Skip to content

omamobile-device-husky 0.1.3: the display at 120 Hz, with brightness - #228

Merged
derekross merged 1 commit into
derekross:mainfrom
greenbrettmichael:husky-display-start
Nov 1, 2026
Merged

derekross merged 1 commit into
derekross:mainfrom
greenbrettmichael:husky-display-start

Conversation

@greenbrettmichael

@greenbrettmichael greenbrettmichael commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

omamobile-device-husky 0.1.3 brings up the Pixel 8 Pro's display at 120 Hz, with brightness, at every boot. derekross/linux-omarchy-mobile#160 runs the screen on gs-decon, with DSIM0 and an HK3 panel driver. As on the Pixel 9a, gs-decon decides on 120 Hz when it probes, so gs_dsim_cmd.allow_tx has to be on before gs_decon loads.

What it adds:

  • omamobile-husky-display-start, hang-safe like omamobile-tegu-graphics-start:

    • it writes and syncs a marker, then loads gs_decon stage=2 fbdev=0 before the display manager. A load that never returns leaves the marker, so the next boot stays on simple-framebuffer until --retry;
    • before that it allows the panel's DSI commands, only when the bootloader names the google-hk3 panel on the command line (exynos_drm.panel_name=google-hk3.20800a on the first phone);
    • a load that returns but fails is logged, and the display stays on simple-framebuffer.
  • omamobile-husky-display.service, enabled by a vendor multi-user.target.wants link, plus display-manager.service.d/10-omamobile-husky.conf, so the session starts after it.

  • modprobe.d/omamobile-husky-display.conf: blacklist gs_decon, so only the start loads it. There's no install line.

  • The saved brightness: systemd-backlight restores it in sysinit, before the start allows the panel's commands. docs/review: lessons from review round 29 #160's HK3 driver keeps that level instead of failing, and the start sends it again once allow_tx is on.

    Ordering systemd-backlight after the start would be a cycle, because sysinit comes before the start. I tried it, and systemd broke the cycle. The other way to fix it, moving the start before sysinit, would let a hung load block the whole boot.

  • options gs_decon screen_off=2, as cheetah and tegu ship it: on DPMS off, DECON stops its frames and the HK3 goes to sleep, and DPMS on wakes it. Without it, the panel stays awake on black frames.

  • kernel/configs/omamobile.config: CONFIG_DRM_PANEL_GOOGLE_HK3=m, a module that binds only to its compatible. On a kernel without docs/review: lessons from review round 29 #160, olddefconfig drops the unknown symbol, and the unit is skipped by its DECON0 condition.

Tests:

  • test/display-start runs the start in a scratch root, with modprobe, sync and sleep faked:
    • the marker is down before the first load;
    • allow_tx is 1 when gs_decon loads for google-hk3 and google-hk3.<id>, and stays 0 for google-tg4c, google-hk3x or no name;
    • the brightness is sent again only after allow_tx;
    • a failed load is logged and clears the marker;
    • a stale marker keeps gs_decon off until --retry;
    • a second start and a DTB without DECON0 do nothing.
  • test/all checks the staged files, the enable link, both orderings, and that gs_decon is the only blacklisted module.

tools/pr-check and packages/omamobile-device-husky/test/all pass.

Tested on the first husky (the kernel at #160's tag husky-test-20261031-d3, these files installed under /etc and /usr/local), over two reboots:

  • the start allowed DSI commands for google-hk3.20800a before gs_decon loaded at about 4.5 s, and SDDM started after it;
  • Hyprland came up on gs-decon and switched to 120 Hz;
  • a brightness of 700 set before the reboot came back, and the systemd-backlight unit stays active, so it saves the level at shutdown;
  • omamobile-device-husky 0.1.2: let the GPU clock scale to 890 MHz #221's zuma_gpu_max_khz=890000 took effect too.

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thank you, this is a careful piece of work: the marker, the panel-gated allow_tx and the test that checks the order with a fake modprobe are just what a boot-time display load should look like, and the note on why systemd-backlight can't simply be ordered after the start saved me the same experiment. A few things before it lands:

  1. Could you run one boot from the built 0.1.3 package, with the copies under /etc and /usr/local removed first (a unit or modprobe.d file in /etc overrides the package's), and post tools/phone-smoke with uname -r, the start's journal lines and systemctl status omamobile-husky-display?
  2. It depends on kernel docs/review: lessons from review round 29 #160 (and tools/power-ab: a systemd suspend mode, and a gauge without charge_counter #157 for the GPU clock). On a kernel without docs/review: lessons from review round 29 #160 the unit fails at every boot, so I'd like to merge after those, or have the README say what a published kernel shows.
  3. CONFIG_DRM_PANEL_GOOGLE_HK3=m doesn't hurt any other phone, so it can go in kernel/configs/omamobile.config; the device fragment is only for options that would break another device during bring-up.
  4. Do all Pixel 8 Pros use the HK3 panel? If another panel ships, please say what gs_decon does there with commands off.

#221 and #214 are merged now, so this needs a rebase on version and device/husky/kernel.config.

The kernel display PR runs husky's screen on gs-decon, with DSIM0 and the
HK3 panel driver for brightness and 120 Hz. Those need
gs_dsim_cmd.allow_tx set before gs_decon probes, since gs-decon decides on
120 Hz at probe, as on the Pixel 9a.

- omamobile-husky-display-start loads gs_decon (stage 2) before the
  display manager, hang-safe like omamobile-tegu-graphics-start: a
  synced marker that a load which never returns leaves behind keeps it
  off on the next boot, until --retry. It allows the panel's DSI
  commands first, only when the bootloader names the google-hk3 panel
  on the command line (exynos_drm.panel_name=google-hk3.<id>);
- omamobile-husky-display.conf blacklists gs_decon's autoload, so only
  the start loads it, and sets screen_off=2, as cheetah and tegu do: on
  DPMS off DECON stops its frames and the HK3 sleeps, and DPMS on wakes
  it;
- display-manager.service.d orders the session after it;
- the saved brightness: systemd-backlight restores it in sysinit, before
  the start allows the panel's commands, and the HK3 driver keeps that
  level. The start sends it again once allow_tx is on. Ordering
  systemd-backlight after the start instead would be a cycle (sysinit is
  before the start) or would move the start before sysinit, where a hung
  load blocks the whole boot;
- kernel/configs/omamobile.config builds the HK3 driver: a module that
  binds only to its compatible, so it is safe on every phone. On a
  kernel without the display PR the DTB has no DECON0, and the unit is
  skipped by its condition.

test/display-start runs the start in a scratch root: the marker before
the load, allow_tx on for google-hk3 and google-hk3.<id> only and before
gs_decon, the brightness sent again only then, a failed load, a stale
marker and --retry.

Tested on the first husky with these files installed under /etc and
/usr/local: on two reboots the start allowed DSI commands for
google-hk3.20800a before gs_decon, the session came up at 120 Hz on
gs-decon, and a brightness of 700 set before the reboot came back.
test/all checks the staged files, the enable link, the orderings and
the unit's DECON0 condition.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
@greenbrettmichael

Copy link
Copy Markdown
Contributor Author

Thanks Derek. Rebased onto main after #221 and #214 (6d57e14).

  1. The package boot: not done yet. It needs the built 0.1.3 package installed and my /etc and /usr/local copies removed, and I'll post the report when I have it. Meanwhile, from today's e1 kernel boot with the copies (same files, husky-test-20261101-e1), the start's journal:
    omamobile-husky-display: panel google-hk3.20800a: DSI commands allowed (gs_dsim_cmd.allow_tx=1)
    omamobile-husky-display: brightness of 19440000.dsi.0: level 700 sent
    omamobile-husky-display: up: display
    
    systemctl status: active (exited), status=0/SUCCESS.
  2. On a kernel without docs/review: lessons from review round 29 #160 the unit doesn't fail. Its ConditionPathExists=/sys/bus/platform/devices/19470000.display-controller is false when the DTB has no DECON0, so systemd skips it, and the screen stays on simple-framebuffer. The README row says so now, and test/all checks the condition. I'm happy for this to merge after docs/review: lessons from review round 29 #160 and tools/power-ab: a systemd suspend mode, and a gauge without charge_counter #157 anyway.
  3. CONFIG_DRM_PANEL_GOOGLE_HK3=m is in kernel/configs/omamobile.config now, next to TG4C, and the device fragment is back to main's.
  4. The panel: stock husky has one panel node, google,hk3 (Google's zuma-husky-display.dtsi). Google's two HK3 drivers are the same panel, and its revisions differ only in the id the bootloader appends (google-hk3.20800a here). If another name ever shows up, the start leaves allow_tx off, and gs-decon still loads. It runs flip-only at 60 Hz, as the d1 kernel did before the panel driver existed: no commands are sent, so no brightness and no 120 Hz.

New in this push: options gs_decon screen_off=2, as cheetah and tegu ship it. Your sleep question on kernel #160 showed that with the default, DPMS off leaves the HK3 awake on black frames. With it, four off/on cycles stopped DECON, slept the panel and woke it at 120 Hz, with no errors (details on #160). test/all checks the option.

tools/pr-check and packages/omamobile-device-husky/test/all pass.

@greenbrettmichael greenbrettmichael changed the title omamobile-device-husky 0.1.3: the display at 120 Hz, with brightness (stacked on #221) omamobile-device-husky 0.1.3: the display at 120 Hz, with brightness Nov 1, 2026
@greenbrettmichael

Copy link
Copy Markdown
Contributor Author

The boot from the package, as you asked:

  • omamobile-device-husky 0.1.3-1, built from this branch (6d57e14) with tools/phone-install-package;
  • my copies removed first: the unit, its drop-in and enable link under /etc/systemd/system, both files in /etc/modprobe.d, and /usr/local/bin/omamobile-husky-display-start. After the removal, /etc/modprobe.d holds only hid_apple.conf, and /usr/local/bin is empty;
  • kernel tag husky-test-20261101-e1, re-flashed to slot B, then a normal boot.

uname -r: 7.3.0-rc6-omamobile-husky-wifi (the test kernels pin the release name).

The start's journal (monotonic time):

[4.122510] systemd[1]: Starting Bring up the Pixel 8 Pro display driver (hang-safe)...
[4.346863] omamobile-husky-display-start[735]: omamobile-husky-display: panel google-hk3.20800a: DSI commands allowed (gs_dsim_cmd.allow_tx=1)
[4.361469] omamobile-husky-display-start[735]: omamobile-husky-display: brightness of 19440000.dsi.0: level 700 sent
[4.527514] omamobile-husky-display-start[735]: omamobile-husky-display: up: display
[4.564618] systemd[1]: Finished Bring up the Pixel 8 Pro display driver (hang-safe).
[4.736686] systemd[1]: Started Simple Desktop Display Manager.

systemctl status omamobile-husky-display:

● omamobile-husky-display.service - Bring up the Pixel 8 Pro display driver (hang-safe)
     Loaded: loaded (/usr/lib/systemd/system/omamobile-husky-display.service; disabled; preset: disabled)
     Active: active (exited)
    Process: 735 ExecStart=/usr/bin/omamobile-husky-display-start (code=exited, status=0/SUCCESS)

"disabled" is only is-enabled looking at /etc. multi-user.target wants the unit through the package's /usr/lib/systemd/system/multi-user.target.wants link, which is why it ran.

What the boot shows:

  • allow_tx Y, screen_off 2, zuma_gpu_max_khz 890000;
  • gs-decon at 120 Hz;
  • the saved brightness of 700 came back, and systemd-backlight@ stays active;
  • one DPMS off/on cycle: the panel slept with no frames, then woke at 120 Hz, with no flip timeouts and none late.
tools/phone-smoke
Check Result Detail
boot (required) ok kernel 7.3.0-rc6-omamobile-husky-wifi, slot b, up 56s
storage guard (required) ok 62 partitions; only userdata and slot B's boot images writable
root (required) ok ext4 on userdata, 97G free
kernel health (required) ok no oops or BUG; 0 WARN, 33 error lines in dmesg
systemd units ok none failed
USB gadget (required) ok bound to 11210000.usb
display ok card2-DSI-1 connected, 1344x2992
GPU ok render node on panthor
session (required) ok Hyprland and the Omarchy shell running
lock ok PAM file present, the shell can lock (passwordPam true, locked true)
touch ok SALIXP Touchscreen
buttons ok gpio-keys
Wi-Fi ok wlp1s0 on brcmfmac, link up
Bluetooth skip no controller
battery skip no battery power supply
sound cards skip none
sensors skip no sensor proxy
cameras skip no libcamera tools
thermal ok hottest zone 51 °C

The 33 error lines are the BCM4398 bring-up's (#156: unexpected monitor packets, firmware channels 169-177, PKTID, the P2P interface, one join retried before it succeeded), plus the ones listed on kernel #157: panthor's missing regulator, the HK3's one refused write before allow_tx, and exynos-clkout from the zuma PMU on omamobile.

@derekross

Copy link
Copy Markdown
Owner

Thanks, Brett. Every point is answered, and I checked each one against the new head (6d57e14):

  1. The package boot. It's from the built 0.1.3, with your /etc and /usr/local copies gone. The journal shows the start allowing DSI commands at 4.35 s, gs-decon up by 4.53 s, and SDDM after it. You're right about the "disabled" in systemctl status: the package's multi-user.target.wants link is what pulls the unit in. Thanks also for checking the e1 tag. It's on your fork and contains docs/review: lessons from review round 29 #160's head.
  2. Without docs/review: lessons from review round 29 #160, ConditionPathExists=…/19470000.display-controller skips the unit (that's DECON0's address, and no zuma DTB on omamobile has the node yet). The README row says so, and test/all checks it. That was the "or" in my request, so this doesn't have to wait for docs/review: lessons from review round 29 #160 and tools/power-ab: a systemd suspend mode, and a gauge without charge_counter #157.
  3. CONFIG_DRM_PANEL_GOOGLE_HK3=m is in the shared fragment, and husky's kernel.config matches main again.
  4. The panel. One google,hk3 node on stock, and an unknown name leaves allow_tx off with gs-decon still flip-only at 60 Hz. That's the answer I wanted.

screen_off=2 is a good addition, and the four DPMS cycles on #160 cover it. Policy and tests pass, and the branch merges cleanly with main.

One thing for the merge, and it isn't yours to fix: Lef's USB-rescue PR (#250) also takes omamobile-device-husky 0.1.3, and the two conflict in install and test/all. I'll merge this one first, and #250 moves to 0.1.4 on a rebase.

Merging. Thanks for a really well-tested piece of boot plumbing.

@derekross
derekross merged commit eb567ad into derekross:main Nov 1, 2026
2 checks passed
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…erekross#228)

Reviewed in round 32: see the pull request for the tests.
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