We also need to add SYM_TYPED_FUNC_START() to util/include/linux/linkage.h
and update tools/perf/check_headers.sh to ignore the include cfi_types.h
line when checking if the kernel original files drifted from the copies
we carry.
This is to get the changes from:
ccace936ee ("x86: Add types to indirectly called assembly functions")
Addressing these tools/perf build warnings:
Warning: Kernel ABI header at 'tools/arch/x86/lib/memcpy_64.S' differs from latest version at 'arch/x86/lib/memcpy_64.S'
diff -u tools/arch/x86/lib/memcpy_64.S arch/x86/lib/memcpy_64.S
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/lkml/Y1f3VRIec9EBgX6F@kernel.org/
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
During the transition to libbpf 1.0 some functions that perf used were
deprecated and finally removed from libbpf, so bpf_program__set_insns()
was introduced for perf to continue to use its bpf loader.
But when build with LIBBPF_DYNAMIC=1 we now need to check if that
function is available so that perf can build with older libbpf versions,
even if the end result is emitting a warning to the user that the use
of the perf BPF loader requires a newer libbpf, since bpf_program__set_insns()
touches libbpf objects internal state.
This affects only 'perf trace' when using bpf C code or pre-compiled
bytecode as an event.
Noticed on RHEL9, that has libbpf 0.7.0, where bpf_program__set_insns()
isn't available.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The bpf_load_program() prototype appeared in tools/lib/bpf/bpf.h as
deprecated, but nowadays its completely removed, so add it back for
building with the system libbpf when using 'make LIBBPF_DYNAMIC=1'.
This is a stop gap hack till we do like tools/bpf does with bpftool,
i.e. bootstrap the libbpf build and install it in the perf build
directory when not using 'make LIBBPF_DYNAMIC=1'.
That has to be done to all libraries in tools/lib/, so tha we can
remove -Itools/lib/ from the tools/perf CFLAGS.
Noticed when building with LIBBPF_DYNAMIC=1 and libbpf 0.7.0 on RHEL9.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
If the kernel exposes a new perf_event_attr field in a format attr, perf
will return an error stating the specified PMU can't be found. For
example, a format attr with 'config3:0-63' causes an error as config3 is
unknown to perf. This causes a compatibility issue between a newer
kernel with older perf tool.
Before this change with a kernel adding 'config3' I get:
$ perf record -e arm_spe// -- true
event syntax error: 'arm_spe//'
\___ Cannot find PMU `arm_spe'. Missing kernel support?
Run 'perf list' for a list of valid events
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-e, --event <event> event selector. use 'perf list' to list
available events
After this change, I get:
$ perf record -e arm_spe// -- true
WARNING: 'arm_spe_0' format 'inv_event_filter' requires 'perf_event_attr::config3' which is not supported by this version of perf!
[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.091 MB perf.data ]
To support unknown configN formats, rework the YACC implementation to
pass any config[0-9]+ format to perf_pmu__new_format() to handle with a
warning.
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Rob Herring <robh@kernel.org>
Tested-by: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@arm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20220914-arm-perf-tool-spe1-2-v2-v4-1-83c098e6212e@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf tools updates from Arnaldo Carvalho de Melo:
- Add support for AMD on 'perf mem' and 'perf c2c', the kernel
enablement patches went via tip.
Example:
$ sudo perf mem record -- -c 10000
^C[ perf record: Woken up 227 times to write data ]
[ perf record: Captured and wrote 58.760 MB perf.data (836978 samples) ]
$ sudo perf mem report -F mem,sample,snoop
Samples: 836K of event 'ibs_op//', Event count (approx.): 8418762
Memory access Samples Snoop
N/A 700620 N/A
L1 hit 126675 N/A
L2 hit 424 N/A
L3 hit 664 HitM
L3 hit 10 N/A
Local RAM hit 2 N/A
Remote RAM (1 hop) hit 8558 N/A
Remote Cache (1 hop) hit 3 N/A
Remote Cache (1 hop) hit 2 HitM
Remote Cache (2 hops) hit 10 HitM
Remote Cache (2 hops) hit 6 N/A
Uncached hit 4 N/A
$
- "perf lock" improvements:
- Add -E/--entries option to limit the number of entries to
display, say to ask for just the top 5 contended locks.
- Add -q/--quiet option to suppress header and debug messages.
- Add a 'perf test' kernel lock contention entry to test 'perf
lock'.
- "perf lock contention" improvements:
- Ask BPF's bpf_get_stackid() to skip some callchain entries.
The ones closer to the tooling are bpf related and not that
interesting, the ones calling the locking function are the ones
we're interested in, example of a full, unskipped callstack:
- Allow changing the callstack depth and number of entries to skip.
1 10.74 us 10.74 us 10.74 us spinlock __bpf_trace_contention_begin+0xb
0xffffffffc03b5c47 bpf_prog_bf07ae9e2cbd02c5_contention_begin+0x117
0xffffffffc03b5c47 bpf_prog_bf07ae9e2cbd02c5_contention_begin+0x117
0xffffffffbb8b8e75 bpf_trace_run2+0x35
0xffffffffbb7eab9b __bpf_trace_contention_begin+0xb
0xffffffffbb7ebe75 queued_spin_lock_slowpath+0x1f5
0xffffffffbc1c26ff _raw_spin_lock+0x1f
0xffffffffbb841015 tick_do_update_jiffies64+0x25
0xffffffffbb8409ee tick_irq_enter+0x9e
- Show full callstack in verbose mode (-v option), sometimes this
is desirable instead of showing just one callstack entry.
- Allow multiple time ranges in 'perf record --delay' to help in
reducing the amount of data collected from hardware tracing (Intel
PT, etc) when there is a rough idea of periods of time where events
of interest take time.
- Add Intel PT to record only decoder debug messages when error
happens.
- Improve layout of Intel PT man page.
- Add new branch types: alignment, data and inst faults and arch
specific ones, such as fiq, debug_halt, debug_exit, debug_inst and
debug_data on arm64.
Kernel enablement went thru the tip tree.
- Fix 'perf probe' error log check in 'perf test' when no debuginfo is
available.
- Fix 'perf stat' aggregation mode logic, it should be looking at the
CPU not at the core number.
- Fix flags parsing in 'perf trace' filters.
- Introduce compact encoding of CPU range encoding on perf.data, to
avoid having a bitmap with all the CPUs.
- Improvements to the 'perf stat' metrics, including adding
"core_wide", and computing "smt" from the CPU topology.
- Add support to the new PERF_FORMAT_LOST perf_event_attr.read_format,
that allows tooling to ask for the precise number of lost samples for
a given event.
- Add 'addr' sort key to see just the address of sampled instructions:
$ perf record -o- true | perf report -i- -s addr
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.000 MB - ]
# Samples: 12 of event 'cycles:u'
# Event count (approx.): 252512
#
# Overhead Address
# ........ ..................
42.96% 0x7f96f08443d7
29.55% 0x7f96f0859b50
14.76% 0x7f96f0852e02
8.30% 0x7f96f0855028
4.43% 0xffffffff8de01087
perf annotate: Toggle full address <-> offset display
- Add 'f' hotkey to the 'perf annotate' TUI interface when in
'disassembler output' mode ('o' hotkey) to toggle showing full
virtual address or just the offset.
- Cache DSO build-ids when synthesizing PERF_RECORD_MMAP records for
pre-existing threads, at the start of a 'perf record' session,
speeding up that record startup phase.
- Add a command line option to specify build ids in 'perf inject'.
- Update JSON event files for the Intel alderlake, broadwell,
broadwellde, broadwellx, cascadelakex, haswell, haswellx, icelake,
icelakex, ivybridge, ivytown, jaketown, sandybridge, sapphirerapids,
skylake, skylakex, and tigerlake processors.
- Update vendor JSON event files for the ARM Neoverse V1 and E1
platforms.
- Add a 'perf test' entry for 'perf mem' where a struct has false
sharing and this gets detected in the 'perf mem' output, tested with
Intel, AMD and ARM64 systems.
- Add a 'perf test' entry to test the resolution of java symbols, where
an output like this is expected:
8.18% jshell jitted-50116-29.so [.] Interpreter
0.75% Thread-1 jitted-83602-1670.so [.] jdk.internal.jimage.BasicImageReader.getString(int)
- Add tests for the ARM64 CoreSight hardware tracing feature, with
specially crafted pureloop, memcpy, thread loop and unroll tread that
then gets traced and the output compared with expected output.
Documentation explaining it is also included.
- Add per thread Intel PT 'perf test' entry to check that
PERF_RECORD_TEXT_POKE events are recorded per CPU, resulting in a
mixture of per thread and per CPU events and mmaps, verify that this
gets all recorded correctly.
- Introduce pthread mutex wrappers to allow for building with clang's
-Wthread-safety, i.e. using the "guarded_by" "pt_guarded_by"
"lockable", "exclusive_lock_function", "exclusive_trylock_function",
"exclusive_locks_required", and "no_thread_safety_analysis" compiler
function attributes.
- Fix empty version number when building outside of a git repo.
- Improve feature detection display when multiple versions of a feature
are present, such as for binutils libbfd, that has a mix of possible
ways to detect according to the Linux distribution.
Previously in some cases we had:
Auto-detecting system features
<SNIP>
... libbfd: [ on ]
... libbfd-liberty: [ on ]
... libbfd-liberty-z: [ on ]
<SNIP>
Now for this case we show just the main feature:
Auto-detecting system features
<SNIP>
... libbfd: [ on ]
<SNIP>
- Remove some unused structs, variables, macros, function prototypes
and includes from various places.
* tag 'perf-tools-for-v6.1-1-2022-10-07' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (169 commits)
perf script: Add missing fields in usage hint
perf mem: Print "LFB/MAB" for PERF_MEM_LVLNUM_LFB
perf mem/c2c: Avoid printing empty lines for unsupported events
perf mem/c2c: Add load store event mappings for AMD
perf mem/c2c: Set PERF_SAMPLE_WEIGHT for LOAD_STORE events
perf mem: Add support for printing PERF_MEM_LVLNUM_{CXL|IO}
perf amd ibs: Sync arch/x86/include/asm/amd-ibs.h header with the kernel
tools headers UAPI: Sync include/uapi/linux/perf_event.h header with the kernel
perf stat: Fix cpu check to use id.cpu.cpu in aggr_printout()
perf test coresight: Add relevant documentation about ARM64 CoreSight testing
perf test: Add git ignore for tmp and output files of ARM CoreSight tests
perf test coresight: Add unroll thread test shell script
perf test coresight: Add unroll thread test tool
perf test coresight: Add thread loop test shell scripts
perf test coresight: Add thread loop test tool
perf test coresight: Add memcpy thread test shell script
perf test coresight: Add memcpy thread test tool
perf test: Add git ignore for perf data generated by the ARM CoreSight tests
perf test: Add arm64 asm pureloop test shell script
perf test: Add asm pureloop test tool
...
Pull cgroup updates from Tejun Heo:
- cpuset now support isolated cpus.partition type, which will enable
dynamic CPU isolation
- pids.peak added to remember the max number of pids used
- holes in cgroup namespace plugged
- internal cleanups
* tag 'cgroup-for-6.1' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup: (25 commits)
cgroup: use strscpy() is more robust and safer
iocost_monitor: reorder BlkgIterator
cgroup: simplify code in cgroup_apply_control
cgroup: Make cgroup_get_from_id() prettier
cgroup/cpuset: remove unreachable code
cgroup: Remove CFTYPE_PRESSURE
cgroup: Improve cftype add/rm error handling
kselftest/cgroup: Add cpuset v2 partition root state test
cgroup/cpuset: Update description of cpuset.cpus.partition in cgroup-v2.rst
cgroup/cpuset: Make partition invalid if cpumask change violates exclusivity rule
cgroup/cpuset: Relocate a code block in validate_change()
cgroup/cpuset: Show invalid partition reason string
cgroup/cpuset: Add a new isolated cpus.partition type
cgroup/cpuset: Relax constraints to partition & cpus changes
cgroup/cpuset: Allow no-task partition to have empty cpuset.cpus.effective
cgroup/cpuset: Miscellaneous cleanups & add helper functions
cgroup/cpuset: Enable update_tasks_cpumask() on top_cpuset
cgroup: add pids.peak interface for pids controller
cgroup: Remove data-race around cgrp_dfl_visible
cgroup: Fix build failure when CONFIG_SHRINKER_DEBUG
...
When perf is built in a full source tree that is not a git repository,
e.g. from a kernel source tarball, `perf version` will print empty tag
and commit strings:
$ perf version
perf version
Currently the tag version is only generated from the root Makefile when
building in a git repository. If PERF-VERSION-FILE has not been
generated and the source tree is not in a git repository, then
PERF-VERSION-GEN will return an empty version.
The problem can be reproduced with the following steps:
$ wget https://git.kernel.org/torvalds/t/linux-6.0-rc7.tar.gz
$ tar -xf linux-6.0-rc7.tar.gz && cd linux-6.0-rc7
$ make -C tools/perf
$ tools/perf/perf -v
perf version
Builds from tarballs generated with `make perf-tar-src-pkg` are not
impacted by this issue as PERF-VERSION-FILE is included in the archive.
The perf RPM provided by Fedora for 5.18+ is experiencing this problem.
Package build logs[0] show that the build is attempting to fall back on
PERF-VERSION-FILE, but it is not present.
To resolve this, revert back to the previous logic of using the kernel
Makefile version if not in a git repository and PERF-VERSION-FILE does
not exist.
[0] https://kojipkgs.fedoraproject.org/packages/kernel-tools/5.19.4/200.fc36/data/logs/x86_64/build.log
Fixes: 7572733b84 ("perf tools: Fix version kernel tag")
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Will Chandler <wfc@wfchandler.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: John Garry <john.garry@huawei.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20220930151157.529674-1-wfc@wfchandler.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Sometimes users want to see actual (virtual) address of sampled instructions.
Add a new 'addr' sort key to display the raw addresses.
$ perf record -o- true | perf report -i- -s addr
# To display the perf.data header info, please use --header/--header-only options.
#
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.000 MB - ]
#
# Total Lost Samples: 0
#
# Samples: 12 of event 'cycles:u'
# Event count (approx.): 252512
#
# Overhead Address
# ........ ..................
#
42.96% 0x7f96f08443d7
29.55% 0x7f96f0859b50
14.76% 0x7f96f0852e02
8.30% 0x7f96f0855028
4.43% 0xffffffff8de01087
Note that it just compares and displays the sample ip. Each process can
have a different memory layout and the ip will be different even if they run
the same binary. So this sort key is mostly meaningful for per-process
profile data.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: https://lore.kernel.org/r/20220923173142.805896-4-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Currently it collects stack traces to max size then skip entries.
Because we don't have control how to skip perf callchains. But BPF can
do it with bpf_get_stackid() with a flag.
Say we have max-stack=4 and stack-skip=2, we get these stack traces.
Before: After:
.---> +---+ <--. .---> +---+ <--.
| | | | | | | |
| +---+ usable | +---+ |
max | | | max | | |
stack +---+ <--' stack +---+ usable
| | X | | | | |
| +---+ skip | +---+ |
| | X | | | | |
`---> +---+ `---> +---+ <--' <=== collection
| X |
+---+ skip
| X |
+---+
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Cc: bpf@vger.kernel.org
Link: https://lore.kernel.org/r/20220912055314.744552-5-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It needs stack traces to find callers of locks. To minimize the
performance overhead it only collects up to 8 entries for each stack
trace. And it skips first 3 entries as they came from BPF, tracepoint
and lock functions which are not interested for most users.
But it turned out that those numbers are different in some
configuration. Using fixed number can result in non meaningful caller
names. Let's make them adjustable with --stack-depth and --skip-stack
options.
On my setup, the default output is like below:
# /perf lock con -ab -F contended,wait_total sleep 3
contended total wait type caller
28 4.55 ms rwlock:W __bpf_trace_contention_begin+0xb
33 1.67 ms rwlock:W __bpf_trace_contention_begin+0xb
12 580.28 us spinlock __bpf_trace_contention_begin+0xb
60 240.54 us rwsem:R __bpf_trace_contention_begin+0xb
27 64.45 us spinlock __bpf_trace_contention_begin+0xb
If I change the stack skip to 5, the result will be like:
# perf lock con -ab -F contended,wait_total --stack-skip 5 sleep 3
contended total wait type caller
32 715.45 us spinlock folio_lruvec_lock_irqsave+0x61
26 550.22 us spinlock folio_lruvec_lock_irqsave+0x61
15 486.93 us rwsem:R mmap_read_lock+0x13
12 139.66 us rwsem:W vm_mmap_pgoff+0x93
1 7.04 us spinlock tick_do_update_jiffies64+0x25
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Cc: bpf@vger.kernel.org
Link: https://lore.kernel.org/r/20220912055314.744552-4-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Currently it shows a caller function for each entry, but users need to see
the full call stacks sometimes. Use -v/--verbose option to do that.
# perf lock con -a -b -v sleep 3
Looking at the vmlinux_path (8 entries long)
symsrc__init: cannot get elf header.
Using /proc/kcore for kernel data
Using /proc/kallsyms for symbols
contended total wait max wait avg wait type caller
1 10.74 us 10.74 us 10.74 us spinlock __bpf_trace_contention_begin+0xb
0xffffffffc03b5c47 bpf_prog_bf07ae9e2cbd02c5_contention_begin+0x117
0xffffffffc03b5c47 bpf_prog_bf07ae9e2cbd02c5_contention_begin+0x117
0xffffffffbb8b8e75 bpf_trace_run2+0x35
0xffffffffbb7eab9b __bpf_trace_contention_begin+0xb
0xffffffffbb7ebe75 queued_spin_lock_slowpath+0x1f5
0xffffffffbc1c26ff _raw_spin_lock+0x1f
0xffffffffbb841015 tick_do_update_jiffies64+0x25
0xffffffffbb8409ee tick_irq_enter+0x9e
1 7.70 us 7.70 us 7.70 us spinlock __bpf_trace_contention_begin+0xb
0xffffffffc03b5c47 bpf_prog_bf07ae9e2cbd02c5_contention_begin+0x117
0xffffffffc03b5c47 bpf_prog_bf07ae9e2cbd02c5_contention_begin+0x117
0xffffffffbb8b8e75 bpf_trace_run2+0x35
0xffffffffbb7eab9b __bpf_trace_contention_begin+0xb
0xffffffffbb7ebe75 queued_spin_lock_slowpath+0x1f5
0xffffffffbc1c26ff _raw_spin_lock+0x1f
0xffffffffbb7bc27e raw_spin_rq_lock_nested+0xe
0xffffffffbb7cef9c load_balance+0x66c
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Cc: bpf@vger.kernel.org
Link: https://lore.kernel.org/r/20220912055314.744552-3-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
