Mirrors_Framework/rt-thread

Fork 0

mirror of https://github.com/RT-Thread/rt-thread.git synced 2026-08-18 20:48:41 +08:00

Files

T

GuEe-GUI cfda3b3d1a

doc_doxygen / doxygen_doc generate (push) Has been cancelled

Details

doc_doxygen / deploy (push) Has been cancelled

Details

[docs] device-driver: add DM subsystem docs and expand INDEX

Add Doxygen pages for RT-Thread device-model drivers: platform/OFW,
DM core (bus, dm, power), clk/regulator/reset/power-domain, PCI/PIC,
Phye, DMA/hwcache, block/SCSI/ATA/NVMe/UFS/SDIO, SCMI, RPMsg,
mailbox, NVMem, NUMA, syscon, thermal, input/LED/IIO/graphic, and
related power/charger/supply docs.
Expand device-driver INDEX.md with ~90 @subpage entries so all new
pages appear in the driver chapter navigation.
Update existing UART/SPI/RTC/pin/framework/dtc pages: UART docs moved
from serial/ to uart/ (uart + uart_dm + uart_earlycon); SPI and RTC
gain DM companion pages; pin links to pin_dm; framework points at DM
topics.
Focus on registration/probe flow and in-tree APIs; no standalone
VirtIO subsystem page (transport code still incomplete).
Test plan: build Doxygen for documentation/ and verify device-driver
INDEX links resolve without missing @ref/@subpage warnings.

Signed-off-by: GuEe-GUI <2991707448@qq.com>

2026-05-28 16:09:55 +08:00

3.1 KiB

Executable File

Raw Blame History

@page page_device_input Input subsystem

Device-tree binding and DM registration: @ref page_device_input_dm.

Topic	Page
Userspace API	@ref page_device_input_uapi
Touch class bridge	@ref page_device_input_touch
Power / restart keys	@ref page_device_input_power

Input subsystem (API)

Header: drivers/input.h. Typical stack: low-level driver → rt_input_event / reporters → handlers (GUI, shell) or /dev/inputN (@ref page_device_input_uapi).

Path	API
In-kernel listener	`rt_input_add_handler`
POSIX process (Smart)	`open` / `read` / `ioctl` / `poll` on `/dev/inputN` — @ref page_device_input_uapi
Legacy touch GUI	`struct rt_touch_device` — @ref page_device_input_touch

When to use the input core

Use *`rt_input_`** when…	Consider raw `rt_device` read when…
You want Linux-like `EV_KEY` / `EV_REL` / `EV_ABS` / multitouch slot semantics and multiple listeners.	You have a single proprietary app talking to a vendor IOCTL-only touch stack.
You need polling (`rt_input_setup_polling`) for GPIO keys without IRQ storms.	GPIO IRQ latency is fine and you do not need unified event routing.

Event contract

rt_input_set_capability before rt_input_device_register so users know which type/code pairs are valid.
For each hardware change, call rt_input_report_* then rt_input_sync to close the frame (SYN_REPORT). Missing sync merges events poorly for consumers.
rt_input_trigger routes through optional idev->trigger—use when you wrap another input device; otherwise rt_input_event is enough.

Callback / context rules

rt_input_add_handler → callback: invoked from rt_input_event path—assume thread or workqueue unless your driver documents IRQ delivery; keep callback short.
poller->poll: runs on timer tick—must not block or take mutexes that can deadlock with the same thread calling rt_input_sync.
Touch helpers: rt_input_setup_touch / rt_input_report_touch_* require RT_INPUT_TOUCHSCREEN — @ref page_device_input_touch.

Registration lifecycle

rt_input_device_register: registers underlying rt_device; failures leave partially init state—clean up partial setup in error paths.
rt_input_remove_config: tears down polling/touch dynamic config before unregister if you toggled features at runtime.

Pitfalls

ABS without rt_input_set_absinfo: consumers cannot clamp—set min/max/resolution for axes you report.
Key autorepeat: not automatic unless a layer implements it—do not assume Linux repeat behavior.
Re-entrancy: posting events from inside a handler for the same device can recurse—prefer queueing.

3.1 KiB Executable File Raw Blame History