kernel/linux.git/tools/perf/util/session.c, branch v5.19

perf tools: Add guest_code support

2022-05-23T13:18:38+00:00

A common case for KVM test programs is that the test program acts as the hypervisor, creating, running and destroying the virtual machine, and providing the guest object code from its own object code. In this case, the VM is not running an OS, but only the functions loaded into it by the hypervisor test program, and conveniently, loaded at the same virtual addresses. Normally to resolve addresses, MMAP events are needed to map addresses back to the object code and debug symbols for that object code. Currently, there is no way to get such mapping information from guests but, in the scenario described above, the guest has the same mappings as the hypervisor, so support for that scenario can be achieved. To support that, copy the host thread's maps to the guest thread's maps. Note, we do not discover the guest until we encounter a guest event, which works well because it is not until then that we know that the host thread's maps have been set up. Typically the main function for the guest object code is called "guest_code", hence the name chosen for this feature. Note, that is just a convention, the function could be named anything, and the tools do not care. This is primarily aimed at supporting Intel PT, or similar, where trace data can be recorded for a guest. Refer to the final patch in this series "perf intel-pt: Add guest_code support" for an example. Signed-off-by: Adrian Hunter Cc: Andi Kleen Cc: Ian Rogers Cc: Jiri Olsa Cc: Leo Yan Cc: Namhyung Kim Cc: kvm@vger.kernel.org Link: https://lore.kernel.org/r/20220517131011.6117-4-adrian.hunter@intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf session: Fix Intel LBR callstack entries and nr print message

2022-05-21T17:56:24+00:00

When generating callstack information from branch_stack(Intel LBR), the actual number of callstack entry should be bigger than the number of branch_stack, for example: branch_stack records: B() -> C() A() -> B() converted callstack records should be: C() B() A() though, the number of callstack equals to the number of branch stack plus 1. This patch fixes above issue in branch_stack__printf(). For example, # echo 'scale=2000; 4*a(1)' > cmd # perf record --call-graph lbr bc -l < cmd Before applying this patch, `perf script -D` output: 1220022677386876 0x2a40 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x4002): 17990/17990: 0x40a6d6 period: 894172 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 000000000040a410 ..... 2: 000000000040573c ..... 3: 0000000000408650 ..... 4: 00000000004022f2 ..... 5: 00000000004015f5 ..... 6: 00007f5ed6dcb553 ..... 7: 0000000000401698 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 000000000040a6d8 ... branch callstack: nr:6 # which is not consistent with LBR records. ..... 0: 000000000040a410 ..... 1: 0000000000408650 # ditto ..... 2: 00000000004022f2 ..... 3: 00000000004015f5 ..... 4: 00007f5ed6dcb553 ..... 5: 0000000000401698 ... thread: bc:17990 ...... dso: /usr/bin/bc bc 17990 1220022.677386: 894172 cycles: 40a410 [unknown] (/usr/bin/bc) 40573c [unknown] (/usr/bin/bc) 408650 [unknown] (/usr/bin/bc) 4022f2 [unknown] (/usr/bin/bc) 4015f5 [unknown] (/usr/bin/bc) 7f5ed6dcb553 __libc_start_main+0xf3 (/usr/lib64/libc-2.17.so) 401698 [unknown] (/usr/bin/bc) After applied: 1220022677386876 0x2a40 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x4002): 17990/17990: 0x40a6d6 period: 894172 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 000000000040a410 ..... 2: 000000000040573c ..... 3: 0000000000408650 ..... 4: 00000000004022f2 ..... 5: 00000000004015f5 ..... 6: 00007f5ed6dcb553 ..... 7: 0000000000401698 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 000000000040a6d8 ... branch callstack: nr:7 ..... 0: 000000000040a410 ..... 1: 000000000040573c ..... 2: 0000000000408650 ..... 3: 00000000004022f2 ..... 4: 00000000004015f5 ..... 5: 00007f5ed6dcb553 ..... 6: 0000000000401698 ... thread: bc:17990 ...... dso: /usr/bin/bc bc 17990 1220022.677386: 894172 cycles: 40a410 [unknown] (/usr/bin/bc) 40573c [unknown] (/usr/bin/bc) 408650 [unknown] (/usr/bin/bc) 4022f2 [unknown] (/usr/bin/bc) 4015f5 [unknown] (/usr/bin/bc) 7f5ed6dcb553 __libc_start_main+0xf3 (/usr/lib64/libc-2.17.so) 401698 [unknown] (/usr/bin/bc) Change from v1: - refined code style according to Jiri's review comments. Signed-off-by: Chengdong Li Acked-by: Namhyung Kim Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Alexey Bayduraev Cc: Andi Kleen Cc: German Gomez Cc: Ian Rogers Cc: Ingo Molnar Cc: Jiri Olsa Cc: Mark Rutland Cc: Peter Zijlstra Cc: Riccardo Mancini Cc: likexu@tencent.com Link: https://lore.kernel.org/r/20220517015726.96131-1-chengdongli@tencent.com Signed-off-by: Arnaldo Carvalho de Melo

perf intel-pt: Fix timeless decoding with perf.data directory

2022-04-28T13:20:52+00:00

Intel PT does not capture data in separate directories, so do not use separate directory processing because it doesn't work for timeless decoding. It also looks like it doesn't support one_mmap handling. Example: Before: # perf record --kcore -a -e intel_pt/tsc=0/k sleep 0.1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.799 MB perf.data ] # perf script --itrace=bep | head # After: # perf script --itrace=bep | head perf 21073 [000] psb: psb offs: 0 ffffffffaa68faf4 native_write_msr+0x4 ([kernel.kallsyms]) perf 21073 [000] cbr: cbr: 45 freq: 4505 MHz (161%) ffffffffaa68faf4 native_write_msr+0x4 ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: 0 [unknown] ([unknown]) => ffffffffaa68faf6 native_write_msr+0x6 ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: ffffffffaa68faf8 native_write_msr+0x8 ([kernel.kallsyms]) => ffffffffaa61aab0 pt_config_start+0x60 ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: ffffffffaa61aabd pt_config_start+0x6d ([kernel.kallsyms]) => ffffffffaa61b8ad pt_event_start+0x27d ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: ffffffffaa61b8bb pt_event_start+0x28b ([kernel.kallsyms]) => ffffffffaa61ba60 pt_event_add+0x40 ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: ffffffffaa61ba76 pt_event_add+0x56 ([kernel.kallsyms]) => ffffffffaa880e86 event_sched_in+0xc6 ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: ffffffffaa880e9b event_sched_in+0xdb ([kernel.kallsyms]) => ffffffffaa880ea5 event_sched_in+0xe5 ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: ffffffffaa880eba event_sched_in+0xfa ([kernel.kallsyms]) => ffffffffaa880f96 event_sched_in+0x1d6 ([kernel.kallsyms]) perf 21073 [000] 1 branches:k: ffffffffaa880fc8 event_sched_in+0x208 ([kernel.kallsyms]) => ffffffffaa880ec0 event_sched_in+0x100 ([kernel.kallsyms]) Fixes: bb6be405c4a2a5 ("perf session: Load data directory files for analysis") Cc: stable@vger.kernel.org Signed-off-by: Adrian Hunter Signed-off-by: Arnaldo Carvalho de Melo Link: https://lore.kernel.org/r/20220428093109.274641-1-adrian.hunter@intel.com Cc: Ian Rogers Cc: Arnaldo Carvalho de Melo Cc: Namhyung Kim Cc: Alexey Bayduraev Cc: Jiri Olsa Cc: linux-kernel@vger.kernel.org

perf session: Remap buf if there is no space for event

2022-04-09T17:20:59+00:00

If a perf event doesn't fit into remaining buffer space return NULL to remap buf and fetch the event again. Keep the logic to error out on inadequate input from fuzzing. This fixes perf failing on ChromeOS (with 32b userspace): $ perf report -v -i perf.data ... prefetch_event: head=0x1fffff8 event->header_size=0x30, mmap_size=0x2000000: fuzzed or compressed perf.data? Error: failed to process sample Fixes: 57fc032ad643ffd0 ("perf session: Avoid infinite loop when seeing invalid header.size") Reviewed-by: James Clark Signed-off-by: Denis Nikitin Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Alexey Budankov Cc: Namhyung Kim Link: https://lore.kernel.org/r/20220330031130.2152327-1-denik@chromium.org Signed-off-by: Arnaldo Carvalho de Melo

perf session: Print branch stack entry type in --dump-raw-trace

2022-03-07T17:51:19+00:00

This can help with debugging issues. It only prints when -j save_type is used otherwise an empty string is printed. Before the change: 101603801707130 0xa70 [0x630]: PERF_RECORD_SAMPLE(IP, 0x2): 1108/1108: 0xffff9c1df24c period: 10694 addr: 0 ... branch stack: nr:64 ..... 0: 0000ffff9c26029c -> 0000ffff9c26f340 0 cycles P 0 ..... 1: 0000ffff9c2601bc -> 0000ffff9c26f340 0 cycles P 0 After the change: 101603801707130 0xa70 [0x630]: PERF_RECORD_SAMPLE(IP, 0x2): 1108/1108: 0xffff9c1df24c period: 10694 addr: 0 ... branch stack: nr:64 ..... 0: 0000ffff9c26029c -> 0000ffff9c26f340 0 cycles P 0 CALL ..... 1: 0000ffff9c2601bc -> 0000ffff9c26f340 0 cycles P 0 IND_CALL Signed-off-by: James Clark Cc: Alexander Shishkin Cc: Anshuman Khandual Cc: German Gomez Cc: Jiri Olsa Cc: Leo Yan Cc: Mark Rutland Cc: Namhyung Kim Link: https://lore.kernel.org/r/20220307171917.2555829-3-james.clark@arm.com Signed-off-by: Arnaldo Carvalho de Melo

perf report: Output data file name in raw trace dump

2022-02-10T19:27:34+00:00

Print path and name of a data file into raw dump (-D) @: 0x2226a@perf.data [0x30]: event: 9 or 0x15cc36@perf.data/data.7 [0x30]: event: 9 Reviewed-by: Riccardo Mancini Signed-off-by: Alexey Bayduraev Tested-by: Jiri Olsa Tested-by: Riccardo Mancini Acked-by: Andi Kleen Acked-by: Namhyung Kim Cc: Adrian Hunter Cc: Alexander Antonov Cc: Alexander Shishkin Cc: Alexei Budankov Cc: Ingo Molnar Cc: Peter Zijlstra Link: https://lore.kernel.org/r/e8378fd4910c10751b001be880705653989283c2.1642440724.git.alexey.v.bayduraev@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf session: Load data directory files for analysis

2022-02-10T19:27:28+00:00

Load data directory files and provide basic raw dump and aggregated analysis support of data directories in report mode, still with no memory consumption optimizations. READER_MAX_SIZE is chosen based on the results of measurements on different machines on perf.data directory sizes >1GB. On machines with big core count (192 cores) the difference between 1MB and 2MB is about 4%. Other sizes (>2MB) are quite equal to 2MB. On machines with small core count (4-24) there is no differences between 1-16 MB sizes. So this constant is 2MB. Suggested-by: Jiri Olsa Reviewed-by: Riccardo Mancini Signed-off-by: Alexey Bayduraev Tested-by: Jiri Olsa Tested-by: Riccardo Mancini Acked-by: Namhyung Kim Cc: Adrian Hunter Cc: Alexander Antonov Cc: Alexander Shishkin Cc: Alexei Budankov Cc: Andi Kleen Cc: Ingo Molnar Cc: Namhyung Kim Cc: Peter Zijlstra Link: https://lore.kernel.org/r/3f10c13a226c0ceb53e88a082f847b91c1ae2c25.1642440724.git.alexey.v.bayduraev@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf session: Check for NULL pointer before dereference

2022-02-06T12:03:06+00:00

Move NULL pointer check before dereferencing the variable. Addresses-Coverity: 1497622 ("Derereference before null check") Reviewed-by: James Clark Signed-off-by: Ameer Hamza Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Alexey Bayduraev Cc: German Gomez Cc: Ingo Molnar Cc: Jiri Olsa Cc: Leo Yan Cc: Mark Rutland Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Riccardo Mancini Link: https://lore.kernel.org/r/20220125121141.18347-1-amhamza.mgc@gmail.com Signed-off-by: Arnaldo Carvalho de Melo

perf cpumap: Migrate to libperf cpumap api

2022-01-22T20:08:42+00:00

Switch from directly accessing the perf_cpu_map to using the appropriate libperf API when possible. Using the API simplifies the job of refactoring use of perf_cpu_map. Signed-off-by: Ian Rogers Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Alexey Bayduraev Cc: Andi Kleen Cc: Andrew Morton Cc: André Almeida Cc: Andy Shevchenko Cc: Darren Hart Cc: Davidlohr Bueso Cc: Dmitriy Vyukov Cc: Eric Dumazet Cc: German Gomez Cc: James Clark Cc: Jin Yao Cc: Jiri Olsa Cc: John Garry Cc: Kajol Jain Cc: Kan Liang Cc: Leo Yan Cc: Madhavan Srinivasan Cc: Mark Rutland Cc: Masami Hiramatsu Cc: Miaoqian Lin Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Riccardo Mancini Cc: Shunsuke Nakamura Cc: Song Liu Cc: Stephane Eranian Cc: Stephen Brennan Cc: Steven Rostedt (VMware) Cc: Thomas Gleixner Cc: Thomas Richter Cc: Yury Norov Link: http://lore.kernel.org/lkml/20220122045811.3402706-3-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo

perf cpumap: Give CPUs their own type

2022-01-12T17:28:23+00:00

A common problem is confusing CPU map indices with the CPU, by wrapping the CPU with a struct then this is avoided. This approach is similar to atomic_t. Committer notes: To make it build with BUILD_BPF_SKEL=1 these files needed the conversions to 'struct perf_cpu' usage: tools/perf/util/bpf_counter.c tools/perf/util/bpf_counter_cgroup.c tools/perf/util/bpf_ftrace.c Also perf_env__get_cpu() was removed back in "perf cpumap: Switch cpu_map__build_map to cpu function". Additionally these needed to be fixed for the ARM builds to complete: tools/perf/arch/arm/util/cs-etm.c tools/perf/arch/arm64/util/pmu.c Suggested-by: John Garry Signed-off-by: Ian Rogers Cc: Alexander Shishkin Cc: Andi Kleen Cc: Ingo Molnar Cc: James Clark Cc: Jiri Olsa Cc: Kajol Jain Cc: Kan Liang Cc: Leo Yan Cc: Mark Rutland Cc: Mathieu Poirier Cc: Mike Leach Cc: Namhyung Kim Cc: Paul Clarke Cc: Peter Zijlstra Cc: Riccardo Mancini Cc: Stephane Eranian Cc: Suzuki Poulouse Cc: Vineet Singh Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: zhengjun.xing@intel.com Link: https://lore.kernel.org/r/20220105061351.120843-49-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo