i2c: fix slave clock-stretch gate and master STOP-settle race The slave_read drain re-armed the receive path unconditionally on every call, with no concept of whether the consumer had actually taken the previous frame -- a frame arriving before the client drained the prior one just overwrote it in the server's one-frame latch. Gate the re-arm on a per-frame credit the consumer grants by draining, so the controller clock-stretches a slow client instead of losing data. write_buffer_mode returned as soon as PKT_DONE was observed, but the packet state machine (I2CM14[31:28]) was still reading STOP at that moment -- the controller was still physically driving the stop condition. The next write's command register landed on a busy controller and was silently dropped. Add a settle loop that waits for the state machine to return to idle before returning. Reading back what DMA wrote requires invalidating the CM4F's write-through cache first, which touches the SCU directly. Map scu_regs into every i2c_server consumer of this driver, not just the one that exercised the path when this was written -- a config that lacked the mapping would MemoryManagement fault the moment a slave frame actually arrived.
The OpenPRoT Technical Charter can be found at https://github.com/OpenPRoT/.github/blob/main/GOVERNANCE.md
NOTE: We are converting our build system to bazel. We recommend installing bazelisk to automatically manage bazel versions.
You can run tasks using the Pigweed workflow launcher pw or bazel.
./pw presubmit - Run presubmit checks: formatting, license checks, C/C++ header checks and clippy../pw format - Run the code formatters.bazel test //... - Run all tests.bazel build //docs - Build documentation.The project is structured as a bazel module.
No additional tools are required - all dependencies are managed by bazel.