangle-java-test.o
> CC util/strbuf.o
> CC util/string.o
> CC tests/demangle-ocaml-test.o
> CC util/strlist.o
> CC tests/demangle-rust-v0-test.o
> CC tests/pfm.o
> CC tests/parse-metric.o
> CC util/strfilter.o
> CC tests/pe-file-parsing.o
> util/llvm.c: In function ‘init_llvm’:
> util/llvm.c:78:17: error: implicit declaration of function ‘LLVMInitializeAllTargetInfos’ [-Wimplicit-function-declaration]
> 78 | LLVMInitializeAllTargetInfos();
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> util/llvm.c:79:17: error: implicit declaration of function ‘LLVMInitializeAllTargetMCs’ [-Wimplicit-function-declaration]
> 79 | LLVMInitializeAllTargetMCs();
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~
> util/llvm.c:80:17: error: implicit declaration of function ‘LLVMInitializeAllDisassemblers’ [-Wimplicit-function-declaration]
> 80 | LLVMInitializeAllDisassemblers();
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> util/llvm.c: At top level:
> util/llvm.c:73:13: error: ‘init_llvm’ defined but not used [-Werror=unused-function]
> 73 | static void init_llvm(void)
> | ^~~~~~~~~
> cc1: all warnings being treated as errors
> CC tests/expand-cgroup.o
> CC util/top.o
> CC tests/perf-time-to-tsc.o
> CC util/usage.o
> make[6]: *** [/home/acme/git/perf-tools-next/tools/build/Makefile.build:86: util/llvm.o] Error 1
> make[6]: *** Waiting for unfinished jobs....
> CC tests/dlfilter-test.o
> CC tests/sigtrap.o
> CC tests/event_groups.o
Guess this will be enough:
diff --git a/tools/perf/util/llvm.c b/tools/perf/util/llvm.c
index 565cad1969e5e51f..2ebf1f5f65bf77c7 100644
--- a/tools/perf/util/llvm.c
+++ b/tools/perf/util/llvm.c
@@ -70,6 +70,7 @@ int llvm__addr2line(const char *dso_name __maybe_unused, u64 addr __maybe_unused
#endif
}
+#ifdef HAVE_LIBLLVM_SUPPORT
static void init_llvm(void)
{
static bool init;
@@ -90,7 +91,6 @@ static void init_llvm(void)
* should add some textual annotation for after the instruction. The caller
* will use this information to add the actual annotation.
*/
-#ifdef HAVE_LIBLLVM_SUPPORT
struct symbol_lookup_storage {
u64 branch_addr;
u64 pcrel_load_addr;
On Mon, Oct 6, 2025 at 11:41 AM Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
>
> On Mon, Oct 06, 2025 at 03:39:13PM -0300, Arnaldo Carvalho de Melo wrote:
> > On Sun, Oct 05, 2025 at 02:22:01PM -0700, Ian Rogers wrote:
> > > v7: Refactor now the first 5 patches, that largely moved code around,
> > > have landed. Move the dlopen code to the end of the series so that
> > > the first 8 patches can be picked improving capstone/LLVM support
>
> > So I tentatively picked the first 8 patches, will test it now, hopefully
> > we can go with it to have BPF annotation...
>
> > Wait, will try to fix this one:
>
> > ⬢ [acme@toolbx perf-tools-next]$ git log --oneline -1 ; time make -C tools/perf build-test
> > make_static: cd . && make LDFLAGS=-static NO_PERF_READ_VDSO32=1 NO_PERF_READ_VDSOX32=1 NO_JVMTI=1 NO_LIBTRACEEVENT=1 NO_LIBELF=1 -j32 DESTDIR=/tmp/tmp.w26bDGykTM
> > cd . && make LDFLAGS=-static NO_PERF_READ_VDSO32=1 NO_PERF_READ_VDSOX32=1 NO_JVMTI=1 NO_LIBTRACEEVENT=1 NO_LIBELF=1 -j32 DESTDIR=/tmp/tmp.w26bDGykTM
> > BUILD: Doing 'make -j32' parallel build
> > <SNIP>
> > Auto-detecting system features:
> > ... libdw: [ OFF ]
> > ... glibc: [ on ]
> > ... libelf: [ OFF ]
> > ... libnuma: [ OFF ]
> > ... numa_num_possible_cpus: [ OFF ]
> > ... libpython: [ OFF ]
> > ... libcapstone: [ OFF ]
> > ... llvm-perf: [ OFF ]
> > ... zlib: [ OFF ]
> > ... lzma: [ OFF ]
> > ... get_cpuid: [ on ]
> > ... bpf: [ on ]
> > ... libaio: [ on ]
> > ... libzstd: [ OFF ]
> > <SNIP>
> > CC tests/api-io.o
> > CC util/sha1.o
> > CC util/smt.o
> > LD util/intel-pt-decoder/perf-util-in.o
> > CC tests/demangle-java-test.o
> > CC util/strbuf.o
> > CC util/string.o
> > CC tests/demangle-ocaml-test.o
> > CC util/strlist.o
> > CC tests/demangle-rust-v0-test.o
> > CC tests/pfm.o
> > CC tests/parse-metric.o
> > CC util/strfilter.o
> > CC tests/pe-file-parsing.o
> > util/llvm.c: In function ‘init_llvm’:
> > util/llvm.c:78:17: error: implicit declaration of function ‘LLVMInitializeAllTargetInfos’ [-Wimplicit-function-declaration]
> > 78 | LLVMInitializeAllTargetInfos();
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > util/llvm.c:79:17: error: implicit declaration of function ‘LLVMInitializeAllTargetMCs’ [-Wimplicit-function-declaration]
> > 79 | LLVMInitializeAllTargetMCs();
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~
> > util/llvm.c:80:17: error: implicit declaration of function ‘LLVMInitializeAllDisassemblers’ [-Wimplicit-function-declaration]
> > 80 | LLVMInitializeAllDisassemblers();
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > util/llvm.c: At top level:
> > util/llvm.c:73:13: error: ‘init_llvm’ defined but not used [-Werror=unused-function]
> > 73 | static void init_llvm(void)
> > | ^~~~~~~~~
> > cc1: all warnings being treated as errors
> > CC tests/expand-cgroup.o
> > CC util/top.o
> > CC tests/perf-time-to-tsc.o
> > CC util/usage.o
> > make[6]: *** [/home/acme/git/perf-tools-next/tools/build/Makefile.build:86: util/llvm.o] Error 1
> > make[6]: *** Waiting for unfinished jobs....
> > CC tests/dlfilter-test.o
> > CC tests/sigtrap.o
> > CC tests/event_groups.o
>
> Guess this will be enough:
>
> diff --git a/tools/perf/util/llvm.c b/tools/perf/util/llvm.c
> index 565cad1969e5e51f..2ebf1f5f65bf77c7 100644
> --- a/tools/perf/util/llvm.c
> +++ b/tools/perf/util/llvm.c
> @@ -70,6 +70,7 @@ int llvm__addr2line(const char *dso_name __maybe_unused, u64 addr __maybe_unused
> #endif
> }
>
> +#ifdef HAVE_LIBLLVM_SUPPORT
> static void init_llvm(void)
> {
> static bool init;
> @@ -90,7 +91,6 @@ static void init_llvm(void)
> * should add some textual annotation for after the instruction. The caller
> * will use this information to add the actual annotation.
> */
> -#ifdef HAVE_LIBLLVM_SUPPORT
> struct symbol_lookup_storage {
> u64 branch_addr;
> u64 pcrel_load_addr;
Ah crap. Yeah, it's a PITA trying to keep LLVM and not LLVM builds
happy. perf_LLVM* fixes this by possibly always punting to the
dlopen/dlsym version when HAVE_LIBLLVM_SUPPORT isn't there - hence a
lot of the comments I've been making. I forgot to rebuild without
libLLVM and so missed needing the extra #ifdefs. I agree with the fix
but the fix can be removed in the (now) later patches that use
dlopen/dlsym.
Thanks,
Ian
© 2016 - 2025 Red Hat, Inc.