Claude review: gpu: nova-core: send UNLOADING_GUEST_DRIVER GSP command upon unloading

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

Patch Review

This is the core patch. It adds:

1. **Bindings** (`gsp/fw/r570_144/bindings.rs`): Adds `rpc_unloading_guest_driver_v1F_07` struct.

2. **Firmware command wrapper** (`gsp/fw/commands.rs`):
```rust
+pub(crate) struct UnloadingGuestDriver(bindings::rpc_unloading_guest_driver_v1F_07);
+
+impl UnloadingGuestDriver {
+    pub(crate) fn new(suspend: bool) -> Self {
+        Self(bindings::rpc_unloading_guest_driver_v1F_07 {
+            bInPMTransition: u8::from(suspend),
+            bGc6Entering: 0,
+            newLevel: if suspend { 3 } else { 0 },
+            ..Zeroable::zeroed()
+        })
+    }
+}
```

The magic number `3` for `newLevel` when suspending appears to correspond to a D3 power state. This matches what OpenRM does (`NV2080_CTRL_GPU_SET_POWER_STATE_GPU_LEVEL_3`). A named constant would be nice but not critical since this is matching hardware/firmware protocol values.

3. **GSP command** (`gsp/commands.rs`): Implements `CommandToGsp` and `MessageFromGsp` for the new command. The reply type `UnloadingGuestDriverReply` has `type Message = ()` — the reply payload is empty/ignored, which is correct for this RPC.

4. **Shutdown logic** (`gsp/boot.rs`):
```rust
+    fn shutdown_gsp(
+        cmdq: &Cmdq,
+        bar: &Bar0,
+        gsp_falcon: &Falcon<Gsp>,
+        suspend: bool,
+    ) -> Result<()> {
+        cmdq.send_command(bar, commands::UnloadingGuestDriver::new(suspend))?;
+
+        const LIBOS_INTERRUPT_PROCESSOR_SUSPENDED: u32 = 0x8000_0000;
+        read_poll_timeout(
+            || Ok(gsp_falcon.read_mailbox0(bar)),
+            |&mb0| mb0 == LIBOS_INTERRUPT_PROCESSOR_SUSPENDED,
+            Delta::from_millis(10),
+            Delta::from_secs(5),
+        )
+        .map(|_| ())
+    }
```

The 5-second timeout with 10ms polling interval seems reasonable. The `LIBOS_INTERRUPT_PROCESSOR_SUSPENDED` constant matches what OpenRM uses (`NV_LIBOS_INTERRUPT_PROCESSOR_SUSPENDED`).

5. **Integration** (`gpu.rs`):
```rust
+        let _ = kernel::warn_on_err!(self.gsp.unload(dev, bar, &self.gsp_falcon));
```

The `let _ =` discard after `warn_on_err!` means if unload fails, we continue with `sysmem_flush.unregister()`. This is the right approach for an unbind path — best-effort teardown.

**Minor observation:** `shutdown_gsp` is a static method (`&Cmdq` parameter instead of `&self`) while `unload` takes `&self`. This is fine since `shutdown_gsp` is a private helper, but worth noting that `cmdq` comes from `self.cmdq` in the caller.

No blocking issues.

---
Generated by Claude Code Patch Reviewer