-
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 n sys/types.h n sys/systm.h n 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" n 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. l -tag -width indent t 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. t Fn counter_u64_free c Free the previously allocated counter .Fa c . t Fn counter_u64_add c v Add .Fa v to .Fa c . The KPI does not guarantee any protection from wraparound. t 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 . t Fn counter_exit Exit mode for updating several counters. t Fn counter_u64_add_protected c v Same as .Fn counter_u64_add , but should be preceded by .Fn counter_enter . t 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. t Fn counter_u64_zero c Clear the counter .Fa c and set it to zero. t 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. t 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.
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 n sys/types.h n sys/systm.h n 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" n 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. l -tag -width indent t 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. t Fn counter_u64_free c Free the previously allocated counter .Fa c . t Fn counter_u64_add c v Add .Fa v to .Fa c . The KPI does not guarantee any protection from wraparound. t 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 . t Fn counter_exit Exit mode for updating several counters. t Fn counter_u64_add_protected c v Same as .Fn counter_u64_add , but should be preceded by .Fn counter_enter . t 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. t Fn counter_u64_zero c Clear the counter .Fa c and set it to zero. t 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. t 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.
p 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.
p 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.
p 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 .