[dylanneve1]

Adds docs/AUDIO.md: how iOS 14 on the iPhone 11 plays audio (MCA I2S clusters over SIO DMA, CS42L77 codec on SPI3, CS35L27 amp on I2C2, AOP audio service), which kexts match which device-tree nodes, what Inferno already models (MCA/codec/amp models exist upstream, but the DT nodes are deleted), the Android backend choice, work items and effort.

Recommendation: an AAudio driver inside QEMU (-audiodev aaudio), not a PCM socket to AudioTrack. Since #66 the app can run QEMU in-process, so the 'everything over unix sockets' design only applies to the child-process path. AAudio works the same in both modes, keeps PCM in one process and has the lowest latency. If AAudio turns out not to open from the exec'd child, the socket + AudioTrack route is the fallback for that mode.

Effort: about 1.5–3 weeks to hear the speaker on a Pixel. Most of the uncertainty is in milestone 2 (how much of the AOP speaker path iOS needs).

• Milestone 1: t8030: opt-in audio hardware (audio=on, audiodev=) pyre#24 (draft). Drivers probe with audio=on, nothing played yet.
• Issues: Audio M1: expose the t8030 audio hardware behind audio=on #91 Audio M2: make iOS play the speaker route into the MCA #92 Audio M3: AAudio audio driver for QEMU (-audiodev aaudio) #93 Audio M4: build + app wiring for guest audio #94 Audio M5: pacing and underruns on the phone #95 Audio M6 (later): microphone input #96.

Docs only. No Apple firmware or binaries in this PR.

🤖 Generated with Claude Code