RP2350 PIOs - Help Getting Data from the DMA

[TheFargo]

It's me again, having more troubles with the DMA system! I appreciate everyone's help.
I'm trying to set up a DMA channel that'll feed data to a PIO's TX-FIFO, which should be pretty straightforward.

In my sample code below I've got a PIO that takes in two pulls of data to blink the on-board LED - you can send it an "on" delay and an "off" delay. The PIO definitely works if the CPU manually put()s the data into the FIFO, as demonstrated in the test below. But the DMA channel remains silent. I'm certain it's a formatting or setup issue, but I haven't been able to get a clean transfer. I've tried setting the write address to the state machine object (which should work, according to the documentation) or to the physical address of the FIFO (0x50200010 unless I'm wrong), and nothing seems to kick off the transfer. Am I missing something on the DMA side? The PIO setup? Thanks in advance!

##### DMA EXAMPLE #####

import time
from machine import Pin
import rp2
import array
import uctypes

##### CONFIGURATION DATA #####

LED_PIN = 25

DMA_BASE = 0x50000000   # Physical location of the DMA in memory
PIO0_SM0_TX = 0x50200010 # PIO0, State Machine 0 TX-Fifo

##### PIO PROGRAMS #####

@rp2.asm_pio(set_init=rp2.PIO.OUT_LOW)
def blink_loop():
    """
    Continuously input pairs of data: Light on pause length, light off pause length.  
    """
    wrap_target()
    #### Get Data 1 ######################################################### 
    pull(block)                     # Pull "on" count from TX-FIFO
    mov(x, osr)                     # Store count 
    #### LED ON #############################################################
    set(pins, 1)                    # Flip the LED pin ON
    label("on_loop")                # Loop X number times
    jmp(x_dec, "on_loop")  [6]      # Count 7 cycles for each loop (jmp + 6)
    #### Get Data 2 #########################################################
    pull(block)                     # Pull "off" count from TX-FIFO
    mov(x, osr)                     # Store count
    #### LED OFF ############################################################
    set(pins, 0)                    # Flip the LED pin OFF
    label("off_loop")               # Loop X number times
    jmp(x_dec, "off_loop")  [6]     # Count 7 cycles for each loop (jmp + 6)
    #### LOOP ENDS ########################################################## 
    wrap()                          # Off cycle complete; return to loop


##### Initialization #####

# Prepare DMA Data 
blink_data = array.array('L', [6000000,3000000])
blink_data_addr = uctypes.addressof(blink_data)
# Allocate DMA channel
dma_blink_data = rp2.DMA()
# Pack our Control Word
dma_blink_data_ctrl = dma_blink_data.pack_ctrl(
    treq_sel=0,                  # DREQ for PIO0, SM0
    inc_read=True,               # Increment read address each time you send
    inc_write=False,             # Write address stays the same (the Pio TX-FIFO)
    size=2,                      # 4-byte transfers
    enable=True)

# Attach communication pins
LED_output_pin = Pin(LED_PIN, Pin.OUT)

# Set up the state machines
sm_blinker = rp2.StateMachine(0, blink_loop, freq=150000000, set_base=LED_output_pin)

##### MAIN PROGRAM #####

print ("Begin Test!")
sm_blinker.active(1)
print (f"  Blinker FIFO before DMA: {sm_blinker.tx_fifo()}")

# Begin data transfer:
dma_blink_data.config(
    read=blink_data_addr,     # Our data address
    write=sm_blinker,         # State machine object. This should go to the TX-FIFO.
    count=2,                  # Two chunks of data
    ctrl=dma_blink_data_ctrl, # Control word configured above
    trigger=True)             # Start immediately

print (f"  Blinker FIFO After DMA: {sm_blinker.tx_fifo()}")

time.sleep(2)

print ("Manually placing data:")
sm_blinker.put(6000000)
sm_blinker.put(3000000)
print (f"  Blinker FIFO After Manual Put(): {sm_blinker.tx_fifo()}")

time.sleep(10)

sm_blinker.active(0)
print ("Test Complete.")

[TheFargo]

Been doing some experimenting.
Adding "Autopull=true" in the Pio assembly decorator does not fix my DMA, but it does appear to make the manual data entry "loop" - it keeps pulling the same data over and over, so the pin blinks. So it seems like that data just stays there in the FIFO buffer when I do that. (Not sure what this has to do with my problem, but maybe it's useful to know.)

[GitHubsSilverBullet]

It works. Probably not like you expected, but it does work.
Changes to illustrate:
blink_data = array.array('L', [6000000,3000000]*10)

# Begin data transfer:
dma_blink_data.config(
    read=blink_data_addr,     # Our data address
    write=sm_blinker,         # State machine object. This should go to the TX-FIFO.
    count=len(blink_data),       
    ctrl=dma_blink_data_ctrl, # Control word configured above
    trigger=True)             # Start immediately

while dma_blink_data.count:
    time.sleep_ms(10)
while sm_blinker.tx_fifo():
    print (f"  Blinker FIFO After DMA: {sm_blinker.tx_fifo()}")
    time.sleep_ms(10)

[TheFargo]

Interesting, your example code hangs here:

while dma_blink_data.count:
    time.sleep_ms(10)

The only thing I can see changed in your example code is that we put more data into the array and tried to send more data. In either example, it's not seeing it. What am I doing wrong? Is it here:

@rp2.asm_pio(set_init=rp2.PIO.OUT_LOW)

Or is it when I initialize the state machine?

sm_blinker = rp2.StateMachine(0, blink_loop, freq=150000000, set_base=LED_output_pin)

My DMA code is identical to your example.

[GitHubsSilverBullet]

Especially when toying with DMA code, one should always do a hard reset before trying the next iteration.
Or do a proper DMA subsystem reset early in your code.

[TheFargo]

Ahhh, the code was working after a hard reset. Thank you! I'm certain this was causing me all sorts of headaches!

Is there a way to hard reset the DMA controller through MicroPython code? I don't see anything in machine or rp2. Thanks!

[GitHubsSilverBullet]

Either Viper code or mem32 and the appropriate registers pulled from the datasheet. Very simple, actually.
Be cautious, rp2040 and rp2350 do it slightly different.

[GitHubsSilverBullet]

Actually that's better ---> (Viper code or mem32) and ...
AND binds stronger than OR ;-)