- 28 Jul, 2017 8 commits
-
-
Geneviève Bastien authored
This adds the mmap and mmap2 events to the CTF trace obtained from perf data. These events will allow CTF trace visualization tools like Trace Compass to automatically resolve the symbols of the callchain to the corresponding function or origin library. To include those events, one needs to convert with the --all option. Here follows an output of babeltrace: $ sudo perf data convert --all --to-ctf myctftrace $ babeltrace ./myctftrace [19:00:00.000000000] (+0.000000000) perf_mmap2: { cpu_id = 0 }, { pid = 638, tid = 638, start = 0x7F54AE39E000, filename = "/usr/lib/ld-2.25.so" } [19:00:00.000000000] (+0.000000000) perf_mmap2: { cpu_id = 0 }, { pid = 638, tid = 638, start = 0x7F54AE565000, filename = "/usr/lib/libudev.so.1.6.6" } [19:00:00.000000000] (+0.000000000) perf_mmap2: { cpu_id = 0 }, { pid = 638, tid = 638, start = 0x7FFC093EA000, filename = "[vdso]" } Signed-off-by:Geneviève Bastien <gbastien@versatic.net> Acked-by:
Jiri Olsa <jolsa@kernel.org> Tested-by:
Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Francis Deslauriers <francis.deslauriers@efficios.com> Cc: Julien Desfossez <jdesfossez@efficios.com> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/20170727181205.24843-2-gbastien@versatic.netSigned-off-by:
-
Geneviève Bastien authored
The field perf_callchain, if available, is added to the sampling events during the CTF conversion. It is an array of u64 values. The perf_callchain_size field contains the size of the array. It will allow the analysis of sampling data in trace visualization tools like Trace Compass. Possible analyses with those data: dynamic flamegraphs, correlation with other tracing data like a userspace trace. Here follows a babeltrace CTF output of a trace with callchain: $ babeltrace ./myctftrace [17:38:45.672760285] (+?.?????????) cycles:ppp: { cpu_id = 0 }, { perf_ip = 0xFFFFFFFF81063EE4, perf_tid = 25841, perf_pid = 25774, perf_period = 1, perf_callchain_size = 7, perf_callchain = [ [0] = 0xFFFFFFFFFFFFFF80, [1] = 0xFFFFFFFF81063EE4, [2] = 0xFFFFFFFF8100C770, [3] = 0xFFFFFFFF81006EC6, [4] = 0xFFFFFFFF8118245E, [5] = 0xFFFFFFFF810A9224, [6] = 0xFFFFFFFF8164A4C6 ] } [17:38:45.672777672] (+0.000017387) cycles:ppp: { cpu_id = 0 }, { perf_ip = 0xFFFFFFFF81063EE4, perf_tid = 25841, perf_pid = 25774, perf_period = 1, perf_callchain_size = 8, perf_callchain = [ [0] = 0xFFFFFFFFFFFFFF80, [1] = 0xFFFFFFFF81063EE4, [2] = 0xFFFFFFFF8100C770, [3] = 0xFFFFFFFF81006EC6, [4] = 0xFFFFFFFF8118245E, [5] = 0xFFFFFFFF810A9224, [6] = 0xFFFFFFFF8164A4C6, [7] = 0xFFFFFFFF8164ABAD ] } [17:38:45.672786700] (+0.000009028) cycles:ppp: { cpu_id = 0 }, { perf_ip = 0xFFFFFFFF81063EE4, perf_tid = 25841, perf_pid = 25774, perf_period = 70, perf_callchain_size = 3, perf_callchain = [ [0] = 0xFFFFFFFFFFFFFF80, [1] = 0xFFFFFFFF81063EE4, [2] = 0xFFFFFFFF8100C770 ] } Signed-off-by: -
Arnaldo Carvalho de Melo authored
Either when we start 'perf annotate' or 'perf report' with --show-total-period or when we, in the annotate browser, press 't' to toggle period/percent for the first column, we need to adjust the width for the 'period' case. Based-on-a-patch-by:
Taeung Song <treeze.taeung@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-n2np5qcs20u6qjdr9orygne6@git.kernel.orgSigned-off-by:
-
Taeung Song authored
We have the 't' hotkey to toggle showing either the total period or the percentage of samples for a given line, but we forgot to toggle as well the column header, always showing "Percent", even when showing the period, fix it. Signed-off-by:
-
Arnaldo Carvalho de Melo authored
In commit f8f4aaea ("perf annotate: Finally display IPC and cycle accounting") the 'pcnt_width' variable was abused in a few places to also include the optional width of the "IPC" and "cycles" columns, while in other places we stopped using 'pcnt_width' and instead its previous equation... Now that we need to tap into annotate_browser__pcnt_width() to consider if --show-total-period is being used and instead of that hardcoded 7 (strlen("Percent")) we need to use it or strlen("Event count") we need this properly clarified to avoid having to touch all the (7 * nr_events) places. Clarify this by introducing a separate annotate_browser__cycles_width() to leave the pcnt_width calculate just what its name implies. Cc: Taeung Song <treeze.taeung@gmail.com> Cc: Milian Wolff <milian.wolff@kdab.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Andi Kleen <ak@linux.intel.com> Link: http://lkml.kernel.org/n/tip-szgb07t4k5wtvks8nzwkg710@git.kernel.orgSigned-off-by:
-
Taeung Song authored
We were showing the number of samples, not the total period, fix it. Reported-by:
Namhyung Kim <namhyung@kernel.org> Signed-off-by:
-
Arnaldo Carvalho de Melo authored
Just paving the way to fix --show-total-period in the TUI, i.e. now we save in struct disasm_line_samples not just the number of samples, but also the total period. Based-on-a-patch-by:
-
Arnaldo Carvalho de Melo authored
The existing loop incremented the offset while using it as the array index, when we went to an array of sym_hist_entry instances, we should've moved the increment to outside of the array element reference, oops, fix it. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Taeung Song <treeze.taeung@gmail.com> Cc: Wang Nan <wangnan0@huawei.com> Fixes: 461c17f0 ("perf annotate: Store the sample period in each histogram bucket") Link: http://lkml.kernel.org/n/tip-s3dm6uyrazlpag3f0psfia07@git.kernel.orgSigned-off-by:
-
- 26 Jul, 2017 19 commits
-
-
Arnaldo Carvalho de Melo authored
Now that we set the first column header according to wether --show-total-period is being used, we need to size it accordingly. Based-on-a-patch-by:
-
David Carrillo-Cisneros authored
Fixes bug noted by Jiri in https://lkml.org/lkml/2017/6/13/755 and caused by commit d49dadea ("perf tools: Make 'trace' or 'trace_fields' sort key default for tracepoint events") not taking into account that evlist is empty in pipe-mode. Before this commit, pipe mode will only show bogus "100.00% N/A" instead of correct output as follows: $ perf record -o - sleep 1 | perf report -i - # To display the perf.data header info, please use --header/--header-only options. # [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.000 MB - ] # # Total Lost Samples: 0 # # Samples: 8 of event 'cycles:ppH' # Event count (approx.): 145658 # # Overhead Trace output # ........ ............ # 100.00% N/A Correct output, after patch: $ perf record -o - sleep 1 | perf report -i - # To display the perf.data header info, please use --header/--header-only options. # [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.000 MB - ] # # Total Lost Samples: 0 # # Samples: 8 of event 'cycles:ppH' # Event count (approx.): 191331 # # Overhead Command Shared Object Symbol # ........ ....... ................. ................................. # 81.63% sleep libc-2.19.so [.] _exit 13.58% sleep ld-2.19.so [.] do_lookup_x 2.34% sleep [kernel.kallsyms] [k] context_switch 2.34% sleep libc-2.19.so [.] __GI___libc_nanosleep 0.11% perf [kernel.kallsyms] [k] __intel_pmu_enable_a Reported-by:
David Carrillo-Cisneros <davidcc@google.com> Tested-by:
-
Arnaldo Carvalho de Melo authored
When we parse an event we may get a value from the kernel in response to PERF_SAMPLE_WEIGHT being set in perf_event_attr->sample_type, and if it is not set, then perf_sample->weight will be set to zero, which should be ok according to a discussion with Andi Kleen [1]: 1: https://lkml.kernel.org/r/20170724174637.GS3044@two.firstfloor.orgAcked-by:
Andi Kleen <andi@firstfloor.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Milian Wolff <milian.wolff@kdab.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Taeung Song <treeze.taeung@gmail.com> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-8ev8ufk3lzmvgz37yg9nv3qz@git.kernel.orgSigned-off-by:
-
Jiri Olsa authored
Make perf stat use group read if there are groups defined. The group read will get the values for all member of groups within a single syscall instead of calling read syscall for every event. We can see considerable less amount of kernel cycles spent on single group read, than reading each event separately, like for following perf stat command: # perf stat -e {cycles,instructions} -I 10 -a sleep 1 Monitored with "perf stat -r 5 -e '{cycles:u,cycles:k}'" Before: 24,325,676 cycles:u 297,040,775 cycles:k 1.038554134 seconds time elapsed After: 25,034,418 cycles:u 158,256,395 cycles:k 1.036864497 seconds time elapsed The perf_evsel__open fallback changes contributed by Andi Kleen. Signed-off-by: -
Jiri Olsa authored
Add perf_evsel__read_counter() to read single or group counter. After calling this function the counter's evsel::counts struct is filled with values for the counter and member of its group if there are any. Signed-off-by:
-
Jiri Olsa authored
Currently we use the size of struct perf_counts_values to read the event, which prevents us to put any new member to the struct. Adding perf_evsel__read_size to return size of the buffer needed for event read. Signed-off-by:
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo-4.14-20170725' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core Pull perf/core improvemends and fixes for v4.14: New features: - Filter out 'sshd' in the tracer ancestry in 'perf trace' syswide tracing, to elliminate tracing loops (Arnaldo Carvalho de Melo) - Support lookup of symbols in other mount namespaces in 'perf top' (Krister Johansen) - Initial 'clone' syscall args beautifier in 'perf trace' (Arnaldo Carvalho de Melo) User visible changes: - Ignore 'fd' and 'offset' args for MAP_ANONYMOUS in 'perf trace' (Arnaldo Carvalho de Melo) - Process tracing data in 'perf annotate' pipe mode (David Carrillo-Cisneros) - Make 'perf report --branch-history' work without callgraphs(-g) option in perf record (Jin Yao) - Tag branch type/flag on "to" and tag cycles on "from" in 'perf report' (Jin Yao) Fixes: - Fix jvmti linker error when libelf config is disabled (Sudeep Holla) - Fix cgroups refcount usage (Arnaldo Carvalho de Melo) - Fix kernel symbol adjustment for s390x (Thomas Richter) - Fix 'perf report --stdio --show-total-period', it was showing the number of samples, not the total period (Taeung Song) Infrastructure changes: - Add perf_sample dictionary to tracepoint handlers in 'perf script' python, which were already present for other types of events (hardware, etc) (Arun Kalyanasundaram) - Make build fail on vendor events JSON parse error (Andi Kleen) - Adopt strstarts() from the kernel (Arnaldo Carvalho de Melo) Arch specific changes: - Set no_aux_samples for the tracking event in Intel PT (Kan Liang) - Always set no branch for Intel PT dummy event (Kan Liang) Trivial changes: - Simplify some error handlers in 'perf script' (Dan Carpenter) - Add EXCLUDE_EXTLIBS and EXTRA_PERFLIBS to makefile (David Carrillo-Cisneros) Signed-off-by:
Ingo Molnar <mingo@kernel.org>
-
Arnaldo Carvalho de Melo authored
This file was copied from the kernel so that we could build tools/perf/ on older systems where some newer defines, such as these are available: CC trace/beauty/fcntl.o trace/beauty/fcntl.c: In function ‘syscall_arg__scnprintf_fcntl_arg’: trace/beauty/fcntl.c:93:13: error: ‘F_OFD_SETLK’ undeclared (first use in this function) cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW || cmd == F_OFD_GETLK || ^ trace/beauty/fcntl.c:93:13: note: each undeclared identifier is reported only once for each function it appears in trace/beauty/fcntl.c:93:35: error: ‘F_OFD_SETLKW’ undeclared (first use in this function) cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW || cmd == F_OFD_GETLK || ^ trace/beauty/fcntl.c:93:58: error: ‘F_OFD_GETLK’ undeclared (first use in this function) cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW || cmd == F_OFD_GETLK || ^ mv: cannot stat ‘trace/beauty/.fcntl.o.tmp’: No such file or directory make[4]: *** [trace/beauty/fcntl.o] Error 1 make[3]: *** [trace/beauty] Error 2 make[3]: *** Waiting for unfinished jobs.... CC tests/llvm.o But we need to make sure that it is also in the tools/perf/MANIFEST file, that is used to build a tarball for detached (from the kernel sources) compilation, which was failing, with the above message, on a RHEL7.4 system, fix it. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Fixes: 84d1d8a1 ("tools include uapi asm-generic: Grab a copy of fcntl.h") Link: http://lkml.kernel.org/n/tip-2d5px7aq5stbwi24pgirwtlm@git.kernel.orgSigned-off-by: -
Taeung Song authored
Currently the first column header is always "Percent", fix it to show correct column name based on given options, i.e. if using --show-total-period, show "Event count" as a first column. Reported-by:
Milian Wolff <milian.wolff@kdab.com> Signed-off-by:
-
Andi Kleen authored
Today, when a JSON file fails parsing the build continues, but there are no json files built in, which is difficult to debug later. Make the build stop on a parse error instead. v2: Add fixes from Sukadev. Now we handle architectures with no JSON events correctly. And fix some stale comments. Committer note: Tested by running the cross build container tests, that were all failing for v1. Signed-off-by:Andi Kleen <ak@linux.intel.com> Acked-by:
-
Jin Yao authored
Current --branch-history LBR annotation displays confused data. For example, each cycles report is duplicated on both "from" and "to" entries. For example: perf report --branch-history --no-children --stdio --2.32%--main div.c:39 (COND_BWD CROSS_2M predicted:49.7% cycles:1) main div.c:44 (predicted:49.7% cycles:1) main div.c:42 (RET CROSS_2M cycles:2) compute_flag div.c:28 (cycles:2) compute_flag div.c:27 (RET CROSS_2M cycles:1) rand rand.c:28 (cycles:1) rand rand.c:28 (RET CROSS_2M cycles:1) __random random.c:298 (cycles:1) __random random.c:297 (COND_BWD CROSS_2M cycles:1) __random random.c:295 (cycles:1) __random random.c:295 (COND_BWD CROSS_2M cycles:1) __random random.c:295 (cycles:1) __random random.c:295 (RET CROSS_2M cycles:9) The cycles should be tagged only on the "from". It's for the code block that ends with "from", not for "to". Another issue is the "predicted:49.7%" is duplicated too (tag on both "from" and "to"). This patch tags the branch type/flag on "to" and tag the cycles on "from". For example: --2.32%--main div.c:39 (COND_BWD CROSS_2M predicted:49.7%) main div.c:44 (cycles:1) main div.c:42 (RET CROSS_2M) compute_flag div.c:28 (cycles:2) compute_flag div.c:27 (RET CROSS_2M) rand rand.c:28 (cycles:1) rand rand.c:28 (RET CROSS_2M) __random random.c:298 (cycles:1) __random random.c:297 (COND_BWD CROSS_2M) __random random.c:295 (cycles:1) __random random.c:295 (COND_BWD CROSS_2M) __random random.c:295 (cycles:1) __random random.c:295 (RET CROSS_2M) | --2.23%--__random_r random_r.c:392 (cycles:9) In this example, The "main div.c:39 (COND_BWD CROSS_2M predicted:49.7%)" is "to" of branch and "main div.c:44 (cycles:1)" is "from" of branch. It should be easier for understanding than before. Signed-off-by:Yao Jin <yao.jin@linux.intel.com> Reviewed-by:
-
Jin Yao authored
perf record -b -g <command> perf report --branch-history This merges the LBRs with the callgraphs. However it would be nice if it also works without callgraphs (-g) set in perf record, so that only the LBRs are displayed. But currently perf report errors in this case. For example, perf record -b <command> perf report --branch-history Error: Selected -g or --branch-history but no callchain data. Did you call 'perf record' without -g? This patch displays the LBRs only even if callgraphs(-g) is not enabled in perf record. Change log: v2: According to Milian Wolff's comment, change the obsolete error message. Now the error message is: ┌─Error:─────────────────────────────────────┐ │Selected -g or --branch-history. │ │But no callchain or branch data. │ │Did you call 'perf record' without -g or -b?│ │ │ │ │ │Press any key... │ └────────────────────────────────────────────┘ When passing the last parameter to hists__fprintf, changes "|" to "||". hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, symbol_conf.use_callchain || symbol_conf.show_branchflag_count); Signed-off-by: -
Arun Kalyanasundaram authored
Modify the signature of tracepoint specific and trace_unhandled hooks to add the perf_sample dict as a new argument. Create a python helper function to print a dictionary. Signed-off-by:
Arun Kalyanasundaram <arunkaly@google.com> Acked-by:
-
Arun Kalyanasundaram authored
The process_event python hook receives a dict with all perf_sample entries, but the tracepoint specific and trace_unhandled hooks predate the introduction of this dict, and do not receive it. Add the aforementioned dict as an additional argument to the affected handlers. To keep backwards compatibility (and avoid unnecessary work), do not pass the dict if the number of arguments signals that handler version predates this change. Signed-off-by:
-
Arun Kalyanasundaram authored
Provide time_enabled, time_running and counter value in the perf_sample dict. Signed-off-by:
-
Arun Kalyanasundaram authored
Move the creation of the dict containing perf_sample entries into a helper function to enable its reuse in other sample processing routines. Signed-off-by:
-
Arun Kalyanasundaram authored
Avoid allocating memory if hook handler is not available. This saves unused memory allocation and simplifies error path. Let handler in python_process_tracepoint point to either tracepoint specific or trace_unhandled hook. Use dict to check if handler points to trace_unhandled. Remove the exit label in python_process_general_event and return when no handler is available. Signed-off-by:
-
Dan Carpenter authored
If script_desc__new() fails then the current code has a NULL dereference. We don't actually need to do any cleanup, we can just return NULL. Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: kernel-janitors@vger.kernel.org Link: http://lkml.kernel.org/r/20170722073610.nnsyiwdcfl6bhn4t@mwandaSigned-off-by:
-
Krister Johansen authored
The perf top command needs to unshare its fs from the helper threads in order to successfully setns(2) during its symbol lookup. It also needs to impelement a force flag to ignore ownership of perf-<pid>.map files. Signed-off-by:
Krister Johansen <kjlx@templeofstupid.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Brendan Gregg <brendan.d.gregg@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Link: http://lkml.kernel.org/r/1499305693-1599-6-git-send-email-kjlx@templeofstupid.comSigned-off-by:
-
- 25 Jul, 2017 7 commits
-
-
Jiri Olsa authored
Adding info about what is being switched off in the sys_perf_event_open fallback. New output (notice the 'switching off' lines): $ perf stat -e '{cycles,instructions}' -vvv ls Using CPUID GenuineIntel-6-3D intel_pt default config: tsc ------------------------------------------------------------ perf_event_attr: size 112 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP disabled 1 inherit 1 enable_on_exec 1 exclude_guest 1 ------------------------------------------------------------ sys_perf_event_open: pid 3591 cpu -1 group_fd -1 flags 0x8 sys_perf_event_open failed, error -22 switching off cloexec flag ------------------------------------------------------------ perf_event_attr: size 112 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP disabled 1 inherit 1 enable_on_exec 1 exclude_guest 1 ------------------------------------------------------------ sys_perf_event_open: pid 3591 cpu -1 group_fd -1 flags 0 sys_perf_event_open failed, error -22 switching off exclude_guest, exclude_host ------------------------------------------------------------ perf_event_attr: size 112 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP disabled 1 inherit 1 enable_on_exec 1 ------------------------------------------------------------ sys_perf_event_open: pid 3591 cpu -1 group_fd -1 flags 0 sys_perf_event_open failed, error -22 switching off sample_id_all ------------------------------------------------------------ perf_event_attr: size 112 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP disabled 1 inherit 1 enable_on_exec 1 ... Signed-off-by: -
Sudeep Holla authored
When libelf is disabled in the configuration, we get the following linker error: LINK libperf-jvmti.so ld: cannot find -lelf Makefile.perf:515: recipe for target 'libperf-jvmti.so' failed Jiri pointed out that both librt and libelf are not really required. So this patch fixes the linker error by getting rid of unwanted libraries in the linker stage. Signed-off-by:
Sudeep Holla <sudeep.holla@arm.com> Acked-by:
-
David Carrillo-Cisneros authored
'perf annotate' was missing the handler for tracing data records. Prior to this patch we obtained "unhandled" records when piping trace events to perf annotate (using -D option to show the dump_printf messages in process_event_synth_tracing_data_stub): $ perf record -o - -e block:bio_free sleep 2 | perf annotate -D --stdio ... 0x78 [0xc]: PERF_RECORD_TRACING_DATA: unhandled! ... Signed-off-by:
-
David Carrillo-Cisneros authored
The goal is to allow users to override linking of libraries that were automatically added to PERFLIBS. EXCLUDE_EXTLIBS contains linker flags to be removed from LIBS while EXTRA_PERFLIBS contains linker flags to be added. My use case is to force certain library to be build statically, e.g. for libelf: EXCLUDE_EXTLIBS=-lelf EXTRA_PERFLIBS=path/libelf.a Signed-off-by:
-
Arnaldo Carvalho de Melo authored
When converting from atomic_t to refcount_t we didn't follow the usual step of initializing it to one before taking any new reference, which trips over checking if taking a reference for a freed refcount_t, fix it. Brendan's report: --- It's 4.12-rc7, with node v4.4.1. I'm building 4.13-rc1 now, as I hit what I think is another unrelated perf bug and I'm starting to wonder what else is broken on that version: (root) /mnt/src/linux-4.12-rc7/tools/perf # ./perf record -F 99 -a -e cpu-clock --cgroup=docker/f9e9d5df065b14646e8a11edc837a13877fd90c171137b2ba3feb67a0201cb65 -g perf: /mnt/src/linux-4.12-rc7/tools/include/linux/refcount.h:108: refcount_inc: Assertion `!(!refcount_inc_not_zero(r))' failed. Aborted that used to work... --- Testing it: Before: # perf stat -e cycles -C 0 --cgroup / perf: /home/acme/git/linux/tools/include/linux/refcount.h:108: refcount_inc: Assertion `!(!refcount_inc_not_zero(r))' failed. Aborted (core dumped) # After: # perf stat -e cycles -C 0 --cgroup / ^C Performance counter stats for 'CPU(s) 0': 132,081,393 cycles / 2.492942763 seconds time elapsed # Reported-by:Brendan Gregg <brendan.d.gregg@gmail.com> Acked-by:
Elena Reshetova <elena.reshetova@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: David Carrillo-Cisneros <davidcc@google.com> Cc: Kees Kook <keescook@chromium.org> Cc: Krister Johansen <kjlx@templeofstupid.com> Cc: Paul Turner <pjt@google.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Sudeep Holla <Sudeep.Holla@arm.com> Cc: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Cc: Wang Nan <wangnan0@huawei.com> Fixes: 79c5fe6d ("perf cgroup: Convert cgroup_sel.refcnt from atomic_t to refcount_t") Link: http://lkml.kernel.org/n/tip-l7ovfblq14ip2i08m1g0fkhv@git.kernel.orgSigned-off-by:
-
Thomas Richter authored
On s390x the kernel text segment starts at address 0x0. When perf report reads kernel symbols from vmlinux file it adds an offset of 0x1000. For example see symbol set_reset_devices: [root@s8360047 linux-devel]# nm -A vmlinux| fgrep set_reset_devices vmlinux:0000000001379000 t set_reset_devices [root@s8360047 linux-devel]# [root@s8360047 linux-devel]# fgrep set_reset_devices /proc/kallsyms 0000000001379000 t set_reset_devices [root@s8360047 linux-devel]# The kernel symbol table and the vmlinux file have the same address for symbol set_reset_devices namely 1379000. When perf report reads this symbols it displays it with address symbol__new: set_reset_devices 0x137a000-0x137a018 There is a difference between perf report and vmlinux of 0x1000. The reason for the difference is at kernel symbol load time in function dso__load_sym(). The vmlinux file is investigated with its ELF header. Command readelf shows this: Section Headers: [Nr] Name Type Address Offset Size EntSize Flags Link Info Align [ 0] NULL 0000000000000000 00000000 0000000000000000 0000000000000000 0 0 0 [ 1] .text PROGBITS 0000000000000000 00001000 0000000000b0e0c2 0000000000000000 AX 0 0 128 This leads to an invalid calculation of the symbol start address, see file utit/symbol-elf.c line 974: /* Adjust symbol to map to file offset */ if (adjust_kernel_syms) sym.st_value -= shdr.sh_addr - shdr.sh_offset; With shdr.sh_addr set to 0x0 and shdr.sh_offset set to 0x1000 as read from the ELF .text section 0x1000 is added to the symbol address. I would like to fix this by introducing an archticture specific function named elf__needs_adjust_symbols(). This is the same approach as done by PowerPC. The function currently does not exist for s390x and the default weak one is used. The s390x specific one returns false when symsrc_init() is invoked for kernel symbols and results in variable adjust_kernel_syms being false. This omits the adjustment and the correct address is displayed (when symbol resolvement does not work). The s390x specific function returns false for kernel symbol adjustment and returns true for kernel modules, processes and shared libraries. Signed-off-by:Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Cc: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> LPU-Reference: 20170713130252.6167-1-tmricht@linux.vnet.ibm.com Signed-off-by:
-
Taeung Song authored
We were showing the total number of samples, not the total period as asked by the user, fix it. Reported-by:
-
- 21 Jul, 2017 5 commits
-
-
Taeung Song authored
In fixing the --show-total-period option it was noticed that the value of sample->period was being overwritten, fix it. Signed-off-by:
-
Taeung Song authored
We'll use it soon, when fixing --show-total-period. Signed-off-by:
-
Taeung Song authored
To pave the way to use perf_sample fields in the annotate code, storing sample->period in sym_hist->addr->period and its sum in sym_hist->period. Signed-off-by:
-
Taeung Song authored
To make it more clear that it is the sum of all the nr_samples fields in the addr[] entries, i.e.: sym_hist->nr_samples = sum(sym_hist->addr[0 .. symbol__size(sym)]->nr_samples) Committer notes: Taeung had renamed it to total_samples, but using nr_samples, as in the added explanation above, looks clearer and establishes the direct connection, making clear it is about the _number_ of samples. Signed-off-by:
-
Taeung Song authored
struct sym_hist has addr[] but it should have not only number of samples but also the sample period. So use new struct symhist_entry to pave the way to have that. Committer notes: This initial patch will only introduce the struct sym_hist_entry and use only the nr_samples member, which makes the code clearer and paves the way to save the period as well. Signed-off-by:
-
- 20 Jul, 2017 1 commit
-
-
Arnaldo Carvalho de Melo authored
Getting support for "on", "off" introduced in a81a5a17 ("lib: add "on"/"off" support to kstrtobool") and making it check for NULL, introduced in ef951599 ("lib: move strtobool() to kstrtobool()"). Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Cc: Kees Cook <keescook@chromium.org> Link: http://lkml.kernel.org/n/tip-mu8ghin4rklacmmubzwv8td7@git.kernel.orgSigned-off-by:
-
