[ehb]

A sound card for S/PDIF output straight from GPIO21 (PCM_DOUT), without a transceiver chip:
a TOSLINK module or a coaxial driver on header pin 40 is all the hardware. As suggested on
the forum (https://forums.raspberrypi.com/viewtopic.php?p=2392395), this is an overlay of its
own instead of borrowing hifiberry-dac.

The overlay only wires up what the kernel already has: the generic linux,spdif-dit codec on a
simple-audio-card, the Pi as bit and frame clock producer, 2 x 32-bit slots per frame, and
GPIO21 alone switched to its I2S function, so GPIOs 18-20 stay free. The S/PDIF line coding
(biphase-mark, at twice the audio rate as S32_LE) happens in userspace; the README entry says
so, as plain PCM played to this card does not produce audio. The userspace side is an ALSA
plugin behind alsa-lib's iec958 plugin: https://github.com/ehb/rpi-gpio-spdif

The same overlay should work on a Pi 5: the pin group uses the legacy brcm,pins/brcm,function properties, which the RP1 pin driver maps (GPIO21 ALT0 to i2s0).

Tested on a Pi 4B (LibreELEC 12.2.1, kernel 6.12), applied with dtoverlay at runtime (also on
every boot, by a service before the media player starts; not through config.txt):

• the card appears as spdifgpio21, GPIO18-20 stay inputs, GPIO21 is PCM_DOUT
• 44.1/48/96 kHz audio (I2S at 88.2/96/192 kHz) plays; a receiver decodes PCM, AC3 and DTS
  passthrough
• lip sync checked with a flash/click test video

Checked, not run: both overlays and overlay_map built from this branch (cpp + dtc -@) and
applied with fdtoverlay to the base trees of Pi 1 B+, Zero W, Zero 2 W, Pi 3 B, Pi 4 B and
Pi 5 B, then read back (pins, clock producer, card). The Pi 5 variant has not run on
hardware; I have no Pi 5. Feedback or a test by someone with a Pi 5 is welcome.

[pelwell]

I wouldn't be surprised if your non-pi5 overlay works on Pi 5 as is - the pinctrl/GPIO driver for RP1 has support for the legacy brcm, properties, mapping the ALT functions where possible - but I can try both.

[ehb]

Good point, thanks. With the RP1 legacy mapping it gets simpler, so I've reworked it: the overlay now adds its own pin group under &gpio (brcm,pins 21, brcm,function ALT0) and points the I2S block at it, instead of rewriting i2s_pins (which the Pi 5 tree doesn't have). That makes it one overlay for all models; the -pi5 variant and the overlay_map entry are gone. Applied to the Pi 1/Zero/3/4/5 base trees it resolves everywhere, and on my Pi 4 pin 21 is now owned by the I2S block with 18-20 unclaimed. A run on a Pi 5 would be very welcome.