From mboxrd@z Thu Jan  1 00:00:00 1970
From: Claude Code Review Bot <claude-review@example.com>
To: dri-devel-reviews@example.com
Subject: Claude review: gpu: nova-core: add non-sec2 unload path
Date: Sun, 12 Apr 2026 11:19:37 +1000
Message-ID: <review-patch3-20260409-b4-blackwell-unload-v1-3-0f5a2ff838dd@nvidia.com>
In-Reply-To: <20260409-b4-blackwell-unload-v1-3-0f5a2ff838dd@nvidia.com>
References: <20260409-b4-blackwell-unload-v1-0-0f5a2ff838dd@nvidia.com> <20260409-b4-blackwell-unload-v1-3-0f5a2ff838dd@nvidia.com>
X-Mailer: Claude Code Patch Reviewer
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0

Patch Review

**Purpose:** Split the `unload()` function so that SEC2-based architectures=
 run the full FWSEC-SB + booter unloader sequence, while non-SEC2 architect=
ures (Hopper/Blackwell using FSP) just wait for the GSP falcon to halt.

```rust
+        if chipset.arch().uses_sec2() {
+            /* Run FWSEC-SB to reset the GSP falcon to its pre-libos state=
. */
             ...existing FWSEC-SB + booter unloader code...
+        } else {
+            // GSP falcon does most of the work of resetting, so just wait=
 for it to finish.
+            read_poll_timeout(
+                || Ok(gsp_falcon.is_riscv_active(bar)),
+                |&active| !active,
+                Delta::from_millis(10),
+                Delta::from_secs(5),
+            )
+            .inspect_err(|_| dev_err!(dev, "GSP falcon failed to halt\n"))=
?;
+        }
```

**Observations:**

- **Correct polling logic**: The poll waits for `is_riscv_active()` to retu=
rn `false`, meaning the GSP falcon has halted. This is the inverse of the b=
oot-time poll which waits for RISC-V to become active. The logic `|&active|=
 !active` correctly negates the condition.

- **Timeout values**: 10ms sleep interval with 5 second timeout. The boot-t=
ime active poll uses the same values (`Delta::from_millis(10)`, `Delta::fro=
m_secs(5)`). This is consistent and seems reasonable for waiting for GSP sh=
utdown.

- **Error logging**: Good =E2=80=94 `.inspect_err()` provides a clear messa=
ge on timeout, unlike patch 1.

- **Mixed comment styles**: Within this patch, the SEC2 path uses C-style `=
/* ... */` comments (preserved from the original code), while the new else =
branch uses Rust `//` comments. This is a very minor inconsistency, but it'=
s understandable since the SEC2 side is just indented existing code. Not wo=
rth changing.

- **Pure indentation change on the SEC2 path**: The SEC2 side is just the e=
xisting code wrapped in an `if` block with one additional indentation level=
. This is clean =E2=80=94 no behavioral changes to the existing path.

- **`read_poll_timeout` import**: This function is already imported in `boo=
t.rs` (line 6 of the current tree shows `io::poll::read_poll_timeout`), and=
 `Delta` is also already imported (line 10). The prerequisite unload series=
 presumably keeps these imports intact, so no additional imports are needed=
 for patch 3.

**Verdict:** Correct and clean. The approach of simply waiting for GSP falc=
on halt on non-SEC2 architectures makes hardware sense =E2=80=94 on Hopper+=
, the FSP-based boot path means GSP handles its own cleanup during shutdown.

---
Generated by Claude Code Patch Reviewer