SoftIRQ

Softirq is a mechanism in the Linux kernel for deferring work from the hard interrupt context to a later, lower-priority execution path. It is part of the kernel’s bottom-half processing and is designed to handle high-frequency interrupt workloads without spending too long in the initial interrupt handler. Softirq handlers run in a special softirq context, which cannot sleep, and they execute with interrupts enabled to allow progress on other interrupts.

Softirq processing is per-CPU and uses a queue of pending softirq vectors. When a hard interrupt finishes, the kernel checks for pending softirqs and executes the corresponding handlers by invoking do_softirq. If the backlog grows too large or system load is high, the kernel may defer work to a per-CPU ksoftirqd thread, which runs in process context and can sleep, helping to avoid long softirq latencies.

There are several predefined softirq types used by different subsystems. Networking relies on NET_RX and NET_TX to process received packets and transmitted data. The BLOCK softirq handles block I/O layer tasks, while other softirqs cover system timer processing and various subsystems. In addition to softirqs, the kernel historically used tasklets as a simpler, less scalable bottom-half mechanism, but tasklets are now largely superseded by the softirq and workqueue infrastructure.

Softirq design emphasizes short, non-blocking handlers to minimize interrupt latency. Handlers should avoid sleeping, allocating large amounts of memory, or performing long I/O operations. If mismanaged, heavy softirq activity can lead to latency spikes or degraded throughput, which is why the kernel provides mechanisms like ksoftirqd to balance processing load and maintain responsiveness.