Print the missed-branches percentage with different warning level ASCII colors,
as the percentage passes the 5%/10%/20% thresholds.
These thresholds are set to relatively low levels, because on most CPUs even a
moderate percentage of branch-misses already shows up as a slowdown.
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/n/tip-ybqukg7p86leiup7gl03ecgk@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Print the stalled-cycles percentage with different warning level ASCII colors,
as the percentage passes the 25%/50%/75% thresholds.
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/n/tip-e25zz44rcms7mu9az4fu5zp0@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The new default output looks like this:
Performance counter stats for './loop_1b_instructions':
236.010686 task-clock # 0.996 CPUs utilized
0 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
99 page-faults # 0.000 M/sec
756,487,646 cycles # 3.205 GHz
354,938,996 stalled-cycles # 46.92% of all cycles are idle
1,001,403,797 instructions # 1.32 insns per cycle
# 0.35 stalled cycles per insn
100,279,773 branches # 424.895 M/sec
12,646 branch-misses # 0.013 % of all branches
0.236902540 seconds time elapsed
We dropped cache-refs and cache-misses and added stalled-cycles - this is a
more generic "how well utilized is the CPU" metric.
If the stalled-cycles ratio is too high then more specific measurements can be
taken to figure out the source of the inefficiency.
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/n/tip-pbpl2l4mn797s69bclfpwkwn@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add stalled cycles accounting and use it to print the "cycles stalled per
instruction" value.
Also change the unit of the cycles output from M/sec to GHz - this is more
intuitive.
Prettify the output to:
Performance counter stats for './loop_1b_instructions':
239.775036 task-clock # 0.997 CPUs utilized
761,903,912 cycles # 3.178 GHz
356,620,620 stalled-cycles # 46.81% of all cycles are idle
1,001,578,351 instructions # 1.31 insns per cycle
# 0.36 stalled cycles per insn
14,782 cache-references # 0.062 M/sec
5,694 cache-misses # 38.520 % of all cache refs
0.240493656 seconds time elapsed
Also adjust the --repeat output to make the percentages align vertically:
Performance counter stats for './loop_1b_instructions' (10 runs):
236.096793 task-clock # 0.997 CPUs utilized ( +- 0.011% )
756,553,086 cycles # 3.204 GHz ( +- 0.002% )
354,942,692 stalled-cycles # 46.92% of all cycles are idle ( +- 0.008% )
1,001,389,700 instructions # 1.32 insns per cycle
# 0.35 stalled cycles per insn ( +- 0.000% )
10,166 cache-references # 0.043 M/sec ( +- 0.742% )
468 cache-misses # 4.608 % of all cache refs ( +- 13.385% )
0.236874136 seconds time elapsed ( +- 0.01% )
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/n/tip-uapziqny39601apdmmhoz7hk@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Create update_shadow_stats() which is then used in both read_counter_aggr()
and read_counter().
This not only simplifies the code but also fixes a bug: HW_CACHE_REFERENCES
was not updated in read_counter().
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/n/tip-9uc55z3g88r47exde7zxjm6p@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf stat doesn't mmap and its perfectly fine for it to use task-bound
counters with inheritance.
So set the attr.inherit on the caller and leave the syscall itself to
validate it.
When the mmap fails perf_evlist__mmap will just emit a warning if this
is the failure reason.
Reported-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110414170121.GC3229@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Now the --filter option is usable with perf stat too.
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <1300117230-8404-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch adds the ability to filter monitoring based on container groups
(cgroups) for both perf stat and perf record. It is possible to monitor
multiple cgroup in parallel. There is one cgroup per event. The cgroups to
monitor are passed via a new -G option followed by a comma separated list of
cgroup names.
The cgroup filesystem has to be mounted. Given a cgroup name, the perf tool
finds the corresponding directory in the cgroup filesystem and opens it. It
then passes that file descriptor to the kernel.
Example:
$ perf stat -B -a -e cycles:u,cycles:u,cycles:u -G test1,,test2 -- sleep 1
Performance counter stats for 'sleep 1':
2,368,667,414 cycles test1
2,369,661,459 cycles
<not counted> cycles test2
1.001856890 seconds time elapsed
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <4d590290.825bdf0a.7d0a.4890@mx.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
That was causing a SEGV on selected old distros.
Problem introduced in 7e2ed09.
Reported-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So that we don't have to pass it around to the several methods that
needs it, simplifying usage.
There is one case where we don't have the thread/cpu map in advance,
which is in the parsing routines used by top, stat, record, that we have
to wait till all options are parsed to know if a cpu or thread list was
passed to then create those maps.
For that case consolidate the cpu and thread map creation via
perf_evlist__create_maps() out of the code in top and record, while also
providing a perf_evlist__set_maps() for cases where multiple evlists
share maps or for when maps that represent CPU sockets, for instance,
get crafted out of topology information or subsets of threads in a
particular application are to be monitored, providing more granularity
in specifying which cpus and threads to monitor.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To untangle it from struct thread handling, that is tied to symbols, etc.
Right now in the python bindings I'm working on I need just a subset of
the util/ files, untangling it allows me to do that.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
As this is a per-cpu attribute, we can't set it up in advance and use it
for all the calls.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf_evsel__open now have an extra boolean argument specifying if
event grouping is desired.
The first file descriptor created on a CPU becomes the group leader.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Killing two more perf wide global variables: nr_counters and evsel_list
as a list_head.
There are more operations that will need more fields in perf_evlist,
like the pollfd for polling all the fds in a list of evsel instances.
Use option->value to pass the evsel_list to parse_{events,filters}.
LKML-Reference: <new-submission>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Using %L[uxd] has issues in some architectures, like on ppc64. Fix it
by making our 64 bit integers typedefs of stdint.h types and using
PRI[ux]64 like, for instance, git does.
Reported by Denis Kirjanov that provided a patch for one case, I went
and changed all cases.
Reported-by: Denis Kirjanov <dkirjanov@kernel.org>
Tested-by: Denis Kirjanov <dkirjanov@kernel.org>
LKML-Reference: <20110120093246.GA8031@hera.kernel.org>
Cc: Denis Kirjanov <dkirjanov@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Pingtian Han <phan@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We need to defer calling perf_evsel_list__delete() till after atexit
registered routines, because we need to traverse the events being
recorded at that time at least on 'perf record'.
This fixes the problem reported by Thomas Renninger where cmd_record
called by cmd_timechart would not write the tracing data to the perf.data
file header because the evsel_list at atexit (control+C on 'perf timechart
record') time would be empty, being already deleted by run_builtin(),
and thus 'perf timechart' when trying to process such perf.data file would
die with:
"no trace data in the file"
Problem introduced in 70d544d.
Reported-by: Thomas Renninger <trenn@suse.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
For unsupported events (e.g., H/W events when running in a VM)
perf stat currently fails with the error message:
Error: open_counter returned with 2 (No such file or directory).
/bin/dmesg may provide additional information.
Fatal: Not all events could be opened.
dmesg is of no help and it is not clear as to why it fails to
open the counter. This patch changes the error message to
Error: cache-misses event is not supported.
Fatal: Not all events could be opened.
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: a.p.zijlstra@chello.nl
LPU-Reference: <1294597272-17335-1-git-send-email-daahern@cisco.com>
Signed-off-by: David Ahern <daahern@cisco.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Since commit 69aad6f1(perf tools: Introduce event selectors), only
perf_event_attr::type and ::config are passed to event selector, which
makes perf tool not work correctly.
For example, PEBS does not work because perf_event_attr::precise_ip is
not passed to the syscall.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1294369869.20563.19.camel@minggr.sh.intel.com>
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So that later, we can pass the thread_map instance instead of
(thread_num, thread_map) for things like perf_evsel__open and friends,
just like was done with cpu_map.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So that later, we can pass the cpu_map instance instead of (nr_cpus, cpu_map)
for things like perf_evsel__open and friends.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Abstracting away the loops needed to create the various event fd handlers.
The users have to pass a confiruged perf->evsel.attr field, which is already
usable after perf_evsel__new (constructor) time, using defaults.
Comes out of the ad-hoc routines in builtin-stat, that now uses it.
Fixed a small silly bug where we were die()ing before killing our
children, dysfunctional family this one 8-)
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Making them hopefully generic enough to be used in 'perf test',
well see.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Out of ad-hoc code and global arrays with hard coded sizes.
This is the first step on having a library that will be first
used on regression tests in the 'perf test' tool.
[acme@felicio linux]$ size /tmp/perf.before
text data bss dec hex filename
1273776 97384 5104416 6475576 62cf38 /tmp/perf.before
[acme@felicio linux]$ size /tmp/perf.new
text data bss dec hex filename
1275422 97416 1392416 2765254 2a31c6 /tmp/perf.new
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch adds an option (-x/--field-separator) to print counts using a
CSV-style output. The user can pass a custom separator. This makes it very easy
to import counts directly into your favorite spreadsheet without having to
write scripts.
Example:
$ perf stat --field-separator=, -a -- sleep 1
4009.961740,task-clock-msecs
13,context-switches
2,CPU-migrations
189,page-faults
9596385684,cycles
3493659441,instructions
872897069,branches
41562,branch-misses
22424,cache-references
1289,cache-misses
Works also in non-aggregated mode:
$ perf stat -x , -a -A -- sleep 1
CPU0,1002.526168,task-clock-msecs
CPU1,1002.528365,task-clock-msecs
CPU2,1002.523360,task-clock-msecs
CPU3,1002.519878,task-clock-msecs
CPU0,1,context-switches
CPU1,5,context-switches
CPU2,5,context-switches
CPU3,6,context-switches
CPU0,0,CPU-migrations
CPU1,1,CPU-migrations
CPU2,0,CPU-migrations
CPU3,1,CPU-migrations
CPU0,2,page-faults
CPU1,6,page-faults
CPU2,9,page-faults
CPU3,174,page-faults
CPU0,2399439771,cycles
CPU1,2380369063,cycles
CPU2,2399142710,cycles
CPU3,2373161192,cycles
CPU0,872900618,instructions
CPU1,873030960,instructions
CPU2,872714525,instructions
CPU3,874460580,instructions
CPU0,221556839,branches
CPU1,218134342,branches
CPU2,218161730,branches
CPU3,218284093,branches
CPU0,18556,branch-misses
CPU1,1449,branch-misses
CPU2,3447,branch-misses
CPU3,12714,branch-misses
CPU0,8330,cache-references
CPU1,313844,cache-references
CPU2,47993728,cache-references
CPU3,826481,cache-references
CPU0,272,cache-misses
CPU1,5360,cache-misses
CPU2,1342193,cache-misses
CPU3,13992,cache-misses
This second version adds the ability to name a separator and uses
field-separator as the long option to be consistent with perf report.
Commiter note: Since we enabled --big-num by default in 201e0b0 and -x can't be
used with it, we need to notice if the user explicitely enabled or disabled -B,
add code to disable big_num if the user didn't explicitely set --big_num when
-x is used.
Cc: David S. Miller <davem@davemloft.net>
Cc: Frederik Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: paulus@samba.org
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Richter <robert.richter@amd.com>
LKML-Reference: <4cf68aa7.0fedd80a.5294.1203@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch makes several changes to "perf stat":
- "perf stat" will no longer go ahead and run the application when one or
more of the specified events could not be opened.
- Use error() and die() instead of pr_err() so that the output is more
consistent with "perf top" and "perf record".
- Handle permission errors in a more robust way, and in a similar way to
"perf record" and "perf top".
In addition, the sys_perf_event_open() error handling of "perf top" and "perf
record" is made more consistent and adds the following phrase when an event
doesn't open (with something ther than an access or permission error):
"/bin/dmesg may provide additional information."
This is added because kernel code doesn't have a good way of expressing
detailed errors to user space, so its only avenue is to use printk's. However,
many users may not think of looking at dmesg to find out why an event is being
rejected.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <fweisbec@gmail.com>
Cc: Ian Munsie <ianmunsi@au1.ibm.com>
Cc: Michael Ellerman <michaele@au1.ibm.com>
LKML-Reference: <1290217044-26293-1-git-send-email-cjashfor@linux.vnet.ibm.com>
Signed-off-by: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch adds a new -A option to perf stat. If specified then perf stat does
not aggregate counts across all monitored CPUs in system-wide mode, i.e., when
using -a. This option is not supported in per-thread mode.
Being able to get a per-cpu breakdown is useful to detect imbalances between
CPUs when running a uniform workload than spans all monitored CPUs.
The second version corrects the missing cpumap[] support, so that it works when
the -C option is used.
The third version fixes a missing cpumap[] in print_counter() and removes a
stray patch in builtin-trace.c.
Examples on a 4-way system:
# perf stat -a -e cycles,instructions -- sleep 1
Performance counter stats for 'sleep 1':
9592808135 cycles
3490380006 instructions # 0.364 IPC
1.001584632 seconds time elapsed
# perf stat -a -A -e cycles,instructions -- sleep 1
Performance counter stats for 'sleep 1':
CPU0 2398163767 cycles
CPU1 2398180817 cycles
CPU2 2398217115 cycles
CPU3 2398247483 cycles
CPU0 872282046 instructions # 0.364 IPC
CPU1 873481776 instructions # 0.364 IPC
CPU2 872638127 instructions # 0.364 IPC
CPU3 872437789 instructions # 0.364 IPC
1.001556052 seconds time elapsed
Cc: David S. Miller <davem@davemloft.net>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Richter <robert.richter@amd.com>
LKML-Reference: <4ce257b5.1e07e30a.7b6b.3aa9@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch adds a -C option to stat, record, top to designate a list of CPUs to
monitor. CPUs can be specified as a comma-separated list or ranges, no space
allowed.
Examples:
$ perf record -a -C0-1,4-7 sleep 1
$ perf top -C0-4
$ perf stat -a -C1,2,3,4 sleep 1
With perf record in per-thread mode with inherit mode on, samples are collected
only when the thread runs on the designated CPUs.
The -C option does not turn on system-wide mode automatically.
Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <4bff9496.d345d80a.41fe.7b00@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It is hard to read very large numbers so provide an option to perf stat
to separate thousands using a separator. The patch leverages the locale
support of stdio. You need to set your LC_NUMERIC appropriately, for
instance LC_NUMERIC=en_US.UTF8. You need to pass -B to activate this
feature. This way existing scripts parsing the output do not need to be
changed. Here is an example.
$ perf stat noploop 2
noploop for 2 seconds
Performance counter stats for 'noploop 2':
1998.347031 task-clock-msecs # 0.998 CPUs
61 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
118 page-faults # 0.000 M/sec
4,138,410,900 cycles # 2070.917 M/sec (scaled from 70.01%)
2,062,650,268 instructions # 0.498 IPC (scaled from 70.01%)
2,057,653,466 branches # 1029.678 M/sec (scaled from 70.01%)
40,267 branch-misses # 0.002 % (scaled from 30.04%)
2,055,961,348 cache-references # 1028.831 M/sec (scaled from 30.03%)
53,725 cache-misses # 0.027 M/sec (scaled from 30.02%)
2.001393933 seconds time elapsed
$ perf stat -B noploop 2
noploop for 2 seconds
Performance counter stats for 'noploop 2':
1998.297883 task-clock-msecs # 0.998 CPUs
59 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
119 page-faults # 0.000 M/sec
4,131,380,160 cycles # 2067.450 M/sec (scaled from 70.01%)
2,059,096,507 instructions # 0.498 IPC (scaled from 70.01%)
2,054,681,303 branches # 1028.216 M/sec (scaled from 70.01%)
25,650 branch-misses # 0.001 % (scaled from 30.05%)
2,056,283,014 cache-references # 1029.017 M/sec (scaled from 30.03%)
47,097 cache-misses # 0.024 M/sec (scaled from 30.02%)
2.001391016 seconds time elapsed
Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <4bf28fe8.914ed80a.01ca.fffff5f5@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
By default, event inheritance across fork and pthread_create was on but the -i
option of stat and record, which enabled inheritance, led to believe it was off
by default.
This patch fixes this logic by inverting the meaning of the -i option. By
default inheritance is on whether you attach to a process (-p), a thread (-t)
or start a process. If you pass -i, then you turn off inheritance. Turning off
inheritance if you don't need it, helps limit perf resource usage as well.
The patch also fixes perf stat -t xxxx and perf record -t xxxx which did not
start the counters.
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <4bea9d2f.d60ce30a.0b5b.08e1@mx.google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Parsing an option from the command line with OPT_BOOLEAN on a
bool data type would not work on a big-endian machine due to the
manner in which the boolean was being cast into an int and
incremented. For example, running 'perf probe --list' on a
PowerPC machine would fail to properly set the list_events bool
and would therefore print out the usage information and
terminate.
This patch makes OPT_BOOLEAN work as expected with a bool
datatype. For cases where the original OPT_BOOLEAN was
intentionally being used to increment an int each time it was
passed in on the command line, this patch introduces OPT_INCR
with the old behaviour of OPT_BOOLEAN (the verbose variable is
currently the only such example of this).
I have reviewed every use of OPT_BOOLEAN to verify that a true
C99 bool was passed. Where integers were used, I verified that
they were only being used for boolean logic and changed them to
bools to ensure that they would not be mistakenly used as ints.
The major exception was the verbose variable which now uses
OPT_INCR instead of OPT_BOOLEAN.
Signed-off-by: Ian Munsie <imunsie@au.ibm.com>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: <stable@kernel.org> # NOTE: wont apply to .3[34].x cleanly, please backport
Cc: Git development list <git@vger.kernel.org>
Cc: Ian Munsie <imunsie@au1.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Eric B Munson <ebmunson@us.ibm.com>
Cc: Valdis.Kletnieks@vt.edu
Cc: WANG Cong <amwang@redhat.com>
Cc: Thiago Farina <tfransosi@gmail.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Cc: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: John Kacur <jkacur@redhat.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1271147857-11604-1-git-send-email-imunsie@au.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Parameter --pid (or -p) of perf currently means a thread-wide
collection. For exmaple, if a process whose id is 8888 has 10
threads, 'perf top -p 8888' just collects the main thread
statistics. That's misleading. Users are used to attach a whole
process when debugging a process by gdb. To follow normal usage
style, the patch change --pid to process-wide collection and add
--tid (-t) to mean a thread-wide collection.
Usage example is:
# perf top -p 8888
# perf record -p 8888 -f sleep 10
# perf stat -p 8888 -f sleep 10
Above commands collect the statistics of all threads of process
8888.
Signed-off-by: Zhang Yanmin <yanmin_zhang@linux.intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Sheng Yang <sheng@linux.intel.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Jes Sorensen <Jes.Sorensen@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: zhiteng.huang@intel.com
Cc: Zachary Amsden <zamsden@redhat.com>
LKML-Reference: <1268922965-14774-3-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Command 'perf stat' doesn't enable counters when collecting an
existing (by -p) process or system-wide statistics. Fix the
issue.
Change the condition of fork/exec subcommand. If there is a
subcommand parameter, perf always forks/execs it. The usage
example is:
# perf stat -a sleep 10
So this command could collect statistics for 10 seconds
precisely. User still could stop it by CTRL+C. Without the new
capability, user could only use CTRL+C to stop it without
precise time clock.
Another issue is 'perf stat -a' consumes 100% time of a full
single logical cpu. It has a bad impact on running workload.
Fix it by adding a sleep(1) in the while(!done) loop in function
run_perf_stat.
Signed-off-by: Zhang Yanmin <yanmin_zhang@linux.intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Sheng Yang <sheng@linux.intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Jes Sorensen <Jes.Sorensen@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Zachary Amsden <zamsden@redhat.com>
Cc: <zhiteng.huang@intel.com>
LKML-Reference: <1268922965-14774-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
At present, the perf subcommands that do system-wide monitoring
(perf stat, perf record and perf top) don't work properly unless
the online cpus are numbered 0, 1, ..., N-1. These tools ask
for the number of online cpus with sysconf(_SC_NPROCESSORS_ONLN)
and then try to create events for cpus 0, 1, ..., N-1.
This creates problems for systems where the online cpus are
numbered sparsely. For example, a POWER6 system in
single-threaded mode (i.e. only running 1 hardware thread per
core) will have only even-numbered cpus online.
This fixes the problem by reading the /sys/devices/system/cpu/online
file to find out which cpus are online. The code that does that is in
tools/perf/util/cpumap.[ch], and consists of a read_cpu_map()
function that sets up a cpumap[] array and returns the number of
online cpus. If /sys/devices/system/cpu/online can't be read or
can't be parsed successfully, it falls back to using sysconf to
ask how many cpus are online and sets up an identity map in cpumap[].
The perf record, perf stat and perf top code then calls
read_cpu_map() in the system-wide monitoring case (instead of
sysconf) and uses cpumap[] to get the cpu numbers to pass to
perf_event_open.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
current pid option doesn't work for perf stat. Change it to what
perf record --pid acts as.
Signed-off-by: Liming Wang <liming.wang@windriver.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1262246750-2191-1-git-send-email-liming.wang@windriver.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The ratio between the number of events and the time elapsed makes
sense only if task-clock event is counted. Otherwise it will be
simply a (confusing)
# 0.000 M/sec
This patch outputs the ratio only if task-clock event is counted.
Some test examples of before and after:
Before:
[lucas@skywalker linux.trees.git]$ sudo perf stat -e branch-misses -a -- sleep 1
Performance counter stats for 'sleep 1':
1367818 branch-misses # 0.000 M/sec
1.001494325 seconds time elapsed
After (without task-clock):
[lucas@skywalker perf]$ sudo ./perf stat -e branch-misses -a -- sleep 1
Performance counter stats for 'sleep 1':
1135044 branch-misses
1.001370775 seconds time elapsed
After (with task-clock):
[lucas@skywalker perf]$ sudo ./perf stat -e branch-misses -e task-clock -a -- sleep 1
Performance counter stats for 'sleep 1':
1070111 branch-misses # 0.534 M/sec
2002.730893 task-clock-msecs # 1.999 CPUs
1.001640292 seconds time elapsed
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20091115140507.GB21561@skywalker.lan>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Adds performance event information about branches
and branch misses to the default output of perf stat.
Signed-off-by: Tim Blechmann <tim@klingt.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <4ADC3975.8050109@klingt.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When we count both branches and branch-misses it is useful to
print out the percentage of branch-misses:
# perf stat -e branches -e branch-misses /bin/true
Performance counter stats for '/bin/true':
401684 branches # 0.000 M/sec
23301 branch-misses # 5.801 %
Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: paulus@samba.org
Cc: a.p.zijlstra@chello.nl
LKML-Reference: <20091018112923.GQ4808@kryten>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
If we launch the child on behalf of the user, ensure that it dies
along with ourselves when we are interrupted.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
LKML-Reference: <1254616502-4728-1-git-send-email-chris@chris-wilson.co.uk>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Remove some, now useless, global storage.
Don't calculate the stddev when not needed.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Use the more advanced single pass variance algorithm outlined
on the wikipedia page. This is numerically more stable for
larger sample sets.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When we're computing the mean by sampling the distribution,
then the std dev of the mean is related to the std dev of the
sample set by:
stddev_mean = std_dev / sqrt(N)
Which is exactly what we want.
This results in the error on the mean decreasing with
increasing number of samples.
Also fix the scaled == -1, aka not counted case.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Since we don't need all the individual samples to calculate the
error remove both the limit and the storage overhead associated
with that.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The current noise computation does:
\Sum abs(n_i - avg(n)) * N^-1.5
Which is (afaik) not a regular noise function, and needs the
complete sample set available to post-process.
Change this to use a regular stddev computation which can be
done by keeping a two sums:
stddev = sqrt( 1/N (\Sum n_i^2) - avg(n)^2 )
For which we only need to keep \Sum n_i and \Sum n_i^2.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <stable@kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Librarize trace_event() helper so that perf trace can use it
too. Also clean up the debug.h includes a bit.
It's not good to have it included in perf.h because it doesn't
make it flexible against other headers it may need (headers
that can also depend on perf.h and then create a recursive
header dependency).
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1250453149-664-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Factorize multiple definitions of high level dso helpers into the
symbol source file.
The side effect is a general export of the verbose and eprintf
debugging helpers into a new file dedicated to debugging purposes.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Brice Goglin <Brice.Goglin@inria.fr>
We want to use a coherent flag for -S/--stat across all tools,
so free up -S in perf stat.
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: paulus@samba.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf stat and perf record currently look for all options on the command
line. This can lead to some confusion:
# perf stat ls -l
Error: unknown switch `l'
While we can work around this by adding '--' before the command, the git
option parsing code can stop at the first non option:
# perf stat ls -l
Performance counter stats for 'ls -l':
....
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20090722130412.GD9029@kryten>
Building builtin-stat.c reports the following errors:
cc1: warnings being treated as errors
builtin-stat.c: In function ‘run_perf_stat’:
builtin-stat.c:242: erreur: ignoring return value of ‘read’, declared with attribute warn_unused_result
builtin-stat.c:255: erreur: ignoring return value of ‘read’, declared with attribute warn_unused_result
make: *** [builtin-stat.o] Erreur 1
This patch handles the possible pipe read failures.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <1246474930-6088-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
MATCH_EVENT is useful:
1. for multiple attrs checking
2. avoid repetition of PERF_TYPE_ and PERF_COUNT_ and save space
3. avoids line breakage
Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1246440909.3403.5.camel@hpdv5.satnam>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Enable -Wextra. This found a few real bugs plus a number
of signed/unsigned type mismatches/uncleanlinesses. It
also required a few annotations
All things considered it was still worth it so lets try with
this enabled for now.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This provides a way to mark a counter to be enabled on the next
exec. This is useful for measuring the total activity of a
program without including overhead from the process that
launches it.
This also changes the perf stat command to use this new
facility.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <19017.43927.838745.689203@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Vince Weaver reported a 'perf stat' measurement overhead in the
count of retired instructions, which can amount to a +6000
instructions inflated count in the reported count.
At present, perf stat creates its counters on the perf process. Thus
the counters count the fork and various other activity in both the
parent and child, such as the resolver overhead for resolving PLT
entries for any libc functions that haven't been called before, such
as execvp.
This reduces the overhead by creating the counters on the child process
after the fork, using a couple of pipes to synchronize so that the
child process waits until the parent has created the counters before
doing the exec. To eliminate the PLT resolution overhead on calling
execvp, this does a dummy execvp first which will always fail.
With this, the overhead of executing a program goes down from over
4800 instructions to about 90 instructions on powerpc (32-bit).
This was measured with a statically-linked program written in
assembler which only does the 3 instructions needed to call _exit(0).
Before:
$ perf stat -e 0:1:u ./three
Performance counter stats for './three':
4858 instructions
0.001274523 seconds time elapsed
After:
$ perf stat -e 0:1:u ./three
Performance counter stats for './three':
92 instructions
0.000468153 seconds time elapsed
Reported-by: Vince Weaver <vince@deater.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <19016.41425.814043.870352@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter expressed a strong preference for percentage based
display of scaled values - so revert to that from the
recently introduced multiplication-factor unit.
Reported-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Jaswinder Singh Rajput <jaswinder@kernel.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Set attrs and nr_counters if no event is selected and !null_run.
Setting of attrs should depend on number of counters,
so we need to memcpy only for sizeof(default_attrs)
Also set nr_counters as ARRAY_SIZE(default_attrs) in place of
hardcoded value.
Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1246126749.32198.16.camel@hpdv5.satnam>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In multi-run (-r/--repeat) printouts, print out the noise of
the wall-clock average as well.
Also, fix a bug in printing out scaled counters: if it was not
scaled then we should not update the average with -1.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Allow a no-counters run. This can be useful to measure just
elapsed wall-clock time - or to assess the raw overhead of perf
stat itself, without running any counters.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Remove dead code and do some code alignment.
Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1245847774.2681.2.camel@ht.satnam>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Error message should use stderr for verbose (-v), otherwise
message will be lost for:
$ ./perf stat -v <cmd> > /dev/null
For example on AMD bus-cycles event is not available so now
it looks like:
$ ./perf stat -v -e bus-cycles ls > /dev/null
Error: counter 0, sys_perf_counter_open() syscall returned with -1 (Invalid argument)
Performance counter stats for 'ls':
<not counted> bus-cycles
0.006765877 seconds time elapsed.
Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <1245757369.3776.1.camel@localhost.localdomain>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
On 64-bit powerpc, __u64 is defined to be unsigned long rather than
unsigned long long. This causes compiler warnings every time we
print a __u64 value with %Lx.
Rather than changing __u64, we define our own u64 to be unsigned long
long on all architectures, and similarly s64 as signed long long.
For consistency we also define u32, s32, u16, s16, u8 and s8. These
definitions are put in a new header, types.h, because these definitions
are needed in util/string.h and util/symbol.h.
The main change here is the mechanical change of __[us]{64,32,16,8}
to remove the "__". The other changes are:
* Create types.h
* Include types.h in perf.h, util/string.h and util/symbol.h
* Add types.h to the LIB_H definition in Makefile
* Added (u64) casts in process_overflow_event() and print_sym_table()
to kill two remaining warnings.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: benh@kernel.crashing.org
LKML-Reference: <19003.33494.495844.956580@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add the --repeat <n> feature to perf stat, which repeats a given
command up to a 100 times, collects the stats and calculates an
average and a stddev.
For example, the following oneliner 'perf stat' command runs hackbench
5 times and prints a tabulated result of all metrics, with averages
and noise levels (in percentage) printed:
aldebaran:~/linux/linux/tools/perf> ./perf stat --repeat 5 ~/hackbench 10
Time: 0.117
Time: 0.108
Time: 0.089
Time: 0.088
Time: 0.100
Performance counter stats for '/home/mingo/hackbench 10' (5 runs):
1243.989586 task-clock-msecs # 10.460 CPUs ( +- 4.720% )
47706 context-switches # 0.038 M/sec ( +- 19.706% )
387 CPU-migrations # 0.000 M/sec ( +- 3.608% )
17793 page-faults # 0.014 M/sec ( +- 0.354% )
3770941606 cycles # 3031.329 M/sec ( +- 4.621% )
1566372416 instructions # 0.415 IPC ( +- 2.703% )
16783421 cache-references # 13.492 M/sec ( +- 5.202% )
7128590 cache-misses # 5.730 M/sec ( +- 7.420% )
0.118924455 seconds time elapsed.
The goal of this feature is to allow the reliance on these accurate
statistics and to know how many times a command has to be repeated
for the noise to go down to an acceptable level.
(The -v option can be used to see a line printed out as each run progresses.)
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
- use IPC for the instruction normalization output
- CPUs for the CPU utilization factor value.
- print out time elapsed like the other rows
- tidy up the task-clocks/cpu-clocks printout
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Pure renames only, to PERF_COUNT_HW_* and PERF_COUNT_SW_*.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Currently report and stat catch SIGINT (and others) without altering
their exit state. This means that things like:
while :; do perf stat ./foo ; done
Loops become hard-to-interrupt, because bash never sees perf terminate
due to interruption. Fix this.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Before:
$ perf stat ~/hackbench 5
error: syscall returned with -1 (No such device)
After:
$ perf stat ~/hackbench 5
Time: 1.640
Performance counter stats for '/home/mingo/hackbench 5':
6524.570382 task-clock-ticks # 3.838 CPU utilization factor
35704 context-switches # 0.005 M/sec
191 CPU-migrations # 0.000 M/sec
8958 page-faults # 0.001 M/sec
<not counted> cycles
<not counted> instructions
<not counted> cache-references
<not counted> cache-misses
Wall-clock time elapsed: 1699.999995 msecs
Also add -v (--verbose) option to allow the printing of failed
counter opens.
Plus dont print 'inf' if wall-time is zero (due to jiffies granularity),
instead skip the printing of the CPU utilization factor.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Several people have suggested that 'perf' has become a full-fledged
tool that should be moved out of Documentation/. Move it to the
(new) tools/ directory.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
