PIO push() does not show up in rx_fifo()

[ghfbsd]

Dear All -

Debugging a new PIO input program, I've encountered a puzzle where pin input does not show up when pushed to the rx_fifo. Here is the problem distilled to its essence:

# piotest.py
# PIO test program
   
from machine import Pin
from time import sleep, ticks_ms
from rp2 import StateMachine, PIO, asm_pio
   
@asm_pio()      
def test():
   nop()                # section delay 1000 ticks
   set(x, 31)   [6]
   label('a')
   nop()        [29]
   jmp(x_dec, 'a')                
   
   set(x, 31)   [7]     # section delay 1000 ticks
   label('b')
   nop()        [29]
   jmp(x_dec, 'b')

   mov(isr,invert(null))
   push(noblock)
   irq(0)               # data ready interrupt 
 
led = Pin('LED', Pin.OUT)
def ticker(irq, buf=bytearray(8)):
   led.toggle()
   n = sm.rx_fifo() 
   print('data %d' % n)
   if n > 0:
      for i in range(n):
         buf[i] = sm.get()
      print('got %d bytes: %s' % (n,buf[0:n].hex()))
      
sm = StateMachine(0, test, freq=2_000)
sm.irq(handler=ticker)
sm.active(1)
try:
   while True:
      sleep(1)
except KeyboardInterrupt:
   sm.active(0)
   PIO(0).remove_program()        # prevent ENOMEM on restart: PIO mem full

It produces the following output:

Entering REPL. Use Control-X to exit.
>
MicroPython v1.24.1 on 2024-11-29; Raspberry Pi Pico with RP2040
Type "help()" for more information.
>>> 
>>> execfile('piotest.py')
data 0
data 0
data 0
data 0
data 0
data 0
data 0
data 0
data 0
data 0
data 0
<keyboard interrupt>
>>>

My question is why does the pushed word not show up in the rx_fifo?

Signed, Puzzled PIO Programmer

[GitHubsSilverBullet]

Works here, but your timing is a bit off.

MicroPython v1.25.0 on 2025-04-15; Raspberry Pi Pico with RP2040

You should bracket your PIO code with wrap_target() and wrap()

@asm_pio()
def test2():
    wrap_target()           #───────────────────────────────────
    set(x, 30)         [12] # 13    section delay 2000 ticks
    label('loop')           #
    nop()              [31] # 32    31×(32+32)+13 ──▶ 1997
    jmp(x_dec, 'loop') [31] # 32
    mov(isr,invert(null))   #  1
    push(noblock)           #  1
    irq(0)                  #  1    1997+3 ──▶ 2000
    wrap()                  #───────────────────────────────────

def ticker(irq, buf=bytearray(8)):
    z = ticks_us()
    s = divmod(z, 1_000_000)
    mil,mic = divmod(z-s[0]*1000000, 1_000)
    print(f'Timing: {s[0]}s {mil}ms {mic}µs')
    led.toggle()
    n = sm.rx_fifo()
    print('data %d' % n)
    if n > 0:
        for i in range(n):
            buf[i] = sm.get()
        print('got %d bytes: %s' % (n,buf[0:n].hex()))

[ghfbsd]

Thanks for the feedback - after power cycling my pico, it now works for me too (odd, that).

Signed, Pleased PIO Programmer (but dismayed)

[GitHubsSilverBullet]

Thanks for the feedback - after power cycling my pico, it now works for me too (odd, that).

Leftover stuff in the PIO EXEC memory. Without bracketing, the code runs into those areas and hopefully (but uncontrolled) wraps at endaddress 32.

[ghfbsd]

It turns out the PIO push() failure happens after a DMA engine is configured to take input from the PIO state machine output. Even after the engine involved is explicitly close()d, the PIO push() no longer works.

[GitHubsSilverBullet]

Yeah … happens. Transaction finished? DMA counter down to ›0‹?
I have a habit to nuke DMA with a viper function, manually clearing all up once done.
And of course a subsystem reset at the program start.

[ghfbsd]

As a followup note, setting the appropriate bit(s) in the DMA_ABORT register for the running engine(s) before calling close() on them clears the problem with the PIO push() being ignored subsequently. It might be useful to incorporate this into micropython's DMA.close() code since the engine "can not be used" after this according to the docs.