Claude review: drm: bridge: dw_hdmi: Merge top and bottom half IRQ handlers

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

Patch Review

With HPD handling moved to `delayed_work`, the bottom half (threaded IRQ) only calls `mod_delayed_work()`, which is safe from hardirq context. The two handlers are merged into a single hardirq handler, and `devm_request_threaded_irq()` is replaced with `devm_request_irq()`.

```c
+	intr_stat = hdmi_readb(hdmi, HDMI_IH_PHY_STAT0) & HDMI_IH_PHY_STAT0_HPD;
 	if (intr_stat) {
+		hdmi_writeb(hdmi, ~0, HDMI_IH_MUTE_PHY_STAT0);
```

The muting (`~0` to MUTE register) followed by unmuting (`~HDMI_IH_PHY_STAT0_HPD`) within the same handler is correct — it prevents re-entry while the handler processes the interrupt. The `mod_delayed_work()` documentation confirms it is safe in atomic context.

One small note: the register reads and writes within the hardirq handler use `regmap_write`/`regmap_read` underneath (via `hdmi_writeb`/`hdmi_readb`). If the regmap uses a bus that can sleep (e.g., I2C), this would be problematic in hardirq context. However, dw-hdmi uses MMIO regmap, so this is fine. **No Reviewed-by yet.**

**Summary**: The series is well-decomposed and correct. The new v4 patches (12-17) are the ones that need review attention. The main risk area is patch 14's removal of RXSENSE handling, which changes the HPD detection behavior to be purely HPD-based with debouncing. This should work well for HDMI 1.4+ sinks but could potentially affect edge cases with older or non-compliant sinks that rely on RXSENSE for detection. The cover letter notes testing was done on Rockchip RK3328/RK3399/RK3568, and requests more testing with bridge connector users.

---
Generated by Claude Code Patch Reviewer