Claude review: soc/tegra: pmc: Create PMC context dynamically

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

Patch Review

This is the most substantial patch. It renames the global `pmc` to `early_pmc`, allocates a new `struct tegra_pmc` dynamically in `tegra_pmc_probe()`, and extracts a `tegra_pmc_init_common()` helper shared between early init and probe.

The `tegra_pmc_detect_tz_only()` function is moved from `__init` to non-`__init` since it's now called from both early init and probe. This is correct.

In `tegra_pmc_probe()`, the new dynamic PMC's `base` is remapped via `devm_platform_ioremap_resource()`, and the `early_pmc->base` is updated to point to the new mapping (under `powergates_lock`). The early ioremap'd region is freed. This ensures legacy callers that still go through `early_pmc` continue to work.

One observation: `tegra_pmc_init_common()` unconditionally calls `tegra_pmc_detect_tz_only()`, but in the early init path, the original code only did so when `pmc->soc->maybe_tz_only` was true:

```c
-		if (pmc->soc->maybe_tz_only)
-			pmc->tz_only = tegra_pmc_detect_tz_only(pmc);
```

The new `tegra_pmc_init_common()` does not check `maybe_tz_only`:

```c
+static void tegra_pmc_init_common(struct tegra_pmc *pmc)
+{
+	pmc->tz_only = tegra_pmc_detect_tz_only(pmc);
```

This means on SoCs where `maybe_tz_only` is false, the scratch register probe-by-write will now be executed unnecessarily. For most SoCs this is harmless (the write-readback test is benign), but it's a behavioral change worth noting. If scratch0 is protected on some SoC where `maybe_tz_only` was intentionally false, this could cause unexpected side effects.

The whitespace fix aligning `"scratch"` in the platform_get_resource_byname call is fine as a drive-by cleanup.

Also, `tegra_pmc_sync_state()` correctly retrieves the dynamically allocated PMC via `dev_get_drvdata()` instead of using the old global.

---
Generated by Claude Code Patch Reviewer