Linux Kernel Architecture

Chapter 14: Kernel Activities

Note that the prototype of ISRs was changed during the development of 2.6.19! Before, the arguments of

the handler routine also included a pointer to the saved registers:

<interrupt.h>

irqreturn_t (*handler)(int irq, void *dev_id, struct pt_regs *regs);

Interrupt handlers are obviously hot code paths, and time is very critical. Although most handlers do not

need the register state, time and stack space is required to pass a pointer to it to every ISR. Removing this

pointer from the prototype is thus a good idea.^16

Handlers that need the register set can still access it. The kernel defines a global per-CPU array that stores

the registers, andget_irq_regsfrom>include/asm-generic/irq_regs.h>can be used to retrieve a

pointer to thept_regsinstance. This instance contains the register setting that was active when the

switch to kernel mode was made. The information is not used by normal device drivers but sometimes

comes in useful when debugging kernel problems.

Again we emphasize that interrupt handlers can only use two return values:IRQ_HANDLEDif the IRQ was

handled correctly, orIRQ_NONEif the ISR did not feel responsible for the IRQ.

What are the tasks of a handler routine? To service a shared interrupt, the routine must first check

whether the IRQ is intended for it. If the peripheral device is of a more modern design, the hardware

offers a simple method of performing this check, usually by means of a special device register. If the

device has caused an interrupt, the register value is set to 1. In this case, the handler routine must restore

the value to its default (usually 0) and then start normal servicing of the interrupt. If it finds the value 0,

it can be sure that the managed device is not the source of the interrupt, and control can be returned to

the higher-level code.

If a device does not have a state register of this kind, the option of manual polling still remains. Each

time an interrupt occurs, the handler checks whetherdata are available for the device. If so, the data are

processed. If not, the routine is terminated.

A handler routine can, of course, be responsible for several devices at the same time, for example, two

network cards of the same type. If an IRQ is received, the same code is executed on both cards because

both handler functions point to the same position in the kernel code. If the two devices use different IRQ

numbers, the handler routine can differentiate between them. If they share a common IRQ, reference can

still be made to the device-specificdev_idfield to uniquely identify each card.

14.2 Software Interrupts

Software interrupts enable the kernel to defer tasks. Because they function in a similar way to the inter-

rupts described above but are implemented fully inthe software, they are logically enough known as

software interruptsorsoftIRQs.

The kernel is informed of an anomalous condition by means of a software interrupt, and the situation is

resolved at some later time by special handler routines. As already noted, the kernel services all pending

software interrupts at the end ofdo_IRQso that regular activation is ensured.

(^16) Since the patch that introduced the change had to change every ISR, it might well be the one to touch most files at a single blow
in the kernel history.