From mboxrd@z Thu Jan 1 00:00:00 1970 From: Claude Code Review Bot To: dri-devel-reviews@example.com Subject: Claude review: Re: [PATCH v2] cgroup/dmem: implement dmem.high soft limit via prioritized eviction Date: Mon, 25 May 2026 19:12:32 +1000 Message-ID: In-Reply-To: References: 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: Re: [PATCH v2] cgroup/dmem: implement dmem.high soft limit via prioritized eviction Author: Natalie Vock Patches: 2 Reviewed: 2026-05-25T19:12:32.597474 --- This is a single-patch series (v2) adding a `dmem.high` soft limit to the dmem cgroup v2 controller. When device memory eviction is triggered (e.g., by hitting the hard `max` limit), buffer objects from cgroups exceeding their `high` limit are evicted first via a new prioritized pass in `ttm_bo_evict_alloc()`. **v2 improves significantly over v1** by replacing sleep-on-allocation throttling (which would be catastrophic for GPU submission) with prioritized eviction. The approach reuses existing `page_counter` infrastructure (`page_counter.high`, `page_counter_set_high()`) and follows the cgroup v2 naming convention (min < low < high < max). However, there are design concerns around protection bypass, missing documentation/selftests, and a semantic inconsistency with how the memory cgroup controller handles `high` vs protection guarantees. --- --- Generated by Claude Code Patch Reviewer