linux/tools/perf, branch v5.19

perf bpf: Remove undefined behavior from bpf_perf_object__next()

2022-07-27T14:19:39Z

bpf_perf_object__next() folded the last element in the list test with the empty list test. However, this meant that offsets were computed against null and that a struct list_head was compared against a 'struct bpf_perf_object'. Working around this with clang's undefined behavior sanitizer required -fno-sanitize=null and -fno-sanitize=object-size. Remove the undefined behavior by using the regular Linux list APIs and handling the starting case separately from the end testing case. Looking at uses like bpf_perf_object__for_each(), as the constant NULL or non-NULL argument can be constant propagated, the code is no less efficient. Signed-off-by: Ian Rogers Cc: Alexander Shishkin Cc: Alexei Starovoitov Cc: Andrii Nakryiko Cc: Christy Lee Cc: Ingo Molnar Cc: Jiri Olsa Cc: Mark Rutland Cc: Miaoqian Lin Cc: Namhyung Kim Cc: Nathan Chancellor Cc: Nick Desaulniers Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Tom Rix Cc: bpf@vger.kernel.org Cc: llvm@lists.linux.dev Link: https://lore.kernel.org/r/20220726220921.2567761-1-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo

perf symbol: Skip symbols if SHF_ALLOC flag is not set

2022-07-27T14:17:50Z

Some symbols are observed with the 'st_value' field zeroed. E.g. libc.so.6 in Ubuntu contains a symbol '__evoke_link_warning_getwd' which resides in the '.gnu.warning.getwd' section. Unlike normal sections, such kind of sections are used for linker warning when a file calls deprecated functions, but they are not part of memory images, the symbols in these sections should be dropped. This patch checks the section attribute SHF_ALLOC bit, if the bit is not set, it skips symbols to avoid spurious ones. Suggested-by: Fangrui Song Signed-off-by: Leo Yan Acked-by: Namhyung Kim Cc: Alexander Shishkin Cc: Chang Rui Cc: Ian Rogers Cc: Ingo Molnar Cc: Jiri Olsa Cc: Mark Rutland Cc: Peter Zijlstra Link: https://lore.kernel.org/r/20220724060013.171050-3-leo.yan@linaro.org Signed-off-by: Arnaldo Carvalho de Melo

perf symbol: Correct address for bss symbols

2022-07-27T14:17:50Z

When using 'perf mem' and 'perf c2c', an issue is observed that tool reports the wrong offset for global data symbols. This is a common issue on both x86 and Arm64 platforms. Let's see an example, for a test program, below is the disassembly for its .bss section which is dumped with objdump: ... Disassembly of section .bss: 0000000000004040 : ... 0000000000004080 : ... 00000000000040c0 : ... 0000000000004100 : ... First we used 'perf mem record' to run the test program and then used 'perf --debug verbose=4 mem report' to observe what's the symbol info for 'buf1' and 'buf2' structures. # ./perf mem record -e ldlat-loads,ldlat-stores -- false_sharing.exe 8 # ./perf --debug verbose=4 mem report ... dso__load_sym_internal: adjusting symbol: st_value: 0x40c0 sh_addr: 0x4040 sh_offset: 0x3028 symbol__new: buf2 0x30a8-0x30e8 ... dso__load_sym_internal: adjusting symbol: st_value: 0x4080 sh_addr: 0x4040 sh_offset: 0x3028 symbol__new: buf1 0x3068-0x30a8 ... The perf tool relies on libelf to parse symbols, in executable and shared object files, 'st_value' holds a virtual address; 'sh_addr' is the address at which section's first byte should reside in memory, and 'sh_offset' is the byte offset from the beginning of the file to the first byte in the section. The perf tool uses below formula to convert a symbol's memory address to a file address: file_address = st_value - sh_addr + sh_offset ^ ` Memory address We can see the final adjusted address ranges for buf1 and buf2 are [0x30a8-0x30e8) and [0x3068-0x30a8) respectively, apparently this is incorrect, in the code, the structure for 'buf1' and 'buf2' specifies compiler attribute with 64-byte alignment. The problem happens for 'sh_offset', libelf returns it as 0x3028 which is not 64-byte aligned, combining with disassembly, it's likely libelf doesn't respect the alignment for .bss section, therefore, it doesn't return the aligned value for 'sh_offset'. Suggested by Fangrui Song, ELF file contains program header which contains PT_LOAD segments, the fields p_vaddr and p_offset in PT_LOAD segments contain the execution info. A better choice for converting memory address to file address is using the formula: file_address = st_value - p_vaddr + p_offset This patch introduces elf_read_program_header() which returns the program header based on the passed 'st_value', then it uses the formula above to calculate the symbol file address; and the debugging log is updated respectively. After applying the change: # ./perf --debug verbose=4 mem report ... dso__load_sym_internal: adjusting symbol: st_value: 0x40c0 p_vaddr: 0x3d28 p_offset: 0x2d28 symbol__new: buf2 0x30c0-0x3100 ... dso__load_sym_internal: adjusting symbol: st_value: 0x4080 p_vaddr: 0x3d28 p_offset: 0x2d28 symbol__new: buf1 0x3080-0x30c0 ... Fixes: f17e04afaff84b5c ("perf report: Fix ELF symbol parsing") Reported-by: Chang Rui Suggested-by: Fangrui Song Signed-off-by: Leo Yan Acked-by: Namhyung Kim Cc: Alexander Shishkin Cc: Ian Rogers Cc: Ingo Molnar Cc: Jiri Olsa Cc: Mark Rutland Cc: Peter Zijlstra Link: https://lore.kernel.org/r/20220724060013.171050-2-leo.yan@linaro.org Signed-off-by: Arnaldo Carvalho de Melo

perf scripts python: Let script to be python2 compliant

2022-07-27T14:17:50Z

The mainline kernel can be used for relative old distros, e.g. RHEL 7. The distro doesn't upgrade from python2 to python3, this causes the building error that the python script is not python2 compliant. To fix the building failure, this patch changes from the python f-string format to traditional string format. Fixes: 12fdd6c009da0d02 ("perf scripts python: Support Arm CoreSight trace data disassembly") Reported-by: Akemi Yagi Signed-off-by: Leo Yan Cc: Alexander Shishkin Cc: ElRepo Cc: Ian Rogers Cc: Ingo Molnar Cc: Jiri Olsa Cc: Leo Yan Cc: Mark Rutland Cc: Namhyung Kim Cc: Peter Zijlstra Link: https://lore.kernel.org/r/20220725104220.1106663-1-leo.yan@linaro.org Signed-off-by: Arnaldo Carvalho de Melo

perf trace: Fix SIGSEGV when processing syscall args

2022-07-17T13:59:52Z

On powerpc, 'perf trace' is crashing with a SIGSEGV when trying to process a perf.data file created with 'perf trace record -p': #0 0x00000001225b8988 in syscall_arg__scnprintf_augmented_string at builtin-trace.c:1492 #1 syscall_arg__scnprintf_filename at builtin-trace.c:1492 #2 syscall_arg__scnprintf_filename at builtin-trace.c:1486 #3 0x00000001225bdd9c in syscall_arg_fmt__scnprintf_val at builtin-trace.c:1973 #4 syscall__scnprintf_args at builtin-trace.c:2041 #5 0x00000001225bff04 in trace__sys_enter at builtin-trace.c:2319 That points to the below code in tools/perf/builtin-trace.c: /* * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible * arguments, even if the syscall being handled, say "openat", uses only 4 arguments * this breaks syscall__augmented_args() check for augmented args, as we calculate * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file, * so when handling, say the openat syscall, we end up getting 6 args for the * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly * thinking that the extra 2 u64 args are the augmented filename, so just check * here and avoid using augmented syscalls when the evsel is the raw_syscalls one. */ if (evsel != trace->syscalls.events.sys_enter) augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size); As the comment points out, we should not be trying to augment the args for raw_syscalls. However, when processing a perf.data file, we are not initializing those properly. Fix the same. Reported-by: Claudio Carvalho Signed-off-by: Naveen N. Rao Cc: Jiri Olsa Cc: Namhyung Kim Link: http://lore.kernel.org/lkml/20220707090900.572584-1-naveen.n.rao@linux.vnet.ibm.com Signed-off-by: Arnaldo Carvalho de Melo

perf tests: Fix Convert perf time to TSC test for hybrid

2022-07-17T13:57:07Z

The test does not always correctly determine the number of events for hybrids, nor allow for more than 1 evsel when parsing. Fix by iterating the events actually created and getting the correct evsel for the events processed. Fixes: d9da6f70eb235110 ("perf tests: Support 'Convert perf time to TSC' test for hybrid") Reviewed-by: Kan Liang Signed-off-by: Adrian Hunter Cc: Ian Rogers Cc: Jin Yao Cc: Jiri Olsa Cc: Namhyung Kim Cc: Thomas Richter Link: https://lore.kernel.org/r/20220713123459.24145-3-adrian.hunter@intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf tests: Stop Convert perf time to TSC test opening events twice

2022-07-17T13:56:58Z

Do not call evlist__open() twice. Fixes: 5bb017d4b97a0f13 ("perf test: Fix error message for test case 71 on s390, where it is not supported") Reviewed-by: Kan Liang Signed-off-by: Adrian Hunter Cc: Ian Rogers Cc: Jiri Olsa Cc: Namhyung Kim Cc: Thomas Richter Link: https://lore.kernel.org/r/20220713123459.24145-2-adrian.hunter@intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf synthetic-events: Ignore dead threads during event synthesis

2022-07-02T12:22:26Z

When it synthesize various task events, it scans the list of task first and then accesses later. There's a window threads can die between the two and proc entries may not be available. Instead of bailing out, we can ignore that thread and move on. Signed-off-by: Namhyung Kim Acked-by: Ian Rogers Cc: Jiri Olsa Cc: Peter Zijlstra Link: http://lore.kernel.org/lkml/20220701205458.985106-2-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf synthetic-events: Don't sort the task scan result from /proc

2022-07-02T12:21:41Z

It should not sort the result as procfs already returns a proper ordering of tasks. Actually sorting the order caused problems that it doesn't guararantee to process the main thread first. Signed-off-by: Namhyung Kim Acked-by: Ian Rogers Cc: Jiri Olsa Cc: Peter Zijlstra Link: http://lore.kernel.org/lkml/20220701205458.985106-1-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf unwind: Fix unitialized 'offset' variable on aarch64

2022-07-02T12:16:52Z

Commit dc2cf4ca866f5715 ("perf unwind: Fix segbase for ld.lld linked objects") uncovered the following issue on aarch64: util/unwind-libunwind-local.c: In function 'find_proc_info': util/unwind-libunwind-local.c:386:28: error: 'offset' may be used uninitialized in this function [-Werror=maybe-uninitialized] 386 | if (ofs > 0) { | ^ util/unwind-libunwind-local.c:199:22: note: 'offset' was declared here 199 | u64 address, offset; | ^~~~~~ util/unwind-libunwind-local.c:371:20: error: 'offset' may be used uninitialized in this function [-Werror=maybe-uninitialized] 371 | if (ofs <= 0) { | ^ util/unwind-libunwind-local.c:199:22: note: 'offset' was declared here 199 | u64 address, offset; | ^~~~~~ util/unwind-libunwind-local.c:363:20: error: 'offset' may be used uninitialized in this function [-Werror=maybe-uninitialized] 363 | if (ofs <= 0) { | ^ util/unwind-libunwind-local.c:199:22: note: 'offset' was declared here 199 | u64 address, offset; | ^~~~~~ In file included from util/libunwind/arm64.c:37: Fixes: dc2cf4ca866f5715 ("perf unwind: Fix segbase for ld.lld linked objects") Signed-off-by: Ivan Babrou Cc: Alexander Shishkin Cc: Fangrui Song Cc: Ian Rogers Cc: James Clark Cc: Jiri Olsa Cc: kernel-team@cloudflare.com Cc: Mark Rutland Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lore.kernel.org/lkml/20220701182046.12589-1-ivan@cloudflare.com Signed-off-by: Arnaldo Carvalho de Melo