After upgrading a Toshiba Satellite Z30-A laptop from Debian 9 (Stretch) with Linux 4.9 to Debian 10 (Buster) which is using the Linux 4.19 kernel the external monitor, a Dell U2715H, would no longer be driven at its native resolution of 2560x1440 and a reduced resolution of 2048x1152 was used instead. Already with Debian Stretch a custom modeline in the X.Org configuration was required to compel the software into using the appropriate resolution:
# 2560x1440 59.95 Hz (CVT 3.69M9-R) hsync: 88.79 kHz; pclk: 241.50 MHz
Modeline "2560x1440R" 241.50 2560 2608 2640 2720 1440 1443 1448 1481 +hsync -vsync
For the investigation with Debian Buster I enabled verbose logging:
# Enable DRM driver and KMS debug messages
echo 0x6 > /sys/module/drm/parameters/debug
# Disable
echo 0 > /sys/module/drm/parameters/debug
The messages showed that the pixel clock rate was being limited to 165 MHz:
[ 1428.752898] [drm:drm_dp_dual_mode_detect [drm_kms_helper]] DP dual mode HDMI ID: DP-HDMI ADAPTOR\004 (err 0)
[ 1428.753676] [drm:drm_dp_dual_mode_detect [drm_kms_helper]] DP dual mode adaptor ID: ff (err 0)
[ 1428.753728] [drm:intel_hdmi_set_edid [i915]] DP dual mode adaptor (type 1 HDMI) detected (max TMDS clock: 165000 kHz)
After reading plenty of Linux kernel code (primarily in drivers/gpu/drm and
drivers/gpu/drm/i915) and comparing changes between Linux 4.9
and 4.19, as well as in xf86-video-intel from Debian Stretch,
I came up with an experimental kernel code change which made the native
resolution work again.
As I wasn't sure whether I was dealing with a software bug or have been driving
the hardware out of specs for over two years I filed bug #111553 with the
i915 graphics driver developers. Upstream developer Ville Syrjälä
confirmed that the hardware indeed is supposed to only support pixel clock
frequencies up to 165 MHz. Despite the monitor being attached via a HDMI-A
cable there must be a converter from DisplayPort to HDMI or the other way
around in the signal path (possibly in the monitor as it also has a DisplayPort
connector).
Considering that everything was working well before the upgrade I wanted to
achieve a working configuration under Debian Buster without changing hardware.
One way would've been to rebuild the Debian kernel package with
a patch as already done for my experiments (albeit for those
the i915.ko module was sufficient). Maintaining a local kernel package is
a maintenance burden I'd rather avoid and thus looked for another solution.
The module uses Kernel probes to hooks itself into the return path
of the drm_dp_dual_mode_detect function. This
function is used by the i915 driver as part of the decision whether the pixel
clock rate should be restricted and is, this is critical, exported.
The fake_mode parameter defaults to DRM_DP_DUAL_MODE_NONE ("there is no
DisplayPort conversion"), but can be set to any value.
There is packaging information for Debian using DKMS.