Weird "Uncaught exception in IRQ callback handler" when using 32bit values via PIO's FIFO

[mungewell]

I am using a 'raw' register to hold 32bit of information. When this is accessed in my IRQ handler I get an exception.

MPY: soft reboot
Pico-Timecode v2.1+
www.github.com/mungewell/pico-timecode
RX: 00:00:00:00
Uncaught exception in IRQ callback handler

The IRQ code reads the value from a FIFO, and if the upper 2 bits are set I get the error. The value is being passed to PIO block's out FIFO and returned via in FIFO.

def irq_handler(m):
    global eng, stop
    global tx_raw, tx_ticks_us, rx_ticks_us, sl_ticks_us
    global core_dis

    core_dis[machine.mem32[0xd0000000]] = machine.disable_irq()
    ticks = utime.ticks_us()

    if m==eng.sm[0]:
        sl_ticks_us = ticks

    if m==eng.sm[1]:
        if eng.sm[0].rx_fifo() > 0:
            tx_raw = eng.sm[0].get()

Trigger can be manually set by forcing the bits...

        # Wait for TX FIFO to be empty enough to accept more
        while eng.mode <= MONITOR and eng.sm[2].tx_fifo() < 5:
            if eng.sm[0].tx_fifo() < 3:
                eng.sm[0].put(eng.tc.to_raw() | 0x80000000)     # ??? IRQ issue
            for w in eng.tc.to_ltc_packet(send_sync, False):
                eng.sm[2].put(w)
            eng.tc.release()
            send_sync = (send_sync + 1) & 1

The PIO block is just returning what it is sent... I am using this to keep track of which frame is transiting through the various FIFO's on multiple PIO blocks.

@rp2.asm_pio(autopull=True, autopush=True)

def auto_start():
    irq(clear, 4)                   # immediately Trigger Sync...
    
    label("halt")         
    out(x, 32)                      # will 'block' waiting for TX timecode
    in_(x, 32)                      # and then write back into FIFO
    jmp("halt") [31]

Any suggestions?

[mungewell]

Occurs with MicroPython 1.22, 1.23, 1.24 and a recent (self built) 1.25.

This script can run standalone on stock hardware, and if the line is changed to set upper bits will cause issue:
https://github.com/mungewell/pico-timecode/blob/main/pico_timecode.py#L953

[mungewell]

Just for clarity, the ISR and the FIFO access work correctly if the MSBs are NOT set. The correct value is pulled from the FIFO and put into the global variable tx_raw.

I do have following, but unsure how to access further information on the exception.

import micropython
micropython.alloc_emergency_exception_buf(100)

[mungewell]

Maybe this... Unsigned to Signed?
https://github.com/micropython/micropython/blob/master/ports/rp2/rp2_pio.c#L820

        if (dest == NULL) {
            return mp_obj_new_int_from_uint(value);
        }

[mungewell]

Simplified example, actually gives more information (presumably as it's not threaded).
uncaught_exception_test.py.txt

MPY: soft reboot
Uncaught exception in IRQ callback handler
Traceback (most recent call last):
  File "<stdin>", line 33, in irq_handler
MemoryError: memory allocation failed, heap is locked
0x00000000
0x00000000

In the working case (MSB not set) I get:

MPY: soft reboot
0x12345678
0x12345678

[mungewell]

reported as bug:
#16221