NAME

splbio, splclock, splhigh, splimp, splnet, splsoftclock, splsofttty, splstatclock, spltty, splvm, spl0, splx — manipulate interrupt priorities

SYNOPSIS

#include <sys/types.h>
#include <sys/systm.h>

intrmask_t
splbio(void);

intrmask_t
splclock(void);

intrmask_t
splhigh(void);

intrmask_t
splimp(void);

intrmask_t
splnet(void);

intrmask_t
splsoftclock(void);

intrmask_t
splsofttty(void);

intrmask_t
splstatclock(void);

intrmask_t
spltty(void);

void
spl0(void);

void
splx(intrmask_t ipl);

DESCRIPTION

The spl() function family sets the interrupt priority “level” of the CPU. This prevents interrupt handlers of the blocked priority level from being run. This is used in the “synchronous” part of a driver (the part that runs on behalf of the user process) to examine or modify data areas that might be examined or modified by interrupt handlers.

Each driver that uses interrupts is normally assigned to an interrupt priority group by a keyword in its config line. For example:

device foo0 at isa? port 0x0815 irq 12 tty

assigns interrupt 12 to the “tty” priority group. The system automatically arranges for interrupts in the xxx group to be called at a priority >= splxxx()

The function splx() sets the interrupt priority to an absolute value. The intent is that the value returned by the other functions should be saved in a local variable, and later passed to splx() in order to restore the previous priority.

The function spl0() lowers the priority to a value where all interrupt handlers are unblocked, but ASTs (asynchronous system traps) remain blocked until the system is about to return to user mode.

The traditional assignment of the various device drivers to the interrupt priority groups can be roughly classified as:

splnet()

Software part of the network interface drivers.

splimp()

All network interface drivers.

splbio()

All buffered IO (i.e., disk and the like) drivers.

spltty()

Basically, all non-network communications devices, but effectively used for all drivers that are neither network nor disks.

RETURN VALUES

All functions except splx() and spl0() return the previous priority value.

EXAMPLES

This is a typical example demonstrating the usage:

struct foo_softc {
	...
	int flags;
#define	FOO_ASLEEP	1
#define	FOO_READY	2

} foo_softc[NFOO];

int
foowrite(...)
{
	struct foo_softc *sc;
	int s, error;

	...
	s = spltty();
	if (!(sc->flags & FOO_READY)) {
		/* Not ready, must sleep on resource. */
		sc->flags |= FOO_ASLEEP;
		error = tsleep(sc, PZERO, "foordy", 0);
		sc->flags &= ~FOO_ASLEEP;
	}
	sc->flags &= ~FOO_READY;
	splx(s);

	...
}

void
foointr(...)
{
	struct foo_softc *sc;

	...
	sc->flags |= FOO_READY;
	if (sc->flags & FOO_ASLEEP)
		/* Somebody was waiting for us, awake him. */
		wakeup(sc);
	...
}

HISTORY

The interrupt priority levels appeared in a very early version of UNIX. They have been traditionally known by number instead of by names, and were inclusive up to higher priority levels (i.e., priority 5 has been blocking everything up to level 5). This is no longer the case in FreeBSD. The traditional name ‘level’ for them is still reflected in the letter ‘l’ of the respective functions and variables, although they are not really levels anymore, but rather different (partially inclusive) sets of functions to be blocked during some periods of the life of the system. The historical number scheme can be considered as a simple linearly ordered set of interrupt priority groups.

FreeBSD 5.0 eliminated spl entirely in favor of locking primitives which scale to more than one processor.

AUTHORS

This manual page was written by Jörg Wunsch.