kernel/linux.git/tools/perf/util/srcline.c, branch v7.1

perf tools: Replace basename() calls with perf_basename()

2026-04-09T02:21:05+00:00

As noticed in a sashiko review for a patch adding a missing libgen.h in a file using basename(): https://sashiko.dev/#/patchset/20260402001740.2220481-1-acme%40kernel.org So avoid these subtleties and instead reuse the gnu_basename() function we had in srcline.c, renaming it to perf_basename() and replace basename() calls with it, simplifying several cases by removing now needless strdups. Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

perf tools: Use calloc() where applicable

2026-04-09T02:21:05+00:00

Instead of using zalloc(nr_entries * sizeof_entry) that is what calloc() does. In some places where linux/zalloc.h isn't needed, remove it, add when needed and was getting it indirectly. Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

perf powerpc: Unify the skip-callchain-idx libdw with that for addr2line

2026-01-20T16:04:38+00:00

Rather than have 2 Dwfl unify the Dwfl in skip-callchain-idx with that is used by libdw__addr2line(). Rename that variable in 'struct dso' from 'a2l_libdw' to just 'libdw' as it is now used in more than addr2line. The Dwfl in skip-callchain-idx uses a map address when being read with dwfl_report_elf (rather than dwfl_report_offline that addr2line uses). skip-callchain-idx is wrong as the map address can vary between processes because of ASLR, ie it should need a different Dwfl per process. In the code after this patch the base address becomes 0 and the mapped PC is used with the dwfl functions. This should increase the accuracy of skip-callchain-idx, but the impact has only been build tested. Signed-off-by: Ian Rogers Cc: Aditya Bodkhe Cc: Adrian Hunter Cc: Albert Ou Cc: Alexandre Ghiti Cc: Andi Kleen Cc: Athira Rajeev Cc: Chun-Tse Shao Cc: Dmitriy Vyukov Cc: Dr. David Alan Gilbert Cc: Guo Ren Cc: Haibo Xu Cc: Howard Chu Cc: Ingo Molnar Cc: James Clark Cc: Jiri Olsa Cc: John Garry Cc: Krzysztof Łopatowski Cc: Leo Yan Cc: Mark Wielaard Cc: Namhyung Kim Cc: Palmer Dabbelt Cc: Paul Walmsley Cc: Peter Zijlstra Cc: Sergei Trofimovich Cc: Shimin Guo Cc: Stephen Brennan Cc: Thomas Falcon Cc: Will Deacon Signed-off-by: Arnaldo Carvalho de Melo

perf srcline: Add configuration support for the addr2line style

2026-01-13T19:09:14+00:00

Allow the addr2line style to be specified on the `perf report` command line or in the .perfconfig file. Committer testing: The methods: # perf probe -x ~/bin/perf -F *__addr2line cmd__addr2line libbfd__addr2line libdw__addr2line llvm__addr2line # So if we configure one of them, say 'addr2line': # perf config addr2line.style=addr2line # perf config addr2line.style addr2line.style=addr2line # And have probes on all of them: # perf probe -x ~/bin/perf *__addr2line Added new events: probe_perf:cmd__addr2line (on *__addr2line in /home/acme/bin/perf) probe_perf:llvm__addr2line (on *__addr2line in /home/acme/bin/perf) probe_perf:libbfd__addr2line (on *__addr2line in /home/acme/bin/perf) probe_perf:libdw__addr2line (on *__addr2line in /home/acme/bin/perf) You can now use it in all perf tools, such as: perf record -e probe_perf:libdw__addr2line -aR sleep 1 # Only the selected method should be used: # perf stat -e probe_perf:*_addr2line perf report -f --dso perf --stdio -s srcfile,srcline # Total Lost Samples: 0 # # Samples: 4K of event 'cpu/cycles/Pu' # Event count (approx.): 5535180842 # # Overhead Source File Source:Line # ........ ............ ............... # 99.04% inlineloop.c inlineloop.c:21 0.46% inlineloop.c inlineloop.c:20 # # (Tip: For hierarchical output, try: perf report --hierarchy) # Performance counter stats for 'perf report -f --dso perf --stdio -s srcfile,srcline': 44 probe_perf:cmd__addr2line 0 probe_perf:llvm__addr2line 0 probe_perf:libbfd__addr2line 0 probe_perf:libdw__addr2line 0.035915611 seconds time elapsed 0.028008000 seconds user 0.009051000 seconds sys # I checked and that is the case for the other methods. Also when using: # perf config addr2line.style=libdw,llvm Performance counter stats for 'perf report -f --dso perf --stdio -s srcfile,srcline': 0 probe_perf:cmd__addr2line 23 probe_perf:llvm__addr2line 0 probe_perf:libbfd__addr2line 44 probe_perf:libdw__addr2line Reviewed-by: James Clark Signed-off-by: Ian Rogers Tested-by: Arnaldo Carvalho de Melo Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Howard Chu Cc: Ingo Molnar Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephen Brennan Cc: Tony Jones Signed-off-by: Arnaldo Carvalho de Melo

perf addr2line: Add a libdw implementation

2026-01-12T19:23:06+00:00

Add an implementation of addr2line that uses libdw. Other addr2line implementations are slow, particularly in the case of forking addr2line. Add an implementation that caches the libdw information in the dso and uses it to find the file and line number information. Inline information is supported but because cu_walk_functions_at visits the leaf function last add a inline_list__append_tail to reverse the lists order. Committer testing: # perf probe -x ~/bin/perf libdw__addr2line Added new event: probe_perf:libdw_addr2line (on libdw__addr2line in /home/acme/bin/perf) You can now use it in all perf tools, such as: perf record -e probe_perf:libdw_addr2line -aR sleep 1 # # perf stat -e probe_perf:libdw_addr2line perf report -f --dso perf --stdio -s srcfile,srcline # To display the perf.data header info, please use --header/--header-only options. # # # Total Lost Samples: 0 # # Samples: 4K of event 'cpu/cycles/Pu' # Event count (approx.): 5535180842 # # Overhead Source File Source:Line # ........ ............ ............... # 99.04% inlineloop.c inlineloop.c:21 0.46% inlineloop.c inlineloop.c:20 # # (Tip: For tracepoint events, try: perf report -s trace_fields) # Performance counter stats for 'perf report -f --dso perf --stdio -s srcfile,srcline': 44 probe_perf:libdw_addr2line 0.037260744 seconds time elapsed 0.025299000 seconds user 0.011918000 seconds sys # Adding probes to the other addr2line implementations (llvm__addr2line, libbfd__addr2line and cmd__addr2line) I noticed some fallbacks to the llvm one: Performance counter stats for 'perf report -f --dso perf --stdio -s srcfile,srcline': 44 probe_perf:libdw_addr2line 23 probe_perf:llvm_addr2line 0 probe_perf:libbfd_addr2line 0 probe_perf:cmd_addr2line Something to investigate further, but at least we don't fallback to the cmd based one :-) Reviewed-by: James Clark Signed-off-by: Ian Rogers Tested-by: Arnaldo Carvalho de Melo Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Howard Chu Cc: Ingo Molnar Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephen Brennan Cc: Tony Jones Signed-off-by: Arnaldo Carvalho de Melo

perf srcline: Fallback between addr2line implementations

2025-10-06T19:47:49+00:00

Factor the addr2line function implementation into separate source files (addr2line.[ch]) and rename the addr2line function cmd__addr2line. In srcline replace the ifdef-ed addr2line implementations with one that first tries the llvm__addr2line implementation, then the deprecated libbfd__addr2line function and on failure uses cmd__addr2line. If HAVE_LIBLLVM_SUPPORT is enabled the llvm__addr2line will execute against the libLLVM.so it is linked against. If HAVE_LIBLLVM_DYNAMIC is enabled then libperf-llvm.so (that links against libLLVM.so) will be dlopened. If the dlopen succeeds then the behavior should match HAVE_LIBLLVM_SUPPORT. On failure cmd__addr2line is used. The dlopen is only tried once. If HAVE_LIBLLVM_DYNAMIC isn't enabled then llvm__addr2line immediately fails and cmd__addr2line is used. Clean up the dso__free_a2l logic, which is only needed in the non-LLVM version and moved to addr2line.c. Signed-off-by: Ian Rogers Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Alexandre Ghiti Cc: Andi Kleen Cc: Athira Rajeev Cc: Bill Wendling Cc: Charlie Jenkins Cc: Collin Funk Cc: Dmitriy Vyukov Cc: Dr. David Alan Gilbert Cc: Eric Biggers Cc: Haibo Xu Cc: Ingo Molnar Cc: James Clark Cc: Jiri Olsa Cc: Justin Stitt Cc: Li Huafei Cc: Masami Hiramatsu Cc: Namhyung Kim Cc: Nathan Chancellor Cc: Nick Desaulniers Cc: Peter Zijlstra Cc: Song Liu Cc: Stephen Brennan Signed-off-by: Arnaldo Carvalho de Melo

perf libbfd: Move libbfd functionality to its own file

2025-10-02T18:39:44+00:00

Move symbolization and srcline libbfd dependencies to a separate libbfd.c. This mirrors moving llvm and capstone code. While this code is deprecated as it is part of BUILD_NONDISTRO license incompatible code, moving the code to its own file minimizes disruption in the main files. disasm_bpf.c is moved to libbfd.c also except for symbol__disassemble_bpf_image which is currently more of a placeholder function rather than something that provides disassembly support. demangle-cxx.cpp code isn't migrated as it is very limited. Signed-off-by: Ian Rogers Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Alexandre Ghiti Cc: Andi Kleen Cc: Athira Rajeev Cc: Bill Wendling Cc: Charlie Jenkins Cc: Collin Funk Cc: Dmitriy Vyukov Cc: Dr. David Alan Gilbert Cc: Eric Biggers Cc: Haibo Xu Cc: Ingo Molnar Cc: James Clark Cc: Jiri Olsa Cc: Justin Stitt Cc: Li Huafei Cc: Mark Rutland Cc: Masami Hiramatsu Cc: Namhyung Kim Cc: Nathan Chancellor Cc: Nick Desaulniers Cc: Peter Zijlstra Cc: Song Liu Cc: Stephen Brennan Signed-off-by: Arnaldo Carvalho de Melo

perf llvm: Move llvm functionality into its own file

2025-10-02T18:38:23+00:00

LLVM disassembly support was in disasm.c and addr2line support in srcline.c. Move support out of these files into llvm.[ch] and remove LLVM includes from those files. As disassembly routines can fail, make failure the only option without HAVE_LIBLLVM_SUPPORT. For simplicity's sake, duplicate the read_symbol utility function. The intent with moving LLVM support into a single file is that dynamic support, using dlopen for libllvm, can be added in later patches. This can potentially always succeed or fail, so relying on ifdefs isn't sufficient. Using dlopen is a useful option to minimize the perf tools dependencies and potentially size. Signed-off-by: Ian Rogers Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Alexandre Ghiti Cc: Andi Kleen Cc: Athira Rajeev Cc: Bill Wendling Cc: Charlie Jenkins Cc: Collin Funk Cc: Dmitriy Vyukov Cc: Dr. David Alan Gilbert Cc: Eric Biggers Cc: Haibo Xu Cc: Ingo Molnar Cc: James Clark Cc: Jiri Olsa Cc: Justin Stitt Cc: Li Huafei Cc: Mark Rutland Cc: Masami Hiramatsu Cc: Namhyung Kim Cc: Nathan Chancellor Cc: Nick Desaulniers Cc: Peter Zijlstra Cc: Song Liu Cc: Stephen Brennan Signed-off-by: Arnaldo Carvalho de Melo

perf srcline: Lower verbosity on addr2line debug messages

2025-06-24T17:27:49+00:00

Lower non-error debug messages to verbose 3 or larger. Signed-off-by: Ian Rogers Link: https://lore.kernel.org/r/20250623161930.1421216-2-irogers@google.com Signed-off-by: Namhyung Kim

perf report: Support LLVM for addr2line()

2024-09-03T13:15:16+00:00

In addition to the existing support for libbfd and calling out to an external addr2line command, add support for using libllvm directly. This is both faster than libbfd, and can be enabled in distro builds (the LLVM license has an explicit provision for GPLv2 compatibility). Thus, it is set as the primary choice if available. As an example, running 'perf report' on a medium-size profile with DWARF-based backtraces took 58 seconds with LLVM, 78 seconds with libbfd, 153 seconds with external llvm-addr2line, and I got tired and aborted the test after waiting for 55 minutes with external bfd addr2line (which is the default for perf as compiled by distributions today). Evidently, for this case, the bfd addr2line process needs 18 seconds (on a 5.2 GHz Zen 3) to load the .debug ELF in question, hits the 1-second timeout and gets killed during initialization, getting restarted anew every time. Having an in-process addr2line makes this much more robust. As future extensions, libllvm can be used in many other places where we currently use libbfd or other libraries: - Symbol enumeration (in particular, for PE binaries). - Demangling (including non-Itanium demangling, e.g. Microsoft or Rust). - Disassembling (perf annotate). However, these are much less pressing; most people don't profile PE binaries, and perf has non-bfd paths for ELF. The same with demangling; the default _cxa_demangle path works fine for most users, and while bfd objdump can be slow on large binaries, it is possible to use --objdump=llvm-objdump to get the speed benefits. (It appears LLVM-based demangling is very simple, should we want that.) Tested with LLVM 14, 15, 16, 18 and 19. For some reason, LLVM 12 was not correctly detected using feature_check, and thus was not tested. Committer notes: Added the name and a __maybe_unused to address: 1 13.50 almalinux:8 : FAIL gcc version 8.5.0 20210514 (Red Hat 8.5.0-22) (GCC) util/srcline.c: In function 'dso__free_a2l': util/srcline.c:184:20: error: parameter name omitted void dso__free_a2l(struct dso *) ^~~~~~~~~~~~ make[3]: *** [/git/perf-6.11.0-rc3/tools/build/Makefile.build:158: util] Error 2 Signed-off-by: Steinar H. Gunderson Tested-by: Arnaldo Carvalho de Melo Cc: Ian Rogers Link: https://lore.kernel.org/r/20240803152008.2818485-1-sesse@google.com Signed-off-by: Arnaldo Carvalho de Melo