Claude review: phy: qualcomm: add MSM8974 HDMI PHY support

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

Patch Review

Adds PLL rate calculation and programming to the MSM8974 (28HPM) HDMI PHY driver. This is the most algorithmically complex patch.

**Critical: Likely unit mismatch for `tmds_char_rate`**

The `struct phy_configure_opts_hdmi` defines `tmds_char_rate` in Hz (unsigned long long). However, the PLL calculation code treats it as kHz:

```c
unsigned long pixclk = opts->hdmi.tmds_char_rate;
...
if (pixclk == 297000) {
```

297000 Hz is 297 kHz, not the 297 MHz (297000000 Hz) that 4K HDMI requires. Either the comparisons should be against `297000000` (Hz), or `pixclk` should be explicitly divided: `pixclk = opts->hdmi.tmds_char_rate / 1000`. The VCO calculations later also appear to assume kHz:

```c
unsigned long vco_rate = pixclk * 10;
...
unsigned long rem = vco_rate % ref_clk;
```

If `pixclk` were in Hz (~148500000 for 1080p), then `vco_rate = pixclk * 10` would overflow a 32-bit intermediate on some platforms, and the divisor values would be wrong. This needs verification against the hardware reference.

**Good use of FIELD_PREP for register programming**

```c
writel(FIELD_PREP(UNIPHY_PLL_SDM_CFG0_BYP, !sdm_byp) |
       FIELD_PREP(UNIPHY_PLL_SDM_CFG0_BYP_DBLR, sdm_byp),
       pll_base + UNIPHY_PLL_SDM_CFG0);
```

This is much more readable than the raw magic number writes in the old driver.

**`of_device_get_match_data` in probe path**

Same issue as patch 1 — should use `device_get_match_data()`.

**PLL lock polling**

```c
while (!(readl(pll_base + UNIPHY_PLL_STATUS) & UNIPHY_PLL_STATUS_PLL_LOCK)) {
	udelay(1);
	if (timeout_us-- == 0)
		return -ETIMEDOUT;
}
```

This is a common pattern but `readl_poll_timeout()` would be the idiomatic kernel helper for this, providing consistent timeout behavior and reducing boilerplate.

---

**Summary**: The series is architecturally sound but has two critical issues (missing compatible strings for APQ8084/MSM8998, and a likely Hz/kHz unit mismatch in PLL calculations) that must be resolved before merge. The minor issues (typos, API preferences, code duplication) are worth fixing but non-blocking.

---
Generated by Claude Code Patch Reviewer