Day 9 - GPIO Driver with gpiod and Device Tree¶

🎯 Objective¶

Learn how to control real hardware (GPIO) from a Linux kernel driver using:

Platform driver
Device Tree (DT)
Descriptor-based GPIO API (gpiod_*)
Character device interface (/dev/mygpio)

🧠 Key Concepts¶

1. From Legacy GPIO to gpiod¶

Old (deprecated):

gpio_request(17, "mygpio");
gpio_set_value(17, 1);

Problems: - GPIO number is not portable - Platform dependent - Hard to integrate with Device Tree

New (recommended):

gpiod = devm_gpiod_get(dev, "led", GPIOD_OUT_LOW);
gpiod_set_value(gpiod, 1);

Advantages: - No global GPIO number dependency - Works with Device Tree - Portable across platforms

2. Device Tree Binding¶

Device Tree defines hardware mapping:

mygpio_led {
    compatible = "myvendor,mygpio-led";
    led-gpios = <&gpio 17 0>;
};

Driver retrieves GPIO using:

devm_gpiod_get(dev, "led", GPIOD_OUT_LOW);

Mapping rule:

DT property	gpiod_get argument
`led-gpios`	`"led"`

3. Platform Driver Flow¶

Device Tree
    ↓
platform device
    ↓
driver probe()
    ↓
gpiod_get()
    ↓
character device (/dev/mygpio)

4. Character Device Interface¶

User space interacts via:

echo 1 > /dev/mygpio
cat /dev/mygpio

Driver implements:

write() → control GPIO
read() → read GPIO state

🔧 Implementation Highlights¶

write()¶

gpiod_set_value(mydev->led_gpiod, value);

Accepts 0 or 1
Controls GPIO output

read()¶

value = gpiod_get_value(mydev->led_gpiod);

Returns:

"0\n" or "1\n"

EOF Handling¶

if (*offset != 0)
    return 0;

Required for cat to terminate
Standard char driver pattern

🧪 Testing¶

Setup¶

./cleanup.sh
./rebuild.sh
./setup.sh

GPIO Control¶

echo 1 | sudo tee /dev/mygpio
echo 0 | sudo tee /dev/mygpio

GPIO Read¶

cat /dev/mygpio

Live Debug¶

sudo dmesg -w

🐞 Debug Notes¶

1. Legacy GPIO issue on Raspberry Pi 5¶

Using:

gpio_request(17, ...)

failed because:

GPIO number ≠ BCM number
Requires global GPIO index
Not reliable on modern platforms

2. "write() not working" false alarm¶

Observed:

No GPIO set to X in dmesg
Assumed write() was not called

Actual cause:

dmesg | tail missed logs
Driver was reloaded multiple times
Logs overwritten by probe/remove messages

3. System state inconsistency¶

Repeated:

dtoverlay
insmod / rmmod

caused:

Multiple probe/remove cycles
Confusing runtime state

4. Reboot fixed the issue¶

After reboot:

GPIO behavior returned to normal
Driver worked as expected

Conclusion:

Issue was caused by stale runtime state, not code logic.

📊 What I Learned¶

Modern Linux uses gpiod, not gpio_request
Device Tree defines hardware, driver should not hardcode GPIO
Platform driver + DT is the standard model
Char device bridges user space and kernel
Debugging requires careful log observation (dmesg -w)

🚀 Next Steps¶

Add interrupt support (GPIO button)
Learn IRQ handling
Explore poll / event-driven driver design

🧩 Summary¶

Day 9 marks the transition from:

"virtual driver"
→
"real hardware driver"

Key achievement:

✔ Successfully controlled GPIO via Linux kernel driver
✔ Integrated Device Tree + platform driver
✔ Implemented read/write interface
✔ Understood modern GPIO architecture