From mboxrd@z Thu Jan 1 00:00:00 1970 From: Claude Code Review Bot To: dri-devel-reviews@example.com Subject: Claude review: drm/ttm/pool: back up at native page order Date: Tue, 05 May 2026 08:38:59 +1000 Message-ID: In-Reply-To: <20260504042619.2896273-1-matthew.brost@intel.com> References: <20260504042619.2896273-1-matthew.brost@intel.com> X-Mailer: Claude Code Patch Reviewer Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Overall Series Review Subject: drm/ttm/pool: back up at native page order Author: Matthew Brost Patches: 5 Reviewed: 2026-05-05T08:38:59.897610 --- This is a single patch that modifies TTM pool backup to preserve compound p= age integrity during the shrinker backup path, avoiding memory fragmentatio= n caused by unconditionally splitting higher-order allocations to order-0 b= efore writing to shmem. The motivation is sound and well-documented: under = kswapd pressure, the existing `split_page()` approach shatters higher-order= blocks, amplifying fragmentation and preventing the system from recovering. The core approach =E2=80=94 back up each sub-page to shmem but free the all= ocation atomically at native order =E2=80=94 is clean. The rollback logic f= or partial failures is correct and ensures compound-level atomicity. Howeve= r, there is one significant issue: the patch uses `__free_pages_gpu_account= ()`, which does not exist in the current drm-next tree. This is an unstated= build dependency on another pending series (likely [1] referenced in the c= over notes). The patch cannot compile standalone against drm-next. The commit message is thorough, and the `Cc: stable` tag is present, but th= e stable backport would also need the dependency or an adaptation. --- --- Generated by Claude Code Patch Reviewer