Claude review: drm/msm/a6xx: Add HFI support for CLX feature

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

Patch Review

**The HFI plumbing is correct.** The `a6xx_hfi_enable_clx()` / `a6xx_hfi_enable_iff_pclx()` pair follows the same pattern as existing `a6xx_hfi_enable_acd()`. The size calculation for the dynamically-allocated `a6xx_hfi_table` is correct, and `a6xx_hfi_send_msg()` fills in the header word.

**Issue: `a6xx_limits_tbl.tbl` should be `const`-qualified.**

In `a6xx_gpu.h`:
```c
struct a6xx_limits_tbl {
	/** @tbl: List of limits table **/
	struct a6xx_hfi_limits_tbl *tbl;
	...
};
```

The `tbl` pointer should be `const struct a6xx_hfi_limits_tbl *tbl`, since:
1. The `DECLARE_ADRENO_LIMITS_TABLE` macro creates a `static const struct a6xx_limits_tbl`, so the compiler needs the member to accept a pointer to const data.
2. The data is only ever read (`memcpy` source in `a6xx_hfi_enable_iff_pclx`), never written.

Without `const`, the assignment from the `static const` table data in patches 2/3 will produce a warning (discarding `const` qualifier) — or worse, silently strip the constness. This should be:
```c
const struct a6xx_hfi_limits_tbl *tbl;
```

**Minor: memcpy from const source is fine.** The `memcpy(&cmd, info->clx_tbl, sizeof(cmd))` in `a6xx_hfi_enable_clx()` correctly copies the const catalog data into a stack-local `cmd` before `a6xx_hfi_send_msg` mutates the header — good pattern.

**Ordering is correct.** CLX is enabled after ACD and before IFPC in `a6xx_hfi_start()`, which makes sense — current limiting should be established before entering IFPC.

---
Generated by Claude Code Patch Reviewer