[PATCH v2 00/22] Initial USB4/Thunderbolt support for Apple M1/M2/M3 SoCs

[Joshua Peisach]

On Sun Sep 6, 2026 at 2:36 PM EDT, Sven Peter wrote:
> Hi,
>
> This series adds initial USB4/Thunderbolt support for Apple Silicon
> M1, M2 and M3 SoCs. Each USB4 capable Type-C port comes with an ACIO
> ("Apple Converged I/O") block which contains a Cortex-M3 co-processor
> and the hardware implementing the USB4 router.
>
> The NHI and DART IOMMU are part of this ACIO block as well. Instead of
> being exposed as independent always-accessible devices the main SoC bus
> only gets a 16 MiB window into ACIO's MMIO space while its co-processor
> is running. It is not entirely clear whether this is ACIO's own address
> space or a window into the co-processor's address space.
>
> This is why the device tree represents the NHI and DART as children of
> ACIO with addresses relative to that window. The driver only populates
> these children after booting the co-processor and removes them again
> before powering ACIO down. This setup is slightly unusual but matches
> both the MMIO layout and the lifetime of the devices.
>
> ACIO also has rather strict ordering requirements: it can only be
> started after the Type-C PHY has been configured for USB4/Thunderbolt
> (and ideally has to be stopped again before the PHY is powered down but
> this doesn't break anything unlike getting startup wrong). It also
> needs cable details which are only known to the Type-C port controller.
>
> The first two patches add synchronous Type-C alternate mode state
> notifiers and support for representing USB4 connections on the Type-C
> bus. USB4 uses Enter_USB instead of Enter Mode and has no SVID, so it
> gets its own mode kind and stores the negotiated EUDO directly.
> The next two patches register a USB4 port mode and publish
> the negotiated Thunderbolt and USB4 partner modes on the Type-C bus,
> along with the Thunderbolt cable mode. These modes are marked active
> after the mux has been configured and deactivated before it is switched
> back to its safe state.
>
> The following two patches add the bindings for the NHI and ACIO. The
> Thunderbolt core patches prepare for the different NHI register layout
> and interrupt handling, read the USB4 router DROM from the device tree,
> unlock the host router ports and find the Apple VSE capability. They
> also add a hook for device links to tunneled native ports, a quirk for
> USB3 bandwidth allocation not implemented by the Apple hardware and
> the symbol exports needed by the driver.
>
> The main driver registers the newly introduced notifications for the
> negotiated Type-C modes, powers up ACIO, boots the co-processor,
> populates the DART and NHI children once their MMIO regions are
> accessible and registers the NHI with the software connection
> manager. The last three patches add the device tree nodes
> for the M1 and M2. M3 nodes will follow once the rest of the stack is
> upstream, but the current Thunderbolt/USB4 code has already been
> successfully tested on those.
>
> Right now only XDomain connections and USB3-via-USB4 tunnels are
> supported. Both PCIe and DisplayPort tunnels will come later since those
> require additional work and reverse engineering that is not done yet.
> Suspend is also not supported yet and we actually have to prevent
> suspending whenever an active connection is present because we would
> otherwise resume into an SError.
>
> This series depends on the Apple CIO reset controller [1], the ATCPHY
> USB3-via-4 pipehandler support [2], support for specifying the DART DMA
> aperture in the device tree [3], a fix for CD321x Thunderbolt
> altmode VDOs [4], and fixes for making the DPTX capabilities read fail
> gracefully [5]. I couldn't split it any further since the remaining
> patches now directly depend upon each other.
> Except for the dts changes the series applies cleanly without these
> dependencies though.
>
> For merging we'll probably need an immutable brach with the first two
> commits after they've been reviewed which is then merged in both the usb
> and thunderbolt tree sincethere are other tipd changes in flight as well.
>
> Happy to split this series in two as well if you prefer that.
>
> Best,
>
> Sven
>
> [1] https://patch.msgid.link/20260821-b4-cio-reset-v2-0-2d045f80a424@kernel.org
> [2] https://patch.msgid.link/20260821-b4-atcphy-usb4-v1-0-45c0b741e0c0@kernel.org
> [3] https://patch.msgid.link/20260819-iommu-apple-dart-aperture-v1-0-252703f381aa@jannau.net
> [4] https://patch.msgid.link/20260813-b4-tipd-vdo-fix-v1-1-70317f2cd554@kernel.org
> [5] https://patch.msgid.link/20260829-b4-tbt-fixes-v3-0-e1fab6ac54fe@kernel.org
>
> ---
> Changes in v2:
> - Dropped thunderbolt switch in favor of the typec bus and an additional
>   notifer
> - Add USB4 to the Type-C bus with its own mode kind instead of an SVID
> - Register port/partner altmodes for CD321x
> - Replaced the struct with ring MMIO offsets with two callbacks insid
>   tb_nhi_ops
> - Moved port unlock sequence and finding the Apple VSE capability into
>   core tb code
> - Added device links for USB3 tunnels
> - Move the USB3 BW quirk from an NHI quirk to a router quirk
> - Added Joshua's rb to patches which didn't change
> - Link to v1: https://patch.msgid.link/20260830-b4-apple-soc-tbt-v1-0-44bc9348683c@kernel.org
>
>
> ---
> Sven Peter (22):
>       usb: typec: Add alternate mode state notifiers
>       usb: typec: Represent USB4 on the Type-C bus
>       usb: typec: tipd: Register a USB4 port mode for CD321x
>       usb: typec: tipd: Publish CD321x partner alternate modes
>       dt-bindings: thunderbolt: Add Apple USB4/Thunderbolt NHI
>       dt-bindings: thunderbolt: Add Apple USB4/Thunderbolt ACIO block
>       thunderbolt: Try reading host DROM from device tree first
>       thunderbolt: Don't read the UID if we already know it
>       thunderbolt: Allocate ring HopID before requesting the ring interrupt
>       thunderbolt: Unlock host router ports during startup
>       thunderbolt: Find Apple VSE capability during startup
>       thunderbolt: Add ring_interrupt_active to tb_nhi_ops
>       thunderbolt: Add ring register accessors to tb_nhi_ops
>       thunderbolt: Add ring_interrupt_mask to tb_nhi_ops
>       thunderbolt: Add ring_configure to tb_nhi_ops
>       thunderbolt: Add add_links to tb_nhi_ops
>       thunderbolt: Add QUIRK_NO_USB3_BW_ALLOC
>       thunderbolt: Export symbols required by the Apple Silicon driver
>       thunderbolt: Add Apple Silicon support
>       arm64: dts: apple: t8103: Add USB4 ACIO and NHI
>       arm64: dts: apple: t8112: Add USB4 ACIO and NHI
>       arm64: dts: apple: t60xx: Add USB4 ACIO and NHI
>

For patches

3: usb: typec: tipd: Register a USB4 port mode for CD321x
4: usb: typec: tipd: Publish CD321x partner alternate modes
10: thunderbolt: Unlock host router ports during startup
13: thunderbolt: Add ring register accessors to tb_nhi_ops
15: thunderbolt: Add ring_configure to tb_nhi_ops
16: thunderbolt: Add add_links to tb_nhi_ops
17: thunderbolt: Add QUIRK_NO_USB3_BW_ALLOC
19: thunderbolt: Add Apple Silicon support

Reviewed-by: Joshua Peisach <jpeisach at ubuntu.com>

Side note: I'm probably going to stop reviewing devicetrees as I am
not really an expert in the full hardware definitions and how it
should work.