Skip to content

device-lynx 0.2.3: persist the DSIM transmit enable - #438

Merged
derekross merged 8 commits into
derekross:mainfrom
cx3301J:lynx-display-package
Nov 18, 2026
Merged

derekross merged 8 commits into
derekross:mainfrom
cx3301J:lynx-display-package

Conversation

@cx3301J

@cx3301J cx3301J commented Nov 5, 2026

Copy link
Copy Markdown
Contributor

The panel driver can only send DCS writes after gs-dsim-cmd has allow_tx=1; without it, brightness and DSI commands return -EPERM. The display stack now takes over the connector on lynx, so the package should ship the option instead of relying on a manual /etc/modprobe.d file.

  • �endor/etc/modprobe.d/gs-dsim-cmd.conf: options gs-dsim-cmd allow_tx=1
  • install stages it at /usr/lib/modprobe.d/gs-dsim-cmd.conf
  • Version 0.2.3.

Verified on the phone: card2-DSI-1 enabled at 1080x2400, Hyprland running, brightness writable.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thanks for keeping the display work moving, and for putting the option in the package rather than leaving it as a hand-made file in /etc. Before it can go in:

  1. Lynx ships with more than one panel. Google's lynx tree has two panel drivers (panel-samsung-s6e3fc3-l10.c and panel-samsung-s6e3fc5.c in kernel/devices/google/lynx, display/), and its display DT lists the S6E3FC3 variants and the S6E3FC5, picked by the bootloader. linux-omarchy-mobile#202 drives the S6E3FC5 only, and allow_tx=1 for everyone would let its commands go to an S6E3FC3 too. The Pixel 8 Pro's package solves this by allowing DSI commands only when the bootloader names its panel on the command line (packages/omamobile-device-husky/bin/omamobile-husky-display-start). Could you do the same here? It would also help to know which panel name your unit's /proc/cmdline shows.
  2. It goes in after the kernel side. The published kernel has no display node for lynx, so this option does nothing until device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202 lands; I'll merge it with or after that.
  3. The file needs our SPDX header and a line saying why, like the other files in modprobe.d (# SPDX-License-Identifier: MIT). Please also put it in vendor/modprobe.d/ with an omamobile- name, as the other device packages do, and add a line to test/all that checks it's staged.
  4. A smoke report: please add tools/phone-smoke from the unit, with the kernel release it ran (CONTRIBUTING §6.5).

On versions: #436 also needs a bump, so whichever lands second takes the next number; I'll sort that out at merge.

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

Thanks, and agreed on the panel gating.

  • I will gate �llow_tx on the bootloader-selected panel name from /proc/cmdline, following the husky package, so S6E3FC3 units never get DSI commands.
  • I will move it to �endor/modprobe.d/ with an omamobile- name and the MIT SPDX header plus rationale, and add the est/all staging check and phone-smoke report.
  • I will hold it until device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202 lands, and report the panel name from this unit once Arch is back on the phone.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

Reworked as requested:

  • gs-dsim-cmd no longer gets a global �llow_tx=1. The option file is now �endor/modprobe.d/omamobile-display.conf with the MIT SPDX header and the reason.
  • �in/omamobile-lynx-display-start reads exynos_drm.panel_name= from /proc/cmdline and allows DSI commands only for sdc-s6e3fc5*; anything else leaves �llow_tx off.
  • A Before=display-manager.service unit runs it, and install/ est/all stage and check the script, unit, symlink and modprobe file.
  • Version 0.2.3, Signed-off-by included.

Tested on this unit (S6E3FC5): the service logs "panel sdc-s6e3fc5, DSI commands allowed", �llow_tx reads Y, and a brightness write takes. The gating script also reports "DSI commands stay off" for a fake sdc-s6e3fc3 cmdline.

Still to add: ools/phone-smoke. I will attach it once the package is installed from the staged tree rather than by hand.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author
ools/phone-smoke --here on the unit, with omamobile-device-lynx 0.2.3-1 installed and the gating service active:

| Check | Result | Detail | | boot (required) | skip | no slot on the command line | | storage guard (required) | ok | 55 partitions | | root (required) | ok | ext4 on userdata, 99G free | | kernel health (required) | ok | no oops or BUG | | USB gadget (required) | ok | bound to 11210000.usb | | display | ok | card2-DSI-1 connected, 1080x2400 | | GPU | ok | render node on panthor | | session (required) | not ok | Hyprland runs, the shell isn't running | | touch | ok | Goodix Berlin Capacitive TouchScreen | | Bluetooth | ok | hci0 |

Package: omamobile-device-lynx 0.2.3-1; �llow_tx=Y; brightness write to 1200/2047 took. The session-shell failure is this image's shell setup, not the DSIM option.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thanks for the rework. Gating on the bootloader's panel name is exactly what I hoped for, and it's good to know your unit has the S6E3FC5. The package test passes here, and the modprobe file now has its header.

A few things before it can go in:

  1. The unit can run before the DSI driver is loaded. gs_dsim_cmd autoloads from its device-tree node, and nothing orders your unit after that, so on some boots /sys/module/gs_dsim_cmd/parameters/allow_tx won't exist yet and the write fails (safely, but DSI commands then stay off for that boot). The Pixel 8 Pro's script runs modprobe gs_dsim_cmd 2>/dev/null || true first, for exactly this reason, and its unit has After=local-fs.target systemd-modules-load.service, RemainAfterExit=yes and a ConditionPathExists= on the display controller (packages/omamobile-device-husky/). Could you do the same? Please also give the .service file our SPDX header and mode 644.
  2. A test for the gate. The script already takes OMAMOBILE_LYNX_ROOT, so test/all could run it against a fake root for sdc-s6e3fc5.<id>, sdc-s6e3fc3-l10 and no panel name, and check what ends up in the fake allow_tx. The Pixel 8 Pro's test/display-start does this.
  3. It still goes in after linux-omarchy-mobile#202, and device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202 needs more work first. Its panel driver still sends the Pixel 7 Pro panel's init commands, including a 1440x3120 window, to the S6E3FC5, and its reset line is the touchscreen's. I've written that up on device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202. This switch is what lets those commands through, so I'd like the driver right before the package turns it on.
  4. Version: device-lynx: Bluetooth on the APM UART #98 also takes 0.2.3, and device-lynx 0.2.2: accept devinfo 3.11 for boot-success #436 is at 0.2.4. Since this one waits for the kernel side, it will land last and take the next free number then. I'll set it and the title at merge.

Smaller: systemd-backlight restores the saved brightness before your script allows commands, so that level never reaches the panel; the Pixel 8 Pro's script sends it again after the write. And when you next post tools/phone-smoke, please include its header line with the kernel release.

@cx3301J

cx3301J commented Nov 9, 2026

Copy link
Copy Markdown
Contributor Author

All the host-side points are in:

  • The script now runs modprobe gs_dsim_cmd 2>/dev/null || true before touching �llow_tx.
  • The unit has After=local-fs.target systemd-modules-load.service, Before=display-manager.service, RemainAfterExit=yes, a ConditionPathExists on the lynx DECON device, and the MIT SPDX header; the file is mode 644.
  • est/all now runs the script in a fake root with a fake modprobe for sdc-s6e3fc5.20800a, sdc-s6e3fc3-l10 and no panel name, checking the resulting �llow_tx.

Package tests pass. It still waits for #202 and will take the next free version at merge.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you for the quick rework. Loading gs_dsim_cmd first, the ordering and the fake-root test of the gate are exactly what I asked for, and the package test passes here, also merged with #436.

One small thing breaks it, though. The new condition in omamobile-lynx-display.service is cut short:

ConditionPathExists=/sys/bus/platform/devices/1c240000.display-controlle

The device is 1c240000.display-controller (decon0: display-controller@1c240000 in gs201.dtsi), so as it stands the unit would be skipped on every boot and allow_tx would never be set. That's safe, but it would silently do nothing. Could you fix the name and add a line to test/all that checks the unit's condition? The Pixel 8 Pro's test has one (grep -qx 'ConditionPathExists=…'). Your gate test runs the script, not the unit, so it couldn't have caught this.

Two smaller ones from last time are still open:

  • Brightness. systemd-backlight restores the saved level before your script allows commands, so that level never reaches the panel. The Pixel 8 Pro's script writes it again right after enabling (restore_brightness() in omamobile-husky-display-start).
  • README. Please add rows for the new script, unit and modprobe file, and drop the Documentation= line or point it at the repository, since the package doesn't install NOTES.md.

And a question for when linux-omarchy-mobile#202 is ready: on the Pixel 8 Pro, gs_decon's autoload is blacklisted, and the script enables commands before loading it, because gs-decon decides on 120 Hz at probe. On lynx, gs_decon autoloads, so it may probe before your unit runs. If #202's lynx path decides anything at probe, it would need the same order.

This still waits for #202. When that's in, a tools/phone-smoke report from the installed package, with its header line, will be the last piece. I'll set the version and title at merge. Thanks again!

@cx3301J

cx3301J commented Nov 10, 2026

Copy link
Copy Markdown
Contributor Author

Fixed: the unit now carries the full 1c240000.display-controller path, and est/all asserts that exact ConditionPathExists line so a truncated value fails the package test.

Head is �2d9cda; package tests pass.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thanks, that fixes it. The unit now names 1c240000.display-controller in full, the test checks that exact line on the staged file, and the package test passes here on today's main, also together with #436 (the two only meet in version). Your last push is waiting for my approval before CI runs, which is why no checks show; I'll start it.

Nothing else blocks the code. What's left:

  • It still waits for linux-omarchy-mobile#202. This switch is what lets the panel driver's commands through, and device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202 hasn't moved since my notes of the 7th (the init sequence and the reset line). When that's in, a tools/phone-smoke report from the installed package, with its header line, is the last piece.
  • Brightness. systemd-backlight restores the saved level before your script allows commands, so that level never reaches the panel. The Pixel 8 Pro's script sends it again right after enabling (restore_brightness() in omamobile-husky-display-start); the same few lines would do here.
  • README. Please add rows for the script, the unit and the modprobe file, and drop the unit's Documentation= line, since the package doesn't install NOTES.md.
  • The modprobe file now holds only comments. That's fine if device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202 ends up needing a blacklist gs_decon there, as on the Pixel 8 Pro (my question from last time about gs_decon probing before your unit); otherwise it can go.

I'll set the version and title at merge. Thanks for the steady work on this.

@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Contributor Author

Built this package from the branch on the phone (aarch64) and installed it over 0.1.0, then enabled the unit:

omamobile-device-lynx 0.2.3-1 omamobile-lynx-display: panel sdc-s6e3fc5, DSI commands allowed allow_tx Y

Smoke report from the installed package:

`

phone-smoke: unknown

Kernel 7.3.0-rc6-omamobile-push1-g788026d954a1-dirty, 2026-11-13T14:30Z
display ok card2-DSI-1 connected, 1080x2400
GPU ok render node on panthor
touch ok Goodix Berlin Capacitive TouchScreen
session not ok Hyprland runs, the shell isn't running
systemd not ok failed: omamobile-firstboot.service
`

The session line is the old rootfs image (omamobile-shell 0.2.16), not this package: the display service is enabled and reaches �llow_tx before the session starts. A full official-package smoke needs a rootfs built from current main; the local package here is the same source, built on-device.

Assisted-by: DeepSeek

@cx3301J
cx3301J force-pushed the lynx-display-package branch from a2d9cda to 325489e Compare November 13, 2026 14:35
@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto current main; the �ersion conflict is resolved by taking main's �.2.4. Package tests pass and the branch is mergeable again. Head 325489e.

The installed-package test above still stands: omamobile-device-lynx 0.2.3 was built from this source on the phone, the unit reaches the panel gate, and the smoke shows display/touch/GPU ok; the session failure is the old rootfs.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J, for the on-phone run and the smoke report. It's good to see the gate log panel sdc-s6e3fc5, DSI commands allowed from the installed package, and the display, GPU and touch rows all ok. I checked the new head: the code is the same as on the 10th, the package test passes on today's main, and it merges cleanly with #436, with the lynx test passing on the two together.

What's left:

  1. The version. The rebase took main's 0.2.4, so the branch now adds the script, the unit and the modprobe file without a new version, and pacman would never install them on a phone that already has 0.2.4. Please don't renumber it yourself: this PR lands after linux-omarchy-mobile#202, and I'll set the next free number then (0.2.5 today) and the title with it.
  2. It still waits for linux-omarchy-mobile#202. Thank you for today's rework there; I'll review it. When it's settled, I'd like one run of this package on device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202's head as it then stands: a cold boot with the unit enabled, then brightness and the 60/90 Hz switch, plus a smoke report. Please also say which device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202 commit the kernel was built from. The release you posted (…-g788026d954a1-dirty) is on no published commit, and it predates your 14:26Z push.
  3. gs-decon at probe. With device/tegu MODEM-PLAN, AGENTS: GNSS gets a fix on the 9a #202's 90 Hz mode, gs-decon picks the rate when it probes, and it autoloads on lynx, so it can probe before your unit sets allow_tx. If the panel then needs commands that weren't allowed yet, the modprobe file is the place for a blacklist gs_decon, with the script loading it after the write, as the Pixel 8 Pro does. If it isn't needed, the comments-only modprobe file can go.
  4. Still open from before: send the saved brightness again right after enabling (the Pixel 8 Pro's restore_brightness()), add README rows for the script, the unit and the modprobe file, and drop the unit's Documentation= line, since the package doesn't install NOTES.md.

One note for your phone. It now has a local "0.2.3-1" built from this branch, but main publishes a different 0.2.4, so the next omarchy update will replace your build and remove the display unit. Until this merges, either expect that or add IgnorePkg = omamobile-device-lynx to /etc/pacman.conf (and take it out again after the merge).

Thanks again for the steady work on this.

@cx3301J

cx3301J commented Nov 14, 2026

Copy link
Copy Markdown
Contributor Author

Addressed the host-side points in 518ad13:

  • gs_decon is now blacklisted in omamobile-display.conf; the start loads it only after �llow_tx is set.
  • The saved brightness is written again right after enabling the panel, as the Pixel 8 Pro's
    estore_brightness() does.
  • README rows for the script, unit and modprobe file; the unit's Documentation= line is gone.
  • est/all now also asserts the brightness write and that no brightness write happens for a non-S6E3FC5 panel.

The version is untouched: main's 0.2.4 stays, so the 0.2.5 number and title are yours at merge. It still waits on #202; after the rebase I will say which #202 commit the kernel was built from and run the cold-boot / brightness / 60?90?60 / smoke set.

My phone's local 0.2.3-1 is expected to be replaced by main's 0.2.4 unless I add IgnorePkg; I'll leave the package alone until this merges.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. All the host-side points are in. gs_decon is blacklisted and loaded after the gate, the brightness is sent again right after enabling, the README rows are there and the Documentation= line is gone. I checked the head: the package test passes on today's main, and the unit is inert on the published kernel, where lynx's DECON node is disabled.

Two things before the cold-boot run you mentioned:

  1. Make the display start hang-safe, as the Pixel 8 Pro's is. With the blacklist, this script is now the only thing that loads gs_decon on lynx, and the unit runs before the display manager. It's a oneshot with no TimeoutStartSec, and systemd gives oneshot units no start timeout by default, so a modprobe gs_decon that never comes back would hold the display manager for good, and nothing would stop the next boot from doing the same. omamobile-husky-display-start guards the same load with a marker written and synced before modprobe (the next boot then leaves gs_decon alone, and --retry clears it), an "already loaded this boot" exit, a check that the DECON device actually bound, and TimeoutStartSec=60 on its unit. Please bring those over. Husky's script and its test/display-start are a good template.
  2. Test the order. The fake modprobe in test/all ignores which module it's asked for. I tried deleting the blacklist line, and then moving the gs_decon load above the allow_tx write, and the test passed both times. If the fake records what it loads, the test can check that gs_decon comes after the write and that the blacklist is in place. While you're there, the install test's "nothing that starts hardware" line isn't quite true any more.

For the run itself, please include the brightness of … sent line from the journal. The panel's backlight comes up with the DSI host and the panel driver rather than with gs_decon, so that line shows whether the panel had bound by the time the restore ran.

It still waits for linux-omarchy-mobile#202, which you pushed again at 13:33 today, after your note here. And as you said, I'll set the version at merge. It will be 0.2.5 or 0.2.6, depending on whether #98 lands first. I'll also write the squash message from the code, since the description still describes the earlier modprobe-options version.

Thanks again for working through all of this so carefully.

@cx3301J

cx3301J commented Nov 15, 2026

Copy link
Copy Markdown
Contributor Author

Hang-safe start is in (3180989):

  • marker written and synced before modprobe gs_decon, stale marker stops the next boot, --retry clears it;
  • early exit when gs_decon is already loaded this boot;
  • DECON bind check, TimeoutStartSec=60;
  • the test's fake modprobe records what it loads, so deleting the blacklist or moving the load before the �llow_tx write now fails; marker and --retry are covered;
  • the install test's "nothing that starts hardware" line now says "and the display gate, and no radio".

The cold-boot report will include the �rightness of . sent journal line.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. All of it is in, and it's right. The start now carries the Pixel 8 Pro's guards line for line: the marker is written and synced before the load, a stale marker leaves gs_decon alone and --retry clears it, there's the once-per-boot exit and the bind check, and the unit has TimeoutStartSec=60. There's no exit trap, so a start that systemd times out keeps its marker for the next boot, which is what we want. The fake modprobe now catches a missing blacklist or a load before the allow_tx write. Plain modprobe gs_decon without husky's stage=2 fbdev=0 is fine too: those are the driver's defaults on the published kernel. I ran the package test on today's main with this merged, and it passes.

Two small things, neither blocking:

  • Nothing yet tests that the marker exists before modprobe gs_decon runs. I deleted the three lines that write it, and the test still passed, because the stale-marker case writes its own. Husky's test/display-start does this with a fake modprobe that fails when the marker isn't there.
  • The README row and the unit's Description= could mention the marker and --retry. That way someone whose display stays on simple-framebuffer after a hang knows how to try again.

It's still waiting on linux-omarchy-mobile#202. Once that's in and you've posted the cold-boot run (with the brightness of … sent line), I'll merge this. I'll set the version then: 0.2.5 if this goes in before #98, 0.2.6 if after.

@cx3301J

cx3301J commented Nov 15, 2026

Copy link
Copy Markdown
Contributor Author

Cold-boot run on the phone, package omamobile-device-lynx 0.2.4-1 installed from this branch, unit enabled, kernel built from 7.3.0-rc6-omamobile-push1-g788026d954a1-dirty (the #202 90 Hz experiment, not a published commit):

Starting Enable lynx S6E3FC5 DSI commands...
omamobile-lynx-display: panel sdc-s6e3fc5, DSI commands allowed
omamobile-lynx-display: brightness of 1c2c0000.dsi.0: level 1023 sent
omamobile-lynx-display: up: display
Finished Enable lynx S6E3FC5 DSI commands.

allow_tx Y, gs_decon loaded, card2-DSI-1 enabled. Brightness write 1200 read back 1200. Connector exposes 1080x2400 60 and 1080x2400 90.

Smoke: display ok, GPU ok, touch ok; session not ok (rootfs still has shell 0.2.16), boot-success failed (3.11 write inert).

The 60 to 90 switch is not demonstrated: Hyprland 0.56 rejects keyword monitor (can't work with non-legacy parsers) and hl.monitor() via eval returns ok without changing hyprctl monitors, at scale 1 and 2. The mode is advertised; only the switch remains unproven.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you for the cold-boot run, @cx3301J. Those are the lines I wanted: the panel named, DSI commands allowed, the brightness line, "up: display", and the brightness read back.

One thing to know: that kernel (788026d954a1, dirty) still matched the DECON by google,gs201-decon-lynx. #202's current head no longer binds the DECON on the 7a; I've written it up there. With that head this unit would log "the display controller did not bind to gs-decon" and leave the display on simple-framebuffer, which is the safe outcome, so nothing here needs to change. It does mean this still waits for #202. Once #202's fixed head is in, a short run on that kernel with the same lines is all I need, and then I'll merge.

Also, your phone now has a local omamobile-device-lynx 0.2.4 that isn't main's 0.2.4. If #98 merges first as 0.2.5, the next omarchy update would replace it and take this unit and the blacklist with it. Adding IgnorePkg = omamobile-device-lynx to pacman.conf until this merges avoids that.

The two small points from last time (a test that the marker exists when gs_decon loads, and the README row and the unit's Description=) are still open, and neither blocks. I'll set the version at merge: 0.2.5 if this lands before #98, 0.2.6 if after.

@cx3301J

cx3301J commented Nov 16, 2026

Copy link
Copy Markdown
Contributor Author

Re-ran on the fixed #202 head 61e15599a (7.3.0-rc6-omamobile-push1-g61e15599a196):

panel sdc-s6e3fc5, DSI commands allowed brightness of 1c2c0000.dsi.0: level 1023 sent up: display

�llow_tx Y, gs_decon loaded, card2-DSI-1 enabled; brightness 1200 read back. Smoke: display/GPU/touch ok, session still not ok (old rootfs shell), boot-success failed (3.11 inert). No custom compatible on this head. The connector now exposes only 60 Hz, as expected until the timers land.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 16, 2026

Copy link
Copy Markdown
Contributor Author

Unit Description= is now "Lynx S6E3FC5 display start" (�ecc22a). The README rows were already there; tell me if you meant a different row.

Assisted-by: DeepSeek

Install gs-dsim-cmd allow_tx=1 so brightness and DSI writes survive a reinstall; the panel needs it once the display stack owns the connector.
Move the stale /etc/modprobe.d option into vendor/modprobe.d with the
package SPDX header, and stop enabling gs-dsim-cmd globally. A boot-time
unit reads exynos_drm.panel_name= from /proc/cmdline and allows the
panel's DSI commands only for sdc-s6e3fc5, so an S6E3FC3 unit stays
read-only.

Tested on the unit: the service logs "panel sdc-s6e3fc5, DSI commands
allowed", allow_tx reads Y, and brightness writes.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
Load gs_dsim_cmd from the script before touching allow_tx, and order the
unit after systemd-modules-load with RemainAfterExit and a DECON
ConditionPathExists, as the Pixel 8 Pro package does. Give the unit its
SPDX header and mode 644, and cover the gate with a fake-root test for the
S6E3FC5 name, an S6E3FC3 name and no panel name.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
The ConditionPathExists value lost its last character, so systemd would
skip the gate on every boot. Restore the full 1c240000.display-controller
path and check it from the package test.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
Blacklist gs_decon so it cannot probe before allow_tx is set, restore the
saved brightness after enabling the panel's commands, then load gs_decon
from the start. Drop the unit's Documentation line, add the three display
files to the README, and test the brightness write as well as the gate.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
Write and sync a marker before loading gs_decon, stop the next boot when
the marker is still there, exit early when gs_decon is already loaded,
check that the DECON device binds, and add TimeoutStartSec=60. The test
now records what modprobe loads, checks that gs_decon comes after the
allow_tx write, and covers the marker and --retry.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
The unit starts the display, it does not only enable DSI commands.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
@cx3301J
cx3301J force-pushed the lynx-display-package branch from aecc22a to 03cba69 Compare November 17, 2026 13:00
@cx3301J

cx3301J commented Nov 17, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto main c462357 (install-on-running-slot changes) and re-ran the package test: all checks pass. Head �3cba69.

Assisted-by: DeepSeek

Set by the maintainer at merge: the branch carried main's 0.2.4.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. The re-run is what I asked for: the three lines, allow_tx Y, the DECON bound and the brightness read back, on a clean build of 61e15599a196. I fetched that commit and compared it with #202's current head: only the panel's LHBM unlock and one binding line differ, and the board file, gs-decon.c and gs-dsim-cmd.c are identical, so the run stands for this unit. The Description= is right now, and the package test passes on today's main.

On your question about the README: I meant the row for bin/omamobile-lynx-display-start. It says what the start does, and not what happens after a load that never returned. One more sentence would do: that it leaves a marker, the next boot stays on simple-framebuffer and says so in the journal, and omamobile-lynx-display-start --retry clears it. That and the test that the marker exists when gs_decon loads are still optional.

Nothing else is needed from you here. Two things are mine:

  • The version. Your branch has 0.2.4, which is main's number. I'll set it when I merge: 0.2.5 if this lands before device-lynx: Bluetooth on the APM UART #98, 0.2.6 if after.
  • The order with device-lynx: Bluetooth on the APM UART #98. Both PRs change the same line of packages/omamobile-device-lynx/test/all (the "install stages…" pass line), so whichever lands second needs a small rebase: keep both sets of checks and one pass line that names both. I'll say when.

Until this is merged, keep IgnorePkg = omamobile-device-lynx on your phone, so an update doesn't replace your local build.

@derekross
derekross merged commit 51995c7 into derekross:main Nov 18, 2026
2 checks passed
@derekross

Copy link
Copy Markdown
Owner

Merged, and a correction to my last comment and to the commit's subject. I set the version myself with one commit on your branch: the package is 0.2.5 (packages/omamobile-device-lynx/version on main says so, and that is what the next development publish builds). The squash subject on main reads "device-lynx 0.2.6", which is my mistake; main's history isn't rewritten, so the subject stays wrong and the file is what counts.

I merged it ahead of #202 because it is inert on the published kernel (the DECON node is disabled, so the unit is skipped) and the panel gate is then in place before any overlay can turn the display path on. #98 takes 0.2.6 and resolves the shared pass line in its rebase.

Once 0.2.5 is in the development repository you can drop IgnorePkg = omamobile-device-lynx on your phone.

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