So right now there's a somewhat inconsistent mess of the benchmarking
code and options sometimes calling benchmarked functions 'functions',
sometimes calling them 'routines'.
Name them 'functions' consistently.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1445241870-24854-14-git-send-email-mingo@kernel.org
[ Updated perf-bench man page, pointed out by David Ahern ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We have three benchmarking subsystems that specify some sort of 'number
of loops' parameter - but all of them do it inconsistently:
numa: -l/--nr_loops
sched messaging: -l/--loops
mem memset/memcpy: -i/--iterations
Harmonize them to -l/--nr_loops by picking the numa variant - which is
also the most likely one to have existing scripting which we don't want
to break.
Plus improve the parameter help texts to indicate the default value for
the nr_loops variable to keep users from guessing ...
Also propagate the naming to internal variables.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1445241870-24854-13-git-send-email-mingo@kernel.org
[ Let the harmonisation reach the perf-bench man page as well ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So 'perf bench mem memcpy/memset' consistently uses 'len' and 'length'
for buffer sizes - while it's really a memory buffer size. (strings have
length.)
Rename all affected variables.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1445241870-24854-10-git-send-email-mingo@kernel.org
[ Update perf-bench man page ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So 'perf bench mem memset/memcpy' has a CPU cycles measurement method,
but calls it 'cycle' (singular) throughout the code, which makes it
harder to read.
Rename all related functions, variables and options to a plural 'cycles'
nomenclature.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1445241870-24854-8-git-send-email-mingo@kernel.org
[ s/--cycle/--cycles/g in perf-bench man page ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So 'perf bench mem memcpy/memset' has elaborate code to measure
memcpy()/memset() performance both with freshly allocated buffers (which
includes initial page fault overhead) and with preallocated buffers.
But the thing is, the resulting bandwidth results are mostly
meaningless, because page faults dominate so much of the cost.
It might make sense to measure cache cold vs. cache hot performance, but
the code does not do this.
So remove this complication, and always prefault the ranges before using
them.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1445241870-24854-6-git-send-email-mingo@kernel.org
[ Remove --no-prefault, --only-prefault from docs, noticed by David Ahern ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Which is the most common default found in other similar tools.
Requested-by: Ingo Molnar <mingo@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Chandler Carruth <chandlerc@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://www.youtube.com/watch?v=nXaxk27zwlk
Link: http://lkml.kernel.org/n/tip-v8lq36aispvdwgxdmt9p9jd9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Adding handling for '-h' and '-v' options to invoke help and version
command respectively.
Current behaviour is:
$ perf -v
Unknown option: -v
Usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS]
$ perf -h
Unknown option: -h
Usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS]
New behaviour:
$ perf -h
usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS]
The most commonly used perf commands are:
annotate Read perf.data (created by perf record) and display annotated code
archive Create archive with object files with build-ids found in perf.data file
bench General framework for benchmark suites
...
$ perf -v
perf version 4.3.rc3.gc99e32
Updated man page.
Requested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-10-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The 'P' will cause the event to get maximum possible detected precise
level.
Following record:
$ perf record -e cycles:P ...
will detect maximum precise level for 'cycles' event and use it.
Commiter note:
Testing it:
$ perf record -e cycles:P usleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.013 MB perf.data (9 samples) ]
$ perf evlist
cycles:P
$ perf evlist -v
cycles:P: size: 112, { sample_period, sample_freq }: 4000, sample_type:
IP|TID|TIME|PERIOD, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1,
enable_on_exec: 1, task: 1, precise_ip: 2, sample_id_all: 1, mmap2: 1,
comm_exec: 1
$
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-6-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The --interval-print parameter was limited to 100ms. However, for
example, 10ms is required to do sophisticated bandwidth analysis using
uncore events.
The test shows that the overhead of the system-wide uncore monitoring
with 10ms interval is only ~2%. So this patch reduces the minimal
interval-print allowd to 10ms.
But 10ms may not work well for all cases. For example, when the
cpus/threads number is very large, for system-wide core event monitoring
the overhead could be high.
To handle this issue, a warning will be displayed when the
interval-print is set between 10ms to 100ms. So users can make a
decision according to their specific cases.
# perf stat -e uncore_imc_1/cas_count_read/ -a --interval-print 10 -- sleep 1
print interval < 100ms. The overhead percentage could be high in some
cases. Please proceed with caution.
# time counts unit events
0.010200451 0.10 MiB uncore_imc_1/cas_count_read/
0.020475117 0.02 MiB uncore_imc_1/cas_count_read/
0.030692800 0.01 MiB uncore_imc_1/cas_count_read/
0.040948161 0.02 MiB uncore_imc_1/cas_count_read/
0.051159564 0.00 MiB uncore_imc_1/cas_count_read/
Signed-off-by: Kan Liang <kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1443776674-42511-1-git-send-email-kan.liang@intel.com
[ Added warning about overhead when using sub 100ms intervals to the man page ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The --max_stack option was added as an optimization to reduce processing time,
so people specifying --max-stack might get a increased processing time if
combined with synthesized callchains, but otherwise no real harm.
A warning about setting both --max_stack and the synthesized callchains max
depth seems like overkill. Amend the documentation.
Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/560A5155.4060105@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
autofdo incorrectly expects branch flags to include either mispred or
predicted. In fact mispred = predicted = 0 is valid and means the flags
are not supported, which they aren't by Intel PT.
To make autofdo work, add a config option which will cause Intel PT
decoder to set the mispred flag on all branches.
Below is an example of using Intel PT with autofdo. The example is
also added to the Intel PT documentation. It requires autofdo
(https://github.com/google/autofdo) and gcc version 5. The bubble
sort example is from the AutoFDO tutorial (https://gcc.gnu.org/wiki/AutoFDO/Tutorial)
amended to take the number of elements as a parameter.
$ gcc-5 -O3 sort.c -o sort_optimized
$ ./sort_optimized 30000
Bubble sorting array of 30000 elements
2254 ms
$ cat ~/.perfconfig
[intel-pt]
mispred-all
$ perf record -e intel_pt//u ./sort 3000
Bubble sorting array of 3000 elements
58 ms
[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 3.939 MB perf.data ]
$ perf inject -i perf.data -o inj --itrace=i100usle --strip
$ ./create_gcov --binary=./sort --profile=inj --gcov=sort.gcov -gcov_version=1
$ gcc-5 -O3 -fauto-profile=sort.gcov sort.c -o sort_autofdo
$ ./sort_autofdo 30000
Bubble sorting array of 30000 elements
2155 ms
Note there is currently no advantage to using Intel PT instead of LBR,
but that may change in the future if greater use is made of the data.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-26-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a new option --strip which is used with --itrace to strip out
non-synthesized events. This results in a perf.data file that is
simpler for external tools to parse. In particular, this can be used to
prepare a perf.data file for consumption by autofdo.
A subsequent patch makes a change to Intel PT also to enable use with
autofdo and gives an example of that use.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-25-git-send-email-adrian.hunter@intel.com
[ Made it use perf_evlist__remove() + perf_evsel__delete() ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add support for generating branch stack context for PT samples. The
decoder reports a configurable number of branches as branch context for
each sample. Internally it keeps track of them by using a simple sliding
window. We also flush the last branch buffer on each sample to avoid
overlapping intervals.
This is useful for:
- Reporting accurate basic block edge frequencies through the perf
report branch view
- Using with --branch-history to get the wider context of samples
- Other users of LBRs
Also the Documentation is updated.
Examples:
Record with Intel PT:
perf record -e intel_pt//u ls
Branch stacks are used by default if synthesized so:
perf report --itrace=ile
is the same as:
perf report --itrace=ile -b
Branch history can be requested also:
perf report --itrace=igle --branch-history
Based-on-patch-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-15-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add AUX area tracing option 'l' to synthesize branch stacks on samples
just like sample type PERF_SAMPLE_BRANCH_STACK. This is taken into use
by Intel PT in a subsequent patch.
Based-on-patch-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-9-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add option --ns to display time to 9 decimal places. That is useful in
some cases, for example when using Intel PT cycle accurate mode.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-6-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Instruction tracing options (i.e. --itrace) include an option for
sampling instructions at an arbitrary period. e.g.
--itrace=i10us
means make an 'instructions' sample for every 10us of trace.
Currently the logic does not distinguish between a period of
zero and no period being specified at all, so it gets treated
as the default period which is 100000. That doesn't really
make sense.
Fix it so that zero period is accepted and treated as meaning
"as often as possible".
In the case of Intel PT that is the same as a period of 1 and
a unit of 'instructions' (i.e. --itrace=i1i).
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-2-git-send-email-adrian.hunter@intel.com
[ Add a few lines describing this in the Documentation/intel-pt.txt file ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
no_force_psb was dropped as a late change to the kernel driver.
Consequently, remove it from the documentation.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443089122-19082-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch modifies the -I/--int-regs option to enablepassing the name
of the registers to sample on interrupt. Registers can be specified by
their symbolic names. For instance on x86, --intr-regs=ax,si.
The motivation is to reduce the size of the perf.data file and the
overhead of sampling by only collecting the registers useful to a
specific analysis. For instance, for value profiling, sampling only the
registers used to passed arguements to functions.
With no parameter, the --intr-regs still records all possible registers
based on the architecture.
To name registers, it is necessary to use the long form of the option,
i.e., --intr-regs:
$ perf record --intr-regs=si,di,r8,r9 .....
To record any possible registers:
$ perf record -I .....
$ perf report --intr-regs ...
To display the register, one can use perf report -D
To list the available registers:
$ perf record --intr-regs=\?
available registers: AX BX CX DX SI DI BP SP IP FLAGS CS SS R8 R9 R10 R11 R12 R13 R14 R15
Signed-off-by: Stephane Eranian <eranian@google.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1441039273-16260-4-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch adds the output of the interrupted machine state (iregs) to
perf script. It presents them as NAME:VALUE so this is easy to parse
during post processing.
To capture the interrupted machine state:
$ perf record -I ....
to display iregs, use the -F option:
$ perf script -F ip,iregs
40afc2 AX:0x6c5770 BX:0x1e CX:0x5f4d80a DX:0x101010101010101 SI:0x1
Signed-off-by: Stephane Eranian <eranian@google.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1441039273-16260-2-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Sometimes when post-processing output from `perf script` one does not
want to demangle C++ symbol names. Add an option to allow this.
Also add --[no-]demangle-kernel to be consistent with top/report/probe.
Signed-off-by: Mark Drayton <mbd@fb.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1440616695-32340-1-git-send-email-scientist@fb.com
Signed-off-by: Yannick Brosseau <scientist@fb.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf script, report and inject all have the same itrace options. Put
them into an asciidoc include file.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1437150840-31811-10-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Now perf can set per-event value of time and (sampling) period. But I
guess most users like me just want to set frequency rather than period.
So add the 'freq' term in the event parser.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1439102724-14079-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
In some cases it's useful to characterize samples by file. This is
useful to get a higher level categorization, for example to map cost to
subsystems.
Add a srcfile sort key to perf report. It builds on top of the existing
srcline support.
Commiter notes:
E.g.:
# perf record -F 10000 usleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.016 MB perf.data (13 samples) ]
[root@zoo ~]# perf report -s srcfile --stdio
# Total Lost Samples: 0
#
# Samples: 13 of event 'cycles'
# Event count (approx.): 869878
#
# Overhead Source File
# ........ ...........
60.99% .
20.62% paravirt.h
14.23% rmap.c
4.04% signal.c
0.11% msr.h
#
The first line is collecting all the files for which srcfiles couldn't somehow
get resolved to:
# perf report -s srcfile,dso --stdio
# Total Lost Samples: 0
#
# Samples: 13 of event 'cycles'
# Event count (approx.): 869878
#
# Overhead Source File Shared Object
# ........ ........... ................
40.97% . ld-2.20.so
20.62% paravirt.h [kernel.vmlinux]
20.02% . libc-2.20.so
14.23% rmap.c [kernel.vmlinux]
4.04% signal.c [kernel.vmlinux]
0.11% msr.h [kernel.vmlinux]
#
XXX: Investigate why that is not resolving on Fedora 21, Andi says he hasn't
seen this on Fedora 22.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438988064-21834-1-git-send-email-andi@firstfloor.org
[ Added column length update, from 0e65bdb3f90f ('perf hists: Update the column width for the "srcline" sort key') ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
For perf report/script srcline currently only the base file name of the
source file is printed. This is a good default because it usually fits
on the screen.
But in some cases we want to know the full file name, for example to
aggregate hits per file.
In the later case we need more than the base file name to resolve file
naming collisions: for example the kernel source has ~70 files named
"core.c"
It's also useful as input to post processing tools which want to point
to the right file.
Add a flag to allow full file name output.
Add an option to perf report/script to enable this option.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438986245-15191-1-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Now that we can process branch data in annotate it makes sense to
support enabling branch recording from top too. Most of the code needed
for this is already in shared code with report. But we need to add:
- The option parsing code (using shared code from the previous patch)
- Document the options
- Set up the IPC/cycles accounting state in the top session
- Call the accounting code in the hist iter callback
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-8-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
cycles is a new branch_info field available on some CPUs that indicates
the time deltas between branches in the LBR.
Add a sort key and output code for the cycles to allow to display the
basic block cycles individually in perf report.
We also pass in the cycles for weight when LBRs are processed, which
allows to get global and local weight, to get an estimate of the total
cost.
And also print the cycles information for perf report -D. I also added
printing for the previously missing LBR flags (mispredict etc.)
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-2-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patchkit adds the ability to turn off time stamps per event.
One usaful case for partial time is to work with per-event callgraph to
enable "PEBS threshold > 1" (https://lkml.org/lkml/2015/5/10/196), which
can significantly reduce the sampling overhead.
The event samples with time stamps off will not be ordered.
Signed-off-by: Kan Liang <kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438677022-34296-2-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Currently the command line option settings beats the per event period
settings:
With no global settings, we get per-event configuration:
$ perf record -e 'cpu/instructions,period=20000/' sleep 1
$ perf evlist -v
... { sample_period, sample_freq }: 20000 ...
With 'c' option period setup, we get 'c' option value:
$ perf record -e 'cpu/instructions,period=20000/' -c 1000 sleep 1
$ perf evlist -v
... { sample_period, sample_freq }: 1000 ...
This patch makes the per-event settings overload the global 'c' option
setup:
$ perf record -e 'cpu/instructions,period=20000/' -c 1000 sleep 1
$ perf evlist -v
... { sample_period, sample_freq }: 20000 ...
I think the making the per-event settings to overload any other config
makes more sense than current state. However it breaks the current
'period' term handling, which might cause some noise.. so let's see ;-).
Also fixing parse event tests with the new behaviour.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438162936-59698-3-git-send-email-kan.liang@intel.com
Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add option --show-switch-events to show switch events in a similar
fashion to --show-task-events and --show-mmap-events.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Jiri Olsa <jolsa@redhat.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Pawel Moll <pawel.moll@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1437471846-26995-6-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Allows a way of measuring low level kernel implementation of FUTEX_LOCK_PI and
FUTEX_UNLOCK_PI.
The program comes in two flavors:
(i) single futex (default), all threads contend on the same uaddr. For the
sake of the benchmark, we call into kernel space even when the lock is
uncontended. The kernel will set it to TID, any waters that come in and
contend for the pi futex will be handled respectively by the kernel.
(ii) -M option for multiple futexes, each thread deals with its own futex. This
is a trivial scenario and only measures kernel handling of 0->TID transition.
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Mel Gorman <mgorman@suse.de>
Link: http://lkml.kernel.org/r/1436259353.12255.78.camel@stgolabs.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch allows 'perf record' to exclude events issued by perf itself
by '--exclude-perf' option.
Before this patch, when doing something like:
# perf record -a -e syscalls:sys_enter_write <cmd>
One could easily get result like this:
# /tmp/perf report --stdio
...
# Overhead Command Shared Object Symbol
# ........ ....... .................. ....................
#
99.99% perf libpthread-2.18.so [.] __write_nocancel
0.01% ls libc-2.18.so [.] write
0.01% sshd libc-2.18.so [.] write
...
Where most events are generated by perf itself.
A shell trick can be done to filter perf itself out:
# cat << EOF > ./tmp
> #!/bin/sh
> exec perf record -e ... --filter="common_pid != \$\$" -a sleep 10
> EOF
# chmod a+x ./tmp
# ./tmp
However, doing so is user unfriendly.
This patch extracts evsel iteration framework introduced by patch 'perf
record: Apply filter to all events in a glob matching' into
foreach_evsel_in_last_glob(), and makes exclude_perf() function append
new filter expression to each evsel selected by a '-e' selector.
To avoid losing filters if user pass '--filter' after '--exclude-perf',
this patch uses perf_evsel__append_filter() in both case, instead of
perf_evsel__set_filter() which removes old filter. As a side effect, now
it is possible to use multiple '--filter' option for one selector. They
are combinded with '&&'.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1436513770-8896-2-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The 'period' param is not defined in
/sys/bus/event_sources/devices/<pmu>/format/*, but can be used, document
it.
Signed-off-by: Kan Liang <kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1436345097-11113-3-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Currently all the -p option PID arguments tasks values get aggregated
and printed as single values.
Adding --per-tasks option to print values per task.
$ perf stat -e cycles,instructions --per-thread -p 30190,30242
^C
Performance counter stats for process id '30190,30242':
cat-30190 0 cycles
yes-30242 3,842,525,421 cycles
cat-30190 0 instructions
yes-30242 10,370,817,010 instructions
1.143155657 seconds time elapsed
Also works under interval mode:
$ perf stat -e cycles,instructions --per-thread -p 30190,30242 -I 1000
# time comm-pid counts unit events
1.000073435 cat-30190 89,058 cycles
1.000073435 yes-30242 3,360,786,902 cycles (100.00%)
1.000073435 cat-30190 14,066 instructions
1.000073435 yes-30242 9,069,937,462 instructions
2.000204830 cat-30190 0 cycles
2.000204830 yes-30242 3,351,667,626 cycles
2.000204830 cat-30190 0 instructions
2.000204830 yes-30242 9,045,796,885 instructions
^C 2.771286639 cat-30190 0 cycles
2.771286639 yes-30242 2,593,884,166 cycles
2.771286639 cat-30190 0 instructions
2.771286639 yes-30242 7,001,171,191 instructions
It works only with -t and -p options, otherwise following error is
printed:
$ perf stat -e cycles --per-thread -I 1000 ls
The --per-thread option is only available when monitoring via -p -t options.
-p, --pid <pid> stat events on existing process id
-t, --tid <tid> stat events on existing thread id
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1435310967-14570-23-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The time out to limit the individual proc map processing was hard code
to 500ms. This patch introduce a new option --proc-map-timeout to make
the time limit configurable.
Signed-off-by: Kan Liang <kan.liang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ying Huang <ying.huang@intel.com>
Link: http://lkml.kernel.org/r/1434549071-25611-2-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Because there's too many options and I cannot read, I frequently get
confused between -c and -P, and try to do things like:
perf record -P 50000 -- foo
Which does not work; try and make the option description slightly longer
and hopefully less confusing.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20150610144850.GP19282@twins.programming.kicks-ass.net
[ Do those changes on the man page as well ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The 'perf record -s' and 'perf report -T' should be used together to see
per-thread event counts. Document the relation of these commands.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1431184784-30525-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add --no-inlines(--inlines) option to avoid searching inline functions.
Searching all functions which matches glob pattern can take a long time
and find a lot of inline functions.
With this option perf-probe searches target on the non-inlined
functions.
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20150508010333.24812.86568.stgit@localhost.localdomain
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The futex-wake benchmark only measures wakeups done within a single
process. While this has value in its own, it does not really generate
any hb->lock contention.
A new benchmark 'wake-parallel' is added, by extending the futex-wake
code such that we can measure parallel waker threads. The program output
shows the avg per-thread latency in order to complete its share of
wakeups:
Run summary [PID 13474]: blocking on 512 threads (at [private] futex 0xa88668), 8 threads waking up 64 at a time.
[Run 1]: Avg per-thread latency (waking 64/512 threads) in 0.6230 ms (+-15.31%)
[Run 2]: Avg per-thread latency (waking 64/512 threads) in 0.5175 ms (+-29.95%)
[Run 3]: Avg per-thread latency (waking 64/512 threads) in 0.7578 ms (+-18.03%)
[Run 4]: Avg per-thread latency (waking 64/512 threads) in 0.8944 ms (+-12.54%)
[Run 5]: Avg per-thread latency (waking 64/512 threads) in 1.1204 ms (+-23.85%)
Avg per-thread latency (waking 64/512 threads) in 0.7826 ms (+-9.91%)
Naturally, different combinations of numbers of blocking and waker
threads will exhibit different information.
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Davidlohr Bueso <dbueso@suse.de>
Link: http://lkml.kernel.org/r/1431110280-20231-1-git-send-email-dave@stgolabs.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
