On Thu, 17 Sep 2026, Theo Andersen Carton <[email protected]> wrote: > On dual-GPU Retina MacBooks the internal panel is muxed to the discrete > GPU at boot, so Apple's firmware never lights up eDP on the Intel side. > As a result it never sets DDI_A_4_LANES, and intel_ddi_max_lanes() reads > back a source maximum of two lanes for port A. > > intel_dp_max_common_lane_count() then takes min3(source_max = 2, > sink_max = 4, ...) = 2, giving 2 x 2.7 Gbps x 8/10 = 4.32 Gbps of link > bandwidth. The panel's native 2880x1800@60 8bpc mode needs 8.1 Gbps, so > every mode is rejected with MODE_CLOCK_HIGH and the connector comes up > with no usable modes at all -- the panel stays dark whenever i915 is > made to drive it, which looks like a hardware limitation but is not. > > The lanes are physically there. Reading the panel's DPCD from the AMD > side, where amdgpu drives it correctly, reports MAX_LANE_COUNT = 0x84, > i.e. 4 lanes, and amdgpu's own link_settings shows it running the panel > at 4 lanes / HBR. macOS likewise drives this panel from the iGPU at its > native resolution, so all four lanes are routed through the gmux to > DDI A. Only the register bit is unset. > > intel_ddi_a_force_4_lanes() already exists for exactly this situation -- > its neighbouring comment notes that "Some BIOS might fail to set this > bit on port A if eDP wasn't lit up at boot" -- but it only covers > Broxton and Geminilake. Extend it with a DMI quirk so affected Macs get > the correct lane count too. > > The DMI match is deliberately narrow. Other dual-GPU models in the same > family are very likely affected identically, but they are untested, so > widen the match only as reports come in. > > Tested on a MacBookPro11,5 (Haswell Crystal Well, Iris Pro 5200 + > Radeon R9 M370X): with the quirk, DDI A comes up as 4 lanes, link > training passes at link rate 270000 / lane count 4, and i915 drives the > internal panel at its native 2880x1800. > > Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/17087 > Signed-off-by: Theo Andersen Carton <[email protected]>
Sorry, this fell through the cracks. It looks good to me, but I kicked the CI to run. We don't actually have Macs in CI, but it's more to ensure we're not regressing anything. Reviewed-by: Jani Nikula <[email protected]> Cc: [email protected] > --- > Numbering note: neither earlier posting was labelled, so to match the > numbering already used on the list this is v3. v1 [1] was pasted into a > mail client by hand, which hard-wrapped the diff and detabbed it, leaving > it unapplicable. v2 [2] was the same patch resent with git send-email, > unlabelled. > > v3: > - No functional change since v2. > - Commit message: the machine is Haswell (Crystal Well), not Broadwell as > v1/v2 claimed. The DMI match was always by model, so no code change. > - Added Closes: tag for the bug report, which carries the failing-case > logs, the VBT dump and the EDID. > > [1] > https://lore.kernel.org/r/canfnrsphik7ii28xpdlkomyhszcj9xz6ct-knx+oiif+8n-...@mail.gmail.com > [2] > https://lore.kernel.org/lkml/[email protected]/ > drivers/gpu/drm/i915/display/intel_ddi.c | 7 +++++ > drivers/gpu/drm/i915/display/intel_quirks.c | 34 +++++++++++++++++++++ > drivers/gpu/drm/i915/display/intel_quirks.h | 1 + > 3 files changed, 42 insertions(+) > > diff --git a/drivers/gpu/drm/i915/display/intel_ddi.c > b/drivers/gpu/drm/i915/display/intel_ddi.c > index 02a53c9848e1..f746287830b3 100644 > --- a/drivers/gpu/drm/i915/display/intel_ddi.c > +++ b/drivers/gpu/drm/i915/display/intel_ddi.c > @@ -4976,6 +4976,13 @@ static bool intel_ddi_a_force_4_lanes(struct > intel_digital_port *dig_port) > if (display->platform.geminilake || display->platform.broxton) > return true; > > + /* > + * Machines whose firmware leaves eDP dark on the Intel side never set > + * the bit, so trust the quirk rather than the register. > + */ > + if (intel_has_quirk(display, QUIRK_DDI_A_FORCE_4_LANES)) > + return true; > + > return false; > } > > diff --git a/drivers/gpu/drm/i915/display/intel_quirks.c > b/drivers/gpu/drm/i915/display/intel_quirks.c > index 33245f44c0d5..feb28956a013 100644 > --- a/drivers/gpu/drm/i915/display/intel_quirks.c > +++ b/drivers/gpu/drm/i915/display/intel_quirks.c > @@ -100,6 +100,20 @@ static void quirk_disable_psr2(struct intel_display > *display) > drm_info(display->drm, "PSR2 support not currently available for this > setup, applying disable PSR2 quirk\n"); > } > > +/* > + * Dual-GPU Macs boot with the internal panel muxed to the discrete GPU, so > + * the firmware never lights up eDP on the Intel side and consequently never > + * sets DDI_A_4_LANES -- even though all four lanes are wired through the > gmux > + * to DDI A. Without the bit, port A is capped at two lanes, which is not > + * enough bandwidth for the panel's native mode, leaving the connector with > no > + * usable modes at all. > + */ > +static void quirk_ddi_a_force_4_lanes(struct intel_display *display) > +{ > + intel_set_quirk(display, QUIRK_DDI_A_FORCE_4_LANES); > + drm_info(display->drm, "Applying DDI A force 4 lanes quirk\n"); > +} > + > struct intel_quirk { > int device; > int subsystem_vendor; > @@ -142,6 +156,13 @@ static int intel_dmi_no_pps_backlight(const struct > dmi_system_id *id) > return 1; > } > > +static int intel_dmi_ddi_a_force_4_lanes(const struct dmi_system_id *id) > +{ > + DRM_INFO("DDI A is 4 lanes despite firmware on %s\n", id->ident); > + > + return 1; > +} > + > static const struct intel_dmi_quirk intel_dmi_quirks[] = { > { > .dmi_id_list = &(const struct dmi_system_id[]) { > @@ -188,6 +209,19 @@ static const struct intel_dmi_quirk intel_dmi_quirks[] = > { > }, > .hook = quirk_no_pps_backlight_power_hook, > }, > + { > + .dmi_id_list = &(const struct dmi_system_id[]) { > + { > + .callback = intel_dmi_ddi_a_force_4_lanes, > + .ident = "Apple MacBookPro11,5", > + .matches = {DMI_EXACT_MATCH(DMI_SYS_VENDOR, > "Apple Inc."), > + DMI_EXACT_MATCH(DMI_PRODUCT_NAME, > "MacBookPro11,5"), > + }, > + }, > + { } > + }, > + .hook = quirk_ddi_a_force_4_lanes, > + }, > }; > > static struct intel_quirk intel_quirks[] = { > diff --git a/drivers/gpu/drm/i915/display/intel_quirks.h > b/drivers/gpu/drm/i915/display/intel_quirks.h > index 970a4fe52faf..d2b8e9183c62 100644 > --- a/drivers/gpu/drm/i915/display/intel_quirks.h > +++ b/drivers/gpu/drm/i915/display/intel_quirks.h > @@ -23,6 +23,7 @@ enum intel_quirk_id { > QUIRK_EDP_LIMIT_RATE_HBR2, > QUIRK_DISABLE_EDP_PANEL_REPLAY, > QUIRK_DISABLE_PSR2, > + QUIRK_DDI_A_FORCE_4_LANES, > }; > > void intel_init_quirks(struct intel_display *display); > > base-commit: 50d05c7c76c96b90462f24debacca971d2e86713 -- Jani Nikula, Intel
