From mboxrd@z Thu Jan 1 00:00:00 1970 From: Claude Code Review Bot To: dri-devel-reviews@example.com Subject: Claude review: Fair(er) DRM scheduler Date: Mon, 09 Mar 2026 08:37:18 +1000 Message-ID: In-Reply-To: <20260306163445.97243-1-tvrtko.ursulin@igalia.com> References: <20260306163445.97243-1-tvrtko.ursulin@igalia.com> X-Mailer: Claude Code Patch Reviewer Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit MIME-Version: 1.0 Overall Series Review Subject: Fair(er) DRM scheduler Author: Tvrtko Ursulin Patches: 31 Reviewed: 2026-03-09T08:37:18.150766 --- This is a well-structured, v7 series by Tvrtko Ursulin that replaces the existing FIFO and round-robin scheduling policies in the DRM GPU scheduler with a new CFS-inspired "fair" scheduling algorithm based on virtual runtime. The series is mature, with acks from Danilo Krummrich and reviews from multiple maintainers across the affected subsystems. **Strengths:** - Clear progression: cleanups/consolidation first, then the new algorithm, then removal of old code and simplification - Extensive unit tests added before the algorithm changes - Well-documented commit messages explaining design trade-offs (particularly the GPU time accounting placement rationale in patch 8) - The splitting point at patch 12 is well-chosen for staged merge strategy - Comprehensive benchmarking data in the cover letter - Clean removal of `num_rqs` across all drivers at the end **Concerns:** - The `container_of(entity->rq, typeof(*sched), rq)` pattern introduced in patch 14 is verbose and appears many times across amdgpu code. Worth noting but this is inherent to the embedding approach. - The `drm_sched_entity_stats` lock granularity could be a future contention point under heavy workloads with many entities, though the current design is reasonable for now. Overall, the series is in good shape and the approach is sound. --- Generated by Claude Code Patch Reviewer