Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion docs/matrix.md
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,7 @@ panel).
| Camera | aura #188 (K116s row), K114s row 0xba in [`patches/aura/camera/0001`](../patches/aura/camera/0001-kboot-T6050-hand-the-K114s-camera-over-on-27.2.patch) (PR #123); linux #1648 camera block, #1831 (27.2 parser, draft) | K114s ADT `isp0` equals the K116s in every value except sensor-type 0xba (agent-run 2026-11-15, `tests/k114s-adt-camera-facts.py`); `isp` node disabled, no `isp` alias, no `/dev/video*` today | evaluated; loader row and DTS draft (`dts/t6050-k114s-isp.dtsi`, off by default) written, not booted (#33, PR #123, [docs/r3-camera.md](r3-camera.md)) | Own one-boot trial after the display hand-off, with linux #1831. Owner: look at a frame (`tests/hardware/k114s-camera-check.sh --capture`). Open: sensor preset (K116s config-index 5 vs K114c 6), 27.2 firmware files unmatched, camera indicator needs the DCP |
| Ambient light, lid and other AOP sensors | linux #1645, #1751 (in base as series 0019); #1983, #1929, #1876, #1924 open; linux #1688, draft (K114c: built-in panel as eDP so lid close suspends) | Draft fragment `dts/t6050-k114s-aop-sensors.dtsi` enables the AOP and its `als` and `las` children; AOP, mailbox, setup port and dart-aop equal the ADT; the ADT has an `als` node, **no `las` node**, and no ALS calibration | **drafted, ADT-checked** (agent-run, 2026-11-15; [docs/r3-parity-drafts.md](r3-parity-drafts.md)); in no tree, never booted | Boot with the sound AOP (enabled once); lid angle may not exist on this firmware; ALS may need #1983/#1929 (driver changes) |
| Touch ID / Secure Enclave | linux #1649 (K116s SEP profile); `aurora-sep`, `libfprint-aurora` in the image | Research only. No enrollment and no Secure Enclave writes | evaluated; ADT sensor and SEP facts all equal the J714s of #1587 (18 facts, agent-run); draft fragment `dts/t6050-k114s-touchid.dtsi` (PR #124, [docs/r3-touchid.md](r3-touchid.md)); installed `apple_sep` does not admit `apple,k114s`; no `/aliases/sep` in the booted tree; omarchy-mac says `unsupported:board` for T6050 (#49) | Upstream adds the K114s line to the SEP profile (docs/r3-touchid.md in PR #124) |
| Battery, charging | K114c reports charge and status; linux #1615, open (uevent fix, in r3a as `sleep-lid/0007`) | Should behave as K114c | **hardware-verified** for charge, state, capacity and cycle count (agent-run); the `manufacture_year` -95 defect is fixed on r3a via `sleep-lid/0007` (#1615, open draft upstream; agent-run 2026-11-15, [r3-power.md](r3-power.md); #52); discharge sysfs and UPower agreement are still owner-run. Charge limit: thresholds present (100/100), no limit ever saved; script and restore pass a fake-sysfs test (replica); the udev rules were only syntax-checked, not exercised; udev now handles `macsmc-battery` on r3a, so the boot-restore blocker is gone, but the owner charge-limit run is still needed (#52); SMC behaviour untested. Charging path: `macsmc-ac` online, input limit 59.8 W, neither USB-C port has a partner and both tps6598x supplies are offline, so the charger is on MagSafe or the left-back port (agent-run); MagSafe has no node. See [r3-power.md](r3-power.md) | Owner runs of `tests/hardware/k114s-charge-limit-test.sh`, `k114s-battery-log.sh`, `k114s-plug-probe.sh` |
| Battery, charging | K114c reports charge and status; linux #1615, open (uevent fix, in r3a as `sleep-lid/0007`) | Should behave as K114c | **hardware-verified** for charge, state, capacity and cycle count (agent-run); the `manufacture_year` -95 defect is fixed on r3a via `sleep-lid/0007` (#1615, open draft upstream; agent-run 2026-11-15, [r3-power.md](r3-power.md); #52); discharge sysfs and UPower agreement are still owner-run. Charge limit: thresholds present (100/100), no limit ever saved; script and restore pass a fake-sysfs test (replica); the udev rules were only syntax-checked, not exercised; udev handles `macsmc-battery` on r3a, but **not on r3b to r3b.c1** (no `sleep-lid/0007`): a saved 80 % limit read 100 after a reboot into r3b.c1 (agent-run, 2026-11-16); r3b.c2 carries the fix ([r3b-kernel.md](r3b-kernel.md) section 10); SMC behaviour untested. Charging path: `macsmc-ac` online, input limit 59.8 W, neither USB-C port has a partner and both tps6598x supplies are offline, so the charger is on MagSafe or the left-back port (agent-run); MagSafe has no node. See [r3-power.md](r3-power.md) | Owner runs of `tests/hardware/k114s-charge-limit-test.sh`, `k114s-battery-log.sh`, `k114s-plug-probe.sh` |
| Fans, thermals (SMC sensors) | linux #1648 hwmon part (K116s) | Ported as a K114s copy | **hardware-verified** for the voltages (agent-run); fans read about 2200 and 2375 RPM (min 2000) at 22-26 W desktop load on r3a (agent-run, #55); sustained-load run still open; TAOC, TAOI, TAOJ, TAOP read -127 °C and TfCB and a few more read 0 (agent-run); no CPU thermal zone; read-only logger `tests/hardware/k114s-thermal-log.sh` tested on a fake tree (replica) | Owner sustained-load run with the logger; mask the TAO* keys after the ADT key list |
| Power management, performance | linux `cd2c9f52` (K116s holds back the CPU PMU and CLPC link); #1789, open (PMP domains) | Follow the K116s | evaluated; profile `performance` on AC is the `omarchy-powerprofiles-set` default from omarchy-dev (`balanced` on battery unless saved), governor `performance` on `apple-cpufreq`, set by `omarchy-power-profiles-daemon` (AC default profile `performance`), not the kernel, whose default is `schedutil` ([r3-cpufreq.md](r3-cpufreq.md), PR #95; #54), `omarchy-powerprofiles-floor` not running (agent-run); re-checked 2026-11-15 on r3a, unchanged (#56); idle draw about 10 to 14 W on AC; profile logic passes a stub test (replica); battery draw and the flip on unplug not measured. See [r3-power.md](r3-power.md) | With r3; owner discharge run with `k114s-battery-log.sh`, and a balanced run with `k114s-cpufreq-log.sh` |
| Sleep, resume, lid | linux #1647, open (K116s s2idle with the N1 up), #1844, #1789, #1688 | With the N1 on, the current kernel refuses suspend. Needs #1647 plus an `apple,k114s` line. Deep sleep does not exist on 27.2 | **blocked on purpose**: r3a booted with `sleep-lid/0001-0003` (#1647 with an `apple,k114s` line), the GPU mailbox interrupt is proven (about 400/s), `suspend_stats` reads 0/0 and the suspend targets are masked (agent-run; PR #109; #50). Only the `pm_test` freezer level is safe now | Open: `sleep-lid/0010` (not in r3a; the interrupt gate that kept it out is met, #34), PMP nodes (0008 inert), the first attended cycle |
Expand Down
31 changes: 30 additions & 1 deletion docs/r3b-kernel.md
Original file line number Diff line number Diff line change
Expand Up @@ -59,7 +59,7 @@ changes. All agent-run.
| `sleep-lid/0002` (devices below a kernel-trained port may leave D0, #1647) | **not needed on a K114s**, not carried | it exists for the K116s's second, kernel-trained port (SD reader). No tree of this repository enables the K114s SD reader (`dts/t6050-k114s-sd.dtsi` is a draft, included nowhere). Needed again only if that draft is enabled |
| `sleep-lid/0003` (K114s admitted to s2idle) | **still needed**, re-cut to apply without 0001 and 0002: `r3b/0005` | forward fails and reverse fails (context from 0001/0002); the K114c is already admitted in the base, the K114s is not (`apple_pcie_check_adopted_s2idle`) |
| `sleep-lid/0004`, `0005`, `0006` (#1688: eDP connector for a boot framebuffer marked `apple,internal-panel`; the K114/K116 dtsi line; binding) | **T6050-generic, missing, not needed to boot: send to betizzel** (section 7) | forward applies, reverse fails, no log hit |
| `sleep-lid/0007` (#1615: unreadable battery manufacture date) | **T6050-generic, missing: send to betizzel** | forward applies; log none. It makes an unreadable manufacture date return `-ENODATA`, which the power-supply core skips. Observed on this K114s: 30 `manufacture_year ... -95` lines in five minutes (`docs/r3-sleep-lid.md`) |
| `sleep-lid/0007` (#1615: unreadable battery manufacture date) | **T6050-generic, missing: send to betizzel** | forward applies; log none. It makes an unreadable manufacture date return `-ENODATA`, which the power-supply core skips. Observed on this K114s: 30 `manufacture_year ... -95` lines in five minutes (`docs/r3-sleep-lid.md`). Not carried in r3b to r3b.c1, which was wrong: it also decides whether a saved charge limit comes back at boot (section 10). Carried from r3b.c2 |
| `sleep-lid/0008` (PMP power domains always on, #1789) | **in v272-11** (their 0028, same patch) | reverse applies cleanly; all 8 added lines present |
| `sleep-lid/0009` (GPU mailbox receive by interrupt, `41c359da`, #1844) | **in v272-11** | log: merge `adb5a7e344` of #1844, commits `41c359da81` and `788441e089` are in the history. See risk 1 |
| `display-gpu/0003` to `0006` (#1868: 120 Hz-only T6050 internal panel modes) | **in v272-11**, in upstream's merged shape | log: merge `1177fdf5f9` (#1868), `3176918ba2`, `84b3b93c04`, `10f14b7ff8`. The base scopes the gate with `dcp->hw.only_120hz_modes`, our `0006` did it with an fw-compat test: same behaviour, different lines (`0006` reverse applies; `0003` to `0005` differ in a few comment and condition lines only) |
Expand Down Expand Up @@ -337,3 +337,32 @@ connector stays `Unknown-1` (the `eDP-1` monitor rule and every `eDP-1 connected
'magicmouse|report_size'` and pointer movement); the lid close tests are blocked by the connector.

Packaging, `check-package.sh` and the Stage 2 builder are reusable as they are.

## 10. r3b.c2: the battery fix the r3b series left out (2026-11-16)

**Seen on hardware (owner session, 2026-11-16).** The charge limit was set to 80 % with `omarchy-battery-charge-limit 80`
(the SMC took it: end 80, start 75; `/etc/udev/macsmc-battery.conf` holds `CHARGE_CONTROL_END_THRESHOLD=80`). After one
reboot into r3b.c1 the control file read **100** again (agent-run). That boot logged `failed to report 'manufacture_year'
property: -95` ten times or more, and `udevadm info` for `macsmc-battery` listed no properties beyond `DEVPATH` and
`SUBSYSTEM`. So no udev rule ran for the battery, neither asahi-scripts' `93-macsmc-battery-charge-control` nor Omarchy's
`94-omarchy-mac-battery-charge-limit`, and nothing restored the saved limit. The r3b.b1 boots before it log the same lines.
r3a carries the fix (`sleep-lid/0007`, aurora-silicon/linux#1615), and r3b did not take it (section 1, "send to betizzel");
since the D9 promotion r3b is what this Mac boots.

**The fix.** `patches/linux/r3b-c2.series`: r3b.c1 plus `sleep-lid/0007`, unchanged (it applies to the r3b.c1 tree
`08ddebb521` without fuzz). The release, the package name and the configuration stay; the version is `7.3.0.r3b.c2-1`, so it
upgrades r3b.c1 in place. `PKGBUILD` refuses `K114S_CAND=r3b-c2` on a tree without the fix.

| Check | Result | Label |
| --- | --- | --- |
| `git am` of `sleep-lid/0007` on r3b.c1 | head `ecbecd26ad`, tree `e078a06e67c5` | agent-run |
| `K114S_LABEL=K114S-R3B-C2 kmake O=build/r3b-c2 Image modules dtbs` (full build, the r3b.c1 `.config`) | exit 0, no warning in `drivers/power/supply` or `drivers/gpu/drm/apple`; release `7.3.0-17.v272-11-k114s-r3b-ARCH` | agent-run |
| `k114s-cmp-modules build/r3b-c2` against the installed r3b.c1 | 1972 of 1972 SAME: `macsmc-power` is built in (`CONFIG_MACSMC_POWER=y`), so only the Image differs | agent-run |
| `K114S_CAND=r3b-c2 makepkg`, then `check-package.sh` against `linux-aurora` and `linux-k114s` | all passed (the brightness floor, the Display P3 default and the 500-nit cap are checked for every `r3b.c<N>`) | agent-run |

Package (outside the repository, `$K114S_BUILD_ROOT/integrate/release-r3b-c2/`, with `SHA256SUMS`):
`linux-k114s-r3b-7.3.0.r3b.c2-1-aarch64.pkg.tar.zst`, sha256 `57dfac86c2a51d73847a7e3e114fc1eb7b378282a73ca4b4c94ab5b6be8ed20f`.
Install as r3b.c1 (`docs/r3-hdr.md`, section 6 step 1, and block P9 before and after); after the reboot,
`charge_control_end_threshold` must read the saved value and the boot log must have no `-95` line. The way back is the r3b.c1
package (`$K114S_BUILD_ROOT/integrate/release-r3b-c1/`) or the r3a entry.

22 changes: 16 additions & 6 deletions packaging/linux-k114s/PKGBUILD
Original file line number Diff line number Diff line change
Expand Up @@ -22,10 +22,14 @@
# r3b-c1 (docs/r3-hdr.md, issue #92) is r3b-b1 plus the HDR series (patches/linux/r3b-c1.series): the same package name
# and release again, version 7.3.0.r3b.c1-1, so it upgrades r3b.b1 in place the same way.
#
# r3b-c2 (docs/r3b-kernel.md "r3b.c2", aurora-silicon/linux#1615) is r3b-c1 plus sleep-lid/0007, the battery
# manufacture-date fix (patches/linux/r3b-c2.series): version 7.3.0.r3b.c2-1, upgrading r3b.c1 in place.
#
# This file does not compile anything. It packages a finished build:
# K114S_CAND the candidate, r3a, r3b, r3b-b1 or r3b-c1 (default r3a, the one the runbook docs/r3-integration.md names;
# K114S_CAND the candidate, r3a, r3b, r3b-b1, r3b-c1 or r3b-c2 (default r3a, the one the runbook docs/r3-integration.md names;
# r3b: docs/r3b-kernel.md, K114S_CAND=r3b makepkg ...; r3b-b1: D5a, K114S_CAND=r3b-b1 makepkg ...;
# r3b-c1: docs/r3-hdr.md, K114S_CAND=r3b-c1 makepkg ...)
# r3b-c1: docs/r3-hdr.md, K114S_CAND=r3b-c1 makepkg ...; r3b-c2: docs/r3b-kernel.md,
# K114S_CAND=r3b-c2 makepkg ...)
# K114S_SRC the kernel source tree (default $K114S_BUILD_ROOT/src/<candidate>)
# K114S_OUT its output directory (default $K114S_BUILD_ROOT/build/<candidate>)
# with the toolchain of $K114S_BUILD_ROOT (default ~/k114s-build; env.sh and bin/kmake there).
Expand All @@ -39,7 +43,8 @@ case $_cand in
r3b) pkgname=linux-k114s-r3b; _kcand=r3b; _ver=r3b; _rel=1 ;;
r3b-b1) pkgname=linux-k114s-r3b; _kcand=r3b; _ver=r3b.b1; _rel=1 ;; # vercmp: 7.3.0.r3b.b1-1 is newer than 7.3.0.r3b-1
r3b-c1) pkgname=linux-k114s-r3b; _kcand=r3b; _ver=r3b.c1; _rel=1 ;; # vercmp: 7.3.0.r3b.c1-1 is newer than 7.3.0.r3b.b1-1
*) echo "K114S_CAND must be r3a, r3b, r3b-b1 or r3b-c1" >&2; exit 1 ;;
r3b-c2) pkgname=linux-k114s-r3b; _kcand=r3b; _ver=r3b.c2; _rel=1 ;; # vercmp: 7.3.0.r3b.c2-1 is newer than 7.3.0.r3b.c1-1
*) echo "K114S_CAND must be r3a, r3b, r3b-b1, r3b-c1 or r3b-c2" >&2; exit 1 ;;
esac
pkgbase=$pkgname
pkgver=7.3.0.$_ver
Expand All @@ -66,16 +71,21 @@ package() {
[[ $_kcand != r3b || $kernel_release != 7.3.0-17.v272-4-k114s-r3a-ARCH ]] || { echo "r3b release equals r3a's" >&2; return 1; }
[[ $kernel_release == *-k114s-${_kcand}-ARCH ]] || { echo "release '$kernel_release' is not a $_kcand release (missing the k114s release suffix?)" >&2; return 1; }
# r3b-b1 must be built from a tree with the brightness floor (patches/linux/brightness 0006), never from plain r3b
if [[ $_cand == r3b-b1 || $_cand == r3b-c1 ]]; then
if [[ $_cand == r3b-b1 || $_cand == r3b-c1 || $_cand == r3b-c2 ]]; then
grep -q 'dcp_bl_set_minimum' "$_src/drivers/gpu/drm/apple/dcp_backlight.h" ||
{ echo "$_src has no brightness series (patches/linux/r3b-b1.series)" >&2; return 1; }
fi
# r3b-c1 must carry the HDR series (patches/linux/r3b-c1.series) with the K114s 500-nit EDR cap
if [[ $_cand == r3b-c1 ]]; then
# r3b-c1 and r3b-c2 must carry the HDR series (patches/linux/r3b-c1.series) with the K114s 500-nit EDR cap
if [[ $_cand == r3b-c1 || $_cand == r3b-c2 ]]; then
grep -q 'fw27_panel_edr_max_nits = 500;' "$_src/drivers/gpu/drm/apple/dcp.c" &&
grep -q 'dcp_panel_color_target' "$_src/drivers/gpu/drm/apple/dcp-display-metadata.c" ||
{ echo "$_src has no HDR series with the 500-nit cap (patches/linux/r3b-c1.series)" >&2; return 1; }
fi
# r3b-c2 must carry sleep-lid/0007: an unreadable manufacture date is -ENODATA, not -EOPNOTSUPP (aurora-silicon/linux#1615)
if [[ $_cand == r3b-c2 ]]; then
grep -A3 '^static int macsmc_battery_get_date' "$_src/drivers/power/supply/macsmc-power.c" | grep -q 'return -ENODATA;' ||
{ echo "$_src has no battery manufacture-date fix (patches/linux/r3b-c2.series, sleep-lid/0007)" >&2; return 1; }
fi
modpath="$pkgdir/usr/lib/modules/$kernel_release"

# vmlinuz: the uncompressed Image, as linux-aurora ships it
Expand Down
31 changes: 31 additions & 0 deletions patches/linux/r3b-c2.series
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
# K114s r3b.c2 kernel (aurora-silicon/linux#1615): r3b.c1 plus the battery manufacture-date fix.
#
# Same base, release suffix and configuration as r3b.c1 (patches/linux/r3b-c1.series): the release stays
# 7.3.0-17.v272-11-k114s-r3b-ARCH and the package stays linux-k114s-r3b (version 7.3.0.r3b.c2-1), so it upgrades the
# installed r3b.c1 package in place and replaces its UKI (omarchy_linux-k114s-r3b.efi). macsmc-power is built in
# (CONFIG_MACSMC_POWER=y), so only the Image differs; every module is the same.
# Build: K114S_LABEL=K114S-R3B-C2 (uname -v), output $K114S_BUILD_ROOT/build/r3b-c2, package K114S_CAND=r3b-c2.
# Why (seen on hardware, 2026-11-16, docs/r3b-kernel.md "r3b.c2"): r3b to r3b.c1 lack sleep-lid/0007, so every boot
# logs `failed to report 'manufacture_year' property: -95`, the battery's uevent fails, udev never handles
# macsmc-battery and a saved charge limit (/etc/udev/macsmc-battery.conf) is not restored at boot.
#
# 1. the r3b.b1 series (patches/linux/r3b-b1.series), unchanged and in its order
r3b/0001-HID-dockchannel-hid-clamp-Report-Size-items-where-th.patch
r3b/0002-HID-dockchannel-hid-pad-short-reports-to-the-clamped.patch
r3b/0003-HID-magicmouse-drop-the-GD-Mouse-usages-on-dockchann.patch
r3b/0004-HID-magicmouse-parse-the-0x5a-MTP-report-of-the-K114.patch
sound/0004-ASoC-apple-macaudio-Add-K114-with-the-speakers-off.patch
r3b/0005-PCI-apple-qualify-the-K114s-for-suspend-to-idle-over.patch
# 2. localversion.10-pkgrel -17.v272-11-k114s-r3b (as r3b; not a patch file)
brightness/0001-drm-apple-answer-the-PMU-temperature-for-every-panel.patch
brightness/0002-drm-apple-apply-the-syslog-debug-filter-to-the-inter.patch
brightness/0003-drm-apple-take-the-27.2-D590-notification-as-a-no-op.patch
brightness/0005-dt-bindings-display-apple-dcp-optional-panel-brightn.patch
brightness/0006-drm-apple-hold-brightness-requests-at-the-panel-s-fl.patch
# 3. the HDR series (docs/r3-hdr.md), applied with `git am` on top of 1 and 2. 0001 and 0002 are offered upstream
# (aurora-silicon/linux#2060); 0003 is K114s-only (the 500-nit default of the cap).
hdr/0001-drm-apple-describe-an-internal-panel-without-a-known.patch
hdr/0002-drm-apple-cap-the-panel-peak-a-firmware-27-panel-s-E.patch
hdr/0003-drm-apple-K114s-no-panel-EDR-above-500-nits-by-defau.patch
# 4. the battery fix, as r3a carries it (r3a.series), applied with `git am` on top of 1 to 3
sleep-lid/0007-power-supply-macsmc-skip-an-unreadable-manufacture-d.patch
7 changes: 5 additions & 2 deletions patches/linux/r3b.series
Original file line number Diff line number Diff line change
Expand Up @@ -30,5 +30,8 @@ r3b/0005-PCI-apple-qualify-the-K114s-for-suspend-to-idle-over.patch
# display-gpu/0009, 0010, sound/0001, usb/0001 (reverse-apply cleanly)
# not needed on a K114s (not carried): sleep-lid/0001, 0002 (K116s, #1647; 0002 only for a kernel-trained second port)
# T6050-generic, missing in v272-11, the K114s boot does not need them (not carried, send to betizzel):
# sleep-lid/0004, 0005, 0006 (#1688, eDP connector for apple,internal-panel), sleep-lid/0007 (#1615,
# battery manufacture_year), sound/0002 (tas2764 bus quiet), sound/0003 (i2c-pasemi NACK)
# sleep-lid/0004, 0005, 0006 (#1688, eDP connector for apple,internal-panel), sound/0002 (tas2764 bus quiet),
# sound/0003 (i2c-pasemi NACK)
# sleep-lid/0007 (#1615, battery manufacture_year) was in this list, and that was wrong: without it the battery's
# uevent fails with -95, so udev never handles macsmc-battery and a saved charge limit is not restored at boot
# (seen on hardware, 2026-11-16). r3b.c2 carries it (patches/linux/r3b-c2.series).