Backport commit e968767: Fix I2C HID command to use write_read for response commands (#867)

[jerrysxie]

Per the HID over I2C Protocol Spec v1.0, commands that require a response (GetReport, GetIdle, GetProtocol) must be performed as a single I2C transaction with a repeated START condition (no STOP between the write and read phases).

The previous implementation used separate write and read operations, which inserted a STOP condition between them, violating the spec. Use write_read with a separate stack-allocated command buffer (max 7 bytes for these commands) so the command is sent and the response is read in a single I2C transaction.

Also simplify the data_reg conditional logic in both branches to remove dead-code conditions.

Assisted-by: GitHub Copilot:claude-opus-4.6

[Copilot]

Pull request overview

Summary of changes

This PR updates the I2C HID device command path to comply with the HID over I2C Protocol v1.0 requirement that response-producing commands (GetReport/GetIdle/GetProtocol) be executed as a single I2C transaction using a repeated START (i.e., no STOP between write and read). It replaces the previous “write then read” sequence with a single write_read call, using a small stack buffer for the encoded command to avoid reusing the shared response buffer for the write phase. It also simplifies the data_reg selection logic by removing conditions that were effectively dead for certain opcode categories.

Changes:

• Switch response commands to write_read (repeated START) using a stack-allocated command buffer.
• Keep non-response commands as write only, with simplified data_reg conditional logic.

Step-by-step review guide

1. Response commands now use a single I2C transaction (write_read)

   • The opcode.has_response() branch encodes the command into temp_w_buf and performs bus.write_read(address, cmd_bytes, buf) under a timeout.
   • The fixed-size [u8; 7] buffer is consistent with existing command serialization tests (GetReport/GetIdle with registers can reach 7 bytes; GetProtocol with registers is 6).

2. Non-response commands remain write

   • The else branch continues to encode directly into the shared buffer and uses bus.write(...).
   • The data_reg inclusion is now based solely on opcode.requires_host_data() in this branch, reducing unnecessary branching.

3. Timeout semantics changed for response commands

   • Previously, the read phase used data_read_timeout; now the combined transaction uses device_response_timeout. This changes behavior and may cause spurious timeouts on large/slow feature reads.

Potential issues

#	Severity	File	Description	Code
1	🟡 Medium	hid-service/src/i2c/device.rs:214-216	write_read for response commands uses device_response_timeout, but feature-data reads were previously governed by data_read_timeout (and Config docs define it that way). This can cause premature timeouts for legitimate response reads.	with_timeout(self.timeout_config.device_response_timeout, bus.write_read(...))