Claude review: Qualcomm Sahara protocol enhancements.

[Claude Code Review Bot <claude-review@example.com>]

Overall Series Review

Subject: Qualcomm Sahara protocol enhancements.
Author: Kishore Batta <kishore.batta@oss.qualcomm.com>
Patches: 10
Reviewed: 2026-03-22T04:44:33.268433

---

This 9-patch series moves the Qualcomm Sahara protocol driver from the QAIC accelerator tree to `drivers/bus/mhi/sahara/`, generalizes device matching, adds QDU100 support, and introduces DDR training data capture via Sahara command mode with sysfs exposure.

**The architecture is sound** — making the Sahara driver a standalone MHI module is the right approach for reuse. However, the series has several significant bugs that need to be fixed before merging:

1. **Bug in `qaic_init()` error handling** (Patch 2) — goto labels are swapped, leading to unregistering drivers that were never registered.
2. **Unsigned comparison always-false** (Patch 7) — `le32_to_cpu()` returns `u32`, comparing `< 0` is a no-op.
3. **Sleeping in atomic context** (Patch 7) — `mutex_lock()` and `devres_alloc(GFP_KERNEL)` called from MHI DL callback which runs in tasklet context.
4. **Data races** (Patch 7) — `context->is_cmd_mode`, `receiving_trng_data`, etc. accessed from both DL callback and work queue without synchronization.
5. **Unbounded allocation from device-controlled size** (Patch 7) — `kzalloc(resp_len)` with no upper bound check.

The documentation patch (1/9) has numerous typos and copy-paste errors. The cover letter changelog for v4 still says "EDITME".

---

---
Generated by Claude Code Patch Reviewer