Re: perf_event_open() manpage

Re: perf_event_open() manpage

[Michael Kerrisk (man-pages)]

Hi Vince.

>> After you've done a (quick?) pass through, we can send any revised
>> version you have out for wider comment. Sound okay?
>
> I've made a quick pass through and fixed some of the FIXME's.  In the
> process I merged in some changes I had been working on since the last time
> I submitted, but they were minor.
>
> One of your FIXMEs was to add the PR_ASK_PERF_EVENT_* prctl() calls
> to the prctl() manpage, but as far as I can tell this was already done.

True. The thing was you had a couple of hanging mentions of these
flags. I'll fix that.

> I'm ready to send things out for further comment if you are.

Who do you propose as individuals and lists?. Ingo Molnar and Peter
Zijlstra seem the obvious people to ask. I'm nor sure if there others
who should be explicitly CCed (Stephane Eranian?). ANy other lists
than linux-kernel?

I'll do some tweaking of the page and have a new version soon.

Thanks,

Michael
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: perf_event_open() manpage

[Michael Kerrisk (man-pages)]

> who should be explicitly CCed (Stephane Eranian?). ANy other lists

D'oh! Stephane is already CCed on this thread :-}

-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Author of "The Linux Programming Interface"; http://man7.org/tlpi/
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: perf_event_open() manpage

[Vince Weaver]

On Mon, 22 Oct 2012, Michael Kerrisk (man-pages) wrote:

> Who do you propose as individuals and lists?. Ingo Molnar and Peter
> Zijlstra seem the obvious people to ask. I'm nor sure if there others
> who should be explicitly CCed (Stephane Eranian?). ANy other lists
> than linux-kernel?

When sending perf_event related patches to the kernel I ususally include
the maintainers plus Stephane:
  
  linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
  Peter Zijlstra <a.p.zijlstra-/NLkJaSkS4VmR6Xm/wNWPw@public.gmane.org>
  Paul Mackerras <paulus-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org>
  Ingo Molnar <mingo-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
  Arnaldo Carvalho de Melo <acme-f8uhVLnGfZaxAyOMLChx1axOck334EZe@public.gmane.org>
  Stephane Eranian <eranian-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>

There's also the 
   linux-perf-users-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
list which was often more responsive than the maintainers when I was 
trying to track down various previously undocumented perf_event corner 
cases.

Thanks,

Vince

--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: perf_event_open() manpage

[Michael Kerrisk (man-pages)]

Hi Vince

I made a handful of minor edits. See the new version below.

On Tue, Oct 23, 2012 at 5:32 AM, Vince Weaver <vincent.weaver-e7X0jjDqjFGHXe+LvDLADg@public.gmane.org> wrote:
> On Mon, 22 Oct 2012, Michael Kerrisk (man-pages) wrote:
>
>> Who do you propose as individuals and lists?. Ingo Molnar and Peter
>> Zijlstra seem the obvious people to ask. I'm nor sure if there others
>> who should be explicitly CCed (Stephane Eranian?). ANy other lists
>> than linux-kernel?
>
> When sending perf_event related patches to the kernel I ususally include
> the maintainers plus Stephane:
>
>   linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>   Peter Zijlstra <a.p.zijlstra-/NLkJaSkS4VmR6Xm/wNWPw@public.gmane.org>
>   Paul Mackerras <paulus-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org>
>   Ingo Molnar <mingo-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>   Arnaldo Carvalho de Melo <acme-f8uhVLnGfZaxAyOMLChx1axOck334EZe@public.gmane.org>
>   Stephane Eranian <eranian-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>
> There's also the
>    linux-perf-users-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> list which was often more responsive than the maintainers when I was
> trying to track down various previously undocumented perf_event corner
> cases.

That sounds like a reasonable list; I'd add linux-man@ of course ;-).

Would you like to send this out, or shall I? I'd prefer the former,
since I have very limited time to drive the review myself.

In the mail out, as well as asking for general review, it would be
good to mention that help is specifically needed for the pieces marked
FIXME and "[To be documented]"

Cheers,

Michael


.\" Hey Emacs! This file is -*- nroff -*- source.
.\"
.\" Copyright (c) 2012, Vincent Weaver
.\"
.\" This is free documentation; you can redistribute it and/or
.\" modify it under the terms of the GNU General Public License as
.\" published by the Free Software Foundation; either version 2 of
.\" the License, or (at your option) any later version.
.\"
.\" The GNU General Public License's references to "object code"
.\" and "executables" are to be interpreted as the output of any
.\" document formatting or typesetting system, including
.\" intermediate and printed output.
.\"
.\" This manual is distributed in the hope that it will be useful,
.\" but WITHOUT ANY WARRANTY; without even the implied warranty of
.\" MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
.\" GNU General Public License for more details.
.\"
.\" You should have received a copy of the GNU General Public
.\" License along with this manual; if not, see
.\" <http://www.gnu.org/licenses/>.
.\"
.\" This document is based on the perf_event.h header file, the
.\" tools/perf/design.txt file, and a lot of bitter experience.
.\"
.TH PERF_EVENT_OPEN 2 2012-10-22 "Linux" "Linux Programmer's Manual"
.SH NAME
perf_event_open \- set up performance monitoring
.SH SYNOPSIS
.nf
.B #include <linux/perf_event.h>
.B #include <linux/hw_breakpoint.h>
.sp
.BI "int perf_event_open(struct perf_event_attr *" hw_event ,
.BI "                    pid_t " pid ", int " cpu ", int " group_fd ,
.BI "                    unsigned long " flags  );
.fi

.IR Note :
There is no glibc wrapper for this system call; see NOTES.
.SH DESCRIPTION
Given a list of parameters,
.BR perf_event_open ()
returns a file descriptor, for use in subsequent system calls
.RB ( read "(2), " mmap "(2), " prctl "(2), " fcntl "(2), etc.)."
.PP
A call to
.BR perf_event_open ()
creates a file descriptor that allows measuring performance
information.
Each file descriptor corresponds to one
event that is measured; these can be grouped together
to measure multiple events simultaneously.
.PP
Events can be enabled and disabled in two ways: via
.BR ioctl (2)
and via
.BR prctl (2) .
When an event is disabled it does not count or generate overflows but does
continue to exist and maintain its count value.
Events come in two flavors: counting and sampled.
A
.I counting
event is one that is used for counting the aggregate number of events
that occur.
In general, counting event results are gathered with a
.BR read (2)
call.
A
.I sampling
event periodically writes measurements to a buffer that can then
be accessed via
.BR  mmap (2) .
.SS Arguments
.P
The argument
.I pid
allows events to be attached to processes in various ways.
If
.I pid
is 0, measurements happen on the current task, if
.I pid
is greater than 0, the process indicated by
.I pid
is measured, and if
.I pid
is less than 0, all processes are counted.

The
.I cpu
argument allows measurements to be specific to a CPU.
If
.I cpu
is greater than or equal to 0,
measurements are restricted to the specified CPU;
if
.I cpu
is \-1, the events are measured on all CPUs.
.P
Note that the combination of
.IR pid " == \-1"
and
.IR cpu " == \-1"
is not valid.
.P
A
.IR pid " > 0"
and
.IR cpu " == \-1"
setting measures per-process and follows that process to whatever CPU the
process gets scheduled to.
Per-process events can be created by any user.
.P
A
.IR pid " == \-1"
and
.IR cpu " >= 0"
setting is per-CPU and measures all processes on the specified CPU.
Per-CPU events need the
.B CAP_SYS_ADMIN
capability.
.P
The
.I group_fd
argument allows counter groups to be set up.
A counter group has one counter which is the group leader.
The leader is created first, with
.IR group_fd " = \-1"
in the
.BR perf_event_open ()
call that creates it.
The rest of the group members are created subsequently, with
.IR group_fd
giving the fd of the group leader.
(A single counter on its own is created with
.IR group_fd " = \-1"
and is considered to be a group with only 1 member.)
.P
A counter group is scheduled onto the CPU as a unit: it will only
be put onto the CPU if all of the counters in the group can be put onto
the CPU.
This means that the values of the member counters can be
meaningfully compared, added, divided (to get ratios), etc., with each
other, since they have counted events for the same set of executed
instructions.
.P
The
.I flags
argument takes one of the following values:
.TP
.BR PERF_FLAG_FD_NO_GROUP
.\" FIXME The following sentence is unclear
This flag allows creating an event as part of an event group but
having no group leader.
It is unclear why this is useful.
.\" FIXME So, why is it useful?
.TP
.BR PERF_FLAG_FD_OUTPUT
This flag re-routes the output from an event to the group leader.
.TP
.BR PERF_FLAG_PID_CGROUP " (Since Linux 2.6.39)."
This flag activates per-container system-wide monitoring.
A container
is an abstraction that isolates a set of resources for finer grain
control (CPUs, memory, etc...).
In this mode, the event is measured
only if the thread running on the monitored CPU belongs to the designated
container (cgroup).
The cgroup is identified by passing a file descriptor
opened on its directory in the cgroupfs filesystem.
For instance, if the
cgroup to monitor is called
.IR test ,
then a file descriptor opened on
.I /dev/cgroup/test
(assuming cgroupfs is mounted on
.IR /dev/cgroup )
must be passed as the
.I pid
parameter.
cgroup monitoring is only available
for system-wide events and may therefore require extra permissions.
.P
The
.I perf_event_attr
structure is what is passed into the
.BR perf_event_open ()
syscall.
It is large and has a complicated set of dependent fields.

.in +4n
.nf
struct perf_event_attr {
    __u32     type;         /* Type of event */
    __u32     size;         /* Size of attribute structure */
    __u64     config;       /* Type-specific configuration */

    union {
        __u64 sample_period;    /* Period of sampling */
        __u64 sample_freq;      /* Frequency of sampling */
    };

    __u64     sample_type;  /* Specifies values included in sample */
    __u64     read_format;  /* Specifies values returned in read */

    __u64     disabled       : 1,   /* off by default         */
              inherit        : 1,   /* children inherit it    */
              pinned         : 1,   /* must always be on PMU  */
              exclusive      : 1,   /* only group on PMU      */
              exclude_user   : 1,   /* don't count user       */
              exclude_kernel : 1,   /* don't count kernel     */
              exclude_hv     : 1,   /* don't count hypervisor */
              exclude_idle   : 1,   /* don't count when idle  */
              mmap           : 1,   /* include mmap data      */
              comm           : 1,   /* include comm data      */
              freq           : 1,   /* use freq, not period   */
              inherit_stat   : 1,   /* per task counts        */
              enable_on_exec : 1,   /* next exec enables      */
              task           : 1,   /* trace fork/exit        */
              watermark      : 1,   /* wakeup_watermark       */
              precise_ip     : 2,   /* skid constraint        */
              mmap_data      : 1,   /* non-exec mmap data     */
              sample_id_all  : 1,   /* sample_type all events */
              exclude_host   : 1,   /* don't count in host    */
              exclude_guest  : 1,   /* don't count in guest   */
              __reserved_1   : 43;

    union {
        __u32 wakeup_events;    /* wakeup every n events */
        __u32 wakeup_watermark; /* bytes before wakeup   */
    };

    __u32     bp_type;          /* breakpoint type */

    union {
        __u64 bp_addr;          /* breakpoint address */
        __u64 config1;          /* extension of config */
    };

    union {
        __u64 bp_len;           /* breakpoint length */
        __u64 config2;          /* extension of config1 */
    };
    __u64     branch_sample_type;   /* enum branch_sample_type */
};
.fi
.in

The fields of the
.I perf_event_attr
structure are described in more detail below.

.TP
.I type
This field specifies the overall event type.
It has one of the following values:
.RS
.TP
.B PERF_TYPE_HARDWARE
This indicates one of the "generalized" hardware events provided
by the kernel.
See the
.I config
field definition for more details.
.TP
.B PERF_TYPE_SOFTWARE
This indicates one of the software-defined events provided by the kernel
(even if no hardware support is available).
.TP
.B PERF_TYPE_TRACEPOINT
This indicates a tracepoint
provided by the kernel tracepoint infrastructure.
.TP
.B PERF_TYPE_HW_CACHE
This indicates a hardware cache event.
This has a special encoding, described in the
.I config
field definition.
.TP
.B PERF_TYPE_RAW
This indicates a "raw" implementation-specific event in the
.IR config " field."
.TP
.BR PERF_TYPE_BREAKPOINT " (Since Linux 2.6.33)"
This indicates a hardware breakpoint as provided by the CPU.
Breakpoints can be read/write accesses to an address as well as
execution of an instruction address.
.TP
.RB "dynamic PMU"
Since Linux 2.6.39,
.BR perf_event_open()
can support multiple PMUs.
To enable this, a value exported by the kernel can be used in the
.I type
field to indicate which PMU to use.
The value to use can be found in the sysfs filesystem:
there is a subdirectory per PMU instance under
.IR /sys/devices .
In each sub-directory there is a
.I type
file whose content is an integer that can be used in the
.I type
field.
For instance,
.I /sys/devices/cpu/type
contains the value for the core CPU PMU, which is usually 4.
.RE

.TP
.I "size"
The size of the
.I perf_event_attr
structure for forward/backward compatibility.
Set this using
.I sizeof(struct perf_event_attr)
to allow the kernel to see
the struct size at the time of compilation.

The related define
.B PERF_ATTR_SIZE_VER0
is set to 64; this was the size of the first published struct.
.B PERF_ATTR_SIZE_VER1
is 72, corresponding to the addition of breakpoints in Linux 2.6.33.
.B PERF_ATTR_SIZE_VER2
is 80 corresponding to the addition of branch sampling in Linux 3.4.

.TP
.I "config"
This specifies which event you want, in conjunction with
the
.I type
field.
The
.IR config1 " and " config2
fields are also taken into account in cases where 64 bits is not
enough to fully specify the event.
The encoding of these fields are event dependent.

The most significant bit (bit 63) of
.I config
signifies CPU-specific (raw) counter configuration data;
if the most significant bit is unset, the next 7 bits are an event
type and the rest of the bits are the event identifier.

There are various ways to set the
.I config
field that are dependent on the value of the previously
described
.I type
field.
What follows are various possible settings for
.I config
separated out by
.IR type .

If
.I type
is
.BR PERF_TYPE_HARDWARE ,
we are measuring one of the generalized hardware CPU events.
Not all of these are available on all platforms.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_HW_CPU_CYCLES
Total cycles.
Be wary of what happens during CPU frequency scaling
.TP
.B PERF_COUNT_HW_INSTRUCTIONS
Retired instructions.
Be careful, these can be affected by various
issues, most notably hardware interrupt counts
.TP
.B PERF_COUNT_HW_CACHE_REFERENCES
Cache accesses.
Usually this indicates Last Level Cache accesses but this may
vary depending on your CPU.
This may include prefetches and coherency messages; again this
depends on the design of your CPU.
.TP
.B PERF_COUNT_HW_CACHE_MISSES
Cache misses.
Usually this indicates Last Level Cache misses; this is intended to be
used in conjunction with the
.B PERF_COUNT_HW_CACHE_REFERENCES
event to calculate cache miss rates.
.TP
.B PERF_COUNT_HW_BRANCH_INSTRUCTIONS
Retired branch instructions.
Prior to Linux 2.6.34, this used
the wrong event on AMD processors.
.TP
.B PERF_COUNT_HW_BRANCH_MISSES
Mispredicted branch instructions.
.TP
.B PERF_COUNT_HW_BUS_CYCLES
Bus cycles, which can be different from total cycles.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_FRONTEND " (Since Linux 3.0)"
Stalled cycles during issue.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_BACKEND  " (Since Linux 3.0)"
Stalled cycles during retirement.
.TP
.BR PERF_COUNT_HW_REF_CPU_CYCLES  " (Since Linux 3.3)"
Total cycles; not affected by CPU frequency scaling.
.RE
.IP
If
.I type
is
.BR PERF_TYPE_SOFTWARE ,
we are measuring software events provided by the kernel.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_SW_CPU_CLOCK
This reports the CPU clock, a high-resolution per-CPU timer.
.TP
.B PERF_COUNT_SW_TASK_CLOCK
This reports a clock count specific to the task that is running.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS
This reports the number of page faults.
.TP
.B PERF_COUNT_SW_CONTEXT_SWITCHES
This counts context switches.
Until Linux 2.6.34, these were all reported as user-space
events, after that they are reported as happening in the kernel.
.TP
.B PERF_COUNT_SW_CPU_MIGRATIONS
This reports the number of times the process
has migrated to a new CPU.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MIN
This counts the number of minor page faults.
These did not require disk I/O to handle.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MAJ
This counts the number of major page faults.
These required disk I/O to handle.
.TP
.BR PERF_COUNT_SW_ALIGNMENT_FAULTS " (Since Linux 2.6.33)"
This counts the number of alignment faults.
These happen when unaligned memory accesses happen; the kernel
can handle these but it reduces performance.
This only happens on some architectures (never on x86).
.TP
.BR PERF_COUNT_SW_EMULATION_FAULTS " (Since Linux 2.6.33)"
This counts the number of emulation faults.
The kernel sometimes traps on unimplemented instructions
and emulates them for userspace.
This can negatively impact performance.
.RE
.RE


.RS
If
.I type
is
.BR PERF_TYPE_TRACEPOINT ,
then we are measuring kernel tracepoints.
The value to use in
.I config
can be obtained from under debugfs
.I tracing/events/*/*/id
if ftrace is enabled in the kernel.

.RE

.RS
If
.I type
is
.BR PERF_TYPE_HW_CACHE ,
then we are measuring a hardware CPU cache event.
To calculate the appropriate
.I config
value use the following equation:
.RS 4
.nf

    (perf_hw_cache_id) | (perf_hw_cache_op_id << 8) |
    (perf_hw_cache_op_result_id << 16)
.fi
.P
where
.I perf_hw_cache_id
is one of:
.RS
.TP
.B PERF_COUNT_HW_CACHE_L1D
for measuring Level 1 Data Cache
.TP
.B PERF_COUNT_HW_CACHE_L1I
for measuring Level 1 Instruction Cache
.TP
.B PERF_COUNT_HW_CACHE_LL
for measuring Last-Level Cache
.TP
.B PERF_COUNT_HW_CACHE_DTLB
for measuring the Data TLB
.TP
.B PERF_COUNT_HW_CACHE_ITLB
for measuring the Instruction TLB
.TP
.B PERF_COUNT_HW_CACHE_BPU
for measuring the branch prediction unit
.TP
.BR PERF_COUNT_HW_CACHE_NODE " (Since Linux 3.0)"
for measuring local memory accesses
.RE

.P
and
.I perf_hw_cache_op_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_OP_READ
for read accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_WRITE
for write accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_PREFETCH
for prefetch accesses
.RE

.P
and
.I perf_hw_cache_op_result_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_RESULT_ACCESS
to measure accesses
.TP
.B PERF_COUNT_HW_CACHE_RESULT_MISS
to measure misses
.RE
.RE

If
.I type
is
.BR PERF_TYPE_RAW ,
then a custom "raw"
.I config
value is needed.
Most CPUs support events that are not covered by the "generalized" events.
These are implementation defined; see your CPU manual (for example
the Intel Volume 3B documentation or the AMD BIOS and Kernel Developer
Guide).
The libpfm4 library can be used to translate from the name in the
architectural manuals to the raw hex value
.BR perf_event_open ()
expects in this field.

If
.I type
is
.BR PERF_TYPE_BREAKPOINT ,
then leave
.I config
set to zero.
Its parameters are set in other places.
.RE
.TP
.IR sample_period ", " sample_freq
A "sampling" counter is one that generates an interrupt
every N events, where N is given by
.IR sample_period .
A sampling counter has
.IR sample_period " > 0."
The
.I sample_type
field controls what data is recorded on each interrupt.

.I sample_freq
can be used if you wish to use frequency rather than period.
In this case you set the
.I freq
flag.
The kernel will adjust the sampling period
to try and achieve the desired rate.
The rate of adjustment is a
timer tick.


.TP
.I "sample_type"
The various bits in this field specify which values to include
in the overflow packets.
They will be recorded in a ring-buffer,
which is available to user-space using
.BR mmap (2).
The order in which the values are saved in the
overflow packets as documented in the MMAP Layout subsection below;
it is not the
.I "enum perf_event_sample_format"
order.
.RS
.TP
.B PERF_SAMPLE_IP
instruction pointer
.TP
.B PERF_SAMPLE_TID
thread id
.TP
.B PERF_SAMPLE_TIME
time
.TP
.B PERF_SAMPLE_ADDR
address
.TP
.B PERF_SAMPLE_READ
[To be documented]
.TP
.B PERF_SAMPLE_CALLCHAIN
[To be documented]
.TP
.B PERF_SAMPLE_ID
[To be documented]
.TP
.B PERF_SAMPLE_CPU
[To be documented]
.TP
.B PERF_SAMPLE_PERIOD
[To be documented]
.TP
.B PERF_SAMPLE_STREAM_ID
[To be documented]
.TP
.B PERF_SAMPLE_RAW
[To be documented]
.TP
.BR PERF_SAMPLE_BRANCH_STACK " (Since Linux 3.4)"
[To be documented]
.RE

.TP
.IR "read_format"
This field specifies the format of the data returned by
.BR read (2)
on a
.BR perf_event_open()
file descriptor.
.RS
.TP
.B PERF_FORMAT_TOTAL_TIME_ENABLED
Adds the 64-bit "time_enabled" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and multiplexing is happening.
.TP
.B PERF_FORMAT_TOTAL_TIME_RUNNING
Adds the 64-bit "time_running" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and  multiplexing is happening.
.TP
.B PERF_FORMAT_ID
Adds a 64-bit unique value that corresponds to the event-group.
.TP
.B PERF_FORMAT_GROUP
Allows all counter values in an event-group to be read with one read.
.RE

.TP
.IR "disabled"
The
.I disabled
bit specifies whether the counter starts out disabled or enabled.
If disabled, the event can later be enabled by
.BR ioctl (2),
.BR prctl (2),
or
.IR enable_on_exec .

.TP
.IR "inherit"
The
.I inherit
bit specifies that this counter should count events of child
tasks as well as the task specified.
This only applies to new children, not to any existing children at
the time the counter is created (nor to any new children of
existing children).

Inherit does not work for some combinations of
.IR read_format s,
such as
.BR PERF_FORMAT_GROUP .

.TP
.IR "pinned"
The
.I pinned
bit specifies that the counter should always be on the CPU if at all
possible.
It only applies to hardware counters and only to group leaders.
If a pinned counter cannot be put onto the CPU (e.g., because there are
not enough hardware counters or because of a conflict with some other
event), then the counter goes into an 'error' state, where reads
return end-of-file (i.e.,
.BR read (2)
returns 0) until the counter is subsequently enabled or disabled.

.TP
.IR "exclusive"
The
.I exclusive
bit specifies that when this counter's group is on the CPU,
it should be the only group using the CPU's counters.
In the future this may allow monitoring programs to
support PMU features that need to run alone so that they do not
disrupt other hardware counters.

.TP
.IR "exclude_user"
If this bit is set, the count excludes events that happen in user-space.

.TP
.IR "exclude_kernel"
If this bit is set, the count excludes events that happen in kernel-space.

.TP
.IR "exclude_hv"
If this bit is set, the count excludes events that happen in the
hypervisor.
This is mainly for PMUs that have built-in support for handling this
(such as POWER).
Extra support is needed for handling hypervisor measurements on most
machines.

.TP
.IR "exclude_idle"
If set, don't count when the CPU is idle.

.TP
.IR "mmap"
The
.I mmap
bit enables recording of extra information to a
mmap'd ring-buffer.
This is
described below in subsection MMAP Layout.

.TP
.IR "comm"
The
.I comm
bit enables tracking of process command name as modified by the
.IR exec (2)
and
.IR prctl (PR_SET_NAME)
system calls.
Unfortunately for tools,
there is no way to distinguish one system call versus the other.

.TP
.IR "freq"
If this bit is set, then
.I sample_frequency
not
.I sample_period
is used when setting up the sampling interval.

.TP
.IR "inherit_stat"
This bit enables saving of event counts on context switch for
inherited tasks.
This is only meaningful if the
.I inherit
field is set.

.TP
.IR "enable_on_exec"
If this bit is set, a counter is automatically
enabled after a call to
.BR exec (2).

.TP
.IR "task"
If this bit is set, then
fork/exit notifications are included in the ring buffer.

.TP
.IR "watermark"
If set, have a sampling interrupt happen when we cross the
wakeup_watermark boundary.

.TP
.IR "precise_ip" " (Since Linux 2.6.35)"
This controls the amount of skid.
Skid is how many instructions
execute between an event of interest happening and the kernel
being able to stop and record the event.
Smaller skid is
better and allows more accurate reporting of which events
correspond to which instructions, but hardware is often limited
with how small this can be.

The values of this are the following:
.RS
.TP
0 -
.B SAMPLE_IP
can have arbitrary skid
.TP
1 -
.B SAMPLE_IP
must have constant skid
.TP
2 -
.B SAMPLE_IP
requested to have 0 skid
.TP
3 -
.B SAMPLE_IP
must have 0 skid.
See also
.BR PERF_RECORD_MISC_EXACT_IP .
.RE

.TP
.IR "mmap_data" " (Since Linux 2.6.36)"
Include mmap events in the ring_buffer.

.TP
.IR "sample_id_all" " (Since Linux 2.6.38)"
If set, then all sample ID info (TID, TIME, ID, CPU, STREAM_ID)
will be provided.

.TP
.IR "exclude_host" " (Since Linux 3.2)"
Do not measure time spent in VM host

.TP
.IR "exclude_guest" " (Since Linux 3.2)"
Do not measure time spent in VM guest


.TP
.IR "wakeup_events" ", " "wakeup_watermark"
This union sets how many events
.RI ( wakeup_events )
or bytes
.RI ( wakeup_watermark )
happen before an overflow signal happens.
Which one is used is selected by the
.I watermark
bitflag.

.TP
.IR "bp_type" " (Since Linux 2.6.33)"
This chooses the breakpoint type.
It is one of:
.RS
.TP
.BR HW_BREAKPOINT_EMPTY
no breakpoint
.TP
.BR HW_BREAKPOINT_R
count when we read the memory location
.TP
.BR HW_BREAKPOINT_W
count when we write the memory location
.TP
.BR HW_BREAKPOINT_RW
count when we read or write the memory location
.TP
.BR HW_BREAKPOINT_X
count when we execute code at the memory location

.LP
The values can be combined via a bitwsie or, but the
combination of
.B HW_BREAKPOINT_R
or
.B HW_BREAKPOINT_W
with
.B HW_BREAKPOINT_X
is not allowed.
.RE

.TP
.IR "bp_addr" " (Since Linux 2.6.33)"
.I bp_addr
address of the breakpoint.
For execution breakpoints this is the memory address of the instruction
of interest; for read and write breakpoints it is the memory address
of the memory location of interest.

.TP
.IR "config1" " (Since Linux 2.6.39)"
.I config1
is used for setting events that need an extra register or otherwise
do not fit in the regular config field.
Raw OFFCORE_EVENTS on Nehalem/Westmere/SandyBridge use this field
on 3.3 and later kernels.

.TP
.IR "bp_len" " (Since Linux 2.6.33)"
.I bp_len
is the length of the breakpoint being measured if
.I type
is
.BR PERF_TYPE_BREAKPOINT .
Options are
.BR HW_BREAKPOINT_LEN_1 ,
.BR HW_BREAKPOINT_LEN_2 ,
.BR HW_BREAKPOINT_LEN_4 ,
.BR HW_BREAKPOINT_LEN_8 .
For an execution breakpoint, set this to
.IR sizeof(long) .

.TP
.IR "config2" " (Since Linux 2.6.39)"

.I config2
is a further extension of the
.I config1
field.

.TP
.IR "branch_sample_type" " (Since Linux 3.4)"
This is used with the CPUs hardware branch sampling, if available.
It can have one of the following values:
.RS
.TP
.B PERF_SAMPLE_BRANCH_USER
Branch target is in user space
.TP
.B PERF_SAMPLE_BRANCH_KERNEL
Branch target is in kernel space
.TP
.B PERF_SAMPLE_BRANCH_HV
Branch target is in hypervisor
.TP
.B PERF_SAMPLE_BRANCH_ANY
Any branch type.
.TP
.B PERF_SAMPLE_BRANCH_ANY_CALL
Any call branch
.TP
.B PERF_SAMPLE_BRANCH_ANY_RETURN
Any return branch
.TP
.BR PERF_SAMPLE_BRANCH_IND_CALL
Indirect calls
.TP
.BR PERF_SAMPLE_BRANCH_PLM_ALL
User, kernel, and hv
.RE


.SS "MMAP Layout"

When using
.BR perf_event_open()
in sampled mode, asynchronous events
(like counter overflow or
.B PROT_EXEC
mmap tracking)
are logged into a ring-buffer.
This ring-buffer is created and accessed through
.BR mmap (2).

The mmap size should be 1+2^n pages, where the first page is a
metadata page
.IR ( "struct perf_event_mmap_page" )
that contains various
bits of information such as where the ring-buffer head is.

Before kernel 2.6.39, there is a bug that means you must allocate a mmap
ring buffer when sampling even if you do not plan to access it.

The structure of the first metadata mmap page is as follows:

.in +4n
.nf
struct perf_event_mmap_page {
    __u32 version;          /* version number of this structure   */
    __u32 compat_version;   /* lowest version this is compat with */
    __u32 lock;             /* seqlock for synchronization        */
    __u32 index;            /* hardware counter identifier        */
    __s64 offset;           /* add to hardware counter value      */
    __u64 time_enabled;     /* time event active                  */
    __u64 time_running;     /* time event on CPU                  */
    union {
        __u64   capabilities;
        __u64   cap_usr_time  : 1,
                cap_usr_rdpmc : 1,
    };
    __u16   pmc_width;
    __u16   time_shift;
    __u32   time_mult;
    __u64   time_offset;
    __u64   __reserved[120];   /* Pad to 1k                */
    __u64   data_head;         /* head in the data section */
    __u64   data_tail;         /* user-space written tail  */
}
.fi
.in



The following looks at the fields in the
.I perf_event_mmap_page
structure in more detail.

.RS

.TP
.I version
Version number of this structure.

.TP
.I compat_version
The lowest version this is compatible with.

.TP
.I lock
A seqlock for synchronization.

.TP
.I index;
A unique hardware counter identifier.

.TP
.I offset
.\" FIXME clarify
Add this to hardware counter value??

.TP
.I time_enabled
Time the event was active.

.TP
.I time_running
Time the event was running.

.TP
.I cap_usr_time
User time capability

.TP
.I cap_usr_rdpmc
If the hardware supports user-space read of performance counters
without syscall (this is the "rdpmc" instruction on x86), then
the following code can be used to do a read:

.in +4n
.nf
u32 seq, time_mult, time_shift, idx, width;
u64 count, enabled, running;
u64 cyc, time_offset;
s64 pmc = 0;

do {
    seq = pc\->lock;
    barrier();
    enabled = pc\->time_enabled;
    running = pc\->time_running;

    if (pc\->cap_usr_time && enabled != running) {
        cyc = rdtsc();
        time_offset = pc\->time_offset;
        time_mult   = pc\->time_mult;
        time_shift  = pc\->time_shift;
    }

    idx = pc\->index;
    count = pc\->offset;

    if (pc\->cap_usr_rdpmc && idx) {
        width = pc\->pmc_width;
        pmc = rdpmc(idx \- 1);
    }

    barrier();
} while (pc\->lock != seq);
.fi
.in



.TP
.I pmc_width
If
.IR cap_usr_rdpmc ,
this field provides the bit-width of the value
read using the rdpmc or equivalent instruction.
This can be used to sign extend the result like:

.in +4n
.nf
pmc <<= 64 \- pmc_width;
pmc >>= 64 \- pmc_width; // signed shift right
count += pmc;
.fi
.in


.TP
.IR time_shift ", " time_mult ", " time_offset

If
.IR cap_usr_time ,
these fields can be used to compute the time
delta since time_enabled (in ns) using rdtsc or similar.
.nf

    u64 quot, rem;
    u64 delta;
    quot = (cyc >> time_shift);
    rem = cyc & ((1 << time_shift) \- 1);
    delta = time_offset + quot * time_mult +
            ((rem * time_mult) >> time_shift);
.fi

Where time_offset,time_mult,time_shift and cyc are read in the
seqcount loop described above.
This delta can then be added to
enabled and possible running (if idx), improving the scaling:
.nf

    enabled += delta;
    if (idx)
        running += delta;
    quot = count / running;
    rem  = count % running;
    count = quot * enabled + (rem * enabled) / running;
.fi

.TP
.I data_head
This points to the head of the data section.
On SMP-capable platforms, after reading the data_head value,
user-space should issue an rmb().

.TP
.I data_tail;
When the mapping is
.BR PROT_WRITE ,
the data_tail value should be written by
userspace to reflect the last read data.
In this case the kernel will not over-write unread data.

.RE


The following 2^n ring-buffer pages have the layout described below.

If
.I perf_event_attr.sample_id_all
is set, then all event types will
have the sample_type selected fields related to where/when (identity)
an event took place (TID, TIME, ID, CPU, STREAM_ID) described in
.B PERF_RECORD_SAMPLE
below, it will be stashed just after the
perf_event_header and the fields already present for the existing
fields, i.e., at the end of the payload.
That way a newer perf.data
file will be supported by older perf tools, with these new optional
fields being ignored.

The mmap values start with a header:

.in +4n
.nf
struct perf_event_header {
    __u32   type;
    __u16   misc;
    __u16   size;
};
.fi
.in

Below, we describe the
.I perf_event_header
fields in more detail.

.TP
.I type
The
.I type
value is one of the below.
The values in the corresponding record (that follows the header)
depend on the
.I type
selected as shown.

.RS
.TP
.B PERF_RECORD_MMAP
The MMAP events record the
.B PROT_EXEC
mappings so that we can correlate
userspace IPs to code.
They have the following structure:

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32    pid, tid;
    u64    addr;
    u64    len;
    u64    pgoff;
    char   filename[];
};
.in

.TP
.B PERF_RECORD_LOST
This record indicates when events are lost.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 id;
    u64 lost;
};
.fi
.in

.TP
.B PERF_RECORD_COMM
This record indicates a change in the process name.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    char comm[];
};
.fi
.in

.TP
.B PERF_RECORD_EXIT
This record indicates a process exit event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.BR PERF_RECORD_THROTTLE ", " PERF_RECORD_UNTHROTTLE
This record indicates a throttle/unthrottle event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 time;
    u64 id;
    u64 stream_id;
};
.fi
.in

.TP
.B PERF_RECORD_FORK
This record indicates a fork event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.B PERF_RECORD_READ
This record indicates a read event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    struct read_format values;
};
.fi
.in

.TP
.B PERF_RECORD_SAMPLE
This record indicates a sample.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64   ip;          /* if PERF_SAMPLE_IP      */
    u32   pid, tid;    /* if PERF_SAMPLE_TID     */
    u64   time;        /* if PERF_SAMPLE_TIME    */
    u64   addr;        /* if PERF_SAMPLE_ADDR    */
    u64   id;          /* if PERF_SAMPLE_ID      */
    u64   stream_id; /* if PERF_SAMPLE_STREAM_ID */
    u32   cpu, res;    /* if PERF_SAMPLE_CPU     */
    u64   period;      /* if PERF_SAMPLE_PERIOD  */
    struct read_format v; /* if PERF_SAMPLE_READ */
    u64   nr;        /* if PERF_SAMPLE_CALLCHAIN */
    u64   ips[nr];   /* if PERF_SAMPLE_CALLCHAIN */
    u32   size;        /* if PERF_SAMPLE_RAW     */
    char  data[size];  /* if PERF_SAMPLE_RAW     */
    u64   from;   /* if PERF_SAMPLE_BRANCH_STACK */
    u64   to;     /* if PERF_SAMPLE_BRANCH_STACK */
    u64   flags;  /* if PERF_SAMPLE_BRANCH_STACK */
    u64   lbr[nr];/* if PERF_SAMPLE_BRANCH_STACK */
};
.fi
.in

The RAW record data is opaque with respect to the ABI.
The ABI doesn't make any promises with respect to the stability
of its content, it may vary depending
on event, hardware, and kernel version.

.RE


.TP
.I misc
The
.I misc
field is one of the following:
.RS
.TP
.B PERF_RECORD_MISC_CPUMODE_MASK
[To be documented]
.TP
.B PERF_RECORD_MISC_CPUMODE_UNKNOWN
[To be documented]
.TP
.B PERF_RECORD_MISC_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_HYPERVISOR
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_EXACT_IP
This indicates that the content of
.B PERF_SAMPLE_IP
points
to the actual instruction that triggered the event.
See also
.IR perf_event_attr.precise_ip .

.RE

.TP
.I size
This indicates the size of the record.


.SS "Signal Overflow"

Counters can be set to signal when a threshold is crossed.
This is set up using traditional the
.BR poll (2),
.BR select (2),
.BR epoll (2)
and
.BR fcntl (2),
system calls.

Normally, a notification is generated for every page filled, however
one can additionally set
.I perf_event_attr.wakeup_events
to generate one every so many counter overflow events.

.SS "Reading Results"
Once a
.BR perf_event_open()
file descriptor  has been opened, the values
of the events can be read from the file descriptor.
The values that are there are specified by the
.I read_format
field in the attr structure at open time.

If you attempt to read into a buffer that is not big enough to hold the
data, an error is returned
.IR ( ENOSPC ).

Here is the layout of the data returned by a read.

If
.B PERF_FORMAT_GROUP
was specified to allow reading all events in a group at once:

.in +4n
.nf
struct {
    u64 nr;             /* The number of events */
    u64 time_enabled;   /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;   /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    struct {
        u64 value;      /* The value of the event */
        u64 id;         /* if PERF_FORMAT_ID */
    } values[nr];
};
.fi
.in

If
.B PERF_FORMAT_GROUP
was
.I not
specified, then the read values look as following:

.in +4n
.nf
struct {
    u64 value;         /* The value of the event */
    u64 time_enabled;  /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;  /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    u64 id;            /* if PERF_FORMAT_ID */
};
.fi
.in

The values read are described in more detail below.
.RS
.TP
.I nr
The number of events in this file descriptor.
Only available if
.B PERF_FORMAT_GROUP
was specified.

.TP
.IR time_enabled ", " time_running
Total time the event was enabled and running.
Normally these are the same.
If more events are started
than available counter slots on the PMU, then multiplexing
happens and events only run part of the time.
In that case the
.I time_enabled
and
.I time running
values can be used to scale an estimated value for the count.

.TP
.I value
An unsigned 64-bit value containing the counter result.

.TP
.I id
A globally unique value for this particular event, only there if
.B PERF_FORMAT_ID
was specified in read_format.

.RE




.RE

.SS "rdpmc instruction"
Starting with Linux 3.4 on x86, you can use the
.I rdpmc
instruction to get low-latency reads without having to enter the kernel.


.SS "perf_event ioctl calls"
.PP
Various ioctls act on
.BR perf_event_open()
file descriptors
.\" FIXME the arguments for these ioctl() operations need to be described
.TP
.B PERF_EVENT_IOC_ENABLE
Enables an individual counter or counter group.

.TP
.B PERF_EVENT_IOC_DISABLE
Disables an individual counter or counter group.

Enabling or disabling the leader of a group enables or disables the
entire group; that is, while the group leader is disabled, none of the
counters in the group will count.
Enabling or disabling a member of a group other than the leader only
affects that counter; disabling a non-leader
stops that counter from counting but doesn't affect any other counter.

.TP
.B PERF_EVENT_IOC_REFRESH
Non-inherited overflow counters can use this
to enable a counter for 'nr' events, after which it gets disabled again.
.\" FIXME the following needs clarification/confirmation
I think the goal of IOC_REFRESH is not to reload the period but simply to
adjust the number of events before the next notifications.

.TP
.B PERF_EVENT_IOC_RESET
Reset the event count to zero.
This only resets the counts; there is no way to reset the
multiplexing
.I time_enabled
or
.I time_running
values.
When sent to a group leader, only
the leader is reset (child events are not).

.TP
.B PERF_EVENT_IOC_PERIOD
IOC_PERIOD is the command to update the period; it
does not update the current period but instead defers until next.

.TP
.B PERF_EVENT_IOC_SET_OUTPUT
This tells the kernel to report event notifications to the specified
file descriptor rather than the default one.
The file descriptors must all be on the same CPU.

.TP
.BR PERF_EVENT_IOC_SET_FILTER " (Since Linux 2.6.33)"
This adds an ftrace filter to this event.

.SS "Using prctl"
A process can enable or disable all the counter groups that are
attached to it using the
.BR prctl (2)
.B PR_TASK_PERF_EVENTS_ENABLE
and
.B PR_TASK_PERF_EVENTS_DISABLE
operations.
This applies to all counters on the current process, whether created by
this process or by another, and does not affect any counters that this
process has created on other processes.
It only enables or disables
the group leaders, not any other members in the groups.

.SS /proc/sys/kernel/perf_event_paranoid

The
.I /proc/sys/kernel/perf_event_paranoid
file can be set to restrict access to the performance counters.
2
means no measurements allowed,
1
means normal counter access,
0
means you can access CPU-specific data, and
\-1
means no restrictions.

The existence of the
.I perf_event_paranoid
file is the official method for determining if a kernel supports
.BR perf_event_open().

.SH "RETURN VALUE"
.BR perf_event_open ()
returns the new file descriptor, or \-1 if an error occurred
(in which case,
.I errno
is set appropriately).
.SH ERRORS
.TP
.B EINVAL
Returned if the specified event is not available.
.TP
.B ENOSPC
Prior to Linux 3.3, if there was no counter room,
.B ENOSPC
was returned.
Linus did not like this, and this was changed to
.BR EINVAL .
.B ENOSPC
is still returned if you try to read results into
too small a buffer.

.SH VERSION

.BR perf_event_open ()
was introduced in Linux 2.6.31 but was called
.BR perf_counter_open () .
It was renamed in Linux 2.6.32.

.SH CONFORMING TO

This call is specific to Linux
and should not be used in programs intended to be portable.

.SH NOTES
Glibc does not provide a wrapper for this system call; call it using
.BR syscall (2).

The official way of knowing if
.BR perf_event_open()
support is enabled is checking
for the existence of the file
.I /proc/sys/kernel/perf_event_paranoid

.SH BUGS

The
.B F_SETOWN_EX
option to
.IR fcntl (2)
is needed to properly get overflow signals in threads.
This was introduced in Linux 2.6.32.

Prior to Linux 2.6.33 (at least for x86) the kernel did not check
if events could be scheduled together until read time.
The same happens on all known kernels if the NMI watchdog is enabled.
This means to see if a given set of events works you have to
.BR perf_event_open (),
start, then read before you know for sure you
can get valid measurements.

Prior to Linux 2.6.34 event constraints were not enforced by the kernel.
In that case, some events would silently return "0" if the kernel
scheduled them in an improper counter slot.

Prior to Linux 2.6.34 there was a bug when multiplexing where the
wrong results could be returned.

Kernels from Linux 2.6.35 to Linux 2.6.39 can quickly crash the kernel if
"inherit" is enabled and many threads are started.

Prior to Linux 2.6.35,
.B PERF_FORMAT_GROUP
did not work with attached processes.

In older Linux 2.6 versions,
refreshing an event group leader refreshed all siblings,
and refreshing with a parameter of 0 enabled infinite refresh.
This behavior is unsupported and should not be relied on.

There is a bug in the kernel code between
Linux 2.6.36 and Linux 3.0 that ignores the
"watermark" field and acts as if a wakeup_event
was chosen if the union has a
non-zero value in it.

Always double-check your results!  Various generalized events
have had wrong values.
For example, retired branches measured
the wrong thing on AMD machines until Linux 2.6.35.

.SH EXAMPLE
The following is a short example that measures the total
instruction count of a call to printf().
.nf

#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include <asm/unistd.h>

long perf_event_open( struct perf_event_attr *hw_event, pid_t pid,
                      int cpu, int group_fd, unsigned long flags )
{
    int ret;

    ret = syscall( __NR_perf_event_open, hw_event, pid, cpu,
                   group_fd, flags );
    return ret;
}


int
main(int argc, char **argv)
{

    struct perf_event_attr pe;
    long long count;
    int fd;

    memset(&pe, 0, sizeof(struct perf_event_attr));
    pe.type = PERF_TYPE_HARDWARE;
    pe.size = sizeof(struct perf_event_attr);
    pe.config = PERF_COUNT_HW_INSTRUCTIONS;
    pe.disabled = 1;
    pe.exclude_kernel = 1;
    pe.exclude_hv = 1;

    fd = perf_event_open(&pe, 0, \-1, \-1, 0);
    if (fd < 0) {
       fprintf(stderr, "Error opening leader %llx\\n", pe.config);
    }

    ioctl(fd, PERF_EVENT_IOC_RESET, 0);
    ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);

    printf("Measuring instruction count for this printf\\n");

    ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
    read(fd, &count, sizeof(long long));

    printf("Used %lld instructions\\n", count);

    close(fd);
}
.fi

.SH "SEE ALSO"
.BR fcntl (2),
.BR mmap (2),
.BR open (2),
.BR prctl (2),
.BR read (2)
-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Author of "The Linux Programming Interface"; http://man7.org/tlpi/
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[RFC] perf: proposed perf_event_open() manpage

[Vince Weaver]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 42248 bytes --]

Hello

attached is a proposed manpage for the perf_event_open() system call.

I'd appreciate any review or comments, especially for the parts marked
as FIXME or "[To be documented]"

This system call has a complicated interface and I'm sure I've missed
or glossed over various important features, so your feedback is needed and 
appreciated.

The eventual goal is to have this included with the Linux man-pages 
project.

Thanks!

Vince Weaver
vincent.weaver@maine.edu


.\" Hey Emacs! This file is -*- nroff -*- source.
.\"
.\" Copyright (c) 2012, Vincent Weaver
.\"
.\" This is free documentation; you can redistribute it and/or
.\" modify it under the terms of the GNU General Public License as
.\" published by the Free Software Foundation; either version 2 of
.\" the License, or (at your option) any later version.
.\"
.\" The GNU General Public License's references to "object code"
.\" and "executables" are to be interpreted as the output of any
.\" document formatting or typesetting system, including
.\" intermediate and printed output.
.\"
.\" This manual is distributed in the hope that it will be useful,
.\" but WITHOUT ANY WARRANTY; without even the implied warranty of
.\" MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
.\" GNU General Public License for more details.
.\"
.\" You should have received a copy of the GNU General Public
.\" License along with this manual; if not, see
.\" <http://www.gnu.org/licenses/>.
.\"
.\" This document is based on the perf_event.h header file, the
.\" tools/perf/design.txt file, and a lot of bitter experience.
.\"
.TH PERF_EVENT_OPEN 2 2012-10-23 "Linux" "Linux Programmer's Manual"
.SH NAME
perf_event_open \- set up performance monitoring
.SH SYNOPSIS
.nf
.B #include <linux/perf_event.h>
.B #include <linux/hw_breakpoint.h>
.sp
.BI "int perf_event_open(struct perf_event_attr *" hw_event ,
.BI "                    pid_t " pid ", int " cpu ", int " group_fd ,
.BI "                    unsigned long " flags  );
.fi

.IR Note :
There is no glibc wrapper for this system call; see NOTES.
.SH DESCRIPTION
Given a list of parameters,
.BR perf_event_open ()
returns a file descriptor, for use in subsequent system calls
.RB ( read "(2), " mmap "(2), " prctl "(2), " fcntl "(2), etc.)."
.PP
A call to
.BR perf_event_open ()
creates a file descriptor that allows measuring performance
information.
Each file descriptor corresponds to one
event that is measured; these can be grouped together
to measure multiple events simultaneously.
.PP
Events can be enabled and disabled in two ways: via
.BR ioctl (2)
and via
.BR prctl (2) .
When an event is disabled it does not count or generate overflows but does
continue to exist and maintain its count value.
Events come in two flavors: counting and sampled.
A
.I counting
event is one that is used for counting the aggregate number of events
that occur.
In general, counting event results are gathered with a
.BR read (2)
call.
A
.I sampling
event periodically writes measurements to a buffer that can then
be accessed via
.BR  mmap (2) .
.SS Arguments
.P
The argument
.I pid
allows events to be attached to processes in various ways.
If
.I pid
is 0, measurements happen on the current task, if
.I pid
is greater than 0, the process indicated by
.I pid
is measured, and if
.I pid
is less than 0, all processes are counted.

The
.I cpu
argument allows measurements to be specific to a CPU.
If
.I cpu
is greater than or equal to 0,
measurements are restricted to the specified CPU;
if
.I cpu
is \-1, the events are measured on all CPUs.
.P
Note that the combination of
.IR pid " == \-1"
and
.IR cpu " == \-1"
is not valid.
.P
A
.IR pid " > 0"
and
.IR cpu " == \-1"
setting measures per-process and follows that process to whatever CPU the
process gets scheduled to.
Per-process events can be created by any user.
.P
A
.IR pid " == \-1"
and
.IR cpu " >= 0"
setting is per-CPU and measures all processes on the specified CPU.
Per-CPU events need the
.B CAP_SYS_ADMIN
capability.
.P
The
.I group_fd
argument allows counter groups to be set up.
A counter group has one counter which is the group leader.
The leader is created first, with
.IR group_fd " = \-1"
in the
.BR perf_event_open ()
call that creates it.
The rest of the group members are created subsequently, with
.IR group_fd
giving the fd of the group leader.
(A single counter on its own is created with
.IR group_fd " = \-1"
and is considered to be a group with only 1 member.)
.P
A counter group is scheduled onto the CPU as a unit: it will only
be put onto the CPU if all of the counters in the group can be put onto
the CPU.
This means that the values of the member counters can be
meaningfully compared, added, divided (to get ratios), etc., with each
other, since they have counted events for the same set of executed
instructions.
.P
The
.I flags
argument takes one of the following values:
.TP
.BR PERF_FLAG_FD_NO_GROUP
.\" FIXME The following sentence is unclear
This flag allows creating an event as part of an event group but
having no group leader.
It is unclear why this is useful.
.\" FIXME So, why is it useful?
.TP
.BR PERF_FLAG_FD_OUTPUT
This flag re-routes the output from an event to the group leader.
.TP
.BR PERF_FLAG_PID_CGROUP " (Since Linux 2.6.39)."
This flag activates per-container system-wide monitoring.
A container
is an abstraction that isolates a set of resources for finer grain
control (CPUs, memory, etc...).
In this mode, the event is measured
only if the thread running on the monitored CPU belongs to the designated
container (cgroup).
The cgroup is identified by passing a file descriptor
opened on its directory in the cgroupfs filesystem.
For instance, if the
cgroup to monitor is called
.IR test ,
then a file descriptor opened on
.I /dev/cgroup/test
(assuming cgroupfs is mounted on
.IR /dev/cgroup )
must be passed as the
.I pid
parameter.
cgroup monitoring is only available
for system-wide events and may therefore require extra permissions.
.P
The
.I perf_event_attr
structure is what is passed into the
.BR perf_event_open ()
syscall.
It is large and has a complicated set of dependent fields.

.in +4n
.nf
struct perf_event_attr {
    __u32     type;         /* Type of event */
    __u32     size;         /* Size of attribute structure */
    __u64     config;       /* Type-specific configuration */

    union {
        __u64 sample_period;    /* Period of sampling */
        __u64 sample_freq;      /* Frequency of sampling */
    };

    __u64     sample_type;  /* Specifies values included in sample */
    __u64     read_format;  /* Specifies values returned in read */

    __u64     disabled       : 1,   /* off by default         */
              inherit        : 1,   /* children inherit it    */
              pinned         : 1,   /* must always be on PMU  */
              exclusive      : 1,   /* only group on PMU      */
              exclude_user   : 1,   /* don't count user       */
              exclude_kernel : 1,   /* don't count kernel     */
              exclude_hv     : 1,   /* don't count hypervisor */
              exclude_idle   : 1,   /* don't count when idle  */
              mmap           : 1,   /* include mmap data      */
              comm           : 1,   /* include comm data      */
              freq           : 1,   /* use freq, not period   */
              inherit_stat   : 1,   /* per task counts        */
              enable_on_exec : 1,   /* next exec enables      */
              task           : 1,   /* trace fork/exit        */
              watermark      : 1,   /* wakeup_watermark       */
              precise_ip     : 2,   /* skid constraint        */
              mmap_data      : 1,   /* non-exec mmap data     */
              sample_id_all  : 1,   /* sample_type all events */
              exclude_host   : 1,   /* don't count in host    */
              exclude_guest  : 1,   /* don't count in guest   */
              __reserved_1   : 43;

    union {
        __u32 wakeup_events;    /* wakeup every n events */
        __u32 wakeup_watermark; /* bytes before wakeup   */
    };

    __u32     bp_type;          /* breakpoint type */

    union {
        __u64 bp_addr;          /* breakpoint address */
        __u64 config1;          /* extension of config */
    };

    union {
        __u64 bp_len;           /* breakpoint length */
        __u64 config2;          /* extension of config1 */
    };
    __u64     branch_sample_type;   /* enum branch_sample_type */
};
.fi
.in

The fields of the
.I perf_event_attr
structure are described in more detail below.

.TP
.I type
This field specifies the overall event type.
It has one of the following values:
.RS
.TP
.B PERF_TYPE_HARDWARE
This indicates one of the "generalized" hardware events provided
by the kernel.
See the
.I config
field definition for more details.
.TP
.B PERF_TYPE_SOFTWARE
This indicates one of the software-defined events provided by the kernel
(even if no hardware support is available).
.TP
.B PERF_TYPE_TRACEPOINT
This indicates a tracepoint
provided by the kernel tracepoint infrastructure.
.TP
.B PERF_TYPE_HW_CACHE
This indicates a hardware cache event.
This has a special encoding, described in the
.I config
field definition.
.TP
.B PERF_TYPE_RAW
This indicates a "raw" implementation-specific event in the
.IR config " field."
.TP
.BR PERF_TYPE_BREAKPOINT " (Since Linux 2.6.33)"
This indicates a hardware breakpoint as provided by the CPU.
Breakpoints can be read/write accesses to an address as well as
execution of an instruction address.
.TP
.RB "dynamic PMU"
Since Linux 2.6.39,
.BR perf_event_open()
can support multiple PMUs.
To enable this, a value exported by the kernel can be used in the
.I type
field to indicate which PMU to use.
The value to use can be found in the sysfs filesystem:
there is a subdirectory per PMU instance under
.IR /sys/devices .
In each sub-directory there is a
.I type
file whose content is an integer that can be used in the
.I type
field.
For instance,
.I /sys/devices/cpu/type
contains the value for the core CPU PMU, which is usually 4.
.RE

.TP
.I "size"
The size of the
.I perf_event_attr
structure for forward/backward compatibility.
Set this using
.I sizeof(struct perf_event_attr)
to allow the kernel to see
the struct size at the time of compilation.

The related define
.B PERF_ATTR_SIZE_VER0
is set to 64; this was the size of the first published struct.
.B PERF_ATTR_SIZE_VER1
is 72, corresponding to the addition of breakpoints in Linux 2.6.33.
.B PERF_ATTR_SIZE_VER2
is 80 corresponding to the addition of branch sampling in Linux 3.4.

.TP
.I "config"
This specifies which event you want, in conjunction with
the
.I type
field.
The
.IR config1 " and " config2
fields are also taken into account in cases where 64 bits is not
enough to fully specify the event.
The encoding of these fields are event dependent.

The most significant bit (bit 63) of
.I config
signifies CPU-specific (raw) counter configuration data;
if the most significant bit is unset, the next 7 bits are an event
type and the rest of the bits are the event identifier.

There are various ways to set the
.I config
field that are dependent on the value of the previously
described
.I type
field.
What follows are various possible settings for
.I config
separated out by
.IR type .

If
.I type
is
.BR PERF_TYPE_HARDWARE ,
we are measuring one of the generalized hardware CPU events.
Not all of these are available on all platforms.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_HW_CPU_CYCLES
Total cycles.
Be wary of what happens during CPU frequency scaling
.TP
.B PERF_COUNT_HW_INSTRUCTIONS
Retired instructions.
Be careful, these can be affected by various
issues, most notably hardware interrupt counts
.TP
.B PERF_COUNT_HW_CACHE_REFERENCES
Cache accesses.
Usually this indicates Last Level Cache accesses but this may
vary depending on your CPU.
This may include prefetches and coherency messages; again this
depends on the design of your CPU.
.TP
.B PERF_COUNT_HW_CACHE_MISSES
Cache misses.
Usually this indicates Last Level Cache misses; this is intended to be
used in conjunction with the
.B PERF_COUNT_HW_CACHE_REFERENCES
event to calculate cache miss rates.
.TP
.B PERF_COUNT_HW_BRANCH_INSTRUCTIONS
Retired branch instructions.
Prior to Linux 2.6.34, this used
the wrong event on AMD processors.
.TP
.B PERF_COUNT_HW_BRANCH_MISSES
Mispredicted branch instructions.
.TP
.B PERF_COUNT_HW_BUS_CYCLES
Bus cycles, which can be different from total cycles.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_FRONTEND " (Since Linux 3.0)"
Stalled cycles during issue.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_BACKEND  " (Since Linux 3.0)"
Stalled cycles during retirement.
.TP
.BR PERF_COUNT_HW_REF_CPU_CYCLES  " (Since Linux 3.3)"
Total cycles; not affected by CPU frequency scaling.
.RE
.IP
If
.I type
is
.BR PERF_TYPE_SOFTWARE ,
we are measuring software events provided by the kernel.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_SW_CPU_CLOCK
This reports the CPU clock, a high-resolution per-CPU timer.
.TP
.B PERF_COUNT_SW_TASK_CLOCK
This reports a clock count specific to the task that is running.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS
This reports the number of page faults.
.TP
.B PERF_COUNT_SW_CONTEXT_SWITCHES
This counts context switches.
Until Linux 2.6.34, these were all reported as user-space
events, after that they are reported as happening in the kernel.
.TP
.B PERF_COUNT_SW_CPU_MIGRATIONS
This reports the number of times the process
has migrated to a new CPU.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MIN
This counts the number of minor page faults.
These did not require disk I/O to handle.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MAJ
This counts the number of major page faults.
These required disk I/O to handle.
.TP
.BR PERF_COUNT_SW_ALIGNMENT_FAULTS " (Since Linux 2.6.33)"
This counts the number of alignment faults.
These happen when unaligned memory accesses happen; the kernel
can handle these but it reduces performance.
This only happens on some architectures (never on x86).
.TP
.BR PERF_COUNT_SW_EMULATION_FAULTS " (Since Linux 2.6.33)"
This counts the number of emulation faults.
The kernel sometimes traps on unimplemented instructions
and emulates them for userspace.
This can negatively impact performance.
.RE
.RE


.RS
If
.I type
is
.BR PERF_TYPE_TRACEPOINT ,
then we are measuring kernel tracepoints.
The value to use in
.I config
can be obtained from under debugfs
.I tracing/events/*/*/id
if ftrace is enabled in the kernel.

.RE

.RS
If
.I type
is
.BR PERF_TYPE_HW_CACHE ,
then we are measuring a hardware CPU cache event.
To calculate the appropriate
.I config
value use the following equation:
.RS 4
.nf

    (perf_hw_cache_id) | (perf_hw_cache_op_id << 8) |
    (perf_hw_cache_op_result_id << 16)
.fi
.P
where
.I perf_hw_cache_id
is one of:
.RS
.TP
.B PERF_COUNT_HW_CACHE_L1D
for measuring Level 1 Data Cache
.TP
.B PERF_COUNT_HW_CACHE_L1I
for measuring Level 1 Instruction Cache
.TP
.B PERF_COUNT_HW_CACHE_LL
for measuring Last-Level Cache
.TP
.B PERF_COUNT_HW_CACHE_DTLB
for measuring the Data TLB
.TP
.B PERF_COUNT_HW_CACHE_ITLB
for measuring the Instruction TLB
.TP
.B PERF_COUNT_HW_CACHE_BPU
for measuring the branch prediction unit
.TP
.BR PERF_COUNT_HW_CACHE_NODE " (Since Linux 3.0)"
for measuring local memory accesses
.RE

.P
and
.I perf_hw_cache_op_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_OP_READ
for read accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_WRITE
for write accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_PREFETCH
for prefetch accesses
.RE

.P
and
.I perf_hw_cache_op_result_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_RESULT_ACCESS
to measure accesses
.TP
.B PERF_COUNT_HW_CACHE_RESULT_MISS
to measure misses
.RE
.RE

If
.I type
is
.BR PERF_TYPE_RAW ,
then a custom "raw"
.I config
value is needed.
Most CPUs support events that are not covered by the "generalized" events.
These are implementation defined; see your CPU manual (for example
the Intel Volume 3B documentation or the AMD BIOS and Kernel Developer
Guide).
The libpfm4 library can be used to translate from the name in the
architectural manuals to the raw hex value
.BR perf_event_open ()
expects in this field.

If
.I type
is
.BR PERF_TYPE_BREAKPOINT ,
then leave
.I config
set to zero.
Its parameters are set in other places.
.RE
.TP
.IR sample_period ", " sample_freq
A "sampling" counter is one that generates an interrupt
every N events, where N is given by
.IR sample_period .
A sampling counter has
.IR sample_period " > 0."
The
.I sample_type
field controls what data is recorded on each interrupt.

.I sample_freq
can be used if you wish to use frequency rather than period.
In this case you set the
.I freq
flag.
The kernel will adjust the sampling period
to try and achieve the desired rate.
The rate of adjustment is a
timer tick.


.TP
.I "sample_type"
The various bits in this field specify which values to include
in the overflow packets.
They will be recorded in a ring-buffer,
which is available to user-space using
.BR mmap (2).
The order in which the values are saved in the
overflow packets as documented in the MMAP Layout subsection below;
it is not the
.I "enum perf_event_sample_format"
order.
.RS
.TP
.B PERF_SAMPLE_IP
instruction pointer
.TP
.B PERF_SAMPLE_TID
thread id
.TP
.B PERF_SAMPLE_TIME
time
.TP
.B PERF_SAMPLE_ADDR
address
.TP
.B PERF_SAMPLE_READ
[To be documented]
.TP
.B PERF_SAMPLE_CALLCHAIN
[To be documented]
.TP
.B PERF_SAMPLE_ID
[To be documented]
.TP
.B PERF_SAMPLE_CPU
[To be documented]
.TP
.B PERF_SAMPLE_PERIOD
[To be documented]
.TP
.B PERF_SAMPLE_STREAM_ID
[To be documented]
.TP
.B PERF_SAMPLE_RAW
[To be documented]
.TP
.BR PERF_SAMPLE_BRANCH_STACK " (Since Linux 3.4)"
[To be documented]
.RE

.TP
.IR "read_format"
This field specifies the format of the data returned by
.BR read (2)
on a
.BR perf_event_open()
file descriptor.
.RS
.TP
.B PERF_FORMAT_TOTAL_TIME_ENABLED
Adds the 64-bit "time_enabled" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and multiplexing is happening.
.TP
.B PERF_FORMAT_TOTAL_TIME_RUNNING
Adds the 64-bit "time_running" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and  multiplexing is happening.
.TP
.B PERF_FORMAT_ID
Adds a 64-bit unique value that corresponds to the event-group.
.TP
.B PERF_FORMAT_GROUP
Allows all counter values in an event-group to be read with one read.
.RE

.TP
.IR "disabled"
The
.I disabled
bit specifies whether the counter starts out disabled or enabled.
If disabled, the event can later be enabled by
.BR ioctl (2),
.BR prctl (2),
or
.IR enable_on_exec .

.TP
.IR "inherit"
The
.I inherit
bit specifies that this counter should count events of child
tasks as well as the task specified.
This only applies to new children, not to any existing children at
the time the counter is created (nor to any new children of
existing children).

Inherit does not work for some combinations of
.IR read_format s,
such as
.BR PERF_FORMAT_GROUP .

.TP
.IR "pinned"
The
.I pinned
bit specifies that the counter should always be on the CPU if at all
possible.
It only applies to hardware counters and only to group leaders.
If a pinned counter cannot be put onto the CPU (e.g., because there are
not enough hardware counters or because of a conflict with some other
event), then the counter goes into an 'error' state, where reads
return end-of-file (i.e.,
.BR read (2)
returns 0) until the counter is subsequently enabled or disabled.

.TP
.IR "exclusive"
The
.I exclusive
bit specifies that when this counter's group is on the CPU,
it should be the only group using the CPU's counters.
In the future this may allow monitoring programs to
support PMU features that need to run alone so that they do not
disrupt other hardware counters.

.TP
.IR "exclude_user"
If this bit is set, the count excludes events that happen in user-space.

.TP
.IR "exclude_kernel"
If this bit is set, the count excludes events that happen in kernel-space.

.TP
.IR "exclude_hv"
If this bit is set, the count excludes events that happen in the
hypervisor.
This is mainly for PMUs that have built-in support for handling this
(such as POWER).
Extra support is needed for handling hypervisor measurements on most
machines.

.TP
.IR "exclude_idle"
If set, don't count when the CPU is idle.

.TP
.IR "mmap"
The
.I mmap
bit enables recording of extra information to a
mmap'd ring-buffer.
This is
described below in subsection MMAP Layout.

.TP
.IR "comm"
The
.I comm
bit enables tracking of process command name as modified by the
.IR exec (2)
and
.IR prctl (PR_SET_NAME)
system calls.
Unfortunately for tools,
there is no way to distinguish one system call versus the other.

.TP
.IR "freq"
If this bit is set, then
.I sample_frequency
not
.I sample_period
is used when setting up the sampling interval.

.TP
.IR "inherit_stat"
This bit enables saving of event counts on context switch for
inherited tasks.
This is only meaningful if the
.I inherit
field is set.

.TP
.IR "enable_on_exec"
If this bit is set, a counter is automatically
enabled after a call to
.BR exec (2).

.TP
.IR "task"
If this bit is set, then
fork/exit notifications are included in the ring buffer.

.TP
.IR "watermark"
If set, have a sampling interrupt happen when we cross the
wakeup_watermark boundary.

.TP
.IR "precise_ip" " (Since Linux 2.6.35)"
This controls the amount of skid.
Skid is how many instructions
execute between an event of interest happening and the kernel
being able to stop and record the event.
Smaller skid is
better and allows more accurate reporting of which events
correspond to which instructions, but hardware is often limited
with how small this can be.

The values of this are the following:
.RS
.TP
0 -
.B SAMPLE_IP
can have arbitrary skid
.TP
1 -
.B SAMPLE_IP
must have constant skid
.TP
2 -
.B SAMPLE_IP
requested to have 0 skid
.TP
3 -
.B SAMPLE_IP
must have 0 skid.
See also
.BR PERF_RECORD_MISC_EXACT_IP .
.RE

.TP
.IR "mmap_data" " (Since Linux 2.6.36)"
Include mmap events in the ring_buffer.

.TP
.IR "sample_id_all" " (Since Linux 2.6.38)"
If set, then all sample ID info (TID, TIME, ID, CPU, STREAM_ID)
will be provided.

.TP
.IR "exclude_host" " (Since Linux 3.2)"
Do not measure time spent in VM host

.TP
.IR "exclude_guest" " (Since Linux 3.2)"
Do not measure time spent in VM guest


.TP
.IR "wakeup_events" ", " "wakeup_watermark"
This union sets how many events
.RI ( wakeup_events )
or bytes
.RI ( wakeup_watermark )
happen before an overflow signal happens.
Which one is used is selected by the
.I watermark
bitflag.

.TP
.IR "bp_type" " (Since Linux 2.6.33)"
This chooses the breakpoint type.
It is one of:
.RS
.TP
.BR HW_BREAKPOINT_EMPTY
no breakpoint
.TP
.BR HW_BREAKPOINT_R
count when we read the memory location
.TP
.BR HW_BREAKPOINT_W
count when we write the memory location
.TP
.BR HW_BREAKPOINT_RW
count when we read or write the memory location
.TP
.BR HW_BREAKPOINT_X
count when we execute code at the memory location

.LP
The values can be combined via a bitwsie or, but the
combination of
.B HW_BREAKPOINT_R
or
.B HW_BREAKPOINT_W
with
.B HW_BREAKPOINT_X
is not allowed.
.RE

.TP
.IR "bp_addr" " (Since Linux 2.6.33)"
.I bp_addr
address of the breakpoint.
For execution breakpoints this is the memory address of the instruction
of interest; for read and write breakpoints it is the memory address
of the memory location of interest.

.TP
.IR "config1" " (Since Linux 2.6.39)"
.I config1
is used for setting events that need an extra register or otherwise
do not fit in the regular config field.
Raw OFFCORE_EVENTS on Nehalem/Westmere/SandyBridge use this field
on 3.3 and later kernels.

.TP
.IR "bp_len" " (Since Linux 2.6.33)"
.I bp_len
is the length of the breakpoint being measured if
.I type
is
.BR PERF_TYPE_BREAKPOINT .
Options are
.BR HW_BREAKPOINT_LEN_1 ,
.BR HW_BREAKPOINT_LEN_2 ,
.BR HW_BREAKPOINT_LEN_4 ,
.BR HW_BREAKPOINT_LEN_8 .
For an execution breakpoint, set this to
.IR sizeof(long) .

.TP
.IR "config2" " (Since Linux 2.6.39)"

.I config2
is a further extension of the
.I config1
field.

.TP
.IR "branch_sample_type" " (Since Linux 3.4)"
This is used with the CPUs hardware branch sampling, if available.
It can have one of the following values:
.RS
.TP
.B PERF_SAMPLE_BRANCH_USER
Branch target is in user space
.TP
.B PERF_SAMPLE_BRANCH_KERNEL
Branch target is in kernel space
.TP
.B PERF_SAMPLE_BRANCH_HV
Branch target is in hypervisor
.TP
.B PERF_SAMPLE_BRANCH_ANY
Any branch type.
.TP
.B PERF_SAMPLE_BRANCH_ANY_CALL
Any call branch
.TP
.B PERF_SAMPLE_BRANCH_ANY_RETURN
Any return branch
.TP
.BR PERF_SAMPLE_BRANCH_IND_CALL
Indirect calls
.TP
.BR PERF_SAMPLE_BRANCH_PLM_ALL
User, kernel, and hv
.RE


.SS "MMAP Layout"

When using
.BR perf_event_open()
in sampled mode, asynchronous events
(like counter overflow or
.B PROT_EXEC
mmap tracking)
are logged into a ring-buffer.
This ring-buffer is created and accessed through
.BR mmap (2).

The mmap size should be 1+2^n pages, where the first page is a
metadata page
.IR ( "struct perf_event_mmap_page" )
that contains various
bits of information such as where the ring-buffer head is.

Before kernel 2.6.39, there is a bug that means you must allocate a mmap
ring buffer when sampling even if you do not plan to access it.

The structure of the first metadata mmap page is as follows:

.in +4n
.nf
struct perf_event_mmap_page {
    __u32 version;          /* version number of this structure   */
    __u32 compat_version;   /* lowest version this is compat with */
    __u32 lock;             /* seqlock for synchronization        */
    __u32 index;            /* hardware counter identifier        */
    __s64 offset;           /* add to hardware counter value      */
    __u64 time_enabled;     /* time event active                  */
    __u64 time_running;     /* time event on CPU                  */
    union {
        __u64   capabilities;
        __u64   cap_usr_time  : 1,
                cap_usr_rdpmc : 1,
    };
    __u16   pmc_width;
    __u16   time_shift;
    __u32   time_mult;
    __u64   time_offset;
    __u64   __reserved[120];   /* Pad to 1k                */
    __u64   data_head;         /* head in the data section */
    __u64   data_tail;         /* user-space written tail  */
}
.fi
.in



The following looks at the fields in the
.I perf_event_mmap_page
structure in more detail.

.RS

.TP
.I version
Version number of this structure.

.TP
.I compat_version
The lowest version this is compatible with.

.TP
.I lock
A seqlock for synchronization.

.TP
.I index;
A unique hardware counter identifier.

.TP
.I offset
.\" FIXME clarify
Add this to hardware counter value??

.TP
.I time_enabled
Time the event was active.

.TP
.I time_running
Time the event was running.

.TP
.I cap_usr_time
User time capability

.TP
.I cap_usr_rdpmc
If the hardware supports user-space read of performance counters
without syscall (this is the "rdpmc" instruction on x86), then
the following code can be used to do a read:

.in +4n
.nf
u32 seq, time_mult, time_shift, idx, width;
u64 count, enabled, running;
u64 cyc, time_offset;
s64 pmc = 0;

do {
    seq = pc\->lock;
    barrier();
    enabled = pc\->time_enabled;
    running = pc\->time_running;

    if (pc\->cap_usr_time && enabled != running) {
        cyc = rdtsc();
        time_offset = pc\->time_offset;
        time_mult   = pc\->time_mult;
        time_shift  = pc\->time_shift;
    }

    idx = pc\->index;
    count = pc\->offset;

    if (pc\->cap_usr_rdpmc && idx) {
        width = pc\->pmc_width;
        pmc = rdpmc(idx \- 1);
    }

    barrier();
} while (pc\->lock != seq);
.fi
.in



.TP
.I pmc_width
If
.IR cap_usr_rdpmc ,
this field provides the bit-width of the value
read using the rdpmc or equivalent instruction.
This can be used to sign extend the result like:

.in +4n
.nf
pmc <<= 64 \- pmc_width;
pmc >>= 64 \- pmc_width; // signed shift right
count += pmc;
.fi
.in


.TP
.IR time_shift ", " time_mult ", " time_offset

If
.IR cap_usr_time ,
these fields can be used to compute the time
delta since time_enabled (in ns) using rdtsc or similar.
.nf

    u64 quot, rem;
    u64 delta;
    quot = (cyc >> time_shift);
    rem = cyc & ((1 << time_shift) \- 1);
    delta = time_offset + quot * time_mult +
            ((rem * time_mult) >> time_shift);
.fi

Where time_offset,time_mult,time_shift and cyc are read in the
seqcount loop described above.
This delta can then be added to
enabled and possible running (if idx), improving the scaling:
.nf

    enabled += delta;
    if (idx)
        running += delta;
    quot = count / running;
    rem  = count % running;
    count = quot * enabled + (rem * enabled) / running;
.fi

.TP
.I data_head
This points to the head of the data section.
On SMP-capable platforms, after reading the data_head value,
user-space should issue an rmb().

.TP
.I data_tail;
When the mapping is
.BR PROT_WRITE ,
the data_tail value should be written by
userspace to reflect the last read data.
In this case the kernel will not over-write unread data.

.RE


The following 2^n ring-buffer pages have the layout described below.

If
.I perf_event_attr.sample_id_all
is set, then all event types will
have the sample_type selected fields related to where/when (identity)
an event took place (TID, TIME, ID, CPU, STREAM_ID) described in
.B PERF_RECORD_SAMPLE
below, it will be stashed just after the
perf_event_header and the fields already present for the existing
fields, i.e., at the end of the payload.
That way a newer perf.data
file will be supported by older perf tools, with these new optional
fields being ignored.

The mmap values start with a header:

.in +4n
.nf
struct perf_event_header {
    __u32   type;
    __u16   misc;
    __u16   size;
};
.fi
.in

Below, we describe the
.I perf_event_header
fields in more detail.

.TP
.I type
The
.I type
value is one of the below.
The values in the corresponding record (that follows the header)
depend on the
.I type
selected as shown.

.RS
.TP
.B PERF_RECORD_MMAP
The MMAP events record the
.B PROT_EXEC
mappings so that we can correlate
userspace IPs to code.
They have the following structure:

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32    pid, tid;
    u64    addr;
    u64    len;
    u64    pgoff;
    char   filename[];
};
.in

.TP
.B PERF_RECORD_LOST
This record indicates when events are lost.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 id;
    u64 lost;
};
.fi
.in

.TP
.B PERF_RECORD_COMM
This record indicates a change in the process name.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    char comm[];
};
.fi
.in

.TP
.B PERF_RECORD_EXIT
This record indicates a process exit event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.BR PERF_RECORD_THROTTLE ", " PERF_RECORD_UNTHROTTLE
This record indicates a throttle/unthrottle event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 time;
    u64 id;
    u64 stream_id;
};
.fi
.in

.TP
.B PERF_RECORD_FORK
This record indicates a fork event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.B PERF_RECORD_READ
This record indicates a read event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    struct read_format values;
};
.fi
.in

.TP
.B PERF_RECORD_SAMPLE
This record indicates a sample.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64   ip;          /* if PERF_SAMPLE_IP      */
    u32   pid, tid;    /* if PERF_SAMPLE_TID     */
    u64   time;        /* if PERF_SAMPLE_TIME    */
    u64   addr;        /* if PERF_SAMPLE_ADDR    */
    u64   id;          /* if PERF_SAMPLE_ID      */
    u64   stream_id; /* if PERF_SAMPLE_STREAM_ID */
    u32   cpu, res;    /* if PERF_SAMPLE_CPU     */
    u64   period;      /* if PERF_SAMPLE_PERIOD  */
    struct read_format v; /* if PERF_SAMPLE_READ */
    u64   nr;        /* if PERF_SAMPLE_CALLCHAIN */
    u64   ips[nr];   /* if PERF_SAMPLE_CALLCHAIN */
    u32   size;        /* if PERF_SAMPLE_RAW     */
    char  data[size];  /* if PERF_SAMPLE_RAW     */
    u64   from;   /* if PERF_SAMPLE_BRANCH_STACK */
    u64   to;     /* if PERF_SAMPLE_BRANCH_STACK */
    u64   flags;  /* if PERF_SAMPLE_BRANCH_STACK */
    u64   lbr[nr];/* if PERF_SAMPLE_BRANCH_STACK */
};
.fi
.in

The RAW record data is opaque with respect to the ABI.
The ABI doesn't make any promises with respect to the stability
of its content, it may vary depending
on event, hardware, and kernel version.

.RE


.TP
.I misc
The
.I misc
field is one of the following:
.RS
.TP
.B PERF_RECORD_MISC_CPUMODE_MASK
[To be documented]
.TP
.B PERF_RECORD_MISC_CPUMODE_UNKNOWN
[To be documented]
.TP
.B PERF_RECORD_MISC_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_HYPERVISOR
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_EXACT_IP
This indicates that the content of
.B PERF_SAMPLE_IP
points
to the actual instruction that triggered the event.
See also
.IR perf_event_attr.precise_ip .

.RE

.TP
.I size
This indicates the size of the record.


.SS "Signal Overflow"

Counters can be set to signal when a threshold is crossed.
This is set up using traditional the
.BR poll (2),
.BR select (2),
.BR epoll (2)
and
.BR fcntl (2),
system calls.

Normally, a notification is generated for every page filled, however
one can additionally set
.I perf_event_attr.wakeup_events
to generate one every so many counter overflow events.

.SS "Reading Results"
Once a
.BR perf_event_open()
file descriptor  has been opened, the values
of the events can be read from the file descriptor.
The values that are there are specified by the
.I read_format
field in the attr structure at open time.

If you attempt to read into a buffer that is not big enough to hold the
data, an error is returned
.IR ( ENOSPC ).

Here is the layout of the data returned by a read.

If
.B PERF_FORMAT_GROUP
was specified to allow reading all events in a group at once:

.in +4n
.nf
struct {
    u64 nr;             /* The number of events */
    u64 time_enabled;   /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;   /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    struct {
        u64 value;      /* The value of the event */
        u64 id;         /* if PERF_FORMAT_ID */
    } values[nr];
};
.fi
.in

If
.B PERF_FORMAT_GROUP
was
.I not
specified, then the read values look as following:

.in +4n
.nf
struct {
    u64 value;         /* The value of the event */
    u64 time_enabled;  /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;  /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    u64 id;            /* if PERF_FORMAT_ID */
};
.fi
.in

The values read are described in more detail below.
.RS
.TP
.I nr
The number of events in this file descriptor.
Only available if
.B PERF_FORMAT_GROUP
was specified.

.TP
.IR time_enabled ", " time_running
Total time the event was enabled and running.
Normally these are the same.
If more events are started
than available counter slots on the PMU, then multiplexing
happens and events only run part of the time.
In that case the
.I time_enabled
and
.I time running
values can be used to scale an estimated value for the count.

.TP
.I value
An unsigned 64-bit value containing the counter result.

.TP
.I id
A globally unique value for this particular event, only there if
.B PERF_FORMAT_ID
was specified in read_format.

.RE




.RE

.SS "rdpmc instruction"
Starting with Linux 3.4 on x86, you can use the
.I rdpmc
instruction to get low-latency reads without having to enter the kernel.


.SS "perf_event ioctl calls"
.PP
Various ioctls act on
.BR perf_event_open()
file descriptors
.\" FIXME the arguments for these ioctl() operations need to be described
.TP
.B PERF_EVENT_IOC_ENABLE
Enables an individual counter or counter group.

.TP
.B PERF_EVENT_IOC_DISABLE
Disables an individual counter or counter group.

Enabling or disabling the leader of a group enables or disables the
entire group; that is, while the group leader is disabled, none of the
counters in the group will count.
Enabling or disabling a member of a group other than the leader only
affects that counter; disabling a non-leader
stops that counter from counting but doesn't affect any other counter.

.TP
.B PERF_EVENT_IOC_REFRESH
Non-inherited overflow counters can use this
to enable a counter for 'nr' events, after which it gets disabled again.
.\" FIXME the following needs clarification/confirmation
I think the goal of IOC_REFRESH is not to reload the period but simply to
adjust the number of events before the next notifications.

.TP
.B PERF_EVENT_IOC_RESET
Reset the event count to zero.
This only resets the counts; there is no way to reset the
multiplexing
.I time_enabled
or
.I time_running
values.
When sent to a group leader, only
the leader is reset (child events are not).

.TP
.B PERF_EVENT_IOC_PERIOD
IOC_PERIOD is the command to update the period; it
does not update the current period but instead defers until next.

.TP
.B PERF_EVENT_IOC_SET_OUTPUT
This tells the kernel to report event notifications to the specified
file descriptor rather than the default one.
The file descriptors must all be on the same CPU.

.TP
.BR PERF_EVENT_IOC_SET_FILTER " (Since Linux 2.6.33)"
This adds an ftrace filter to this event.

.SS "Using prctl"
A process can enable or disable all the counter groups that are
attached to it using the
.BR prctl (2)
.B PR_TASK_PERF_EVENTS_ENABLE
and
.B PR_TASK_PERF_EVENTS_DISABLE
operations.
This applies to all counters on the current process, whether created by
this process or by another, and does not affect any counters that this
process has created on other processes.
It only enables or disables
the group leaders, not any other members in the groups.

.SS /proc/sys/kernel/perf_event_paranoid

The
.I /proc/sys/kernel/perf_event_paranoid
file can be set to restrict access to the performance counters.
2
means no measurements allowed,
1
means normal counter access,
0
means you can access CPU-specific data, and
\-1
means no restrictions.

The existence of the
.I perf_event_paranoid
file is the official method for determining if a kernel supports
.BR perf_event_open().

.SH "RETURN VALUE"
.BR perf_event_open ()
returns the new file descriptor, or \-1 if an error occurred
(in which case,
.I errno
is set appropriately).
.SH ERRORS
.TP
.B EINVAL
Returned if the specified event is not available.
.TP
.B ENOSPC
Prior to Linux 3.3, if there was no counter room,
.B ENOSPC
was returned.
Linus did not like this, and this was changed to
.BR EINVAL .
.B ENOSPC
is still returned if you try to read results into
too small a buffer.

.SH VERSION

.BR perf_event_open ()
was introduced in Linux 2.6.31 but was called
.BR perf_counter_open () .
It was renamed in Linux 2.6.32.

.SH CONFORMING TO

This call is specific to Linux
and should not be used in programs intended to be portable.

.SH NOTES
Glibc does not provide a wrapper for this system call; call it using
.BR syscall (2).

The official way of knowing if
.BR perf_event_open()
support is enabled is checking
for the existence of the file
.I /proc/sys/kernel/perf_event_paranoid

.SH BUGS

The
.B F_SETOWN_EX
option to
.IR fcntl (2)
is needed to properly get overflow signals in threads.
This was introduced in Linux 2.6.32.

Prior to Linux 2.6.33 (at least for x86) the kernel did not check
if events could be scheduled together until read time.
The same happens on all known kernels if the NMI watchdog is enabled.
This means to see if a given set of events works you have to
.BR perf_event_open (),
start, then read before you know for sure you
can get valid measurements.

Prior to Linux 2.6.34 event constraints were not enforced by the kernel.
In that case, some events would silently return "0" if the kernel
scheduled them in an improper counter slot.

Prior to Linux 2.6.34 there was a bug when multiplexing where the
wrong results could be returned.

Kernels from Linux 2.6.35 to Linux 2.6.39 can quickly crash the kernel if
"inherit" is enabled and many threads are started.

Prior to Linux 2.6.35,
.B PERF_FORMAT_GROUP
did not work with attached processes.

In older Linux 2.6 versions,
refreshing an event group leader refreshed all siblings,
and refreshing with a parameter of 0 enabled infinite refresh.
This behavior is unsupported and should not be relied on.

There is a bug in the kernel code between
Linux 2.6.36 and Linux 3.0 that ignores the
"watermark" field and acts as if a wakeup_event
was chosen if the union has a
non-zero value in it.

Always double-check your results!  Various generalized events
have had wrong values.
For example, retired branches measured
the wrong thing on AMD machines until Linux 2.6.35.

.SH EXAMPLE
The following is a short example that measures the total
instruction count of a call to printf().
.nf

#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include <asm/unistd.h>

long perf_event_open( struct perf_event_attr *hw_event, pid_t pid,
                      int cpu, int group_fd, unsigned long flags )
{
    int ret;

    ret = syscall( __NR_perf_event_open, hw_event, pid, cpu,
                   group_fd, flags );
    return ret;
}


int
main(int argc, char **argv)
{

    struct perf_event_attr pe;
    long long count;
    int fd;

    memset(&pe, 0, sizeof(struct perf_event_attr));
    pe.type = PERF_TYPE_HARDWARE;
    pe.size = sizeof(struct perf_event_attr);
    pe.config = PERF_COUNT_HW_INSTRUCTIONS;
    pe.disabled = 1;
    pe.exclude_kernel = 1;
    pe.exclude_hv = 1;

    fd = perf_event_open(&pe, 0, \-1, \-1, 0);
    if (fd < 0) {
       fprintf(stderr, "Error opening leader %llx\\n", pe.config);
    }

    ioctl(fd, PERF_EVENT_IOC_RESET, 0);
    ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);

    printf("Measuring instruction count for this printf\\n");

    ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
    read(fd, &count, sizeof(long long));

    printf("Used %lld instructions\\n", count);

    close(fd);
}
.fi

.SH "SEE ALSO"
.BR fcntl (2),
.BR mmap (2),
.BR open (2),
.BR prctl (2),
.BR read (2)

[-- Attachment #2: Type: TEXT/PLAIN, Size: 43520 bytes --]

.\" Hey Emacs! This file is -*- nroff -*- source.
.\"
.\" Copyright (c) 2012, Vincent Weaver
.\"
.\" This is free documentation; you can redistribute it and/or
.\" modify it under the terms of the GNU General Public License as
.\" published by the Free Software Foundation; either version 2 of
.\" the License, or (at your option) any later version.
.\"
.\" The GNU General Public License's references to "object code"
.\" and "executables" are to be interpreted as the output of any
.\" document formatting or typesetting system, including
.\" intermediate and printed output.
.\"
.\" This manual is distributed in the hope that it will be useful,
.\" but WITHOUT ANY WARRANTY; without even the implied warranty of
.\" MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
.\" GNU General Public License for more details.
.\"
.\" You should have received a copy of the GNU General Public
.\" License along with this manual; if not, see
.\" <http://www.gnu.org/licenses/>.
.\"
.\" This document is based on the perf_event.h header file, the
.\" tools/perf/design.txt file, and a lot of bitter experience.
.\"
.TH PERF_EVENT_OPEN 2 2012-10-23 "Linux" "Linux Programmer's Manual"
.SH NAME
perf_event_open \- set up performance monitoring
.SH SYNOPSIS
.nf
.B #include <linux/perf_event.h>
.B #include <linux/hw_breakpoint.h>
.sp
.BI "int perf_event_open(struct perf_event_attr *" hw_event ,
.BI "                    pid_t " pid ", int " cpu ", int " group_fd ,
.BI "                    unsigned long " flags  );
.fi

.IR Note :
There is no glibc wrapper for this system call; see NOTES.
.SH DESCRIPTION
Given a list of parameters,
.BR perf_event_open ()
returns a file descriptor, for use in subsequent system calls
.RB ( read "(2), " mmap "(2), " prctl "(2), " fcntl "(2), etc.)."
.PP
A call to
.BR perf_event_open ()
creates a file descriptor that allows measuring performance
information.
Each file descriptor corresponds to one
event that is measured; these can be grouped together
to measure multiple events simultaneously.
.PP
Events can be enabled and disabled in two ways: via
.BR ioctl (2)
and via
.BR prctl (2) .
When an event is disabled it does not count or generate overflows but does
continue to exist and maintain its count value.
Events come in two flavors: counting and sampled.
A
.I counting
event is one that is used for counting the aggregate number of events
that occur.
In general, counting event results are gathered with a
.BR read (2)
call.
A
.I sampling
event periodically writes measurements to a buffer that can then
be accessed via
.BR  mmap (2) .
.SS Arguments
.P
The argument
.I pid
allows events to be attached to processes in various ways.
If
.I pid
is 0, measurements happen on the current task, if
.I pid
is greater than 0, the process indicated by
.I pid
is measured, and if
.I pid
is less than 0, all processes are counted.

The
.I cpu
argument allows measurements to be specific to a CPU.
If
.I cpu
is greater than or equal to 0,
measurements are restricted to the specified CPU;
if
.I cpu
is \-1, the events are measured on all CPUs.
.P
Note that the combination of
.IR pid " == \-1"
and
.IR cpu " == \-1"
is not valid.
.P
A
.IR pid " > 0"
and
.IR cpu " == \-1"
setting measures per-process and follows that process to whatever CPU the
process gets scheduled to.
Per-process events can be created by any user.
.P
A
.IR pid " == \-1"
and
.IR cpu " >= 0"
setting is per-CPU and measures all processes on the specified CPU.
Per-CPU events need the
.B CAP_SYS_ADMIN
capability.
.P
The
.I group_fd
argument allows counter groups to be set up.
A counter group has one counter which is the group leader.
The leader is created first, with
.IR group_fd " = \-1"
in the
.BR perf_event_open ()
call that creates it.
The rest of the group members are created subsequently, with
.IR group_fd
giving the fd of the group leader.
(A single counter on its own is created with
.IR group_fd " = \-1"
and is considered to be a group with only 1 member.)
.P
A counter group is scheduled onto the CPU as a unit: it will only
be put onto the CPU if all of the counters in the group can be put onto
the CPU.
This means that the values of the member counters can be
meaningfully compared, added, divided (to get ratios), etc., with each
other, since they have counted events for the same set of executed
instructions.
.P
The
.I flags
argument takes one of the following values:
.TP
.BR PERF_FLAG_FD_NO_GROUP
.\" FIXME The following sentence is unclear
This flag allows creating an event as part of an event group but
having no group leader.
It is unclear why this is useful.
.\" FIXME So, why is it useful?
.TP
.BR PERF_FLAG_FD_OUTPUT
This flag re-routes the output from an event to the group leader.
.TP
.BR PERF_FLAG_PID_CGROUP " (Since Linux 2.6.39)."
This flag activates per-container system-wide monitoring.
A container
is an abstraction that isolates a set of resources for finer grain
control (CPUs, memory, etc...).
In this mode, the event is measured
only if the thread running on the monitored CPU belongs to the designated
container (cgroup).
The cgroup is identified by passing a file descriptor
opened on its directory in the cgroupfs filesystem.
For instance, if the
cgroup to monitor is called
.IR test ,
then a file descriptor opened on
.I /dev/cgroup/test
(assuming cgroupfs is mounted on
.IR /dev/cgroup )
must be passed as the
.I pid
parameter.
cgroup monitoring is only available
for system-wide events and may therefore require extra permissions.
.P
The
.I perf_event_attr
structure is what is passed into the
.BR perf_event_open ()
syscall.
It is large and has a complicated set of dependent fields.

.in +4n
.nf
struct perf_event_attr {
    __u32     type;         /* Type of event */
    __u32     size;         /* Size of attribute structure */
    __u64     config;       /* Type-specific configuration */

    union {
        __u64 sample_period;    /* Period of sampling */
        __u64 sample_freq;      /* Frequency of sampling */
    };

    __u64     sample_type;  /* Specifies values included in sample */
    __u64     read_format;  /* Specifies values returned in read */

    __u64     disabled       : 1,   /* off by default         */
              inherit        : 1,   /* children inherit it    */
              pinned         : 1,   /* must always be on PMU  */
              exclusive      : 1,   /* only group on PMU      */
              exclude_user   : 1,   /* don't count user       */
              exclude_kernel : 1,   /* don't count kernel     */
              exclude_hv     : 1,   /* don't count hypervisor */
              exclude_idle   : 1,   /* don't count when idle  */
              mmap           : 1,   /* include mmap data      */
              comm           : 1,   /* include comm data      */
              freq           : 1,   /* use freq, not period   */
              inherit_stat   : 1,   /* per task counts        */
              enable_on_exec : 1,   /* next exec enables      */
              task           : 1,   /* trace fork/exit        */
              watermark      : 1,   /* wakeup_watermark       */
              precise_ip     : 2,   /* skid constraint        */
              mmap_data      : 1,   /* non-exec mmap data     */
              sample_id_all  : 1,   /* sample_type all events */
              exclude_host   : 1,   /* don't count in host    */
              exclude_guest  : 1,   /* don't count in guest   */
              __reserved_1   : 43;

    union {
        __u32 wakeup_events;    /* wakeup every n events */
        __u32 wakeup_watermark; /* bytes before wakeup   */
    };

    __u32     bp_type;          /* breakpoint type */

    union {
        __u64 bp_addr;          /* breakpoint address */
        __u64 config1;          /* extension of config */
    };

    union {
        __u64 bp_len;           /* breakpoint length */
        __u64 config2;          /* extension of config1 */
    };
    __u64     branch_sample_type;   /* enum branch_sample_type */
};
.fi
.in

The fields of the
.I perf_event_attr
structure are described in more detail below.

.TP
.I type
This field specifies the overall event type.
It has one of the following values:
.RS
.TP
.B PERF_TYPE_HARDWARE
This indicates one of the "generalized" hardware events provided
by the kernel.
See the
.I config
field definition for more details.
.TP
.B PERF_TYPE_SOFTWARE
This indicates one of the software-defined events provided by the kernel
(even if no hardware support is available).
.TP
.B PERF_TYPE_TRACEPOINT
This indicates a tracepoint
provided by the kernel tracepoint infrastructure.
.TP
.B PERF_TYPE_HW_CACHE
This indicates a hardware cache event.
This has a special encoding, described in the
.I config
field definition.
.TP
.B PERF_TYPE_RAW
This indicates a "raw" implementation-specific event in the
.IR config " field."
.TP
.BR PERF_TYPE_BREAKPOINT " (Since Linux 2.6.33)"
This indicates a hardware breakpoint as provided by the CPU.
Breakpoints can be read/write accesses to an address as well as
execution of an instruction address.
.TP
.RB "dynamic PMU"
Since Linux 2.6.39,
.BR perf_event_open()
can support multiple PMUs.
To enable this, a value exported by the kernel can be used in the
.I type
field to indicate which PMU to use.
The value to use can be found in the sysfs filesystem:
there is a subdirectory per PMU instance under
.IR /sys/devices .
In each sub-directory there is a
.I type
file whose content is an integer that can be used in the
.I type
field.
For instance,
.I /sys/devices/cpu/type
contains the value for the core CPU PMU, which is usually 4.
.RE

.TP
.I "size"
The size of the
.I perf_event_attr
structure for forward/backward compatibility.
Set this using
.I sizeof(struct perf_event_attr)
to allow the kernel to see
the struct size at the time of compilation.

The related define
.B PERF_ATTR_SIZE_VER0
is set to 64; this was the size of the first published struct.
.B PERF_ATTR_SIZE_VER1
is 72, corresponding to the addition of breakpoints in Linux 2.6.33.
.B PERF_ATTR_SIZE_VER2
is 80 corresponding to the addition of branch sampling in Linux 3.4.

.TP
.I "config"
This specifies which event you want, in conjunction with
the
.I type
field.
The
.IR config1 " and " config2
fields are also taken into account in cases where 64 bits is not
enough to fully specify the event.
The encoding of these fields are event dependent.

The most significant bit (bit 63) of
.I config
signifies CPU-specific (raw) counter configuration data;
if the most significant bit is unset, the next 7 bits are an event
type and the rest of the bits are the event identifier.

There are various ways to set the
.I config
field that are dependent on the value of the previously
described
.I type
field.
What follows are various possible settings for
.I config
separated out by
.IR type .

If
.I type
is
.BR PERF_TYPE_HARDWARE ,
we are measuring one of the generalized hardware CPU events.
Not all of these are available on all platforms.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_HW_CPU_CYCLES
Total cycles.
Be wary of what happens during CPU frequency scaling
.TP
.B PERF_COUNT_HW_INSTRUCTIONS
Retired instructions.
Be careful, these can be affected by various
issues, most notably hardware interrupt counts
.TP
.B PERF_COUNT_HW_CACHE_REFERENCES
Cache accesses.
Usually this indicates Last Level Cache accesses but this may
vary depending on your CPU.
This may include prefetches and coherency messages; again this
depends on the design of your CPU.
.TP
.B PERF_COUNT_HW_CACHE_MISSES
Cache misses.
Usually this indicates Last Level Cache misses; this is intended to be
used in conjunction with the
.B PERF_COUNT_HW_CACHE_REFERENCES
event to calculate cache miss rates.
.TP
.B PERF_COUNT_HW_BRANCH_INSTRUCTIONS
Retired branch instructions.
Prior to Linux 2.6.34, this used
the wrong event on AMD processors.
.TP
.B PERF_COUNT_HW_BRANCH_MISSES
Mispredicted branch instructions.
.TP
.B PERF_COUNT_HW_BUS_CYCLES
Bus cycles, which can be different from total cycles.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_FRONTEND " (Since Linux 3.0)"
Stalled cycles during issue.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_BACKEND  " (Since Linux 3.0)"
Stalled cycles during retirement.
.TP
.BR PERF_COUNT_HW_REF_CPU_CYCLES  " (Since Linux 3.3)"
Total cycles; not affected by CPU frequency scaling.
.RE
.IP
If
.I type
is
.BR PERF_TYPE_SOFTWARE ,
we are measuring software events provided by the kernel.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_SW_CPU_CLOCK
This reports the CPU clock, a high-resolution per-CPU timer.
.TP
.B PERF_COUNT_SW_TASK_CLOCK
This reports a clock count specific to the task that is running.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS
This reports the number of page faults.
.TP
.B PERF_COUNT_SW_CONTEXT_SWITCHES
This counts context switches.
Until Linux 2.6.34, these were all reported as user-space
events, after that they are reported as happening in the kernel.
.TP
.B PERF_COUNT_SW_CPU_MIGRATIONS
This reports the number of times the process
has migrated to a new CPU.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MIN
This counts the number of minor page faults.
These did not require disk I/O to handle.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MAJ
This counts the number of major page faults.
These required disk I/O to handle.
.TP
.BR PERF_COUNT_SW_ALIGNMENT_FAULTS " (Since Linux 2.6.33)"
This counts the number of alignment faults.
These happen when unaligned memory accesses happen; the kernel
can handle these but it reduces performance.
This only happens on some architectures (never on x86).
.TP
.BR PERF_COUNT_SW_EMULATION_FAULTS " (Since Linux 2.6.33)"
This counts the number of emulation faults.
The kernel sometimes traps on unimplemented instructions
and emulates them for userspace.
This can negatively impact performance.
.RE
.RE


.RS
If
.I type
is
.BR PERF_TYPE_TRACEPOINT ,
then we are measuring kernel tracepoints.
The value to use in
.I config
can be obtained from under debugfs
.I tracing/events/*/*/id
if ftrace is enabled in the kernel.

.RE

.RS
If
.I type
is
.BR PERF_TYPE_HW_CACHE ,
then we are measuring a hardware CPU cache event.
To calculate the appropriate
.I config
value use the following equation:
.RS 4
.nf

    (perf_hw_cache_id) | (perf_hw_cache_op_id << 8) |
    (perf_hw_cache_op_result_id << 16)
.fi
.P
where
.I perf_hw_cache_id
is one of:
.RS
.TP
.B PERF_COUNT_HW_CACHE_L1D
for measuring Level 1 Data Cache
.TP
.B PERF_COUNT_HW_CACHE_L1I
for measuring Level 1 Instruction Cache
.TP
.B PERF_COUNT_HW_CACHE_LL
for measuring Last-Level Cache
.TP
.B PERF_COUNT_HW_CACHE_DTLB
for measuring the Data TLB
.TP
.B PERF_COUNT_HW_CACHE_ITLB
for measuring the Instruction TLB
.TP
.B PERF_COUNT_HW_CACHE_BPU
for measuring the branch prediction unit
.TP
.BR PERF_COUNT_HW_CACHE_NODE " (Since Linux 3.0)"
for measuring local memory accesses
.RE

.P
and
.I perf_hw_cache_op_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_OP_READ
for read accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_WRITE
for write accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_PREFETCH
for prefetch accesses
.RE

.P
and
.I perf_hw_cache_op_result_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_RESULT_ACCESS
to measure accesses
.TP
.B PERF_COUNT_HW_CACHE_RESULT_MISS
to measure misses
.RE
.RE

If
.I type
is
.BR PERF_TYPE_RAW ,
then a custom "raw"
.I config
value is needed.
Most CPUs support events that are not covered by the "generalized" events.
These are implementation defined; see your CPU manual (for example
the Intel Volume 3B documentation or the AMD BIOS and Kernel Developer
Guide).
The libpfm4 library can be used to translate from the name in the
architectural manuals to the raw hex value
.BR perf_event_open ()
expects in this field.

If
.I type
is
.BR PERF_TYPE_BREAKPOINT ,
then leave
.I config
set to zero.
Its parameters are set in other places.
.RE
.TP
.IR sample_period ", " sample_freq
A "sampling" counter is one that generates an interrupt
every N events, where N is given by
.IR sample_period .
A sampling counter has
.IR sample_period " > 0."
The
.I sample_type
field controls what data is recorded on each interrupt.

.I sample_freq
can be used if you wish to use frequency rather than period.
In this case you set the
.I freq
flag.
The kernel will adjust the sampling period
to try and achieve the desired rate.
The rate of adjustment is a
timer tick.


.TP
.I "sample_type"
The various bits in this field specify which values to include
in the overflow packets.
They will be recorded in a ring-buffer,
which is available to user-space using
.BR mmap (2).
The order in which the values are saved in the
overflow packets as documented in the MMAP Layout subsection below;
it is not the
.I "enum perf_event_sample_format"
order.
.RS
.TP
.B PERF_SAMPLE_IP
instruction pointer
.TP
.B PERF_SAMPLE_TID
thread id
.TP
.B PERF_SAMPLE_TIME
time
.TP
.B PERF_SAMPLE_ADDR
address
.TP
.B PERF_SAMPLE_READ
[To be documented]
.TP
.B PERF_SAMPLE_CALLCHAIN
[To be documented]
.TP
.B PERF_SAMPLE_ID
[To be documented]
.TP
.B PERF_SAMPLE_CPU
[To be documented]
.TP
.B PERF_SAMPLE_PERIOD
[To be documented]
.TP
.B PERF_SAMPLE_STREAM_ID
[To be documented]
.TP
.B PERF_SAMPLE_RAW
[To be documented]
.TP
.BR PERF_SAMPLE_BRANCH_STACK " (Since Linux 3.4)"
[To be documented]
.RE

.TP
.IR "read_format"
This field specifies the format of the data returned by
.BR read (2)
on a
.BR perf_event_open()
file descriptor.
.RS
.TP
.B PERF_FORMAT_TOTAL_TIME_ENABLED
Adds the 64-bit "time_enabled" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and multiplexing is happening.
.TP
.B PERF_FORMAT_TOTAL_TIME_RUNNING
Adds the 64-bit "time_running" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and  multiplexing is happening.
.TP
.B PERF_FORMAT_ID
Adds a 64-bit unique value that corresponds to the event-group.
.TP
.B PERF_FORMAT_GROUP
Allows all counter values in an event-group to be read with one read.
.RE

.TP
.IR "disabled"
The
.I disabled
bit specifies whether the counter starts out disabled or enabled.
If disabled, the event can later be enabled by
.BR ioctl (2),
.BR prctl (2),
or
.IR enable_on_exec .

.TP
.IR "inherit"
The
.I inherit
bit specifies that this counter should count events of child
tasks as well as the task specified.
This only applies to new children, not to any existing children at
the time the counter is created (nor to any new children of
existing children).

Inherit does not work for some combinations of
.IR read_format s,
such as
.BR PERF_FORMAT_GROUP .

.TP
.IR "pinned"
The
.I pinned
bit specifies that the counter should always be on the CPU if at all
possible.
It only applies to hardware counters and only to group leaders.
If a pinned counter cannot be put onto the CPU (e.g., because there are
not enough hardware counters or because of a conflict with some other
event), then the counter goes into an 'error' state, where reads
return end-of-file (i.e.,
.BR read (2)
returns 0) until the counter is subsequently enabled or disabled.

.TP
.IR "exclusive"
The
.I exclusive
bit specifies that when this counter's group is on the CPU,
it should be the only group using the CPU's counters.
In the future this may allow monitoring programs to
support PMU features that need to run alone so that they do not
disrupt other hardware counters.

.TP
.IR "exclude_user"
If this bit is set, the count excludes events that happen in user-space.

.TP
.IR "exclude_kernel"
If this bit is set, the count excludes events that happen in kernel-space.

.TP
.IR "exclude_hv"
If this bit is set, the count excludes events that happen in the
hypervisor.
This is mainly for PMUs that have built-in support for handling this
(such as POWER).
Extra support is needed for handling hypervisor measurements on most
machines.

.TP
.IR "exclude_idle"
If set, don't count when the CPU is idle.

.TP
.IR "mmap"
The
.I mmap
bit enables recording of extra information to a
mmap'd ring-buffer.
This is
described below in subsection MMAP Layout.

.TP
.IR "comm"
The
.I comm
bit enables tracking of process command name as modified by the
.IR exec (2)
and
.IR prctl (PR_SET_NAME)
system calls.
Unfortunately for tools,
there is no way to distinguish one system call versus the other.

.TP
.IR "freq"
If this bit is set, then
.I sample_frequency
not
.I sample_period
is used when setting up the sampling interval.

.TP
.IR "inherit_stat"
This bit enables saving of event counts on context switch for
inherited tasks.
This is only meaningful if the
.I inherit
field is set.

.TP
.IR "enable_on_exec"
If this bit is set, a counter is automatically
enabled after a call to
.BR exec (2).

.TP
.IR "task"
If this bit is set, then
fork/exit notifications are included in the ring buffer.

.TP
.IR "watermark"
If set, have a sampling interrupt happen when we cross the
wakeup_watermark boundary.

.TP
.IR "precise_ip" " (Since Linux 2.6.35)"
This controls the amount of skid.
Skid is how many instructions
execute between an event of interest happening and the kernel
being able to stop and record the event.
Smaller skid is
better and allows more accurate reporting of which events
correspond to which instructions, but hardware is often limited
with how small this can be.

The values of this are the following:
.RS
.TP
0 -
.B SAMPLE_IP
can have arbitrary skid
.TP
1 -
.B SAMPLE_IP
must have constant skid
.TP
2 -
.B SAMPLE_IP
requested to have 0 skid
.TP
3 -
.B SAMPLE_IP
must have 0 skid.
See also
.BR PERF_RECORD_MISC_EXACT_IP .
.RE

.TP
.IR "mmap_data" " (Since Linux 2.6.36)"
Include mmap events in the ring_buffer.

.TP
.IR "sample_id_all" " (Since Linux 2.6.38)"
If set, then all sample ID info (TID, TIME, ID, CPU, STREAM_ID)
will be provided.

.TP
.IR "exclude_host" " (Since Linux 3.2)"
Do not measure time spent in VM host

.TP
.IR "exclude_guest" " (Since Linux 3.2)"
Do not measure time spent in VM guest


.TP
.IR "wakeup_events" ", " "wakeup_watermark"
This union sets how many events
.RI ( wakeup_events )
or bytes
.RI ( wakeup_watermark )
happen before an overflow signal happens.
Which one is used is selected by the
.I watermark
bitflag.

.TP
.IR "bp_type" " (Since Linux 2.6.33)"
This chooses the breakpoint type.
It is one of:
.RS
.TP
.BR HW_BREAKPOINT_EMPTY
no breakpoint
.TP
.BR HW_BREAKPOINT_R
count when we read the memory location
.TP
.BR HW_BREAKPOINT_W
count when we write the memory location
.TP
.BR HW_BREAKPOINT_RW
count when we read or write the memory location
.TP
.BR HW_BREAKPOINT_X
count when we execute code at the memory location

.LP
The values can be combined via a bitwsie or, but the
combination of
.B HW_BREAKPOINT_R
or
.B HW_BREAKPOINT_W
with
.B HW_BREAKPOINT_X
is not allowed.
.RE

.TP
.IR "bp_addr" " (Since Linux 2.6.33)"
.I bp_addr
address of the breakpoint.
For execution breakpoints this is the memory address of the instruction
of interest; for read and write breakpoints it is the memory address
of the memory location of interest.

.TP
.IR "config1" " (Since Linux 2.6.39)"
.I config1
is used for setting events that need an extra register or otherwise
do not fit in the regular config field.
Raw OFFCORE_EVENTS on Nehalem/Westmere/SandyBridge use this field
on 3.3 and later kernels.

.TP
.IR "bp_len" " (Since Linux 2.6.33)"
.I bp_len
is the length of the breakpoint being measured if
.I type
is
.BR PERF_TYPE_BREAKPOINT .
Options are
.BR HW_BREAKPOINT_LEN_1 ,
.BR HW_BREAKPOINT_LEN_2 ,
.BR HW_BREAKPOINT_LEN_4 ,
.BR HW_BREAKPOINT_LEN_8 .
For an execution breakpoint, set this to
.IR sizeof(long) .

.TP
.IR "config2" " (Since Linux 2.6.39)"

.I config2
is a further extension of the
.I config1
field.

.TP
.IR "branch_sample_type" " (Since Linux 3.4)"
This is used with the CPUs hardware branch sampling, if available.
It can have one of the following values:
.RS
.TP
.B PERF_SAMPLE_BRANCH_USER
Branch target is in user space
.TP
.B PERF_SAMPLE_BRANCH_KERNEL
Branch target is in kernel space
.TP
.B PERF_SAMPLE_BRANCH_HV
Branch target is in hypervisor
.TP
.B PERF_SAMPLE_BRANCH_ANY
Any branch type.
.TP
.B PERF_SAMPLE_BRANCH_ANY_CALL
Any call branch
.TP
.B PERF_SAMPLE_BRANCH_ANY_RETURN
Any return branch
.TP
.BR PERF_SAMPLE_BRANCH_IND_CALL
Indirect calls
.TP
.BR PERF_SAMPLE_BRANCH_PLM_ALL
User, kernel, and hv
.RE


.SS "MMAP Layout"

When using
.BR perf_event_open()
in sampled mode, asynchronous events
(like counter overflow or
.B PROT_EXEC
mmap tracking)
are logged into a ring-buffer.
This ring-buffer is created and accessed through
.BR mmap (2).

The mmap size should be 1+2^n pages, where the first page is a
metadata page
.IR ( "struct perf_event_mmap_page" )
that contains various
bits of information such as where the ring-buffer head is.

Before kernel 2.6.39, there is a bug that means you must allocate a mmap
ring buffer when sampling even if you do not plan to access it.

The structure of the first metadata mmap page is as follows:

.in +4n
.nf
struct perf_event_mmap_page {
    __u32 version;          /* version number of this structure   */
    __u32 compat_version;   /* lowest version this is compat with */
    __u32 lock;             /* seqlock for synchronization        */
    __u32 index;            /* hardware counter identifier        */
    __s64 offset;           /* add to hardware counter value      */
    __u64 time_enabled;     /* time event active                  */
    __u64 time_running;     /* time event on CPU                  */
    union {
        __u64   capabilities;
        __u64   cap_usr_time  : 1,
                cap_usr_rdpmc : 1,
    };
    __u16   pmc_width;
    __u16   time_shift;
    __u32   time_mult;
    __u64   time_offset;
    __u64   __reserved[120];   /* Pad to 1k                */
    __u64   data_head;         /* head in the data section */
    __u64   data_tail;         /* user-space written tail  */
}
.fi
.in



The following looks at the fields in the
.I perf_event_mmap_page
structure in more detail.

.RS

.TP
.I version
Version number of this structure.

.TP
.I compat_version
The lowest version this is compatible with.

.TP
.I lock
A seqlock for synchronization.

.TP
.I index;
A unique hardware counter identifier.

.TP
.I offset
.\" FIXME clarify
Add this to hardware counter value??

.TP
.I time_enabled
Time the event was active.

.TP
.I time_running
Time the event was running.

.TP
.I cap_usr_time
User time capability

.TP
.I cap_usr_rdpmc
If the hardware supports user-space read of performance counters
without syscall (this is the "rdpmc" instruction on x86), then
the following code can be used to do a read:

.in +4n
.nf
u32 seq, time_mult, time_shift, idx, width;
u64 count, enabled, running;
u64 cyc, time_offset;
s64 pmc = 0;

do {
    seq = pc\->lock;
    barrier();
    enabled = pc\->time_enabled;
    running = pc\->time_running;

    if (pc\->cap_usr_time && enabled != running) {
        cyc = rdtsc();
        time_offset = pc\->time_offset;
        time_mult   = pc\->time_mult;
        time_shift  = pc\->time_shift;
    }

    idx = pc\->index;
    count = pc\->offset;

    if (pc\->cap_usr_rdpmc && idx) {
        width = pc\->pmc_width;
        pmc = rdpmc(idx \- 1);
    }

    barrier();
} while (pc\->lock != seq);
.fi
.in



.TP
.I pmc_width
If
.IR cap_usr_rdpmc ,
this field provides the bit-width of the value
read using the rdpmc or equivalent instruction.
This can be used to sign extend the result like:

.in +4n
.nf
pmc <<= 64 \- pmc_width;
pmc >>= 64 \- pmc_width; // signed shift right
count += pmc;
.fi
.in


.TP
.IR time_shift ", " time_mult ", " time_offset

If
.IR cap_usr_time ,
these fields can be used to compute the time
delta since time_enabled (in ns) using rdtsc or similar.
.nf

    u64 quot, rem;
    u64 delta;
    quot = (cyc >> time_shift);
    rem = cyc & ((1 << time_shift) \- 1);
    delta = time_offset + quot * time_mult +
            ((rem * time_mult) >> time_shift);
.fi

Where time_offset,time_mult,time_shift and cyc are read in the
seqcount loop described above.
This delta can then be added to
enabled and possible running (if idx), improving the scaling:
.nf

    enabled += delta;
    if (idx)
        running += delta;
    quot = count / running;
    rem  = count % running;
    count = quot * enabled + (rem * enabled) / running;
.fi

.TP
.I data_head
This points to the head of the data section.
On SMP-capable platforms, after reading the data_head value,
user-space should issue an rmb().

.TP
.I data_tail;
When the mapping is
.BR PROT_WRITE ,
the data_tail value should be written by
userspace to reflect the last read data.
In this case the kernel will not over-write unread data.

.RE


The following 2^n ring-buffer pages have the layout described below.

If
.I perf_event_attr.sample_id_all
is set, then all event types will
have the sample_type selected fields related to where/when (identity)
an event took place (TID, TIME, ID, CPU, STREAM_ID) described in
.B PERF_RECORD_SAMPLE
below, it will be stashed just after the
perf_event_header and the fields already present for the existing
fields, i.e., at the end of the payload.
That way a newer perf.data
file will be supported by older perf tools, with these new optional
fields being ignored.

The mmap values start with a header:

.in +4n
.nf
struct perf_event_header {
    __u32   type;
    __u16   misc;
    __u16   size;
};
.fi
.in

Below, we describe the
.I perf_event_header
fields in more detail.

.TP
.I type
The
.I type
value is one of the below.
The values in the corresponding record (that follows the header)
depend on the
.I type
selected as shown.

.RS
.TP
.B PERF_RECORD_MMAP
The MMAP events record the
.B PROT_EXEC
mappings so that we can correlate
userspace IPs to code.
They have the following structure:

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32    pid, tid;
    u64    addr;
    u64    len;
    u64    pgoff;
    char   filename[];
};
.in

.TP
.B PERF_RECORD_LOST
This record indicates when events are lost.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 id;
    u64 lost;
};
.fi
.in

.TP
.B PERF_RECORD_COMM
This record indicates a change in the process name.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    char comm[];
};
.fi
.in

.TP
.B PERF_RECORD_EXIT
This record indicates a process exit event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.BR PERF_RECORD_THROTTLE ", " PERF_RECORD_UNTHROTTLE
This record indicates a throttle/unthrottle event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 time;
    u64 id;
    u64 stream_id;
};
.fi
.in

.TP
.B PERF_RECORD_FORK
This record indicates a fork event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.B PERF_RECORD_READ
This record indicates a read event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    struct read_format values;
};
.fi
.in

.TP
.B PERF_RECORD_SAMPLE
This record indicates a sample.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64   ip;          /* if PERF_SAMPLE_IP      */
    u32   pid, tid;    /* if PERF_SAMPLE_TID     */
    u64   time;        /* if PERF_SAMPLE_TIME    */
    u64   addr;        /* if PERF_SAMPLE_ADDR    */
    u64   id;          /* if PERF_SAMPLE_ID      */
    u64   stream_id; /* if PERF_SAMPLE_STREAM_ID */
    u32   cpu, res;    /* if PERF_SAMPLE_CPU     */
    u64   period;      /* if PERF_SAMPLE_PERIOD  */
    struct read_format v; /* if PERF_SAMPLE_READ */
    u64   nr;        /* if PERF_SAMPLE_CALLCHAIN */
    u64   ips[nr];   /* if PERF_SAMPLE_CALLCHAIN */
    u32   size;        /* if PERF_SAMPLE_RAW     */
    char  data[size];  /* if PERF_SAMPLE_RAW     */
    u64   from;   /* if PERF_SAMPLE_BRANCH_STACK */
    u64   to;     /* if PERF_SAMPLE_BRANCH_STACK */
    u64   flags;  /* if PERF_SAMPLE_BRANCH_STACK */
    u64   lbr[nr];/* if PERF_SAMPLE_BRANCH_STACK */
};
.fi
.in

The RAW record data is opaque with respect to the ABI.
The ABI doesn't make any promises with respect to the stability
of its content, it may vary depending
on event, hardware, and kernel version.

.RE


.TP
.I misc
The
.I misc
field is one of the following:
.RS
.TP
.B PERF_RECORD_MISC_CPUMODE_MASK
[To be documented]
.TP
.B PERF_RECORD_MISC_CPUMODE_UNKNOWN
[To be documented]
.TP
.B PERF_RECORD_MISC_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_HYPERVISOR
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_EXACT_IP
This indicates that the content of
.B PERF_SAMPLE_IP
points
to the actual instruction that triggered the event.
See also
.IR perf_event_attr.precise_ip .

.RE

.TP
.I size
This indicates the size of the record.


.SS "Signal Overflow"

Counters can be set to signal when a threshold is crossed.
This is set up using traditional the
.BR poll (2),
.BR select (2),
.BR epoll (2)
and
.BR fcntl (2),
system calls.

Normally, a notification is generated for every page filled, however
one can additionally set
.I perf_event_attr.wakeup_events
to generate one every so many counter overflow events.

.SS "Reading Results"
Once a
.BR perf_event_open()
file descriptor  has been opened, the values
of the events can be read from the file descriptor.
The values that are there are specified by the
.I read_format
field in the attr structure at open time.

If you attempt to read into a buffer that is not big enough to hold the
data, an error is returned
.IR ( ENOSPC ).

Here is the layout of the data returned by a read.

If
.B PERF_FORMAT_GROUP
was specified to allow reading all events in a group at once:

.in +4n
.nf
struct {
    u64 nr;             /* The number of events */
    u64 time_enabled;   /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;   /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    struct {
        u64 value;      /* The value of the event */
        u64 id;         /* if PERF_FORMAT_ID */
    } values[nr];
};
.fi
.in

If
.B PERF_FORMAT_GROUP
was
.I not
specified, then the read values look as following:

.in +4n
.nf
struct {
    u64 value;         /* The value of the event */
    u64 time_enabled;  /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;  /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    u64 id;            /* if PERF_FORMAT_ID */
};
.fi
.in

The values read are described in more detail below.
.RS
.TP
.I nr
The number of events in this file descriptor.
Only available if
.B PERF_FORMAT_GROUP
was specified.

.TP
.IR time_enabled ", " time_running
Total time the event was enabled and running.
Normally these are the same.
If more events are started
than available counter slots on the PMU, then multiplexing
happens and events only run part of the time.
In that case the
.I time_enabled
and
.I time running
values can be used to scale an estimated value for the count.

.TP
.I value
An unsigned 64-bit value containing the counter result.

.TP
.I id
A globally unique value for this particular event, only there if
.B PERF_FORMAT_ID
was specified in read_format.

.RE




.RE

.SS "rdpmc instruction"
Starting with Linux 3.4 on x86, you can use the
.I rdpmc
instruction to get low-latency reads without having to enter the kernel.


.SS "perf_event ioctl calls"
.PP
Various ioctls act on
.BR perf_event_open()
file descriptors
.\" FIXME the arguments for these ioctl() operations need to be described
.TP
.B PERF_EVENT_IOC_ENABLE
Enables an individual counter or counter group.

.TP
.B PERF_EVENT_IOC_DISABLE
Disables an individual counter or counter group.

Enabling or disabling the leader of a group enables or disables the
entire group; that is, while the group leader is disabled, none of the
counters in the group will count.
Enabling or disabling a member of a group other than the leader only
affects that counter; disabling a non-leader
stops that counter from counting but doesn't affect any other counter.

.TP
.B PERF_EVENT_IOC_REFRESH
Non-inherited overflow counters can use this
to enable a counter for 'nr' events, after which it gets disabled again.
.\" FIXME the following needs clarification/confirmation
I think the goal of IOC_REFRESH is not to reload the period but simply to
adjust the number of events before the next notifications.

.TP
.B PERF_EVENT_IOC_RESET
Reset the event count to zero.
This only resets the counts; there is no way to reset the
multiplexing
.I time_enabled
or
.I time_running
values.
When sent to a group leader, only
the leader is reset (child events are not).

.TP
.B PERF_EVENT_IOC_PERIOD
IOC_PERIOD is the command to update the period; it
does not update the current period but instead defers until next.

.TP
.B PERF_EVENT_IOC_SET_OUTPUT
This tells the kernel to report event notifications to the specified
file descriptor rather than the default one.
The file descriptors must all be on the same CPU.

.TP
.BR PERF_EVENT_IOC_SET_FILTER " (Since Linux 2.6.33)"
This adds an ftrace filter to this event.

.SS "Using prctl"
A process can enable or disable all the counter groups that are
attached to it using the
.BR prctl (2)
.B PR_TASK_PERF_EVENTS_ENABLE
and
.B PR_TASK_PERF_EVENTS_DISABLE
operations.
This applies to all counters on the current process, whether created by
this process or by another, and does not affect any counters that this
process has created on other processes.
It only enables or disables
the group leaders, not any other members in the groups.

.SS /proc/sys/kernel/perf_event_paranoid

The
.I /proc/sys/kernel/perf_event_paranoid
file can be set to restrict access to the performance counters.
2
means no measurements allowed,
1
means normal counter access,
0
means you can access CPU-specific data, and
\-1
means no restrictions.

The existence of the
.I perf_event_paranoid
file is the official method for determining if a kernel supports
.BR perf_event_open().

.SH "RETURN VALUE"
.BR perf_event_open ()
returns the new file descriptor, or \-1 if an error occurred
(in which case,
.I errno
is set appropriately).
.SH ERRORS
.TP
.B EINVAL
Returned if the specified event is not available.
.TP
.B ENOSPC
Prior to Linux 3.3, if there was no counter room,
.B ENOSPC
was returned.
Linus did not like this, and this was changed to
.BR EINVAL .
.B ENOSPC
is still returned if you try to read results into
too small a buffer.

.SH VERSION

.BR perf_event_open ()
was introduced in Linux 2.6.31 but was called
.BR perf_counter_open () .
It was renamed in Linux 2.6.32.

.SH CONFORMING TO

This call is specific to Linux
and should not be used in programs intended to be portable.

.SH NOTES
Glibc does not provide a wrapper for this system call; call it using
.BR syscall (2).

The official way of knowing if
.BR perf_event_open()
support is enabled is checking
for the existence of the file
.I /proc/sys/kernel/perf_event_paranoid

.SH BUGS

The
.B F_SETOWN_EX
option to
.IR fcntl (2)
is needed to properly get overflow signals in threads.
This was introduced in Linux 2.6.32.

Prior to Linux 2.6.33 (at least for x86) the kernel did not check
if events could be scheduled together until read time.
The same happens on all known kernels if the NMI watchdog is enabled.
This means to see if a given set of events works you have to
.BR perf_event_open (),
start, then read before you know for sure you
can get valid measurements.

Prior to Linux 2.6.34 event constraints were not enforced by the kernel.
In that case, some events would silently return "0" if the kernel
scheduled them in an improper counter slot.

Prior to Linux 2.6.34 there was a bug when multiplexing where the
wrong results could be returned.

Kernels from Linux 2.6.35 to Linux 2.6.39 can quickly crash the kernel if
"inherit" is enabled and many threads are started.

Prior to Linux 2.6.35,
.B PERF_FORMAT_GROUP
did not work with attached processes.

In older Linux 2.6 versions,
refreshing an event group leader refreshed all siblings,
and refreshing with a parameter of 0 enabled infinite refresh.
This behavior is unsupported and should not be relied on.

There is a bug in the kernel code between
Linux 2.6.36 and Linux 3.0 that ignores the
"watermark" field and acts as if a wakeup_event
was chosen if the union has a
non-zero value in it.

Always double-check your results!  Various generalized events
have had wrong values.
For example, retired branches measured
the wrong thing on AMD machines until Linux 2.6.35.

.SH EXAMPLE
The following is a short example that measures the total
instruction count of a call to printf().
.nf

#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include <asm/unistd.h>

long perf_event_open( struct perf_event_attr *hw_event, pid_t pid,
                      int cpu, int group_fd, unsigned long flags )
{
    int ret;

    ret = syscall( __NR_perf_event_open, hw_event, pid, cpu,
                   group_fd, flags );
    return ret;
}


int
main(int argc, char **argv)
{

    struct perf_event_attr pe;
    long long count;
    int fd;

    memset(&pe, 0, sizeof(struct perf_event_attr));
    pe.type = PERF_TYPE_HARDWARE;
    pe.size = sizeof(struct perf_event_attr);
    pe.config = PERF_COUNT_HW_INSTRUCTIONS;
    pe.disabled = 1;
    pe.exclude_kernel = 1;
    pe.exclude_hv = 1;

    fd = perf_event_open(&pe, 0, \-1, \-1, 0);
    if (fd < 0) {
       fprintf(stderr, "Error opening leader %llx\\n", pe.config);
    }

    ioctl(fd, PERF_EVENT_IOC_RESET, 0);
    ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);

    printf("Measuring instruction count for this printf\\n");

    ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
    read(fd, &count, sizeof(long long));

    printf("Used %lld instructions\\n", count);

    close(fd);
}
.fi

.SH "SEE ALSO"
.BR fcntl (2),
.BR mmap (2),
.BR open (2),
.BR prctl (2),
.BR read (2)

Re: [RFC] perf: proposed perf_event_open() manpage

[Namhyung Kim]

Hi Vince,

Great work!

On Tue, 23 Oct 2012 11:35:13 -0400 (EDT), Vince Weaver wrote:
> Hello
>
> attached is a proposed manpage for the perf_event_open() system call.
>
> I'd appreciate any review or comments, especially for the parts marked
> as FIXME or "[To be documented]"
>
> This system call has a complicated interface and I'm sure I've missed
> or glossed over various important features, so your feedback is needed and 
> appreciated.
>
> The eventual goal is to have this included with the Linux man-pages 
> project.
[snip]
> .BI "int perf_event_open(struct perf_event_attr *" hw_event ,

hw_event?  Looks unusual.. how about 'attr'?


> .BI "                    pid_t " pid ", int " cpu ", int " group_fd ,
> .BI "                    unsigned long " flags  );
> .fi
[snip]
> .SS Arguments
> .P
> The argument
> .I pid
> allows events to be attached to processes in various ways.
> If
> .I pid
> is 0, measurements happen on the current task, if
> .I pid
> is greater than 0, the process indicated by
> .I pid
> is measured, and if
> .I pid
> is less than 0, all processes are counted.

Is that true?  Shouldn't pid be -1?


>
> The
> .I cpu
> argument allows measurements to be specific to a CPU.
> If
> .I cpu
> is greater than or equal to 0,
> measurements are restricted to the specified CPU;
> if
> .I cpu
> is \-1, the events are measured on all CPUs.
> .P
> Note that the combination of
> .IR pid " == \-1"
> and
> .IR cpu " == \-1"
> is not valid.
> .P
> A
> .IR pid " > 0"

s/>/>=/ ?


> and
> .IR cpu " == \-1"
> setting measures per-process and follows that process to whatever CPU the
> process gets scheduled to.
> Per-process events can be created by any user.
> .P
> A
> .IR pid " == \-1"
> and
> .IR cpu " >= 0"
> setting is per-CPU and measures all processes on the specified CPU.
> Per-CPU events need the
> .B CAP_SYS_ADMIN
> capability.

Or value of perf_event_paranoid is less than 1.


> .TP
> .RB "dynamic PMU"
> Since Linux 2.6.39,
> .BR perf_event_open()
> can support multiple PMUs.
> To enable this, a value exported by the kernel can be used in the
> .I type
> field to indicate which PMU to use.
> The value to use can be found in the sysfs filesystem:
> there is a subdirectory per PMU instance under
> .IR /sys/devices .

/sys/bus/event_source/devices will be the right place.


> In each sub-directory there is a
> .I type
> file whose content is an integer that can be used in the
> .I type
> field.
> For instance,
> .I /sys/devices/cpu/type

/sys/bus/event_source/devices/cpu/type


> contains the value for the core CPU PMU, which is usually 4.
> .RE
>
[snip]
> .TP
> .IR sample_period ", " sample_freq
> A "sampling" counter is one that generates an interrupt
> every N events, where N is given by
> .IR sample_period .
> A sampling counter has
> .IR sample_period " > 0."

How about adding this here:

"When an (overflow) interrupt generated, requested data (sample) would
be recorded."

> The
> .I sample_type
> field controls what data is recorded on each interrupt.
>
> .I sample_freq
> can be used if you wish to use frequency rather than period.
> In this case you set the
> .I freq
> flag.
> The kernel will adjust the sampling period
> to try and achieve the desired rate.
> The rate of adjustment is a
> timer tick.

Is that true?  I thought it'd be adjusted whenever overflow occures.


>
>
> .TP
> .I "sample_type"
> The various bits in this field specify which values to include
> in the overflow packets.

I guess the overflow packets here means samples.  It'd be better if we
use a consistent word for specifying a thing.


> They will be recorded in a ring-buffer,
> which is available to user-space using
> .BR mmap (2).
> The order in which the values are saved in the
> overflow packets as documented in the MMAP Layout subsection below;
> it is not the
> .I "enum perf_event_sample_format"
> order.
> .RS
> .TP
> .B PERF_SAMPLE_IP
> instruction pointer
> .TP
> .B PERF_SAMPLE_TID
> thread id
> .TP
> .B PERF_SAMPLE_TIME
> time
> .TP
> .B PERF_SAMPLE_ADDR
> address
> .TP
> .B PERF_SAMPLE_READ
> [To be documented]

It's for an event group to sample leader only.  Values of other members
will be read when an interrupt occurred on the leader.

Jiri is working on it.

> .TP
> .B PERF_SAMPLE_CALLCHAIN
> [To be documented]

callchain (or stack backtrace)

> .TP
> .B PERF_SAMPLE_ID
> [To be documented]

unique(?) id for the opened event.

> .TP
> .B PERF_SAMPLE_CPU
> [To be documented]

cpu number

> .TP
> .B PERF_SAMPLE_PERIOD
> [To be documented]

event count

> .TP
> .B PERF_SAMPLE_STREAM_ID
> [To be documented]
> .TP
> .B PERF_SAMPLE_RAW
> [To be documented]

additional data - usually for tracepoint events

> .TP
> .BR PERF_SAMPLE_BRANCH_STACK " (Since Linux 3.4)"
> [To be documented]

requested branch stack - only supported on intel machines which has LBR
feature(?).  See branch_sample_type.

> .RE
[snip]
> .SS /proc/sys/kernel/perf_event_paranoid
>
> The
> .I /proc/sys/kernel/perf_event_paranoid
> file can be set to restrict access to the performance counters.
> 2
> means no measurements allowed,

This is not true.  It only allows user mode measurements.

$ cat /proc/sys/kernel/perf_event_paranoid 
2

$ perf stat usleep 1
  Error: You may not have permission to collect stats.
	 Consider tweaking /proc/sys/kernel/perf_event_paranoid or running as root.
Not all events could be opened.

$ perf stat -e cycles:u usleep 1

 Performance counter stats for 'usleep 1':

           253,055 cycles:u                  #    0.000 GHz                    

       0.001988538 seconds time elapsed


> 1
> means normal counter access,

This includes kernel mode measurements.

> 0
> means you can access CPU-specific data, and

But cannot access raw tracepoint samples.

> \-1
> means no restrictions.


Thanks,
Namhyung

Re: [RFC] perf: proposed perf_event_open() manpage

[Vince Weaver]

On Wed, 24 Oct 2012, Namhyung Kim wrote:

> > .BI "int perf_event_open(struct perf_event_attr *" hw_event ,
> 
> hw_event?  Looks unusual.. how about 'attr'?

this (and some of the other stuff) is because the manpage used the 
somewhat out of date "tools/perf/design.txt" as a reference.

It looks like the perf tool uses "attr" here, so I'll make that change.

> > is measured, and if
> > .I pid
> > is less than 0, all processes are counted.
> 
> Is that true?  Shouldn't pid be -1?

tools/perf/design.txt claims less than 0, but you're right, in
kernel/events/core.c there are a lot of explicit checks for pid==-1

I'll fix this.

> > Note that the combination of
> > .IR pid " == \-1"
> > and
> > .IR cpu " == \-1"
> > is not valid.
> > .P
> > A
> > .IR pid " > 0"
> 
> s/>/>=/ ?

Again, from tools/perf/design.txt
Is it meaningful to monitor pid 0?
I tried using perf stat to measure pid 0 and it just reports
"Problems finding threads of monitor"

> > Per-CPU events need the
> > .B CAP_SYS_ADMIN
> > capability.
> 
> Or value of perf_event_paranoid is less than 1.

I'll add that.

> > .TP
> > .RB "dynamic PMU"
> > Since Linux 2.6.39,
> > .BR perf_event_open()
> > can support multiple PMUs.
> > To enable this, a value exported by the kernel can be used in the
> > .I type
> > field to indicate which PMU to use.
> > The value to use can be found in the sysfs filesystem:
> > there is a subdirectory per PMU instance under
> > .IR /sys/devices .
> 
> /sys/bus/event_source/devices will be the right place.

I'll update that.

> > In each sub-directory there is a
> > .I type
> > file whose content is an integer that can be used in the
> > .I type
> > field.
> > For instance,
> > .I /sys/devices/cpu/type
> 
> /sys/bus/event_source/devices/cpu/type

Well, the former works too, but I guess the latter is more clear.

> > .TP
> > .IR sample_period ", " sample_freq
> > A "sampling" counter is one that generates an interrupt
> > every N events, where N is given by
> > .IR sample_period .
> > A sampling counter has
> > .IR sample_period " > 0."
> 
> How about adding this here:
> 
> "When an (overflow) interrupt generated, requested data (sample) would
> be recorded."

OK.


> > The kernel will adjust the sampling period
> > to try and achieve the desired rate.
> > The rate of adjustment is a
> > timer tick.
> 
> Is that true?  I thought it'd be adjusted whenever overflow occures.

I was told that during an e-mail discussion I was having once about why 
IOC_REFRESH as used by PAPI gives weird results.  I can't seem to find the 
exact reference though.  It would be nice to have an official 
clarification.

> > .TP
> > .I "sample_type"
> > The various bits in this field specify which values to include
> > in the overflow packets.
> 
> I guess the overflow packets here means samples.  It'd be better if we
> use a consistent word for specifying a thing.

I'll try to make things more consistent.

> > .TP
> > .B PERF_SAMPLE_READ
> > [To be documented]
> 
> It's for an event group to sample leader only.  Values of other members
> will be read when an interrupt occurred on the leader.

I'll add that.

> > .TP
> > .B PERF_SAMPLE_CALLCHAIN
> > [To be documented]
> 
> callchain (or stack backtrace)

are the values stored in the sample buffer for all of these documented 
somewhere?

> > .TP
> > .B PERF_SAMPLE_ID
> > [To be documented]
> 
> unique(?) id for the opened event.

Is this the same ID as that when using PERF_FORMAT_ID?

> > .TP
> > .B PERF_SAMPLE_CPU
> > [To be documented]
> 
> cpu number

OK

> > .TP
> > .B PERF_SAMPLE_PERIOD
> > [To be documented]
> 
> event count

What event count?  The count that caused the sample to happen?

> > .TP
> > .B PERF_SAMPLE_RAW
> > [To be documented]
> 
> additional data - usually for tracepoint events

What type of additional data?

> > .TP
> > .BR PERF_SAMPLE_BRANCH_STACK " (Since Linux 3.4)"
> > [To be documented]
> 
> requested branch stack - only supported on intel machines which has LBR
> feature(?).  See branch_sample_type.

I'll add.
 
> > .RE
> [snip]
> > .SS /proc/sys/kernel/perf_event_paranoid
> >
> > The
> > .I /proc/sys/kernel/perf_event_paranoid
> > file can be set to restrict access to the performance counters.
> > 2
> > means no measurements allowed,
> 
> This is not true.  It only allows user mode measurements.

Interesting.  Is there some way to totally disable perf_events?
It is a security hole, and it's not easy to configure an x86 kernel
w/o perf_event support.

I'll update with expanded descriptions.

In addition, would it be useful to include documentation on the files in
/sys/bus/event_source/devices/
 such as
   type
   format/
   uevent
   rdpmc
or would these get documented elsewhere?

Thanks for the valuable feedback!

Vince Weaver
vincent.weaver-e7X0jjDqjFGHXe+LvDLADg@public.gmane.org
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC] perf: proposed perf_event_open() manpage

[Vince Weaver]

below is an updated version of the manpage.  It has the changes you 
suggested plus a few other additions.

Please verify the "sample_type" changes.  I did some digging in the
kernel source and added some info to some of the descriptions.

I also added
  /proc/sys/kernel/
and
  /sys/bus/event_source/devices/
subsections near the end to document the files pertaining to perf_event.

Thanks

Vince Weaver
vincent.weaver-e7X0jjDqjFGHXe+LvDLADg@public.gmane.org

.\" Hey Emacs! This file is -*- nroff -*- source.
.\"
.\" Copyright (c) 2012, Vincent Weaver
.\"
.\" This is free documentation; you can redistribute it and/or
.\" modify it under the terms of the GNU General Public License as
.\" published by the Free Software Foundation; either version 2 of
.\" the License, or (at your option) any later version.
.\"
.\" The GNU General Public License's references to "object code"
.\" and "executables" are to be interpreted as the output of any
.\" document formatting or typesetting system, including
.\" intermediate and printed output.
.\"
.\" This manual is distributed in the hope that it will be useful,
.\" but WITHOUT ANY WARRANTY; without even the implied warranty of
.\" MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
.\" GNU General Public License for more details.
.\"
.\" You should have received a copy of the GNU General Public
.\" License along with this manual; if not, see
.\" <http://www.gnu.org/licenses/>.
.\"
.\" This document is based on the perf_event.h header file, the
.\" tools/perf/design.txt file, and a lot of bitter experience.
.\"
.TH PERF_EVENT_OPEN 2 2012-10-24 "Linux" "Linux Programmer's Manual"
.SH NAME
perf_event_open \- set up performance monitoring
.SH SYNOPSIS
.nf
.B #include <linux/perf_event.h>
.B #include <linux/hw_breakpoint.h>
.sp
.BI "int perf_event_open(struct perf_event_attr *" attr ,
.BI "                    pid_t " pid ", int " cpu ", int " group_fd ,
.BI "                    unsigned long " flags  );
.fi

.IR Note :
There is no glibc wrapper for this system call; see NOTES.
.SH DESCRIPTION
Given a list of parameters,
.BR perf_event_open ()
returns a file descriptor, for use in subsequent system calls
.RB ( read "(2), " mmap "(2), " prctl "(2), " fcntl "(2), etc.)."
.PP
A call to
.BR perf_event_open ()
creates a file descriptor that allows measuring performance
information.
Each file descriptor corresponds to one
event that is measured; these can be grouped together
to measure multiple events simultaneously.
.PP
Events can be enabled and disabled in two ways: via
.BR ioctl (2)
and via
.BR prctl (2) .
When an event is disabled it does not count or generate overflows but does
continue to exist and maintain its count value.
Events come in two flavors: counting and sampled.
A
.I counting
event is one that is used for counting the aggregate number of events
that occur.
In general, counting event results are gathered with a
.BR read (2)
call.
A
.I sampling
event periodically writes measurements to a buffer that can then
be accessed via
.BR  mmap (2) .
.SS Arguments
.P
The argument
.I pid
allows events to be attached to processes in various ways.
If
.I pid
is 0, measurements happen on the current task, if
.I pid
is greater than 0, the process indicated by
.I pid
is measured, and if
.I pid
is \-1, all processes are counted.

The
.I cpu
argument allows measurements to be specific to a CPU.
If
.I cpu
is greater than or equal to 0,
measurements are restricted to the specified CPU;
if
.I cpu
is \-1, the events are measured on all CPUs.
.P
Note that the combination of
.IR pid " == \-1"
and
.IR cpu " == \-1"
is not valid.
.P
A
.IR pid " > 0"
and
.IR cpu " == \-1"
setting measures per-process and follows that process to whatever CPU the
process gets scheduled to.
Per-process events can be created by any user.
.P
A
.IR pid " == \-1"
and
.IR cpu " >= 0"
setting is per-CPU and measures all processes on the specified CPU.
Per-CPU events need the
.B CAP_SYS_ADMIN
capability or a 
.I /proc/sys/kernel/perf_event_paranoid 
value of less than 1.
.P
The
.I group_fd
argument allows counter groups to be set up.
A counter group has one counter which is the group leader.
The leader is created first, with
.IR group_fd " = \-1"
in the
.BR perf_event_open ()
call that creates it.
The rest of the group members are created subsequently, with
.IR group_fd
giving the fd of the group leader.
(A single counter on its own is created with
.IR group_fd " = \-1"
and is considered to be a group with only 1 member.)
.P
A counter group is scheduled onto the CPU as a unit: it will only
be put onto the CPU if all of the counters in the group can be put onto
the CPU.
This means that the values of the member counters can be
meaningfully compared, added, divided (to get ratios), etc., with each
other, since they have counted events for the same set of executed
instructions.
.P
The
.I flags
argument takes one of the following values:
.TP
.BR PERF_FLAG_FD_NO_GROUP
.\" FIXME The following sentence is unclear
This flag allows creating an event as part of an event group but
having no group leader.
It is unclear why this is useful.
.\" FIXME So, why is it useful?
.TP
.BR PERF_FLAG_FD_OUTPUT
This flag re-routes the output from an event to the group leader.
.TP
.BR PERF_FLAG_PID_CGROUP " (Since Linux 2.6.39)."
This flag activates per-container system-wide monitoring.
A container
is an abstraction that isolates a set of resources for finer grain
control (CPUs, memory, etc...).
In this mode, the event is measured
only if the thread running on the monitored CPU belongs to the designated
container (cgroup).
The cgroup is identified by passing a file descriptor
opened on its directory in the cgroupfs filesystem.
For instance, if the
cgroup to monitor is called
.IR test ,
then a file descriptor opened on
.I /dev/cgroup/test
(assuming cgroupfs is mounted on
.IR /dev/cgroup )
must be passed as the
.I pid
parameter.
cgroup monitoring is only available
for system-wide events and may therefore require extra permissions.
.P
The
.I perf_event_attr
structure is what is passed into the
.BR perf_event_open ()
syscall.
It is large and has a complicated set of dependent fields.

.in +4n
.nf
struct perf_event_attr {
    __u32     type;         /* Type of event */
    __u32     size;         /* Size of attribute structure */
    __u64     config;       /* Type-specific configuration */

    union {
        __u64 sample_period;    /* Period of sampling */
        __u64 sample_freq;      /* Frequency of sampling */
    };

    __u64     sample_type;  /* Specifies values included in sample */
    __u64     read_format;  /* Specifies values returned in read */

    __u64     disabled       : 1,   /* off by default         */
              inherit        : 1,   /* children inherit it    */
              pinned         : 1,   /* must always be on PMU  */
              exclusive      : 1,   /* only group on PMU      */
              exclude_user   : 1,   /* don't count user       */
              exclude_kernel : 1,   /* don't count kernel     */
              exclude_hv     : 1,   /* don't count hypervisor */
              exclude_idle   : 1,   /* don't count when idle  */
              mmap           : 1,   /* include mmap data      */
              comm           : 1,   /* include comm data      */
              freq           : 1,   /* use freq, not period   */
              inherit_stat   : 1,   /* per task counts        */
              enable_on_exec : 1,   /* next exec enables      */
              task           : 1,   /* trace fork/exit        */
              watermark      : 1,   /* wakeup_watermark       */
              precise_ip     : 2,   /* skid constraint        */
              mmap_data      : 1,   /* non-exec mmap data     */
              sample_id_all  : 1,   /* sample_type all events */
              exclude_host   : 1,   /* don't count in host    */
              exclude_guest  : 1,   /* don't count in guest   */
              __reserved_1   : 43;

    union {
        __u32 wakeup_events;    /* wakeup every n events */
        __u32 wakeup_watermark; /* bytes before wakeup   */
    };

    __u32     bp_type;          /* breakpoint type */

    union {
        __u64 bp_addr;          /* breakpoint address */
        __u64 config1;          /* extension of config */
    };

    union {
        __u64 bp_len;           /* breakpoint length */
        __u64 config2;          /* extension of config1 */
    };
    __u64     branch_sample_type;   /* enum branch_sample_type */
};
.fi
.in

The fields of the
.I perf_event_attr
structure are described in more detail below.

.TP
.I type
This field specifies the overall event type.
It has one of the following values:
.RS
.TP
.B PERF_TYPE_HARDWARE
This indicates one of the "generalized" hardware events provided
by the kernel.
See the
.I config
field definition for more details.
.TP
.B PERF_TYPE_SOFTWARE
This indicates one of the software-defined events provided by the kernel
(even if no hardware support is available).
.TP
.B PERF_TYPE_TRACEPOINT
This indicates a tracepoint
provided by the kernel tracepoint infrastructure.
.TP
.B PERF_TYPE_HW_CACHE
This indicates a hardware cache event.
This has a special encoding, described in the
.I config
field definition.
.TP
.B PERF_TYPE_RAW
This indicates a "raw" implementation-specific event in the
.IR config " field."
.TP
.BR PERF_TYPE_BREAKPOINT " (Since Linux 2.6.33)"
This indicates a hardware breakpoint as provided by the CPU.
Breakpoints can be read/write accesses to an address as well as
execution of an instruction address.
.TP
.RB "dynamic PMU"
Since Linux 2.6.39,
.BR perf_event_open()
can support multiple PMUs.
To enable this, a value exported by the kernel can be used in the
.I type
field to indicate which PMU to use.
The value to use can be found in the sysfs filesystem:
there is a subdirectory per PMU instance under
.IR /sys/bus/event_source/devices .
In each sub-directory there is a
.I type
file whose content is an integer that can be used in the
.I type
field.
For instance,
.I /sys/bus/event_source/devices/cpu/type
contains the value for the core CPU PMU, which is usually 4.
.RE

.TP
.I "size"
The size of the
.I perf_event_attr
structure for forward/backward compatibility.
Set this using
.I sizeof(struct perf_event_attr)
to allow the kernel to see
the struct size at the time of compilation.

The related define
.B PERF_ATTR_SIZE_VER0
is set to 64; this was the size of the first published struct.
.B PERF_ATTR_SIZE_VER1
is 72, corresponding to the addition of breakpoints in Linux 2.6.33.
.B PERF_ATTR_SIZE_VER2
is 80 corresponding to the addition of branch sampling in Linux 3.4.

.TP
.I "config"
This specifies which event you want, in conjunction with
the
.I type
field.
The
.IR config1 " and " config2
fields are also taken into account in cases where 64 bits is not
enough to fully specify the event.
The encoding of these fields are event dependent.

The most significant bit (bit 63) of
.I config
signifies CPU-specific (raw) counter configuration data;
if the most significant bit is unset, the next 7 bits are an event
type and the rest of the bits are the event identifier.

There are various ways to set the
.I config
field that are dependent on the value of the previously
described
.I type
field.
What follows are various possible settings for
.I config
separated out by
.IR type .

If
.I type
is
.BR PERF_TYPE_HARDWARE ,
we are measuring one of the generalized hardware CPU events.
Not all of these are available on all platforms.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_HW_CPU_CYCLES
Total cycles.
Be wary of what happens during CPU frequency scaling
.TP
.B PERF_COUNT_HW_INSTRUCTIONS
Retired instructions.
Be careful, these can be affected by various
issues, most notably hardware interrupt counts
.TP
.B PERF_COUNT_HW_CACHE_REFERENCES
Cache accesses.
Usually this indicates Last Level Cache accesses but this may
vary depending on your CPU.
This may include prefetches and coherency messages; again this
depends on the design of your CPU.
.TP
.B PERF_COUNT_HW_CACHE_MISSES
Cache misses.
Usually this indicates Last Level Cache misses; this is intended to be
used in conjunction with the
.B PERF_COUNT_HW_CACHE_REFERENCES
event to calculate cache miss rates.
.TP
.B PERF_COUNT_HW_BRANCH_INSTRUCTIONS
Retired branch instructions.
Prior to Linux 2.6.34, this used
the wrong event on AMD processors.
.TP
.B PERF_COUNT_HW_BRANCH_MISSES
Mispredicted branch instructions.
.TP
.B PERF_COUNT_HW_BUS_CYCLES
Bus cycles, which can be different from total cycles.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_FRONTEND " (Since Linux 3.0)"
Stalled cycles during issue.
.TP
.BR PERF_COUNT_HW_STALLED_CYCLES_BACKEND  " (Since Linux 3.0)"
Stalled cycles during retirement.
.TP
.BR PERF_COUNT_HW_REF_CPU_CYCLES  " (Since Linux 3.3)"
Total cycles; not affected by CPU frequency scaling.
.RE
.IP
If
.I type
is
.BR PERF_TYPE_SOFTWARE ,
we are measuring software events provided by the kernel.
Set
.I config
to one of the following:
.RS 12
.TP
.B PERF_COUNT_SW_CPU_CLOCK
This reports the CPU clock, a high-resolution per-CPU timer.
.TP
.B PERF_COUNT_SW_TASK_CLOCK
This reports a clock count specific to the task that is running.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS
This reports the number of page faults.
.TP
.B PERF_COUNT_SW_CONTEXT_SWITCHES
This counts context switches.
Until Linux 2.6.34, these were all reported as user-space
events, after that they are reported as happening in the kernel.
.TP
.B PERF_COUNT_SW_CPU_MIGRATIONS
This reports the number of times the process
has migrated to a new CPU.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MIN
This counts the number of minor page faults.
These did not require disk I/O to handle.
.TP
.B PERF_COUNT_SW_PAGE_FAULTS_MAJ
This counts the number of major page faults.
These required disk I/O to handle.
.TP
.BR PERF_COUNT_SW_ALIGNMENT_FAULTS " (Since Linux 2.6.33)"
This counts the number of alignment faults.
These happen when unaligned memory accesses happen; the kernel
can handle these but it reduces performance.
This only happens on some architectures (never on x86).
.TP
.BR PERF_COUNT_SW_EMULATION_FAULTS " (Since Linux 2.6.33)"
This counts the number of emulation faults.
The kernel sometimes traps on unimplemented instructions
and emulates them for userspace.
This can negatively impact performance.
.RE
.RE


.RS
If
.I type
is
.BR PERF_TYPE_TRACEPOINT ,
then we are measuring kernel tracepoints.
The value to use in
.I config
can be obtained from under debugfs
.I tracing/events/*/*/id
if ftrace is enabled in the kernel.

.RE

.RS
If
.I type
is
.BR PERF_TYPE_HW_CACHE ,
then we are measuring a hardware CPU cache event.
To calculate the appropriate
.I config
value use the following equation:
.RS 4
.nf

    (perf_hw_cache_id) | (perf_hw_cache_op_id << 8) |
    (perf_hw_cache_op_result_id << 16)
.fi
.P
where
.I perf_hw_cache_id
is one of:
.RS
.TP
.B PERF_COUNT_HW_CACHE_L1D
for measuring Level 1 Data Cache
.TP
.B PERF_COUNT_HW_CACHE_L1I
for measuring Level 1 Instruction Cache
.TP
.B PERF_COUNT_HW_CACHE_LL
for measuring Last-Level Cache
.TP
.B PERF_COUNT_HW_CACHE_DTLB
for measuring the Data TLB
.TP
.B PERF_COUNT_HW_CACHE_ITLB
for measuring the Instruction TLB
.TP
.B PERF_COUNT_HW_CACHE_BPU
for measuring the branch prediction unit
.TP
.BR PERF_COUNT_HW_CACHE_NODE " (Since Linux 3.0)"
for measuring local memory accesses
.RE

.P
and
.I perf_hw_cache_op_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_OP_READ
for read accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_WRITE
for write accesses
.TP
.B PERF_COUNT_HW_CACHE_OP_PREFETCH
for prefetch accesses
.RE

.P
and
.I perf_hw_cache_op_result_id
is one of
.RS
.TP
.B PERF_COUNT_HW_CACHE_RESULT_ACCESS
to measure accesses
.TP
.B PERF_COUNT_HW_CACHE_RESULT_MISS
to measure misses
.RE
.RE

If
.I type
is
.BR PERF_TYPE_RAW ,
then a custom "raw"
.I config
value is needed.
Most CPUs support events that are not covered by the "generalized" events.
These are implementation defined; see your CPU manual (for example
the Intel Volume 3B documentation or the AMD BIOS and Kernel Developer
Guide).
The libpfm4 library can be used to translate from the name in the
architectural manuals to the raw hex value
.BR perf_event_open ()
expects in this field.

If
.I type
is
.BR PERF_TYPE_BREAKPOINT ,
then leave
.I config
set to zero.
Its parameters are set in other places.
.RE
.TP
.IR sample_period ", " sample_freq
A "sampling" counter is one that generates an interrupt
every N events, where N is given by
.IR sample_period .
A sampling counter has
.IR sample_period " > 0."
When an overflow interrupt occurs, requested data is recorded
in the mmap buffer.
The
.I sample_type
field controls what data is recorded on each interrupt.

.I sample_freq
can be used if you wish to use frequency rather than period.
In this case you set the
.I freq
flag.
The kernel will adjust the sampling period
to try and achieve the desired rate.
The rate of adjustment is a
timer tick.


.TP
.I "sample_type"
The various bits in this field specify which values to include
in the sample.
They will be recorded in a ring-buffer,
which is available to user-space using
.BR mmap (2).
The order in which the values are saved in the
sample are documented in the MMAP Layout subsection below;
it is not the
.I "enum perf_event_sample_format"
order.
.RS
.TP
.B PERF_SAMPLE_IP
instruction pointer
.TP
.B PERF_SAMPLE_TID
thread id
.TP
.B PERF_SAMPLE_TIME
time
.TP
.B PERF_SAMPLE_ADDR
address
.TP
.B PERF_SAMPLE_READ
record values for all events in a group, not just the group leader
.TP
.B PERF_SAMPLE_CALLCHAIN
callchain (stack backtrace)
.TP
.B PERF_SAMPLE_ID
a unique ID for the opened event.
If the event is a member of an event group, the group leader
ID is returned.
This ID is the same as the one returned by PERF_FORMAT_ID.
.TP
.B PERF_SAMPLE_CPU
cpu number
.TP
.B PERF_SAMPLE_PERIOD
current sampling period
.TP
.B PERF_SAMPLE_STREAM_ID
a unique ID for the opened event.
Unlike
.B PERF_SAMPLE_ID
the actual ID is returned, not the group leader.
This ID is the same as the one returned by PERF_FORMAT_ID.
.TP
.B PERF_SAMPLE_RAW
additional data.  Usually returned by tracepoint events.
.TP
.BR PERF_SAMPLE_BRANCH_STACK " (Since Linux 3.4)"
requested branch stack.  See branch_sample_type.
.RE

.TP
.IR "read_format"
This field specifies the format of the data returned by
.BR read (2)
on a
.BR perf_event_open()
file descriptor.
.RS
.TP
.B PERF_FORMAT_TOTAL_TIME_ENABLED
Adds the 64-bit "time_enabled" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and multiplexing is happening.
.TP
.B PERF_FORMAT_TOTAL_TIME_RUNNING
Adds the 64-bit "time_running" field.
This can be used to calculate estimated totals if
the PMU is overcommitted and  multiplexing is happening.
.TP
.B PERF_FORMAT_ID
Adds a 64-bit unique value that corresponds to the event-group.
.TP
.B PERF_FORMAT_GROUP
Allows all counter values in an event-group to be read with one read.
.RE

.TP
.IR "disabled"
The
.I disabled
bit specifies whether the counter starts out disabled or enabled.
If disabled, the event can later be enabled by
.BR ioctl (2),
.BR prctl (2),
or
.IR enable_on_exec .

.TP
.IR "inherit"
The
.I inherit
bit specifies that this counter should count events of child
tasks as well as the task specified.
This only applies to new children, not to any existing children at
the time the counter is created (nor to any new children of
existing children).

Inherit does not work for some combinations of
.IR read_format s,
such as
.BR PERF_FORMAT_GROUP .

.TP
.IR "pinned"
The
.I pinned
bit specifies that the counter should always be on the CPU if at all
possible.
It only applies to hardware counters and only to group leaders.
If a pinned counter cannot be put onto the CPU (e.g., because there are
not enough hardware counters or because of a conflict with some other
event), then the counter goes into an 'error' state, where reads
return end-of-file (i.e.,
.BR read (2)
returns 0) until the counter is subsequently enabled or disabled.

.TP
.IR "exclusive"
The
.I exclusive
bit specifies that when this counter's group is on the CPU,
it should be the only group using the CPU's counters.
In the future this may allow monitoring programs to
support PMU features that need to run alone so that they do not
disrupt other hardware counters.

.TP
.IR "exclude_user"
If this bit is set, the count excludes events that happen in user-space.

.TP
.IR "exclude_kernel"
If this bit is set, the count excludes events that happen in kernel-space.

.TP
.IR "exclude_hv"
If this bit is set, the count excludes events that happen in the
hypervisor.
This is mainly for PMUs that have built-in support for handling this
(such as POWER).
Extra support is needed for handling hypervisor measurements on most
machines.

.TP
.IR "exclude_idle"
If set, don't count when the CPU is idle.

.TP
.IR "mmap"
The
.I mmap
bit enables recording of extra information to a
mmap'd ring-buffer.
This is
described below in subsection MMAP Layout.

.TP
.IR "comm"
The
.I comm
bit enables tracking of process command name as modified by the
.IR exec (2)
and
.IR prctl (PR_SET_NAME)
system calls.
Unfortunately for tools,
there is no way to distinguish one system call versus the other.

.TP
.IR "freq"
If this bit is set, then
.I sample_frequency
not
.I sample_period
is used when setting up the sampling interval.

.TP
.IR "inherit_stat"
This bit enables saving of event counts on context switch for
inherited tasks.
This is only meaningful if the
.I inherit
field is set.

.TP
.IR "enable_on_exec"
If this bit is set, a counter is automatically
enabled after a call to
.BR exec (2).

.TP
.IR "task"
If this bit is set, then
fork/exit notifications are included in the ring buffer.

.TP
.IR "watermark"
If set, have a sampling interrupt happen when we cross the
wakeup_watermark boundary.

.TP
.IR "precise_ip" " (Since Linux 2.6.35)"
This controls the amount of skid.
Skid is how many instructions
execute between an event of interest happening and the kernel
being able to stop and record the event.
Smaller skid is
better and allows more accurate reporting of which events
correspond to which instructions, but hardware is often limited
with how small this can be.

The values of this are the following:
.RS
.TP
0 -
.B SAMPLE_IP
can have arbitrary skid
.TP
1 -
.B SAMPLE_IP
must have constant skid
.TP
2 -
.B SAMPLE_IP
requested to have 0 skid
.TP
3 -
.B SAMPLE_IP
must have 0 skid.
See also
.BR PERF_RECORD_MISC_EXACT_IP .
.RE

.TP
.IR "mmap_data" " (Since Linux 2.6.36)"
Include mmap events in the ring_buffer.

.TP
.IR "sample_id_all" " (Since Linux 2.6.38)"
If set, then all sample ID info (TID, TIME, ID, CPU, STREAM_ID)
will be provided.

.TP
.IR "exclude_host" " (Since Linux 3.2)"
Do not measure time spent in VM host

.TP
.IR "exclude_guest" " (Since Linux 3.2)"
Do not measure time spent in VM guest


.TP
.IR "wakeup_events" ", " "wakeup_watermark"
This union sets how many events
.RI ( wakeup_events )
or bytes
.RI ( wakeup_watermark )
happen before an overflow signal happens.
Which one is used is selected by the
.I watermark
bitflag.

.TP
.IR "bp_type" " (Since Linux 2.6.33)"
This chooses the breakpoint type.
It is one of:
.RS
.TP
.BR HW_BREAKPOINT_EMPTY
no breakpoint
.TP
.BR HW_BREAKPOINT_R
count when we read the memory location
.TP
.BR HW_BREAKPOINT_W
count when we write the memory location
.TP
.BR HW_BREAKPOINT_RW
count when we read or write the memory location
.TP
.BR HW_BREAKPOINT_X
count when we execute code at the memory location

.LP
The values can be combined via a bitwsie or, but the
combination of
.B HW_BREAKPOINT_R
or
.B HW_BREAKPOINT_W
with
.B HW_BREAKPOINT_X
is not allowed.
.RE

.TP
.IR "bp_addr" " (Since Linux 2.6.33)"
.I bp_addr
address of the breakpoint.
For execution breakpoints this is the memory address of the instruction
of interest; for read and write breakpoints it is the memory address
of the memory location of interest.

.TP
.IR "config1" " (Since Linux 2.6.39)"
.I config1
is used for setting events that need an extra register or otherwise
do not fit in the regular config field.
Raw OFFCORE_EVENTS on Nehalem/Westmere/SandyBridge use this field
on 3.3 and later kernels.

.TP
.IR "bp_len" " (Since Linux 2.6.33)"
.I bp_len
is the length of the breakpoint being measured if
.I type
is
.BR PERF_TYPE_BREAKPOINT .
Options are
.BR HW_BREAKPOINT_LEN_1 ,
.BR HW_BREAKPOINT_LEN_2 ,
.BR HW_BREAKPOINT_LEN_4 ,
.BR HW_BREAKPOINT_LEN_8 .
For an execution breakpoint, set this to
.IR sizeof(long) .

.TP
.IR "config2" " (Since Linux 2.6.39)"

.I config2
is a further extension of the
.I config1
field.

.TP
.IR "branch_sample_type" " (Since Linux 3.4)"
This is used with the CPUs hardware branch sampling, if available.
It can have one of the following values:
.RS
.TP
.B PERF_SAMPLE_BRANCH_USER
Branch target is in user space
.TP
.B PERF_SAMPLE_BRANCH_KERNEL
Branch target is in kernel space
.TP
.B PERF_SAMPLE_BRANCH_HV
Branch target is in hypervisor
.TP
.B PERF_SAMPLE_BRANCH_ANY
Any branch type.
.TP
.B PERF_SAMPLE_BRANCH_ANY_CALL
Any call branch
.TP
.B PERF_SAMPLE_BRANCH_ANY_RETURN
Any return branch
.TP
.BR PERF_SAMPLE_BRANCH_IND_CALL
Indirect calls
.TP
.BR PERF_SAMPLE_BRANCH_PLM_ALL
User, kernel, and hv
.RE


.SS "MMAP Layout"

When using
.BR perf_event_open()
in sampled mode, asynchronous events
(like counter overflow or
.B PROT_EXEC
mmap tracking)
are logged into a ring-buffer.
This ring-buffer is created and accessed through
.BR mmap (2).

The mmap size should be 1+2^n pages, where the first page is a
metadata page
.IR ( "struct perf_event_mmap_page" )
that contains various
bits of information such as where the ring-buffer head is.

Before kernel 2.6.39, there is a bug that means you must allocate a mmap
ring buffer when sampling even if you do not plan to access it.

The structure of the first metadata mmap page is as follows:

.in +4n
.nf
struct perf_event_mmap_page {
    __u32 version;          /* version number of this structure   */
    __u32 compat_version;   /* lowest version this is compat with */
    __u32 lock;             /* seqlock for synchronization        */
    __u32 index;            /* hardware counter identifier        */
    __s64 offset;           /* add to hardware counter value      */
    __u64 time_enabled;     /* time event active                  */
    __u64 time_running;     /* time event on CPU                  */
    union {
        __u64   capabilities;
        __u64   cap_usr_time  : 1,
                cap_usr_rdpmc : 1,
    };
    __u16   pmc_width;
    __u16   time_shift;
    __u32   time_mult;
    __u64   time_offset;
    __u64   __reserved[120];   /* Pad to 1k                */
    __u64   data_head;         /* head in the data section */
    __u64   data_tail;         /* user-space written tail  */
}
.fi
.in



The following looks at the fields in the
.I perf_event_mmap_page
structure in more detail.

.RS

.TP
.I version
Version number of this structure.

.TP
.I compat_version
The lowest version this is compatible with.

.TP
.I lock
A seqlock for synchronization.

.TP
.I index;
A unique hardware counter identifier.

.TP
.I offset
.\" FIXME clarify
Add this to hardware counter value??

.TP
.I time_enabled
Time the event was active.

.TP
.I time_running
Time the event was running.

.TP
.I cap_usr_time
User time capability

.TP
.I cap_usr_rdpmc
If the hardware supports user-space read of performance counters
without syscall (this is the "rdpmc" instruction on x86), then
the following code can be used to do a read:

.in +4n
.nf
u32 seq, time_mult, time_shift, idx, width;
u64 count, enabled, running;
u64 cyc, time_offset;
s64 pmc = 0;

do {
    seq = pc\->lock;
    barrier();
    enabled = pc\->time_enabled;
    running = pc\->time_running;

    if (pc\->cap_usr_time && enabled != running) {
        cyc = rdtsc();
        time_offset = pc\->time_offset;
        time_mult   = pc\->time_mult;
        time_shift  = pc\->time_shift;
    }

    idx = pc\->index;
    count = pc\->offset;

    if (pc\->cap_usr_rdpmc && idx) {
        width = pc\->pmc_width;
        pmc = rdpmc(idx \- 1);
    }

    barrier();
} while (pc\->lock != seq);
.fi
.in



.TP
.I pmc_width
If
.IR cap_usr_rdpmc ,
this field provides the bit-width of the value
read using the rdpmc or equivalent instruction.
This can be used to sign extend the result like:

.in +4n
.nf
pmc <<= 64 \- pmc_width;
pmc >>= 64 \- pmc_width; // signed shift right
count += pmc;
.fi
.in


.TP
.IR time_shift ", " time_mult ", " time_offset

If
.IR cap_usr_time ,
these fields can be used to compute the time
delta since time_enabled (in ns) using rdtsc or similar.
.nf

    u64 quot, rem;
    u64 delta;
    quot = (cyc >> time_shift);
    rem = cyc & ((1 << time_shift) \- 1);
    delta = time_offset + quot * time_mult +
            ((rem * time_mult) >> time_shift);
.fi

Where time_offset,time_mult,time_shift and cyc are read in the
seqcount loop described above.
This delta can then be added to
enabled and possible running (if idx), improving the scaling:
.nf

    enabled += delta;
    if (idx)
        running += delta;
    quot = count / running;
    rem  = count % running;
    count = quot * enabled + (rem * enabled) / running;
.fi

.TP
.I data_head
This points to the head of the data section.
On SMP-capable platforms, after reading the data_head value,
user-space should issue an rmb().

.TP
.I data_tail;
When the mapping is
.BR PROT_WRITE ,
the data_tail value should be written by
userspace to reflect the last read data.
In this case the kernel will not over-write unread data.

.RE


The following 2^n ring-buffer pages have the layout described below.

If
.I perf_event_attr.sample_id_all
is set, then all event types will
have the sample_type selected fields related to where/when (identity)
an event took place (TID, TIME, ID, CPU, STREAM_ID) described in
.B PERF_RECORD_SAMPLE
below, it will be stashed just after the
perf_event_header and the fields already present for the existing
fields, i.e., at the end of the payload.
That way a newer perf.data
file will be supported by older perf tools, with these new optional
fields being ignored.

The mmap values start with a header:

.in +4n
.nf
struct perf_event_header {
    __u32   type;
    __u16   misc;
    __u16   size;
};
.fi
.in

Below, we describe the
.I perf_event_header
fields in more detail.

.TP
.I type
The
.I type
value is one of the below.
The values in the corresponding record (that follows the header)
depend on the
.I type
selected as shown.

.RS
.TP
.B PERF_RECORD_MMAP
The MMAP events record the
.B PROT_EXEC
mappings so that we can correlate
userspace IPs to code.
They have the following structure:

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32    pid, tid;
    u64    addr;
    u64    len;
    u64    pgoff;
    char   filename[];
};
.in

.TP
.B PERF_RECORD_LOST
This record indicates when events are lost.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 id;
    u64 lost;
};
.fi
.in

.TP
.B PERF_RECORD_COMM
This record indicates a change in the process name.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    char comm[];
};
.fi
.in

.TP
.B PERF_RECORD_EXIT
This record indicates a process exit event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.BR PERF_RECORD_THROTTLE ", " PERF_RECORD_UNTHROTTLE
This record indicates a throttle/unthrottle event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64 time;
    u64 id;
    u64 stream_id;
};
.fi
.in

.TP
.B PERF_RECORD_FORK
This record indicates a fork event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, ppid;
    u32 tid, ptid;
    u64 time;
};
.fi
.in

.TP
.B PERF_RECORD_READ
This record indicates a read event.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u32 pid, tid;
    struct read_format values;
};
.fi
.in

.TP
.B PERF_RECORD_SAMPLE
This record indicates a sample.

.in +4n
.nf
struct {
    struct perf_event_header header;
    u64   ip;          /* if PERF_SAMPLE_IP      */
    u32   pid, tid;    /* if PERF_SAMPLE_TID     */
    u64   time;        /* if PERF_SAMPLE_TIME    */
    u64   addr;        /* if PERF_SAMPLE_ADDR    */
    u64   id;          /* if PERF_SAMPLE_ID      */
    u64   stream_id; /* if PERF_SAMPLE_STREAM_ID */
    u32   cpu, res;    /* if PERF_SAMPLE_CPU     */
    u64   period;      /* if PERF_SAMPLE_PERIOD  */
    struct read_format v; /* if PERF_SAMPLE_READ */
    u64   nr;        /* if PERF_SAMPLE_CALLCHAIN */
    u64   ips[nr];   /* if PERF_SAMPLE_CALLCHAIN */
    u32   size;        /* if PERF_SAMPLE_RAW     */
    char  data[size];  /* if PERF_SAMPLE_RAW     */
    u64   from;   /* if PERF_SAMPLE_BRANCH_STACK */
    u64   to;     /* if PERF_SAMPLE_BRANCH_STACK */
    u64   flags;  /* if PERF_SAMPLE_BRANCH_STACK */
    u64   lbr[nr];/* if PERF_SAMPLE_BRANCH_STACK */
};
.fi
.in

The RAW record data is opaque with respect to the ABI.
The ABI doesn't make any promises with respect to the stability
of its content, it may vary depending
on event, hardware, and kernel version.

.RE


.TP
.I misc
The
.I misc
field is one of the following:
.RS
.TP
.B PERF_RECORD_MISC_CPUMODE_MASK
[To be documented]
.TP
.B PERF_RECORD_MISC_CPUMODE_UNKNOWN
[To be documented]
.TP
.B PERF_RECORD_MISC_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_HYPERVISOR
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_KERNEL
[To be documented]
.TP
.B PERF_RECORD_MISC_GUEST_USER
[To be documented]
.TP
.B PERF_RECORD_MISC_EXACT_IP
This indicates that the content of
.B PERF_SAMPLE_IP
points
to the actual instruction that triggered the event.
See also
.IR perf_event_attr.precise_ip .

.RE

.TP
.I size
This indicates the size of the record.


.SS "Signal Overflow"

Counters can be set to signal when a threshold is crossed.
This is set up using traditional the
.BR poll (2),
.BR select (2),
.BR epoll (2)
and
.BR fcntl (2),
system calls.

Normally, a notification is generated for every page filled, however
one can additionally set
.I perf_event_attr.wakeup_events
to generate one every so many counter overflow events.

.SS "Reading Results"
Once a
.BR perf_event_open()
file descriptor  has been opened, the values
of the events can be read from the file descriptor.
The values that are there are specified by the
.I read_format
field in the attr structure at open time.

If you attempt to read into a buffer that is not big enough to hold the
data, an error is returned
.IR ( ENOSPC ).

Here is the layout of the data returned by a read.

If
.B PERF_FORMAT_GROUP
was specified to allow reading all events in a group at once:

.in +4n
.nf
struct {
    u64 nr;             /* The number of events */
    u64 time_enabled;   /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;   /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    struct {
        u64 value;      /* The value of the event */
        u64 id;         /* if PERF_FORMAT_ID */
    } values[nr];
};
.fi
.in

If
.B PERF_FORMAT_GROUP
was
.I not
specified, then the read values look as following:

.in +4n
.nf
struct {
    u64 value;         /* The value of the event */
    u64 time_enabled;  /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
    u64 time_running;  /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
    u64 id;            /* if PERF_FORMAT_ID */
};
.fi
.in

The values read are described in more detail below.
.RS
.TP
.I nr
The number of events in this file descriptor.
Only available if
.B PERF_FORMAT_GROUP
was specified.

.TP
.IR time_enabled ", " time_running
Total time the event was enabled and running.
Normally these are the same.
If more events are started
than available counter slots on the PMU, then multiplexing
happens and events only run part of the time.
In that case the
.I time_enabled
and
.I time running
values can be used to scale an estimated value for the count.

.TP
.I value
An unsigned 64-bit value containing the counter result.

.TP
.I id
A globally unique value for this particular event, only there if
.B PERF_FORMAT_ID
was specified in read_format.

.RE




.RE

.SS "rdpmc instruction"
Starting with Linux 3.4 on x86, you can use the
.I rdpmc
instruction to get low-latency reads without having to enter the kernel.


.SS "perf_event ioctl calls"
.PP
Various ioctls act on
.BR perf_event_open()
file descriptors
.\" FIXME the arguments for these ioctl() operations need to be described
.TP
.B PERF_EVENT_IOC_ENABLE
Enables an individual counter or counter group.

.TP
.B PERF_EVENT_IOC_DISABLE
Disables an individual counter or counter group.

Enabling or disabling the leader of a group enables or disables the
entire group; that is, while the group leader is disabled, none of the
counters in the group will count.
Enabling or disabling a member of a group other than the leader only
affects that counter; disabling a non-leader
stops that counter from counting but doesn't affect any other counter.

.TP
.B PERF_EVENT_IOC_REFRESH
Non-inherited overflow counters can use this
to enable a counter for 'nr' events, after which it gets disabled again.
.\" FIXME the following needs clarification/confirmation
I think the goal of IOC_REFRESH is not to reload the period but simply to
adjust the number of events before the next notifications.

.TP
.B PERF_EVENT_IOC_RESET
Reset the event count to zero.
This only resets the counts; there is no way to reset the
multiplexing
.I time_enabled
or
.I time_running
values.
When sent to a group leader, only
the leader is reset (child events are not).

.TP
.B PERF_EVENT_IOC_PERIOD
IOC_PERIOD is the command to update the period; it
does not update the current period but instead defers until next.

.TP
.B PERF_EVENT_IOC_SET_OUTPUT
This tells the kernel to report event notifications to the specified
file descriptor rather than the default one.
The file descriptors must all be on the same CPU.

.TP
.BR PERF_EVENT_IOC_SET_FILTER " (Since Linux 2.6.33)"
This adds an ftrace filter to this event.

.SS "Using prctl"
A process can enable or disable all the counter groups that are
attached to it using the
.BR prctl (2)
.B PR_TASK_PERF_EVENTS_ENABLE
and
.B PR_TASK_PERF_EVENTS_DISABLE
operations.
This applies to all counters on the current process, whether created by
this process or by another, and does not affect any counters that this
process has created on other processes.
It only enables or disables
the group leaders, not any other members in the groups.

.SS Files in /proc/sys/kernel/

.RS
.TP
.I
/proc/sys/kernel/perf_event_paranoid

The
.I perf_event_paranoid
file can be set to restrict access to the performance counters.

2 - only allow userspace measurements

1 - (default) allow both kernel and user measurements

0 - allow access to CPU-specific data but not raw tracepoint samples

\-1 - no restrictions

The existence of the
.I perf_event_paranoid
file is the official method for determining if a kernel supports
.BR perf_event_open().

.TP
.I /proc/sys/kernel/perf_event_max_sample_rate

This sets the maximum sample rate.  Setting this too high can allow
users to sample at a rate that impacts overall machine performance
and potentially lock up the machine.  The default value is
100000 (samples per second).

.TP
.I /proc/sys/kernel/perf_event_mlock_kb

Maximum number of pages an unprivledged user can mlock (2) .
The default is 129 (kB).
.RE

.SS Files in /sys/bus/event_source/devices/

Since Linux 2.6.34 the kernel supports having multiple PMUs
available for monitoring.  
Information on how to program these PMUs can be found under
.IR /sys/bus/event_source/devices/ .
Each subdirectory corresponds to a different PMU.

.RS
.TP
.I /sys/bus/event_source/devices/*/type
This contains an integer that can be used in the
.I type
field of perf_event_attr to indicate you wish to use this PMU.

.TP
.I /sys/bus/event_source/devices/*/rdpmc
[To be documented]

.TP
.I /sys/bus/event_source/devices/*/format/
This sub-directory contains information on what bits in the
.I config
field of perf_event_attr correspond to.

.TP
.I /sys/bus/event_source/devices/*/uevent
[To be documented]

.RE


.SH "RETURN VALUE"
.BR perf_event_open ()
returns the new file descriptor, or \-1 if an error occurred
(in which case,
.I errno
is set appropriately).
.SH ERRORS
.TP
.B EINVAL
Returned if the specified event is not available.
.TP
.B ENOSPC
Prior to Linux 3.3, if there was no counter room,
.B ENOSPC
was returned.
Linus did not like this, and this was changed to
.BR EINVAL .
.B ENOSPC
is still returned if you try to read results into
too small a buffer.

.SH VERSION

.BR perf_event_open ()
was introduced in Linux 2.6.31 but was called
.BR perf_counter_open () .
It was renamed in Linux 2.6.32.

.SH CONFORMING TO

This call is specific to Linux
and should not be used in programs intended to be portable.

.SH NOTES
Glibc does not provide a wrapper for this system call; call it using
.BR syscall (2).

The official way of knowing if
.BR perf_event_open()
support is enabled is checking
for the existence of the file
.I /proc/sys/kernel/perf_event_paranoid

.SH BUGS

The
.B F_SETOWN_EX
option to
.IR fcntl (2)
is needed to properly get overflow signals in threads.
This was introduced in Linux 2.6.32.

Prior to Linux 2.6.33 (at least for x86) the kernel did not check
if events could be scheduled together until read time.
The same happens on all known kernels if the NMI watchdog is enabled.
This means to see if a given set of events works you have to
.BR perf_event_open (),
start, then read before you know for sure you
can get valid measurements.

Prior to Linux 2.6.34 event constraints were not enforced by the kernel.
In that case, some events would silently return "0" if the kernel
scheduled them in an improper counter slot.

Prior to Linux 2.6.34 there was a bug when multiplexing where the
wrong results could be returned.

Kernels from Linux 2.6.35 to Linux 2.6.39 can quickly crash the kernel if
"inherit" is enabled and many threads are started.

Prior to Linux 2.6.35,
.B PERF_FORMAT_GROUP
did not work with attached processes.

In older Linux 2.6 versions,
refreshing an event group leader refreshed all siblings,
and refreshing with a parameter of 0 enabled infinite refresh.
This behavior is unsupported and should not be relied on.

There is a bug in the kernel code between
Linux 2.6.36 and Linux 3.0 that ignores the
"watermark" field and acts as if a wakeup_event
was chosen if the union has a
non-zero value in it.

Always double-check your results!  Various generalized events
have had wrong values.
For example, retired branches measured
the wrong thing on AMD machines until Linux 2.6.35.

.SH EXAMPLE
The following is a short example that measures the total
instruction count of a call to printf().
.nf

#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include <asm/unistd.h>

long perf_event_open( struct perf_event_attr *hw_event, pid_t pid,
                      int cpu, int group_fd, unsigned long flags )
{
    int ret;

    ret = syscall( __NR_perf_event_open, hw_event, pid, cpu,
                   group_fd, flags );
    return ret;
}


int
main(int argc, char **argv)
{

    struct perf_event_attr pe;
    long long count;
    int fd;

    memset(&pe, 0, sizeof(struct perf_event_attr));
    pe.type = PERF_TYPE_HARDWARE;
    pe.size = sizeof(struct perf_event_attr);
    pe.config = PERF_COUNT_HW_INSTRUCTIONS;
    pe.disabled = 1;
    pe.exclude_kernel = 1;
    pe.exclude_hv = 1;

    fd = perf_event_open(&pe, 0, \-1, \-1, 0);
    if (fd < 0) {
       fprintf(stderr, "Error opening leader %llx\\n", pe.config);
    }

    ioctl(fd, PERF_EVENT_IOC_RESET, 0);
    ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);

    printf("Measuring instruction count for this printf\\n");

    ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
    read(fd, &count, sizeof(long long));

    printf("Used %lld instructions\\n", count);

    close(fd);
}
.fi

.SH "SEE ALSO"
.BR fcntl (2),
.BR mmap (2),
.BR open (2),
.BR prctl (2),
.BR read (2)
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html