perf record
-----------
* Introduce latency profiling using scheduler information. The latency
profiling is to show impacts on wall-time rather than cpu-time. By
tracking context switches, it can weight samples and find which part
of the code contributed more to the execution latency.
The value (period) of the sample is weighted by dividing it by the
number of parallel execution at the moment. The parallelism is
tracked in perf report with sched-switch records. This will reduce
the portion that are run in parallel and in turn increase the portion
of serial executions.
For now, it's limited to profile processes, IOW system-wide profiling
is not supported. You can add --latency option to enable this.
$ perf record --latency -- make -C tools/perf
I've run the above command for perf build which adds -j option to
make with the number of CPUs in the system internally. Normally
it'd show something like below:
$ perf report -F overhead,comm
...
#
# Overhead Command
# ........ ...............
#
78.97% cc1
6.54% python3
4.21% shellcheck
3.28% ld
1.80% as
1.37% cc1plus
0.80% sh
0.62% clang
0.56% gcc
0.44% perl
0.39% make
...
The cc1 takes around 80% of the overhead as it's the actual compiler.
However it runs in parallel so its contribution to latency may be less
than that. Now, perf report will show both overhead and latency (if
--latency was given at record time) like below:
$ perf report -s comm
...
#
# Overhead Latency Command
# ........ ........ ...............
#
78.97% 48.66% cc1
6.54% 25.68% python3
4.21% 0.39% shellcheck
3.28% 13.70% ld
1.80% 2.56% as
1.37% 3.08% cc1plus
0.80% 0.98% sh
0.62% 0.61% clang
0.56% 0.33% gcc
0.44% 1.71% perl
0.39% 0.83% make
...
You can see latency of cc1 goes down to around 50% and python3 and ld
contribute a lot more than their overhead. You can use --latency
option in perf report to get the same result but ordered by latency.
$ perf report --latency -s comm
perf report
-----------
* As a side effect of the latency profiling work, it adds a new output
field 'latency' and a sort key 'parallelism'. The below is a result
from my system with 64 CPUs. The build was well-parallelized but
contained some serial portions.
$ perf report -s parallelism
...
#
# Overhead Latency Parallelism
# ........ ........ ...........
#
16.95% 1.54% 62
13.38% 1.24% 61
12.50% 70.47% 1
11.81% 1.06% 63
7.59% 0.71% 60
4.33% 12.20% 2
3.41% 0.33% 59
2.05% 0.18% 64
1.75% 1.09% 9
1.64% 1.85% 5
...
* Support Feodra mini-debuginfo which is a LZMA compressed symbol table
inside ".gnu_debugdata" ELF section.
perf annotate
-------------
* Add --code-with-type option to enable data-type profiling with the
usual annotate output. Instead of focusing on data structure, it
shows code annotation together with data type it accesses in case the
instruction refers to a memory location (and it was able to resolve
the target data type). Currently it only works with --stdio.
$ perf annotate --stdio --code-with-type
...
Percent | Source code & Disassembly of vmlinux for cpu/mem-loads,ldlat=30/pp (18 samples, percent: local period)
----------------------------------------------------------------------------------------------------------------------
: 0 0xffffffff81050610 <__fdget>:
0.00 : ffffffff81050610: callq 0xffffffff81c01b80 <__fentry__> # data-type: (stack operation)
0.00 : ffffffff81050615: pushq %rbp # data-type: (stack operation)
0.00 : ffffffff81050616: movq %rsp, %rbp
0.00 : ffffffff81050619: pushq %r15 # data-type: (stack operation)
0.00 : ffffffff8105061b: pushq %r14 # data-type: (stack operation)
0.00 : ffffffff8105061d: pushq %rbx # data-type: (stack operation)
0.00 : ffffffff8105061e: subq $0x10, %rsp
0.00 : ffffffff81050622: movl %edi, %ebx
0.00 : ffffffff81050624: movq %gs:0x7efc4814(%rip), %rax # 0x14e40 <current_task> # data-type: struct task_struct* +0
0.00 : ffffffff8105062c: movq 0x8d0(%rax), %r14 # data-type: struct task_struct +0x8d0 (files)
0.00 : ffffffff81050633: movl (%r14), %eax # data-type: struct files_struct +0 (count.counter)
0.00 : ffffffff81050636: cmpl $0x1, %eax
0.00 : ffffffff81050639: je 0xffffffff810506a9 <__fdget+0x99>
0.00 : ffffffff8105063b: movq 0x20(%r14), %rcx # data-type: struct files_struct +0x20 (fdt)
0.00 : ffffffff8105063f: movl (%rcx), %eax # data-type: struct fdtable +0 (max_fds)
0.00 : ffffffff81050641: cmpl %ebx, %eax
0.00 : ffffffff81050643: jbe 0xffffffff810506ef <__fdget+0xdf>
0.00 : ffffffff81050649: movl %ebx, %r15d
5.56 : ffffffff8105064c: movq 0x8(%rcx), %rdx # data-type: struct fdtable +0x8 (fd)
...
The "# data-type:" part was added with this change. The first few
entries are not very interesting. But later you can it accesses
a couple of fields in the task_struct, files_struct and fdtable.
perf trace
----------
* Support syscall tracing for different ABI. For example it can trace
system calls for 32-bit applications on 64-bit kernel transparently.
* Add --summary-mode=total option to show global syscall summary. The
default is 'thread' to show per-thread syscall summary.
Python support
--------------
* Add more interfaces to 'perf' module to parse events, and config,
enable or disable the event list properly so that it can implement
basic functionalities purely in Python. There is an example code
for these new interfaces in python/tracepoint.py.
* Add mypy and pylint support to enable build time checking. Fix
some code based on the findings from these tools.
Internals
---------
* Introduce io_dir__readdir() API to make directory traveral (usually
for proc or sysfs) efficient with less memory footprint.
JSON vendor events
------------------
* Add events and metrics for ARM Neoverse N3 and V3
* Update events and metrics on various Intel CPUs
* Add/update events for a number of SiFive processors
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQSo2x5BnqMqsoHtzsmMstVUGiXMgwUCZ+Y/YQAKCRCMstVUGiXM
g64CAP9sdUgKCe66JilZRAEy9d3Cp835v7KjHSlCXLXkRSoy6wD+KH96Y0OqVtGw
GYON4s9ZoXAoH+J0lKIFSffVL9MhYgY=
=7L7B
-----END PGP SIGNATURE-----
Merge tag 'perf-tools-for-v6.15-2025-03-27' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools
Pull perf tools updates from Namhyung Kim:
"perf record:
- Introduce latency profiling using scheduler information.
The latency profiling is to show impacts on wall-time rather than
cpu-time. By tracking context switches, it can weight samples and
find which part of the code contributed more to the execution
latency.
The value (period) of the sample is weighted by dividing it by the
number of parallel execution at the moment. The parallelism is
tracked in perf report with sched-switch records. This will reduce
the portion that are run in parallel and in turn increase the
portion of serial executions.
For now, it's limited to profile processes, IOW system-wide
profiling is not supported. You can add --latency option to enable
this.
$ perf record --latency -- make -C tools/perf
I've run the above command for perf build which adds -j option to
make with the number of CPUs in the system internally. Normally
it'd show something like below:
$ perf report -F overhead,comm
...
#
# Overhead Command
# ........ ...............
#
78.97% cc1
6.54% python3
4.21% shellcheck
3.28% ld
1.80% as
1.37% cc1plus
0.80% sh
0.62% clang
0.56% gcc
0.44% perl
0.39% make
...
The cc1 takes around 80% of the overhead as it's the actual
compiler. However it runs in parallel so its contribution to
latency may be less than that. Now, perf report will show both
overhead and latency (if --latency was given at record time) like
below:
$ perf report -s comm
...
#
# Overhead Latency Command
# ........ ........ ...............
#
78.97% 48.66% cc1
6.54% 25.68% python3
4.21% 0.39% shellcheck
3.28% 13.70% ld
1.80% 2.56% as
1.37% 3.08% cc1plus
0.80% 0.98% sh
0.62% 0.61% clang
0.56% 0.33% gcc
0.44% 1.71% perl
0.39% 0.83% make
...
You can see latency of cc1 goes down to around 50% and python3 and
ld contribute a lot more than their overhead. You can use --latency
option in perf report to get the same result but ordered by
latency.
$ perf report --latency -s comm
perf report:
- As a side effect of the latency profiling work, it adds a new
output field 'latency' and a sort key 'parallelism'. The below is a
result from my system with 64 CPUs. The build was well-parallelized
but contained some serial portions.
$ perf report -s parallelism
...
#
# Overhead Latency Parallelism
# ........ ........ ...........
#
16.95% 1.54% 62
13.38% 1.24% 61
12.50% 70.47% 1
11.81% 1.06% 63
7.59% 0.71% 60
4.33% 12.20% 2
3.41% 0.33% 59
2.05% 0.18% 64
1.75% 1.09% 9
1.64% 1.85% 5
...
- Support Feodra mini-debuginfo which is a LZMA compressed symbol
table inside ".gnu_debugdata" ELF section.
perf annotate:
- Add --code-with-type option to enable data-type profiling with the
usual annotate output.
Instead of focusing on data structure, it shows code annotation
together with data type it accesses in case the instruction refers
to a memory location (and it was able to resolve the target data
type). Currently it only works with --stdio.
$ perf annotate --stdio --code-with-type
...
Percent | Source code & Disassembly of vmlinux for cpu/mem-loads,ldlat=30/pp (18 samples, percent: local period)
----------------------------------------------------------------------------------------------------------------------
: 0 0xffffffff81050610 <__fdget>:
0.00 : ffffffff81050610: callq 0xffffffff81c01b80 <__fentry__> # data-type: (stack operation)
0.00 : ffffffff81050615: pushq %rbp # data-type: (stack operation)
0.00 : ffffffff81050616: movq %rsp, %rbp
0.00 : ffffffff81050619: pushq %r15 # data-type: (stack operation)
0.00 : ffffffff8105061b: pushq %r14 # data-type: (stack operation)
0.00 : ffffffff8105061d: pushq %rbx # data-type: (stack operation)
0.00 : ffffffff8105061e: subq $0x10, %rsp
0.00 : ffffffff81050622: movl %edi, %ebx
0.00 : ffffffff81050624: movq %gs:0x7efc4814(%rip), %rax # 0x14e40 <current_task> # data-type: struct task_struct* +0
0.00 : ffffffff8105062c: movq 0x8d0(%rax), %r14 # data-type: struct task_struct +0x8d0 (files)
0.00 : ffffffff81050633: movl (%r14), %eax # data-type: struct files_struct +0 (count.counter)
0.00 : ffffffff81050636: cmpl $0x1, %eax
0.00 : ffffffff81050639: je 0xffffffff810506a9 <__fdget+0x99>
0.00 : ffffffff8105063b: movq 0x20(%r14), %rcx # data-type: struct files_struct +0x20 (fdt)
0.00 : ffffffff8105063f: movl (%rcx), %eax # data-type: struct fdtable +0 (max_fds)
0.00 : ffffffff81050641: cmpl %ebx, %eax
0.00 : ffffffff81050643: jbe 0xffffffff810506ef <__fdget+0xdf>
0.00 : ffffffff81050649: movl %ebx, %r15d
5.56 : ffffffff8105064c: movq 0x8(%rcx), %rdx # data-type: struct fdtable +0x8 (fd)
...
The "# data-type:" part was added with this change. The first few
entries are not very interesting. But later you can it accesses a
couple of fields in the task_struct, files_struct and fdtable.
perf trace:
- Support syscall tracing for different ABI. For example it can trace
system calls for 32-bit applications on 64-bit kernel
transparently.
- Add --summary-mode=total option to show global syscall summary. The
default is 'thread' to show per-thread syscall summary.
Python support:
- Add more interfaces to 'perf' module to parse events, and config,
enable or disable the event list properly so that it can implement
basic functionalities purely in Python. There is an example code
for these new interfaces in python/tracepoint.py.
- Add mypy and pylint support to enable build time checking. Fix some
code based on the findings from these tools.
Internals:
- Introduce io_dir__readdir() API to make directory traveral (usually
for proc or sysfs) efficient with less memory footprint.
JSON vendor events:
- Add events and metrics for ARM Neoverse N3 and V3
- Update events and metrics on various Intel CPUs
- Add/update events for a number of SiFive processors"
* tag 'perf-tools-for-v6.15-2025-03-27' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools: (229 commits)
perf bpf-filter: Fix a parsing error with comma
perf report: Fix a memory leak for perf_env on AMD
perf trace: Fix wrong size to bpf_map__update_elem call
perf tools: annotate asm_pure_loop.S
perf python: Fix setup.py mypy errors
perf test: Address attr.py mypy error
perf build: Add pylint build tests
perf build: Add mypy build tests
perf build: Rename TEST_LOGS to SHELL_TEST_LOGS
tools/build: Don't pass test log files to linker
perf bench sched pipe: fix enforced blocking reads in worker_thread
perf tools: Fix is_compat_mode build break in ppc64
perf build: filter all combinations of -flto for libperl
perf vendor events arm64 AmpereOneX: Fix frontend_bound calculation
perf vendor events arm64: AmpereOne/AmpereOneX: Mark LD_RETIRED impacted by errata
perf trace: Fix evlist memory leak
perf trace: Fix BTF memory leak
perf trace: Make syscall table stable
perf syscalltbl: Mask off ABI type for MIPS system calls
perf build: Remove Makefile.syscalls
...
The previous change to support cgroup filters introduced a bug that
pathname can include commas. It confused the lexer to treat an item and
the trailing comma as a single token. And it resulted in a parse error:
$ sudo perf record -e cycles:P --filter 'period > 0, ip > 64' -- true
perf_bpf_filter: Error: Unexpected item: 0,
perf_bpf_filter: syntax error, unexpected BFT_ERROR, expecting BFT_NUM
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
--filter <filter>
event filter
It should get "0" and "," separately.
An easiest fix would be to remove "," from the possible pathname
characters. As it's for cgroup names, probably ok to assume it won't
have commas in the pathname.
I found that the existing BPF filtering test didn't have any complex
filter condition with commas. Let's update the group filter test which
is supposed to test filter combinations like this.
Link: https://lore.kernel.org/r/20250307220922.434319-1-namhyung@kernel.org
Fixes: 91e88437d5156b20 ("perf bpf-filter: Support filtering on cgroups")
Reported-by: Sally Shi <sshii@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The env.pmu_mapping can be leaked when it reads data from a pipe on AMD.
For a pipe data, it reads the header data including pmu_mapping from
PERF_RECORD_HEADER_FEATURE runtime. But it's already set in:
perf_session__new()
__perf_session__new()
evlist__init_trace_event_sample_raw()
evlist__has_amd_ibs()
perf_env__nr_pmu_mappings()
Then it'll overwrite that when it processes the HEADER_FEATURE record.
Here's a report from address sanitizer.
Direct leak of 2689 byte(s) in 1 object(s) allocated from:
#0 0x7fed8f814596 in realloc ../../../../src/libsanitizer/lsan/lsan_interceptors.cpp:98
#1 0x5595a7d416b1 in strbuf_grow util/strbuf.c:64
#2 0x5595a7d414ef in strbuf_init util/strbuf.c:25
#3 0x5595a7d0f4b7 in perf_env__read_pmu_mappings util/env.c:362
#4 0x5595a7d12ab7 in perf_env__nr_pmu_mappings util/env.c:517
#5 0x5595a7d89d2f in evlist__has_amd_ibs util/amd-sample-raw.c:315
#6 0x5595a7d87fb2 in evlist__init_trace_event_sample_raw util/sample-raw.c:23
#7 0x5595a7d7f893 in __perf_session__new util/session.c:179
#8 0x5595a7b79572 in perf_session__new util/session.h:115
#9 0x5595a7b7e9dc in cmd_report builtin-report.c:1603
#10 0x5595a7c019eb in run_builtin perf.c:351
#11 0x5595a7c01c92 in handle_internal_command perf.c:404
#12 0x5595a7c01deb in run_argv perf.c:448
#13 0x5595a7c02134 in main perf.c:556
#14 0x7fed85833d67 in __libc_start_call_main ../sysdeps/nptl/libc_start_call_main.h:58
Let's free the existing pmu_mapping data if any.
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Link: https://lore.kernel.org/r/20250311000416.817631-1-namhyung@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
In linux-next
commit c760174401f6 ("perf cpumap: Reduce cpu size from int to int16_t")
causes the perf tests 100 126 to fail on s390:
Output before:
# ./perf test 100
100: perf trace BTF general tests : FAILED!
#
The root cause is the change from int to int16_t for the
cpu maps. The size of the CPU key value pair changes from
four bytes to two bytes. However a two byte key size is
not supported for bpf_map__update_elem().
Note: validate_map_op() in libbpf.c emits warning
libbpf: map '__augmented_syscalls__': \
unexpected key size 2 provided, expected 4
when key size is set to int16_t.
Therefore change to variable size back to 4 bytes for
invocation of bpf_map__update_elem().
Output after:
# ./perf test 100
100: perf trace BTF general tests : Ok
#
Fixes: c760174401f6 ("perf cpumap: Reduce cpu size from int to int16_t")
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Acked-by: Howard Chu <howardchu95@gmail.com>
Cc: James Clark <james.clark@linaro.org>
Link: https://lore.kernel.org/r/20250324152756.3879571-1-tmricht@linux.ibm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Annotate so it is built with non-executable stack.
Fixes: 8b97519711c3 ("perf test: Add asm pureloop test tool")
Signed-off-by: Marcus Meissner <meissner@suse.de>
Reviewed-by: Leo Yan <leo.yan@arm.com>
Link: https://lore.kernel.org/r/20250323085410.23751-1-meissner@suse.de
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
getenv may return None, so assert it isn't None for CC and srctree
environmental variables required for the script.
Disable an optional warning related to Popen.
Reviewed-by: James Clark <james.clark@linaro.org>
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250311213628.569562-7-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
ConfigParser existed in python2 but not in python3 causing mypy to
fail.
Whilst removing a python2 workaround remove reference to __future__.
Reviewed-by: James Clark <james.clark@linaro.org>
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250311213628.569562-6-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
If PYLINT=1 is passed to the build then run pylint over python code in
perf. Unlike shellcheck this isn't default on as there are currently
too many errors.
An example of an error:
```
************* Module setup
util/setup.py:19:0: C0301: Line too long (127/100) (line-too-long)
util/setup.py:20:0: C0301: Line too long (138/100) (line-too-long)
util/setup.py:63:0: C0301: Line too long (106/100) (line-too-long)
util/setup.py:1:0: C0114: Missing module docstring (missing-module-docstring)
util/setup.py:24:4: W0622: Redefining built-in 'vars' (redefined-builtin)
util/setup.py:11:4: C0103: Constant name "cc_options" doesn't conform to UPPER_CASE naming style (invalid-name)
util/setup.py:13:4: C0103: Constant name "cc_options" doesn't conform to UPPER_CASE naming style (invalid-name)
util/setup.py:15:34: R1732: Consider using 'with' for resource-allocating operations (consider-using-with)
util/setup.py:18:0: C0116: Missing function or method docstring (missing-function-docstring)
util/setup.py:19:16: R1732: Consider using 'with' for resource-allocating operations (consider-using-with)
util/setup.py:44:0: C0413: Import "from setuptools import setup, Extension" should be placed at the top of the module (wrong-import-position)
util/setup.py:46:0: C0413: Import "from setuptools.command.build_ext import build_ext as _build_ext" should be placed at the top of the module (wrong-import-position)
util/setup.py:47:0: C0413: Import "from setuptools.command.install_lib import install_lib as _install_lib" should be placed at the top of the module (wrong-import-position)
util/setup.py:49:0: C0115: Missing class docstring (missing-class-docstring)
util/setup.py:49:0: C0103: Class name "build_ext" doesn't conform to PascalCase naming style (invalid-name)
util/setup.py:52:8: W0201: Attribute 'build_lib' defined outside __init__ (attribute-defined-outside-init)
util/setup.py:53:8: W0201: Attribute 'build_temp' defined outside __init__ (attribute-defined-outside-init)
util/setup.py:55:0: C0115: Missing class docstring (missing-class-docstring)
util/setup.py:55:0: C0103: Class name "install_lib" doesn't conform to PascalCase naming style (invalid-name)
util/setup.py:58:8: W0201: Attribute 'build_dir' defined outside __init__ (attribute-defined-outside-init)
*-----------------------------------------------------------------
Your code has been rated at 6.67/10 (previous run: 6.51/10, +0.16)
make[4]: *** [util/Build:442: util/setup.py.pylint_log] Error 1
```
Reviewed-by: James Clark <james.clark@linaro.org>
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250311213628.569562-5-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
If MYPY=1 is passed to the build then run mypy over python code in
perf. Unlike shellcheck this isn't default on as there are currently
too many errors.
An example of an error:
```
util/setup.py:8: error: Item "None" of "str | None" has no attribute "split" [union-attr]
util/setup.py:15: error: Item "None" of "IO[bytes] | None" has no attribute "readline" [union-attr]
util/setup.py:15: error: List item 0 has incompatible type "str | None"; expected "str | bytes | PathLike[str] | PathLike[bytes]" [list-item]
util/setup.py:16: error: Unsupported left operand type for + ("None") [operator]
util/setup.py:16: note: Left operand is of type "str | None"
util/setup.py:74: error: Unsupported left operand type for + ("None") [operator]
util/setup.py:74: note: Left operand is of type "str | None"
Found 5 errors in 1 file (checked 1 source file)
make[4]: *** [util/Build:430: util/setup.py.mypy_log] Error 1
```
Reviewed-by: James Clark <james.clark@linaro.org>
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250311213628.569562-4-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Rename TEST_LOGS to SHELL_TEST_LOGS as later changes will add more
kinds of test logs.
Minor comment tweak in Makefile.perf as more than just test shell
tests are checked.
Reviewed-by: James Clark <james.clark@linaro.org>
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250311213628.569562-3-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The function worker_thread() is programmed in a way that roughly
doubles the number of expectable context switches, because it enforces
blocking reads:
Performance counter stats for 'perf bench sched pipe':
2,000,004 context-switches
11.859548321 seconds time elapsed
0.674871000 seconds user
8.076890000 seconds sys
The result of this behavior is that the blocking reads by far dominate
the performance analysis of 'perf bench sched pipe':
Samples: 78K of event 'cycles:P', Event count (approx.): 27964965844
Overhead Command Shared Object Symbol
25.28% sched-pipe [kernel.kallsyms] [k] read_hpet
8.11% sched-pipe [kernel.kallsyms] [k] retbleed_untrain_ret
2.82% sched-pipe [kernel.kallsyms] [k] pipe_write
From the code, it is unclear if that behavior is wanted but the log
says that at least Ingo Molnar aims to mimic lmbench's lat_ctx, that
doesn't handle the pipe ends that way
(https://sourceforge.net/p/lmbench/code/HEAD/tree/trunk/lmbench2/src/lat_ctx.c)
Fix worker_thread() by always first feeding the write ends of the pipes
and then trying to read.
This roughly halves the context switches and runtime of pure
'perf bench sched pipe':
Performance counter stats for 'perf bench sched pipe':
1,005,770 context-switches
6.033448041 seconds time elapsed
0.423142000 seconds user
4.519829000 seconds sys
And the blocking reads do no longer dominate the analysis at the above
extreme:
Samples: 40K of event 'cycles:P', Event count (approx.): 14309364879
Overhead Command Shared Object Symbol
12.20% sched-pipe [kernel.kallsyms] [k] read_hpet
9.23% sched-pipe [kernel.kallsyms] [k] retbleed_untrain_ret
3.68% sched-pipe [kernel.kallsyms] [k] pipe_write
Signed-off-by: Dirk Gouders <dirk@gouders.net>
Acked-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20250323140316.19027-2-dirk@gouders.net
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Commit 54f9aa1092457 ("tools/perf/powerpc/util: Add support to
handle compatible mode PVR for perf json events") introduced
to select proper JSON events in case of compat mode using
auxiliary vector. But this caused a compilation error in ppc64
Big Endian.
arch/powerpc/util/header.c: In function 'is_compat_mode':
arch/powerpc/util/header.c:20:21: error: cast to pointer from
integer of different size [-Werror=int-to-pointer-cast]
20 | if (!strcmp((char *)platform, (char *)base_platform))
| ^
arch/powerpc/util/header.c:20:39: error: cast to pointer from
integer of different size [-Werror=int-to-pointer-cast]
20 | if (!strcmp((char *)platform, (char *)base_platform))
|
Commit saved the getauxval(AT_BASE_PLATFORM) and getauxval(AT_PLATFORM)
return values in u64 which causes the compilation error.
Patch fixes this issue by changing u64 to "unsigned long".
Fixes: 54f9aa1092457 ("tools/perf/powerpc/util: Add support to handle compatible mode PVR for perf json events")
Signed-off-by: Likhitha Korrapati <likhitha@linux.ibm.com>
Reviewed-by: Athira Rajeev <atrajeev@linux.ibm.com>
Link: https://lore.kernel.org/r/20250321100726.699956-1-likhitha@linux.ibm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
When enabling the libperl feature the build uses perl's build flags
(ccopts) but filters out various flags, e.g. for LTO.
While this is conceptually correct, it is insufficient in practice,
since only "-flto=auto" is filtered out. When perl itself is built with
"-flto" this can cause parts of perf being built with LTO and others
without, giving exciting build errors like e.g.:
../tools/perf/pmu-events/pmu-events.c:72851:(.text+0xb79): undefined
reference to `strcmp_cpuid_str' collect2: error: ld returned 1 exit status
Fix this by filtering all matching flag values of -flto{=n,auto,..}.
Signed-off-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Link: https://lore.kernel.org/r/20250321082038.27901-2-holger@applied-asynchrony.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
frontend_bound metrics was miscalculated due to different scaling in
a couple of metrics it depends on. Change the scaling to match with
AmpereOne.
Fixes: 16438b652b46 ("perf vendor events arm64 AmpereOneX: Add core PMU events and metrics")
Signed-off-by: Ilkka Koskinen <ilkka@os.amperecomputing.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Link: https://lore.kernel.org/r/20250313201559.11332-3-ilkka@os.amperecomputing.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Atomic instructions are both memory-reading and memory-writing
instructions and so should be counted by both LD_RETIRED and ST_RETIRED
performance monitoring events. However LD_RETIRED does not count atomic
instructions.
Signed-off-by: Ilkka Koskinen <ilkka@os.amperecomputing.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Link: https://lore.kernel.org/r/20250313201559.11332-2-ilkka@os.amperecomputing.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leak sanitizer was reporting a memory leak in the "perf record and
replay" test. Add evlist__delete to trace__exit, also ensure
trace__exit is called after trace__record.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-15-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add missing btf__free in trace__exit.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-14-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Namhyung fixed the syscall table being reallocated and moving by
reloading the system call pointer after a move:
https://lore.kernel.org/lkml/Z9YHCzINiu4uBQ8B@google.com/
This could be brittle so this patch changes the syscall table to be an
array of pointers of "struct syscall" that don't move. Remove
unnecessary copies and searches with this change.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-13-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Now a single beauty file is generated and used by all architectures,
remove the per-architecture Makefiles, Kbuild files and previous
generator script.
Note: there was conversation with Charlie Jenkins
<charlie@rivosinc.com> and they'd written an alternate approach to
support multiple architectures:
https://lore.kernel.org/all/20250114-perf_syscall_arch_runtime-v1-1-5b304e408e11@rivosinc.com/
It would have been better to have helped Charlie fix their series (my
apologies) but they agreed that the approach taken here was likely
best for longer term maintainability:
https://lore.kernel.org/lkml/Z6Jk_UN9i69QGqUj@ghost/
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Howard Chu <howardchu95@gmail.com>
Reviewed-by: Charlie Jenkins <charlie@rivosinc.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-11-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Rather than generating individual syscall header files generate a
single trace/beauty/generated/syscalltbl.c. In a syscalltbls array
have references to each architectures tables along with the
corresponding e_machine. When the 32-bit or 64-bit table is ambiguous,
match the perf binary's type. For ARM32 don't use the arm64 32-bit
table which is smaller. EM_NONE is present for is no machine matches.
Conditionally compile the tables, only having the appropriate 32 and
64-bit table. If ALL_SYSCALLTBL is defined all tables can be
compiled.
Add comment for noreturn column suggested by Arnd Bergmann:
https://lore.kernel.org/lkml/d47c35dd-9c52-48e7-a00d-135572f11fbb@app.fastmail.com/
and added in commit 9142be9e6443 ("x86/syscall: Mark exit[_group]
syscall handlers __noreturn").
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Howard Chu <howardchu95@gmail.com>
Reviewed-by: Charlie Jenkins <charlie@rivosinc.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-9-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
First try to read the e_machine from the dsos associated with the
thread's maps. If live use the executable from /proc/pid/exe and read
the e_machine from the ELF header. On failure use EM_HOST. Change
builtin-trace syscall functions to pass e_machine from the thread
rather than EM_HOST, so that in later patches when syscalltbl can use
the e_machine the system calls are specific to the architecture.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-8-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
For ELF file dsos read the e_machine from the ELF header. For kernel
types assume the e_machine matches the perf tool. In other cases
return EM_NONE.
When reading from the ELF header use DSO__SWAP that may need
dso->needs_swap initializing. Factor out dso__swap_init to allow this.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-7-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The syscalltbl held entries of system call name and number pairs,
generated from a native syscalltbl at start up. As there are gaps in
the system call number there is a notion of index into the
table. Going forward we want the system call table to be identifiable
by a machine type, for example, i386 vs x86-64. Change the interface
to the syscalltbl so (1) a (currently unused machine type of EM_HOST)
is passed (2) the index to syscall number and system call name mapping
is computed at build time.
Two tables are used for this, an array of system call number to name,
an array of system call numbers sorted by the system call name. The
sorted array doesn't store strings in part to save memory and
relocations. The index notion is carried forward and is an index into
the sorted array of system call numbers, the data structures are
opaque (held only in syscalltbl.c), and so the number of indices for a
machine type is exposed as a new API.
The arrays are computed in the syscalltbl.sh script and so no start-up
time computation and storage is necessary.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Howard Chu <howardchu95@gmail.com>
Reviewed-by: Charlie Jenkins <charlie@rivosinc.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-6-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Identify struct syscall information in the syscalls table by a machine
type and syscall number, not just system call number. Having the
machine type means that 32-bit system calls can be differentiated from
64-bit ones on a machine capable of both. Having a table for all
machine types and all system call numbers would be too large, so
maintain a sorted array of system calls as they are encountered.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Howard Chu <howardchu95@gmail.com>
Reviewed-by: Charlie Jenkins <charlie@rivosinc.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-5-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The definition of "static const char *const syscalltbl[] = {" is done
in a generated syscalls_32.h or syscalls_64.h that is architecture
dependent. In order to include the appropriate file a syscall_table.h
is found via the perf include path and it includes the syscalls_32.h
or syscalls_64.h as appropriate.
To support having multiple syscall tables, one for 32-bit and one for
64-bit, or for different architectures, an include path cannot be
used. Remove syscall_table.h because of this and inline what it does
into syscalltbl.c.
For architectures without a syscall_table.h this will cause a failure
to include either syscalls_32.h or syscalls_64.h rather than a failure
to include syscall_table.h. For architectures that only included one
or other, the behavior matches BITS_PER_LONG as previously done on
architectures supporting both syscalls_32.h and syscalls_64.h.
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Howard Chu <howardchu95@gmail.com>
Reviewed-by: Charlie Jenkins <charlie@rivosinc.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-4-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
There are many and non-obvious meanings to the dso_binary_type enum
values. Add kernel-doc to speed interpretting their meanings.
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-3-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The variables elf_base_addr, debug_frame_offset, eh_frame_hdr_addr and
eh_frame_hdr_offset are only accessed in unwind-libunwind-local.c
which is conditionally built on having libunwind support. Make the
variables conditional on libunwind support too.
Reviewed-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: https://lore.kernel.org/r/20250319050741.269828-2-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
I've realized that it doesn't make sense to accumulate the samples to
parent in the callchain when data type profiling is enabled. Because it
won't have the same data type access in the parent. Otherwise it'd see
something like this:
$ perf report -s type --stdio -g none
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 2K of event 'cycles:Pu'
# Event count (approx.): 8266456478
#
# Children Latency Self Latency Data Type
# ........ ....... ........ ........ .........
#
698.97% 697.72% 99.80% 99.61% (unknown)
0.09% 0.18% 0.09% 0.18% Elf64_Rela
0.05% 0.10% 0.05% 0.10% unsigned char
0.05% 0.10% 0.05% 0.10% struct exit_function_list
0.00% 0.01% 0.00% 0.01% struct rtld_global
Link: https://lore.kernel.org/r/20250307080829.354947-3-namhyung@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
This is useful for hierarchy output mode where the first level is
considered as output fields. We want them in the same level so that it
can show only the remaining groups in the hierarchy.
Before:
$ perf report -s overhead,sample,period,comm,dso -H --stdio
...
# Overhead Samples / Period / Command / Shared Object
# ................. ..........................................
#
100.00% 4035
100.00% 3835883066
100.00% perf
99.37% perf
0.50% ld-linux-x86-64.so.2
0.06% [unknown]
0.04% libc.so.6
0.02% libLLVM-16.so.1
After:
$ perf report -s overhead,sample,period,comm,dso -H --stdio
...
# Overhead Samples Period Command / Shared Object
# ....................................... .......................
#
100.00% 4035 3835883066 perf
99.37% 4005 3811826223 perf
0.50% 19 19210014 ld-linux-x86-64.so.2
0.06% 8 2367089 [unknown]
0.04% 2 1720336 libc.so.6
0.02% 1 759404 libLLVM-16.so.1
Acked-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20250307080829.354947-1-namhyung@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
On linux-next
commit 72c6f57a4193 ("perf pmu: Dynamically allocate tool PMU")
allocated PMU named "tool" dynamicly. However that allocation
can fail and a NULL pointer is returned. That case is currently
not handled and would result in an invalid address reference.
Add a check for NULL pointer.
Fixes: 72c6f57a4193 ("perf pmu: Dynamically allocate tool PMU")
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Link: https://lore.kernel.org/r/20250319122820.2898333-1-tmricht@linux.ibm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
zfree() requires an address otherwise it frees what's in name, rather
than name itself. Pass the address of name to fix it.
This was the only incorrect occurrence in Perf found using a search.
Fixes: 8db5cabcf1b6 ("perf stat: Fork and launch 'perf record' when 'perf stat' needs to get retire latency value for a metric.")
Signed-off-by: James Clark <james.clark@linaro.org>
Link: https://lore.kernel.org/r/20250319101614.190922-1-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Kernel modules for which we cannot find a file on-disk will have a
dso->long_name that looks like "[module_name]". Prior to the commit
listed in the fixes, the dso->kernel field would be zero (for user
space), so dso__is_kallsyms() would return false. After the commit,
kernel module DSOs are correctly labeled, but the result is that
dso__is_kallsyms() erroneously returns true for those modules without a
filesystem path.
Later, build_id_cache__add() consults this value of is_kallsyms, and
when true, it copies /proc/kallsyms into the cache. Users with many
kernel modules without a filesystem path (e.g. ksplice or possibly
kernel live patch modules) have reported excessive disk space usage in
the build ID cache directory due to this behavior.
To reproduce the issue, it's enough to build a trivial out-of-tree hello
world kernel module, load it using insmod, and then use:
perf record -ag -- sleep 1
In the build ID directory, there will be a directory for your module
name containing a kallsyms file.
Fix this up by changing dso__is_kallsyms() to consult the
dso_binary_type enumeration, which is also symmetric to the above checks
for dso__is_vmlinux() and dso__is_kcore(). With this change, kallsyms is
not cached in the build-id cache for out-of-tree modules.
Fixes: 02213cec64bbe ("perf maps: Mark module DSOs with kernel type")
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
Link: https://lore.kernel.org/r/20250318230012.2038790-1-stephen.s.brennan@oracle.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The functionalities of {disabled,required}-features.h have been replaced with
the auto-generated generated/<asm/cpufeaturemasks.h> header.
Thus they are no longer needed and can be removed.
None of the macros defined in {disabled,required}-features.h is used in tools,
delete them too.
Signed-off-by: Xin Li (Intel) <xin@zytor.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://lore.kernel.org/r/20250305184725.3341760-4-xin@zytor.com
When s2[i] = '\0', if s1[i] != '\0', it will be judged by ret,
and if s1[i] = '\0', it will be judegd by !s1[i].
So in reality, s2 [i] will never make a judgment
Signed-off-by: Feng Yang <yangfeng@kylinos.cn>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250314031013.94480-1-yangfeng59949@163.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The pyrf_event__new() method copies the event obtained from the perf
ring buffer to a structure that will then be turned into a python object
for further consumption, so it copies perf_event.header.size bytes to
its 'event' member:
$ pahole -C pyrf_event /tmp/build/perf-tools-next/python/perf.cpython-312-x86_64-linux-gnu.so
struct pyrf_event {
PyObject ob_base; /* 0 16 */
struct evsel * evsel; /* 16 8 */
struct perf_sample sample; /* 24 312 */
/* XXX last struct has 7 bytes of padding, 2 holes */
/* --- cacheline 5 boundary (320 bytes) was 16 bytes ago --- */
union perf_event event; /* 336 4168 */
/* size: 4504, cachelines: 71, members: 4 */
/* member types with holes: 1, total: 2 */
/* paddings: 1, sum paddings: 7 */
/* last cacheline: 24 bytes */
};
$
It was doing so without checking if the event just obtained has more
than that space, fix it.
This isn't a proper, final solution, as we need to support larger
events, but for the time being we at least bounds check and document it.
Fixes: 877108e42b1b9ba6 ("perf tools: Initial python binding")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250312203141.285263-7-acme@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
When processing tracepoints the perf python binding was parsing the
event before calling perf_mmap__consume(&md->core) in
pyrf_evlist__read_on_cpu().
But part of this event parsing was to set the perf_sample->raw_data
pointer to the payload of the event, which then could be overwritten by
other event before tracepoint fields were asked for via event.prev_comm
in a python program, for instance.
This also happened with other fields, but strings were were problems
were surfacing, as there is UTF-8 validation for the potentially garbled
data.
This ended up showing up as (with some added debugging messages):
( field 'prev_comm' ret=0x7f7c31f65110, raw_size=68 ) ( field 'prev_pid' ret=0x7f7c23b1bed0, raw_size=68 ) ( field 'prev_prio' ret=0x7f7c239c0030, raw_size=68 ) ( field 'prev_state' ret=0x7f7c239c0250, raw_size=68 ) time 14771421785867 prev_comm= prev_pid=1919907691 prev_prio=796026219 prev_state=0x303a32313175 ==>
( XXX '��' len=16, raw_size=68) ( field 'next_comm' ret=(nil), raw_size=68 ) Traceback (most recent call last):
File "/home/acme/git/perf-tools-next/tools/perf/python/tracepoint.py", line 51, in <module>
main()
File "/home/acme/git/perf-tools-next/tools/perf/python/tracepoint.py", line 46, in main
event.next_comm,
^^^^^^^^^^^^^^^
AttributeError: 'perf.sample_event' object has no attribute 'next_comm'
When event.next_comm was asked for, the PyUnicode_FromString() python
API would fail and that tracepoint field wouldn't be available, stopping
the tools/perf/python/tracepoint.py test tool.
But, since we already do a copy of the whole event in pyrf_event__new,
just use it and while at it remove what was done in in e8968e654191390a
("perf python: Fix pyrf_evlist__read_on_cpu event consuming") because we
don't really need to wait for parsing the sample before declaring the
event as consumed.
This copy is questionable as is now, as it limits the maximum event +
sample_type and tracepoint payload to sizeof(union perf_event), this all
has been "working" because 'struct perf_event_mmap2', the largest entry
in 'union perf_event' is:
$ pahole -C perf_event ~/bin/perf | grep mmap2
struct perf_record_mmap2 mmap2; /* 0 4168 */
$
Fixes: bae57e3825a3dded ("perf python: Add support to resolve tracepoint fields")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250312203141.285263-6-acme@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
To avoid a leak if we have the python object but then something happens
and we need to return the operation, decrement the offset of the newly
created object.
Fixes: 377f698db12150a1 ("perf python: Add struct evsel into struct pyrf_event")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250312203141.285263-5-acme@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Otherwise when debugging we see just "python" in perf, top, etc.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250312203141.285263-4-acme@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
When python2 support was removed in e7e9943c87d857da ("perf python:
Remove python 2 scripting support"), all use of the
_PyUnicode_FromString(arg), _PyUnicode_FromFormat(...), and
_PyLong_FromLong(arg) macros was removed as well, so remove it.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250312203141.285263-3-acme@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Some old cut'n'paste error, its "ip", so the description should be
"event ip", not "event type".
Fixes: 877108e42b1b9ba6 ("perf tools: Initial python binding")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250312203141.285263-2-acme@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The DSO data read test opens a file but as dsos__exit is used the test
file isn't closed. This causes the subsequent subtests in don't fork
(-F) mode to fail as one more than expected file descriptor is open.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250318043151.137973-4-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
dso__list_del with address sanitizer and/or reference count checking
will call dso__put that can call dso__data_close reentrantly trying to
lock the dso__data_open_lock and deadlocking. Switch from pthread
mutexes to perf's mutex so that lock checking is performed in debug
builds. Add lock annotations that diagnosed the problem. Release the
dso__data_open_lock around the dso__put to avoid the deadlock.
Change the declaration of dso__data_get_fd to return a boolean,
indicating the fd is valid and the lock is held, to make it compatible
with the thread safety annotations as a try lock.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250318043151.137973-3-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Used to annotate when locks shouldn't be held for a function or if a
function returns a lock that's used by later mutex lock unlock
operations.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250318043151.137973-2-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Parameterize the basic testing to generate directly a perf.data file
or to generate/use one from pipe input or output. To simplify the
refactor move some of the head/grep logic around. Use "-q" with grep
to make the test output cleaner.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250311211635.541090-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
When make_data fails its error message needs to go to stderr rather
than stdout and the stdout value is captured in a variable. Quote the
$err value so that it is always a valid input for test. This error is
commonly encountered if no sample data is gathered by the test.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250312001841.1515779-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The dc6d2bc2d893a878 ("perf sample: Make user_regs and intr_regs optional") misses
the changes to a file, resulting in this problem:
$ make LIBUNWIND=1 -C tools/perf O=/tmp/build/perf-tools-next install-bin
<SNIP>
CC /tmp/build/perf-tools-next/util/unwind-libunwind-local.o
CC /tmp/build/perf-tools-next/util/unwind-libunwind.o
<SNIP>
util/unwind-libunwind-local.c: In function ‘access_mem’:
util/unwind-libunwind-local.c:582:56: error: ‘ui->sample->user_regs’ is a pointer; did you mean to use ‘->’?
582 | if (__write || !stack || !ui->sample->user_regs.regs) {
| ^
| ->
util/unwind-libunwind-local.c:587:38: error: passing argument 2 of ‘perf_reg_value’ from incompatible pointer type [-Wincompatible-pointer-types]
587 | ret = perf_reg_value(&start, &ui->sample->user_regs,
| ^~~~~~~~~~~~~~~~~~~~~~
| |
| struct regs_dump **
<SNIP>
⬢ [acme@toolbox perf-tools-next]$ git bisect bad
dc6d2bc2d893a878e7b58578ff01b4738708deb4 is the first bad commit
commit dc6d2bc2d893a878e7b58578ff01b4738708deb4 (HEAD)
Author: Ian Rogers <irogers@google.com>
Date: Mon Jan 13 11:43:45 2025 -0800
perf sample: Make user_regs and intr_regs optional
Detected using:
make -C tools/perf build-test
Fixes: dc6d2bc2d893a878 ("perf sample: Make user_regs and intr_regs optional")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250313033121.758978-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
