Peter suggested that using the exclusive mode in perf could avoid some
problems with bad scheduling of groups. Exclusive is implemented in the
kernel, but wasn't exposed by the perf tool, so hard to use without
custom low level API users.
Add support for marking groups or events with :e for exclusive in the
perf tool. The implementation is basically the same as the existing
pinned attribute.
Committer testing:
# perf test "parse event"
6: Parse event definition strings : Ok
# perf test -v "parse event" |& grep :u*e
running test 56 'instructions:uep'
running test 57 '{cycles,cache-misses,branch-misses}:e'
#
#
# grep "model name" -m1 /proc/cpuinfo
model name : AMD Ryzen 9 3900X 12-Core Processor
#
# perf stat -a -e '{cycles,cache-misses,branch-misses}:e' sleep 1
Performance counter stats for 'system wide':
<not counted> cycles (0.00%)
<not counted> cache-misses (0.00%)
<not counted> branch-misses (0.00%)
1.001269893 seconds time elapsed
Some events weren't counted. Try disabling the NMI watchdog:
echo 0 > /proc/sys/kernel/nmi_watchdog
perf stat ...
echo 1 > /proc/sys/kernel/nmi_watchdog
# echo 0 > /proc/sys/kernel/nmi_watchdog
# perf stat -a -e '{cycles,cache-misses,branch-misses}:e' sleep 1
Performance counter stats for 'system wide':
1,298,663,141 cycles
30,962,215 cache-misses
5,325,150 branch-misses
1.001474934 seconds time elapsed
#
# The output for asking for precise events on AMD needs to improve, it
# supposedly works only for system wide or per CPU
#
# perf stat -a -e '{cycles,cache-misses,branch-misses}:uep' sleep 1
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (cycles).
/bin/dmesg | grep -i perf may provide additional information.
# perf stat -a -e '{cycles,cache-misses,branch-misses}:ue' sleep 1
Performance counter stats for 'system wide':
746,363,126 cycles
16,881,611 cache-misses
2,871,259 branch-misses
1.001636066 seconds time elapsed
#
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20201014144255.22699-1-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The --for-each-cgroup option is a syntax sugar to monitor large number
of cgroups easily. Current command line requires to list all the events
and cgroups even if users want to monitor same events for each cgroup.
This patch addresses that usage by copying given events for each cgroup
on user's behalf.
For instance, if they want to monitor 6 events for 200 cgroups each they
should write 1200 event names (with -e) AND 1200 cgroup names (with -G)
on the command line. But with this change, they can just specify 6
events and 200 cgroups with a new option.
A simpler example below: It wants to measure 3 events for 2 cgroups ('A'
and 'B'). The result is that total 6 events are counted like below.
$ perf stat -a -e cpu-clock,cycles,instructions --for-each-cgroup A,B sleep 1
Performance counter stats for 'system wide':
988.18 msec cpu-clock A # 0.987 CPUs utilized
3,153,761,702 cycles A # 3.200 GHz (100.00%)
8,067,769,847 instructions A # 2.57 insn per cycle (100.00%)
982.71 msec cpu-clock B # 0.982 CPUs utilized
3,136,093,298 cycles B # 3.182 GHz (99.99%)
8,109,619,327 instructions B # 2.58 insn per cycle (99.99%)
1.001228054 seconds time elapsed
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20200924124455.336326-3-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add 'snapshot' control command to create an AUX area tracing snapshot
the same as if sending SIGUSR2. The advantage of the FIFO is that access
is governed by access to the FIFO.
Example:
$ mkfifo perf.control
$ mkfifo perf.ack
$ cat perf.ack &
[1] 15235
$ sudo ~/bin/perf record --control fifo:perf.control,perf.ack -S -e intel_pt//u -- sleep 60 &
[2] 15243
$ ps -e | grep perf
15244 pts/1 00:00:00 perf
$ kill -USR2 15244
bash: kill: (15244) - Operation not permitted
$ echo snapshot > perf.control
ack
$
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Acked-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lore.kernel.org/lkml/20200901093758.32293-6-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This adds an option '-m/--buffer-size' to allow us set the size of per-cpu
tracing buffer.
Committer testing:
Before running with this option:
# find /sys/kernel/tracing/ -name buffer_size_kb | xargs cat
1408
1408
1408
1408
1408
1408
1408
1408
1408
#
Then, run:
# perf ftrace -m 2048K | head -10
2) | mutex_unlock() {
2) ==========> |
2) | smp_irq_work_interrupt() {
2) | irq_enter() {
2) 0.121 us | rcu_irq_enter();
2) 0.128 us | irqtime_account_irq();
2) 0.719 us | }
2) | __wake_up() {
2) | __wake_up_common_lock() {
2) 0.105 us | _raw_spin_lock_irqsave();
#
Now look at those tracefs knobs:
# find /sys/kernel/tracing/ -name buffer_size_kb | xargs cat
2048
2048
2048
2048
2048
2048
2048
2048
2048
#
This should be similar to the -m option in the other perf tools, such as
'perf record', 'perf trace', etc.
Signed-off-by: Changbin Du <changbin.du@gmail.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt (VMware) <rostedt@goodmis.org>
Link: http://lore.kernel.org/lkml/20200808023141.14227-5-changbin.du@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This adds an option '-F/--funcs' to list all available functions to
trace, which is read from tracing file 'available_filter_functions'.
$ sudo ./perf ftrace -F | head
trace_initcall_finish_cb
initcall_blacklisted
do_one_initcall
do_one_initcall
trace_initcall_start_cb
run_init_process
try_to_run_init_process
match_dev_by_label
match_dev_by_uuid
rootfs_init_fs_context
$
Committer notes:
This is the same command line option and for the same purpose as in
'perf probe'.
Signed-off-by: Changbin Du <changbin.du@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt (VMware) <rostedt@goodmis.org>
Link: http://lore.kernel.org/lkml/20200808023141.14227-3-changbin.du@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The '-g/-G' options have already implied function_graph tracer should be
used instead of function tracer. So we don't need extra option
'--tracer' in this case.
This patch changes the behavior as below:
- If '-g' or '-G' option is on, then function_graph tracer is used.
- If '-T' or '-N' option is on, then function tracer is used.
- The function_graph has priority over function tracer.
- The option '--tracer' only take effect if neither -g/-G nor -T/-N
is specified.
Here are some examples.
This will start tracing all functions using default tracer:
$ sudo perf ftrace
This will trace all functions using function graph tracer:
$ sudo perf ftrace -G '*'
This will trace function vfs_read using function graph tracer:
$ sudo perf ftrace -G vfs_read
This will trace function vfs_read using function tracer:
$ sudo perf ftrace -T vfs_read
Committer notes:
Using '-h -G' will tell what that option is about, so to further clarify
the above examples:
# perf ftrace -h -G
-G, --graph-funcs <func> Set graph filter on given functions
# perf ftrace -h -g
-g, --nograph-funcs <func> Set nograph filter on given functions
# perf ftrace -h -T
-T, --trace-funcs <func> trace given functions only
# perf ftrace -h -N
-N, --notrace-funcs <func> do not trace given functions
#
Signed-off-by: Changbin Du <changbin.du@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt (VMware) <rostedt@goodmis.org>
Link: http://lore.kernel.org/lkml/20200808023141.14227-2-changbin.du@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Adad support to convert and store time of day in CTF data conversion for
'perf data convert' subcommand.
The perf.data used for conversion needs to have clock data information -
must be recorded with -k/--clockid option).
New --tod option is added to 'perf data convert' subcommand to convert
data with timestamps converted to wall clock time.
Record data with clockid set:
# perf record -k CLOCK_MONOTONIC kill
kill: not enough arguments
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.033 MB perf.data (8 samples) ]
Convert data with TOD timestamps:
# perf data convert --tod --to-ctf ./ctf
[ perf data convert: Converted 'perf.data' into CTF data './ctf' ]
[ perf data convert: Converted and wrote 0.000 MB (8 samples) ]
Display data in perf script:
# perf script -F+tod --ns
perf 262150 2020-07-13 18:38:50.097678523 153633.958246159: 1 cycles: ...
perf 262150 2020-07-13 18:38:50.097682941 153633.958250577: 1 cycles: ...
perf 262150 2020-07-13 18:38:50.097684997 153633.958252633: 7 cycles: ...
...
Display data in babeltrace:
# babeltrace --clock-date ./ctf
[2020-07-13 18:38:50.097678523] (+?.?????????) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ...
[2020-07-13 18:38:50.097682941] (+0.000004418) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ...
[2020-07-13 18:38:50.097684997] (+0.000002056) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ...
...
It's available only for recording with clockid specified, because it's
the only case where we can get reference time to wallclock time. It's
can't do that with perf clock yet.
Error is display if you want to use --tod on data without clockid
specified:
# perf data convert --tod --to-ctf ./ctf
Can't provide --tod time, missing clock data. Please record with -k/--clockid option.
Failed to setup CTF writer.
Error during conversion setup.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Geneviève Bastien <gbastien@versatic.net>
Cc: Ian Rogers <irogers@google.com>
Cc: Jeremie Galarneau <jgalar@efficios.com>
Cc: Michael Petlan <mpetlan@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lore.kernel.org/lkml/20200805093444.314999-6-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a new CLOCK_DATA feature that stores reference times when
-k/--clockid option is specified.
It contains the clock id and its reference time together with wall clock
time taken at the 'same time', both values are in nanoseconds.
The format of data is as below:
struct {
u32 version; /* version = 1 */
u32 clockid;
u64 wall_clock_ns;
u64 clockid_time_ns;
};
This clock reference times will be used in following changes to display
wall clock for perf events.
It's available only for recording with clockid specified, because it's
the only case where we can get reference time to wallclock time. It's
can't do that with perf clock yet.
Committer testing:
$ perf record -h -k
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-k, --clockid <clockid>
clockid to use for events, see clock_gettime()
$ perf record -k monotonic sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.017 MB perf.data (8 samples) ]
$ perf report --header-only | grep clockid -A1
# event : name = cycles:u, , id = { 88815, 88816, 88817, 88818, 88819, 88820, 88821, 88822 }, size = 120, { sample_period, sample_freq } = 4000, sample_type = IP|TID|TIME|PERIOD, read_format = ID, disabled = 1, inherit = 1, exclude_kernel = 1, mmap = 1, comm = 1, freq = 1, enable_on_exec = 1, task = 1, precise_ip = 3, sample_id_all = 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, use_clockid = 1, ksymbol = 1, bpf_event = 1, clockid = 1
# CPU_TOPOLOGY info available, use -I to display
--
# clockid frequency: 1000 MHz
# cpu pmu capabilities: branches=32, max_precise=3, pmu_name=skylake
# clockid: monotonic (1)
# reference time: 2020-08-06 09:40:21.619290 = 1596717621.619290 (TOD) = 21931.077673635 (monotonic)
$
Original-patch-by: David Ahern <dsahern@gmail.com>
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: David Ahern <dsahern@gmail.com>
Cc: Geneviève Bastien <gbastien@versatic.net>
Cc: Ian Rogers <irogers@google.com>
Cc: Jeremie Galarneau <jgalar@efficios.com>
Cc: Michael Petlan <mpetlan@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lore.kernel.org/lkml/20200805093444.314999-4-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
A single q option decodes ip from only FUP/TIP packets. Make it so that
repeating the q option (i.e. qq) decodes only PSB+, getting ip if there
is a FUP packet within PSB+ (i.e. between PSB and PSBEND).
Example:
$ perf record -e intel_pt//u grep -rI pudding drivers
[ perf record: Woken up 52 times to write data ]
[ perf record: Captured and wrote 57.870 MB perf.data ]
$ time perf script --itrace=bi | wc -l
58948289
real 1m23.863s
user 1m23.251s
sys 0m7.452s
$ time perf script --itrace=biq | wc -l
3385694
real 0m4.453s
user 0m4.455s
sys 0m0.328s
$ time perf script --itrace=biqq | wc -l
1883
real 0m0.047s
user 0m0.043s
sys 0m0.009s
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lore.kernel.org/lkml/20200710151104.15137-13-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Use the new itrace 'q' option to add support for a mode of decoding that
ignores TNT, does not walk object code, but gets the ip from FUP and TIP
packets.
Example:
$ perf record -e intel_pt//u grep -rI pudding drivers
[ perf record: Woken up 52 times to write data ]
[ perf record: Captured and wrote 57.870 MB perf.data ]
$ time perf script --itrace=bi | wc -l
58948289
real 1m23.863s
user 1m23.251s
sys 0m7.452s
$ time perf script --itrace=biq | wc -l
3385694
real 0m4.453s
user 0m4.455s
sys 0m0.328s
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lore.kernel.org/lkml/20200710151104.15137-12-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The itrace "e" option may be followed by flags which affect what errors
will or will not be reported. Each flag must be preceded by either '+' or '-'.
The flags supported by Intel PT are:
-o Suppress overflow errors
-l Suppress trace data lost errors
For example, for errors but not overflow or data lost errors:
--itrace=e-o-l
Suppressing those errors can be useful for testing and debugging because
they are not due to decoding.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lore.kernel.org/lkml/20200710151104.15137-7-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The usefulness of having a standard way of testing syscall performance
has come up from time to time[0]. Furthermore, some of our testing
machinery (such as 'mmtests') already makes use of a simplified version
of the microbenchmark. This patch mainly takes the same idea to measure
syscall throughput compatible with 'perf-bench' via getppid(2), yet
without any of the additional template stuff from Ingo's version (based
on numa.c). The code is identical to what mmtests uses.
[0] https://lore.kernel.org/lkml/20160201074156.GA27156@gmail.com/
Committer notes:
Add mising stdlib.h and unistd.h to get the prototypes for exit() and
getppid().
Committer testing:
$ perf bench
Usage:
perf bench [<common options>] <collection> <benchmark> [<options>]
# List of all available benchmark collections:
sched: Scheduler and IPC benchmarks
syscall: System call benchmarks
mem: Memory access benchmarks
numa: NUMA scheduling and MM benchmarks
futex: Futex stressing benchmarks
epoll: Epoll stressing benchmarks
internals: Perf-internals benchmarks
all: All benchmarks
$
$ perf bench syscall
# List of available benchmarks for collection 'syscall':
basic: Benchmark for basic getppid(2) calls
all: Run all syscall benchmarks
$ perf bench syscall basic
# Running 'syscall/basic' benchmark:
# Executed 10000000 getppid() calls
Total time: 3.679 [sec]
0.367957 usecs/op
2717708 ops/sec
$ perf bench syscall all
# Running syscall/basic benchmark...
# Executed 10000000 getppid() calls
Total time: 3.644 [sec]
0.364456 usecs/op
2743815 ops/sec
$
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lore.kernel.org/lkml/20190308181747.l36zqz2avtivrr3c@linux-r8p5
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Context switch events are added automatically by Intel PT and Coresight.
Make it possible to suppress them. That is useful for tracing the
scheduler without the disturbance that the switch event processing
creates.
Example:
Prerequisites:
$ which perf
~/bin/perf
$ sudo setcap "cap_sys_rawio,cap_sys_admin,cap_sys_ptrace,cap_syslog,cap_ipc_lock=ep" ~/bin/perf
$ sudo chmod +r /proc/kcore
Before:
$ perf record --no-switch-events --kcore -a -e intel_pt//k -- sleep 0.001
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.938 MB perf.data ]
$ perf script -D | grep PERF_RECORD_SWITCH | wc -l
572
After:
$ perf record --no-switch-events --kcore -a -e intel_pt//k -- sleep 0.001
Warning:
Intel Processor Trace decoding will not be possible except for kernel tracing!
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.838 MB perf.data ]
$ perf script -D | grep PERF_RECORD_SWITCH | wc -l
0
$ sudo chmod go-r /proc/kcore
$ sudo setcap -r ~/bin/perf
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: http://lore.kernel.org/lkml/20200528120859.21604-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
