From mboxrd@z Thu Jan 1 00:00:00 1970 From: Claude Code Review Bot To: dri-devel-reviews@example.com Subject: Claude review: gpu: nova-core: gsp: add mechanism to wait for space on command queue Date: Mon, 09 Mar 2026 09:11:26 +1000 Message-ID: In-Reply-To: <20260306-cmdq-continuation-v6-2-cc7b629200ee@nvidia.com> References: <20260306-cmdq-continuation-v6-0-cc7b629200ee@nvidia.com> <20260306-cmdq-continuation-v6-2-cc7b629200ee@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 Replaces the immediate `EAGAIN` failure in `allocate_command` with a `read_= poll_timeout` loop, adding a 1-second timeout. This is required because con= tinuation records can fill up the queue and need the GSP to drain it. The new `driver_write_area_size()` method computes the writable area size w= ithout materializing the slices, which is the right approach for a polling = check. The poll delay of `Delta::from_micros(1)` (changed from zero in v5 to 1us i= n v6) is reasonable =E2=80=94 it avoids a busy spin while still being respo= nsive. The `ALLOCATE_TIMEOUT` of 1 second matches nouveau's behavior per the commi= t message. One minor note: the existing `driver_write_area` method has essentially the= same slot calculation logic. There's some duplication but it's justified s= ince `driver_write_area` returns actual slices while `driver_write_area_siz= e` just returns the size, and combining them would be awkward. No issues. --- Generated by Claude Code Patch Reviewer