| #
f085df1b |
| 07-May-2023 |
Linus Torvalds <torvalds@linux-foundation.org> |
Merge tag 'perf-tools-for-v6.4-3-2023-05-06' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tool updates from Arnaldo Carvalho de Melo:
"Third version of perf tool updates, w
Merge tag 'perf-tools-for-v6.4-3-2023-05-06' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tool updates from Arnaldo Carvalho de Melo:
"Third version of perf tool updates, with the build problems with with
using a 'vmlinux.h' generated from the main build fixed, and the bpf
skeleton build disabled by default.
Build:
- Require libtraceevent to build, one can disable it using
NO_LIBTRACEEVENT=1.
It is required for tools like 'perf sched', 'perf kvm', 'perf
trace', etc.
libtraceevent is available in most distros so installing
'libtraceevent-devel' should be a one-time event to continue
building perf as usual.
Using NO_LIBTRACEEVENT=1 produces tooling that is functional and
sufficient for lots of users not interested in those libtraceevent
dependent features.
- Allow Python support in 'perf script' when libtraceevent isn't
linked, as not all features requires it, for instance Intel PT does
not use tracepoints.
- Error if the python interpreter needed for jevents to work isn't
available and NO_JEVENTS=1 isn't set, preventing a build without
support for JSON vendor events, which is a rare but possible
condition. The two check error messages:
$(error ERROR: No python interpreter needed for jevents generation. Install python or build with NO_JEVENTS=1.)
$(error ERROR: Python interpreter needed for jevents generation too old (older than 3.6). Install a newer python or build with NO_JEVENTS=1.)
- Make libbpf 1.0 the minimum required when building with out of
tree, distro provided libbpf.
- Use libsdtc++'s and LLVM's libcxx's __cxa_demangle, a portable C++
demangler, add 'perf test' entry for it.
- Make binutils libraries opt in, as distros disable building with it
due to licensing, they were used for C++ demangling, for instance.
- Switch libpfm4 to opt-out rather than opt-in, if libpfm-devel (or
equivalent) isn't installed, we'll just have a build warning:
Makefile.config:1144: libpfm4 not found, disables libpfm4 support. Please install libpfm4-dev
- Add a feature test for scandirat(), that is not implemented so far
in musl and uclibc, disabling features that need it, such as
scanning for tracepoints in /sys/kernel/tracing/events.
perf BPF filters:
- New feature where BPF can be used to filter samples, for instance:
$ sudo ./perf record -e cycles --filter 'period > 1000' true
$ sudo ./perf script
perf-exec 2273949 546850.708501: 5029 cycles: ffffffff826f9e25 finish_wait+0x5 ([kernel.kallsyms])
perf-exec 2273949 546850.708508: 32409 cycles: ffffffff826f9e25 finish_wait+0x5 ([kernel.kallsyms])
perf-exec 2273949 546850.708526: 143369 cycles: ffffffff82b4cdbf xas_start+0x5f ([kernel.kallsyms])
perf-exec 2273949 546850.708600: 372650 cycles: ffffffff8286b8f7 __pagevec_lru_add+0x117 ([kernel.kallsyms])
perf-exec 2273949 546850.708791: 482953 cycles: ffffffff829190de __mod_memcg_lruvec_state+0x4e ([kernel.kallsyms])
true 2273949 546850.709036: 501985 cycles: ffffffff828add7c tlb_gather_mmu+0x4c ([kernel.kallsyms])
true 2273949 546850.709292: 503065 cycles: 7f2446d97c03 _dl_map_object_deps+0x973 (/usr/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2)
- In addition to 'period' (PERF_SAMPLE_PERIOD), the other
PERF_SAMPLE_ can be used for filtering, and also some other sample
accessible values, from tools/perf/Documentation/perf-record.txt:
Essentially the BPF filter expression is:
<term> <operator> <value> (("," | "||") <term> <operator> <value>)*
The <term> can be one of:
ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
mem_dtlb, mem_blk, mem_hops
The <operator> can be one of:
==, !=, >, >=, <, <=, &
The <value> can be one of:
<number> (for any term)
na, load, store, pfetch, exec (for mem_op)
l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
remote (for mem_remote)
na, locked (for mem_locked)
na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
na, by_data, by_addr (for mem_blk)
hops0, hops1, hops2, hops3 (for mem_hops)
perf lock contention:
- Show lock type with address.
- Track and show mmap_lock, siglock and per-cpu rq_lock with address.
This is done for mmap_lock by following the current->mm pointer:
$ sudo ./perf lock con -abl -- sleep 10
contended total wait max wait avg wait address symbol
...
16344 312.30 ms 2.22 ms 19.11 us ffff8cc702595640
17686 310.08 ms 1.49 ms 17.53 us ffff8cc7025952c0
3 84.14 ms 45.79 ms 28.05 ms ffff8cc78114c478 mmap_lock
3557 76.80 ms 68.75 us 21.59 us ffff8cc77ca3af58
1 68.27 ms 68.27 ms 68.27 ms ffff8cda745dfd70
9 54.53 ms 7.96 ms 6.06 ms ffff8cc7642a48b8 mmap_lock
14629 44.01 ms 60.00 us 3.01 us ffff8cc7625f9ca0
3481 42.63 ms 140.71 us 12.24 us ffffffff937906ac vmap_area_lock
16194 38.73 ms 42.15 us 2.39 us ffff8cd397cbc560
11 38.44 ms 10.39 ms 3.49 ms ffff8ccd6d12fbb8 mmap_lock
1 5.43 ms 5.43 ms 5.43 ms ffff8cd70018f0d8
1674 5.38 ms 422.93 us 3.21 us ffffffff92e06080 tasklist_lock
581 4.51 ms 130.68 us 7.75 us ffff8cc9b1259058
5 3.52 ms 1.27 ms 703.23 us ffff8cc754510070
112 3.47 ms 56.47 us 31.02 us ffff8ccee38b3120
381 3.31 ms 73.44 us 8.69 us ffffffff93790690 purge_vmap_area_lock
255 3.19 ms 36.35 us 12.49 us ffff8d053ce30c80
- Update default map size to 16384.
- Allocate single letter option -M for --map-nr-entries, as it is
proving being frequently used.
- Fix struct rq lock access for older kernels with BPF's CO-RE
(Compile once, run everywhere).
- Fix problems found with MSAn.
perf report/top:
- Add inline information when using --call-graph=fp or lbr, as was
already done to the --call-graph=dwarf callchain mode.
- Improve the 'srcfile' sort key performance by really using an
optimization introduced in 6.2 for the 'srcline' sort key that
avoids calling addr2line for comparision with each sample.
perf sched:
- Make 'perf sched latency/map/replay' to use "sched:sched_waking"
instead of "sched:sched_waking", consistent with 'perf record'
since d566a9c2d482 ("perf sched: Prefer sched_waking event when it
exists").
perf ftrace:
- Make system wide the default target for latency subcommand, run the
following command then generate some network traffic and press
control+C:
# perf ftrace latency -T __kfree_skb
^C
DURATION | COUNT | GRAPH |
0 - 1 us | 27 | ############# |
1 - 2 us | 22 | ########### |
2 - 4 us | 8 | #### |
4 - 8 us | 5 | ## |
8 - 16 us | 24 | ############ |
16 - 32 us | 2 | # |
32 - 64 us | 1 | |
64 - 128 us | 0 | |
128 - 256 us | 0 | |
256 - 512 us | 0 | |
512 - 1024 us | 0 | |
1 - 2 ms | 0 | |
2 - 4 ms | 0 | |
4 - 8 ms | 0 | |
8 - 16 ms | 0 | |
16 - 32 ms | 0 | |
32 - 64 ms | 0 | |
64 - 128 ms | 0 | |
128 - 256 ms | 0 | |
256 - 512 ms | 0 | |
512 - 1024 ms | 0 | |
1 - ... s | 0 | |
#
perf top:
- Add --branch-history (LBR: Last Branch Record) option, just like
already available for 'perf record'.
- Fix segfault in thread__comm_len() where thread->comm was being
used outside thread->comm_lock.
perf annotate:
- Allow configuring objdump and addr2line in ~/.perfconfig., so that
you can use alternative binaries, such as llvm's.
perf kvm:
- Add TUI mode for 'perf kvm stat report'.
Reference counting:
- Add reference count checking infrastructure to check for use after
free, done to the 'cpumap', 'namespaces', 'maps' and 'map' structs,
more to come.
To build with it use -DREFCNT_CHECKING=1 in the make command line
to build tools/perf. Documented at:
https://perf.wiki.kernel.org/index.php/Reference_Count_Checking
- The above caught, for instance, fix, present in this series:
- Fix maps use after put in 'perf test "Share thread maps"':
'maps' is copied from leader, but the leader is put on line 79
and then 'maps' is used to read the reference count below - so
a use after put, with the put of maps happening within
thread__put.
Fixed by reversing the order of puts so that the leader is put
last.
- Also several fixes were made to places where reference counts were
not being held.
- Make this one of the tests in 'make -C tools/perf build-test' to
regularly build test it and to make sure no direct access to the
reference counted structs are made, doing that via accessors to
check the validity of the struct pointer.
ARM64:
- Fix 'perf report' segfault when filtering coresight traces by
sparse lists of CPUs.
- Add support for 'simd' as a sort field for 'perf report', to show
ARM's NEON SIMD's predicate flags: "partial" and "empty".
arm64 vendor events:
- Add N1 metrics.
Intel vendor events:
- Add graniterapids, grandridge and sierraforrest events.
- Refresh events for: alderlake, aldernaken, broadwell, broadwellde,
broadwellx, cascadelakx, haswell, haswellx, icelake, icelakex,
jaketown, meteorlake, knightslanding, sandybridge, sapphirerapids,
silvermont, skylake, tigerlake and westmereep-dp
- Refresh metrics for alderlake-n, broadwell, broadwellde,
broadwellx, haswell, haswellx, icelakex, ivybridge, ivytown and
skylakex.
perf stat:
- Implement --topdown using JSON metrics.
- Add TopdownL1 JSON metric as a default if present, but disable it
for now for some Intel hybrid architectures, a series of patches
addressing this is being reviewed and will be submitted for v6.5.
- Use metrics for --smi-cost.
- Update topdown documentation.
Vendor events (JSON) infrastructure:
- Add support for computing and printing metric threshold values. For
instance, here is one found in thesapphirerapids json file:
{
"BriefDescription": "Percentage of cycles spent in System Management Interrupts.",
"MetricExpr": "((msr@aperf@ - cycles) / msr@aperf@ if msr@smi@ > 0 else 0)",
"MetricGroup": "smi",
"MetricName": "smi_cycles",
"MetricThreshold": "smi_cycles > 0.1",
"ScaleUnit": "100%"
},
- Test parsing metric thresholds with the fake PMU in 'perf test
pmu-events'.
- Support for printing metric thresholds in 'perf list'.
- Add --metric-no-threshold option to 'perf stat'.
- Add rand (reverse and) and has_pmem (optane memory) support to
metrics.
- Sort list of input files to avoid depending on the order from
readdir() helping in obtaining reproducible builds.
S/390:
- Add common metrics: - CPI (cycles per instruction), prbstate (ratio
of instructions executed in problem state compared to total number
of instructions), l1mp (Level one instruction and data cache misses
per 100 instructions).
- Add cache metrics for z13, z14, z15 and z16.
- Add metric for TLB and cache.
ARM:
- Add raw decoding for SPE (Statistical Profiling Extension) v1.3 MTE
(Memory Tagging Extension) and MOPS (Memory Operations) load/store.
Intel PT hardware tracing:
- Add event type names UINTR (User interrupt delivered) and UIRET
(Exiting from user interrupt routine), documented in table 32-50
"CFE Packet Type and Vector Fields Details" in the Intel Processor
Trace chapter of The Intel SDM Volume 3 version 078.
- Add support for new branch instructions ERETS and ERETU.
- Fix CYC timestamps after standalone CBR
ARM CoreSight hardware tracing:
- Allow user to override timestamp and contextid settings.
- Fix segfault in dso lookup.
- Fix timeless decode mode detection.
- Add separate decode paths for timeless and per-thread modes.
auxtrace:
- Fix address filter entire kernel size.
Miscellaneous:
- Fix use-after-free and unaligned bugs in the PLT handling routines.
- Use zfree() to reduce chances of use after free.
- Add missing 0x prefix for addresses printed in hexadecimal in 'perf
probe'.
- Suppress massive unsupported target platform errors in the unwind
code.
- Fix return incorrect build_id size in elf_read_build_id().
- Fix 'perf scripts intel-pt-events.py' IPC output for Python 2 .
- Add missing new parameter in kfree_skb tracepoint to the python
scripts using it.
- Add 'perf bench syscall fork' benchmark.
- Add support for printing PERF_MEM_LVLNUM_UNC (Uncached access) in
'perf mem'.
- Fix wrong size expectation for perf test 'Setup struct
perf_event_attr' caused by the patch adding
perf_event_attr::config3.
- Fix some spelling mistakes"
* tag 'perf-tools-for-v6.4-3-2023-05-06' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (365 commits)
Revert "perf build: Make BUILD_BPF_SKEL default, rename to NO_BPF_SKEL"
Revert "perf build: Warn for BPF skeletons if endian mismatches"
perf metrics: Fix SEGV with --for-each-cgroup
perf bpf skels: Stop using vmlinux.h generated from BTF, use subset of used structs + CO-RE
perf stat: Separate bperf from bpf_profiler
perf test record+probe_libc_inet_pton: Fix call chain match on x86_64
perf test record+probe_libc_inet_pton: Fix call chain match on s390
perf tracepoint: Fix memory leak in is_valid_tracepoint()
perf cs-etm: Add fix for coresight trace for any range of CPUs
perf build: Fix unescaped # in perf build-test
perf unwind: Suppress massive unsupported target platform errors
perf script: Add new parameter in kfree_skb tracepoint to the python scripts using it
perf script: Print raw ip instead of binary offset for callchain
perf symbols: Fix return incorrect build_id size in elf_read_build_id()
perf list: Modify the warning message about scandirat(3)
perf list: Fix memory leaks in print_tracepoint_events()
perf lock contention: Rework offset calculation with BPF CO-RE
perf lock contention: Fix struct rq lock access
perf stat: Disable TopdownL1 on hybrid
perf stat: Avoid SEGV on counter->name
...
show more ...
|
| #
aa4800e3 |
| 16-Dec-2022 |
Linus Torvalds <torvalds@linux-foundation.org> |
Merge tag 'perf-tools-for-v6.2-1-2022-12-16' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tools updates from Arnaldo Carvalho de Melo:
"Libraries:
- Drop the old copy o
Merge tag 'perf-tools-for-v6.2-1-2022-12-16' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tools updates from Arnaldo Carvalho de Melo:
"Libraries:
- Drop the old copy of libtraceevent in tools/lib/traceevent/ now
that all major distros ship it from its external repository.
This is now just another feature detection, emitting a warning when
the libtraceevent-dev[el] package isn't installed, disabling the
build of perf features and tools that strictly require parsing
things from tracefs while keeping the core functionality present
and working with a subset of the events, the most used ones like
CPU cycles, hardware cache and also vendor events, etc.
This was tested with lots of containers for Fedora, Debian,
OpenSUSE, Alpine Linux, Ubuntu, with cross builds, etc.
Build:
- Update to C standard to gnu11, like was done for the kernel.
- Install the tools/lib/ libraries locally instead of having headers
searched directly from the source code directories, to help the
cases where we can build either from in-kernel source libraries or
from the same library shipped as a distro package, as is the case
with libbpf and was the case with libtraceevent.
perf stat:
- Do not delay the workload with --delay, the delay is just for
starting to count the events, to skip noise at workload startup.
- When we have events for each cgroup, the metric should be printed
for each cgroup separately.
$ perf stat -a --for-each-cgroup system.slice,user.slice --metric-only sleep 1
Performance counter stats for 'system wide':
GHz insn per cycle branch-misses of all branches
system.slice 3.792 0.61 3.24%
user.slice 3.661 2.32 0.37%
- Fix printing field separator in CSV metrics output.
- Fix --metric-only --json output.
- Fix summary output in CSV with --metric-only.
- Update event group check for support of uncore event.
perf test:
- Stop requiring a C toolchain in shell tests, instead add a workload
option that has all the previously C snippets built as part of
'perf test -w' that then get used in the 'perf test' shell scripts.
- Add event group test for events in multiple PMUs
- The "kernel lock contention analysis" test should not print
warnings in quiet mode.
- Add attr tests for ARM64's new VG register.
- Fix record test on KVM guests, as using precise flag with the
br_inst_retired.near_call event causes the test fail on KVM guests,
even when the guests have PMU forwarding enabled and the event
itself is supported, so just remove the precise flag from the
event.
- Add mechanism for skipping attr tests on specific kernel versions
where it is known that these checks will fail.
- Skip watchpoint tests if no watchpoints available.
- Add more Intel PT 'perf test' entries: hybrid CPUs, split the
packet decoder into a suite of subtests.
perf script:
- Introduce task analyzer python script, where one first records some events:
Recording can be done in two ways:
$ perf script record tasks-analyzer -- sleep 10
$ perf record -e sched:sched_switch -a -- sleep 10
The script can parse any perf.data files, as long as it has
sched:sched_switch events, other events will be ignored.
The most simple report use case is to just call the script without
arguments.
Runtime is the time the task was running on the CPU, Time Out-In is
the time between the process being scheduled *out* and scheduled
back *in*. So the last time span between two executions:
$ perf script report tasks-analyzer
Switched-In Switched-Out CPU PID TID Comm Runtime Time Out-In
15576.658891407 15576.659156086 4 2412 2428 gdbus 265 1949
15576.659111320 15576.659455410 0 2412 2412 gnome-shell 344 2267
15576.659491326 15576.659506173 2 74 74 kworker/2:1 15 13145
15576.659506173 15576.659825748 2 2858 2858 gnome-terminal- 320 63263
15576.659871270 15576.659902872 6 20932 20932 kworker/u16:0 32 2314582
15576.659909951 15576.659945501 3 27264 27264 sh 36 -1
15576.659853285 15576.659971052 7 27265 27265 perf 118 5050741
[...]
perf lock:
- Allow concurrent record and report to support live monitoring of
kernel lock contention without BPF:
# perf lock record -a -o- sleep 1 | perf lock contention -i-
contended total wait max wait avg wait type caller
2 10.27 us 6.17 us 5.13 us spinlock load_balance+0xc03
1 5.29 us 5.29 us 5.29 us rwlock:W ep_scan_ready_list+0x54
1 4.12 us 4.12 us 4.12 us spinlock smpboot_thread_fn+0x116
1 3.28 us 3.28 us 3.28 us mutex pipe_read+0x50
- Implement -t/--threads option when using BPF:
$ sudo ./perf lock contention -abt -E 5 sleep 1
contended total wait max wait avg wait pid comm
1 740.66 ms 740.66 ms 740.66 ms 1950 nv_queue
3 305.50 ms 298.19 ms 101.83 ms 1884 nvidia-modeset/
1 25.14 us 25.14 us 25.14 us 2725038 EventManager_De
12 23.09 us 9.30 us 1.92 us 0 swapper
1 20.18 us 20.18 us 20.18 us 2725033 EventManager_De
- Add -l/--lock-addr to aggregate per-lock-instance contention:
$ sudo ./perf lock contention -abl sleep 1
contended total wait max wait avg wait address symbol
1 36.28 us 36.28 us 36.28 us ffff92615d6448b8
9 10.91 us 1.84 us 1.21 us ffffffffbaed50c0 rcu_state
1 10.49 us 10.49 us 10.49 us ffff9262ac4f0c80
8 4.68 us 1.67 us 585 ns ffffffffbae07a40 jiffies_lock
3 3.03 us 1.45 us 1.01 us ffff9262277861e0
1 924 ns 924 ns 924 ns ffff926095ba9d20
1 436 ns 436 ns 436 ns ffff9260bfda4f60
perf record:
- Add remaining branch filters: "no_cycles", "no_flags" & "hw_index",
to be used with hardware such as Intel's LBR that allows things
like stitching stacks of two samples to overcome the limits of the
number of LBR registers.
Symbol resolution:
- Handle .debug files created with 'objcopy --only-keep-debug', where
program headers are zeroed and thus can't be used for adjustments,
use the info in the runtime_ss (runtime ELF) instead.
perf trace:
- Add BPF based augmenter for the 'perf_event_open's 'struct
perf_event_attr' argument.
- Add BPF based augmenter for the 'clock_gettime's 'struct timespec'
argument.
- In both cases the syscall tracepoint has just the pointer value, we
need to hook a BPF program to collect the pointer contents, and
then, in userspace, pretty print it in 'perf trace'.
perf list:
- Introduce JSON output of events.
- Streamline how the expression specifying what events should be
shown is handled, fixing several corner cases, such as the metric
filter that is specified as a glob but was using strstr().
perf probe:
- Fix to avoid crashing if DW_AT_decl_file is NULL, coping with clang
generating DWARF5 like that.
- Use dwarf_attr_integrate() as generic DWARF attr accessor as it
supersedes dwarf_attr(), supporting abstact origin DIEs.
perf inject:
- Set PERF_RECORD_MISC_BUILD_ID_SIZE in the PERF_RECORD_HEADER_BUILD_ID
so that perf.data readers can get the real build-id size and avoid
trailing zeroes.
perf data:
- Add tracepoint fields when converting a perf.data file to JSON.
arm64:
- Fix mksyscalltbl, don't lose syscalls due to sort -nu.
- Add Arm Neoverse V2 PMU events.
riscv:
- Add riscv sbi firmware std event files.
- Add Sifive U74 vendor events (JSON) file.
- Add some more events and metrics for Alderlake/Alderlake-N.
Documentation:
- Add data documentation for the PMU structs in the C source code.
Miscellaneous:
- Periodic sanitization of headers, adding missing includes, removing
needless ones, creating new ones, etc.
- Use sig_atomic_t for signal handlers to avoid undefined behaviour
in all perf tools.
- Fixes for libbpf 1.0+ compatibility (maps, etc) on 'perf trace' BPF
examples.
- Remove some old perf bpf examples, leave the best ones that
demonstrate how to associate BPF functions to points in the kernel.
- Make quiet mode consistent between tools.
- Use dedicated non-atomic clear/set bit helpers.
- Use "grep -E" instead of "egrep" as recommended by warning emitted
by GNU grep since at least version 3.8.
- Complete list of supported subcommands in the 'perf daemon' help
message.
- Update John Garry's email address for arm64 perf tooling on the
MAINTAINERS file, he moved from Huawei to Oracle"
* tag 'perf-tools-for-v6.2-1-2022-12-16' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (239 commits)
libperf: Fix install_pkgconfig target
perf tools: Use "grep -E" instead of "egrep"
perf stat: Do not delay the workload with --delay
perf evlist: Remove group option.
perf build: Fix python/perf.so library's name
perf test arm64: Add attr tests for new VG register
perf test: Add mechanism for skipping attr tests on kernel versions
perf test: Add mechanism for skipping attr tests on auxiliary vector values
perf test: Add ability to test exit code for attr tests
perf test: add new task-analyzer tests
perf script: task-analyzer add csv support
perf script: Introduce task analyzer python script
perf cs-etm: Print auxtrace info even if OpenCSD isn't linked
perf cs-etm: Cleanup cs_etm__process_auxtrace_info()
perf cs-etm: Tidy up auxtrace info header printing
perf cs-etm: Remove unused stub methods
perf cs-etm: Print unknown header version as an error
perf test: Update perf lock contention test
perf lock contention: Add -l/--lock-addr option
perf lock contention: Implement -t/--threads option for BPF
...
show more ...
|