The evlist__for_each_cpu iterator will call sched_setaffitinity when
moving between CPUs to avoid IPIs.
If only 1 IPI is saved then this may be unprofitable as the delay to get
scheduled may be considerable.
This may be particularly true if reading an event group in `perf stat`
in interval mode.
Move the affinity handling completely into the iterator so that a single
evlist__use_affinity can determine whether CPU affinities will be used.
For `perf record` the change is minimal as the dummy event and the real
event will always make the use of affinities the thing to do.
In `perf stat`, tool events are ignored and affinities only used if >1
event on the same CPU occur.
Determining if affinities are useful is done by evlist__use_affinity
which tests per-event whether the event's PMU benefits from affinity use
- it is assumed only perf event using PMUs do.
Fix a bug where when there are no affinities that the CPU map iterator
may reference a CPU not present in the initial evsel. Fix by making the
iterator and non-iterator code common.
Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andres Freund <andres@anarazel.de>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Dr. David Alan Gilbert <linux@treblig.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Test it first by having rust installed, then removing it and building again.
Fixes: 6a32fa5ccd ("tools build: Add a feature test for rust compiler")
Signed-off-by: Dmitry Dolgov <9erthalion6@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
test-rust.bin is missing from the list of FILES, and thus is not removed by the
clean target. This could lead to a false feature detection, since the binary
stays there. Fix it.
Fixes: 6a32fa5ccd ("tools build: Add a feature test for rust compiler")
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Test case perftool-testsuite_report fails on s390 for some time
now.
Root cause is a time out which is too tight for large s390 machines.
The time out value addr2line_timeout_ms is per default set to 1 second.
This is the maximum time the function read_addr2line_record() waits for
a reply from the forked off tool addr2line, which is started as a child
in interactive mode.
It reads stdin (an address in hexadecimal) and replies on stdout with
function name, file name and line number. This might take more than one
second.
However one second is not always enough and the reply from addr2line
tool is not received. Function read_addr2line_record() fails and emits
a warning, which is not expected by the test case. It fails.
Output before:
# perf test -F 133
-- [ PASS ] -- perf_report :: setup :: prepare the perf.data file
==================
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.087 MB \
/tmp/perftool-testsuite_report.FHz/perf_report/perf.data.1 \
(207 samples) ]
==================
-- [ PASS ] -- perf_report :: setup :: prepare the perf.data.1 file
## [ PASS ] ## perf_report :: setup SUMMARY
-- [ SKIP ] -- perf_report :: test_basic :: help message :: testcase skipped
Line did not match any pattern: "cmd__addr2line /usr/lib/debug/lib/modules/
6.19.0-20260205.rc8.git366.9845cf73f7db.300.fc43.s390x+next/
vmlinux: could not read first record"
Line did not match any pattern: "cmd__addr2line /usr/lib/debug/lib/modules/
6.19.0-20260205.rc8.git366.9845cf73f7db.300.fc43.s390x+next/
vmlinux: could not read first record"
-- [ FAIL ] -- perf_report :: test_basic :: basic execution
(output regexp parsing)
....
133: perftool-testsuite_report : FAILED!
Output after:
# ./perf test -F 133
-- [ PASS ] -- perf_report :: setup :: prepare the perf.data file
==================
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.087 MB \
/tmp/perftool-testsuite_report.Mlp/perf_report/perf.data.1
(188 samples) ]
==================
-- [ PASS ] -- perf_report :: setup :: prepare the perf.data.1 file
## [ PASS ] ## perf_report :: setup SUMMARY
-- [ SKIP ] -- perf_report :: test_basic :: help message :: testcase skipped
-- [ PASS ] -- perf_report :: test_basic :: basic execution
-- [ PASS ] -- perf_report :: test_basic :: number of samples
-- [ PASS ] -- perf_report :: test_basic :: header
-- [ PASS ] -- perf_report :: test_basic :: header timestamp
-- [ PASS ] -- perf_report :: test_basic :: show CPU utilization
-- [ PASS ] -- perf_report :: test_basic :: pid
-- [ PASS ] -- perf_report :: test_basic :: non-existing symbol
-- [ PASS ] -- perf_report :: test_basic :: symbol filter
-- [ PASS ] -- perf_report :: test_basic :: latency header
-- [ PASS ] -- perf_report :: test_basic :: default report for latency profile
-- [ PASS ] -- perf_report :: test_basic :: latency report for latency profile
-- [ PASS ] -- perf_report :: test_basic :: parallelism histogram
## [ PASS ] ## perf_report :: test_basic SUMMARY
133: perftool-testsuite_report : Ok
#
Fixes: 257046a367 ("perf srcline: Fallback between addr2line implementations")
Reviewed-by: Jan Polensky <japo@linux.ibm.com>
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Ian Rogers <irogers@google.com>
Cc: linux-s390@vger.kernel.org
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Sumanth Korikkar <sumanthk@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
$ perf test 'perf data type profiling tests'
83: perf data type profiling tests : Skip
$ perf test -vv 'perf data type profiling tests'
83: perf data type profiling tests:
--- start ---
test child forked, pid 977213
Skip: code_with_type workload not built in 'perf test'
---- end(-2) ----
83: perf data type profiling tests : Skip
$
Cc: Dmitrii Dolgov <9erthalion6@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Features in FEATURE_TESTS_BASIC will be set as being available if
test-all.c builds, so since the rust test isn't included in test-all.c,
we can't have 'rust' in there, remove it from FEATURE_TESTS_BASIC and
use feature-check so that it tries to build test-rust.bin, doing the
actual feature detection.
On a system lacking a rust compiler:
Makefile.config:1158: Rust is not found. Test workloads with rust are disabled.
Auto-detecting system features:
... libdw: [ on ]
... glibc: [ on ]
... libelf: [ on ]
... libnuma: [ on ]
... numa_num_possible_cpus: [ on ]
... libpython: [ on ]
... libcapstone: [ on ]
... llvm-perf: [ on ]
... zlib: [ on ]
... lzma: [ on ]
... bpf: [ on ]
... libaio: [ on ]
... libzstd: [ on ]
... libopenssl: [ on ]
... rust: [ OFF ]
$ cat /tmp/build/perf-tools-next/feature/test-rust.make.output
/bin/sh: line 1: rustc: command not found
$ file /tmp/build/perf-tools-next/feature/test-rust.bin
/tmp/build/perf-tools-next/feature/test-rust.bin: cannot open `/tmp/build/perf-tools-next/feature/test-rust.bin' (No such file or directory)
$
$ perf -vv | grep RUST
rust: [ OFF ] # HAVE_RUST_SUPPORT
$
And after installing it:
... rust: [ on ]
$ cat /tmp/build/perf-tools-next/feature/test-rust.make.output
$ file /tmp/build/perf-tools-next/feature/test-rust.bin
/tmp/build/perf-tools-next/feature/test-rust.bin: ELF 64-bit LSB pie executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, BuildID[sha1]=9c416edf673ee3705b97bae893a99a6fcf1ee258, for GNU/Linux 3.2.0, with debug_info, not stripped
$
$ perf -vv | grep RUST
rust: [ on ] # HAVE_RUST_SUPPORT
$
Fixes: 6a32fa5ccd ("tools build: Add a feature test for rust compiler")
Cc: Dmitrii Dolgov <9erthalion6@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Exercise the annotate command with data type profiling feature with C.
For that extend the existing data type profiling shell test to profile
the datasym workload, then annotate the result expecting to see some
data structures from the C code.
Committer testing:
root@number:~# perf test 'perf data type profiling tests'
83: perf data type profiling tests : Ok
root@number:~# perf test -vv 'perf data type profiling tests'
83: perf data type profiling tests:
--- start ---
test child forked, pid 125028
Basic Rust perf annotate test
Basic annotate test [Success]
Pipe Rust perf annotate test
Pipe annotate test [Success]
Basic C perf annotate test
Basic annotate test [Success]
Pipe C perf annotate test
Pipe annotate test [Success]
---- end(0) ----
83: perf data type profiling tests : Ok
root@number:~#
Signed-off-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Exercise the annotate command with data type profiling feature on the
rust runtime. For that add a new shell test, which will profile the
code_with_type workload, then annotate the result expecting to see some
data structures from the rust code.
Committer testing:
root@number:~# perf test 'perf data type profiling tests'
83: perf data type profiling tests : Ok
root@number:~# perf test -v 'perf data type profiling tests'
83: perf data type profiling tests : Ok
root@number:~# perf test -vv 'perf data type profiling tests'
83: perf data type profiling tests:
--- start ---
test child forked, pid 111044
Basic perf annotate test
Basic annotate test [Success]
Pipe perf annotate test
Pipe annotate test [Success]
---- end(0) ----
83: perf data type profiling tests : Ok
root@number:~#
Signed-off-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The purpose of the workload is to gather samples of rust runtime. To
achieve that it has a dummy rust library linked with it.
Per recommendations for such scenarios [1], the rust library is
statically linked.
An example:
$ perf record perf test -w code_with_type
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.160 MB perf.data (4074 samples) ]
$ perf report --stdio --dso perf -s srcfile,srcline
45.16% ub_checks.rs ub_checks.rs:72
6.72% code_with_type.rs code_with_type.rs:15
6.64% range.rs range.rs:767
4.26% code_with_type.rs code_with_type.rs:21
4.23% range.rs range.rs:0
3.99% code_with_type.rs code_with_type.rs:16
[...]
[1]: https://doc.rust-lang.org/reference/linkage.html#mixed-rust-and-foreign-codebases
Signed-off-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a feature test to identify if the rust compiler is available, so
that perf could build rust based worloads based on that.
Signed-off-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add quotes to avoid the following warning:
```
In tests/shell/record.sh line 264:
[ $(uname -m) = "s390x" ] && {
^---------^ SC2046 (warning): Quote this to prevent word splitting.
For more information:
https://www.shellcheck.net/wiki/SC2046 -- Quote this to prevent word splitt...
```
Fixes: c73a56ed3c ("perf test: Fix test case Leader sampling on s390")
Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When building to an output directory the previous code would remove
files and then copy the source files over.
Each source file copy would have a rule to make its directory. All JSON
for every architecture was considered a source file.
This led to unnecessary copying as a file would be deleted and then the
same file copied again, unnecessary directory making, and copying of
files not used in the build.
A side-effect would be a lot of build messages.
This change makes it so that all computed output files are created and
then compared to all files in the OUTPUT directory.
By filtering out the files that would be copied, unnecessary files can
be determined and then deleted - note, this is a phony target which
would remake the pmu-events.c if always depended upon, and so the
dependency is conditional on there being files to remove.
This has some overhead as the $(OUTPUT)/pmu-events is "find" over rather
than just "rm -fr", but the savings from unnecessary copying, etc.
should make up for this new make overhead.
The copy target just does copying but has a dependency on the directory
it needs being built, avoiding repetitive mkdirs.
The source files for copying only consider the JEVENTS_ARCH unless the
JEVENTS_ARCH is all.
The metric JSON is only generated if appropriate, rather than always
being generated and jevents.py deciding whether or not to use the files.
The mypy and pylint targets are fixed as variable names had changed but
the rules not updated.
The line count of a build with "make -C tools/perf O=/tmp/perf clean all"
prior to this change was 2181 lines, after this change it is 1596
lines.
This is a reduction of 585 lines or about 27%.
The generated pmu-events.c for JEVENTS_ARCH "x86" and "all" were
validated as being identical after this change.
Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Leo Yan <leo.yan@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When run on a kernel without BTF info, perf crashes:
libbpf: kernel BTF is missing at '/sys/kernel/btf/vmlinux', was CONFIG_DEBUG_INFO_BTF enabled?
libbpf: failed to find valid kernel BTF
Program received signal SIGSEGV, Segmentation fault.
0x00005555556915b7 in btf.type_cnt ()
(gdb) bt
#0 0x00005555556915b7 in btf.type_cnt ()
#1 0x0000555555691fbc in btf_find_by_name_kind ()
#2 0x00005555556920d0 in btf.find_by_name_kind ()
#3 0x00005555558a1b7c in init_numa_data (con=0x7fffffffd0a0) at util/bpf_lock_contention.c:125
#4 0x00005555558a264b in lock_contention_prepare (con=0x7fffffffd0a0) at util/bpf_lock_contention.c:313
#5 0x0000555555620702 in __cmd_contention (argc=0, argv=0x7fffffffea10) at builtin-lock.c:2084
#6 0x0000555555622c8d in cmd_lock (argc=0, argv=0x7fffffffea10) at builtin-lock.c:2755
#7 0x0000555555651451 in run_builtin (p=0x555556104f00 <commands+576>, argc=3, argv=0x7fffffffea10)
at perf.c:349
#8 0x00005555556516ed in handle_internal_command (argc=3, argv=0x7fffffffea10) at perf.c:401
#9 0x000055555565184e in run_argv (argcp=0x7fffffffe7fc, argv=0x7fffffffe7f0) at perf.c:445
#10 0x0000555555651b9f in main (argc=3, argv=0x7fffffffea10) at perf.c:553
Check if btf loading failed, and don't do anything with it in
init_numa_data(). This leads to the following error message, instead of
just a crash:
libbpf: kernel BTF is missing at '/sys/kernel/btf/vmlinux', was CONFIG_DEBUG_INFO_BTF enabled?
libbpf: failed to find valid kernel BTF
libbpf: kernel BTF is missing at '/sys/kernel/btf/vmlinux', was CONFIG_DEBUG_INFO_BTF enabled?
libbpf: failed to find valid kernel BTF
libbpf: Error loading vmlinux BTF: -ESRCH
libbpf: failed to load BPF skeleton 'lock_contention_bpf': -ESRCH
Failed to load lock-contention BPF skeleton
lock contention BPF setup failed
Signed-off-by: Tycho Andersen (AMD) <tycho@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Chun-Tse Shao <ctshao@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: K Prateek Nayak <kprateek.nayak@amd.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Testing:
- Built perf
- Executed perf mem record and report
Committer notes:
This addresses a TODO and improves the situation where record and
report/c2c are performed on the same machine or in machines with the
same cacheline size, but the proper way is to store the cacheline size
in the perf.data header at 'record' time and then use it at post
processing time.
Signed-off-by: Ricky Ringler <ricky.ringler@proton.me>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20260129004223.26799-1-ricky.ringler@proton.me
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The subtest 'Leader sampling' some time fails on s390.
- for z/VM guest: Disable the test for z/VM guest. There is no
CPU Measurement facility to run the test successfully.
- for LPAR: Use correct event names.
A detailed analysis follows here:
Now to the debugging and investigation:
1. With command
perf record -e '{cycles,cycles}:S' -- ....
the first cycles event starts sampling.
On s390 this sets up sampling with a frequency of 4000 Hz.
This translates to hardware sample rate of 1377000 instructions per
micro-second to meet a frequency of 4000 HZ.
2. With first event cycles now sampling into a hardware buffer, an
interrupt is triggered each time a sampling buffer gets full.
The interrupt handler is then invoked and debug output shows the
processing of samples. The size of one hardware sample is 32 bytes.
With an interrupt triggered when the hardware buffer page of 4KB
gets full, the interrupt handler processes 128 samples.
(This is taken from s390 specific fast debug data gathering)
2025-11-07 14:35:51.977248 000003ffe013cbfa \
perf_event_count_update event->count 0x0 count 0x1502e8
2025-11-07 14:35:51.977248 000003ffe013cbfa \
perf_event_count_update event->count 0x1502e8 count 0x1502e8
2025-11-07 14:35:51.977248 000003ffe013cbfa \
perf_event_count_update event->count 0x2a05d0 count 0x1502e8
2025-11-07 14:35:51.977252 000003ffe013cbfa \
perf_event_count_update event->count 0x3f08b8 count 0x1502e8
2025-11-07 14:35:51.977252 000003ffe013cbfa \
perf_event_count_update event->count 0x540ba0 count 0x1502e8
2025-11-07 14:35:51.977253 000003ffe013cbfa \
perf_event_count_update event->count 0x690e88 count 0x1502e8
2025-11-07 14:35:51.977254 000003ffe013cbfa \
perf_event_count_update event->count 0x7e1170 count 0x1502e8
2025-11-07 14:35:51.977254 000003ffe013cbfa \
perf_event_count_update event->count 0x931458 count 0x1502e8
2025-11-07 14:35:51.977254 000003ffe013cbfa \
perf_event_count_update event->count 0xa81740 count 0x1502e8
3. The value is constantly increasing by the number of instructions
executed to generate a sample entry. This is the first line of the
pairs of lines. count 0x1502e8 --> 1377000
# perf script | grep 1377000 | wc -l
214
# perf script | wc -l
428
#
That is 428 lines in total, and half of the lines contain value
1377000.
4. The second event cycles is opened against the counting PMU, which
is an independent PMU and is not interrupt driven. Once enabled it
runs in the background and keeps running, incrementing silently
about 400+ counters. The counter values are read via assembly
instructions.
This second counter PMU's read call back function is called when the
interrupt handler of the sampling facility processes each sample. The
function call sequence is:
perf_event_overflow()
+--> __perf_event_overflow()
+--> __perf_event_output()
+--> perf_output_sample()
+--> perf_output_read()
+--> perf_output_read_group()
for_each_sibling_event(sub, leader) {
values[n++] = perf_event_count(sub, self);
printk("%s sub %p values %#lx\n", __func__, sub, values[n-1]);
}
The last function perf_event_count() is invoked on the second event
cylces *on* the counting PMU. An added printk statement shows the
following lines in the dmesg output:
# dmesg|grep perf_output_read_group |head -10
[ 332.368620] perf_output_read_group sub 00000000d80b7c1f values 0x3a80917 (1)
[ 332.368624] perf_output_read_group sub 00000000d80b7c1f values 0x3a86c7f (2)
[ 332.368627] perf_output_read_group sub 00000000d80b7c1f values 0x3a89c15 (3)
[ 332.368629] perf_output_read_group sub 00000000d80b7c1f values 0x3a8c895 (4)
[ 332.368631] perf_output_read_group sub 00000000d80b7c1f values 0x3a8f569 (5)
[ 332.368633] perf_output_read_group sub 00000000d80b7c1f values 0x3a9204b
[ 332.368635] perf_output_read_group sub 00000000d80b7c1f values 0x3a94790
[ 332.368637] perf_output_read_group sub 00000000d80b7c1f values 0x3a9704b
[ 332.368638] perf_output_read_group sub 00000000d80b7c1f values 0x3a99888
#
This correlates with the output of
# perf report -D | grep 'id 00000000000000'|head -10
..... id 0000000000000006, value 00000000001502e8, lost 0
..... id 000000000000000e, value 0000000003a80917, lost 0 --> line (1) above
..... id 0000000000000006, value 00000000002a05d0, lost 0
..... id 000000000000000e, value 0000000003a86c7f, lost 0 --> line (2) above
..... id 0000000000000006, value 00000000003f08b8, lost 0
..... id 000000000000000e, value 0000000003a89c15, lost 0 --> line (3) above
..... id 0000000000000006, value 0000000000540ba0, lost 0
..... id 000000000000000e, value 0000000003a8c895, lost 0 --> line (4) above
..... id 0000000000000006, value 0000000000690e88, lost 0
..... id 000000000000000e, value 0000000003a8f569, lost 0 --> line (5) above
Summary:
- Above command starts the CPU sampling facility, with runs interrupt
driven when a 4KB page is full. An interrupt processes the 128 samples
and calls eventually perf_output_read_group() for each sample to save it
in the event's ring buffer.
- At that time the CPU counting facility is invoked to read the value of
the event cycles. This value is saved as the second value in the
sample_read structure.
- The first and odd lines in the perf script output displays the period
value between 2 samples being created by hardware. It is the number
of instructions executes before the hardware writes a sample.
- The second and even lines in the perf script output displays the number
of CPU cycles needed to process each sample and save it in the event's
ring buffer.
These 2 different values can never be identical on s390.
Since event leader sampling is not possible on s390 the perf tool will
return EOPNOTSUPP soon. Perpare the test case for that.
Suggested-by: James Clark <james.clark@linaro.org>
Reviewed-by: Jan Polensky <japo@linux.ibm.com>
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Tested-by: Jan Polensky <japo@linux.ibm.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Sumanth Korikkar <sumanthk@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
On data type profiling, it tried to match register name with a partial
string. For example, it allowed to match with "%rbp)" or "%rdi,8)".
But with recent change in the area, it doesn't match anymore and break
the data type profiling.
Let's pass the correct register name by removing the unwanted part.
Add arch__dwarf_regnum() to handle it in a single place.
Closes: 7d3n23li6drroxrdlpxn7ixehdeszkjdftah3zyngjl2qs22ef@yelcjv53v42o
Reported-by: Dmitry Dolgov <9erthalion6@gmail.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zecheng Li <zli94@ncsu.edu>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Currently, `perf stat` skips or hides metrics when the underlying
hardware events cannot be counted (e.g., due to insufficient permissions
or unsupported events).
In `--metric-only` mode, this often results in missing columns or blank
spaces, making the output difficult to parse.
Modify the logic to ensure metrics are consistently displayed by
propagating NAN (Not a Number) through the expression evaluator.
Specifically:
1. Update `prepare_metric()` in stat-shadow.c to treat uncounted events
(where `run == 0`) as NAN. This leverages the existing math in expr.y
to propagate NAN through metric expressions.
2. Remove the early return in the display logic's `printout()` function
that was previously skipping metrics in `--metric-only` mode for
failed events.
l
3. Simplify `perf_stat__skip_metric_event()` to no longer depend on
event runtime.
Tested:
1. `perf all metrics test` did not crash while paranoid is 2.
2. Multiple combinations with `CPUs_utilized` while paranoid is 2.
$ ./perf stat -M CPUs_utilized -a -- sleep 1
Performance counter stats for 'system wide':
<not supported> msec cpu-clock:u # nan CPUs CPUs_utilized
1,006,356,120 duration_time
1.004375550 seconds time elapsed
$ ./perf stat -M CPUs_utilized -a -j -- sleep 1
{"counter-value" : "<not supported>", "unit" : "msec", "event" : "cpu-clock:u", "event-runtime" : 0, "pcnt-running" : 100.00, "metric-value" : "nan", "metric-unit" : "CPUs CPUs_utilized"}
{"counter-value" : "1006642462.000000", "unit" : "", "event" : "duration_time", "event-runtime" : 1, "pcnt-running" : 100.00}
$ ./perf stat -M CPUs_utilized -a --metric-only -- sleep 1
Performance counter stats for 'system wide':
CPUs CPUs_utilized
nan
1.004424652 seconds time elapsed
$ ./perf stat -M CPUs_utilized -a --metric-only -j -- sleep 1
{"CPUs CPUs_utilized" : "none"}
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Chun-Tse Shao <ctshao@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The current IP of a leaf function when reported from a perf record with
"--call-graph lbr" is the "to" field of the LBR branch stack record.
The sample for the event being recorded may be further into the function
and there may be inlining information associated with it.
Rather than use the branch stack "to" field in this case switch to the
callchain appending the sample->ip and thereby allowing the inline
information to show.
Before this change:
```
$ perf record --call-graph lbr perf test -w inlineloop
...
$ perf script --fields +srcline
...
perf-inlineloop 467586 4649.344493: 950905 cpu_core/cycles/P:
55dfda2829c0 parent+0x0 (perf)
inlineloop.c:31
55dfda282a96 inlineloop+0x86 (perf)
inlineloop.c:47
55dfda236420 run_workload+0x59 (perf)
builtin-test.c:715
55dfda236b03 cmd_test+0x413 (perf)
builtin-test.c:825
...
```
After this change:
```
$ perf record --call-graph lbr perf test -w inlineloop
...
$ perf script --fields +srcline
...
perf-inlineloop 529703 11878.680815: 950905 cpu_core/cycles/P:
555ce86be9e6 leaf+0x26
inlineloop.c:20 (inlined)
555ce86be9e6 middle+0x26
inlineloop.c:27 (inlined)
555ce86be9e6 parent+0x26 (perf)
inlineloop.c:32
555ce86bea96 inlineloop+0x86 (perf)
inlineloop.c:47
555ce8672420 run_workload+0x59 (perf)
builtin-test.c:715
555ce8672b03 cmd_test+0x413 (perf)
builtin-test.c:825
...
```
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Dmitriy Vyukov <dvyukov@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Krzysztof Łopatowski <krzysztof.m.lopatowski@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Weilin Wang <weilin.wang@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Currently, some architecture-specific perf-regs functions, such as
arch__intr_reg_mask() and arch__user_reg_mask(), are defined with the
__weak attribute.
This approach ensures that only functions matching the architecture of
the build/run host are compiled and executed, reducing build time and
binary size.
However, this __weak attribute restricts these functions to be called
only on the same architecture, preventing cross-architecture
functionality.
For example, a perf.data file captured on x86 cannot be parsed on an ARM
platform.
To address this limitation, this patch removes the __weak attribute from
these perf-regs functions.
The architecture-specific code is moved from the arch/ directory to the
util/perf-regs-arch/ directory.
The appropriate architectural functions are then called based on the
EM_HOST.
No functional changes are intended.
Suggested-by: Ian Rogers <irogers@google.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Guo Ren <guoren@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: John Garry <john.g.garry@oracle.com>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <pjw@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Xudong Hao <xudong.hao@intel.com>
Cc: Zide Chen <zide.chen@intel.com>
[ Fixed up somme fuzz with s390 and riscv Build files wrt removing perf_regs.o ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Since commit ee27476fa3 ("perf record: Skip don't fail for events
that don't open"), if a user does not have permission to access a PMU
event, perf reports:
perf record -e cs_etm// -C 3 -- ls
Error:
Failure to open event 'cs_etm//u' on PMU 'cs_etm' which will be removed.
No fallback found for 'cs_etm//u' for error 13
Error:
Failure to open event 'dummy:u' on PMU 'software' which will be removed.
No fallback found for 'dummy:u' for error 13
Error:
Failure to open any events for recording.
The log is not very helpful, as no clear indication of what "error 13"
means or how to address the issue.
This commit restores evsel__open_strerror() to generate a readable error
message and print it out:
perf record -e cs_etm// -C 3 -- ls
Error:
Failure to open event 'cs_etm//' on PMU 'cs_etm' which will be removed.
Access to performance monitoring and observability operations is limited.
Consider adjusting /proc/sys/kernel/perf_event_paranoid setting to open
access to performance monitoring and observability operations for processes
without CAP_PERFMON, CAP_SYS_PTRACE or CAP_SYS_ADMIN Linux capability.
More information can be found at 'Perf events and tool security' document:
https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html
perf_event_paranoid setting is 1:
-1: Allow use of (almost) all events by all users
Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK
>= 0: Disallow raw and ftrace function tracepoint access
>= 1: Disallow CPU event access
>= 2: Disallow kernel profiling
To make the adjusted perf_event_paranoid setting permanent preserve it
in /etc/sysctl.conf (e.g. kernel.perf_event_paranoid = <setting>)
Error:
Failure to open event 'dummy:u' on PMU 'software' which will be removed.
Access to performance monitoring and observability operations is limited.
Consider adjusting /proc/sys/kernel/perf_event_paranoid setting to open
access to performance monitoring and observability operations for processes
without CAP_PERFMON, CAP_SYS_PTRACE or CAP_SYS_ADMIN Linux capability.
More information can be found at 'Perf events and tool security' document:
https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html
perf_event_paranoid setting is 1:
-1: Allow use of (almost) all events by all users
Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK
>= 0: Disallow raw and ftrace function tracepoint access
>= 1: Disallow CPU event access
>= 2: Disallow kernel profiling
To make the adjusted perf_event_paranoid setting permanent preserve it
in /etc/sysctl.conf (e.g. kernel.perf_event_paranoid = <setting>)
Error:
Failure to open any events for recording.
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Leo Yan <leo.yan@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When building tools/perf the CFLAGS can contain a directory for the
installed headers.
As the headers may be being installed while building libperf.a this can
cause headers to be partially installed and found in the include path
while building an object file for libperf.a.
The installed header may reference other installed headers that are
missing given the partial nature of the install and then the build fails
with a missing header file.
Avoid this by ensuring the libperf source headers are always first in
the CFLAGS.
Fixes: 3143504918 ("libperf: Make libperf.a part of the perf build")
Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
