Claude review: ufs: exynos: stop poking into struct phy guts

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

Patch Review

This is a behavior change that replaces `generic_phy->power_count` checks with a local `phy_powered_on` boolean. The commit message correctly identifies that reading `power_count` without holding `phy->mutex` is racy, and that a single consumer shouldn't rely on a shared refcount.

The implementation looks correct:

- `exynos_ufs_phy_init()`: sets `phy_powered_on = false` on power-down, `true` after successful power-on
- `exynos_ufs_exit()`: early return if `!phy_powered_on` — **minor concern**: the original code unconditionally called `phy_power_off()`/`phy_exit()`. Adding the early return changes behavior if `exynos_ufs_exit()` is called without a prior successful `phy_init()`. However, since `phy_powered_on` starts as `false` (zero-init of kzalloc'd struct), this should be safe and actually prevents potential underrun.
- `exynos_ufs_suspend()`: correctly gates `phy_power_off()` with `phy_powered_on`
- `exynos_ufs_resume()`: now checks error from `phy_power_on()` — good improvement

No issues. Well-reasoned change.

---
Generated by Claude Code Patch Reviewer