[PATCH 0/2] drm: Populate ELD Port_ID to identify audio endpoints

[PATCH 0/2] drm: Populate ELD Port_ID to identify audio endpoints

[Kai Vehmanen]

The ELD Port_ID field (bytes 8..15 of the baseline block) is typically
zero on Linux today: drm_edid_to_eld() leaves it empty and no in-tree driver
fills it in. amdgpu, radeon and nouveau pass the raw connector->eld to
the audio component. On the amdgpu DC Azalia register path the value
is a hardcoded constant, the same for every port.
Only closed display drivers fill the value.

Pipewire added support for "Port_ID" in merge request:
https://gitlab.freedesktop.org/pipewire/pipewire/-/merge_requests/3025

As a result, when two identical monitors are connected their ELDs are
byte-identical, and userspace (e.g. PipeWire/WirePlumber) cannot
passively tell which HDMI/DP PCM belongs to which DRM connector.

This series adds a DRM helper that fills Port_ID with a common layout,
and uses it in i915:

  bits 63:48 - PCI domain of the DRM device
  bits 47:32 - PCI bus/device/function of the DRM device
  bits 31:0  - DRM connector object ID

The upper half follows the layout documented for the AMD Azalia sink
info registers (dce_audio.c: "PORT_ID1 = 16bit BDF"). The aim is for
this to become a convention other drivers can opt into. For drivers
that copy connector->eld to the audio component (amdgpu DM, radeon,
nouveau), adopting it would be a one-line change. Comments from the
AMD and nouveau folks on whether the layout works for them are very
welcome.

An alternative would be to fill Port_ID directly in drm_edid_to_eld(),
so every driver gets it without changes. I went with an opt-in helper
to avoid changing the ELD contents seen by all drivers and existing
userspace at once.

The value is exposed to userspace via the ALSA ELD control and
/proc/asound/cardX/eld#Y.

Kai Vehmanen (2):
  drm/eld: Add drm_eld_set_port_id() helper
  drm/i915/audio: Populate ELD Port_ID

 drivers/gpu/drm/drm_eld.c                  | 42 ++++++++++++++++++++++
 drivers/gpu/drm/i915/display/intel_audio.c |  2 ++
 include/drm/drm_eld.h                      |  3 ++
 3 files changed, 47 insertions(+)

-- 
2.43.0

[PATCH 1/2] drm/eld: Add drm_eld_set_port_id() helper

[Kai Vehmanen]

The ELD baseline block reserves bytes 8..15 for a 64-bit Port_ID that
identifies the source port of the audio endpoint. drm_edid_to_eld()
leaves the field zero, and no driver fills it in, so it carries no
information.

When two identical monitors are connected, their ELDs are then
byte-identical, and userspace has no passive way to tell which HDMI/DP
PCM belongs to which DRM connector. The only workaround is to toggle
the connector audio property and watch which ELD goes away, which
disrupts the endpoint.

Add a helper that drivers can use to fill in Port_ID with a common
layout:

  bits 63:48 - PCI domain of the DRM device
  bits 47:32 - PCI bus/device/function of the DRM device
  bits 31:0  - DRM connector object ID

Userspace can read the connector object ID directly from KMS, and the
PCI address tells GPUs apart in multi-GPU systems. For non-PCI devices
the upper half is left zero.

The split between a 16-bit PCI BDF in the upper half and a per-device
display identifier in the lower half follows the Port_ID layout
documented for the AMD Azalia sink info registers in dce_audio.c.

Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
---
 drivers/gpu/drm/drm_eld.c | 42 +++++++++++++++++++++++++++++++++++++++
 include/drm/drm_eld.h     |  3 +++
 2 files changed, 45 insertions(+)

diff --git a/drivers/gpu/drm/drm_eld.c b/drivers/gpu/drm/drm_eld.c
index c0428d07de53..baebece7b4ed 100644
--- a/drivers/gpu/drm/drm_eld.c
+++ b/drivers/gpu/drm/drm_eld.c
@@ -4,7 +4,11 @@
  */
 
 #include <linux/export.h>
+#include <linux/pci.h>
+#include <linux/unaligned.h>
 
+#include <drm/drm_connector.h>
+#include <drm/drm_device.h>
 #include <drm/drm_edid.h>
 #include <drm/drm_eld.h>
 
@@ -55,3 +59,41 @@ int drm_eld_sad_set(u8 *eld, int sad_index, const struct cea_sad *cta_sad)
 	return 0;
 }
 EXPORT_SYMBOL(drm_eld_sad_set);
+
+/**
+ * drm_eld_set_port_id - set the ELD Port_ID for a connector
+ * @eld: ELD buffer
+ * @connector: connector the ELD belongs to
+ *
+ * drm_edid_to_eld() leaves the 64-bit Port_ID field of the ELD baseline
+ * block zero. Drivers can use this helper to fill it in with a value
+ * that identifies the audio endpoint, so that userspace can associate an
+ * ELD, and the PCM device it belongs to, with a DRM connector, even when
+ * identical sinks are connected to multiple outputs.
+ *
+ * The Port_ID is stored in little-endian byte order, with the layout:
+ *
+ * - bits 63:48 - PCI domain of the DRM device, truncated to 16 bits
+ * - bits 47:32 - PCI bus/device/function of the DRM device
+ * - bits 31:0  - DRM connector object ID
+ *
+ * For non-PCI devices bits 63:32 are zero.
+ *
+ * The connector object ID is only stable for the lifetime of the
+ * connector, so userspace must look it up at runtime, and not store it.
+ */
+void drm_eld_set_port_id(u8 *eld, const struct drm_connector *connector)
+{
+	struct device *dev = connector->dev->dev;
+	u64 port_id = connector->base.id;
+
+	if (dev && dev_is_pci(dev)) {
+		struct pci_dev *pdev = to_pci_dev(dev);
+
+		port_id |= (u64)(pci_domain_nr(pdev->bus) & 0xffff) << 48 |
+			   (u64)pci_dev_id(pdev) << 32;
+	}
+
+	put_unaligned_le64(port_id, &eld[DRM_ELD_PORT_ID]);
+}
+EXPORT_SYMBOL(drm_eld_set_port_id);
diff --git a/include/drm/drm_eld.h b/include/drm/drm_eld.h
index 0a88d10b28b0..de0478720721 100644
--- a/include/drm/drm_eld.h
+++ b/include/drm/drm_eld.h
@@ -9,6 +9,7 @@
 #include <linux/types.h>
 
 struct cea_sad;
+struct drm_connector;
 
 /* ELD Header Block */
 #define DRM_ELD_HEADER_BLOCK_SIZE	4
@@ -80,6 +81,8 @@ static inline int drm_eld_mnl(const u8 *eld)
 int drm_eld_sad_get(const u8 *eld, int sad_index, struct cea_sad *cta_sad);
 int drm_eld_sad_set(u8 *eld, int sad_index, const struct cea_sad *cta_sad);
 
+void drm_eld_set_port_id(u8 *eld, const struct drm_connector *connector);
+
 /**
  * drm_eld_sad - Get ELD SAD structures.
  * @eld: pointer to an eld memory structure with sad_count set
-- 
2.43.0

[PATCH 2/2] drm/i915/audio: Populate ELD Port_ID

[Kai Vehmanen]

i915 never fills in the ELD Port_ID field, so it is always zero, and
the ELDs of identical monitors connected to different outputs are
byte-identical. Userspace then cannot tell which HDMI/DP PCM belongs to
which DRM connector without toggling the connector audio property.

Use drm_eld_set_port_id() when computing the audio config, so the value
reaches the audio driver through every path: the audio component
get_eld() callback, the hardware ELD buffer, SDVO and LPE audio.

The value is fixed for the lifetime of the connector, so it does not
cause extra audio re-enables when the ELD is compared across commits,
and state readout stays consistent because it is copied from the
programmed ELD.

Link: https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/16693
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
---
 drivers/gpu/drm/i915/display/intel_audio.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/gpu/drm/i915/display/intel_audio.c b/drivers/gpu/drm/i915/display/intel_audio.c
index 32f8c5e55319..3277e67748ee 100644
--- a/drivers/gpu/drm/i915/display/intel_audio.c
+++ b/drivers/gpu/drm/i915/display/intel_audio.c
@@ -881,6 +881,8 @@ bool intel_audio_compute_config(struct intel_encoder *encoder,
 	crtc_state->eld[6] = drm_av_sync_delay(connector, adjusted_mode) / 2;
 	mutex_unlock(&connector->eld_mutex);
 
+	drm_eld_set_port_id(crtc_state->eld, connector);
+
 	return true;
 }
 
-- 
2.43.0

Re: [PATCH 1/2] drm/eld: Add drm_eld_set_port_id() helper

[Ville Syrjälä]

On Thu, Nov 05, 2026 at 06:02:39PM +0200, Kai Vehmanen wrote:
> The ELD baseline block reserves bytes 8..15 for a 64-bit Port_ID that
> identifies the source port of the audio endpoint. drm_edid_to_eld()
> leaves the field zero, and no driver fills it in, so it carries no
> information.
> 
> When two identical monitors are connected, their ELDs are then
> byte-identical, and userspace has no passive way to tell which HDMI/DP
> PCM belongs to which DRM connector. The only workaround is to toggle
> the connector audio property and watch which ELD goes away, which
> disrupts the endpoint.
> 
> Add a helper that drivers can use to fill in Port_ID with a common
> layout:
> 
>   bits 63:48 - PCI domain of the DRM device
>   bits 47:32 - PCI bus/device/function of the DRM device
>   bits 31:0  - DRM connector object ID
> 
> Userspace can read the connector object ID directly from KMS, and the
> PCI address tells GPUs apart in multi-GPU systems. For non-PCI devices
> the upper half is left zero.
> 
> The split between a 16-bit PCI BDF in the upper half and a per-device
> display identifier in the lower half follows the Port_ID layout
> documented for the AMD Azalia sink info registers in dce_audio.c.
> 
> Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
> ---
>  drivers/gpu/drm/drm_eld.c | 42 +++++++++++++++++++++++++++++++++++++++
>  include/drm/drm_eld.h     |  3 +++
>  2 files changed, 45 insertions(+)
> 
> diff --git a/drivers/gpu/drm/drm_eld.c b/drivers/gpu/drm/drm_eld.c
> index c0428d07de53..baebece7b4ed 100644
> --- a/drivers/gpu/drm/drm_eld.c
> +++ b/drivers/gpu/drm/drm_eld.c
> @@ -4,7 +4,11 @@
>   */
>  
>  #include <linux/export.h>
> +#include <linux/pci.h>
> +#include <linux/unaligned.h>
>  
> +#include <drm/drm_connector.h>
> +#include <drm/drm_device.h>
>  #include <drm/drm_edid.h>
>  #include <drm/drm_eld.h>
>  
> @@ -55,3 +59,41 @@ int drm_eld_sad_set(u8 *eld, int sad_index, const struct cea_sad *cta_sad)
>  	return 0;
>  }
>  EXPORT_SYMBOL(drm_eld_sad_set);
> +
> +/**
> + * drm_eld_set_port_id - set the ELD Port_ID for a connector
> + * @eld: ELD buffer
> + * @connector: connector the ELD belongs to
> + *
> + * drm_edid_to_eld() leaves the 64-bit Port_ID field of the ELD baseline
> + * block zero. Drivers can use this helper to fill it in with a value
> + * that identifies the audio endpoint, so that userspace can associate an
> + * ELD, and the PCM device it belongs to, with a DRM connector, even when
> + * identical sinks are connected to multiple outputs.
> + *
> + * The Port_ID is stored in little-endian byte order, with the layout:
> + *
> + * - bits 63:48 - PCI domain of the DRM device, truncated to 16 bits
> + * - bits 47:32 - PCI bus/device/function of the DRM device
> + * - bits 31:0  - DRM connector object ID
> + *
> + * For non-PCI devices bits 63:32 are zero.
> + *
> + * The connector object ID is only stable for the lifetime of the
> + * connector, so userspace must look it up at runtime, and not store it.
> + */
> +void drm_eld_set_port_id(u8 *eld, const struct drm_connector *connector)
> +{
> +	struct device *dev = connector->dev->dev;
> +	u64 port_id = connector->base.id;
> +
> +	if (dev && dev_is_pci(dev)) {
> +		struct pci_dev *pdev = to_pci_dev(dev);
> +
> +		port_id |= (u64)(pci_domain_nr(pdev->bus) & 0xffff) << 48 |
> +			   (u64)pci_dev_id(pdev) << 32;

The numbers look correct.

> +	}
> +
> +	put_unaligned_le64(port_id, &eld[DRM_ELD_PORT_ID]);

Endianness agrees with the HDA spec.

The implementation looks good
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

But I'll leave it to someone else to judge the actual
usefulness of the provided information.

> +}
> +EXPORT_SYMBOL(drm_eld_set_port_id);
> diff --git a/include/drm/drm_eld.h b/include/drm/drm_eld.h
> index 0a88d10b28b0..de0478720721 100644
> --- a/include/drm/drm_eld.h
> +++ b/include/drm/drm_eld.h
> @@ -9,6 +9,7 @@
>  #include <linux/types.h>
>  
>  struct cea_sad;
> +struct drm_connector;
>  
>  /* ELD Header Block */
>  #define DRM_ELD_HEADER_BLOCK_SIZE	4
> @@ -80,6 +81,8 @@ static inline int drm_eld_mnl(const u8 *eld)
>  int drm_eld_sad_get(const u8 *eld, int sad_index, struct cea_sad *cta_sad);
>  int drm_eld_sad_set(u8 *eld, int sad_index, const struct cea_sad *cta_sad);
>  
> +void drm_eld_set_port_id(u8 *eld, const struct drm_connector *connector);
> +
>  /**
>   * drm_eld_sad - Get ELD SAD structures.
>   * @eld: pointer to an eld memory structure with sad_count set
> -- 
> 2.43.0

-- 
Ville Syrjälä
Intel

Re: [PATCH 2/2] drm/i915/audio: Populate ELD Port_ID

[Ville Syrjälä]

On Thu, Nov 05, 2026 at 06:02:40PM +0200, Kai Vehmanen wrote:
> i915 never fills in the ELD Port_ID field, so it is always zero, and
> the ELDs of identical monitors connected to different outputs are
> byte-identical. Userspace then cannot tell which HDMI/DP PCM belongs to
> which DRM connector without toggling the connector audio property.
> 
> Use drm_eld_set_port_id() when computing the audio config, so the value
> reaches the audio driver through every path: the audio component
> get_eld() callback, the hardware ELD buffer, SDVO and LPE audio.
> 
> The value is fixed for the lifetime of the connector, so it does not
> cause extra audio re-enables when the ELD is compared across commits,
> and state readout stays consistent because it is copied from the
> programmed ELD.
> 
> Link: https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/16693
> Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

> ---
>  drivers/gpu/drm/i915/display/intel_audio.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_audio.c b/drivers/gpu/drm/i915/display/intel_audio.c
> index 32f8c5e55319..3277e67748ee 100644
> --- a/drivers/gpu/drm/i915/display/intel_audio.c
> +++ b/drivers/gpu/drm/i915/display/intel_audio.c
> @@ -881,6 +881,8 @@ bool intel_audio_compute_config(struct intel_encoder *encoder,
>  	crtc_state->eld[6] = drm_av_sync_delay(connector, adjusted_mode) / 2;
>  	mutex_unlock(&connector->eld_mutex);
>  
> +	drm_eld_set_port_id(crtc_state->eld, connector);
> +
>  	return true;
>  }
>  
> -- 
> 2.43.0

-- 
Ville Syrjälä
Intel