Table of Contents
Advertisement
38. Additional Notes on DMA Features
In the feature set listed on page 6-1, bullet six is "Operand Packing and Unpacking for Dual-Address Trans-
fers". This packing is for transfers between different port sizes selected in the DMA channel control register,
e.g. Byte <> Word transfers. The DMA controller does not do packing for byte > byte transfers, eliminating the
problem of residual bytes left in the controller when a channel is stopped after an odd byte transfer count.
39. Additional Note on Internal Request Generation
Add to the Internal Request Generation section on page 6-5: For internal request operation, DACKx and DON-
Ex are not active as outputs during transfers. DONEx is valid as an input though and will terminate channel
operation if asserted - pull up if not used.
40. Additional Note on DMA Transfer Latency from DREQ
Add to the External Request Generation section beginning 6-5: DREQx assertions require two clocks for input
synchronization and IMB bus arbitration activity before the resulting DMA bus cycle can start. A DREQx as-
sertion will preempt the next CPU bus cycle if it is recognized two or more clocks before the end of the current
bus cycle, unless the current cycle is not the last cycle of an operand transfer, or is the read of an RMC cycle.
Operand transfers and RMC read/write sequences are indivisible to guarantee data coherency - the bus can-
not be arbitrated from the CPU until the complete operand transfer completes, even if operand and memory
sizing results in multiple bus cycles.
For a DREQx assertion during an idle bus period, bus state S0 of the DMA bus cycle starts 2.5 clocks after the
clock falling edge which DREQx is recognized on. The maximum latency from the clock falling edge that
DREQx is recognized on to the falling edge that AS for the DMA cycle asserts from is shown in the following
table for various memory speeds.
DREQ Latency (Clocks) vs. Bus Width and Access Times
Access Type
41. Additional Note on Burst Transfer DREQx Negation and Overhead
On page 6-5, replace the 2nd paragraph of 6.3.2.1 External Burst Mode with the following: DREQx must be
negated one clock before the end of the last DMA bus cycle of a burst to prevent another DMA transfer from
being generated. Also, DREQx must be negated two clocks before the end of the last DMA bus cycle to prevent
an idle clock between that transfer and the following CPU access.
42. Additional Note on Cycle steal DMA arbitration overhead
Add to the External Cycle Steal Mode description on page 6-6: In general, DMA arbitration occurs transpar-
ently. However, for some 2-clock accesses using cycle steal an idle clock can follow the DMA transfer due to
incomplete overlap of the DMA transfer with internal IMB arbitration. Specifically, an idle clock can follow 1)
single address 2-clock transfers and 2) dual address transfers from memory to 2-clock devices. Arbitration is
completely overlapped for all other cases.
MOTOROLA
Maximum DREQ Latency (Clocks)
16-Bit Bus
Clocks/Bus Cycle
2
Longword
7
RMC (TAS)
10
12
MC68341 USER'S MANUAL ADDENDUM
Clocks/Bus Cycle
3
4
5
2
9
11
13
11
14
16
10
8-Bit Bus
3
4
5
15
19
23
12
14
16
11
Advertisement
Table of Contents
