Hi Luca,
On 6/27/2026 12:29 AM, Luca Ceresoli wrote:
On Thu, 04 Jun 2026 16:52:19 +0800, Damon Ding <[email protected]>
wrote:
Hello Damon,
diff --git a/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c
b/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c
index 7a85774aaac1..e120ef3320c1 100644
--- a/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c
+++ b/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c
@@ -1261,8 +1262,11 @@ static int analogix_dp_dt_parse_pdata(struct
analogix_dp_device *dp)
*/
of_property_read_u32(dp_node, "samsung,link-rate",
&video_info->max_link_rate);
- of_property_read_u32(dp_node, "samsung,lane-count",
- &video_info->max_lane_count);
+ ret = of_property_read_u32(dp_node, "samsung,lane-count",
+ &video_info->max_lane_count);
+ if (ret ||
!drm_dp_lane_count_is_valid(video_info->max_lane_count))
+ return dev_err_probe(dp->dev, ret ? ret : -EINVAL,
+ "failed to parse
samsung,lane-count\n");
I think this report by sashiko makes sense:
> [email protected] <[email protected]>:
>
> [Severity: High]
> Does this make the optional and deprecated samsung,lane-count property a
> strict requirement?
>
> If samsung,lane-count is absent from the device tree,
of_property_read_u32()
> returns -EINVAL. This causes the condition to evaluate to true, aborting
the
> probe with an error.
>
> According to the device tree bindings
>
(Documentation/devicetree/bindings/display/samsung/samsung,exynos5-dp.yaml),
> this property is marked as deprecated and explicitly optional because the
> lane count can be read from the monitor. Does this patch break
compatibility
> with device trees that rightfully omit this deprecated property?
(via: https://patch.msgid.link/[email protected])
Can you comment on this?
I was also confused about this handling at first. From commit
0d0abd894ead ("drm: bridge: analogix/dp: add max link rate and lane
count limit for RK3288"), its commit message does not explain why
samsung,lane-count was changed from a mandatory to optional property.
After digging into the code flow, I found that
analogix_dp_full_link_train() picks the smaller value between the
platform-supported lane count and the lane count retrieved from sink
DPCD for link training. If the samsung,lane-count property is
missing/invalid here, link training will end up using an unexpected lane
count configuration.
Additionally, I checked Samsung’s upstream device trees that utilize
this DP controller, and all of them carry the lane-count property. Based
on this observation, I believe restoring the strict mandatory
requirement for this property makes sense.
Should I create an independent fix patch to revert these two properties
to mandatory, instead of bundling the fix in this series?
Best regards,
Damon
