freebsd-nq/share/man/man9/counter.9
Gleb Smirnoff 4e76af6a41 Merge from projects/counters: counter(9).
Introduce counter(9) API, that implements fast and raceless counters,
provided (but not limited to) for gathering of statistical data.

See http://lists.freebsd.org/pipermail/freebsd-arch/2013-April/014204.html
for more details.

In collaboration with:	kib
Reviewed by:		luigi
Tested by:		ae, ray
Sponsored by:		Nginx, Inc.
2013-04-08 19:40:53 +00:00

201 lines
5.4 KiB
Groff

.\"-
.\" Copyright (c) 2013 Gleb Smirnoff <glebius@FreeBSD.org>
.\" All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\" notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\" notice, this list of conditions and the following disclaimer in the
.\" documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.\" $FreeBSD$
.\"
.Dd April 3, 2013
.Dt COUNTER 9
.Os
.Sh NAME
.Nm counter
.Nd "SMP-friendly kernel counter implementation"
.Sh SYNOPSIS
.In sys/types.h
.In sys/counter.h
.Ft counter_u64_t
.Fn counter_u64_alloc "int wait"
.Ft void
.Fn counter_u64_free "counter_u64_t c"
.Ft void
.Fn counter_u64_add "counter_u64_t c" "int64_t v"
.Ft void
.Fn counter_enter
.Ft void
.Fn counter_exit
.Ft void
.Fn counter_u64_add_protected "counter_u64_t c" "int64_t v"
.Ft uint64_t
.Fn counter_u64_fetch "counter_u64_t c"
.Ft void
.Fn counter_u64_zero "counter_u64_t c"
.In sys/sysctl.h
.Fn SYSCTL_COUNTER_U64 parent nbr name access ptr val descr
.Fn SYSCTL_ADD_COUNTER_U64 ctx parent nbr name access ptr descr
.Sh DESCRIPTION
.Nm
is a generic facility to create counters
that can be utilized for any purpose (such as collecting statistical
data).
A
.Nm
is guaranteed to be lossless when several kernel threads do simultaneous
updates.
However,
.Nm
does not block the calling thread,
also no
.Xr atomic 9
operations are used for the update, therefore the counters
can be used in any non-interrupt context.
Moreover,
.Nm
has special optimisations for SMP environments, making
.Nm
update faster than simple arithmetic on the global variable.
Thus
.Nm
is considered suitable for accounting in the performance-critical
code pathes.
.Bl -tag -width indent
.It Fn counter_u64_alloc how
Allocate a new 64-bit unsigned counter.
The
.Fa wait
argument is the
.Xr malloc 9
wait flag, should be either
.Va M_NOWAIT
or
.Va M_WAITOK .
If
.Va M_WAITOK
is specified the operation may fail.
.It Fn counter_u64_free c
Free the previously allocated counter
.Fa c .
.It Fn counter_u64_add c v
Add
.Fa v
to
.Fa c .
The KPI does not guarantee any protection from wraparound.
.It Fn counter_enter
Enter mode that would allow to safely update several counters via
.Fn counter_u64_add_protected .
On some machines this expands to
.Xr critical 9
section, while on other is a nop.
See
.Sx IMPLEMENTATION DETAILS .
.It Fn counter_exit
Exit mode for updating several counters.
.It Fn counter_u64_add_protected c v
Same as
.Fn counter_u64_add ,
but should be preceded by
.Fn counter_enter .
.It Fn counter_u64_fetch c
Take a snapshot of counter
.Fa c .
The data obtained is not guaranteed to reflect the real cumulative
value for any moment.
.It Fn counter_u64_zero c
Clear the counter
.Fa c
and set it to zero.
.It Fn SYSCTL_COUNTER_U64 parent nbr name access ptr val descr
Declare a static
.Xr sysctl
oid that would represent a
.Nm .
The
.Fa ptr
argument should be a pointer to allocated
.Vt counter_u64_t .
A read of the oid returns value obtained through
.Fn counter_u64_fetch .
Any write to the oid zeroes it.
.It Fn SYSCTL_ADD_COUNTER_U64 ctx parent nbr name access ptr descr
Create a
.Xr sysctl
oid that would represent a
.Nm .
The
.Fa ptr
argument should be a pointer to allocated
.Vt counter_u64_t .
A read of the oid returns value obtained through
.Fn counter_u64_fetch .
Any write to the oid zeroes it.
.El
.Sh IMPLEMENTATION DETAILS
On all architectures
.Nm
is implemented using per-CPU data fields that are specially aligned
in memory, to avoid inter-CPU bus traffic due to shared use
of the variables between CPUs.
These are allocated using
.Va UMA_ZONE_PCPU
.Xr uma 9
zone.
The update operation only touches the field that is private to current CPU.
Fetch operation loops through all per-CPU fields and obtains a snapshot
sum of all fields.
.Pp
On amd64 a
.Nm counter
update is implemented as a single instruction without lock semantics,
operating on the private data for the current CPU,
which is safe against preemption and interrupts.
.Pp
On i386 architecture, when machine supports the cmpxchg8 instruction,
this instruction is used.
The multi-instruction sequence provides the same guarantees as the
amd64 single-instruction implementation.
.Pp
On some architectures updating a counter require a
.Xr critical 9
section.
.Sh SEE ALSO
.Xr atomic 9 ,
.Xr critical 9 ,
.Xr locking 9 ,
.Xr malloc 9 ,
.Xr sysctl 9 ,
.Xr uma 9
.Sh HISTORY
The
.Nm
facility first appeared in
.Fx 10.0 .
.Sh AUTHORS
.An -nosplit
The
.Nm
facility was written by
.An Gleb Smirnoff
and
.An Konstantin Belousov .