Repository navigation
arm64: dts: exynos: google: zuma-husky: give the framebuffer a panel size - #154
Conversation
|
Thanks Lef, nice find, and the write-up made it easy to follow. Good news on the numbers: Google's panel driver for the 8 Pro ( |
2f015a2 to
dbb5ddb
Compare
|
Thanks Derek, all three done and force-pushed. The panel node is at the root now, not under /chosen, and the splash comment is back directly above the framebuffer node. I checked the number myself too: HK3_WIDTH_MM 70 and HK3_HEIGHT_MM 155 in panel-gs-hk3.c, so the comment and the commit message cite that instead of "approximate". dtbs_check on zuma-husky.dtb gives no error for the node (it only shows the two eusb2-phy supply warnings, which I haven't compared against the base). Still not booted as a built-in node, only as the overlay before. Next time I flash the phone I'll boot it that way and say what the keyboard does. |
|
Thanks Lef, all three points are in, and the commit message is honest about what was and wasn't booted. I validated the DTB against the schemas and the panel node adds nothing; simpledrm and fw_devlink behave exactly as you describe. I've approved the CI run and will merge once it's green. When you next flash, a one-line note here that the built-in node gives the same 70 x 155 mm and keyboard height would be nice to have, but it doesn't hold this up. |
…size The framebuffer node has no panel, so simpledrm estimates the display's physical size from its pixels at 96 dpi. On the 1344x2992 panel that is 355x791 mm instead of about 70x155 mm. Hyprland reports it, and squeekboard sizes the on-screen keyboard with it: 105 of 935 logical pixels (11% of the screen) instead of 196 (21%) with the real size. Add a panel node at the root with width-mm and height-mm and point the framebuffer at it. The node is disabled and not under /chosen, whose schema only allows framebuffer and pinctrl children. The simple-framebuffer binding has simpledrm read the size from the node it points to, but with an enabled node without a driver fw_devlink makes the framebuffer wait for ever (no DRM device appears and the screen stays on the boot logo), and booting with fw_devlink=permissive instead takes away the ordering that acpm-protocol needs after exynos-mailbox (16 channels, 0 usable, so no PMIC and no touch). of_link_to_phandle() creates no link to a supplier that has no device and is not available, and simpledrm reads the two properties without looking at the status. 70x155 mm is Google's own value for this panel: HK3_WIDTH_MM and HK3_HEIGHT_MM in panel-gs-hk3.c (kernel/devices/google/shusky). Tested on a Pixel 8 Pro (Kioxia UFS 4.0) on the husky-test-20261031-c6 kernel with the same node and property added by a device tree overlay: simpledrm initialises, acpm-protocol reports 16 channels, 16 usable, salixp binds, Hyprland reports 70x155 mm and the keyboard layer is 196 px tall. The commit as it stands (the node at the root) builds and adds no dtbs_check error for the node; it has not been booted as a built-in node. Signed-off-by: Lef <me@lefteros.com> Assisted-by: LLM
dbb5ddb to
0b83d97
Compare
|
Pushed again, no code change: the commit message now has the Assisted-by tag the kernel doc asks for, so the hash changed (0b83d97). On what Derek asked for last time (a note after a boot): I've been running the same 70 x 155 mm panel node as a DT overlay on the 8 Pro all day, with the normal boot arguments. Hyprland reports 70x155 mm instead of 355x791, the ACPM channels come up as 16 usable and the touchscreen still passes (salixp: sense echo yes events 2). It is the overlay and not the built-in node, so I still can't say I booted it built in, I'll say so here when I do. |
|
Thanks Lef. I compared the new head with the one I approved last night: the tree is identical, and only the commit message changed (the Assisted-by line, in the form the kernel's coding-assistants document gives). That earlier head passed CI, and the gate merges this one cleanly onto today's Thanks for the note on the overlay boot as well. The built-in node and the overlay give simpledrm the same two properties, so I'm happy with that as the evidence. A line here after your next flash would still be nice, but it doesn't hold this up. |
This is the keyboard fix from #1 (derekross/omarchy-mobile#1 (comment)), Brett asked for it as a PR.
The framebuffer node in zuma-husky.dts has no panel, so simpledrm estimates the display's size from its pixels at 96 dpi: 355x791 mm instead of about 70x155. The on-screen keyboard is sized from that and comes out 105 px tall on a 935 px screen (11%) instead of 196 (21%) with the real size.
It adds a disabled panel node with width-mm and height-mm and points the framebuffer at it. Disabled on purpose: an enabled node with no driver makes fw_devlink wait for ever, so the framebuffer never binds and the screen stays on the boot logo; booting with fw_devlink=permissive instead takes away the ordering acpm-protocol needs after exynos-mailbox; and of_link_to_phandle() makes no link to a supplier that isn't available, while simpledrm reads the two properties without looking at the status.
70x155 mm is approximate, from the panel's published 489 ppi, not measured. I looked in Google's display module tree for the real value and didn't find a 1344x2992 panel there, so it would need another source.
Phone: Pixel 8 Pro (Kioxia UFS 4.0).
What I ran: tools/kernel-pr-check with --build --merge in the project container: it merges cleanly onto 1c5ba1a, signed off, checkpatch --strict clean, 8 device trees build with W=1 and no new warnings. I also test-merged it with #148 and #153 and it merges cleanly with both. dtbs_check didn't run locally (no dtschema).
On the phone I tested the same node and property added by an overlay on top of Brett's husky-test-20261031-c6 build: simpledrm initialises, acpm-protocol says 16 channels, 16 usable, salixp binds, Hyprland reports 70x155 mm and the keyboard layer is 196 px tall.
What I didn't test: this commit's exact DTB. The node sits under /chosen here and was at the root in the overlay, I only checked that it builds and decompiles to the right values. I also haven't looked at other simple-framebuffer phones.