Skip to content

gs201-panther: the display through gs-decon, with the panel asleep when the screen is off - #181

Open
nesmeck wants to merge 5 commits into
derekross:omamobilefrom
nesmeck:omamobile-gs201-panther-decon
Open

nesmeck wants to merge 5 commits into
derekross:omamobilefrom
nesmeck:omamobile-gs201-panther-decon

Conversation

@nesmeck

@nesmeck nesmeck commented Nov 1, 2026 •

Copy link
Copy Markdown
Contributor

The Pixel 7's display through gs-decon, as on the Pixel 7 Pro: flip-only KMS on ABL's pipeline, vblanks from DECON's frame start, and the panel put to sleep with the screen off. It builds on #177 (the panel driver and the DSI command path). Five commits:

  1. drm/panel: samsung-s6e3fc3: support sleep and a cold restart from KMS. This is the handed-over panel's state you asked for on gs201-panther: the S6E3FC3 panel and the DSI command path, for brightness #177, as in today's S6E3HC4 driver:
    • a panel taken over from the bootloader is marked prepared and enabled, so drm_panel_disable()/unprepare() reach it;
    • while it is asleep, backlight changes are only recorded;
    • the panel's lock covers every command sequence.
  2. drm: google: gs-dsim-cmd: recognise the Pixel 7's link. The bootloader's 60 Hz pair (UNDERRUN_CTRL 0x1cf8, CMD_TE_CTRL1 0x2a82306d, read on the phone), with no second rate, so gs-decon gets the link and the panel but never switches the rate. 90 Hz comes later.
  3. drm: google: gs-decon: pick the modes by the bootloader's scanout size.
    • Each SoC gets a list of mode sets, one per panel, chosen by the size ABL scans out.
    • gs201 gains the S6E3FC3 P10 at 1080x2400 and 60 Hz, with the timings of Google's panel driver.
    • The Pixel 7 Pro's and the Pixel 9a's modes don't change.
  4. gs201-panther: enable DECON0 for gs-decon. DECON0 gets a display_pool (five 1080x2400 XRGB8888 buffers, below 4 GiB, reusable) and the boot framebuffer, as on the Pixel 7 Pro.
  5. drm: google: gs-decon: scan out padded rows on gs201 too. This one was found on the phone.
    • The problem: Hyprland's buffers for 1080-pixel rows have a 4352-byte pitch (padded to 64 bytes), while the boot framebuffer has 4320. gs201's plane check took only the boot pitch, so every modeset failed with -EINVAL and the screen froze on the boot console.
    • The fix: DPP0's source width is the row length in pixels on gs201, as on G4, except that gs201 packs the source height into the same register. So gs201 now takes padded rows too, and writes the height back with the width.
    • ⚠️ This touches the Pixel 7 Pro's path:
      • padded_pitch is now set for all of gs201, so each flip writes RDMA_SRC_SIZE.
      • With 1440-pixel rows (5760 bytes, no padding), that is the same value ABL leaves.
      • I haven't run it on a Pixel 7 Pro. Could you check it there, per CONTRIBUTING §6.2?

Tested on a Pixel 7 (slot B), staged as in device/panther/DISPLAY-PLAN.md D3, one change per boot. Kernel 7.3.0-rc6-omamobile-push1-g53fab72d1489 was omamobile caf5f8a with #133, #171, #170, #177 and this branch. After the rebase onto today's omamobile (8698185), git range-diff shows all five commits unchanged, and every panther file they touch is the same as on that kernel.

  • Stage 0 (bring-up option gs_decon stage=0):
    • gs-decon binds on the 1080x2400 set and logs ABL's state: command mode, TE trigger on, 1080x2400, DPP0 at the boot framebuffer;
    • gs-dsim-cmd logs "link timers for 60 Hz".
  • Stage 1: 601 frame starts and 601 frame dones in 10.0 s (60.02 Hz), with no storm.
  • Stage 2:
    • without commit 5, Hyprland's modesets all failed as described;
    • with it, simpledrm hands over, Hyprland runs at 1080x2400@60 and the desktop is smooth;
    • 2624 flips in ten minutes, with no flip timeouts and no late flips (latency 8.9 ms on average, 16.8 ms at most).
  • Screen off with Hyprland's DPMS:
    • screen_off=0 sends black frames;
    • screen_off=1 stops the frames, with the trigger masked;
    • screen_off=2 also puts the panel to sleep, and DPMS on wakes it through the panel's cold init with the picture correct. That was the cold init's first run.
    • With screen_off=2, the power key's lock turns the screen off, a tap wakes it, and the PIN works.
    • No DSIM errors.
  • After a reboot, with omamobile-device-panther 0.1.5 setting allow_tx=1 and screen_off=2 and no bring-up file, everything above comes up on its own. That's omarchy-mobile#281.
phone-smoke from that boot

phone-smoke: panther

Kernel 7.3.0-rc6-omamobile-push1-g53fab72d1489, tools at 05b472c, 2026-11-01T18:20Z.

Check Result Detail
boot (required) ok kernel 7.3.0-rc6-omamobile-push1-g53fab72d1489, slot b, up 2191s
storage guard (required) ok 55 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, 9 error lines in dmesg
systemd units ok none failed
USB gadget (required) ok bound to 11210000.usb
display ok card2-DSI-1 connected, 1080x2400
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 false)
touch ok FocalTech FT3658
buttons ok gpio-keys
Wi-Fi skip no wireless interface
Bluetooth skip no controller
battery ok 91% Discharging, 31 °C
sound cards skip none
sensors skip no sensor proxy
cameras skip no libcamera tools
thermal ok hottest zone 45 °C

Packages: hyprland 0.56.2-5.1, libcamera 0.7.2-4.21, mesa 1:26.2.4-1.4, omacam 0.3.3-1, omamobile-apps 0.2.0-1, omamobile-device-panther 0.1.5-1, omamobile-firstboot 0.1.2-1, omamobile-keyring 20261026-1, omamobile-mirror 0.1.1-1, omamobile-omarchy 0.1.12-1, omamobile-owl-widgets 0.1.1-1, omamobile-shell 0.2.28-1, omarchy 4.0.6-1, quickshell 0.3.2.r7.g2b4501a-1

Checks: tools/kernel-pr-check --build OUT --merge origin/omamobile passes everything:

  • a clean merge onto 8698185;
  • the sign-offs;
  • checkpatch --strict with no warnings;
  • W=1 builds of the 3 touched C files;
  • the 8 device trees with W=1 and dtbs_check, with no new warnings.

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck. This is a careful piece of work: staged one change per boot, with the runbook written down before each step, and the padded-row bug found and explained from DRM's own debugfs. The handed-over panel's state is the first commit, as you said it would be on #177, and it matches what the S6E3HC4 driver does today: prepared/enabled set at handover, backlight changes only recorded while the panel is asleep, and the lock around every sequence.

Our gate passes on today's omamobile (b39c04d, which now has the BOE panel series): a clean merge, the sign-offs on all five commits, checkpatch --strict with no warnings, W=1 builds of the three C files, and all eight device trees with no new warnings. CI is green, including dtbs_check. I also checked a few things by hand:

  • I ran dt-validate on gs201-panther.dtb, before and after. The only message is the PMU clock one that the base already has, and the new DECON0 node is checked against google,gs201-decon.yaml.
  • The Pixel 7 Pro's, Fold's and 7a's DTBs are byte-identical before and after.
  • Commit 3 builds on its own. Kernel PRs land commit by commit, so that matters.
  • The new DSIM entry only matches on your exact pair (0x1cf8 with 0x2a82306d). The 9a shares the TE value but has a different underrun, so there's no overlap.
  • Your D3a log line SRC_SIZE 0x09600438 confirms the gs201 layout the fix relies on: the height in the upper half, the width in the lower.

One change, please, and then one thing on my side.

1. Please put the padded-row commit before the DECON0 enable. As the series stands, a bisect that stops on "gs201-panther: enable DECON0 for gs-decon" gets a Pixel 7 whose Hyprland modesets all fail and whose screen freezes on the console. That's exactly your D3c first try. The padded-row change only touches gs-decon.c and doesn't depend on the DT, so the order 1, 2, 3, 5, 4 (enable last) works. If you reorder, please also update the list of changes in D3 of the plan in omarchy-mobile#281, or I'll do it when I merge that one.

2. The Pixel 7 Pro run is on me. You're right that the last commit changes the reference phone. With padded_pitch set for gs201, every flip now writes RDMA_SRC_SIZE, and the plane check accepts padded pitches there too. With 1440-pixel rows the value is the one ABL leaves, so I don't expect anything to change. But per CONTRIBUTING §6.2 it gets a run on my phone before it merges, and the mode-set rework goes through the same boot. My integration branch has an older gs-decon.c than omamobile, so I'll test a kernel built from omamobile with your branch merged. On that boot I'll check:

  • the probe picks the 1440x3120 set and the link timers;
  • Hyprland's framebuffer pitch is 5760;
  • the 60/120 Hz switch and the frame rate at each;
  • a DPMS off/on with screen_off=2;
  • the picture after each of those;
  • a smoke report.

I'll post the results here.

One small thing for the last commit's message, not a blocker. Google's gs201 DPP code programs the source width from fb->width (dpp_convert_plane_state_to_config() in exynos_drm_dpp.c), so it never uses a padded stride, and the fact that DPP0 reads the source width as the stride on gs201 comes from your phone. It's convincing: an 8-pixel mismatch per row would have sheared the whole desktop. Please say so in the message ("measured on the Pixel 7: …"), so the next reader doesn't go looking for it in Google's driver. While you're there, the first paragraph says the failure left "the screen dark", but the Tested paragraph and the plan say it froze on the boot console.

Thanks again. It's great to see the Pixel 7's screen go dark properly when it's off.

@nesmeck

nesmeck commented Nov 2, 2026 •

Copy link
Copy Markdown
Contributor Author

Thanks for the thorough check. Both points are done, and the new head is 0a132eb92, on today's omamobile (91d3f15):

  1. The order is now: the panel's state, the DSIM link, the mode sets, the padded rows, and DECON0's enable last. A bisect can't land on the frozen screen any more.

    • git range-diff shows all five commits' code unchanged.
    • The padded-row commit touches only gs-decon.c, which no later commit changes, so it builds where it now sits. Commit 3's gs-decon.c is the one the D3a kernel ran.
  2. The padded-row message now says:

    • DPP0 reading its source width as the row length is measured on the Pixel 7, since Google's driver programs fb->width and never scans out a padded stride, and an 8-pixel mismatch per row would have sheared every line;
    • the screen froze on the boot console, not "dark".

    It's re-signed, and it also notes that the next commit is the one that enables DECON0.

The base now includes gs-dsim-cmd: Support DCS reads from the BOE series. The branch rebased onto it without conflicts, and our DSIM commit only adds the Pixel 7's table entry.

kernel-pr-check passes on the new head:

  • a clean merge;
  • the five sign-offs;
  • checkpatch --strict with no warnings;
  • W=1 for the three C files;
  • the 8 device trees with no new warnings.

omarchy-mobile#281's list of changes follows the new order.

Thanks for taking the Pixel 7 Pro run.

🤖 Generated with Claude Code

@nesmeck
nesmeck force-pushed the omamobile-gs201-panther-decon branch from da2e805 to 0a132eb Compare November 2, 2026 01:06
@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck, both points are done exactly as asked. I checked the new head:

  • git range-diff against the previous head shows all five commits' code unchanged. Only the order and the padded-row commit's message changed.
  • The padded-row commit is now fourth. Its tree differs from the head only in gs201-panther.dts, so it builds where it sits, and a bisect can no longer stop on the frozen console.
  • Its message now says that DPP0 reading the source width as the row length was measured on the Pixel 7, why Google's driver doesn't show it, and that the screen froze on the boot console.
  • Our gate passes on today's omamobile (2219cc6, the republished base): a clean merge, the sign-offs on all five commits, checkpatch --strict with no warnings, W=1 builds of the three C files, and all eight device trees with no new warnings. CI is green, including dtbs_check.
  • dt-validate on gs201-panther.dtb has no messages, before or after. The Pixel 7 Pro's, Fold's and 7a's DTBs are byte-identical.

Nothing more is needed from you. The Pixel 7 Pro run I described last time is next on my side: a kernel built from omamobile with your branch merged, with the 1440x3120 probe, the 5760 pitch, 60/120 Hz, DPMS with screen_off=2 and a smoke report. I'll post the results here and merge after that, then omarchy-mobile#281.

@angelrenefm

Copy link
Copy Markdown
Contributor

D3 on a second Pixel 7 (stock factory unit, CP3A.261005.002.A1 on both slots, devinfo 3.15). Kernel
7.3.0-rc6-omamobile-push1-g0a132eb92793 (this PR's head), flashed with derekross/omarchy-mobile#396's AVB footers.

Step Result
D3a, stage 0 pass: "link timers for 60 Hz"; DECON0/DPP0 as ABL left them (1080x2400, RDMA 0xfac00000, TRIG_CON 0x3061); panel taken over from the bootloader; simpledrm kept card0; allow_tx Y; no errors
D3b, stage 1 pass: 601 frame starts and 601 frame dones in 10.01 s (60.1 Hz), no storm
D3c, stage 2 pass: gs-decon on card2, simpledrm gone, Hyprland at 1080x2400@60. 10 min: 871 flips, 0 timeouts, 0 late, latency 9.0 ms avg / 16.7 ms max. "No stripes or tearing, much faster than before"
D3d, screen_off written at runtime, DPMS through hl.dsp.dpms 0: frames keep counting while dark (181 in 3 s), desktop back. 1: frame_start frozen for 3 s, trigger masked, desktop back. 2: panel_asleep 1 and the panel's TE count frozen while dark; after DPMS on, TE back at 60 Hz (181 per 3 s), desktop back. Repeated 5 more times: 6/6, no DSIM timeouts, no "cold init failed", nothing in dmesg (cleared before each step)
Boot with stage=2 screen_off=2 from the bring-up file gs-decon at 60.12 Hz, Hyprland up, 0 flip timeouts; the user reported it booted fine

Not checked: the picture right after the runtime cold inits (measured from SSH only: counters and dmesg).

I'm working on this with AI help (Claude).

@derekross

Copy link
Copy Markdown
Owner

Thank you, @angelrenefm. A full D3 run on a second Pixel 7 with this PR's exact head is just what this series needed. Stage 0 found ABL's state as on nesmeck's unit, the frame counters show 60 Hz with no storm, 871 flips had no timeouts, and the screen_off=2 cold init worked six times out of six. Saying what you didn't check (the picture right after the runtime cold inits) helps too.

@nesmeck, nothing more is needed from you here. The head hasn't moved since my last look, and on today's omamobile (fe9627c) it still merges cleanly and passes our gate: the sign-offs, checkpatch with no warnings, W=1 for the three C files, the device trees with no new warnings, and dt-validate on gs201-panther.dtb with no messages. The Pixel 7 Pro's DTB is byte-identical. When you next update omarchy-mobile#281, could you add angelrenefm's run to DISPLAY-PLAN D3 as the second unit's results, with his kernel release?

What's left is mine: the Pixel 7 Pro run I described, a kernel from omamobile with this branch merged, checking the 1440x3120 probe, the 5760 pitch, 60/120 Hz and DPMS with screen_off=2, plus a smoke report. I'll post it here and merge after that.

@nesmeck

nesmeck commented Nov 11, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto today's omamobile (fc7086a): #160 had made this conflict. Three of the five commits had conflicts:

  • gs-decon: pick the modes by the bootloader's scanout size. Pixel 8 Pro display: gs-decon, DSIM0 and the HK3 panel at 120 Hz #160 added the Pixel 8 Pro's HK3 modes and gs101's own match data, both with the old single .modes list. They now get one mode set each (gs101 the 1440x3120 set, zuma HK3's), so the Pixel 6 Pro, 7 Pro, 8 Pro and 9a behave as before: one set, matched by the bootloader's size. The commit message says so.
  • gs201-panther: enable DECON0. Only the header comment, which now lists Bluetooth and the DSI command path too.
  • gs-dsim-cmd: recognise the Pixel 7's link. The Pixel 7's 60 Hz pair now sits after HK3's in the rate table; the entry itself is unchanged.

tools/kernel-pr-check --build OUT --merge origin/omamobile passes everything: a clean merge onto fc7086a, the sign-offs, checkpatch --strict with no warnings, the policy checks, W=1 builds of the three touched C files, and 10 device trees with W=1 and dtbs_check.

Booted on my Pixel 7 from slot B (7.3.0-rc6-omamobile-push1-ga170455e0402, this head on omamobile alone): gs-decon recognised the link ("link timers for 60 Hz"), bound on the 1080x2400 set ("flip-only KMS on DECON0, 1080x2400, up to 60 Hz"), and Hyprland ran on card2-DSI-1 at 60.0 Hz with no flip timeouts or late flips. With screen_off=2, screen off put the panel to sleep and masked the trigger (frame starts stopped), and Power brought the screen back correctly, at 60 Hz again (301 frame starts in 5 s).

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck, for rebasing over #160 and for booting the result. I read the rebase with git range-diff against the head I last checked (0a132eb92): the panel commit is unchanged, the DSIM and padded-row commits changed only in context, and the mode-set commit gives gs101 and zuma one set each, picked by the same condition as before, so the Pixel 6 Pro, 7 Pro, 8 Pro and 9a keep their behaviour. The order is still right, with DECON0's enable last. Our gate passes on today's omamobile (fc7086a68): a clean merge, the sign-offs, checkpatch --strict with nothing to read, W=1 for the three C files, and the ten device trees with no new warnings. CI is green, including dtbs_check.

One fix, please, and one question:

1. The last commit's message has a diff in it. git log -1 --format=%B f763cfe7a is 95 lines: your message, Assisted-by:, then diff --git a/arch/arm64/boot/dts/exynos/google/gs201-panther.dts … with five hunks, and only then Signed-off-by:. It looks like a conflict resolution pasted into the editor, and it is the old version of the diff. Kernel commits land as they are, so that text would stay in the history, and git am would cut the message at the diff. Could you reword the commit so it ends with the two trailers together and nothing after the test paragraph? Our gate didn't catch this, which is on us, and I'll fix it.

2. Which tree booted? Your comment names ga170455e0402, and the head is f763cfe7a. I assume the difference is that message edit. Could you confirm with git rev-parse a170455e0402^{tree}? The head's tree is f0f1d3e17209.

Two notes for later, nothing to do now:

The Pixel 7 Pro run is still mine, and I'm sorry it has taken this long. It's the same list as before, now on omamobile with this head: the 1440x3120 probe and link timers, the 5760 pitch, 60/120 Hz, DPMS with screen_off=2, and a smoke report. I'll post it here and merge after that and your reword.

@nesmeck
nesmeck force-pushed the omamobile-gs201-panther-decon branch from f763cfe to db288b7 Compare November 13, 2026 05:16
@nesmeck

nesmeck commented Nov 13, 2026

Copy link
Copy Markdown
Contributor Author

Thank you for catching that. The diff was left over from the conflict resolution in my rebase. I've reworded the last commit, and it now ends with Assisted-by: and Signed-off-by: together, right after the test paragraph. Only the message changed: the new head db288b74f has the same tree as before.

The tree that booted is this one. git rev-parse a170455e0402^{tree} gives f0f1d3e1720975483d3c6bbd4d6dd721a8c30070, the same as f763cfe7a and the new head. The difference between them was the sign-off.

tools/kernel-pr-check --build OUT --merge origin/omamobile passes again on today's omamobile (2db2bc35b).

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck. Both points are done. The last commit's message now ends with the two trailers, right after the test paragraph, and none of the five messages has diff text left in it. The tree you booted, f0f1d3e17209, is the tree of the new head, so your 11-11 boot is this code.

I re-ran our gate on today's omamobile (2db2bc3): a clean merge, the five sign-offs, checkpatch --strict with nothing to read, W=1 for the three C files, and the ten device trees with no new warnings. dt-validate on gs201-panther.dtb prints nothing, before or after. CI is green on this head. None of your four files changed on omamobile since you rebased, so nothing you tested has moved under you.

Nothing more is needed from you. What's left is my Pixel 7 Pro run, on today's omamobile with this head merged: the 1440x3120 probe and link timers, the 5760 pitch, 60 and 120 Hz with the frame rate at each, DPMS off and on with screen_off=2, and a smoke report. I'm sorry it has taken this long. I'll post the results here and merge after that.

One note for afterwards: #202 (the Pixel 7a) now also reads 0x1cf8 / 0x2a82306d from its bootloader, the same link timers as your Pixel 7, as well as the same 1080x2400 size. I've asked there for a way to tell the two boards apart that doesn't depend on what the bootloader leaves. This PR merges first and #202 rebases onto it. #195 rebases after it too.

Mark a panel taken over from the bootloader as prepared and enabled, so
drm_panel_disable() and drm_panel_unprepare() reach it, and track whether
it is out of reset and sleep. While it is not, backlight updates only
record the new state (drm_panel_enable() applies it through
backlight_enable()). prepare() leaves the display off, which the reset
does. Take the panel's lock in prepare/enable/disable/unprepare and in
backlight updates, so their command sequences don't interleave.

This follows the same change to the S6E3HC4 driver, for the Pixel 7
Pro, and lets gs-decon put the Pixel 7's panel to sleep with the screen
off.

Tested on a Pixel 7 from slot B (7.3.0-rc6), with gs-decon driving the
screen: with gs_decon.screen_off=2, DPMS off put the panel to sleep and
DPMS on woke it through the cold init (reset, the vendor init at 60 Hz,
TE2 and DSC) with the picture correct; the power key's lock turned the
screen off and a tap brought it back. No DSIM errors.

Assisted-by: Claude Opus 5.5
Signed-off-by: Alejandro Garcia <skipper.pelygro@gmail.com>
Add the timers the Pixel 7's bootloader leaves on DSIM0 for its S6E3FC3
P10 panel at 60 Hz (UNDERRUN_CTRL 0x1cf8, CMD_TE_CTRL1 0x2a82306d, read
on the phone), so gs-decon gets the link and the panel and can put the
panel to sleep with the screen off. The set has no second rate: the
panel's 90 Hz has no timers yet, so the link is never switched.

Tested on a Pixel 7 from slot B (7.3.0-rc6): the driver logs "link
timers for 60 Hz", and gs-decon takes the link and puts the panel to
sleep and wakes it with the screen off.

Assisted-by: Claude Opus 5.5
Signed-off-by: Alejandro Garcia <skipper.pelygro@gmail.com>
gs201 offered only the Pixel 7 Pro's 1440x3120 modes and refused to run
when the bootloader scanned out any other size. Give each SoC a list of
mode sets, one per panel, and pick the one whose first mode matches the
size the bootloader left, which the probe already reads. Add the Pixel
7's S6E3FC3 P10 at 1080x2400 and 60 Hz, with the timings of Google's
panel driver; its 90 Hz mode comes with the link timers for it.

Tested on a Pixel 7 from slot B (7.3.0-rc6), with the padded-row change
on top: gs-decon binds on the 1080x2400 set, counts frame starts at
60.02 Hz, and Hyprland runs on it at 1080x2400@60 with no flip
timeouts. The Pixel 6 Pro's, 7 Pro's, 8 Pro's and 9a's sets are
unchanged: one set each.

Assisted-by: Claude Opus 5.5
Signed-off-by: Alejandro Garcia <skipper.pelygro@gmail.com>
gs-decon on gs201 took only framebuffers with the boot framebuffer's
pitch. On the Pixel 7, whose DECON0 the next commit enables,
Hyprland's buffers for the 1080x2400 panel have a 4352-byte pitch (1080
pixels padded to 64 bytes) where the boot framebuffer has 4320, so
every modeset failed the plane check with -EINVAL and the screen froze
on the boot console. The Pixel 7 Pro's 1440-pixel rows need no padding,
which is why it never showed there.

On gs201, DPP0 reads its source width as the row length in pixels,
padding included, as G4 does; gs201 packs the source height into the
upper half of the same register. This is measured on the Pixel 7, not
taken from Google's driver, which programs the source width from
fb->width and never scans out a padded stride: with 4352-byte rows and
the stride written as the source width, the desktop shows unsheared,
where an 8-pixel mismatch per row would shear every line. Set
padded_pitch for gs201, write the height back with the width, and cap
the width at the 16-bit field. With an unpadded buffer the register
gets the same value the bootloader left.

Tested on a Pixel 7 from slot B (7.3.0-rc6), with the next commit on
top: before this change every Hyprland modeset failed and the screen
froze on the boot console; with it the modesets pass, the desktop shows
at 1080x2400@60 and stays smooth, with 2624 flips in ten minutes, no
flip timeouts and no late flips (latency 8.9 ms on average, 16.8 ms at
most).

Assisted-by: Claude Opus 5.5
Signed-off-by: Alejandro Garcia <skipper.pelygro@gmail.com>
Enable DECON0 with a pool for gs-decon's scanout buffers and the boot
framebuffer, as on the Pixel 7 Pro: gs-decon keeps the bootloader's
DECON, DSIM and DSC setup, flips DPP0 between buffers, and takes the
screen over from simpledrm. The pool has room for five 1080x2400
XRGB8888 buffers, below 4 GiB since DPP0 takes 32-bit addresses with
the DPU SysMMU bypassed, and is reusable.

Tested on a Pixel 7 from slot B (7.3.0-rc6): gs-decon allocates its
scanout buffers from the pool, takes the screen over from simpledrm, and
Hyprland flips on it at 60 Hz.

Assisted-by: Claude Opus 5.5
Signed-off-by: Alejandro Garcia <skipper.pelygro@gmail.com>
@nesmeck
nesmeck force-pushed the omamobile-gs201-panther-decon branch from db288b7 to 3c21907 Compare November 16, 2026 21:04
@nesmeck

nesmeck commented Nov 16, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto today's omamobile (26867fc), since 317d87d's MFC pool landed next to this branch's display pool in gs201-panther.dts and the merge stopped being clean. The four driver commits are untouched; in the DECON0 commit the two reserved-memory pools now sit one after the other, mfc_pool first as on omamobile, and nothing else changed. tools/kernel-pr-check --build --merge origin/omamobile passes: a clean merge, the five sign-offs, checkpatch --strict clean, W=1 on the three C files, and the ten device trees with dtbs_check and no new warnings. Nothing to do on your side beyond the 7 Pro run.

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck, for rebasing again. I read it against the head I checked on the 13th: the four driver commits are unchanged (the three files are the same blobs), and in the DECON0 commit the only difference is that display_pool now sits after mfc_pool. The changed lines between your old and new gs201-panther.dts are exactly the lines omamobile itself gained there, so your boot of the 11th still describes this code. Our gate passes on today's omamobile (26867fc4e): a clean merge, the five sign-offs, checkpatch --strict with nothing to read, W=1 for the three C files and the ten device trees with no new warnings. CI is green with dtbs_check.

Nothing is needed from you. What's left is still my Pixel 7 Pro run, on today's omamobile with this head merged: the 1440x3120 probe and link timers, the 5760 pitch, 60 and 120 Hz with the frame rate at each, DPMS off and on with screen_off=2, and a smoke report. You have had to rebase more than once to keep this mergeable while it waited on me, and I'm sorry for that. I'll merge this ahead of the other open Pixel 7 device-tree PRs so you don't have to do it again, and post the results here first.

For the others following: #263 (local HBM) goes in right after this, #268 (GNSS) doesn't conflict with it, and #195 (Wi-Fi) and #202 (the Pixel 7a) rebase onto it.

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.

3 participants