Principles of Operation¶

In the normal, interrupt-based mode, devices generate an interrupt whenever they need attention. This in turn causes a context switch and the execution of an interrupt handler which performs whatever processing is needed by the device. The duration of the interrupt handler is potentially unbounded unless the device driver has been programmed with real-time concerns in mind (which is generally not the case for FreeBSD drivers). Furthermore, under heavy traffic load, the system might be persistently processing interrupts without being able to complete other work, either in the kernel or in userland.

Device polling disables interrupts by polling devices at appropriate times, i.e., on clock interrupts and within the idle loop. This way, the context switch overhead is removed. Furthermore, the operating system can control accurately how much work to spend in handling device events, and thus prevent livelock by reserving some amount of CPU to other tasks.

Enabling polling also changes the way software network interrupts are scheduled, so there is never the risk of livelock because packets are not processed to completion.

Enabling polling¶

Currently only network interface drivers support the polling feature. It is turned on and off with help of ifconfig(8) command.

The historic kern.polling.enable, which enabled polling for all interfaces, can be replaced with the following code:

for i in `ifconfig -l` ;
  do ifconfig $i polling; # use -polling to disable
done

MIB Variables¶

The operation of polling is controlled by the following sysctl(8) MIB variables:

kern.polling.user_frac

When polling is enabled, and provided that there is some work to do, up to this percent of the CPU cycles is reserved to userland tasks, the remaining fraction being available for polling processing. Default is 50.

kern.polling.burst

Maximum number of packets grabbed from each network interface in each timer tick. This number is dynamically adjusted by the kernel, according to the programmed user_frac, burst_max, CPU speed, and system load.

kern.polling.each_burst

The burst above is split into smaller chunks of this number of packets, going round-robin among all interfaces registered for polling. This prevents the case that a large burst from a single interface can saturate the IP interrupt queue (net.inet.ip.intr_queue_maxlen). Default is 5.

kern.polling.burst_max

Upper bound for kern.polling.burst. Note that when polling is enabled, each interface can receive at most (HZ * burst_max) packets per second unless there are spare CPU cycles available for polling in the idle loop. This number should be tuned to match the expected load (which can be quite high with GigE cards). Default is 150 which is adequate for 100Mbit network and HZ=1000.

kern.polling.idle_poll

Controls if polling is enabled in the idle loop. There are no reasons (other than power saving or bugs in the scheduler's handling of idle priority kernel threads) to disable this.

kern.polling.reg_frac

Controls how often (every reg_frac / HZ seconds) the status registers of the device are checked for error conditions and the like. Increasing this value reduces the load on the bus, but also delays the error detection. Default is 20.

kern.polling.handlers

How many active devices have registered for polling.

kern.polling.short_ticks

 

kern.polling.lost_polls

 

kern.polling.pending_polls

 

kern.polling.residual_burst

 

kern.polling.phase

 

kern.polling.suspect

 

kern.polling.stalled

Debugging variables.