; RP2040 PIO program for accelerating SCSI communication
;
; Copyright (c) 2022 Rabbit Hole Computing™
;
; Run "pioasm scsi_accel.pio scsi_accel.pio.h" to regenerate the C header from this.
; GPIO mapping:
; - 0-7: DB0-DB7
; -   8: DBP
; Side set is REQ pin

.define REQ 19  ; was 17
.define ACK 26  ; was 10

; Delay from data setup to REQ assertion.
; deskew delay + cable skew delay = 55 ns minimum
; One clock cycle is 8 ns => delay 7 clocks
.define REQ_DLY 7

; Adds parity to data that is to be written to SCSI
; This works by generating addresses for DMA to fetch data from.
; Register X should be initialized to the base address of the lookup table.
.program scsi_parity
    pull block
    in NULL, 1
    in OSR, 8
    in X, 23

; Write to SCSI bus using asynchronous handshake.
; Data is written as 32-bit words that contain the 8 data bits + 1 parity bit.
; 23 bits in each word are discarded.
; Number of bytes to send must be multiple of 2.
.program scsi_accel_async_write
    .side_set 1

    pull ifempty block          side 1  ; Get data from TX FIFO
    out pins, 9                 side 1  ; Write data and parity bit
    out null, 23 [REQ_DLY-2]    side 1  ; Discard unused bits, wait for data preset time
    wait 1 gpio ACK             side 1  ; Wait for ACK to be inactive
    wait 0 gpio ACK             side 0  ; Assert REQ, wait for ACK low

; Read from SCSI bus using sync or async handshake.
; Data is returned as 32-bit words:
; - bit  0: always zero
; - bits 1-8: data byte
; - bit  9: parity bit
; - bits 10-31: lookup table address
; Lookup table address should be loaded into register Y.
; One dummy word should be written to TX fifo for every byte to receive.
.program scsi_accel_read
    .side_set 1

    pull block                  side 1  ; Pull from TX fifo for counting bytes and pacing sync mode
    wait 1 gpio ACK             side 1  ; Wait for ACK high
    in null, 1                  side 0  ; Zero bit because lookup table entries are 16-bit
    wait 0 gpio ACK             side 0  ; Assert REQ, wait for ACK low
    in pins, 9                  side 1  ; Deassert REQ, read GPIO
    in y, 22                    side 1  ; Copy parity lookup table address

; Data state machine for synchronous writes.
; Takes the lowest 9 bits of each 32 bit word and writes them to bus with REQ pulse.
; The delay times will be rewritten by C code to match the negotiated SCSI sync speed.
;
; Shifts one bit to ISR per every byte transmitted. This is used to control the transfer
; pace, the RX fifo acts as a counter to keep track of unacknowledged bytes. The C code
; can set the syncOffset by changing autopush threshold, e.g. threshold 3 = 12 bytes offset.
.program scsi_sync_write
    .side_set 1

    out pins, 9      [0]        side 1  ; Write data and parity bit, wait for deskew delay
    out null, 23     [0]        side 0  ; Assert REQ, wait for assert time
    in null, 1       [0]        side 1  ; Deassert REQ, wait for transfer period, wait for space in ACK buffer

; Data pacing state machine for synchronous writes.
; Takes one bit from ISR on every falling edge of ACK.
; The C code should set autopull threshold to match scsi_sync_write autopush threshold.
; System DMA will then move words from scsi_sync_write RX fifo to scsi_sync_write_pacer TX fifo.
.program scsi_sync_write_pacer
    wait 1 gpio ACK
    wait 0 gpio ACK   ; Wait for falling edge on ACK
    out null, 1       ; Let scsi_sync_write send one more byte

; Data pacing state machine for synchronous reads.
; The delay times will be rewritten by C code to match the negotiated SCSI sync speed.
; Number of bytes to receive minus one should be loaded into register X.
; In synchronous mode this generates the REQ pulses and dummy words.
; In asynchronous mode it just generates dummy words to feed to scsi_accel_read.
.program scsi_sync_read_pacer
    .side_set 1

start:
    push block      [0]      side 1  ; Send dummy word to scsi_accel_read, wait for transfer period
    jmp x-- start   [0]      side 0  ; Assert REQ, wait for assert time

finish:
    jmp finish      [0]      side 1

; Parity checker for reads from SCSI bus.
; Receives 16-bit words from g_scsi_parity_check_lookup
; Bottom 8 bits are the data byte, which is passed to output FIFO
; The 9th bit is parity valid bit, which is 1 for valid and 0 for parity error.
.program scsi_read_parity
parity_valid:
    out isr, 8                ; Take the 8 data bits for passing to RX fifo
    push block                ; Push the data to RX fifo
    out x, 24                 ; Take the parity valid bit, and the rest of 32-bit word
    jmp x-- parity_valid      ; If parity valid bit is 1, repeat from start
    irq set 0                 ; Parity error, set interrupt flag