[RESEND PATCH v2 0/2] drm/amd: fix HDMI output with non-HDMI EDID overrides

[RESEND PATCH v2 0/2] drm/amd: fix HDMI output with non-HDMI EDID overrides

[Johannes Wüller]

Resend: my first attempt at sending v2 went out without any of the Cc
recipients, including the mailing lists. This is the same series,
unchanged. Sorry for the noise! I'm doing these for the first time.

This series lets amdgpu drive an HDMI display at its native resolution
when a non-HDMI EDID is injected via drm.edid_firmware or debugfs
edid_override.

The motivating case is a Dell U3011 (2560x1600) connected via HDMI,
whose native HDMI EDID only advertises CEA modes up to 1920x1080. By
injecting the monitor's DisplayPort EDID (which contains the native
2560x1600 mode), the correct resolution becomes available. However,
the injected EDID lacks an HDMI VSDB, so amdgpu treats the physical
HDMI connector as a passive HDMI-DVI dongle and rejects the required
pixel clock.

Patch 1 adds an edid_overridden flag to drm_connector, set by the
existing EDID override mechanisms in _drm_do_get_edid(), so that
drivers can detect when an override is in effect and adjust their
behavior accordingly, without duplicating the override mechanism.

Patch 2 uses that flag in amdgpu to trust the physical connector type
over the EDID when selecting the signal type.

Tested on AMD Ryzen 9 7950X (Raphael iGPU, DCN 3.1.5) with a Dell U3011
connected via HDMI and drm.edid_firmware=HDMI-A-1:<DisplayPort EDID>:
- amd-staging-drm-next at 53750a937420: the override EDID is loaded,
  but 2560x1600 is pruned from the connector's mode list.
- with this series applied: 2560x1600@60Hz RGB 8bpc works, alongside
  another Dell U3011 connected via DisplayPort in the same mode.

Changes in v2:
- Rebased onto current amd-staging-drm-next.
- Patch 2: follow the move of create_eml_sink() to
  amdgpu_dm_connector.c.
- Dropped "drm/amd: treat max_tmds_clock==0 as unconstrained": The
  return value of adjust_colour_depth_from_display_info() is no longer
  used to reject the stream, so the patch has no effect anymore.
- Dropped "drm/amd: fix force_yuv420/422_output being ignored": The
  flags were replaced by force_yuv_pixel_format.
- I am (hopefully) sending this version to the right people, since v1's
  cover letter only went to the lists. Please let me know otherwise!

v1: https://lore.kernel.org/amd-gfx/20260315223307.45807-1-johanneswueller@gmail.com/

Johannes Wüller (2):
  drm: add edid_overridden flag to drm_connector
  drm/amd: fix HDMI signal type for EDID overrides

 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_connector.c  | 9 ++++++++-
 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c    | 9 ++++++++-
 drivers/gpu/drm/drm_edid.c                               | 2 ++
 include/drm/drm_connector.h                              | 8 ++++++++
 4 files changed, 26 insertions(+), 2 deletions(-)


base-commit: 53750a937420a01acf41bf74010557666f679aab
-- 
2.56.0

[RESEND PATCH v2 1/2] drm: add edid_overridden flag to drm_connector

[Johannes Wüller]

When an EDID override is in effect, the physical connector may not
match reported capabilities. Adds a drm_connector flag that allows
drivers to detect such mismatches.

Signed-off-by: Johannes Wüller <johanneswueller@gmail.com>
---
v2: rebased, no functional change.

 drivers/gpu/drm/drm_edid.c  | 2 ++
 include/drm/drm_connector.h | 8 ++++++++
 2 files changed, 10 insertions(+)

diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
index 42f4e41c459f..9ea6f5f19131 100644
--- a/drivers/gpu/drm/drm_edid.c
+++ b/drivers/gpu/drm/drm_edid.c
@@ -2424,6 +2424,7 @@ static struct edid *_drm_do_get_edid(struct drm_connector *connector,
 	struct edid *edid, *new;
 	size_t alloc_size = EDID_LENGTH;
 
+	connector->edid_overridden = false;
 	override = drm_edid_override_get(connector);
 	if (override) {
 		alloc_size = override->size;
@@ -2431,6 +2432,7 @@ static struct edid *_drm_do_get_edid(struct drm_connector *connector,
 		drm_edid_free(override);
 		if (!edid)
 			return NULL;
+		connector->edid_overridden = true;
 		goto ok;
 	}
 
diff --git a/include/drm/drm_connector.h b/include/drm/drm_connector.h
index 899643586bb3..d242fb922741 100644
--- a/include/drm/drm_connector.h
+++ b/include/drm/drm_connector.h
@@ -2394,6 +2394,14 @@ struct drm_connector {
 	/** @bad_edid_counter: track sinks that give us an EDID with invalid checksum */
 	unsigned bad_edid_counter;
 
+	/**
+	 * @edid_overridden: Indicates whether the last read EDID was an
+	 * override (e.g. via debugfs edid_override or drm.edid_firmware kernel
+	 * parameter), which can cause the physical connector to differ from
+	 * the advertised capabilities.
+	 */
+	bool edid_overridden;
+
 	/**
 	 * @edid_corrupt: Indicates whether the last read EDID was corrupt. Used
 	 * in Displayport compliance testing - Displayport Link CTS Core 1.2
-- 
2.56.0

[RESEND PATCH v2 2/2] drm/amd: fix HDMI signal type for EDID overrides

[Johannes Wüller]

If an EDID override is in effect (e.g. via debugfs edid_override or
drm.edid_firmware kernel parameter), the overriding EDID may not
advertise capabilities matching the physical connector being used.
In that case, trust the physical connector type over the EDID in
create_eml_sink() and dm_helpers_parse_edid_caps().

Signed-off-by: Johannes Wüller <johanneswueller@gmail.com>
---
v2: rebased; create_eml_sink() now lives in amdgpu_dm_connector.c.

 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_connector.c  | 9 ++++++++-
 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c    | 9 ++++++++-
 2 files changed, 16 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_connector.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_connector.c
index ad3383169d83..2bce57b06037 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_connector.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_connector.c
@@ -2134,7 +2134,14 @@ VISIBLE_IF_KUNIT void create_eml_sink(struct amdgpu_dm_connector *aconnector)
 		return;
 	}
 
-	if (connector->display_info.is_hdmi)
+	/*
+	 * If an EDID override is active, it may not advertise HDMI capability
+	 * even though the physical connector is HDMI. Trust the connector type.
+	 */
+	if (connector->display_info.is_hdmi ||
+	    (connector->edid_overridden &&
+	     (connector->connector_type == DRM_MODE_CONNECTOR_HDMIA ||
+	      connector->connector_type == DRM_MODE_CONNECTOR_HDMIB)))
 		init_params.sink_signal = SIGNAL_TYPE_HDMI_TYPE_A;
 
 	aconnector->drm_edid = drm_edid;
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c
index 5c74f19fc33a..9f4465682088 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c
@@ -223,7 +223,14 @@ enum dc_edid_status dm_helpers_parse_edid_caps(
 				  edid_caps->display_name,
 				  AUDIO_INFO_DISPLAY_NAME_SIZE_IN_CHARS);
 
-	edid_caps->edid_hdmi = connector->display_info.is_hdmi;
+	/*
+	 * If an EDID override is active, it may not advertise HDMI capability
+	 * even though the physical connector is HDMI. Trust the connector type.
+	 */
+	edid_caps->edid_hdmi = connector->display_info.is_hdmi ||
+		(connector->edid_overridden &&
+		 (connector->connector_type == DRM_MODE_CONNECTOR_HDMIA ||
+		  connector->connector_type == DRM_MODE_CONNECTOR_HDMIB));
 
 	if (edid_caps->edid_hdmi) {
 		edid_caps->qs_bit = connector->display_info.rgb_quant_range_selectable;
-- 
2.56.0