Claude review: drm/imagination: Populate FW common context ID before passing to the FW

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

Patch Review

**Verdict: Good fix, minor nit**

This fixes a real bug. Before this patch, `ctx->ctx_id` was allocated via `xa_alloc()` *after* `pvr_context_create_queues()` which calls `pvr_queue_init_fw_context()`, which copies `ctx->ctx_id` into the firmware context at `pvr_queue.c:1097`:

```c
cctx_fw->server_common_context_id = ctx->ctx_id;
```

Since the struct is `kzalloc`'d, the firmware would always see `server_common_context_id = 0` instead of the actual allocated ID. Moving `xa_alloc()` earlier is the correct fix.

The cleanup of the old error path is also good. The old code had a comment about a race where another thread could take a reference on the context via `ctx_ids` between the `ctx_ids` allocation and the `ctx_handles` allocation failure. The new code's comment at `err_free_ctx_id` correctly explains why the erase is safe:

```c
/*
 * Ctx_id is not exposed to userspace and not visible yet within
 * the kernel/FW, plus a matching context handle (exposed to userspace)
 * hasn't been allocated yet, so it is safe to remove ctx_id
 * from the ctx_ids xarray.
 */
xa_erase(&pvr_dev->ctx_ids, ctx->ctx_id);
```

**Nit:** However, note that by moving `xa_alloc` for `ctx_ids` *earlier*, the window during which a partially-initialized context is visible in the `ctx_ids` xarray is now *larger*. If any code path does `xa_load(&pvr_dev->ctx_ids, id)` during this window, it would find a half-built context. The old code acknowledged this race. The commit message should mention why this is safe (e.g., "ctx_ids lookup only happens after the context handle is published to userspace").

---
Generated by Claude Code Patch Reviewer