[PATCH v3] drm/probe-helper: Honour connector->force in drm_helper_probe_detect()

[PATCH v3] drm/probe-helper: Honour connector->force in drm_helper_probe_detect()

[Paul Menzel]

With `video=DP-2:d drm.debug=0x04` the connectors are correctly forced off
at init time:

    [9.742468] [drm] forcing DP-2 connector off

but intel_dp_detect() is still called immediately after:

    [9.908738] [drm:intel_dp_detect] [CONNECTOR:130:DP-2]
    [9.912982] [drm:intel_hotplug_detect_connector] [CONNECTOR:130:DP-2] status updated from unknown to unknown (epoch counter 0->1)

drm_helper_probe_single_connector_modes() already short-circuits when
connector->force is set, returning the forced status without calling any
detect callbacks.  drm_helper_probe_detect() however has no such guard, so
callers that go through it directly – such as i915’s
intel_hotplug_detect_connector() – still run the full hardware probe (AUX
DPCD read, GMBus DDC transaction, ...) even when the user has forced a
connector off via the `video=` command-line parameter.

Add the same connector->force check for forced off connectors at the top of
drm_helper_probe_detect(), so that the hardware probe is skipped regardless
of the entry point.  This makes video=<connector>:d effective at boot time,
not just for userspace GETCONNECTOR ioctls.

DRM_FORCE_ON and DRM_FORCE_ON_DIGITAL are intentionally not short-
circuited here: for those cases the detect callback must still run to
initialise sink capabilities such as DPCD link rate and lane count.

Tested on a Dell XPS 13 9360 (Intel Kaby Lake-U GT2 [HD Graphics 620]
[8086:5916]). With `video=DP-2:d drm.debug=0x04, Linux logs:

    [   11.701243] [drm] forcing DP-2 connector off
    [...]
    [   12.102009] i915 0000:00:02.0: [drm:drm_helper_probe_detect [drm_kms_helper]] [CONNECTOR:130:DP-2] forced off, skipping detect

v2: Do not return early for DRM_FORCE_ON and DRM_FORCE_ON_DIGITAL (Jani,
gemini/gemini-3.1-pro-preview)

v3: Correct if condition

Assisted-by: Claude Sonnet 4.6
Signed-off-by: Paul Menzel <pmenzel@molgen.mpg.de>
---
 drivers/gpu/drm/drm_probe_helper.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c
index d4dc8cb45bce..1e15cb42d50b 100644
--- a/drivers/gpu/drm/drm_probe_helper.c
+++ b/drivers/gpu/drm/drm_probe_helper.c
@@ -401,6 +401,12 @@ drm_helper_probe_detect(struct drm_connector *connector,
 	struct drm_device *dev = connector->dev;
 	int ret;
 
+	if (connector->force == DRM_FORCE_OFF) {
+		drm_dbg_kms(dev, "[CONNECTOR:%d:%s] forced off, skipping detect\n",
+			    connector->base.id, connector->name);
+		return connector_status_disconnected;
+	}
+
 	if (!ctx)
 		return drm_helper_probe_detect_ctx(connector, force);
 
-- 
2.53.0

Claude review: drm/probe-helper: Honour connector->force in drm_helper_probe_detect()

[Claude Code Review Bot]

Overall Series Review

Subject: drm/probe-helper: Honour connector->force in drm_helper_probe_detect()
Author: Paul Menzel <pmenzel@molgen.mpg.de>
Patches: 1
Reviewed: 2026-05-27T13:57:30.178462

---

This is a single well-motivated patch that fixes a real behavioral gap in the DRM probe helper infrastructure. The problem is clear: `drm_helper_probe_single_connector_modes()` (line 595) already respects `connector->force` and skips hardware detection when a connector is forced, but `drm_helper_probe_detect()` does not. Callers that go through `drm_helper_probe_detect()` directly — most notably i915's `intel_hotplug_detect_connector()` — perform unnecessary hardware probes (AUX/DPCD reads, DDC transactions) on connectors the user has explicitly forced off via `video=`.

The fix is minimal, correct in scope, and well-placed. The decision to only short-circuit `DRM_FORCE_OFF` (and not `DRM_FORCE_ON` / `DRM_FORCE_ON_DIGITAL`) is well-reasoned and explicitly documented in the commit message.

**Recommendation: Looks good to merge.**

---
Generated by Claude Code Patch Reviewer

Claude review: drm/probe-helper: Honour connector->force in drm_helper_probe_detect()

[Claude Code Review Bot]

Patch Review

**Correctness: Good**

The early return at line 404-408 is placed before both code paths (`!ctx` and the locked path), so it covers all callers uniformly. The return value `connector_status_disconnected` is consistent with what `drm_helper_probe_single_connector_modes()` sets at line 600 for `DRM_FORCE_OFF`.

**Epoch counter behavior: Acceptable**

The early return bypasses the `epoch_counter` increment at line 419-420. This is fine because:
- For the initial boot probe, `connector->status` starts as `connector_status_unknown`, so the first call returning `connector_status_disconnected` would trigger an epoch bump in the `!ctx` path (via `drm_helper_probe_detect_ctx` at line 377-378). However, with this patch the early return skips that bump. In practice, this is a non-issue: the polling path at line 795 already skips forced connectors entirely (`if (connector->force) continue`), and `drm_helper_probe_single_connector_modes()` at line 595 handles forced connectors before calling `drm_helper_probe_detect()`.
- The i915 caller at `intel_hotplug.c:322` checks `old_epoch_counter != connector->base.epoch_counter` at line 326, but since the connector is forced off, no epoch change is expected and `INTEL_HOTPLUG_UNCHANGED` will be returned, which is the correct result.

**One minor observation on callers:**

The `drm_helper_hpd_irq_event()` caller at line 994 does `connector->status = drm_helper_probe_detect(...)` unconditionally (no `!connector->force` guard like i915 has at line 323). With this patch, a forced-off connector receiving an HPD IRQ would now get `connector_status_disconnected` assigned directly, which is the correct status for a forced-off connector. Previously it would have run the full detect and potentially set `connector_status_connected`, which would then be overridden later by `drm_helper_probe_single_connector_modes()`. So this is actually a slight improvement for that path too.

**Style: Clean**

The debug message format `[CONNECTOR:%d:%s] forced off, skipping detect` follows the existing DRM debug conventions in the file.

**Commit message: Well-written**

Clear problem description, explicit rationale for why `DRM_FORCE_ON`/`DRM_FORCE_ON_DIGITAL` are not short-circuited, and includes test evidence. The version history is helpful.

**Minor nit (non-blocking):** The commit message has a missing closing backtick on the test command line:
```
With `video=DP-2:d drm.debug=0x04, Linux logs:
```
should be:
```
With `video=DP-2:d drm.debug=0x04`, Linux logs:
```

**Verdict: Reviewed-by worthy.** The patch is correct, minimal, and well-explained.

---
Generated by Claude Code Patch Reviewer