1.\" Copyright (c) 2002 Luigi Rizzo 2.\" All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 7.\" 1. Redistributions of source code must retain the above copyright 8.\" notice, this list of conditions and the following disclaimer. 9.\" 2. Redistributions in binary form must reproduce the above copyright 10.\" notice, this list of conditions and the following disclaimer in the 11.\" documentation and/or other materials provided with the distribution. 12.\" 13.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23.\" SUCH DAMAGE. 24.\" 25.\" $FreeBSD: src/share/man/man4/polling.4,v 1.27 2007/04/06 14:25:14 brueffer Exp $ 26.\" $DragonFly: src/share/man/man4/polling.4,v 1.12 2007/10/03 09:55:25 sephe Exp $ 27.\" 28.Dd October 2, 2007 29.Dt POLLING 4 30.Os 31.Sh NAME 32.Nm polling 33.Nd device polling support 34.Sh SYNOPSIS 35.Cd "options DEVICE_POLLING" 36.Sh DESCRIPTION 37Device polling 38.Nm ( 39for brevity) refers to a technique that 40lets the operating system periodically poll devices, instead of 41relying on the devices to generate interrupts when they need attention. 42This might seem inefficient and counterintuitive, but when done 43properly, 44.Nm 45gives more control to the operating system on 46when and how to handle devices, with a number of advantages in terms 47of system responsiveness and performance. 48.Pp 49In particular, 50.Nm 51reduces the overhead for context 52switches which is incurred when servicing interrupts, and 53gives more control on the scheduling of a CPU between various 54tasks (user processes, software interrupts, device handling) 55which ultimately reduces the chances of livelock in the system. 56.Ss Principles of Operation 57In the normal, interrupt-based mode, devices generate an interrupt 58whenever they need attention. 59This in turn causes a 60context switch and the execution of an interrupt handler 61which performs whatever processing is needed by the device. 62The duration of the interrupt handler is potentially unbounded 63unless the device driver has been programmed with real-time 64concerns in mind (which is generally not the case for 65.Dx 66drivers). 67Furthermore, under heavy traffic load, the system might be 68persistently processing interrupts without being able to 69complete other work, either in the kernel or in userland. 70.Pp 71Device polling disables interrupts by polling devices on clock 72interrupts. 73This way, the context switch overhead is removed. 74Furthermore, 75the operating system can control accurately how much work to spend 76in handling device events, and thus prevent livelock by reserving 77some amount of CPU to other tasks. 78.Pp 79Enabling 80.Nm 81also changes the way software network interrupts 82are scheduled, so there is never the risk of livelock because 83packets are not processed to completion. 84.Ss Enabling polling 85Currently only network interface drivers support the 86.Nm 87feature. 88It is turned on and off with help of 89.Xr ifconfig 8 90command. 91.Ss Loader Tunables 92The following tunables can be set from 93.Xr loader.conf 5 : 94.Bl -tag -width indent -compact 95.It Va kern.polling.enable 96If set to non-zero, 97.Nm 98is enabled. 99Default is enabled. 100.Pp 101.It Va kern.polling.cpumask 102A bitmask that controls which CPUs support device polling. 103Default is 0xffffffff. 104.El 105.Ss MIB Variables 106The operation of 107.Nm 108is controlled by the following per CPU 109.Xr sysctl 8 110MIB variables 111.Em ( X 112is the CPU number): 113.Pp 114.Bl -tag -width indent -compact 115.It Va kern.polling.X.enable 116If set to non-zero, 117.Nm 118is enabled. 119Default is enabled. 120.Pp 121.It Va kern.polling.X.pollhz 122The polling frequency, whose range is 1 to 30000. 123Default is 2000. 124.Pp 125.It Va kern.polling.cpumask 126A read only bitmask of the CPUs that support device polling. 127.Pp 128.It Va kern.polling.defcpu 129The default CPU used to run device polling (read only). 130.Pp 131.It Va kern.polling.X.user_frac 132When 133.Nm 134is enabled, and provided that there is some work to do, 135up to this percent of the CPU cycles is reserved to userland tasks, 136the remaining fraction being available for 137.Nm 138processing. 139Default is 50. 140.Pp 141.It Va kern.polling.X.burst 142Maximum number of packets grabbed from each network interface in 143each timer tick. 144This number is dynamically adjusted by the kernel, 145according to the programmed 146.Va user_frac , burst_max , 147CPU speed, and system load. 148.Pp 149.It Va kern.polling.X.each_burst 150The burst above is split into smaller chunks of this number of 151packets, going round-robin among all interfaces registered for 152.Nm . 153This prevents the case that a large burst from a single interface 154can saturate the IP interrupt queue 155.Pq Va net.inet.ip.intr_queue_maxlen . 156Default is 5. 157.Pp 158.It Va kern.polling.X.burst_max 159Upper bound for 160.Va kern.polling.burst . 161Note that when 162.Nm 163is enabled, each interface can receive at most 164.Pq Va pollhz No * Va burst_max 165packets per second unless there are spare CPU cycles available for 166.Nm 167in the idle loop. 168This number should be tuned to match the expected load 169(which can be quite high with GigE cards). 170Default is 150 which is adequate for 100Mbit network and pollhz=1000. 171.Pp 172.It Va kern.polling.X.reg_frac 173Controls how often (every 174.Va reg_frac No / Va pollhz 175seconds) the status registers of the device are checked for error 176conditions and the like. 177Increasing this value reduces the load on the bus, but also delays 178the error detection. 179Default is 20. 180.Pp 181.It Va kern.polling.X.handlers 182How many active devices have registered for 183.Nm . 184.Pp 185.It Va kern.polling.X.short_ticks 186.It Va kern.polling.X.lost_polls 187.It Va kern.polling.X.pending_polls 188.It Va kern.polling.X.residual_burst 189.It Va kern.polling.X.phase 190.It Va kern.polling.X.suspect 191.It Va kern.polling.X.stalled 192Debugging variables. 193.El 194.Sh SUPPORTED DEVICES 195Device polling requires explicit modifications to the device drivers. 196As of this writing, the 197.Xr bce 4 , 198.Xr bge 4 , 199.Xr dc 4 , 200.Xr em 4 , 201.Xr fwe 4 , 202.Xr fxp 4 , 203.Xr nfe 4 , 204.Xr nge 4 , 205.Xr re 4 , 206.Xr rl 4 , 207.Xr sis 4 , 208.Xr stge 4 , 209.Xr vge 4 , 210.Xr vr 4 , 211.Xr wi 4 212and 213.Xr xl 4 214devices are supported, with others in the works. 215The modifications are rather straightforward, consisting in 216the extraction of the inner part of the interrupt service routine 217and writing a callback function, 218.Fn *_poll , 219which is invoked 220to probe the device for events and process them. 221(See the 222conditionally compiled sections of the devices mentioned above 223for more details.) 224.Pp 225In order to reduce the latency in processing packets, 226it is advisable to set the 227.Xr sysctl 8 228variable 229.Va kern.polling.X.pollhz 230to at least 1000. 231.Sh HISTORY 232Device polling first appeared in 233.Fx 4.6 . 234It was rewritten in 235.Dx 1.3 . 236.Sh AUTHORS 237.An -nosplit 238The device polling code was rewritten by 239.An Matt Dillon 240based on the original code by 241.An Luigi Rizzo Aq luigi@iet.unipi.it . 242.An Sepherosa Ziehau 243made the polling frequency settable at runtime and added per CPU polling. 244