Expand description

Abstractions for deferred interrupt tasks, a companion to regular interrupt handlers.

Deferred interrupt tasks are similar to the concept of “top half” and “bottom half” interrupt handlers in other OSes, in which the top half is the short, latency-sensitive function that runs immediately when the interrupt request is serviced, while the bottom half is the more complex function that runs in a deferred manner to handle longer operations. Other terminology is often used, including “first-level” and “second-level” interrupt handlers, or “hard” and “soft” interrupt handlers.

That being said, this implementation of deferred interrupt tasks differs from both tasklets and workqueues in Linux. We also do not use the “top half” or “bottom half” terminology because it is confusing and difficult to remember which is which. Instead, we refer to the first latency-sensitive part as the interrupt handler and the second later part as the deferred task. The “interrupt handler” runs immediately (in a synchronous fashion) when the interrupt occurs, while the deferred task runs asynchronously at some time in the future, ideally as soon as possible.

The general idea is that an interrupt handler should be short and do the minimum amount of work possible in order to keep the system responsive, because all (or most) other interrupts are typically disabled while the interrupt handler executes to completion. Thus, most of the work should be deferred until later, such that the interrupt handler itself only does a couple of quick things:

  • Notifies the deferred task that work is ready to be done, optionally providing details about what work it needs to do,
  • Acknowledges the interrupt such that the hardware knows it was handled.

The deferred task is tied directly to a single interrupt handler in a 1-to-1 manner at the time of creation, which occurs in register_interrupt_handler(). Therefore, it is both efficient and easy to use. In the simplest of cases, such as a serial port device, the interrupt handler only needs to mark the deferred task as unblocked (runnable) and then acknowledge the interrupt. No other data exchange is needed between the interrupt handler and the deferred task. For more complicated cases, the interrupt handler may need to do a minimal amount of bookkeeping tasks (such as advancing a ringbuffer index) and potentially send some information about what the deferred task should do. It is typically best to use a lock-free queue or an interrupt-safe mutex to share such information between the interrupt handler and deferred task.



  • Registers an interrupt handler and spawns a companion “deferred task” that asynchronously handles the longer-running operations related to that interrupt.