openipc_frame_ts: strobe-out — drive a GPIO synchronously with frame events

[widgetii]

Summary

Adds a strobe-out feature to the existing openipc_frame_ts module: drive a SoC GPIO at the exact moment the sensor begins streaming row 0 (MIPI_FS) and/or finishes the last row (ISP_FEND). The expensive part — being inside the kernel at sub-µs precision relative to vsync — already exists from PR #178. This PR adds ~260 lines so that hook can also flip a GPIO without spinning up a userspace daemon.

Two modes:

Mode	Behaviour	Use case
pulse	Assert on MIPI_FS, deassert via hrtimer after strobe_pulse_us µs	Flash / strobe light aligned to sensor row-0 start
hold	Assert on MIPI_FS, deassert on ISP_FEND	Ring light that should be on only while pixels are being received (high window = sensor readout duration, ~10–30 ms depending on mode)
off	No GPIO activity	Default

Configured via sysfs:

/sys/class/misc/openipc-frame-ts/
├── strobe_gpio         (rw int)   GPIO number, -1 = off (default)
├── strobe_mode         (rw str)   off | pulse | hold
├── strobe_pulse_us     (rw uint)  pulse width µs, 1..100000 (default 1000)
├── strobe_active_low   (rw bool)  invert polarity (default 0)
└── strobe_events       (r  u64)   counter of fires since module load

Constraints

• The GPIO must be SoC-direct (memory-mapped). GPIOs that can sleep (I2C / SPI expanders) are rejected at configure time — the firing path is hardirq.
• One global strobe slot for now; per-channel can be added later by indexing the state by vi_chn if anyone needs it.
• Latency from sensor vsync to GPIO assert: a few µs. Pulse-width precision: hrtimer-grade.
• Pure software strobe — physical GPIO must be exposed on the camera board (most IP-camera designs have at least one — the IR-cut filter / day-night sensor pin).

Test plan — validated on hardware

Both lab boards re-flashed with the new module:

ev300 (V4, IMX335) — /tmp/strobe_stress.sh:

• 10 s of pulse mode at strobe_pulse_us=50: 276 strobe_events fired
• Invalid inputs rejected: gpio=9999, pulse_us=0, mode=bogus all returned EINVAL
• dmesg clean: no relevant kernel messages
• echo -1 > strobe_gpio cleanly releases the GPIO

av300 (cv500, IMX415):

• 5 s of pulse: 56 events
• Switch to hold mode while running: another 75 events in 5 s
• Clean teardown

Why bother

The robotics-camera audience documented in the new wiki page (proposed here — not yet merged) lists "hardware trigger / strobe out" as one of the few capabilities that genuinely separate sub-$50 OpenIPC cameras from $500+ industrial gear. With this module the strobe-out half of that gap closes:

• LED ring driven only during exposure (saves power, reduces heat, removes ambient artefacts)
• Camera–IMU calibration: hardware-correlated pulse on every frame
• Triggering downstream measurement equipment at frame rate
• Active-illumination perception (laser line, structured light) gated by the camera's sensor window rather than a free-running timer

Trigger-IN (camera waits for an external pulse before each frame) is a separate, harder problem requiring per-sensor slave-mode driver work; out of scope for this PR.

🤖 Generated with Claude Code