Valve's Steam Frame kernel carries an OLED panel driver, and it is not the 3552x3840 one
The Steam Frame kernel tree does carry an OLED panel driver, and it is not the 3552x3840 panel that a mid-generation OLED Frame was read out of this week. panel-deckard-oled.c is on a branch of its own, declares a single mode of 1776x1920 at 90Hz with a 120Hz variant commented out, and the newest commit on that branch is dated February 10.
Valve's Steam Frame page carries nine short videos, one of them on adjusting the IPD. Image: Valve
The datamine behind the claim is Steam Frame Hub's, published on October 14, and it says the panel "has a resolution of 3552x3840 per eye and appears to be OLED". Those numbers are real, and they belong to a different file, panel-deckard-plano.c, in which the string OLED does not occur once.
Plano's two panel constants are BACKPLANE_HAA 3552 and BACKPLANE_VAA 3840, the active area of a backplane rather than a mode anything can run, and each is exactly twice the 1776x1920 the driver enables. Of its nine mode-enable switches only GLOBAL_2X_MODE_EN is set to true, which leaves 1776x1920 at 90, 72, 60 and 45Hz as the modes on offer, each with display stream compression.
A 30Hz mode is held behind a low-bandwidth test switch and the 2368x2560 modes behind a 1.5X scaling switch, both false, and Valve's comment on the 90Hz version of 2368x2560 says it "needs changes in drivers/gpu/drm/msm/disp/dpu1/dpu_encoder_phys_vid.c to handle fractional DSC width". The seven foveated modes are off as well, each one marked "TODO: WIP -- Not working yet".
The production board file still asks for the LCD
sm8650-mp.dts, the mass-production board, declares compatible = "deckard-panel", and so do the DV2 and EV3D boards. The board that asks for deckard-panel-plano is sm8650-ev3dplano.dts, whose own model string reads "SM8650 EV2 dphy plano rev1 4slam 2et", an engineering revision behind DV1, DV2 and MP.
deckard-panel is the shipping driver, identified in its own module string as the "Sharp LS0DAS7232 DSI Panel Driver with BOE support", and it registers seven modes, every one of them at 2160x2160: 120, 144, 108, 96, 90, 80 and 72Hz. Valve's own Frame page says "2160 x 2160 LCD (per eye)" and "72-144Hz refresh rate (144Hz experimental)", which is the same panel the code describes.
120Hz is the mode the driver flags DRM_MODE_TYPE_PREFERRED, so it is what the panel asks for unless something else chooses. 72Hz is the one conditional entry, added only when the board sets a low-fps-support property, and the mass-production and DV2 boards both set it.
The OLED work and the Plano work never met
On the deckard-oled branch the panel makefile builds panel-deckard-oled.o where the shipping branch builds panel-deckard-plano.o, and neither the Plano driver nor sm8650-ev3dplano.dts exists on that branch at all. The OLED panel's board file is sm8650-ev3doled.dts, model "SM8650 EV2 dphy oled rev1 4slam 2et", and the mass-production board on that same branch still declares deckard-panel.
The two drivers also treat light differently. The OLED one notes that an "OLED display does not use backlight drivers", compiles that code out and switches on the panel's own AVEE and ELVSS rail at 6.3V instead, while Plano registers a backlight device and asks its board for four bias rails named left-pos, left-neg, right-pos and right-neg, the same set the production LCD node carries.
Nothing Plano-specific has landed since July
Plano's history on the shipping branch is four commits. montyg at valvesoftware.com added it on March 5 under the message "Added Plano panel support Global 2.0x working Global 1.5x WIP Foveal modes WIP CPHY support WIP", Linaro's Neil Armstrong posted fixups on April 27 and July 30, and the only change since is a debugfs bounds check across the deckard panels from Valve's Jeremy Selan on October 3.
The OLED branch is quieter still: deckard-oled stops on February 10 and km/deckard-oled on February 5. None of that says Valve has shelved anything, because a kernel tree is where abandoned boards, parked drivers and shipping hardware all sit side by side.
This is a reading of code rather than an announcement, and the provenance needs saying once: the browsable copy of the tree is evlaV's, which labels its SteamOS repositories unofficial mirrors and rebuilds them automatically four times a day. Valve distributes the same kernel itself, as a 2.4GiB source package on its own server, and its public GitLab carries the Frame's CAD and electrical references but no kernel project at all; those CAD files are where the PCIe Gen4 lane in the camera port came from.
What ties the readable copy to Valve is the commits. The package Valve put up on October 14 is named linux-72-deckard-7.2.0+g152ecbcff010, and that commit, the one at the top of the mirrored branch, is a defconfig fix authored from a valvesoftware.com address. Valve has announced no new headset and no new panel, and for anyone deciding whether to wait, the thing with a date on it is the queue: Valve now tells reservation holders whether their order email lands by December or in 2027.
