kernel/tools/perf, branch linux-4.15.y

perf tools: Fix copyfile_offset update of output offset

2018-04-12T10:31:13Z

[ Upstream commit fa1195ccc0af2d121abe0fe266a1caee8c265eea ] We need to increase output offset in each iteration, not decrease it as we currently do. I guess we were lucky to finish in most cases in first iteration, so the bug never showed. However it shows a lot when working with big (~4GB) size data. Signed-off-by: Jiri Olsa Cc: Alexander Shishkin Cc: Andi Kleen Cc: David Ahern Cc: Namhyung Kim Cc: Peter Zijlstra Fixes: 9c9f5a2f1944 ("perf tools: Introduce copyfile_offset() function") Link: http://lkml.kernel.org/r/20180109133923.25406-1-jolsa@kernel.org Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf report: Fix a no annotate browser displayed issue

2018-04-12T10:31:08Z

[ Upstream commit 40c39e3046411f84bab82f66783ff3593e2bcd9b ] When enabling '-b' option in perf record, for example, perf record -b ... perf report and then browsing the annotate browser from perf report (press 'A'), it would fail (annotate browser can't be displayed). It's because the '.add_entry_cb' op of struct report is overwritten by hist_iter__branch_callback() in builtin-report.c. But this function doesn't do something like mapping symbols and sources. So next, do_annotate() will return directly. notes = symbol__annotation(act->ms.sym); if (!notes->src) return 0; This patch adds the lost code to hist_iter__branch_callback (refer to hist_iter__report_callback). v2: Fix a crash bug when perform 'perf report --stdio'. The reason is that we init the symbol annotation only in browser mode, it doesn't allocate/init resources for stdio mode. So now in hist_iter__branch_callback(), it will return directly if it's not in browser mode. Signed-off-by: Jin Yao Tested-by: Arnaldo Carvalho de Melo Cc: Alexander Shishkin Cc: Andi Kleen Cc: Jiri Olsa Cc: Kan Liang Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1514284963-18587-1-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf evsel: Enable ignore_missing_thread for pid option

2018-04-12T10:31:02Z

[ Upstream commit ca8000684ec4e66f965e1f9547a3c6cb834154ca ] While monitoring a multithread process with pid option, perf sometimes may return sys_perf_event_open failure with 3(No such process) if any of the process's threads die before we open the event. However, we want perf continue monitoring the remaining threads and do not exit with error. Here, the patch enables perf_evsel::ignore_missing_thread for -p option to ignore complete failure if any of threads die before we open the event. But it may still return sys_perf_event_open failure with 22(Invalid) if we monitors several event groups. sys_perf_event_open: pid 28960 cpu 40 group_fd 118202 flags 0x8 sys_perf_event_open: pid 28961 cpu 40 group_fd 118203 flags 0x8 WARNING: Ignored open failure for pid 28962 sys_perf_event_open: pid 28962 cpu 40 group_fd [118203] flags 0x8 sys_perf_event_open failed, error -22 That is because when we ignore a missing thread, we change the thread_idx without dealing with its fds, FD(evsel, cpu, thread). Then get_group_fd() may return a wrong group_fd for the next thread and sys_perf_event_open() return with 22. sys_perf_event_open(){ ... if (group_fd != -1) perf_fget_light()//to get corresponding group_leader by group_fd ... if (group_leader) if (group_leader->ctx->task != ctx->task)//should on the same task goto err_context ... } This patch also fixes this bug by introducing perf_evsel__remove_fd() and update_fds to allow removing fds for the missing thread. Changes since v1: - Change group_fd__remove() into a more genetic way without changing code logic - Remove redundant condition Changes since v2: - Use a proper function name and add some comment. - Multiline comment style fixes. Committer testing: Before this patch the recently added 'perf stat --per-thread' for system wide counting would race while enumerating all threads using /proc: [root@jouet ~]# perf stat --per-thread failed to parse CPUs map: No such file or directory Usage: perf stat [] [] -C, --cpu list of cpus to monitor in system-wide -a, --all-cpus system-wide collection from all CPUs [root@jouet ~]# perf stat --per-thread failed to parse CPUs map: No such file or directory Usage: perf stat [] [] -C, --cpu list of cpus to monitor in system-wide -a, --all-cpus system-wide collection from all CPUs [root@jouet ~]# When, say, the kernel was being built, so lots of shortlived threads, after this patch this doesn't happen. Signed-off-by: Mengting Zhang Tested-by: Arnaldo Carvalho de Melo Acked-by: Jiri Olsa Cc: Cheng Jian Cc: Li Bin Cc: Wang Nan Link: http://lkml.kernel.org/r/1513148513-6974-1-git-send-email-zhangmengting@huawei.com [ Remove one use 'evlist' alias variable ] Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf evsel: Fix swap for samples with raw data

2018-04-12T10:31:02Z

[ Upstream commit f9d8adb345d7adbb2d3431eea73beb89c8d6d612 ] When we detect a different endianity we swap event before processing. It's tricky for samples because we have no idea what's inside. We treat it as an array of u64s, swap them and later on we swap back parts which are different. We mangle this way also the tracepoint raw data, which ends up in report showing wrong data: 1.95% comm=Q^B pid=29285 prio=16777216 target_cpu=000 1.67% comm=l^B pid=0 prio=16777216 target_cpu=000 Luckily the traceevent library handles the endianity by itself (thank you Steven!), so we can pass the RAW data directly in the other endianity. 2.51% comm=beah-rhts-task pid=1175 prio=120 target_cpu=002 2.23% comm=kworker/0:0 pid=11566 prio=120 target_cpu=000 The fix is basically to swap back the raw data if different endianity is detected. Signed-off-by: Jiri Olsa Cc: David Ahern Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Steven Rostedt Link: http://lkml.kernel.org/r/20171129184346.3656-1-jolsa@kernel.org [ Add util/memswap.c to python-ext-sources to link missing mem_bswap_64() ] Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf probe: Add warning message if there is unexpected event name

2018-04-12T10:31:02Z

[ Upstream commit 9f5c6d8777a2d962b0eeacb2a16f37da6bea545b ] This improve the error message so that user can know event-name error before writing new events to kprobe-events interface. E.g. ====== #./perf probe -x /lib64/libc-2.25.so malloc_get_state* Internal error: "malloc_get_state@GLIBC_2" is an invalid event name. Error: Failed to add events. ====== Reported-by: Arnaldo Carvalho de Melo Signed-off-by: Masami Hiramatsu Acked-by: Ravi Bangoria Reviewed-by: Thomas Richter Tested-by: Arnaldo Carvalho de Melo Cc: Paul Clarke Cc: bhargavb Cc: linux-rt-users@vger.kernel.org Link: http://lkml.kernel.org/r/151275040665.24652.5188568529237584489.stgit@devbox Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf probe: Find versioned symbols from map

2018-04-12T10:31:02Z

[ Upstream commit 4b3a2716dd785fabb9f6ac80c1d53cb29a88169d ] Commit d80406453ad4 ("perf symbols: Allow user probes on versioned symbols") allows user to find default versioned symbols (with "@@") in map. However, it did not enable normal versioned symbol (with "@") for perf-probe. E.g. ===== # ./perf probe -x /lib64/libc-2.25.so malloc_get_state Failed to find symbol malloc_get_state in /usr/lib64/libc-2.25.so Error: Failed to add events. ===== This solves above issue by improving perf-probe symbol search function, as below. ===== # ./perf probe -x /lib64/libc-2.25.so malloc_get_state Added new event: probe_libc:malloc_get_state (on malloc_get_state in /usr/lib64/libc-2.25.so) You can now use it in all perf tools, such as: perf record -e probe_libc:malloc_get_state -aR sleep 1 # ./perf probe -l probe_libc:malloc_get_state (on malloc_get_state@GLIBC_2.2.5 in /usr/lib64/libc-2.25.so) ===== Signed-off-by: Masami Hiramatsu Reviewed-by: Thomas Richter Acked-by: Ravi Bangoria Tested-by: Arnaldo Carvalho de Melo Cc: Paul Clarke Cc: bhargavb Cc: linux-rt-users@vger.kernel.org Link: http://lkml.kernel.org/r/151275049269.24652.1639103455496216255.stgit@devbox Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf stat: Fix CVS output format for non-supported counters

2018-03-28T16:23:03Z

commit 40c21898ba5372c14ef71717040529794a91ccc2 upstream. When printing stats in CSV mode, 'perf stat' appends extra separators when a counter is not supported: ,,L1-dcache-store-misses,mesos/bd442f34-2b4a-47df-b966-9b281f9f56fc,0,100.00,,,, Which causes a failure when parsing fields. The numbers of separators should be the same for each line, no matter if the counter is or not supported. Signed-off-by: Ilya Pronin Acked-by: Jiri Olsa Cc: Andi Kleen Link: http://lkml.kernel.org/r/20180306064353.31930-1-xiyou.wangcong@gmail.com Fixes: 92a61f6412d3 ("perf stat: Implement CSV metrics output") Signed-off-by: Cong Wang Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Greg Kroah-Hartman

perf annotate: Fix objdump comment parsing for Intel mov dissassembly

2018-03-19T08:09:56Z

[ Upstream commit 35a8a148d8c1ee9e5ae18f9565a880490f816f89 ] The command 'perf annotate' parses the output of objdump and also investigates the comments produced by objdump. For example the output of objdump produces (on x86): 23eee: 4c 8b 3d 13 01 21 00 mov 0x210113(%rip),%r15 # 234008 and the function mov__parse() is called to investigate the complete line. Mov__parse() breaks this line into several parts and finally calls function comment__symbol() to parse the data after the comment character '#'. Comment__symbol() expects a hexadecimal address followed by a symbol in '<' and '>' brackets. However the 2nd parameter given to function comment__symbol() always points to the comment character '#'. The address parsing always returns 0 because the character '#' is not a digit and strtoull() fails without being noticed. Fix this by advancing the second parameter to function comment__symbol() by one byte before invocation and add an error check after strtoull() has been called. Signed-off-by: Thomas Richter Reviewed-by: Hendrik Brueckner Acked-by: Ravi Bangoria Cc: Heiko Carstens Cc: Martin Schwidefsky Fixes: 6de783b6f50f ("perf annotate: Resolve symbols using objdump comment") Link: http://lkml.kernel.org/r/20171128075632.72182-1-tmricht@linux.vnet.ibm.com Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf annotate: Fix unnecessary memory allocation for s390x

2018-03-19T08:09:56Z

[ Upstream commit 36c263607d36c6a3788c09301d9f5fe35404048a ] This patch fixes a bug introduced with commit d9f8dfa9baf9 ("perf annotate s390: Implement jump types for perf annotate"). 'perf annotate' displays annotated assembler output by reading output of command objdump and parsing the disassembled lines. For each shown mnemonic this function sequence is executed: disasm_line__new() | +--> disasm_line__init_ins() | +--> ins__find() | +--> arch->associate_instruction_ops() The s390x specific function assigned to function pointer associate_instruction_ops refers to function s390__associate_ins_ops(). This function checks for supported mnemonics and assigns a NULL pointer to unsupported mnemonics. However even the NULL pointer is added to the architecture dependend instruction array. This leads to an extremely large architecture instruction array (due to array resize logic in function arch__grow_instructions()). Depending on the objdump output being parsed the array can end up with several ten-thousand elements. This patch checks if a mnemonic is supported and only adds supported ones into the architecture instruction array. The array does not contain elements with NULL pointers anymore. Before the patch (With some debug printf output): [root@s35lp76 perf]# time ./perf annotate --stdio > /tmp/xxxbb real 8m49.679s user 7m13.008s sys 0m1.649s [root@s35lp76 perf]# fgrep '__ins__find sorted:1 nr_instructions:' /tmp/xxxbb | tail -1 __ins__find sorted:1 nr_instructions:87433 ins:0x341583c0 [root@s35lp76 perf]# The number of different s390x branch/jump/call/return instructions entered into the array is 87433. After the patch (With some printf debug output:) [root@s35lp76 perf]# time ./perf annotate --stdio > /tmp/xxxaa real 1m24.553s user 0m0.587s sys 0m1.530s [root@s35lp76 perf]# fgrep '__ins__find sorted:1 nr_instructions:' /tmp/xxxaa | tail -1 __ins__find sorted:1 nr_instructions:56 ins:0x3f406570 [root@s35lp76 perf]# The number of different s390x branch/jump/call/return instructions entered into the array is 56 which is sensible. Signed-off-by: Thomas Richter Reviewed-by: Hendrik Brueckner Acked-by: Ravi Bangoria Cc: Heiko Carstens Cc: Martin Schwidefsky Link: http://lkml.kernel.org/r/20171124094637.55558-1-tmricht@linux.vnet.ibm.com Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

perf report: Fix -D output for user metadata events

2018-03-19T08:09:52Z

[ Upstream commit f250b09c779550e4a7a412dae6d3ad34d5201019 ] The PERF_RECORD_USER_ events are synthesized by the tool to assist in processing the PERF_RECORD_ ones generated by the kernel, the printing of that information doesn't come with a perf_sample structure, so, when dumping the event fields using 'perf report -D' there were columns that end up not being printed. To tidy up a bit this, fake a perf_sample structure with zeroes to have the missing columns printed and avoid the occasional surprise with that. Before: 0 0x45b8 [0x68]: PERF_RECORD_MMAP -1/0: [0xffffffffc12ec000(0x4000) @ 0]: x /lib/modules/4.14.0+/kernel/fs/nls/nls_utf8.ko 0x4620 [0x28]: PERF_RECORD_THREAD_MAP nr: 1 thread: 27820 0x4648 [0x18]: PERF_RECORD_CPU_MAP: 0-3 0 0x4660 [0x28]: PERF_RECORD_COMM: perf:27820/27820 0x4a58 [0x8]: PERF_RECORD_FINISHED_ROUND 447723433020976 0x4688 [0x28]: PERF_RECORD_SAMPLE(IP, 0x4001): 27820/27820: 0xffffffff8f1b6d7a period: 1 addr: 0 After: $ perf report -D | grep PERF_RECORD_ | head 0 0xe8 [0x20]: PERF_RECORD_TIME_CONV: unhandled! 0 0x108 [0x28]: PERF_RECORD_THREAD_MAP nr: 1 thread: 32555 0 0x130 [0x18]: PERF_RECORD_CPU_MAP: 0-3 0 0x148 [0x28]: PERF_RECORD_COMM: perf:32555/32555 0 0x4e8 [0x8]: PERF_RECORD_FINISHED_ROUND 448743409421205 0x170 [0x28]: PERF_RECORD_COMM exec: sleep:32555/32555 448743409431883 0x198 [0x68]: PERF_RECORD_MMAP2 32555/32555: [0x55e11d75a000(0x208000) @ 0 fd:00 3147174 2566255743]: r-xp /usr/bin/sleep 448743409443873 0x200 [0x70]: PERF_RECORD_MMAP2 32555/32555: [0x7f0ced316000(0x229000) @ 0 fd:00 3151761 2566238119]: r-xp /usr/lib64/ld-2.25.so 448743409454790 0x270 [0x60]: PERF_RECORD_MMAP2 32555/32555: [0x7ffe84f6d000(0x2000) @ 0 00:00 0 0]: r-xp [vdso] 448743409479500 0x2d0 [0x28]: PERF_RECORD_SAMPLE(IP, 0x4002): 32555/32555: 0xffffffff8f84c7e7 period: 1 addr: 0 $ Cc: Adrian Hunter Cc: Andi Kleen Cc: David Ahern Cc: Jiri Olsa Cc: Namhyung Kim Cc: Wang Nan Cc: Thomas Gleixner Fixes: 9aefcab0de47 ("perf session: Consolidate the dump code") Link: https://lkml.kernel.org/n/tip-todcu15x0cwgppkh1gi6uhru@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman