The struct dump_regs contains 512 bytes of cache_regs, meaning the two
values in perf_sample contribute 1088 bytes of its total 1384 bytes
size. Initializing this much memory has a cost reported by Tavian
Barnes <tavianator@tavianator.com> as about 2.5% when running `perf
script --itrace=i0`:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/
Adrian Hunter <adrian.hunter@intel.com> replied that the zero
initialization was necessary and couldn't simply be removed.
This patch aims to strike a middle ground of still zeroing the
perf_sample, but removing 79% of its size by make user_regs and
intr_regs optional pointers to zalloc-ed memory. To support the
allocation accessors are created for user_regs and intr_regs. To
support correct cleanup perf_sample__init and perf_sample__exit
functions are created and added throughout the code base.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250113194345.1537821-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add samples of APX and other new instructions to the 'x86 instruction
decoder - new instructions' test.
Note the test is only available if the perf tool has been built with
EXTRA_TESTS=1.
Example:
$ make EXTRA_TESTS=1 -C tools/perf
$ tools/perf/perf test -F -v 'new ins' |& grep -i 'jmpabs\|popp\|pushp'
Decoded ok: d5 00 a1 ef cd ab 90 78 56 34 12 jmpabs $0x1234567890abcdef
Decoded ok: d5 08 53 pushp %rbx
Decoded ok: d5 18 50 pushp %r16
Decoded ok: d5 19 57 pushp %r31
Decoded ok: d5 19 5f popp %r31
Decoded ok: d5 18 58 popp %r16
Decoded ok: d5 08 5b popp %rbx
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Chang S. Bae <chang.seok.bae@intel.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Nikolay Borisov <nik.borisov@suse.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: x86@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240502105853.5338-11-adrian.hunter@intel.com
Prior to this patch the first and the last error encountered during
parsing are printed. To see other errors verbose needs
enabling. Unfortunately this can drop useful errors, in particular on
terms. This patch changes the errors so that instead of the first and
last all errors are recorded and printed, the underlying data
structure is changed to a list.
Before:
```
$ perf stat -e 'slots/edge=2/' true
event syntax error: 'slots/edge=2/'
\___ Bad event or PMU
Unable to find PMU or event on a PMU of 'slots'
Initial error:
event syntax error: 'slots/edge=2/'
\___ Cannot find PMU `slots'. Missing kernel support?
Run 'perf list' for a list of valid events
Usage: perf stat [<options>] [<command>]
-e, --event <event> event selector. use 'perf list' to list available events
```
After:
```
$ perf stat -e 'slots/edge=2/' true
event syntax error: 'slots/edge=2/'
\___ Bad event or PMU
Unable to find PMU or event on a PMU of 'slots'
event syntax error: 'slots/edge=2/'
\___ value too big for format (edge), maximum is 1
event syntax error: 'slots/edge=2/'
\___ Cannot find PMU `slots'. Missing kernel support?
Run 'perf list' for a list of valid events
Usage: perf stat [<options>] [<command>]
-e, --event <event> event selector. use 'perf list' to list available events
```
Signed-off-by: Ian Rogers <irogers@google.com>
Reviewed-by: James Clark <james.clark@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: tchen168@asu.edu
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Michael Petlan <mpetlan@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240131134940.593788-3-irogers@google.com
Pull perf tools updates from Namhyung Kim:
"Internal cleanup:
- Refactor PMU data management to handle hybrid systems in a generic
way.
Do more work in the lexer so that legacy event types parse more
easily. A side-effect of this is that if a PMU is specified,
scanning sysfs is avoided improving start-up time.
- Fix hybrid metrics, for example, the TopdownL1 works for both
performance and efficiency cores on Intel machines. To support
this, sort and regroup events after parsing.
- Add reference count checking for the 'thread' data structure.
- Lots of fixes for memory leaks in various places thanks to the ASAN
and Ian's refcount checker.
- Reduce the binary size by replacing static variables with local or
dynamically allocated memory.
- Introduce shared_mutex for annotate data to reduce memory
footprint.
- Make filesystem access library functions more thread safe.
Test:
- Organize cpu_map tests into a single suite.
- Add metric value validation test to check if the values are within
correct value ranges.
- Add perf stat stdio output test to check if event and metric names
match.
- Add perf data converter JSON output test.
- Fix a lot of issues reported by shellcheck(1). This is a
preparation to enable shellcheck by default.
- Make the large x86 new instructions test optional at build time
using EXTRA_TESTS=1.
- Add a test for libpfm4 events.
perf script:
- Add 'dsoff' outpuf field to display offset from the DSO.
$ perf script -F comm,pid,event,ip,dsoff
ls 2695501 cycles: 152cc73ef4b5 (/usr/lib/x86_64-linux-gnu/ld-2.31.so+0x1c4b5)
ls 2695501 cycles: ffffffff99045b3e ([kernel.kallsyms])
ls 2695501 cycles: ffffffff9968e107 ([kernel.kallsyms])
ls 2695501 cycles: ffffffffc1f54afb ([kernel.kallsyms])
ls 2695501 cycles: ffffffff9968382f ([kernel.kallsyms])
ls 2695501 cycles: ffffffff99e00094 ([kernel.kallsyms])
ls 2695501 cycles: 152cc718a8d0 (/usr/lib/x86_64-linux-gnu/libselinux.so.1+0x68d0)
ls 2695501 cycles: ffffffff992a6db0 ([kernel.kallsyms])
- Adjust width for large PID/TID values.
perf report:
- Robustify reading addr2line output for srcline by checking sentinel
output before the actual data and by using timeout of 1 second.
- Allow config terms (like 'name=ABC') with breakpoint events.
$ perf record -e mem:0x55feb98dd169:x/name=breakpoint/ -p 19646 -- sleep 1
perf annotate:
- Handle x86 instruction suffix like 'l' in 'movl' generally.
- Parse instruction operands properly even with a whitespace. This is
needed for llvm-objdump output.
- Support RISC-V binutils lookup using the triplet prefixes.
- Add '<' and '>' key to navigate to prev/next symbols in TUI.
- Fix instruction association and parsing for LoongArch.
perf stat:
- Add --per-cache aggregation option, optionally specify a cache
level like `--per-cache=L2`.
$ sudo perf stat --per-cache -a -e ls_dmnd_fills_from_sys.ext_cache_remote --\
taskset -c 0-15,64-79,128-143,192-207\
perf bench sched messaging -p -t -l 100000 -g 8
# Running 'sched/messaging' benchmark:
# 20 sender and receiver threads per group
# 8 groups == 320 threads run
Total time: 7.648 [sec]
Performance counter stats for 'system wide':
S0-D0-L3-ID0 16 17,145,912 ls_dmnd_fills_from_sys.ext_cache_remote
S0-D0-L3-ID8 16 14,977,628 ls_dmnd_fills_from_sys.ext_cache_remote
S0-D0-L3-ID16 16 262,539 ls_dmnd_fills_from_sys.ext_cache_remote
S0-D0-L3-ID24 16 3,140 ls_dmnd_fills_from_sys.ext_cache_remote
S0-D0-L3-ID32 16 27,403 ls_dmnd_fills_from_sys.ext_cache_remote
S0-D0-L3-ID40 16 17,026 ls_dmnd_fills_from_sys.ext_cache_remote
S0-D0-L3-ID48 16 7,292 ls_dmnd_fills_from_sys.ext_cache_remote
S0-D0-L3-ID56 16 2,464 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID64 16 22,489,306 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID72 16 21,455,257 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID80 16 11,619 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID88 16 30,978 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID96 16 37,628 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID104 16 13,594 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID112 16 10,164 ls_dmnd_fills_from_sys.ext_cache_remote
S1-D1-L3-ID120 16 11,259 ls_dmnd_fills_from_sys.ext_cache_remote
7.779171484 seconds time elapsed
- Change default (no event/metric) formatting for default metrics so
that events are hidden and the metric and group appear.
Performance counter stats for 'ls /':
1.85 msec task-clock # 0.594 CPUs utilized
0 context-switches # 0.000 /sec
0 cpu-migrations # 0.000 /sec
97 page-faults # 52.517 K/sec
2,187,173 cycles # 1.184 GHz
2,474,459 instructions # 1.13 insn per cycle
531,584 branches # 287.805 M/sec
13,626 branch-misses # 2.56% of all branches
TopdownL1 # 23.5 % tma_backend_bound
# 11.5 % tma_bad_speculation
# 39.1 % tma_frontend_bound
# 25.9 % tma_retiring
- Allow --cputype option to have any PMU name (not just hybrid).
- Fix output value not to added when it runs multiple times with -r
option.
perf list:
- Show metricgroup description from JSON file called
metricgroups.json.
- Allow 'pfm' argument to list only libpfm4 events and check each
event is supported before showing it.
JSON vendor events:
- Avoid event grouping using "NO_GROUP_EVENTS" constraints. The
topdown events are correctly grouped even if no group exists.
- Add "Default" metric group to print it in the default output. And
use "DefaultMetricgroupName" to indicate the real metric group
name.
- Add AmpereOne core PMU events.
Misc:
- Define man page date correctly.
- Track exception level properly on ARM CoreSight ETM.
- Allow anonymous struct, union or enum when retrieving type names
from DWARF.
- Fix incorrect filename when calling `perf inject --jit`.
- Handle PLT size correctly on LoongArch"
* tag 'perf-tools-for-v6.5-1-2023-06-28' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next: (269 commits)
perf test: Skip metrics w/o event name in stat STD output linter
perf test: Reorder event name checks in stat STD output linter
perf pmu: Remove a hard coded cpu PMU assumption
perf pmus: Add notion of default PMU for JSON events
perf unwind: Fix map reference counts
perf test: Set PERF_EXEC_PATH for script execution
perf script: Initialize buffer for regs_map()
perf tests: Fix test_arm_callgraph_fp variable expansion
perf symbol: Add LoongArch case in get_plt_sizes()
perf test: Remove x permission from lib/stat_output.sh
perf test: Rerun failed metrics with longer workload
perf test: Add skip list for metrics known would fail
perf test: Add metric value validation test
perf jit: Fix incorrect file name in DWARF line table
perf annotate: Fix instruction association and parsing for LoongArch
perf annotation: Switch lock from a mutex to a sharded_mutex
perf sharded_mutex: Introduce sharded_mutex
tools: Fix incorrect calculation of object size by sizeof
perf subcmd: Fix missing check for return value of malloc() in add_cmdname()
perf parse-events: Remove unneeded semicolon
...
Intel Flexible Return and Event Delivery (FRED) adds instructions ERETS
(return to supervisor) and ERETU (return to user). Intel PT instruction
decoder needs to know about these instructions because they are
branch instructions. Similar to IRET instructions, when the decoder
encounters one of these instructions it will match it to a TIP (target
instruction pointer) packet that informs what the branch destination is.
The existing "x86 instruction decoder - new instructions" test can be
used to test the result e.g.
$ perf test -v ins |& grep eret
Decoded ok: f2 0f 01 ca erets
Decoded ok: f3 0f 01 ca eretu
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Acked-by: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20230320183517.15099-2-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Some places were including event.h just to get 'struct perf_sample',
move it to a separate place so that we speed up a bit the build.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
As of Intel SDM (https://www.intel.com/sdm) version 076, there is a new
Intel PT feature called Event Trace which requires 2 new packets CFE and
EVD. Add them to the packet decoder and packet decoder test.
Committer notes:
I got the "Intel® 64 and IA-32 architectures software developer’s manual
combined volumes: 1, 2A, 2B, 2C, 2D, 3A, 3B, 3C, 3D, and 4" PDF at:
https://cdrdv2.intel.com/v1/dl/getContent/671200
And these new packets are described in page 3951:
<quote>
32.2.4
Event Trace is a capability that exposes details about the asynchronous
events, when they are generated, and when their corresponding software
event handler completes execution. These include:
o Interrupts, including NMI and SMI, including the interrupt vector when
defined.
o Faults, exceptions including the fault vector.
— Page faults additionally include the page fault address, when in context.
o Event handler returns, including IRET and RSM.
o VM exits and VM entries.¹
— VM exits include the values written to the “exit reason” and “exit qualification” VMCS fields.
INIT and SIPI events.
o TSX aborts, including the abort status returned for the RTM instructions.
o Shutdown.
Additionally, it provides indication of the status of the Interrupt Flag
(IF), to indicate when interrupts are masked.
</quote>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: https://lore.kernel.org/r/20220124084201.2699795-4-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The x86 instruction decoder is used for both kernel instructions and
user space instructions (e.g. uprobes, perf tools Intel PT), so it is
good to update it with new instructions.
Add AVX512-FP16 instructions to x86 instruction decoder test.
A subsequent patch adds the instructions to the instruction decoder.
Reference:
Intel AVX512-FP16 Architecture Specification
June 2021
Revision 1.0
Document Number: 347407-001US
Example:
$ perf test -v "x86 instruction decoder" |& grep vfcmaddcph | head -2
Failed to decode: 62 f6 6f 48 56 cb vfcmaddcph %zmm3,%zmm2,%zmm1
Failed to decode: 62 f6 6f 48 56 8c c8 78 56 34 12 vfcmaddcph 0x12345678(%eax,%ecx,8),%zmm2,%zmm1
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lore.kernel.org/r/20211202095029.2165714-6-adrian.hunter@intel.com
The x86 instruction decoder is used for both kernel instructions and
user space instructions (e.g. uprobes, perf tools Intel PT), so it is
good to update it with new instructions.
Add the following instructions to the x86 instruction decoder test:
User Interrupt
clui
senduipi
stui
testui
uiret
Prediction history reset
hreset
Serialize instruction execution
serialize
TSX suspend load address tracking
xresldtrk
xsusldtrk
A subsequent patch adds the instructions to the instruction decoder.
Reference:
Intel Architecture Instruction Set Extensions and Future Features
Programming Reference
May 2021
Document Number: 319433-044
Example:
$ perf test -v "x86 instruction decoder" |& grep -i hreset
Failed to decode length (4 vs expected 6): f3 0f 3a f0 c0 00 hreset $0x0
Failed to decode length (4 vs expected 6): f3 0f 3a f0 c0 00 hreset $0x0
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lore.kernel.org/r/20211202095029.2165714-4-adrian.hunter@intel.com
Pull perf tool updates from Arnaldo Carvalho de Melo:
"perf stat:
- Add support for hybrid PMUs to support systems such as Intel
Alderlake and its BIG/little core/atom cpus.
- Introduce 'bperf' to share hardware PMCs with BPF.
- New --iostat option to collect and present IO stats on Intel
hardware.
This functionality is based on recently introduced sysfs attributes
for Intel® Xeon® Scalable processor family (code name Skylake-SP)
in commit bb42b3d397 ("perf/x86/intel/uncore: Expose an Uncore
unit to IIO PMON mapping")
It is intended to provide four I/O performance metrics in MB per
each PCIe root port:
- Inbound Read: I/O devices below root port read from the host memory
- Inbound Write: I/O devices below root port write to the host memory
- Outbound Read: CPU reads from I/O devices below root port
- Outbound Write: CPU writes to I/O devices below root port
- Align CSV output for summary.
- Clarify --null use cases: Assess raw overhead of 'perf stat' or
measure just wall clock time.
- Improve readability of shadow stats.
perf record:
- Change the COMM when starting tha workload so that --exclude-perf
doesn't seem to be not honoured.
- Improve 'Workload failed' message printing events + what was
exec'ed.
- Fix cross-arch support for TIME_CONV.
perf report:
- Add option to disable raw event ordering.
- Dump the contents of PERF_RECORD_TIME_CONV in 'perf report -D'.
- Improvements to --stat output, that shows information about
PERF_RECORD_ events.
- Preserve identifier id in OCaml demangler.
perf annotate:
- Show full source location with 'l' hotkey in the 'perf annotate'
TUI.
- Add line number like in TUI and source location at EOL to the 'perf
annotate' --stdio mode.
- Add --demangle and --demangle-kernel to 'perf annotate'.
- Allow configuring annotate.demangle{,_kernel} in 'perf config'.
- Fix sample events lost in stdio mode.
perf data:
- Allow converting a perf.data file to JSON.
libperf:
- Add support for user space counter access.
- Update topdown documentation to permit rdpmc calls.
perf test:
- Add 'perf test' for 'perf stat' CSV output.
- Add 'perf test' entries to test the hybrid PMU support.
- Cleanup 'perf test daemon' if its 'perf test' is interrupted.
- Handle metric reuse in pmu-events parsing 'perf test' entry.
- Add test for PE executable support.
- Add timeout for wait for daemon start in its 'perf test' entries.
Build:
- Enable libtraceevent dynamic linking.
- Improve feature detection output.
- Fix caching of feature checks caching.
- First round of updates for tools copies of kernel headers.
- Enable warnings when compiling BPF programs.
Vendor specific events:
- Intel:
- Add missing skylake & icelake model numbers.
- arm64:
- Add Hisi hip08 L1, L2 and L3 metrics.
- Add Fujitsu A64FX PMU events.
- PowerPC:
- Initial JSON/events list for power10 platform.
- Remove unsupported power9 metrics.
- AMD:
- Add Zen3 events.
- Fix broken L2 Cache Hits from L2 HWPF metric.
- Use lowercases for all the eventcodes and umasks.
Hardware tracing:
- arm64:
- Update CoreSight ETM metadata format.
- Fix bitmap for CS-ETM option.
- Support PID tracing in config.
- Detect pid in VMID for kernel running at EL2.
Arch specific updates:
- MIPS:
- Support MIPS unwinding and dwarf-regs.
- Generate mips syscalls_n64.c syscall table.
- PowerPC:
- Add support for PERF_SAMPLE_WEIGH_STRUCT on PowerPC.
- Support pipeline stage cycles for powerpc.
libbeauty:
- Fix fsconfig generator"
* tag 'perf-tools-for-v5.13-2021-04-29' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (132 commits)
perf build: Defer printing detected features to the end of all feature checks
tools build: Allow deferring printing the results of feature detection
perf build: Regenerate the FEATURE_DUMP file after extra feature checks
perf session: Dump PERF_RECORD_TIME_CONV event
perf session: Add swap operation for event TIME_CONV
perf jit: Let convert_timestamp() to be backwards-compatible
perf tools: Change fields type in perf_record_time_conv
perf tools: Enable libtraceevent dynamic linking
perf Documentation: Document intel-hybrid support
perf tests: Skip 'perf stat metrics (shadow stat) test' for hybrid
perf tests: Support 'Convert perf time to TSC' test for hybrid
perf tests: Support 'Session topology' test for hybrid
perf tests: Support 'Parse and process metrics' test for hybrid
perf tests: Support 'Track with sched_switch' test for hybrid
perf tests: Skip 'Setup struct perf_event_attr' test for hybrid
perf tests: Add hybrid cases for 'Roundtrip evsel->name' test
perf tests: Add hybrid cases for 'Parse event definition strings' test
perf record: Uniquify hybrid event name
perf stat: Warn group events from different hybrid PMU
perf stat: Filter out unmatched aggregation for hybrid event
...
In some versions of alpine Linux the perf build is broken since commit
1d509f2a6e ("x86/insn: Support big endian cross-compiles"):
In file included from /usr/include/linux/byteorder/little_endian.h:13,
from /usr/include/asm/byteorder.h:5,
from arch/x86/util/../../../../arch/x86/include/asm/insn.h:10,
from arch/x86/util/archinsn.c:2:
/usr/include/linux/swab.h:161:8: error: unknown type name '__always_inline'
static __always_inline __u16 __swab16p(const __u16 *p)
So move the inclusion of arch/x86/include/asm/insn.h to later in the
places where linux/stddef.h (that conditionally defines
__always_inline) to workaround this problem on Alpine Linux 3.9 to 3.11,
3.12 onwards works.
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The ins_lat of PERF_SAMPLE_WEIGHT_STRUCT stands for the instruction
latency, which is only available for X86. Add a X86 specific test for
the ins_lat and PERF_SAMPLE_WEIGHT_STRUCT type.
The test__x86_sample_parsing() uses the same way as the
test__sample_parsing() to verify a sample type. Since the ins_lat and
PERF_SAMPLE_WEIGHT_STRUCT are the only X86 specific sample type for now,
the test__x86_sample_parsing() only verify the PERF_SAMPLE_WEIGHT_STRUCT
type. Other sample types are still verified in the generic test.
$ perf test 77 -v
77: x86 Sample parsing :
--- start ---
test child forked, pid 102370
test child finished with 0
---- end ----
x86 Sample parsing: Ok
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Sumanth Korikkar <sumanthk@linux.ibm.com>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Link: http://lore.kernel.org/lkml/1614787285-104151-2-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
