Day 18 — Blocking I/O and Event-driven Driver¶
🎯 Goal¶
Understand how Linux drivers handle data flow using:
- blocking read
- wait queue
- poll
- event-driven design
🧠 What I Learned¶
1. Blocking ≠ automatic¶
Even without O_NONBLOCK:
- read() does NOT automatically block
- driver must implement wait_event()
2. Wait queue is the core¶
Used to connect:
- interrupt / workqueue
- user process
3. Proper event flow¶
IRQ → workqueue → enqueue → wake_up → read()
4. poll() integration¶
poll() relies on:
- poll_wait()
- wait queue
5. Event queue importance¶
- avoid losing events
- decouple producer / consumer
6. Lock design¶
- spinlock → event queue
- mutex → control plane
🛠 Implementation¶
- added enqueue / dequeue helper
- refactored workqueue handler
- implemented blocking read
- supported O_NONBLOCK
- verified poll behavior
🧪 Test Result¶
- blocking read works
- non-blocking returns -EAGAIN
- poll wakes correctly
- no busy loop
- debounce works with sysfs / ioctl
💡 Key Insight¶
A correct Linux driver is:
- event-driven
- not polling-based
- not busy-looping
🚀 Next¶
Move to real hardware driver:
- I2C sensor
- subsystem integration (IIO / hwmon)