Patches

Every fix we wrote for T2 Macs, and where it stands upstream. Click a patch for the details.

49patches
11merged
13KAIT2EN only

Under review22

Wi-Fi / brcmfmac BCM4364 and BCM4377 hotspot authenticationv2

The firmware handshake for its access point stalls on the BCM4364 and BCM4377, for WPA2 and WPA3, and clients drop after about 8 s. brcmfmac leaves the handshake to wpa_supplicant or hostapd.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac BCM4364 TX beamforming offv2

With TX beamforming the BCM4364 firmware sends one stream. Uplink rises from about 350 to 620 Mbit/s with beamforming off.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac BCM4355 and BCM4364 ND offload offv2

With ND offload on, downloads dropped to about 55 Mbit/s on the BCM4364 and about 50 Mbit/s on the BCM4355 of the MacBookAir8,1. brcmfmac keeps ARP and ND offload off on these chips on Apple boards while awake, as macOS does. Closes t2linux/kernel#34 and T2-Debian-and-Ubuntu-Kernel#227.

Linux kernel · @deqrocks, @AdityaGarg8 lore.kernel.org ↗
xHCI/USB Card reader power managementv1

We did not support cutting on/off power to the card reader with this patch we add, and we get a slight decrease in power within suspend.

Linux Kernel · @byte lore.kernel.org ↗
apple-gmux / amdgpu Hybrid graphics on T2 MacBook Prosv2

Fixes apple-gmux on T2 Macs: dGPU power sequencing, the MMIO interface, interrupts with hotplug wake of the dGPU for external monitors, and the gpu-active EFI variable the T2 reads. amdgpu gets vga_switcheroo runtime PM on top, so the dGPU powers off when idle. Tested on MacBookPro15,1, 16,1 and 16,4.

Linux kernel · @byte, @deqrocks lore.kernel.org ↗
DRM / i915 Panel Self Refresh on T2 eDP panelsv4

Enables PSR with the sink-specific DPCD setup. Lowers idle display power and allows package C-states up to PC10. KAIT2EN ships it in a patched i915 module until it is merged. Resend by AdityaGarg()

Linux kernel · @4f1sh3r lore.kernel.org ↗
T2 / apple_t2_bridge Upstreamable replacement for apple-bce and t2bce Preparing

Bridge controller, virtual USB host and audio for the T2, laid out for mainline. Suspend and runtime PM work.

Linux kernel · @deqrocks
Bluetooth / hci_bcm4377 Probe with another controller presentv2

With another controller on hci0, a USB dongle for example, probe returned the new index and the PCI core took it as success without the unregister action. The controller then stayed registered after unbind. hci_bcm4377 now treats only negative returns as errors. Reviewed by Sven Peter.

Linux kernel · @deqrocks lore.kernel.org ↗
Bluetooth / hci_bcm4377 Link power managementv2

hci_bcm4377 kept ASPM off on the whole link to the Wi-Fi and Bluetooth chip because of a hardware erratum, which held the MacBookAir9,1 package at PC3. ASPM is now off only while the firmware boots. Acked by Sven Peter.

Linux kernel · @deqrocks lore.kernel.org ↗
Thunderbolt Ice Lake runtime power offv1

The software connection manager enabled runtime PM only for USB4 routers, so the Ice Lake NHI of the MacBookAir9,1 never suspended. With runtime PM for the Ice Lake router PC10 at idle with the display on rises from 33% to 64%. Needs pcie_ports=compat removed, which the installer does.

Linux kernel · @deqrocks lore.kernel.org ↗
x86 / cache Fast CPU bring-up after suspendv1

The firmware leaves the secondary CPUs with invalid MTRRs after resume and x86 restored them only once all CPUs were online, so the CPUs ran uncached and a resume could take up to a minute. Each CPU now gets its MTRRs back as it comes online. Based on Chen Yu's fix.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac Deep sleep on BCM4364 and BCM4377v2

The firmware supports inband deep sleep, but brcmfmac left it disabled, so the chip stayed awake and its latency tolerance held the package out of PC10. brcmfmac now follows the deep sleep handshake of the firmware and the package reaches PC10 with Wi-Fi associated, without ignoring the LTR of the Wi-Fi port.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac Cold resume after S3v2

The chip loses its firmware in S3 but its registers can read as running, so brcmfmac took the hot resume path and commands timed out. It now reloads the firmware when platform firmware handled the resume and Wake-on-WLAN is off. The suspend service keeps the drivers loaded, so sleep starts 10 to 15 s sooner on BCM4377. The BCM4377 D3 acknowledgement uses patch 5 of Michael Reeves' BCM4388 series, carried until merged: https://lore.kernel.org/all/20260925-brcm4388-submit-v1-5-b5cbc2cde2fb@gmail.com/

Linux kernel · @deqrocks lore.kernel.org ↗
Bluetooth / hci_bcm4377 Cold resume after S3v2

The controller loses its firmware in S3. hci_bcm4377 now probes it again after such a resume and skips the ring teardown messages that each waited a second.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac BCM4377 flowring deletev2

The firmware reports queued packets only after the host deletes a flowring. brcmfmac waited up to 100 ms per flowring first, on every disconnect and suspend, and miscounted them. It now deletes right away.

Linux kernel · @deqrocks lore.kernel.org ↗
DRM / i915 Display brightness after screen lock and resume in iGPU modev1

On dual GPU T2 MacBook Pros gmux resets the brightness when i915 powers the panel off, so the display came back at 80% after screen lock, login and resume. i915 now writes the set level back once the panel is on. KAIT2EN ships it in a patched i915 module until it is merged.

Linux kernel · @deqrocks lore.kernel.org ↗
Bluetooth / btbcm LE scanning on Broadcom controllersv2

The controllers set extra bits in the LE advertising report type and the kernel dropped those reports. Controller duplicate filtering hid changed advertisements for up to 11 s. Both drivers turn the filter off, and btbcm clears the extra bits until Fedora ships commit 88397423671f from bluetooth-next. A Mac opening AirDrop shows up within a second.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac AWDL on BCM4364 and BCM4377v1

brcmfmac creates the AWDL interface, passes its firmware events to user space and carries data over awdl0. The t2-drop service drives it as macOS does. Patch 1 is dropped for the same fix in Michael Reeves' BCM4388 series, v2 follows on top of it.

Linux kernel · @deqrocks lore.kernel.org ↗
Bluetooth / btbcm ACL priority on Broadcom controllersv2

With 2.4 GHz Wi-Fi traffic the controller held Bluetooth packets for up to 200 ms and A2DP dropped out. hci_bcm and hci_bcm4377 set high priority for every link on connect, as the Android Broadcom vendor library does.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac Power save timeoutv1

brcmfmac used a 2 s power save timeout when NetworkManager set none. It now uses 200 ms, the macOS value on BCM4377. Idle power on MacBookPro15,1 with Wi-Fi connected dropped from 11.3 W to 9.2 W.

Linux kernel · @deqrocks lore.kernel.org ↗
DRM / amdgpu 5K display on iMac20,1 and iMac20,2v1

Drives the internal 5K panel at full resolution by genlocking its two display streams, including resolution changes. Closes drm/amd#4455 and #4895.

Linux kernel · @byte lore.kernel.org ↗

Merged upstream11

Wi-Fi / brcmfmac BCM4364/3 firmware selection without antenna SKUv1 Merged

ACPI on iMacPro1,1, iMac19,1 and iMac19,2 does not advertise the antenna SKU that brcmfmac requires for Apple firmware selection. The macOS firmware for BCM4364/3 does not depend on it, so the antenna SKU is optional on this chip and used only when ACPI provides it.

Linux kernel · @AdityaGarg8 lore.kernel.org ↗
HID / Input hid-apple ignores the T2 trackpadv1 Merged

Trackpad and keyboard share one VID/PID. Uses the existing APPLE_IGNORE_MOUSE quirk.

Linux kernel · @AdityaGarg8 lore.kernel.org ↗
HID / Touch Bar Runtime PM in suspend and resumev1 Merged

Removes nested runtime PM references in the HID callbacks that stalled Touch Bar suspend and resume.

Linux kernel · @deqrocks lore.kernel.org ↗
udev Display brightness across reboots Merged

Gives the backlight a stable path ID so the brightness setting persists.

systemd · @deqrocks github.com ↗
upower Keyboard backlight after re-enumeration Merged

Restores the keyboard backlight after the keyboard re-enumerates.

upower · @deqrocks gitlab.freedesktop.org ↗
Quick Settings Volume slider on DSP sinks Merged

Fixes crackling while dragging the Quick Settings volume slider on PipeWire DSP sinks.

GNOME Shell · @deqrocks gitlab.gnome.org ↗
PCI / amdgpu dGPU after resume on MacBookPro15,1v1 Merged

Fixes the SMU being unresponsive after resume.

Linux kernel · @FrederickGeek8, @deqrocks lore.kernel.org ↗
HID / Touch Bar Fn double press switches the layerv5 Merged
Linux kernel · @AdityaGarg8, @deqrocks lore.kernel.org ↗
HID / Input Keyboard backlight brightness levelsv1 Merged

Parses the backlight reports with the correct endianness.

Linux kernel · @deqrocks lore.kernel.org ↗
HID / Input Keyboard backlight after resumev1 Merged

Restores the butterfly keyboard backlight level after resume and uses the Magic Keyboard LED naming.

Linux kernel · @deqrocks lore.kernel.org ↗

Shipped in KAIT2EN, upstream pending13

Thunderbolt Titan Ridge runtime power off KAIT2EN

Idle Titan Ridge controllers of T2 MacBook Pros are powered off with the macOS sequence on their PCIe root ports, which lets the package go below PC7. KAIT2EN ships it in a patched thunderbolt module, upstream it needs a PCI quirk for these ports.

KAIT2EN Fedora · @deqrocks
t2-truetone / ambient light sensor True Tone and evening mode KAIT2EN

T2 TrueTone adapts the display white to the color of the ambient light through the panel's gamma LUT, with a strength slider, and warms it on a daily schedule. Replaces GNOME Night Light. Off until turned on in the app.

KAIT2EN Fedora · @deqrocks Documentation
t2-truetone / ambient light sensor Automatic brightness (ALS) KAIT2EN

Part of T2 TrueTone. Follows the ambient light sensor with the macOS curve, a 20 % dead band, slow fades and learning from manual changes. Replaces GNOME automatic brightness.

KAIT2EN Fedora · @deqrocks Documentation
t2-airshare T2 AirShare (AirDrop and Quick Share) KAIT2EN

AirDrop with Apple devices in both directions while the Mac stays on its Wi-Fi network, and Quick Share with Android phones over Bluetooth and a Mac hotspot. One send window, one accept notification and one trusted device list.

KAIT2EN Fedora · @deqrocks Documentation
HID / t2hid Butterfly keyboard debounce KAIT2EN

Optional filter for double-registered keystrokes, enabled with t2hid.debounce_ms=40.

KAIT2EN Fedora · @deqrocks Documentation
t2-sidecar Low-latency video encoding (T2 Sidecar encoder) KAIT2EN

The T2 Sidecar encoder as a V4L2 H.264/HEVC encoder, usable through ffmpeg and the t2-sidecar Sunshine build https://github.com/deqrocks/Sunshine.

KAIT2EN Fedora · @deqrocks Documentation
t2smc / t2-zone-power SMC power limit steps KAIT2EN

The SMC requests lower CPU and GPU limits in steps. Linux ignored them until PROCHOT dropped the CPU to 800 MHz. t2smc exposes the steps and T2 Zone Power applies them as macOS does. An amdgpu patch lets the Navi GPUs go down to 40 % of TDP.

KAIT2EN Fedora · @deqrocks Documentation
t2bdrm / t2touchbar Touch Bar daemon KAIT2EN

Optional daemon with media and F-key rows, gestures for volume and brightness, the Touch ID prompt and media track display. Stays dark until touched.

KAIT2EN Fedora · @deqrocks Documentation
Input / Force Touch Force Click trackpad controls KAIT2EN

Adjustable click and Force Click pressure, Taptic Engine feedback and configurable Force Click actions.

KAIT2EN Fedora · @deqrocks
snd-apple-t2 Audio DSP KAIT2EN

Speaker and microphone DSP graphs per model, derived from the macOS processing.

KAIT2EN Fedora · @deqrocks Documentation
t2-touchid / BiometricKit Touch ID login KAIT2EN

Fingers enrolled under macOS unlock login and sudo through stock fprintd. Enrollment on Linux is open. Target is an upstream libfprint driver.

KAIT2EN Fedora · @deqrocks Documentation
t2-ave Hardware video encoding (T2 AVE) KAIT2EN

HEVC encoding on the T2 Apple Video Encoder, usable through ffmpeg and the t2-ave HandBrake build https://github.com/deqrocks/HandBrake.

KAIT2EN Fedora · @deqrocks Documentation
Power management Package C-states KAIT2EN

Deep package C-states at idle for lower power draw.

KAIT2EN Fedora · @deqrocks

Superseded or withdrawn3

Wi-Fi / brcmfmac Firmware reload after a missing D3 ACKv1 Superseded

Superseded by the cold resume series above. Patch 3 reloads the firmware after a resume through platform firmware and patch 4 drops the tx status wait on flowring delete.

Linux kernel · @deqrocks lore.kernel.org ↗
Wi-Fi / brcmfmac BCM4377 missing D3 ACK workaroundv2 Superseded

No longer needed. With the mailbox over the control ring from Michael Reeves' BCM4388 series the BCM4377 acknowledges D3.

Linux kernel · @deqrocks lore.kernel.org ↗
ACPI / CPU hotplug CPU restore after suspendv1 Superseded

Withdrawn in favor of the MTRR restore on resume below, which fixes the cause instead of offlining the CPUs early.

Linux kernel · @deqrocks lore.kernel.org ↗