* [PATCH] drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR
@ 2026-11-20 5:50 Yitian Shao
2026-11-20 6:15 ` sashiko-bot
2026-11-20 14:51 ` Ville Syrjälä
0 siblings, 2 replies; 4+ messages in thread
From: Yitian Shao @ 2026-11-20 5:50 UTC (permalink / raw)
To: intel-gfx
Cc: dri-devel, ville.syrjala, jani.nikula, sebastian.wick,
uma.shankar, suraj.kandpal
intel_dp_has_gamut_metadata_dip() returns false for port A on pre-ICL
platforms. intel_dp_add_properties() uses that helper to decide whether
to attach the HDR_OUTPUT_METADATA connector property, so eDP panels on
those platforms never expose it even when the panel advertises HDR in
its EDID.
On such panels the HDR metadata is not sent as a gamut metadata SDP, but
programmed through the Intel AUX HDR interface instead. The TCON is
already probed during intel_edp_backlight_setup(), which runs before
intel_dp_add_properties(), so the capability is known at that point.
Attach the property when the panel's TCON advertises PQ (ST2084) decode
support, so that userspace can enable HDR on these panels.
Since no SDP is transmitted in that case, also bail out early from
intel_dp_compute_hdr_metadata_infoframe_sdp() when the encoder cannot
send gamut metadata. Otherwise the computed infoframe state would be
inconsistent with what is actually programmed, and the state checker
would complain.
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/10415
Signed-off-by: Yitian Shao <fighters390@qq.com>
---
Reviewer notes, not part of the commit message:
Tested on a CometLake-U (Gen9.5) laptop with a 3840x2160 10bpc eDP panel that
advertises BT2020 + ST2084 in its EDID. With this change the connector gains
the HDR_OUTPUT_METADATA property, GNOME 50 exposes the HDR switch, and the
panel enters PQ/BT2020 mode (the DP Colorspace property switches to
BT2020_RGB). The Intel TCON HDR backlight interface was already in use before
this change, so the hardware path existed and only the property was missing.
Two things I am not sure about:
1. intel_dp_aux_write_content_luminance() also returns early when
!intel_dp_has_gamut_metadata_dip(), which is exactly the PORT_A case where
the AUX path is used. max_cll/max_fall therefore never reach the TCON. In
my testing Mutter sends all-zero mastering metadata anyway, so I could not
tell whether this shows up in practice. Should this check be relaxed too,
or is it intentional?
2. panel->backlight.edp.intel_cap shares a union with the VESA backlight
info, so supports_2084_decode would read unrelated bits if the VESA
interface had been selected instead of the Intel one. Would you prefer an
explicit "Intel HDR backlight is in use" check here?
drivers/gpu/drm/i915/display/intel_dp.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
index afd4169..49142f4 100644
--- a/drivers/gpu/drm/i915/display/intel_dp.c
+++ b/drivers/gpu/drm/i915/display/intel_dp.c
@@ -3088,6 +3088,9 @@ intel_dp_compute_hdr_metadata_infoframe_sdp(struct intel_dp *intel_dp,
int ret;
struct hdmi_drm_infoframe *drm_infoframe = &crtc_state->infoframes.drm.drm;
+ if (!intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
+ return;
+
if (!conn_state->hdr_output_metadata)
return;
@@ -6715,7 +6718,9 @@ intel_dp_add_properties(struct intel_dp *intel_dp, struct drm_connector *_connec
intel_attach_dp_colorspace_property(&connector->base);
}
- if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
+ if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base) ||
+ (intel_dp_is_edp(intel_dp) &&
+ connector->panel.backlight.edp.intel_cap.supports_2084_decode))
drm_connector_attach_hdr_output_metadata_property(&connector->base);
if (HAS_VRR(display))
--
2.53.0
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH] drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR
2026-11-20 5:50 [PATCH] drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR Yitian Shao
@ 2026-11-20 6:15 ` sashiko-bot
2026-11-20 16:10 ` Ville Syrjälä
2026-11-20 14:51 ` Ville Syrjälä
1 sibling, 1 reply; 4+ messages in thread
From: sashiko-bot @ 2026-11-20 6:15 UTC (permalink / raw)
To: Yitian Shao, intel-gfx; +Cc: dri-devel
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [High] drm/i915/dp: broken LSPCON HDR in
intel_dp_compute_hdr_metadata_infoframe_sdp()
- [Medium] drm/i915/dp: unchecked union read of intel_cap in
intel_dp_add_properties()
- [Medium] drm/i915/dp: fastset skips TCON HDR update when SDP
computation is bypassed
--
commit ee42402ceca82543288507d878602e90d81ac27f
Author: Yitian Shao <fighters390@qq.com>
drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR
This commit attaches the HDR_OUTPUT_METADATA property on eDP connectors
whose TCON supports PQ (ST2084) decoding via the Intel AUX HDR interface,
and skips computing the HDR metadata infoframe SDP when the encoder does
not support gamut metadata DIP.
> diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
> index 47741957935ae..3faa5ddb4c60b 100644
> --- a/drivers/gpu/drm/i915/display/intel_dp.c
> +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> @@ -3239,6 +3239,9 @@ intel_dp_compute_hdr_metadata_infoframe_sdp(struct intel_dp *intel_dp,
> int ret;
> struct hdmi_drm_infoframe *drm_infoframe = &crtc_state->infoframes.drm.drm;
>
> + if (!intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
> + return;
> +
> if (!conn_state->hdr_output_metadata)
> return;
[Severity: High]
Does this break HDR metadata transmission on LSPCON outputs?
intel_dp_has_gamut_metadata_dip() unconditionally returns false when
intel_bios_encoder_is_lspcon(encoder->devdata) is true, because LSPCON
connectors defer attaching HDR_OUTPUT_METADATA until
intel_dp_connector_register() probes intel_lspcon_detect_hdr_capability().
LSPCON encoders still rely on intel_dp_compute_hdr_metadata_infoframe_sdp()
to populate crtc_state->infoframes.drm.drm and enable
HDMI_PACKET_TYPE_GAMUT_METADATA in crtc_state->infoframes.enable for
lspcon_write_infoframe(). With this check, HDR metadata is never computed
or sent on LSPCON ports, and the state checker will report an infoframe
mismatch if VIDEO_DIP_ENABLE_GMP_HSW was already active.
[Severity: Medium]
When intel_dp_compute_hdr_metadata_infoframe_sdp() returns early on pre-ICL
port A eDP panels, can a fastset skip programming the TCON HDR parameters?
When userspace toggles HDR_OUTPUT_METADATA,
intel_digital_connector_atomic_check() sets crtc_state->mode_changed = true,
and on platforms with gamut metadata DIP the resulting change in
crtc_state->infoframes.enable causes intel_pipe_config_compare() to reject
fastset and run a full modeset. Returning early here leaves
crtc_state->infoframes.enable unchanged, so commits without another pipe
config change are downgraded to a fastset (update_pipe). On the fastset
path, intel_ddi_update_pipe_dp() calls intel_backlight_update(), which only
calls __intel_backlight_enable() -> intel_dp_aux_hdr_enable_backlight()
when !panel->backlight.enabled:
drivers/gpu/drm/i915/display/intel_backlight.c:intel_backlight_update() {
...
mutex_lock(&display->backlight.lock);
if (!panel->backlight.enabled)
__intel_backlight_enable(crtc_state, conn_state);
mutex_unlock(&display->backlight.lock);
...
}
Because panel->backlight.enabled is already true on an active panel,
INTEL_EDP_HDR_GETSET_CTRL_PARAMS is never updated in the TCON.
> @@ -7018,7 +7021,9 @@ intel_dp_add_properties(struct intel_dp *intel_dp, struct drm_connector *_connec
> intel_attach_dp_colorspace_property(&connector->base);
> }
>
> - if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
> + if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base) ||
> + (intel_dp_is_edp(intel_dp) &&
> + connector->panel.backlight.edp.intel_cap.supports_2084_decode))
> drm_connector_attach_hdr_output_metadata_property(&connector->base);
>
> if (HAS_VRR(display))
[Severity: Medium]
Can reading connector->panel.backlight.edp.intel_cap.supports_2084_decode
here access the wrong union member when the VESA eDP AUX backlight
interface is selected?
In struct intel_panel, panel.backlight.edp is an untagged union between
vesa and intel_cap:
drivers/gpu/drm/i915/display/intel_display_types.h:struct intel_panel {
...
union {
struct {
struct drm_edp_backlight_info info;
bool luminance_control_support;
} vesa;
struct {
bool sdr_uses_aux;
bool supports_2084_decode;
...
}
supports_2084_decode sits at byte offset 1, overlapping
vesa.info.pwm_freq_pre_divider in struct drm_edp_backlight_info. If
intel_edp_backlight_setup() selected the VESA AUX backlight path instead of
the Intel HDR AUX path, drm_edp_backlight_init() populates vesa.info and a
non-zero pwm_freq_pre_divider will falsely attach HDR_OUTPUT_METADATA (and
trigger an out-of-range bool load under UBSan if the divider is >= 2).
--
Sashiko AI review · https://sashiko.dev/#/patchset/tencent_083AD2514CBC235A9F55EE00BDF72D3B9209@qq.com?part=1
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR
2026-11-20 5:50 [PATCH] drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR Yitian Shao
2026-11-20 6:15 ` sashiko-bot
@ 2026-11-20 14:51 ` Ville Syrjälä
1 sibling, 0 replies; 4+ messages in thread
From: Ville Syrjälä @ 2026-11-20 14:51 UTC (permalink / raw)
To: Yitian Shao
Cc: intel-gfx, dri-devel, jani.nikula, sebastian.wick, uma.shankar,
suraj.kandpal
On Fri, Nov 20, 2026 at 01:50:48PM +0800, Yitian Shao wrote:
> intel_dp_has_gamut_metadata_dip() returns false for port A on pre-ICL
> platforms. intel_dp_add_properties() uses that helper to decide whether
> to attach the HDR_OUTPUT_METADATA connector property, so eDP panels on
> those platforms never expose it even when the panel advertises HDR in
> its EDID.
>
> On such panels the HDR metadata is not sent as a gamut metadata SDP, but
> programmed through the Intel AUX HDR interface instead. The TCON is
> already probed during intel_edp_backlight_setup(), which runs before
> intel_dp_add_properties(), so the capability is known at that point.
>
> Attach the property when the panel's TCON advertises PQ (ST2084) decode
> support, so that userspace can enable HDR on these panels.
>
> Since no SDP is transmitted in that case, also bail out early from
> intel_dp_compute_hdr_metadata_infoframe_sdp() when the encoder cannot
> send gamut metadata. Otherwise the computed infoframe state would be
> inconsistent with what is actually programmed, and the state checker
> would complain.
>
> Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/10415
> Signed-off-by: Yitian Shao <fighters390@qq.com>
> ---
> Reviewer notes, not part of the commit message:
>
> Tested on a CometLake-U (Gen9.5) laptop with a 3840x2160 10bpc eDP panel that
> advertises BT2020 + ST2084 in its EDID. With this change the connector gains
> the HDR_OUTPUT_METADATA property, GNOME 50 exposes the HDR switch, and the
> panel enters PQ/BT2020 mode (the DP Colorspace property switches to
> BT2020_RGB). The Intel TCON HDR backlight interface was already in use before
> this change, so the hardware path existed and only the property was missing.
>
> Two things I am not sure about:
>
> 1. intel_dp_aux_write_content_luminance() also returns early when
> !intel_dp_has_gamut_metadata_dip(), which is exactly the PORT_A case where
> the AUX path is used. max_cll/max_fall therefore never reach the TCON. In
> my testing Mutter sends all-zero mastering metadata anyway, so I could not
> tell whether this shows up in practice. Should this check be relaxed too,
> or is it intentional?
My reading of the "spec" is that we should do the content luminance
write. On the platforms where the SDP is used, I believe the SDP
should actually take precedence over the DPCD based information.
>
> 2. panel->backlight.edp.intel_cap shares a union with the VESA backlight
> info, so supports_2084_decode would read unrelated bits if the VESA
> interface had been selected instead of the Intel one. Would you prefer an
> explicit "Intel HDR backlight is in use" check here?
Yeah, we probably want that. Maybe a small helper function that does all
the checks we need, so as to not clog up the if-statement too much.
>
> drivers/gpu/drm/i915/display/intel_dp.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
> index afd4169..49142f4 100644
> --- a/drivers/gpu/drm/i915/display/intel_dp.c
> +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> @@ -3088,6 +3088,9 @@ intel_dp_compute_hdr_metadata_infoframe_sdp(struct intel_dp *intel_dp,
> int ret;
> struct hdmi_drm_infoframe *drm_infoframe = &crtc_state->infoframes.drm.drm;
>
> + if (!intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
> + return;
> +
> if (!conn_state->hdr_output_metadata)
> return;
>
> @@ -6715,7 +6718,9 @@ intel_dp_add_properties(struct intel_dp *intel_dp, struct drm_connector *_connec
> intel_attach_dp_colorspace_property(&connector->base);
> }
>
> - if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
> + if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base) ||
> + (intel_dp_is_edp(intel_dp) &&
> + connector->panel.backlight.edp.intel_cap.supports_2084_decode))
> drm_connector_attach_hdr_output_metadata_property(&connector->base);
>
> if (HAS_VRR(display))
> --
> 2.53.0
--
Ville Syrjälä
Intel
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR
2026-11-20 6:15 ` sashiko-bot
@ 2026-11-20 16:10 ` Ville Syrjälä
0 siblings, 0 replies; 4+ messages in thread
From: Ville Syrjälä @ 2026-11-20 16:10 UTC (permalink / raw)
To: sashiko-reviews; +Cc: Yitian Shao, intel-gfx, dri-devel
On Fri, Nov 20, 2026 at 06:15:06AM +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 3 potential
> issues to consider:
> - [High] drm/i915/dp: broken LSPCON HDR in
> intel_dp_compute_hdr_metadata_infoframe_sdp()
> - [Medium] drm/i915/dp: unchecked union read of intel_cap in
> intel_dp_add_properties()
> - [Medium] drm/i915/dp: fastset skips TCON HDR update when SDP
> computation is bypassed
> --
>
> commit ee42402ceca82543288507d878602e90d81ac27f
> Author: Yitian Shao <fighters390@qq.com>
>
> drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR
>
> This commit attaches the HDR_OUTPUT_METADATA property on eDP connectors
> whose TCON supports PQ (ST2084) decoding via the Intel AUX HDR interface,
> and skips computing the HDR metadata infoframe SDP when the encoder does
> not support gamut metadata DIP.
>
> > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
> > index 47741957935ae..3faa5ddb4c60b 100644
> > --- a/drivers/gpu/drm/i915/display/intel_dp.c
> > +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> > @@ -3239,6 +3239,9 @@ intel_dp_compute_hdr_metadata_infoframe_sdp(struct intel_dp *intel_dp,
> > int ret;
> > struct hdmi_drm_infoframe *drm_infoframe = &crtc_state->infoframes.drm.drm;
> >
> > + if (!intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
> > + return;
> > +
> > if (!conn_state->hdr_output_metadata)
> > return;
>
> [Severity: High]
> Does this break HDR metadata transmission on LSPCON outputs?
>
> intel_dp_has_gamut_metadata_dip() unconditionally returns false when
> intel_bios_encoder_is_lspcon(encoder->devdata) is true, because LSPCON
> connectors defer attaching HDR_OUTPUT_METADATA until
> intel_dp_connector_register() probes intel_lspcon_detect_hdr_capability().
> LSPCON encoders still rely on intel_dp_compute_hdr_metadata_infoframe_sdp()
> to populate crtc_state->infoframes.drm.drm and enable
> HDMI_PACKET_TYPE_GAMUT_METADATA in crtc_state->infoframes.enable for
> lspcon_write_infoframe(). With this check, HDR metadata is never computed
> or sent on LSPCON ports, and the state checker will report an infoframe
> mismatch if VIDEO_DIP_ENABLE_GMP_HSW was already active.
Looks legit. I think the correct solution would be to drop
the LSPCON check from has_gamut_metadata_dip() since it's
clearly nonsense. But then intel_dp_add_properties()
will need to do the LSPCON check itself so that it doesn't
attache the HDR metadata to the LSPCON there.
And I think for good measure we also want to add a
has_gamut_metadata_dip() check into
intel_lspcon_detect_hdr_capability(), just in case someone
has ever hooked up an on board LSPCON to a port that doesn't
support that DIP. This should be a separate patch though
since it's a separate pre-existing issue.
>
> [Severity: Medium]
> When intel_dp_compute_hdr_metadata_infoframe_sdp() returns early on pre-ICL
> port A eDP panels, can a fastset skip programming the TCON HDR parameters?
>
> When userspace toggles HDR_OUTPUT_METADATA,
> intel_digital_connector_atomic_check() sets crtc_state->mode_changed = true,
> and on platforms with gamut metadata DIP the resulting change in
> crtc_state->infoframes.enable causes intel_pipe_config_compare() to reject
> fastset and run a full modeset. Returning early here leaves
> crtc_state->infoframes.enable unchanged, so commits without another pipe
> config change are downgraded to a fastset (update_pipe). On the fastset
> path, intel_ddi_update_pipe_dp() calls intel_backlight_update(), which only
> calls __intel_backlight_enable() -> intel_dp_aux_hdr_enable_backlight()
> when !panel->backlight.enabled:
This looks correct as well. And I think it highlights that our HDR
metadata fastset support is already broken. Well, it kinda works
for changing the metadata (except when the backlight interface is
involved as is the case here), but not for switching between HDR
and SDR modes.
I say "kinda works" above because in general the whole .update_pipe()
thing is a half baked hack. There is no real consideration for any
differences in double buffering behaviour of DIPs (or a complete lack
of double buffering as is the case for many DIPs). .update_pipe() is
also called outside the vblank evasion critical section, so even
the double buffered DIPs (or anything else we do there) will probably
not sync up with the rest of the atomic commit on the CRTC.
So how to fix this specific AUX backlight problem... Maybe we need
a crtc_state->backlight_changed or something? This might in fact
also be relevant for the recently proposed LUMINANCE connector
property (which I've not had any time to actually consider).
>
> drivers/gpu/drm/i915/display/intel_backlight.c:intel_backlight_update() {
> ...
> mutex_lock(&display->backlight.lock);
> if (!panel->backlight.enabled)
> __intel_backlight_enable(crtc_state, conn_state);
>
> mutex_unlock(&display->backlight.lock);
> ...
> }
>
> Because panel->backlight.enabled is already true on an active panel,
> INTEL_EDP_HDR_GETSET_CTRL_PARAMS is never updated in the TCON.
>
> > @@ -7018,7 +7021,9 @@ intel_dp_add_properties(struct intel_dp *intel_dp, struct drm_connector *_connec
> > intel_attach_dp_colorspace_property(&connector->base);
> > }
> >
> > - if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base))
> > + if (intel_dp_has_gamut_metadata_dip(&dp_to_dig_port(intel_dp)->base) ||
> > + (intel_dp_is_edp(intel_dp) &&
> > + connector->panel.backlight.edp.intel_cap.supports_2084_decode))
> > drm_connector_attach_hdr_output_metadata_property(&connector->base);
> >
> > if (HAS_VRR(display))
>
> [Severity: Medium]
> Can reading connector->panel.backlight.edp.intel_cap.supports_2084_decode
> here access the wrong union member when the VESA eDP AUX backlight
> interface is selected?
>
> In struct intel_panel, panel.backlight.edp is an untagged union between
> vesa and intel_cap:
>
> drivers/gpu/drm/i915/display/intel_display_types.h:struct intel_panel {
> ...
> union {
> struct {
> struct drm_edp_backlight_info info;
> bool luminance_control_support;
> } vesa;
> struct {
> bool sdr_uses_aux;
> bool supports_2084_decode;
> ...
> }
>
> supports_2084_decode sits at byte offset 1, overlapping
> vesa.info.pwm_freq_pre_divider in struct drm_edp_backlight_info. If
> intel_edp_backlight_setup() selected the VESA AUX backlight path instead of
> the Intel HDR AUX path, drm_edp_backlight_init() populates vesa.info and a
> non-zero pwm_freq_pre_divider will falsely attach HDR_OUTPUT_METADATA (and
> trigger an out-of-range bool load under UBSan if the divider is >= 2).
>
> --
> Sashiko AI review · https://sashiko.dev/#/patchset/tencent_083AD2514CBC235A9F55EE00BDF72D3B9209@qq.com?part=1
--
Ville Syrjälä
Intel
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-11-20 16:10 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-20 5:50 [PATCH] drm/i915/dp: Attach HDR_OUTPUT_METADATA on eDP when TCON supports HDR Yitian Shao
2026-11-20 6:15 ` sashiko-bot
2026-11-20 16:10 ` Ville Syrjälä
2026-11-20 14:51 ` Ville Syrjälä
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.