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: drm/fdinfo: Add "evicted" memory accounting
Date: Thu, 23 Apr 2026 08:36:40 +1000
Message-ID: <review-patch1-20260421-panthor-bo-reclaim-observability-v2-1-c9135eedfb6f@collabora.com>
In-Reply-To: <20260421-panthor-bo-reclaim-observability-v2-1-c9135eedfb6f@collabora.com>
References: <20260421-panthor-bo-reclaim-observability-v2-0-c9135eedfb6f@collabora.com> <20260421-panthor-bo-reclaim-observability-v2-1-c9135eedfb6f@collabora.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

This patch adds `DRM_GEM_OBJECT_EVICTED =3D BIT(3)` to the status enum, an =
`evicted` field to `drm_memory_stats`, and wires them into the fdinfo print=
ing/accumulation paths.

**Issue: Missing drm-usage-stats.rst documentation.**
The new `drm-evicted-<region>` key needs an entry in `Documentation/gpu/drm=
-usage-stats.rst` alongside the existing documented keys. This file explici=
tly says driver-specific stats "should be documented above and where possib=
le, aligned with other drivers." Since this is being added to the core DRM =
framework for any driver to use, it's especially important to document the =
semantics =E2=80=94 for example, that evicted is a subset of non-resident m=
emory.

**Minor observation on semantics in `drm_show_memory_stats()`:**
The existing code clears `DRM_GEM_OBJECT_PURGEABLE` when the object is not =
resident:

```c
if (s & DRM_GEM_OBJECT_RESIDENT) {
    status.resident +=3D add_size;
} else {
    s &=3D ~DRM_GEM_OBJECT_PURGEABLE;
}
```

The new `DRM_GEM_OBJECT_EVICTED` is accumulated without any filtering. This=
 is correct =E2=80=94 evicted objects are by definition not resident, so th=
e semantics are different from purgeable (which is defined as resident + id=
le + marked). However, the core code doesn't enforce mutual exclusivity bet=
ween RESIDENT and EVICTED. If a driver were to erroneously return both, bot=
h counters would be incremented. This is arguably fine (it's the driver's j=
ob to set bits correctly), but a brief comment on the enum doc noting that =
EVICTED implies non-resident would be helpful.

**Code is correct as written.** The additions to `drm_memory_stats_is_zero(=
)`, `drm_print_memory_stats()`, and `drm_show_memory_stats()` all follow th=
e established pattern.

---
Generated by Claude Code Patch Reviewer