Repository navigation
HID: apple-mtp-kiwi: the K114c touch screen (needs #1568) - #1605
Draft
joshuaswarren wants to merge 5 commits into
Draft
joshuaswarren wants to merge 5 commits into
joshuaswarren wants to merge 5 commits into
Conversation
joshuaswarren
force-pushed
the
w7c/up-k114c-kiwi
branch
from
November 11, 2026 23:48
13776e2 to
2974ebe
Compare
RyanTheTide
requested changes
Nov 12, 2026
RyanTheTide
left a comment
RyanTheTide
left a comment
Contributor
There was a problem hiding this comment.
Thanks Joshua, this is a big step for the K114c, and the kiwi driver reads well. Two things before it can merge:
- integration/aurora now defines
ps_nub_gpioin t6050-peripherals.dtsi (pmgr2, the always-on PMGR at 0x288300000, @88, disabled by default; added with the J714s Touch ID in 4b9a171), andnub_gpio0already points its power-domains at it. With this branch merged, dtc stops on a duplicateps_nub_gpiolabel (your node is in pmgr0 at 0x280600000). Please drop the t6050-die0.dtsi node and enable the shared one from t6050-k114c.dts the way J714s does:&ps_nub_gpio { status = "okay"; };and&nub_gpio0 { status = "okay"; ... }. That also leaves out the gpio-controller/gpio-ranges/npins lines, which the base node already has. - #1603/#1604 are merged, so t6050-k114c.dts conflicts. Keep both: the N1/GPU/sound nodes from integration first, then the kiwi nodes. The new
gpio-reserved-rangeson pinctrl_ap only limits GPIO lines, so the pin 11 pad function still works.
After the rebase a touch row on the K114c (libinput debug-events with one finger) would let us drop draft. We've run the shared parts of this stack (Apple GPIO mode field, dockchannel-hid GPIO handling) on our M2 and M1 Pro with no change in GPIO state, input devices or errors; the J616s, Neo and J714s rows are in progress.
On the K114c the NUB GPIO block comes out of iBoot with bit 4 of the touch-screen reset and power pins set (0x474210, 0x474212 as read through /dev/mem). With that bit set, the DATA bit does not latch in output mode: writing 0x474213 reads back 0x474212, writing 0x474203 reads back 0x474203. direction_output wrote only bits 1-3 of the mode, so the pin stayed in a state where the driver's output value never took effect. Treat bits 1-4 as the mode, so that direction_output and direction_input clear bit 4. Modes 0-7 are unchanged, and a working output (the PERST# pin of the K114c, register 0x00076a03) has the bit clear. Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
The touch screen of the 14 inch M5 Max MacBook Pro (K114c) sits behind the kiwi interface of the dockchannel MTP. The interface has a 29-byte HID descriptor with one opaque input item and no description of the contacts. The chip reports a raw 82 column by 54 row capacitance image every 8.3 ms and does no contact detection itself. Claim the interface by name (hid-apple and hid-magicmouse already refuse BUS_HOST Apple interfaces that are not a keyboard or mouse), keep the raw pipe on hidraw, and register a direct multi-touch input device with 10 slots over the 3024x1986 panel. The driver: - clamps the Report Size of the descriptor, which is 0x80000 bits and above the 16384 that hid-core accepts, in report_fixup, - skips the whole-image shift frames the chip sends every 2 s and removes a per-row mean from the others, - keeps a fixed-point baseline per cell (fast up when a finger left, slow drift otherwise), - takes 3x3 local maxima 250 counts below baseline, centroids them and tracks them frame to frame with a 2 frame miss debounce. Row 0 of the sensor is the bottom edge of the panel, so y is mirrored; this is a fixed fact of the hardware, there is no option for it. The grid and the panel size are K114 only, and frames that are not 17861 bytes long are ignored. Tested on a K114c: five taps at the corners and the centre, in that order, landed at the matching points on a test page, and evtest shows X/Y changing under a finger. That run used the same parser with a transport that clamped the descriptor and a module option for the mirror; the clamp in this patch is checked in a userland test only. The firmware load and power sequence of the interface are in a separate patch. Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
The touch screen ("kiwi") of the 14 inch M5 Max MacBook Pro rides the
same dockchannel MTP as the multi-touch interface and announces itself
the same way, but it owns its reset and power lines instead of sharing
the AFE reset, and its 1.3 MB firmware image is written by the MTP
bootloader over the touch IC SPI, which takes seconds rather than the
1 s the comm timeout allows.
- A power method 2 interface may decode its own reset and power lines
from its device node; without its own reset line it keeps falling back
to the shared AFE reset, so existing multi-touch boards are unchanged.
- Announced GPIOs are tracked per interface (the MTP announces
clock_24MHz_enable, display_sync, power and reset for kiwi) instead
of failing on the second one, and a GPIO command is matched to its
announced id. The kiwi clock and display_sync pads are pad functions
in the device tree, not host GPIOs: they are left unset and a command
for them is refused.
- While the kiwi firmware load is running, comm commands use a 10 s
timeout, and the kiwi interface waits 10 s for its ready message.
- The kiwi start sequence, taken only when kiwi has a reset line of its
own, power-cycles the chip with a reset pulse (the reset line is active
low in the device tree), settles 50 ms, sends the firmware and then runs
the normal power method 2 transitions. The power line is raised on the
way up and left alone on the way down. The sequence is fixed code, not
tunables.
It needs the device tree side (the kiwi node, its reset and power lines
and the two pad functions), which is a separate patch.
The image installed as apple/tpkiwifw-k114c.bin fails to load at its
operations 26, 29 and 30, which are optional property placeholders
(address 0, not required). The verified runs used an image with those
three operations removed, so whatever installs the firmware needs to
drop them.
Compile tested against this tree with W=1. The start sequence, the
timeouts and the pad function settings are verified on hardware on a
7.2.11 based tree.
Booted on a K114c on this base with the first revision of this series: the
interface binds, reports ready with status 0 and streams 17861-byte frames
at about 123 Hz.
Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
…e properties The kiwi (touch screen) interface of the dockchannel MTP owns its power sequencing lines instead of sharing the multi-touch analog front end reset. Document apple,power-method, apple,reset-gpios and apple,power-gpios on the kiwi child node, mirroring how the multi-touch child documents its power method. The 24 MHz clock and display sync pads are pad functions selected by the transport node pinctrl, not GPIO properties, so they need no binding entries. Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
joshuaswarren
force-pushed
the
w7c/up-k114c-kiwi
branch
from
November 12, 2026 02:14
2974ebe to
81435c7
Compare
Contributor
Author
|
[Joshua's agent] Reworked for both points; the branch is now 81435c7, five commits rebased on integration/aurora de6a8d4.
Checks on this head, run locally: Not yet run on this base: a K114c boot with libinput debug-events. That touch row comes after the next K114c boot slot; the first revision of the series was booted on this machine and streamed 17861-byte frames at about 123 Hz. |
Contributor
|
Thanks Joshua, the rework is what we asked for. The pinctrl, dockchannel-hid and asahi.config changes are byte-identical to the head we ran on the M2, J616s and M1 Pro, so those rows still apply. Only the K114c DT moved, and the die0 node is gone. On the interrupts: please inherit them and drop the |
The touch screen of the 14 inch MacBook Pro with the M5 Max rides the dockchannel MTP as its own "kiwi" interface. It sequences its own power: with power method 2 the host drives the touchscreen reset (nub GPIO 25, active low) and its optional supply enable (nub GPIO 26), instead of the shared multi-touch AFE reset on the SMC. Enable the shared always-on NUB GPIO power domain and controller the way J714s does, with its seven interrupts, mux the 24 MHz clock (nub pin 22) and display sync (AP pin 11) pads through the transport node pinctrl, and enable the interface with its k114c firmware name. The reset polarity, the GPIO numbers and the two pad function values (pin 22 and pin 11, function 1) are those of the device tree that gave touch on the hardware on a kernel based on v7.2.11. Booted on a K114c on an earlier base with the first revision of this series: the interface binds, reports ready with status 0 and streams 17861-byte frames at about 123 Hz. Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
joshuaswarren
force-pushed
the
w7c/up-k114c-kiwi
branch
from
November 12, 2026 02:54
81435c7 to
ea00ed3
Compare
Contributor
|
J714s (M5 Pro) validation of 81435c7 on top of #1568 (1e06558): PASS.
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds the touch screen of the 14-inch PROLED MacBook Pro (K114c, M5 Max). The panel's touch controller is a "kiwi" interface on the same dockchannel MTP transport as the trackpad. It has its own reset and power lines and wants a 24 MHz clock pad and a display-sync pad.
What is in the stack, oldest first:
pinctrl: apple: the GPIO mode field is four bits wide. iBoot leaves bit 4 of the touch reset and power pins set; with it set the DATA bit does not latch in output mode, so the reset never took effect. Bits 1-4 are the mode now. Modes 0-7 are unchanged.HID: apple-mtp-kiwi: add the MTP touch screen driver, with its MAINTAINERS entry andCONFIG_HID_APPLE_MTP_KIWI=minasahi.config.HID: dockchannel-hid: bring up the kiwi (touch screen) interface: the reset, power and clock bring-up, and skipping announced GPIOs the device tree does not wire.dt-bindings: input: apple,dockchannel-hid: document the kiwi interface properties.arm64: dts: apple: t6050-k114c: enable the kiwi touch screen interface, including the pad functions: clock on NUB pin 22, display sync on AP pin 11, reset on NUB pin 25 (active low), power on NUB pin 26.It needs #1568 (the report-size clamp for this MTP firmware) underneath it.
Status, measured on the machine: the firmware is accepted,
Kiwi interface ready with status:0appears,/dev/hidraw2streams 17861-byte frames at about 123 Hz, and the input device "Apple MTP touch screen" registers. Finger events are not verified yet: I have not had a person at the panel since the interface came up. I will update this with alibinput debug-eventscapture. Marked draft until then.