raspberrypi: wake core 0 from idle when core 1 posts USB host events

[mikeysklar]

On rp2, the TinyUSB host task only runs when core 0 drains the background callback queue. The PIO USB host runs on core 1 and posts its events there, but port_wake_main_task() was a no-op, so core 1 could not wake core 0 out of the WFI in port_idle_until_interrupt(). While a program is in time.sleep(), core 0 stays parked in that WFI for the whole sleep, so host events never get serviced. A device plugged in while a program sleeps can fail to enumerate at all.

Implement port_wake_main_task() as SEV and idle with WFE instead of WFI, so core 1 can wake core 0. SEVONPEND keeps the old wake on pending interrupt behavior. It lives in shared port.c, so it covers both rp2040 and rp2350. 12 lines of code.

Start this with nothing in the host port, then hot-plug a USB device while it runs:

import time
import usb.core

print("plug a device into the host port now")
while True:
    devices = list(usb.core.find(find_all=True))
    if devices:
        print("enumerated:", devices[0].product)
        break
    time.sleep(0.5)

On the current main it never prints enumerated, the device never shows up while the program sleeps (over 50 seconds):

With this PR it enumerates within one loop iteration:

plug a device into the host port now
enumerated: Macropad RP2040

Confirmed on a Fruit Jam (RP2350) and a Feather RP2040 USB Host (RP2040).

[dhalbert]

Did you try running something like Fruit Jam OS to smoke test general operation? Thanks if you can.

[mikeysklar]

Dan -

Loaded up Fruit Jam OS (v1.092) + patched with this PR commit. No issues with using the keyboard from boot to navigating the menus and opening apps. I was able to pull the keyboard and re-insert it multiple times. The keyboard continued to work.

>>> import os
>>> os.uname().version
'10.3.0-alpha.3-36-g691004d7b1-dirty on 2026-07-18'
>>> os.uname().machine
'Adafruit Fruit Jam with rp2350b'
>>>

[dhalbert]

Thanks for general testing. This looks OK. I don't know how I might have come up with the this fix, though. Hurrah for LLMs.

[mikeysklar]

The LLM added a lot of instrumentation and at one point looked at how espressif was doing port_wake_main_task() and noticed that only RP2 didn't have a mechanism for core 1 to wake core 0. Anyway...Thanks Claude (Fable 5). That goes away tomorrow.