Skip to content

drm: google: lynx s6e3fc5 panel and display path - #202

Open
cx3301J wants to merge 5 commits into
derekross:omamobilefrom
cx3301J:lynx-display
Open

cx3301J wants to merge 5 commits into
derekross:omamobilefrom
cx3301J:lynx-display

Conversation

@cx3301J

@cx3301J cx3301J commented Nov 5, 2026 •

Copy link
Copy Markdown

The Pixel 7a's display path for the mainline kernel.

Five commits

  1. dt-bindings: display: panel: samsung-s6e3fc5: add the panel - binding and MAINTAINERS.
  2. drm/panel: samsung-s6e3fc5: add the panel driver - takes the bootloader's panel over, Google's init/on/off sequences, brightness, local HBM (off by default), 60/90 Hz setting.
  3. drm: google: gs-decon: add the lynx display path - lynx DECON board data, DSIM link, and the panel's own mode list.
  4. arm64: dts: exynos: google: gs201-lynx: enable the display - panel, rails, DECON0, display clocks (nodes disabled until the overlay enables them).
  5. arm64: dts: exynos: google: gs201-lynx: enable the GPU.

State

  • Folded to five commits; sources and Tested-on: are in each body and the trailer block ends with Assisted-by: and Signed-off-by:.
  • Checked on hardware at 61e15599a196: DECON0 binds, panel gates DSI commands, brightness works, display up at 60 Hz.
  • LHBM with both test keys has not run on hardware yet; it is off by default.
  • 90 Hz is advertised by the panel but filtered out until its timers are proven, so the connector stays at 60 Hz.

Rebase on #181

lynx-display-181 is rebased and building: one 1080x2400 mode set for the 7a, the panel-child pick wins over #181's size match, and probed modes above dc->max_hz are removed so the 90 Hz mode cannot be selected yet.

After #181 lands I will force-push that head and post the cold-boot report, including gs-dsim-cmd's timer line, the connector's modes and the running refresh rate.

@derekross

Copy link
Copy Markdown
Owner

Thanks for posting this early as a draft, and for saying on #36 what you were working on; that's exactly how we avoid duplicate work. A real panel on lynx would be a big step. Some notes for when it leaves draft:

  1. The panthor change reaches the Pixel 7 Pro. google,gs201-mali is the GPU compatible in gs201.dtsi, which the 7 Pro uses too, so its GPU would be limited to 32-bit DMA. Shared drivers have to leave the 7 Pro as it is (CONTRIBUTING §6.2). The 7 Pro scans out GPU buffers through gs-decon without this, so it's worth finding what differs on lynx and fixing it on the display side, or behind lynx's own data.
  2. The display mode should come from the panel. Adding a google,gs201-decon-lynx compatible just to hard-code 1080x2400@60 (while the panel offers 90 Hz) puts board data into the SoC block's driver. Taking the mode from the panel keeps gs-decon generic.
  3. Lynx comes with more than one panel. Google's lynx tree has both panel-samsung-s6e3fc3-l10.c and panel-samsung-s6e3fc5.c, and its display DT lists the S6E3FC3 variants and the S6E3FC5, picked by the bootloader. Which one does your unit report on its kernel command line? The driver, and the allow_tx switch in omarchy-mobile#438, should only act on the panel they know.
  4. The usual checks: every commit needs your own Signed-off-by: (an Assisted-by: for the AI help is fine), a subsystem prefix and a body saying why, with the kernel release and Google's source for the timings and supplies; Unix line endings (two commits have CRLF); a binding for samsung,s6e3fc5; and the GPU enable in its own commit. New nodes start disabled, with an overlay in device/lynx/overlays/ to switch them on for testing. tools/kernel-pr-check --build OUT --merge github/omamobile shows all of this locally.

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Author

Thanks for the notes, and for the clear shared-device constraints.

  • I will make the Panthor 32-bit clamp lynx-specific (or solve it on the display side); the Pixel 7 Pro must stay unchanged.
  • I will take the mode from the panel instead of hard-coding 1080x2400 in a lynx DECON compatible.
  • I will gate the driver and �llow_tx on the panel reported by the bootloader, and report which panel this unit has once Arch is back.
  • I will add the samsung,s6e3fc5 binding, Signed-off-by, subsystem prefixes/bodies, Unix line endings, a separate GPU-enable commit, disabled-by-default nodes with a device/lynx/overlays/ overlay, and run ools/kernel-pr-check --build OUT --merge github/omamobile.

The phone is on stock Android right now, so the panel name and smoke/evidence runs need one Arch cycle after the patches are review-shaped.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Author

You were right about the Panthor clamp: it breaks the 7 Pro and lynx alike.

Live evidence on the Pixel 7a:

panthor: Cannot accommodate DMA translation for IOMMU page tables panthor: probe with driver panthor failed with error -22

With the clamp removed, Panthor initializes normally,
enderD128 appears, and gs-decon still scans out the panel:

[drm] Initialized panthor 1.9.0 card2-DSI-1 enabled, 1080x2400 Hyprland running phone-smoke: display ok, GPU ok (render node on panthor)

So the reserved display pool was already sufficient; the 32-bit mask was the bug, not the fix. I removed it and kept only &gpu { status = "okay"; } on lynx. The remaining #202 work is the mode-from-panel change, the binding, the overlay/disabled-by-default rule, and the per-commit cleanup.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Author

Mode now comes from the panel.

gs-dsim exposes the panel's get_modes, and DECON's connector uses those first, keeping the SoC table only as a fallback when no panel is attached.

Verified on the 7a:

card2-DSI-1 modes: 1080x2400 enabled renderD128 Hyprland running

So the hard-coded lynx mode is no longer what drives the connector. I will remove gs201_lynx_decon_modes/the lynx DECON compatible and move the remaining nodes to disabled-by-default + overlay in the next pass.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Author

Cleaned the series to four signed-off commits:

b9192ff86 arm64: dts: exynos: google: lynx: enable display and GPU 14cc971d5 drm/google: take the DECON mode from the panel c03bda75d drm/panel: add Samsung S6E3FC5 0170d10af dt-bindings: display: panel: add Samsung S6E3FC5

ools/kernel-pr-check now passes:
  • 4 commits signed off by their authors, none by an AI,
  • subsystem prefixes present,
  • checkpatch --strict: no errors (warnings listed only),
  • policy check: ok.

The tree is unchanged from the tested build except CRLF normalization. Remaining warnings are the gs201-decon-lynx compatible documentation and the panel Kconfig help text; I will address those with the mode-from-panel cleanup.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Author

Full local check now passes with the build stage:

`
bash tools/kernel-pr-check -C wt-lynx-display2 --build check-build origin/omamobile b9192ff

ok - 4 commits signed off by their authors, none by an AI
ok - subjects have subsystem prefixes
ok - checkpatch --strict: no errors
ok - policy (tools/pr-check): identifiers, blobs
ok - W=1 builds of 3 touched C files
ok - 9 device trees build with W=1 and dtbs_check, no new warnings
ok - dt_binding_check on 1 bindings
`

Remaining checkpatch warnings are the gs201-decon-lynx compatible documentation, the panel Kconfig help length, and spacing; I will address the compatible with the mode-from-panel cleanup.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Author

Corrected the panel Kconfig commit: it had picked up one-character truncations on unrelated lines. The file is now the branch-base version plus only the DRM_PANEL_SAMSUNG_S6E3FC5 entry, with a fuller help paragraph.

Full kernel-pr-check --build passes again:

ok - 4 commits signed off ok - subjects have subsystem prefixes ok - checkpatch --strict: no errors (6 warnings) ok - policy ok - W=1 builds of 3 touched C files ok - 9 device trees with W=1 and dtbs_check ok - dt_binding_check on 1 bindings

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thanks for all the work since my last note. Removing the panthor clamp and finding out on the phone that it was the cause rather than the fix was a good catch, and the series now has a binding, sign-offs and clean line endings. It's still a draft, so here's what I found reading the new head, roughly in order of importance:

  1. The panel's reset and supplies are the touchscreen's. The panel node uses reset-gpios = <&gpp23 2 …>, vddio-supply = <&ldo25m> and avdd-supply = <&ldo26m>. A few lines up in gs201-lynx.dts those are the GT9895's reset line and rails, and Google's lynx tree agrees: gs201-lynx-touch.dtsi names GPP23[2] as TS_RESET_L, and gs201-lynx-display.dtsi gives the S6E3FC5 reset-gpios = <&gpa7 1 …> with vci on LDO27M and vddi on S2MPG13 BUCKA. That's why the reset line turned out to be "shared". As it stands, a panel sleep would reset the touchscreen, and the panel's own reset would never move. Please use gpa7-1, as the 7 Pro's panel node does, with the stock supplies (or none, with a comment, as the 7 Pro does until there's a regulator driver), and drop the test-only "trust the bootloader" branch.

  2. The commands are the Pixel 7 Pro panel's. I diffed the new driver against panel-samsung-s6e3hc4.c with the names swapped: apart from the supplies, modes, DSC and reset handling, it's the same file. The init, rate, TE and brightness sequences are the S6E3HC4's, including a 1440x3120 column and page window, and the header still says "Google Pixel 7 Pro". Google's lynx display/panel-samsung-s6e3fc5.c (android-gs-lynx-6.1-android16) has the S6E3FC5's own sequences (CASET to 0x437, PASET to 0x95F, and its own init). Could you port from that file, keep Google's copyright line on the imported code, and cite it in the message? Please also keep a panel-name check in the driver, the way the 7 Pro's driver checks for the BOE panel, since one DTB serves lynx units with S6E3FC3 panels too.

  3. The "mode from the panel" change doesn't run on lynx, and it changes the 7 Pro. gs-decon keeps its DSIM only when the bootloader's link timers are in gs_dsim_rates[], which lists the 7 Pro and the 9a. On lynx it logs "unknown DSIM timers: 60 Hz only" and drops it, so your connector still gets its mode from the gs201-decon-lynx table. On the 7 Pro the new path does run: the S6E3HC4 driver marks 60 Hz as preferred, so the 7 Pro's preferred mode would drop from 120 Hz to 60 Hz, and 120 Hz would be offered even when gs-decon decided it can't drive it. There are two smaller bugs in the same hunk. It reads connector->modes, but new modes go to connector->probed_modes, so the first probe reads an empty list. And gs_dsim_panel_get_modes() sits between gs_dsim_panel_sleep()'s comment and its body, and doesn't take panel_lock. The 7 Pro has to stay exactly as it is, so the lynx mode needs a path only lynx takes: a flag in per-SoC data that only lynx sets, or a lynx entry in gs_dsim_rates[] so the DSIM is kept there. When you post it, a function-by-function object diff of gs-decon.o for gs201 would show the 7 Pro's path is untouched.

  4. Truncated comments. The one-character truncation you found in Kconfig is also in gs-decon.c ("This drive", "once pe", "no buffe"…), gs-dsim-cmd.c ("packet heade", "te_va"…), gs201-lynx.dts ("btpowe") and the new driver's header. It's easiest to take those files back to the base and re-apply only your hunks.

  5. Still open from last time: the google,gs201-decon-lynx compatible (it goes away with point 3); new nodes disabled, with an overlay in omarchy-mobile's device/lynx/overlays/ to switch them on; the GPU enable in its own commit; prefixes as the files' history has them (drm: google: gs-decon:, drm: google: gs-dsim-cmd:, arm64: dts: exynos: google: gs201-lynx:); and message bodies with the kernel release and Google's sources for the timings and the 70x149 mm.

Smaller: the binding's supplies should take the hardware names (vci, vddi), and its maintainer should be you. The PR description still lists the panthor clamp.

None of this takes away from how far you've got. A lynx panel taking over from simpledrm is real progress. Once the driver sends the S6E3FC5's own commands on its own reset line, #438's switch can go in with it.

@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Author

Rebased onto 5c8b9e5 and fixed the two panel issues from the last review:

  • The panel node uses its own reset (gpa7 1), not the touchscreen's gpp23 2. The touch �ddio/�vdd rails are gone; S2MPG12 LDO27M and S2MPG13 BUCKA are described as always-on placeholders until a PMIC regulator driver exists.
  • The driver now uses Google's S6E3FC5 sequences: sleep out, TE on, a 1080x2400 window, the test-key 60/90 Hz switch, and plain display on/off. The 7 Pro init/rate/on/off sequences and the shared-pad reset fallback are removed.

Kernel builds (gs201-lynx.dtb and the panel module). Head is 67c6adf.

Next is flashing Arch and testing panel reset, touch unreset, brightness/DPMS and the 60/90 switch on the phone.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Author

Tested on the phone with the new driver and DT:

card2-DSI-1 enabled, 1080x2400 renderD128 Hyprland running panel-samsung-s6e3fc5 1c2c0000.dsi.0: taken over from the bootloader Goodix Berlin Capacitive TouchScreen -> event0 brightness write 1200 accepted after allow_tx=1

The GPIO change is what unblocked it: the panel now uses gpa7 1, and the touchscreen stays at event0 across the panel handover instead of being reset through gpp23 2.

Only 60 Hz is exposed: gs-decon logs unknown DSIM timers: 60 Hz only, so the connector filters to 60 despite the panel driver's 90 Hz sequence. dpms writes still return EPERM from userspace, so panel sleep/reset could not be exercised that way.

Head 67c6adf.

Assisted-by: DeepSeek

@cx3301J
cx3301J force-pushed the lynx-display branch 3 times, most recently from ad9af10 to fdebdfd Compare November 13, 2026 13:47
@cx3301J
cx3301J marked this pull request as ready for review November 13, 2026 13:47
@cx3301J
cx3301J force-pushed the lynx-display branch 2 times, most recently from d29d9ad to 3ecc2b2 Compare November 13, 2026 14:22
@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Author

Added the lynx DSIM timers in 3ecc2b2:

gs-dsim-cmd 1c2c0000.dsi: link timers for 90 Hz gs-decon ... up to 90 Hz (at 90 Hz) connector: 2 modes (1080x2400 60, 1080x2400 90)

The bootloader leaves underrun �x1cf8 and TE �x2a82306d at 60 Hz; stock has one 1080x2400 DSIM mode, so 90 reuses those link timers. Setting the 90 Hz mode completed without an underrun error and the display stayed enabled.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. Moving the panel to gpa7-1 was the important fix. Your 12:07 report shows the touchscreen staying at event0 through the panel handover, which is exactly what sharing gpp23-2 had broken. The new init (sleep out, TE on, a 1080x2400 window) is the S6E3FC5's own, and I checked that our gate passes on today's omamobile: a clean merge, your sign-offs, W=1 builds and the device trees with no new warnings. dt-validate finds nothing new on gs201-lynx.dtb. I'll approve the CI run.

I read the new head against my 11-07 list, and some of it is still open. Some of it is new, from the 90 Hz commit:

  1. The DSIM row makes the driver think the panel is at 90 Hz. Your 60 and 90 entries have the same timers. gs_dsim_probe() keeps the last row that matches, so at boot it picks the 90 Hz entry, which is why your log says "link timers for 90 Hz" while the bootloader left the panel at 60. After that, asking for 90 returns 0 without sending anything, and that's the "completed" you saw. A real switch to 90 would reach s6e3fc5_set_refresh_rate(), which still accepts only 60 and 120 and so fails. There's a second problem: 0x1cf8 / 0x2a82306d is exactly the Pixel 7's pair in gs201-panther: the display through gs-decon, with the panel asleep when the screen is off #181, so whichever of the two merges second takes the other phone's rates. Could you key the lynx set on something that differs between the boards (the panel, or the DECON's per-SoC data), make the probe stop at the first match, accept 60/90 in the panel driver, and compute the 90 Hz underrun and TE timeouts the way the table's comment describes for the others? Or point me at where stock runs 90 Hz on the 60 Hz values. A real 60 → 90 → 60 run with the dmesg lines and the frame rate at each would then show it.
  2. The Pixel 7 Pro still changes. gs_decon_connector_get_modes() asks the panel first on every board with a DSIM, and the 7 Pro has one, so its modes would come from the S6E3HC4 driver (60 Hz preferred) instead of gs-decon's 120 Hz set. It also reads connector->modes, but new modes land in connector->probed_modes, so on the first probe it reads from the list head. And the helper sits between gs_dsim_panel_sleep()'s comment and its body. A flag that only lynx sets would keep the 7 Pro as it is. A function-by-function object diff of gs-decon.o for gs201 would show that.
  3. The cut-off comments are all still there: "This drive", "once pe", "trigge", "no buffe", "powe" in gs-decon.c, "packet heade", "te_va", "neve", "fo" in gs-dsim-cmd.c, "btpowe" in the lynx DTS, and "panel drive" in the new driver. Taking those files back to the base and re-applying only your hunks is the quickest fix.
  4. The panel driver's header and leftovers. It still says "Google Pixel 7 Pro" and has no Google copyright line for the imported sequences. prepare() still sends the HC4's 0xc3 0x0c ("WQHD output"). The LHBM circle is centred at x 720, which is the middle of a 1440-wide panel. Please also add a panel-name check, as the 7 Pro's driver does with exynos_drm.panel_name=, since one DTB serves lynx units with S6E3FC3 panels too.
  5. Two later commits fix earlier ones. "use the panel's own reset and rails" fixes the DTS commit, and "use Google's S6E3FC5 sequences" replaces the first driver commit's sequences. Kernel PRs land commit by commit, so a bisect could stop on a panel that resets the touchscreen and sends the 7 Pro's commands. Could you fold each into the commit it fixes?
  6. Still open from 11-07:
    • the google,gs201-decon-lynx compatible (it goes away with point 2);
    • the new nodes disabled, with an overlay in omarchy-mobile's device/lynx/overlays/ to switch them on, and the GPU enable in its own commit;
    • prefixes as the files' history has them (drm: google: gs-decon:, drm: google: gs-dsim-cmd:, drm/panel: samsung-s6e3fc5:, arm64: dts: exynos: google: gs201-lynx:);
    • message bodies with the kernel release and the sources for the timings and the size.
  7. Kconfig: please add depends on GPIOLIB and select DRM_DISPLAY_HELPER / select DRM_DISPLAY_DSC_HELPER, as S6E3HC4 has. The DSC calls build today only because the 7 Pro's panel selects those.

Smaller:

  • 70 x 149 mm is a 6.5-inch diagonal, and the 7a is 6.1 inches (about 64 x 141 mm at 20:9). Could you check it against Google's lynx panel file or the touch dtsi's panel-height-mm?
  • The binding's supplies should take the hardware names (vci, vddi).
  • The description still starts with "Draft:" and mentions the panthor clamp.

For the order: #181 (the Pixel 7) merges first, and this one rebases onto it, since both edit gs-decon.c and gs-dsim-cmd.c. I know this is a long list again. The hard part, the panel's own reset and its own init, is done.

@cx3301J

cx3301J commented Nov 14, 2026

Copy link
Copy Markdown
Author

Started on the list; head 73ec754.

Done:

  • Removed the unverified lynx DSIM row. It was the Pixel 7's pair and made the probe report 90 at 60. The panel driver now accepts 60/90, so 90 is ready for real timers but not advertised yet.
  • gs_decon_connector_get_modes() now takes modes from the panel only when the soc data sets panel_modes, which only lynx does, and reads connector->probed_modes.
  • Fixed the cut-off comments in gs-decon.c, gs-dsim-cmd.c, the lynx DTS and the panel header.
  • Panel: Pixel 7a header + Google copyright, HC4 �xc3 gone, LHBM centred at 540, �ci/�ddi names, 64x141 mm, S6E3HC4 Kconfig deps; binding supplies/example and DTS reset/rails/size updated.
  • Reworded the two commits whose subjects lost their text.

Still open, and I am not going to claim otherwise:

  • The real 90 Hz DSIM timers and a 60?90?60 run. I removed the guess rather than leave a wrong rate.
  • Removing google,gs201-decon-lynx (the flag currently still keys off it), the disabled-by-default nodes + overlay, prefixes/message sources, and folding the later fixups into the commits they fix.
  • A panel-name check on the DSI RDID rather than the unexported saved_command_line.

Checks pass: W=1 builds, 10 DTs, 2 bindings, one cosmetic MAINTAINERS warning.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. This round fixed the two things that mattered most. Dropping the lynx DSIM row was the right call, since a guessed rate is worse than none. And the panel-modes flag keeps the Pixel 7 Pro as it is: I built gs-decon.o and gs-dsim-cmd.o before and after and compared them function by function. Only gs_decon_connector_get_modes changes, the 7 Pro's data falls through to the old loop there, and gs_dsim_panel_get_modes is new. Reading probed_modes, the Pixel 7a header with Google's copyright, the Kconfig dependencies and the hardware supply names in the DTS are all as asked. Thank you also for listing plainly what's still open.

CI hasn't run on this head yet (it's waiting for my approval), so I ran dt-validate here. Three things need fixing before the rest:

  1. The binding still has the old supplies. dt-validate on gs201-lynx.dtb now reports panel@0 (samsung,s6e3fc5): Unevaluated properties are not allowed ('vci-supply', 'vddi-supply' were unexpected). samsung,s6e3fc5.yaml is as commit 1 added it. It still lists vddio/avdd, and its example still uses gpp23 2 (the touchscreen's line), LDO25M/26M and 70 x 149 mm. Could you rename the supplies to vci-supply/vddi-supply, update the example to gpa7 1, the two rails and 64 x 141 mm, and put that in commit 1?
  2. The LHBM centre is now below the panel. The centre is two packed 12-bit values: the HC4's 0x2d 0x09 0x39 is x 720, y 2361. The new 0x21 0x0c 0x39 decodes to x 528, y 3129, and the panel is 2400 lines tall. The comment still says 0x2d0/0x939, and the area and radius bytes are still the HC4's. It's off by default, so nobody is hurt today, but please either take the 7a's values from Google's lynx panel driver (and cite it) or leave LHBM out of this first version.
  3. Three comment lines in shared files still change. The net diff still turns "display driver" into "display drive" (gs-decon.c:34 and gs-dsim-cmd.c:23) and "timer" into "time" (gs-dsim-cmd.c:20), and rewrites "te_var 1" (Google's variable name) as "TE variant 1". Those lines can stay as the base has them. Please also move gs_dsim_panel_get_modes() above gs_dsim_panel_sleep()'s kernel-doc comment; it currently sits between the comment and the function it describes.

On the panel-name check: the 7 Pro's driver doesn't use saved_command_line. cmdline_names_boe_panel() in panel-samsung-s6e3hc4.c reads /chosen/bootargs with of_property_read_string(), which works from a module, and you can do the same for the 7a's name.

Still open from my last comment, as you said:

  • fold the fix-ups into the commits they fix (there are five now: commits 6 to 10 each rewrite lines an earlier commit added);
  • the nodes disabled, with an overlay in omarchy-mobile's device/lynx/overlays/, and the GPU enable in its own commit;
  • drop google,gs201-decon-lynx in favour of a flag the board's panel or DSIM provides;
  • the files' own prefixes (drm/panel: samsung-s6e3fc5:, drm: google: gs-decon:, drm: google: gs-dsim-cmd:, arm64: dts: exynos: google: gs201-lynx:);
  • message bodies with the kernel release and the sources.

The reply mentions rewording two commit subjects, but commits 1 to 7 have the same messages as before, so that didn't reach the branch.

Smaller: s6e3fc5_reset()'s "shared pad" early return can't happen any more now that probe requires the reset line, and neither can the handover else branch. And the binding's maintainers: could be you, as in MAINTAINERS.

#181 (the Pixel 7) still merges first. Both edit gs-decon.c, so this one rebases onto it afterwards.

@cx3301J

cx3301J commented Nov 15, 2026

Copy link
Copy Markdown
Author

Addressed the three dt-validate points in 29a1123:

  • The binding now has �ci-supply/�ddi-supply, gpa7 1, the 7a rails and 64x141 mm, and the change is in commit 1.
  • LHBM uses Google's 7a location (�x21 0xc6 0xb1 0x63, x 540 y 1713 size 0x63) and the comment cites the stock sequence.
  • The base "display driver"/"timer"/"te_var" lines are restored, and gs_dsim_panel_get_modes() is above gs_dsim_panel_sleep()'s kernel-doc.

Also added the /chosen/bootargs panel-name check, removed the dead reset/handover branches, and kept the decon panel_modes flag. Checks pass: W=1, 10 DTs, 2 bindings; one MAINTAINERS warning.

Still open, as before: folding the fixups, disabled nodes + overlay, dropping google,gs201-decon-lynx, file-specific prefixes and message sources, and the real 90 Hz timers.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. This round fixed the two things dt-validate and the decode showed:

  • The binding: with vci-supply/vddi-supply and the 7a's example, dt-validate on gs201-lynx.dtb now reports only the message the base already has. Against the old schemas it rejects samsung,s6e3fc5, so your binding is what makes it pass.
  • The LHBM centre: 21 C6 B1 63 (x 540, y 1713) is exactly s6e3fc5_lhbm_location_cmds in Google's lynx driver, and thank you for citing it.

The /chosen/bootargs panel check, the removed dead branches and the helper's new position are all good too.

What's left:

  1. One shared comment line still changes. gs-dsim-cmd.c line 20 now reads "it has time values for"; the base says "timer values". The other two lines and te_var are back as they were.
  2. The structural points you listed as open still hold this up, since kernel PRs land commit by commit:
    • fold the fix-ups into the commits they fix (commits 6 to 11 now);
    • the display nodes disabled, with an overlay in omarchy-mobile's device/lynx/overlays/, and the GPU enable in its own commit;
    • drop google,gs201-decon-lynx in favour of a flag the panel or DSIM provides;
    • the files' own prefixes: commit 11's drm/panel: samsung-s6e3fc5: subject also covers gs-decon.c and gs-dsim-cmd.c;
    • message bodies with the kernel release and the sources.

Smaller, for when you're there:

  • Google's location sequence also unlocks with F1 5A 5A (and locks after), and sends it at enable. Worth copying before anyone turns allow_lhbm on.
  • Google's file says 70 x 149 mm, which is bigger than the 6.1-inch glass. I'd keep your 64 x 141 and say why in the message.
  • The lynx DTS framebuffer comment still says there's no panel driver.
  • The binding's maintainers: could be you, as in MAINTAINERS.
  • It would help to quote the exynos_drm.panel_name= your 7a's bootloader passes, since the new check refuses to bind without it.

CI is waiting for my approval on this head; I'll approve it. #181 still merges first, and this one rebases onto it.

@cx3301J

cx3301J commented Nov 15, 2026

Copy link
Copy Markdown
Author

Rebuilt #202 as a clean seven-commit series, head 58dd716:

dt-bindings: display: panel: samsung-s6e3fc5: add the panel drm/panel: samsung-s6e3fc5: add the panel driver dt-bindings: display: samsung: document the lynx DECON drm: google: gs-decon: add the lynx display path arm64: dts: exynos: google: gs201-lynx: enable the display arm64: dts: exynos: google: gs201-lynx: enable the GPU drm: google: gs-dsim-cmd: declare the panel get_modes helper

  • The five fix-ups are gone; each commit now carries its own final content.
  • File-specific prefixes throughout.
  • The DTS ships DSIM0/DECON0/display clocks/GPU disabled; the GPU is its own commit.
  • Overlay for the omarchy-mobile side: #829 (device/lynx: add the display and GPU overlay).

Checks pass: W=1, 10 DTs, 2 bindings, one MAINTAINERS warning.

Still open: drop google,gs201-decon-lynx in favour of a board/panel-provided flag, the real 90 Hz timers, and message bodies with kernel release/source notes.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 15, 2026

Copy link
Copy Markdown
Author

Dropped google,gs201-decon-lynx in �d819c846 (binding) and 12f0b16 (driver/DTS):

  • The DECON probes as plain google,gs201-decon.
  • In gs_decon_get_dsim(), after the DSIM is found, its remote panel node is checked; only a samsung,s6e3fc5 panel selects gs201_lynx_decon_data and panel_modes.
  • The binding no longer lists the lynx compatible, and the DTS no longer sets it.

Checks pass: W=1, 10 DTs, 1 binding, one MAINTAINERS warning.

Remaining: real 90 Hz DSIM timers and the kernel-release/source notes in message bodies.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J, for the rebuild and for opening omarchy-mobile#829. The display nodes are disabled now, the GPU is its own commit, and the prefixes are right. The gate passes: W=1 on the three C files, ten device trees with no new warnings, and only the MAINTAINERS note from checkpatch. But the last two commits broke the display path on the 7a, and there are a few things the gate can't see:

  1. The DECON can no longer pick up the 7a's modes. There are two reasons:

    • gs_decon_get_dsim() looks for the panel with of_graph_get_remote_node(np, 0, 0) on the DSIM node. dsi0's only port is its input from DECON0 (dsi0_in to decon0_out in gs201.dtsi), so that call returns DECON0 itself, never panel@0, which is a child node of dsi0 and not on the graph. I checked it on gs201-lynx.dtb with #829 applied.
    • Even with the right node, the swap comes too late. gs_decon_probe() runs gs_decon_check_boot_state() before gs_decon_get_dsim(). That check compares the bootloader's 1080x2400 with dc->soc->modes[0], which is still gs201's 1440x3120, so probe stops with "unexpected scanout size 1080x2400".

    Your cold-boot run on #438 was on 788026d954a1, which still matched google,gs201-decon-lynx at probe, so it didn't hit this. One way to do it: in probe, before gs_decon_check_boot_state(), find the DSIM through the port as now, then look among its children with of_get_compatible_child(dsim_np, "samsung,s6e3fc5") and set dc->soc there. gs201-panther: the display through gs-decon, with the panel asleep when the screen is off #181, which merges first, picks gs201's modes by the bootloader's scanout size, and the Pixel 7 is also 1080x2400, so after the rebase that panel check is what tells the two phones apart.

  2. The panel driver isn't built any more. The obj-$(CONFIG_DRM_PANEL_SAMSUNG_S6E3FC5) line in drivers/gpu/drm/panel/Makefile was in your previous head (29a11232) and is gone from this one. The gate compiles touched files one at a time, so W=1 still passed. The release kernel will also need CONFIG_DRM_PANEL_SAMSUNG_S6E3FC5=m in omarchy-mobile's kernel/configs/omamobile.config, next to the S6E3FC3 line, as a small PR of its own.

  3. Each commit has to build, since they land one by one. Commit 4 (42b674e20b0c) calls gs_dsim_panel_get_modes(), but the prototype only arrives in commit 7. Building gs-decon.o at commit 4 fails with error: call to undeclared function 'gs_dsim_panel_get_modes'. Commits 3 and 9 add and remove the same compatible, and commit 8 rewrites what commits 4 and 5 added. Please fold 7 and 8 into 4 and 5, and drop 3 and 9.

  4. gs-dsim-cmd.c line 20 still says "time values"; the base says "timer values". The file is the same as in your last head. 788026d954a1's gs-decon.c comments have cut-off words of the same kind ("This drive", "once pe"), so it's worth checking the other files that came from that tree, the panel driver's comments included.

  5. The GPU. Commit 6 sets &gpu { status = "okay"; } in the DTS, while its message and your reply say it stays disabled until the overlay applies. Enabling it in the DTS is fine by me, since the Pixel 7 Pro, the Pixel 7 and the Fold do the same. In that case please say so in the message and drop the GPU fragment from #829. Or keep it disabled and let #829 enable it. Either works, as long as the two agree.

  6. Still open from before: message bodies with the kernel release they were tested on and the sources (Google's file for the sequences and the LHBM location, the phone run), and the real 90 Hz timers. Until there's a lynx entry in gs_dsim_rates[], gs_decon_get_dsim() drops the DSIM ("unknown DSIM timers: 60 Hz only"), so the panel-modes branch doesn't run on the 7a yet and the 60 Hz table mode is used. That's fine for now, and panel_modes could come with the timers commit.

The smaller items from last time are unchanged: the LHBM F1 unlock, the note on 64 x 141 against Google's 70 x 149, the framebuffer comment, the binding's maintainers:, and quoting your bootloader's exynos_drm.panel_name=.

CI is waiting for my approval on this head; I'll approve it. #181 still merges first, and this one rebases onto it.

@cx3301J

cx3301J commented Nov 16, 2026

Copy link
Copy Markdown
Author

Fixed the binding/probe failures in 61e1559, now five commits:

  1. dt-bindings: ... s6e3fc5: add the panel
  2. drm/panel: ... add the panel driver (includes the missing Makefile entry)
  3. drm: google: gs-decon: add the lynx display path (gs-decon, gs-dsim-cmd and the header prototype together, so it builds alone)
  4. �rm64: dts: ... enable the display (nodes disabled)
  5. �rm64: dts: ... enable the GPU

The DECON now calls gs_decon_pick_lynx_soc() before gs_decon_check_boot_state(); it walks DECON's port to DSIM, then of_get_compatible_child(dsim, "samsung,s6e3fc5"). No custom compatible. The base already has the generic DECON binding, so there is no add/remove pair.

Overlay #829 now enables only display clocks/DSIM/DECON; the GPU is in the DTS and its message says so.

Checks pass: W=1, 10 DTs, 1 binding; one MAINTAINERS warning. Still open: 90 Hz timers, LHBM F1 unlock, 64x141 note, framebuffer comment, binding maintainers, panel_name quoting.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 16, 2026

Copy link
Copy Markdown
Author

Hardware check on 61e1559: the DECON binds through gs_decon_pick_lynx_soc() before check_boot_state(), the panel gate allows, brightness restores and the display stays up on a cold boot. The connector exposes only 60 Hz, so the panel-modes branch is not reached until the timers exist.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 16, 2026

Copy link
Copy Markdown
Author

Smalls are in 3831356:

  • LHBM now takes both test keys, F1 on/off around the location sequence.
  • 64x141 mm is noted as the active area, Google's 70x149 as the module.
  • Binding lists cx3301J as a maintainer.
  • Panel driver and binding split so checkpatch is clean apart from the MAINTAINERS note.

Still open: the 90 Hz timers and the kernel-release/source notes in commit bodies.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 16, 2026

Copy link
Copy Markdown
Author

Commit bodies now carry Sources: (Google's panel and DSIM sources) and Tested-on: 7.3.0-rc6-omamobile-push1-g61e15599a196 (Pixel 7a, cold boot), head �579fe1dc.

Only the 90 Hz timers remain open.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 17, 2026

Copy link
Copy Markdown
Author

90 Hz derivation update, from Google's cal_9845/dsim_reg.c:

  • CMD_TE_CTRL1 is built from fixed TE_PROTECT_ON_TIME 158 (15.8 ms) and TE_TIMEOUT_TIME 180 (18 ms), not from refresh: (hs_clk * time) / 16.
  • With hs_clk = 1102, that is exactly the bootloader's �x2a82 / �x306d. So the TE pair is correct for 90 too; only the underrun changes.
  • The 60 Hz bootloader value is �x1cf8 (7416). Scaling by the formula's 60?90 frame-time ratio gives roughly 4400 (�x1130) for 90.

That removes the TE blocker. The remaining step is a phone run of the candidate underrun before the row goes in; I will not guess it into the branch.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. The two things that mattered are fixed, and you showed them on the phone: gs_decon_pick_lynx_soc() runs before the boot-state check and finds the panel as a child of the DSIM, and the panel driver is back in the Makefile. I built all seven commits one by one (the two gs-decon objects and the panel through its Makefile with the symbol on), W=1 clean, and the gate passes on today's base. The cold boot on 61e15599a196 covers everything in this head except the LHBM change. gs-dsim-cmd.c line 20 is right again, the GPU commit and omarchy-mobile#829 now agree, and keeping the 90 Hz underrun out of the branch until it has run on the phone is the right call.

What's left is shape, and then it can go in:

  1. Please fold the two new commits. "unlock F1 for LHBM" only changes the driver that commit 2 adds, and "add the author" only changes the binding that commit 1 adds. They land one by one, so each belongs in the commit it fixes: five commits again.
  2. The sign-off has dropped out of the trailers on every commit. Each message now ends with Signed-off-by: and Assisted-by:, then a blank line, the "Sources:" paragraph, another blank line and Tested-on:. Git takes only the last block as trailers, so git interpret-trailers --parse shows Tested-on: and nothing else. Please move the sources and the test note up into the body, and end each message with Assisted-by: and Signed-off-by: and nothing after them.
  3. The test note on the LHBM change. The F1 keys came after the 61e15599a196 boot, so that commit shouldn't say it was tested on that kernel. In the folded panel commit, a sentence that LHBM with both keys hasn't run yet (it's off by default) is all it needs. The order you send matches Google's s6e3fc5_lhbm_location_cmds.
  4. A leftover in gs_decon_get_dsim(). The earlier lookup is still there (of_graph_get_remote_node(np, 0, 0) on the DSIM, then the samsung,s6e3fc5 test). It returns DECON0 and can never match, and it runs on every phone, so please take it out. And gs_decon_pick_lynx_soc() sits between gs_decon_check_boot_state() and the comment that describes it; move it above the comment.
  5. The source line. Google's panel driver is display/panel-samsung-s6e3fc5.c in kernel/devices/google/lynx (branch android-gs-lynx-6.1-android16), not in kernel/google-modules/display.

Small ones: the framebuffer comment in gs201-lynx.dts still says "There is no panel driver"; the two cmu_dpu/cmu_disp lines repeat the default; and omarchy-mobile#829's title still mentions the GPU.

Then the order: #181 merges first, and this rebases onto it. That rebase will change your gs-decon.c hunk more than a little: #181 picks the mode set by the size the bootloader scans out, and its Pixel 7 set has the same 1080x2400 timings you give the 7a. So after the rebase the 7a matches that set by size, and what your path still adds is the panel's own mode list and the pick by panel child, which has to win over the size match. Once that's done, one cold boot on the rebased head with the same lines as on omarchy-mobile#438 is what I need, and I'll merge.

Two things on my side: CI is waiting for my approval on this head again, which I'll give, and the release kernel's config needs CONFIG_DRM_PANEL_SAMSUNG_S6E3FC5=m in omarchy-mobile. A one-line PR for that next to the S6E3FC3 line is welcome.

Document the S6E3FC5 with its vci/vddi rails, reset line and 64x141 mm
size, and add it to MAINTAINERS.

Tested-on: 7.3.0-rc6-omamobile-push1-g61e15599a196 (Pixel 7a, cold boot).

Assisted-by: DeepSeek
Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
The S6E3FC5 is the Pixel 7a's 1080x2400 command-mode panel. Take the
running panel over from the bootloader, send Google's init/refresh/on/off
sequences, and expose brightness, local HBM and the 60/90 Hz setting.

Sources: Google's display/panel-samsung-s6e3fc5.c in
kernel/devices/google/lynx, branch android-gs-lynx-6.1-android16.

Tested-on: 7.3.0-rc6-omamobile-push1-g61e15599a196 (Pixel 7a, cold boot).
LHBM with both test keys has not run on hardware yet; it is off by default.

Assisted-by: DeepSeek
Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
The Pixel 7a's DECON0 drives a 1080x2400 command-mode panel through
DSIM0. Add the board data, take the mode list from the panel only for
lynx, and let gs-dsim-cmd carry the panel's DSI host and rate switch.

Sources: Google's exynos_drm_dsim.c and cal_9845/dsim_reg.c in
kernel/google-modules/display.

Tested-on: 7.3.0-rc6-omamobile-push1-g61e15599a196 (Pixel 7a, cold boot).

Assisted-by: DeepSeek
Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Describe the S6E3FC5 panel on DSIM0, its vci/vddi rails, DECON0's mode
and the display clocks. The nodes stay disabled until the board's
overlay enables them.

Tested-on: 7.3.0-rc6-omamobile-push1-g61e15599a196 (Pixel 7a, cold boot).

Assisted-by: DeepSeek
Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Turn on the Mali G710, as the Pixel 7 Pro, Pixel 7 and Fold do.

Tested-on: 7.3.0-rc6-omamobile-push1-g61e15599a196 (Pixel 7a, cold boot).

Assisted-by: DeepSeek
Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
@cx3301J

cx3301J commented Nov 18, 2026

Copy link
Copy Markdown
Author

Folded back to five commits, head f1171765f:

  1. binding (now includes the author)
  2. panel driver (now includes F1 LHBM and the active-area note; message says LHBM with both keys has not run)
  3. gs-decon/gs-dsim/header
  4. display DTS
  5. GPU DTS
  • Sources and Tested-on are in the body; every message ends Assisted-by: then Signed-off-by: and nothing after.
  • Removed the stale of_graph_get_remote_node(np, ...) panel lookup from gs_decon_get_dsim().
  • Moved gs_decon_pick_lynx_soc() above the check_boot_state comment.
  • Source line now names display/panel-samsung-s6e3fc5.c in kernel/devices/google/lynx, android-gs-lynx-6.1-android16.
  • Framebuffer comment no longer says there is no panel driver; the two default cmu lines are gone.
  • #829 retitled "device/lynx: add the display overlay".
  • Release config PR opened: #885 (CONFIG_DRM_PANEL_SAMSUNG_S6E3FC5=m).

Checks pass: W=1, 10 DTs, 1 binding; one MAINTAINERS note. #438 is merged, so the next step is #181 then this rebase and one cold boot.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 18, 2026

Copy link
Copy Markdown
Author

Prepared the rebase onto #181 in lynx-display-181 (not #202 yet, since #181 is unmerged):

The moment #181 lands I force-push this as #202 and run the cold boot on the rebased head.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. All five points are done: five commits again, the sign-off is back in the trailer block on every one (git interpret-trailers --parse shows Assisted-by: and Signed-off-by: and nothing else), the panel commit says LHBM with both keys hasn't run, the old lookup is out of gs_decon_get_dsim(), and the source line names the right repository and branch. The gate passes on today's base (W=1 for the three C files, ten device trees, no new warnings), and I'm approving the CI run for this head. #885 is fine and I'm merging it.

I also read your lynx-display-181 branch, so the rebase doesn't cost a round. The resolution is right: one mode set for the 7a, the pick by panel child still runs before the boot-state check, so it wins over #181's size match. It builds clean on #181's head here too. One thing changes there that the cold boot has to show:

  1. Until now the panel-modes branch has never run on the phone. gs_dsim_rates[] has no row for the 7a, so gs_decon_get_dsim() drops the DSIM ("unknown DSIM timers: 60 Hz only") and the 60 Hz table mode is used. gs201-panther: the display through gs-decon, with the panel asleep when the screen is off #181 adds the Pixel 7's row, 0x1cf8 / 0x2a82306d, and your note of the 17th gives the same pair for the 7a. If the 7a's bootloader really leaves those values, gs-dsim-cmd will say "link timers for 60 Hz", the DSIM stays, and the connector takes the panel's list for the first time: 60 Hz and 90 Hz.
  2. The 90 Hz mode can't be set: gs_decon_mode_hz() only accepts modes in dc->modes up to dc->max_hz, and that row has no 90 Hz timers, so atomic_check refuses it. Please keep it off the connector until the timers exist: in the panel_modes branch, skip probed modes above dc->max_hz, as the table loop below does.
  3. Were 0x1cf8 and 0x2a82306d read on your 7a, or taken from the Pixel 7's plan? If they're the 7a's own, a line in the lynx notes would be good, and a word in that row's comment that the 7a shares it.

For the boot report on the rebased head, on top of the lines you posted on omarchy-mobile#438: the gs-dsim-cmd line ("link timers for 60 Hz" or "unknown link timers …"), cat /sys/class/drm/card*-DSI-1/modes, and the mode the session is running.

Two small ones: commit 3's "Sources:" names kernel/google-modules/display without a branch (the two files are under samsung/ there, and that repository has no android-gs-lynx-6.1-android16 branch, so please name the one you read), and the PR description still describes the first version.

The order stays: #181, then your force-push of the rebase, the cold boot, and I merge.

@derekross derekross left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A short update, @cx3301J: the CI run for f1171765f is through and green, with nothing skipped. It is the first run of this head with dtschema, so dtbs_check on the ten device trees (no new warnings) and dt_binding_check on the panel binding are now checked, on top of the W=1 builds of all three C files, the panel driver included now that #885 is merged.

Nothing else has moved: #181 is still open, and I'll write here when it merges. The three points from my last comment stand for the rebase (the 90 Hz mode kept off the connector, whether 0x1cf8 / 0x2a82306d were read on your 7a, and the lines for the boot report). If you fold the first into lynx-display-181 now, the force-push after #181 can be the last one. Your next push will need my click for CI again; I'll watch for it.

@cx3301J

cx3301J commented Nov 19, 2026

Copy link
Copy Markdown
Author

All three points are handled in lynx-display-181 (dad238b):

  1. The panel_modes branch now drops probed modes above dc->max_hz with drm_mode_destroy(), so the connector exposes only 60 Hz until 90 Hz has timers and a phone run.
  2. �x1cf8 and �x2a82306d were read from this phone's DSIM0 registers at 60 Hz (via /dev/mem), not copied from the Pixel 7 plan. I added that as a row in device/lynx/NOTES.md on omarchy-mobile#829 ( d5878e) and noted that gs201-panther: the display through gs-decon, with the panel asleep when the screen is off #181's row is the same pair.
  3. Commit 3's source line now names kernel/google-modules/display, branch �ndroid14-gs-pixel-6.1, instead of a commit id.

PR description is rewritten to match the current five commits. Builds and checks pass on pr181. Ready for the force-push and cold boot as soon as #181 merges.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 19, 2026

Copy link
Copy Markdown
Author

Cold boot on the rebased head found one real blocker and fixed it.

gs-decon deferred forever with gs-decon: mif clock: the 7a tops out below 120 Hz and its ACPM firmware exposes no DVFS clocks, while the probe asked for mif/disp unconditionally for the 120 Hz floors. 2699ca0 skips both clocks on boards where the panel owns the mode list.

After that, on 7.3.0-rc6-omamobile-push1-gae96bf58b2f1:

  • gs-dsim-cmd: link timers for 60 Hz, panel attached.
  • gs-decon: lip-only KMS on DECON0, 1080x2400, up to 60 Hz (at 60 Hz).
  • card1-DSI-1 exists and its modes is 1080x2400; no 90 Hz is exposed, which is the intended state until its timers have run.

So the rebased head boots with display. #181 is still unmerged; when it lands I force-push this and repeat the boot report.

Assisted-by: DeepSeek

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants