Specifications

Device Driver Entry Points

Timeout Handling Routine

Takes whatever action is necessary when a device has not yet responded to a

request for device activity and the time allowed for a response has expired.

Speciﬁed in

Specify the address of the timeout handling routine in the excpt argument to the

WFIKPCH or the WFIRLCH macro.

Called by

The WFIKPCH and WFIRLCH macros use this entry point, but only when the

name of a timeout handling routine is provided in their excpt argument. These

macros are used in the driver’s start-I/O routine; thus, strictly speaking, the

driver itself is the only entity that uses this entry point.

Routines in the system module TIMESCHDL call the timeout handling routine at

the request of the WFIKPCH and WFIRLCH macros.

Synchronization

A timeout handling routine is called at device IPL and must return to its caller at

device IPL. In a multiprocessing environment, the processor holds both the fork

lock and device lock associated with the device at the time of the call.

After taking whatever device-speciﬁc action is necessary at device IPL, a timeout

handling routine can lower IPL to fork IPL to perform less critical activities.

Because its caller restores IPL to fork IPL (and releases the device lock in a

multiprocessing environment), if a timeout handling routine does lower IPL, it

can do so only by forking or by performing the following steps:

• Issue a DEVICEUNLOCK macro to lower to fork level

• Perform timeout handling activities possible at the lower IPL

• Issue a DEVICELOCK macro to again obtain the device lock and raise to

device IPL

• Issue an RSB instruction to return to its caller

Context

Because a timeout handling routine executes in the context of a fork process, it

can access only those virtual addresses that refer to system (S0) space.

A timeout handling routine can use R0, R1, and R2 freely, but must preserve the

contents of all other registers. If a timeout handling routine uses the stack, it

must restore the stack before completing or canceling the current I/O request,

waiting for an interrupt, or returning control to its caller.

4–19