Commits · b20fe10642f90d17c07b5e621bf4c00093c5654b · Kirill Smelkov / linux

20 Feb, 2019 4 commits

perf bpf-event: Add missing new line into pr_debug call · b20fe106

Jiri Olsa authored Feb 20, 2019

Add a missing new line into pr_debug call in perf_event__synthesize_bpf_events(),
so that the error message does not screw the verbose output.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Link: http://lkml.kernel.org/r/20190220122800.864-5-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

b20fe106

perf script: Allow +- operator for type specific fields option · 6ef362fd

Jiri Olsa authored Feb 20, 2019

Add support to add/remove fields for specific event types in -F option.
It's now possible to use '+-' after event type, like:

  # cat > test.c
  #include <stdio.h>

  int main(void)
  {
     printf("Hello world\n");
     while(1) {}
  }
  ^D
  # gcc -g -o test test.c
  # perf probe -x test 'test.c:5'
  # perf record -e '{cpu/cpu-cycles,period=10000/,probe_test:main}:S' ./test
  ...

  # perf script -Ftrace:+period,-cpu
            test  3859 396291.117343:      10275 cpu/cpu-cycles,period=10000/:      7f..
            test  3859 396291.118234:      11041 cpu/cpu-cycles,period=10000/:  ffffff..
            test  3859 396291.118234:          1              probe_test:main:
            test  3859 396291.118248:       8668 cpu/cpu-cycles,period=10000/:  ffffff..
            test  3859 396291.118263:      10139 cpu/cpu-cycles,period=10000/:  ffffff..

Committer testing:

Couldn't make the test above work, but tested it with:

  # perf probe -x hello main
  Added new event:
    probe_hello:main     (on main in /home/acme/c/hello)

  You can now use it in all perf tools, such as:

	  perf record -e probe_hello:main -aR sleep 1

  # perf record -e probe_hello:main ./hello
  hello, world
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.025 MB perf.data (1 samples) ]
  # perf script
           hello 21454 [002] 254116.874005: probe_hello:main: (401126)
  #
  # perf script -Ftrace:+period,-cpu
           hello 21454 254116.874005:          1 probe_hello:main: (401126)
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190220122800.864-4-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

6ef362fd

perf evsel: Force sample_type for slave events · 6e7e8b9f

Jiri Olsa authored Feb 20, 2019

Force sample_type setup for slave events in group leader sessions.

We don't get sample for slave events, we make them when delivering group
leader sample. Set the slave event to follow the master sample_type to
ease up report.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190220122800.864-3-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

6e7e8b9f

perf session: Don't report zero period samples for slave events · 529c1a9e

Jiri Olsa authored Feb 20, 2019

There's no reason to deliver a sample with zero period.  It means there
was no value for slave event since its last group leader sample.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190220122800.864-2-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

529c1a9e

19 Feb, 2019 10 commits

perf trace: Allow dumping a BPF map after setting up BPF events · ff7a4f98

Arnaldo Carvalho de Melo authored Feb 19, 2019

Initial use case:

Dumping the maps setup by tools/perf/examples/bpf/augmented_raw_syscalls.c,
which so far are just booleans, showing just non-zeroed entries:

  # cat ~/.perfconfig
  [llvm]
	dump-obj = true
	clang-opt = -g
  [trace]
	#add_events = /home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.o
	add_events = /wb/augmented_raw_syscalls.o
  $ date
  Tue Feb 19 16:29:33 -03 2019
  $ ls -la /wb/augmented_raw_syscalls.o
  -rwxr-xr-x. 1 root root 14048 Jan 24 12:09 /wb/augmented_raw_syscalls.o
  $ file /wb/augmented_raw_syscalls.o
  /wb/augmented_raw_syscalls.o: ELF 64-bit LSB relocatable, eBPF, version 1 (SYSV), with debug_info, not stripped
  $
  # trace -e recvmmsg,sendmmsg --map-dump foobar
  ERROR: BPF map "foobar" not found
  # trace -e recvmmsg,sendmmsg --map-dump filtered_pids
  ERROR: BPF map "filtered_pids" not found
  # trace -e recvmmsg,sendmmsg --map-dump pids_filtered
  [2583] = 1,
  [2267] = 1,
  ^Z
  [1]+  Stopped                 trace -e recvmmsg,sendmmsg --map-dump pids_filtered
  # pidof trace
  2267
  # ps ax|grep gnome-terminal|grep -v grep
  2583 ?        Ssl   58:33 /usr/libexec/gnome-terminal-server
  ^C
  # trace -e recvmmsg,sendmmsg --map-dump syscalls
  [299] = 1,
  [307] = 1,
  ^C
  # grep x64_recvmmsg arch/x86/entry/syscalls/syscall_64.tbl
  299	64	recvmmsg		__x64_sys_recvmmsg
  # grep x64_sendmmsg arch/x86/entry/syscalls/syscall_64.tbl
  307	64	sendmmsg		__x64_sys_sendmmsg
  #

Next step probably will be something like 'perf stat's --interval-print and
--interval-clear.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Yonghong Song <yhs@fb.com>
Link: https://lkml.kernel.org/n/tip-ztxj25rtx37ixo9cfajt8ocy@git.kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

ff7a4f98

perf bpf: Add bpf_map dumper · d19f8564

Arnaldo Carvalho de Melo authored Feb 19, 2019

At some point I'll suggest moving this to libbpf, for now I'll
experiment with ways to dump BPF maps set by events in 'perf trace',
starting with a very basic dumper for the current very limited needs
of the augmented_raw_syscalls code: dumping booleans.

Having functions that apply to the map keys and values and do table
lookup in things like syscall id to string tables should come next.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Yonghong Song <yhs@fb.com>
Link: https://lkml.kernel.org/n/tip-lz14w0esqyt1333aon05jpwc@git.kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

d19f8564

perf test: Fix failure of 'evsel-tp-sched' test on s390 · 03d30971

Thomas Richter authored Feb 19, 2019

Commit 489338a7 ("perf tests evsel-tp-sched: Fix bitwise operator")
causes test case 14 "Parse sched tracepoints fields" to fail on s390.

This test succeeds on x86.

In fact this test now fails on all architectures with type char treated
as type unsigned char.

The root cause is the signed-ness of character arrays in the tracepoints
sched_switch for structure members prev_comm and next_comm.

On s390 the output of:

 [root@m35lp76 perf]# cat /sys/kernel/debug/tracing/events/sched/sched_switch/format
 name: sched_switch
 ID: 287
 format:
   field:unsigned short common_type; offset:0; size:2;	signed:0;
   ...
   field:char prev_comm[16]; offset:8; size:16;	signed:0;
   ...
   field:char next_comm[16]; offset:40; size:16; signed:0;

reveals the character arrays prev_comm and next_comm are per
default unsigned char and have values in the range of 0..255.

On x86 both fields are signed as this output shows:
 [root@f29]# cat /sys/kernel/debug/tracing/events/sched/sched_switch/format
 name: sched_switch
 ID: 287
 format:
   field:unsigned short common_type; offset:0; size:2;	signed:0;
   ...
   field:char prev_comm[16]; offset:8; size:16;	signed:1;
   ...
   field:char next_comm[16]; offset:40; size:16; signed:1;

and the character arrays prev_comm and next_comm are per default signed
char and have values in the range of -1..127.  The implementation of
type char is architecture specific.

Since the character arrays in both tracepoints sched_switch and
sched_wakeup should contain ascii characters, simply omit the check for
signedness in the test case.

Output before:

  [root@m35lp76 perf]# ./perf test -F 14
  14: Parse sched tracepoints fields                        :
  --- start ---
  sched:sched_switch: "prev_comm" signedness(0) is wrong, should be 1
  sched:sched_switch: "next_comm" signedness(0) is wrong, should be 1
  sched:sched_wakeup: "comm" signedness(0) is wrong, should be 1
  ---- end ----
  14: Parse sched tracepoints fields                        : FAILED!
  [root@m35lp76 perf]#

Output after:

  [root@m35lp76 perf]# ./perf test -Fv 14
  14: Parse sched tracepoints fields                        :
  --- start ---
  ---- end ----
  Parse sched tracepoints fields: Ok
  [root@m35lp76 perf]#

Fixes: 489338a7 ("perf tests evsel-tp-sched: Fix bitwise operator")
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20190219153639.31267-1-tmricht@linux.ibm.comSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

03d30971

perf doc: Fix documentation of the Flags section in perf.data · 8c23a522

Jonas Rabenstein authored Feb 19, 2019

According to the current documentation the flags section is placed after
the file header itself but the code assumes to find the flags section
after the data section. This change updates the documentation to that
assumption.
Signed-off-by: Jonas Rabenstein <jonas.rabenstein@studium.uni-erlangen.de>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Link: http://lkml.kernel.org/r/20190219154515.3954-2-jonas.rabenstein@studium.uni-erlangen.deSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

8c23a522

perf doc: Fix HEADER_CMDLINE description in perf.data documentation · 7a663c0f

Jonas Rabenstein authored Feb 19, 2019

The content of the HEADER_CMDLINE feature header is a perf_header_string_list
of the argument vector and not a perf_header_string of the commandline.
Signed-off-by: Jonas Rabenstein <jonas.rabenstein@studium.uni-erlangen.de>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Link: http://lkml.kernel.org/r/20190219154515.3954-1-jonas.rabenstein@studium.uni-erlangen.deSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

7a663c0f

perf report: Don't shadow inlined symbol with different addr range · 7346195e

He Kuang authored Feb 19, 2019

We can't assume inlined symbols with the same name are equal, because
their address range may be different. This will cause the symbols with
different addresses be shadowed when adding to the hist entry, and lead
to ERANGE error when checking the symbol address during sample parse,
the addr should be within the range of [sym.start, sym.end].

The error message is like: "0x36aea60f [0x8]: failed to process type: 68".

The second parameter of symbol__new() is the length of the fake symbol
for the inline frame, which is the subtraction of the end and start
address of base_sym.
Signed-off-by: He Kuang <hekuang@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: aa441895 ("perf report: Compare symbol name for inlined frames when sorting")
Link: http://lkml.kernel.org/r/20190219130531.15692-1-hekuang@huawei.comSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

7346195e

perf tools: Use sysfs__mountpoint() when reading cpu topology · e19a01c1

Jiri Olsa authored Feb 19, 2019

Use sysfs__mountpoint() when reading sysfs files to obtain cpu/numa
topologies.

Also use scnprintf instead of sprintf as suggested by Namhyung.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190219095815.15931-5-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

e19a01c1

perf tools: Add numa_topology object · 48e6c5ac

Jiri Olsa authored Feb 19, 2019

Add the numa_topology object to return the list of numa nodes together
with their cpus. It will replace the numa code in header.c and will be
used from 'perf record' in the following patches.

Add the following interface functions to load numa details:

  struct numa_topology *numa_topology__new(void);
  void numa_topology__delete(struct numa_topology *tp);

And replace the current (copied) local interface, with no functional
changes.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190219095815.15931-4-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

48e6c5ac

perf tools: Add cpu_topology object · 5135d5ef

Jiri Olsa authored Feb 19, 2019

Make struct cpu_topo global and rename it to 'struct cpu_topology', so
that it can be used from the 'perf record' command in the following
patches.

Add the following interface functions to load/free cpu topology details:

  struct cpu_topology *cpu_topology__new(void);
  void cpu_topology__delete(struct cpu_topology *tp);

Move it to a separate source file cputopo.c together with numa related
object in the following patches.

No functional change, the new interface will be used in upcoming changes.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190219095815.15931-3-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

5135d5ef

perf header: Fix wrong node write in NUMA_TOPOLOGY feature · b00ccb27

Jiri Olsa authored Feb 19, 2019

We are currently passing the node index instead of the real node number.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: fbe96f29 ("perf tools: Make perf.data more self-descriptive (v8)"
Link: http://lkml.kernel.org/r/20190219095815.15931-2-jolsa@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

b00ccb27

15 Feb, 2019 2 commits

perf tests shell: Skip trace+probe_vfs_getname.sh if built without trace support · 83244772

Tommi Rantala authored Feb 15, 2019

If perf was built without trace support, the trace+probe_vfs_getname.sh
'perf test' entry fails:

  # perf trace -h
  perf: 'trace' is not a perf-command. See 'perf --help'

  # perf test 64
  64: Check open filename arg using perf trace + vfs_getname: FAILED!

Check trace support, so that we'll skip the test in that case:

  # perf test 64
  64: Check open filename arg using perf trace + vfs_getname: Skip
Signed-off-by: Tommi Rantala <tommi.t.rantala@nokia.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Hendrik Brueckner <brueckner@linux.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kim Phillips <kim.phillips@arm.com>
Cc: Michael Petlan <mpetlan@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190215134253.11454-1-tt.rantala@gmail.comSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

83244772

Merge tag 'perf-core-for-mingo-5.1-20190214' of... · 43f4e627

Ingo Molnar authored Feb 15, 2019

Merge tag 'perf-core-for-mingo-5.1-20190214' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

perf list:

Jiri Olsa:

- Display metric expressions for --details option

perf record:

Alexey Budankov:

- Implement --affinity=node|cpu option, leftover, the other patches
in this kit were already applied.

perf trace:

Arnaldo Carvalho de Melo:

- Fix segfaults due to not properly handling negative file descriptor syscall args.

- Fix segfault related to the 'waitid' 'options' prefix showing logic.

- Filter out 'gnome-terminal*' if it is a parent of 'perf trace', to reduce the
syscall feedback loop in system wide sessions.

BPF:

Song Liu:

- Silence "Couldn't synthesize bpf events" warning for EPERM.

Build system:

Arnaldo Carvalho de Melo:

- Fix the test-all.c feature detection fast path that was broken for
quite a while leading to longer build times.

Event parsing:

Jiri Olsa:

- Fix legacy events symbol separator parsing

cs-etm:

Mathieu Poirier:

- Fix some error path return errors and plug some memory leaks.

- Add proper header file for symbols

- Remove unused structure fields.

- Modularize auxtrace_buffer fetch, decoder and packet processing loop.

Vendor events:

Paul Clarke:

- Add assorted metrics for the Power8 and Power9 architectures.

perf report:

Thomas Richter:

- Add s390 diagnostic sampling descriptor size
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

43f4e627

14 Feb, 2019 24 commits

tools build feature sched_getcpu: Undef _GNU_SOURCE at the end · 44ec8396

Arnaldo Carvalho de Melo authored Feb 14, 2019

Since this feature test is included in test-all.c, the feature detection
fast path compile/link phase, it can't leave any defines behind, as it
can affect the tests included after it, so remove it.
Reported-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jakub Kicinski <jakub.kicinski@netronome.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Song Liu <songliubraving@fb.com>
Cc: Yonghong Song <yhs@fb.com>
Link: https://lkml.kernel.org/n/tip-lg3kpd9tzypc797vb1f42u6k@git.kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

44ec8396

perf header: Remove unused 'cpu_nr' field from 'struct cpu_topo' · aa4df30d