Intel 80C188XL, 80C186XL user manual Bus Interface Unit, 3.7.1.2Refresh Operation During a Bus HOLD

BUS INTERFACE UNIT

The major factors that influence bus latency are listed below (in order from longest delay to short- est delay).

1.Bus Not Ready — As long as the bus remains not ready, a bus hold request cannot be serviced.

2. Locked Bus Cycle — As long as LOCK remains asserted, a bus hold request cannot be serviced. Performing a locked move string operation can take several thousands of clocks.

3.Completion of Current Bus Cycle — A bus hold request cannot be serviced until the current bus cycle completes. A bus hold request will not separate bus cycles required to move odd-aligned word data. Also, bus cycles with long wait states will delay the servicing of a bus hold request.

4.Interrupt Acknowledge Bus Cycle — A bus hold request is not serviced until after an INTA bus cycle has completed. An INTA bus cycle drives LOCK active.

5.DMA and Refresh Bus Cycles — A bus hold request is not serviced until after the DMA request or refresh bus cycle has completed. Refresh bus cycles have a higher priority than hold bus requests. A bus hold request cannot separate the bus cycles associated with a DMA transfer (worst case is an odd-aligned transfer, which takes four bus cycles to complete).

3.7.1.2Refresh Operation During a Bus HOLD

Under normal operating conditions, once HLDA has been asserted it remains asserted until HOLD is removed. However, when a refresh bus request is generated, the HLDA output is re- moved (driven low) to signal the need for the BIU to regain control of the local bus. The BIU does not gain control of the bus until HOLD is removed. This procedure prevents the BIU from just arbitrarily regaining control of the bus.

Figure 3-35 shows the timing associated with the occurrence of a refresh request while HLDA is active. Note that HLDA can be as short as one clock in duration. This happens when a refresh request occurs just after HLDA is granted. A refresh request has higher priority than a bus hold request; therefore, when the two occur simultaneously, the refresh request occurs before HLDA becomes active.

3-41