SPI is eventually the most simple interface to implement. It was designed especially to work with low end processors. I tend to use SPI for all my micro controller designs (*1).
Unless one intends to use some dedicated hardware, like a shift register (useful if speed is of concern), all it needs are 3 port bits for the data interface, plus one port bit per device connected (*2) for it's basic signals:
- MOSI - Out - Master Out Slave In, the data the CPU sends out
- MISO - In - Master In Slave Out, the data returned by external hardware
- SCLK - Out - Serial Clock, a signal to be toggled once per bit
- SS/CS - Out - Slave Select, also called Chip Select, selecting the external interface.
The great advantage is that SPI does not require any specific timing (*3) as long as the basic sequence is followed. It can be handled as slow as a bit per day. Any interface able to provide three output lines and one input line can be used. Including a classic PC parallel port. Well, or a Z80 PIO.
Let's say we have a Z80-PIO connected to a Z80-CPU at I/O address 20h (*4) and use port A with its first 4 lines assigned as (*4):
- A0 - MISO - In
- A1 - MOSI - Out
- A2 - SCLK - Out
- A3 - SS/CS - Out
We also assume the device to be operated in Mode 0 (CPOL=0, CPHA=0) as the RFM95 does. For other modes this may need to be adapted.
A subroutine to initialize this setup may look like this:
(All references relate to the Z80 Family CPU Peripherals User Manual UM008101-0601)
[Caveat, this has been just hacked down, so no guarantee for this code being bug free or working at all - not at least due to actual Beer-Level :)]
PIO1 EQU 20h ; First PIO
PIO1AD EQU PIO1+0 ; Data Register
PIO1AC EQU PIO1+2 ; Command Register
; Initialized SPI transfer on port A
; A destroyed
LD A,011001111b ; Set Mode 3, Control (see p.189)
LD A,000001110b ; Line 1/2/3 as output all other input (see p.190) (*5)
LD A,000000111b ; Interrupt control, no interrupt (see p.191)
LD A,000001111b ; Set CS high to deselect
By calling SPI_INIT the interface will be put into operating condition. Since sending and receiving is interleaved a single routine can be used:
SPI_CLK EQU 4
SPI_CS EQU 8
; Starts or continues an SPI transaction by SENDing a byte
; while RECeiving the peripherals response at the same time
; Called with byte to send in A
; Returns with byte received in A
LD D,A ; Byte to send in D
LD A,000000000b ; CS Low, Clock Low
LD B,8 ; 8 bits per byte
LD A,000000000 ; Clock Low, Data Low
RL D ; Isolate bit to send
JR NC,BCLR ; Is it a zero bit?
LD A,000000100 ; No? Then set data high
OUT (PIO1AD),A ; Set data value
MOV C,A ; Save value with clock low (*6)
OR A,SPI_CLK ; Set clock
OUT (PIO1AD),A ; Data and Clock
IN A,(PIO1AD) ; Read input data
RRCA ; Isolate input bit to carry
MOV A,C ; Restore Data with clock cleared (*7)
OUT (PIO1AD),A ; Clear Clock (*6)
RL E ; Insert bit from Carry into E
DJNZ BITLOOP ; Next bit?
LD A,E ; Return value in A
The important part is, as mentioned, to stay with the sequence:
- Activate CS
- Apply MOSI
- Raise Clock
- Read MISO
- Lower Clock
- Repeat step 2 thru 5 for all bits/bytes to be transferred
- Deactivate CS
Note that the routine does not perform step 7, it leaves chip select active. This is the base for multi-byte transfer, important for the RFM95, as each of it's transfers always starts with an address followed by one or more data bytes. CS has to stay active during such a transfer over all bytes to be written (and read), acting as transaction marker.
The SENDREC function is written in a way that multiple bytes can be chained by calling it in sequence as often as needed. Of course we now need a another function to end a transfer:
; Ends an SPI transaction
; A destroyed
LD A,000001111b ; Set CS high to deselect
Let's for example put the RFM95 into LoRa mode. This requires to set the high bit of the Operation Mode Register at address 01h needs (See p.102 of the manual). To do so we need to send one byte with the registers address, followed by the new value, all in in one transaction handled by CS:
LD A,01h ; Address of Operation Mode Register
CALL SPI_SENDREC ; Start transaction and send address
LD A,010000000b ; Set LoRa mode + Sleep
CALL SPI_SENDREC ; Continue transaction and send data
CALL SPI_DONE ; End transaction
Now the RFM95 is in LoRa mode (*7) - Easy, isn't it?
So while this function may not be super speed (*8,*9), it might be fast enough for 'some switching and sensor read out'.
Bottom Line: Z80, or any other classic CPU is quite capable of doing SPI (within reason).
P.S.: The very same code could ofc be used with the 245/374 based RC2014 I/O. Addresses need to be adjusted and initialization could be reduced to setting a default state. In fact, by using the same bit (2^0) for MOSI and MISO, one or two instructions could be saved.
And yes, it can be done in Pascal using the
port array (*10). Of course, a bit slower, like 2-5 times. Then again, packaging above assembly into functions would eliminate most of that.
P.P.S.: This made me smile:
is this something you can do with a Z80? Or would an older technology be much more practical?
A Z80 is about as old as it gets (1978), not to mention the TTL used by the 2014's digital I/O
*1 - Foremost to add MMC/SDC support, as the basic interface/protocol spoken even with modern terabyte sized cards is still SPI. But also any other interface or other controllers.
*2 - Using a decoder can reduce this when more than 2-3 devices are connected to a single interface.
*3 - Note that most devices have an upper speed limit, usually in the 5..40 MHz range, so not even remotely relevant (*9).
*4 - Doesn't really matter, just assumed for simplicity of the example code.
*5 - Unused lines should always be set as input.
*6 - With strict Mode 0, this could be deleted, as data is only valid during raising clock.
*7 - Well, in real use you may want to do this twice, just in case that the device was not in Sleep mode first, but that's another story.
*8 - I bet there is still room for improvement - this is, as said, just a quick writeup how it could work, made for the fun of it. Not much thought put into.
*9 - A rough estimate is 7-800 clocks per byte transferred, which gives a maximum transfer speed of around 50 KiB/s on a 4 MHz Z80. Which is about the range one can expect from a Z80 system anyway.
If that is too low, some external shift register and a some glue logic might be the way to go.
*10 - Using the RC2014 digital I/O at address 00h this might look a bit like this:
InPort: Integer = 0;
OutPort: Integer = 0;
MOSI: Byte = 1; /* Bit value for output data */
MISO: Byte = 1; /* Bit value for input data */
CLOCK: Byte = 2; /* Bit value for clock handling */
CS: Byte = 8; /* Bit value for chip select */
function SPI_SendRec(Value: Byte):Byte;
port[OutPort] := 0; /* Clear Clock, activate CS */
for i:= 1 to 8 do
bit := (Value and 128) shl 7; /* isolate next data bit */
Value := Value shl 1; /* prepare follow up bit */
port[OutPort] = bit; /* set MOSI to data bit */
port[OutPort] = bit+ CLOCK; /* Raise CLOCK */
rdval := rdval shl 1; /* Make room for input data */
rdval := rdval or (port[InPort] and 1); /* Move input bit */
port[OutPort] = bit; /* Clear Clock but keep data (*5) */
SPI_SendRec := rdval; /* Retrun what has been read */
/* Procedure to end a transaction */
/* or first time port initialisation */
port[OutPort] := CS + CLOCK + MOSI; /* disable operation */
var Dummy: Byte;
/* Program to set RFM95 into LoRa Mode */
SPI_Done; /* Initialize Port */
Dummy := SPI_SendRec(01); /* Register address */
Dummy := SPI_SendRec(128); /* Value to be set */
SPI_Done; /* End transaction */
(And yes, I was bored :))