Claude review: drm: bridge: Cadence: Add MHDP8501 DP/HDMI driver

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

Patch Review

This is the main new driver (3 source files, ~2000 lines). Overall well-structured with proper use of DRM atomic helpers.

**Issues:**

1. **IRQ vs. removal race**: The `removing` flag uses `WRITE_ONCE`/`READ_ONCE` but this is not sufficient to prevent the IRQ handler from scheduling `hotplug_work` between checking the flag and calling `disable_irq()` in remove. The sequence should be:
```c
// In remove():
WRITE_ONCE(mhdp->removing, true);
disable_irq(mhdp->irq[IRQ_IN]);
disable_irq(mhdp->irq[IRQ_OUT]);
cancel_delayed_work_sync(&mhdp->hotplug_work);
```
rather than relying on `READ_ONCE` in the IRQ handler to prevent scheduling.

2. **HDMI I2C adapter limitation**: The adapter only supports SCDC slave (0x54). While documented, the `i2c_algorithm` should return an error for unsupported addresses rather than silently failing. The code does check:
```c
if (msgs[0].addr != DDC_SEGMENT_ADDR && msgs[0].addr != DDC_ADDR)
```
but returns `-EINVAL` without explanation. A `dev_dbg` message would help.

3. **`dev_set_drvdata(dev, &mhdp->base)`**: The bridge driver stores `&mhdp->base` (not `mhdp` itself) via `dev_set_drvdata` so the PHY child can find it. This is clever but fragile — any caller using `dev_get_drvdata` must know to expect a `cdns_mhdp_base*`, not the full device struct. A comment at the `dev_set_drvdata` call site would prevent confusion.

4. **Debugfs firmware version**: The firmware version is read at debugfs open time via mailbox. If the firmware is not loaded/ready, this will fail silently or block. Consider caching the version after firmware ready is confirmed.

5. **HDMI infoframe handling**: `cdns_hdmi_config_infoframe()` silently returns if `len > 31` without reporting an error. The caller should know the configuration failed.

---
Generated by Claude Code Patch Reviewer