MCTP Over I2C: Migration to Kernel Qdisc Model

[chandu-tmba]

This proposal updates the current MCTP over I2C slave-only implementation to use the in-kernel MCTP networking and qdisc pipeline.This work replaces the legacy slave-mode handling with a unified kernel-routed transport, enabling consistent queuing, routing, and flow control across all MCTP interfaces.

Enhance the transport path to detect and propagate low‑level bus errors (e.g., NACK) from the I2C layer back into the qdisc and up to userspace. This prevents endpoint queues from stalling when a packet is never received and ensures that state machines, retries, and higher-level protocols can correctly handle failed transactions.

please check the flowcharts for our new proposal

[chandu-tmba]

please let us know your feeback on this

[jk-ozlabs]

As a starting point:

This proposal updates the current MCTP over I2C slave-only implementation

I'm not sure what this refers to?

[jk-ozlabs]

please let us know your feeback on this

I'm not sure what this solves. Can you elaborate with any issue you're currently seeing (which this addresses?)

Without knowing the issue, locking the qdisc for further message submission to the netdev's TX path doesn't appear to achieve anything. In the case of the current MCTP-over-I2C driver, it handles queue stop/start when our working queue is full. Implementing this at the qdisc would seem to prevent access to other (completely independent) endpoints on the same i2c bus?