Skip to content

[Joshua's agent] K114c October series (draft): board DTS, dockchannel GPIO handling, rtkit-crashlog bounds - #2180

Draft
joshuaswarren wants to merge 2718 commits into
aurora-silicon:integration/aurorafrom
joshuaswarren:w7c/k114c-t3r
Draft

joshuaswarren wants to merge 2718 commits into
aurora-silicon:integration/aurorafrom
joshuaswarren:w7c/k114c-t3r

Conversation

@joshuaswarren

Copy link
Copy Markdown
Contributor

[Joshua's agent] Preserved series from our October/November K114c work, opened as draft per a patch-id audit: these commits are neither in integration/aurora nor on any PR head. The branch is based on integration/aurora as of 2026-11-10 and is well behind the tip; it needs a rebase before review, and CI should be expected to fail at this base. Draft because it is not rebuilt or booted on the current base.

Series (branch w7c/k114c-t3r, the tip of the t2k2 -> t2k3 -> t3r stack, 27 commits):

  • Board DTS: the K114c board (SMC, keyboard, fans), USB 2.0 host on the three Type-C ports, the N1 Wi-Fi/Bluetooth on apcie0 port 0, the AP pinctrl enabled, the NUB GPIO block with the kiwi touch-screen interface, and a K114c kernel config fragment.
  • Input: dockchannel-hid GPIO handling (announce and request every interface GPIO with the k114c kiwi map, match commands against the announced ids, keep the pm2 afe-reset usable by firmware commands, drive an optional clock-enable line, own reset and power lines for power method 2), clamps for oversized dockchannel Report Size items, magicmouse not marked direct on the MTP trackpad and its REPORT_SIZE clamped to the hid core maximum.
  • wifi: apple: centauri waits for the firmware file before the ROM boot.
  • gpio: macsmc uses the J714c pcIO table on the K114c.
  • PCI: apple: t6050 says why an adopted host rejects its port nodes, and names the missing tunable window.
  • pmdomain: apple: the PMP report is genpd IRQ-safe.
  • soc: apple: rtkit-crashlog bounds (section header and size, mailbox section minimum, decode output) and Ccst/Csta/Crtk/Ccdp/Cpct decoding.

Two commits in the series are patch-id-identical to commits already on open PRs #1568 and #2011; they are kept so the series applies exactly as it ran. Related work that landed separately in evolved form, and is not duplicated here: the pinctrl GPIO-mode width fix (#2004), the speakers/jack/AOP DTS (#1604), and the PCI tunable naming on main (#1606).

Unvoiced; checked by hand.

erykwieliczko and others added 30 commits October 21, 2026 23:26
… 1 MiB

An HDMI 2.1 (FRL) sink's TimingElements arrive from the external
processor as one chunked property transfer of more than 1 MiB (1.38 MB
for a 4K120 TV). The 14.7 client refused any property above 1 MiB, so
such a sink was described with no timings at all.

Let an external processor keep up to 8 MiB per raw property and 24 MiB
in all, with the chunked transfer's announced count bounded the same
way; the panel's properties keep their limits.

Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja>
[port to integration/aurora: the PR's DCP_PROPERTY_RAW_LIMIT/TOTAL for
 the external role, applied to the DCPLink client's raw property store]
Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
(cherry picked from commit 7a99c02)
(cherry picked from commit f37b11c20d12de9b2e4dbaa27d71256682e7f97c)
…bits

sio_send_dt_params() sends each boot parameter with its key split into an
endpoint byte (key >> 8) and a parameter byte (key & 0xff). A key wider
than 16 bits therefore reaches the firmware as its low 16 bits.

The T6030 boot-table check compared whole 32-bit keys. A list could
carry a checked table, such as the map ranges under 0x1a, and then
0x1001a with another page. Both keys were unique and the table check
passed, but the firmware got the second page for the map ranges,
unchecked.

Refuse any key above 0xffff before the duplicate and table checks. Up to
16 bits, every key is its own endpoint and parameter, so the existing
duplicate check covers what the firmware receives. The generic SIO path
is unchanged.

The KUnit test appends, after valid tables, keys that send the same setup
message as a checked table or size key: 0x1001a, 0xff00001a, 0x1000022,
0x10023, 0x1001c and 0x1001d. It checks that each sends that message,
and that each is refused. A wide key aliasing nothing in the list is
refused too, and the 16-bit key 0xffff is still admitted.

Fixes: 215e89d6906c ("dmaengine: apple-sio: Start the T6030 SIO only with protocol-10 tables")
Generated-by: LLM
(cherry picked from commit 621bce2f8d22dc916e7228728e2a9325e125d74e)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
…solved

The T6030 boot-table check reads the IOVA windows from the
iommu-addresses entries of the SIO's memory regions. A short entry
refuses the tables, but an entry whose owner phandle did not resolve
only stopped reading that region. The windows before it still counted,
so tables inside them were admitted with the rest of the entry ignored.

Refuse such an entry, as a short one is. A phandle of 0, or one that no
node has, is malformed either way.

The device-tree KUnit test adds two regions. Each has a good window
holding the tables, followed by an entry owned by phandle 0 in one and by
an unknown phandle in the other. Both are refused.

Fixes: 215e89d6906c ("dmaengine: apple-sio: Start the T6030 SIO only with protocol-10 tables")
Generated-by: LLM
(cherry picked from commit efc2b24864514d3eac2ad61e46f8e8bdb610acd9)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
apple-sio is built on the virt-dma helpers (vchan_init(),
vchan_tx_submit(), vchan_find_desc() and the rest) but never selected
DMA_VIRTUAL_CHANNELS. It only built when another driver selected it at
a compatible level, as APPLE_ADMAC does in the asahi config, where both
are modules.

Any other combination fails to build:
- APPLE_SIO=y with APPLE_ADMAC unset: vmlinux fails to link with
  undefined vchan_* symbols.
- APPLE_SIO=y with APPLE_ADMAC=m: DMA_VIRTUAL_CHANNELS resolves to m,
  which a built-in apple-sio cannot link against.
- APPLE_SIO=m with APPLE_ADMAC unset: modpost finds undefined vchan_*
  symbols.

Select DMA_VIRTUAL_CHANNELS, as the other virt-dma drivers do. Resolved
with allnoconfig plus fragments, DMA_VIRTUAL_CHANNELS now follows
APPLE_SIO in each case. A KUnit kernel with APPLE_SIO=y and no
APPLE_ADMAC links again.

Fixes: 5fdca02 ("dmaengine: apple-sio: Add Apple SIO driver")
Generated-by: LLM
(cherry picked from commit 6b3173bea37605ba6d9ccbb1e672c2bedae40366)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
A T6030 SIO node is compatible with "apple,t6030-sio" and "apple,sio".
Without APPLE_SIO_T6030 the driver matched it as the generic SIO. It
then sent whatever boot tables the boot loader had passed, unchecked,
and configured the channels the generic way, which the T6030 firmware
does not expect.

Refuse such a node at the start of the probe when APPLE_SIO_T6030 is
not built: one message and -ENODEV, before the DMA masks, the registers,
runtime PM, RTKit or any firmware message. The M1 and M2 SIO nodes are
not T6030-compatible and probe as before. With APPLE_SIO_T6030 the
probe is unchanged.

KUnit (APPLE_SIO_KUNIT_TEST) calls the probe on an M1 node and on T6030
nodes with a T6030-aware boot loader's tables and with an earlier
loader's. The devices have no register range, so a probe that goes on
stops there with -EINVAL.
- With APPLE_SIO_T6030: the M1 node and the protocol-10 T6030 node go
  on; the earlier tables are refused.
- Without it: the M1 node still goes on. Both T6030 nodes are refused,
  with the device's DMA mask untouched.

Fixes: e7b14d8b199b ("dmaengine: apple-sio: Configure the T6030 channels as its firmware expects")
Generated-by: LLM
(cherry picked from commit ef76b66ab8bf99818b1c4dd6cbdf3be3d2947b83)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
The G15 generation of the asahi GPU driver, added later in this series,
links page tables into the UAT of the GPU's firmware coprocessor, whose
table walker does not snoop the CPU caches. The tables are ordinary RAM
pages that the shared page-table code allocates and writes through the
cacheable linear map; they are not DMA-mapped, so the DMA API has nothing
to sync them with. Each table, and each changed entry, has to be cleaned
to the point of coherency before the firmware can walk it.

Export the existing dcache_clean_poc() to GPL modules for that, instead
of adding a new cache maintenance helper.

Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja>
[port to integration/aurora: applied unchanged from aurora-silicon/linux PR aurora-silicon#358]
Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
(cherry picked from commit 019af52)
The maintained driver's table excludes T6030/T6031 when the native G15
profile is selected. The native platform driver's private table matches
these devices, but exports no module metadata. With DRM_ASAHI=m, udev
therefore has no device alias with which to load asahi.ko on those Macs.

Export a T6030 alias with the normal of_device_table! macro. Keep the
separate platform driver and its two-device match data; this changes module discovery,
not the probe's supplier admission or firmware-start requirements. The
managed supplier ordering must also be satisfied before qualifying this
profile for automatic boot. T6031 keeps explicit loading until it has
an equivalent display-order admission gate.

Generated-by: LLM
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
The SIO coprocessor (ASC at 0x289c00000, mailbox at +0x8000, interrupts
907..910, stream 0 of the SIO DART) is the DMA controller of the
DisplayPort audio interfaces. As on the other SoCs it stays disabled:
the boot loader reserves its firmware, describes its parameters and
enables it. Its firmware sets up the SPI, UART and audio DMA shims
whichever of them its clients use, so SPI_P, UART_P and AUDIO_P join SIO
as parents of the SIO CPU domain, as t8103, t8112, t600x and t602x
describe them.

The J514S built-in HDMI port's audio is DCPEXT0's audio service with
DPA1, the DisplayPort audio interface linked from DCPEXT0's port@0, whose
samples SIO channel 0x66 carries; DPA1 is in its own power domain. This
is the M1 Pro/Max HDMI port's arrangement (dcpext0, dpaudio1, SIO 0x66).
Enable the SIO mailbox, the alias the boot loader looks the SIO up by,
and DPA1's audio controller.

Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja>
[port to integration/aurora: the DPA1 audio controller, DCPEXT0's port@0
and the SIO alias and mailbox enable go into t603x-j514s-j516s-hdmi.dtsi
against integration's dcpext node (the PR's t6030-j514s-dcpext.dtsi is
not carried); the t6030.dtsi and t6030-pmgr.dtsi hunks are unchanged]
Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
(cherry picked from commit 344f2bd)
(cherry picked from commit 645dac18845071cc7a911ea38efb797980ad1985)
The J514S HDMI port's audio is DCPEXT0's DP/HDMI audio service (EP 0x29,
DCPAVAudioInterface) with DPA1, whose samples the SIO coprocessor
carries. On DCP firmware 14.7's external role the service numbers its
methods as 13.5 does, but differs from what the driver does for M1/M2:

- An element's ElementData, the link data every link method takes, is
  32 bytes, not 24; the last eight are opaque and must be kept.
- StopLink (0/10) and CompleteLink (0/11) carry the link data as
  PrepareLink and StartLink do, and every link method answers an IOReturn
  at body+48.
- One link per stream start: the SIO channel is configured, PrepareLink
  sent; the cyclic DMA is issued before StartLink; on stop the DMA is
  terminated and drained before StopLink and CompleteLink. A link that
  was only prepared (a format probe) is completed at hw_free, the link of
  a connection that has gone is dropped unsent. No DPA reset follows
  CompleteLink: it silenced every later stream on this machine.
- The catalog of a sink's audio elements can be large (114 elements on a
  TV): it is queried with a 256 KiB capacity.
- An open DCPAV interface is a client the firmware waits for, up to 10 s,
  before it removes a terminated interface, and the sink's link with it.
  So the card follows the connection but the service is not opened with
  it: the PCM opens it (Open, command 4, status and opened Boolean
  checked) and its close closes it (Close, command 5). A Terminate
  report on an open service disconnects the card and closes the instance
  from a work item; an instance that goes away takes its open with it,
  and its successor is never closed for it.
- The open, close and link calls wait for their replies without
  srv_rwsem and plug_lock, which the endpoint's receive path takes when
  an instance is published or removed: the instance the PCM opened is
  kept under a mutex of its own. When the card is disconnected for the
  display or the sink, an open PCM's link is completed on the still open
  service first; otherwise the firmware keeps it prepared and refuses
  every later PrepareLink.
- The role's CEC interface (DCPDPCECInterface) broadcasts the same
  reports on the endpoint; it is matched and its reports dropped.

The external role starts the audio endpoint on the driver's opt-in
(appledrm.hdmi_audio), as the other DCPs do, for a controller whose
port@0 links an audio device, once per session and after the sink's
first EDID: with the endpoint's interfaces published, every CopyEDID made
the firmware rebuild them and re-plug the sink. The audio service is
connected when the display is powered on (A472 on) and disconnected when
it is powered off and before the sink is released.

Everything is behind DRM_APPLE_DCP_T6030_EXT and the 14.7 external role:
other configurations build these files unchanged, and an M1/M2 DCP's
24-byte link data still goes out as the same 64 bytes.

Tested on the J514S HDMI port: PCM streams through the card, DPMS cycles
and firmware re-plugs with a PCM open.

Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja>
[port to integration/aurora: keyed to integration's native T6030
external pipe, dcp_is_v14_7_external() (DCP_FIRMWARE_V_14_7 with
external_native), in place of the PR's iomfb_is_v14_7_external(), and
gated by a new DRM_APPLE_DCP_T6030_AUDIO option (default y under
DRM_APPLE_AUDIO) in place of DRM_APPLE_DCP_T6030_EXT. The PR's
dcpext-t6030.c lifecycle hunks become: the AV endpoint is started from
dcp_get_modes() once the first EDID is installed
(dcp_external_audio_start(), external_audio_work in dcp.c, disabled
across unbind and system sleep); the audio service is connected after
A472 on and disconnected before A472 off in dcp_v14_external_modeset()
and iomfb_v14_7_poweroff(), and av_service_connect() waits for
iomfb_v14_7_external_powered(); integration's dcp_poweroff() and DPTX
release paths already disconnect the service. The av.c branches drop
the pinned service through the out label, as integration's service
pinning requires. The AV endpoint's RTKit messages were already routed
to dcp->avep by the DCPLink receive path.]
Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
(cherry picked from commit d71cc2c)
[Chris Kearney: ported onto 94d08ac6, whose dcp.c and iomfb.c were
 reorganised since: the work's enable, disable, init and cancel go into
 the native external blocks of the current bind, unbind, probe, remove,
 suspend and resume; the audio start moves from
 apple_connector_edid_install(), which the base no longer has, to where
 dcp_get_modes() stores an EDID that passed its generation and service
 epoch checks]
Generated-by: LLM

(cherry picked from commit a06ab6c478cb82a49c22c88508288ba8f02adcf9)
A native external pipe's bind enables its audio work, and unbind and
system sleep disable it again; but probe leaves it enabled, so the
first bind's enable underflows the work's disable count and warns, on
every boot of a J514S/J516S (with hdmi_audio off as well), and panics
with panic_on_warn.

Disable it at probe, as the pipe's vblank work is. It is first queued
from the connector, after bind.

Fixes: a06ab6c478cb ("drm: apple: Add the DCP 14.7 external role's HDMI audio")
(cherry picked from commit 3f09f12df6f6c13a6899a7111c0da94f95c1db3c)
[Chris Kearney: Fixes retargeted from c1d99818, the SiliconForest
 variant, to this tree's port of d71cc2c; the context line differs]
Generated-by: LLM
(cherry picked from commit 735cd9c06a754282f23f0f2dc9ab7820e2c1fc06)
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
…et fails early

An external mode set stops the sink's audio before it powers the
display down, and connects it again once the display is on. When the
description never settles, or the mode is gone from it, the mode set
fails before anything is powered down: the display goes on in its old
mode, but its audio stays disconnected. A retry that succeeds connects
it again; once the retries are spent, nothing does, and with no plug
event userspace does not reopen the PCM.

On such a failure, connect the audio again while the display is still
powered and the session running. The audio calls go through two small
helpers, which KUnit stubs to run the production mode set.

Fixes: a06ab6c478cb ("drm: apple: Add the DCP 14.7 external role's HDMI audio")
(cherry picked from commit 55dd126d7f31082a4450022f319845ef559b2fae)
[Chris Kearney: Fixes retargeted from c1d99818, the SiliconForest
 variant, to this tree's port of d71cc2c; the test include block is new
 here, since this iomfb_v14_7.c had none, and kunit/static_stub.h is
 included for the two helpers]
Generated-by: LLM
(cherry picked from commit 9df037b2adc8c687371b686990b271a3af8131f6)
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
…AV service

Two wire facts of the DCP 14.7 external role (T6030 DCPEXT), observed
with its AV service and DPTX port on the J514S:

- Its callback replies carry 8 in byte 35 of the EPIC sub-header. The
  AFK layer zeroed that field for every firmware.

- After a replug it announces its AV service again on the channel the
  old instance had. That is a new instance: the old one is gone with the
  sink it served. The AFK layer refused a second announcement on a
  channel that still had a service, so the host kept the dead instance,
  never opened the new one, and read no EDID through it.

Set the byte on the external role's callback replies, and tear the old
AV instance down before the new announcement is taken, on that role
only.

Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja>
[port to integration/aurora: gated on dcp_is_v14_7_external(), the
 native DCPLink external pipe; integration's handling of the 14.7
 kind-0 replies is kept]
Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
(cherry picked from commit 5033ab8)
(cherry picked from commit 7c5cef7c5f17de4d6f8606616abc58f72e5daa8c)
The DCP 14.7 external role answers EDID requests through an AV service
instance its client has opened. The host opened it with command 6, which
is SetVirtualEDIDMode: with the instance's interfaces present it removes
and recreates them, a software unplug of the sink, and it opens nothing.
DCPAV Open is command 4 (a zero 32-byte body; the reply's status at +0
and its opened Boolean at +16 say whether it took).

An open client is one the firmware waits for, up to 10 s, before it
removes the instance: it holds the reply to the sink's HPD low until the
instance's device has closed ("timeout while waiting for DCPDPDevice
close"), and with it the sink's release and the next sink. The host
never closed its client, so an unplug could leave the controller stuck.

Open the instance with DCPAV Open and check its reply; close the client
(DCPAV Close, command 5) before every HPD low and before a release, and
from a work item when the firmware terminates the instance (report group
2, command 1), after which the instance gets no new use. Close goes only
to the instance that was opened; a removed instance took its open with
it. The EDID read through this role is checked block by block (header
and checksums) before it is published.

Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja>
[port to integration/aurora: dcp_ext_av_work/dcp_ext_av_close and
 dcp_ext_v14_7_av_report re-implemented in dpavservep.c on the
 dcpavserv.opened bookkeeping, the Close hooked into dptxport_set_hpd()
 and dptxport_release_display(); drm_edid_valid() replaces the PR's
 dcpext_edid_object() checks]
Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
(cherry picked from commit 375161e)
[Chris Kearney: ported onto 94d08ac6. The Open replaces the base's
 afk_service_call(service, 4, 6, ...) in dcpavserv_copy_edid(); it is not
 sent in addition to it. The base's enabled-only service pin, the EDID
 generation and service epoch checks and the wire_v2 inline read are kept.
 The opened/terminated resets join the base's init, teardown and detach
 bookkeeping. The Close work is set up once at DCP probe rather than in
 dpavservep_init(), since dpavservep_detach() can run before the endpoint
 was ever started. The dptxep.c hooks go into the base's release and HPD
 functions]
Generated-by: LLM
(cherry picked from commit bcccb5b7c2ce8bb1b2b25fd416c52b1b113a738e)
Comments and two log messages around the 14.7 external role's audio
link, EDID copy and display status read used names that appear nowhere
in the code. Describe them by the commands the code issues instead:
the link's prepare, start, stop and complete (cmds.prepare, start_link,
stop_link, unprepare), dcpavserv_copy_edid(), and A437. No functional
change beyond the two log messages' wording.
(cherry picked from commit 732f8107391c40365a21189f36ba6c225007ef3f)
[Chris Kearney: the audio.c, av.c and dcp.c wording only; the A437
 comment it also reworded comes with the external display status wait
 (45311f2), which is not in this tree yet]
Generated-by: LLM

(cherry picked from commit d0f41527ac4173ef6439bce4ece2dda66ac237a5)
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
…t SIO

Add two KUnit suites for the T6030 external audio path.

The first, apple-dcp-external-audio-work, covers the native external
pipe's audio work, now set up at probe by dcp_external_audio_init(). The
test swaps the work function for a counter, so a start that queued the
work shows. It checks that:
- a first EDID before bind queues nothing;
- bind's single enable is balanced, with no underflow warning;
- suspend, resume, unbind and a second bind keep it so;
- only a native external pipe starts it, and only once per session.

The second, apple-dcpaud-no-sio, covers a DP audio interface whose DMA
controller never registers. This is the T6030 SIO that refuses an earlier
boot loader's tables. The interface gets no TX channel: every PCM open
fails with -ENXIO, and no channel is kept. The display side is not
involved.

Generated-by: LLM
(cherry picked from commit a7a32a6d537850fad3b437b960705650cb1617ed)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
The 14.7 external role's AV client was tracked by service pointer alone,
and its Open was not ordered against a Close or a Terminate. That left
four ways to get the client wrong.

- A Close before an HPD low, or a Terminate, that came while an Open was
  in flight found nothing recorded open. The Open's reply then recorded
  the client open: nothing closed it, and the firmware waits for that
  client before it releases the sink.
- The Close work that a Terminate queued closed whichever instance was
  open when it ran, including a successor published in the same slot
  meanwhile.
- A Close cleared the Terminate mark, so the terminated instance was
  opened and read again before its removal.
- A Terminate that came while no client was recorded open was dropped.

Name each instance by the epoch it was published in. The base already
changes dcpavserv.epoch even when a slot is reused. Then:
- Every Close and Terminate bumps a close generation. An Open that sees
  the generation change is not recorded. If its instance is still the
  published one, its opener closes the client again; an instance gone
  meanwhile took the open with it.
- A Terminate marks its instance's epoch, and that instance gets no new
  use until a successor is published. The Terminate also queues the
  Close for that epoch, and the work closes that instance only.
- An Open in flight keeps a second EDID copy from opening the same
  instance again.

The KUnit suite runs the production open, close, Terminate, Close work and
EDID copy on a fake instance, with only the client calls and the EDID read
stubbed. The Close work runs in test context, and the Open stub acts while
the Open is in flight. It covers:
- Close while Open is in flight, and Terminate while Open is in flight;
- the instance replaced in its slot while Open is in flight;
- an old Close work running after its successor was opened;
- a Terminate before and after its Close, and detach;
- the serial cases, as controls.

Fixes: bcccb5b7c2ce ("drm/apple: dpavservep: open and close the 14.7 external role's AV client")
Generated-by: LLM
(cherry picked from commit 5bf45496cb13766a7ed620e9a6e631c25e200d9d)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
The audio service slots are reusable, but the 14.7 external role's audio
client was tracked by a raw, unpinned service pointer. Once an instance
was removed, its slot could be published again for a successor. Then:
- the successor's Open was skipped, as the stale pointer matched;
- a Close for the old client went to the successor;
- the Terminate work resolved the published service when it ran, so an
  old Terminate disconnected a successor.

Keep the open instance pinned, with an instance generation that changes
whenever an instance is published or removed. The pin keeps its slot from
being reused while it is open.
- An Open on a later generation drops the old client, which went with
  its instance, and opens the new one.
- A Close goes only to the instance still published under the generation
  it was opened in.
- A Terminate records its instance's generation, and its work disconnects
  only that instance.
- The comment that called the slot not reusable is corrected.

The KUnit suite publishes, removes and republishes one reusable slot
through the production init and teardown. It runs the production open,
close, Terminate report and Terminate work, with only the service calls
stubbed and the works run in test context. It covers:
- a successor in the slot being opened, not skipped;
- an old Close not reaching the successor;
- an old Terminate work running after the successor was opened;
- a Terminate from an instance that is not the published one;
- the serial case, as a control.

Fixes: a06ab6c478cb ("drm: apple: Add the DCP 14.7 external role's HDMI audio")
Generated-by: LLM
(cherry picked from commit a7b0d422dce22146ec881663fb47ca006b78c70f)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
The Terminate work checked its instance generation when it started, but
its disconnect then released the PCM's link and closed whichever client
was open when it got there. Within the link release, which runs outside
plug_lock with no open_lock held, the old PCM can close. A successor can
then be published in the same slot and opened. The Terminate's final
Close then closed the successor.

Carry the generation into each step of the disconnect, and check it
where that step is taken, under that step's own lock:
- The card is disconnected only while that instance is still published,
  checked under plug_lock with the generation read under srv_rwsem.
- The link is released only if it was prepared on that instance and that
  instance is still the open one. dcpaud_release_link() asks
  dcp_audiosrv_link_on() under its link_lock, which keeps the existing
  link_lock then open_lock order.
- The final Close checks the generation after taking open_lock.

A link's stop and complete also go only to the instance it was prepared
on: once that instance has gone, the link went with it.

No lock is held across the release or any service call, and the receive
path's locks are not held over a call.

KUnit:
- The old PCM closes within the Terminate's link release, and a
  successor is published in the same slot and opened. The Terminate's
  Close leaves the successor open.
- A Terminate that finds a successor already published leaves the
  successor and the card alone.
- Link stop and complete are refused after the instance changes, and a
  link prepared on the successor is ended normally.
- The production dcpaud_release_link() leaves a link on another instance
  in place, and ends the link of the instance asked for.

Fixes: a7b0d422dce2 ("drm/apple: av: keep the 14.7 audio client on its own instance")
Generated-by: LLM
(cherry picked from commit 387d9ea3bf336e55d7139e1f5616fa090a8b19bd)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
When native G15 explicitly enables APPLE_T6030_GPU_PMP_LINK, use the existing validated halted-PMP startup path for the T6030 report. The actual overlay leaves the PMP, DART and mailbox disabled, and DCP waits on an acknowledged report entry; passive READY polling alone cannot bootstrap that graph.

Route the scoped G15 supplier check through the same entry waiter. Keep the existing power-floor, halted-CPU, UUID and PTD-layout checks, bounded worker lifecycle, deferred external acknowledgements and storage no-ack policy. Other configurations keep the maintained descriptor behavior. No new direct hardware writes are added.

Generated-by: LLM
(cherry picked from commit 0a5be20bea11056a4953d975eeb468004f5b0b76)
Signed-off-by: Chris K <iconidentify@users.noreply.github.com>
The checked translator is an ordinary exported symbol, so symbol_get()
rejects it. DRM_APPLE already depends on OF_ADDRESS and can directly
import the helper. Always pass the remaining tuple length to it instead
of falling back to the unbounded translator after the failed lookup.

Based-on: 388be40927b9c78bef4f9f22eb0a44584da2b163
[Chris: keep the reviewed production fix; omit the OF test that required
additional configuration and positive-control fixes.]
Generated-by: LLM
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
Before 5bf45496, dcpavserv_v14_close() cleared the Terminate mark. After
the Close that a Terminate queued, the next EDID copy then opened and
read the terminated instance again, before its removal. 5bf45496 keeps
the mark until a successor is published, and the Terminate's Close work
no longer goes through dcpavserv_v14_close(). So no test had a Close
reach the instance after its Terminate, and reintroducing the clearing
went unnoticed.

Add a regression test. A Terminate is followed by a Close, as before an
HPD low, in both orders: after the Terminate's queued Close has run, and
before it. In each order:
- the instance gets one Open and one Close;
- the mark stays on its epoch;
- no EDID copy opens or reads it again;
- its successor, a new epoch in the same slot, is opened.

Test only; no production change.

Generated-by: LLM
(cherry picked from commit 66d0fbf8094c1c1017081ce8bc5e7646a332a055)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
Derive DISP, deferred external and storage request metadata from the shipped overlay, then execute the real report probe, startup descriptor/layout validation, child probes and request seeding, managed worker, entry waiter and supplier dispatch. Hardware, work scheduling and OF storage are explicit host boundaries; READY is produced by the simulated firmware bind, never by readiness admission.

Cover missing handoff/layout, missing requests, wrong initial status, failed safety checks/enabling/bind/readiness/acknowledgement, deferred external ACKs, storage no-ack and drained startup ownership. Original 51a9 is a negative control: the real DISP entry waiter fails. Add T6030 worker KUnit cases for deferred external admission; those source cases still need a kernel run.

Host controls: 18 graph cases and 12 compiling runtime mutants; separate config-off/on selection controls and 21 runtime mutants.

Generated-by: LLM

(cherry picked from commit 4bddce608349022a02d952d4d010af8d498c23af)
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
PMP DISP acknowledgement establishes power readiness, not successful native DRM setup. Publish the T6030 component master only after DRM registration and client setup, and withdraw it before unplug/component teardown. Keep a device reference while published, refuse replacement by another master, and serialize admission links against withdrawal with a leaf mutex.

The G15 ADT/PMP admission path gets this optional GPL helper before retaining a PMP supplier or voting for GPU power. An absent module or unpublished display returns EPROBE_DEFER without a hard build dependency. This establishes initial startup order; it does not make retained G15 firmware safe across forced display supplier unbind. T6031 is intentionally unchanged.

Validation: 15 actual-body host controls, including concurrent link/withdraw, and 7 compiling runtime mutants. Publication and withdrawal source ordering checked separately; automatic deferred reprobe and rendering still require hardware qualification.

Generated-by: LLM
(cherry picked from commit 5dd4c76b7e66b496ac38af801344d334814e1caf)
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
A display publisher can run while appledrm is MODULE_STATE_COMING. symbol_get then fails even after successful display registration, and the driver core has no mandatory deferred-probe trigger after the module becomes LIVE. On an otherwise idle system the GPU may remain deferred indefinitely.

Use a direct GPL import and constrain G15 to built-in display support or matching modular GPU/display support. The module loader waits for a COMING dependency to become LIVE before resolving the GPU import, or refuses a failed provider; the existing helper still defers until actual component readiness. No notifier, polling worker or global bus rescan is added. G15-off normal profiles remain unconstrained.

Validation: actual Kconfig conf passes all 18 consumer/provider/request combinations, accepting G15 only for y/y, m/y and m/m. Updated actual-body admission controls pass 14 cases and all 7 compiling runtime mutants. Final modpost, module dependency metadata and hardware startup remain separate gates.

Generated-by: LLM
(cherry picked from commit 8c40970dc1ca5afe43ce98dd1c96d0481e3b0ecb)
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
The production prepare helper now waits for a registered native display. Extend the existing extracted-body fixture with that provider boundary, and require display refusal to leave the PMP link and power votes untouched. Successful display admission must precede each PMP-link attempt.

Generated-by: LLM
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
Register the overlay-derived PMP graph, display publication and native module-linkage checks with the existing hermetic host runner. Run the two source mutation suites, build only the host Kconfig tool for the linkage matrix, and identify the C harnesses as helpers.

Generated-by: LLM
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
Preserve the upstream 13.x panel-property fix and CODEOWNERS update alongside the reviewed 7.2.11 integration. The only production delta is the upstream firmware-version condition in iomfb_template.c.

Upstream: omacom/linux-aurora@70d5dce

Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>

Generated-by: LLM
On the M3 Pro (T6030), a second DP tunnel through a USB4 dock can be
refused a display pipeline for a while. DPIN0 takes the Type-C-only
external pipeline. The hybrid one keeps its HDMI presence settling for
10 s after its HPD-low sample at bind, and dcp_fabric_available() admits
no Type-C stream to it until that ends. The DPIN1 request in that window
gets -EBUSY.

a29fbe8 ("drm/apple: select clockable T6030 Thunderbolt display
routes") gave apple_dpin_m3 capacity_retry and defer_new_bringup, so a
refused T6030 tunnel waits: the DP IN work arms its retry, and
apple_nhi_dp_tunnel_awaits_display() keeps the connection manager from
timing out DPRX and tearing the tunnel down meanwhile. Merge 44d5afc
("Merge pull request aurora-silicon#139 from aurora-silicon/unified/7.2.10") resolved
apple-dpin-state.c, apple-dpin-test.c and two apple.c comments to the
side without that change. Since then a T6030 refusal is final ("dpin1: DP
tunnel setup failed: -16"), and the CM later tears the idle tunnel down,
so the second display behind the dock stays dark.

Restore both flags on apple_dpin_m3, the T6030 ACIO policy. The policies
of the other SoCs are unchanged. Restore a29fbe8's tests along with them:
- the M3 row of apple_dpin_all_policy_fields, plus an assertion that the
  table covers every policy;
- the capacity trace, now run for M3 as well;
- M3 DPRX retention while waiting for a pipeline;
- M3 left out of the policies that a handed sleep leaves unaffected;
- apple_dpin_m3_capacity_suspend.

Restore its two apple.c comments, which name capacity-retry hosts rather
than M1-family ones.

Signed-off-by: Chris Kearney <chris@kearneymail.com>
Generated-by: LLM
(cherry picked from commit 5de33d00761e568788993a7f928716216f9804e8)
Format the five host test sources with clang-format 22 and the kernel
.clang-format, which already leaves includes, comments and string
literals as they are:
- tools/asahi/t6030-clock/fixture.c
- tools/asahi/t6030-wiring/fixture.c
- tools/asahi/tunnel-route/fixture.c
- tools/asahi/t6030-pmp-supplier-tests.c
- tools/testing/selftests/drivers/gpu/asahi-g15/uat-owner.c

Only whitespace changes. Once backslash-newlines are spliced, each file
keeps its clang raw token stream (token kinds and spellings) and its
non-whitespace text. The placeholders the harnesses substitute are
intact.

checkpatch --no-tree -f errors in these files go from 471 to 1. The
remaining one is a __free() declarator inside a macro, which clang-format
spaces as "* node".

The host controls give the same results as before:
- t6030-clock-tests.py, t6030-wiring-tests.py and tunnel-route-tests.py;
- t6030-pmp-supplier-tests.py, the control and its 23 mutation variants;
- asahi-g15 run.py, 61 passed, with the UAT owner check passing.

No driver, header, configuration or Rust source changes.

Signed-off-by: Chris Kearney <chris@kearneymail.com>
Generated-by: LLM
The commit job on PR177 aborts while decoding a patch containing legacy source bytes. Pass the original patch bytes to checkpatch and decode only its diagnostic output, replacing undecodable display characters. Keep commit metadata parsing and sign-off enforcement unchanged.

A real git/Perl regression proves the exact patch bytes reach checkpatch and its warning is retained. The old implementation raises UnicodeDecodeError; all eight commit-checker tests pass with the fix.

Signed-off-by: Chris Kearney <chris@kearneymail.com>

Generated-by: LLM
edimoldovan and others added 29 commits November 8, 2026 14:43
macOS streams the J714c's IMX958 in firmware configuration 6: two
exposures, 3040x5760, merged by the ISP to 3040x2880. Use it for the
1920x1080 and 2560x1440 presets, with the same centre crops; it looks
better than configuration 7. 1280x720 stays on configuration 7.

Signed-off-by: Eduard Moldovan <edimoldovan@gmail.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With SCHEDSTATS off, struct sched_statistics is empty but
____cacheline_aligned. GCC aligns task_struct.stats to 64 bytes; bindgen
emits the empty struct without the alignment, so every Rust task_struct
field after it (mm, pid, tgid, comm) sits 40 bytes early. A Rust driver
reading current->mm then reads an empty list_head instead.

Enabling SCHEDSTATS makes the struct non-empty and the C and Rust
layouts agree. The statistics stay off at run time by default.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add the MacBook Pro 14-inch, M5 Max, OLED (target-type K114c) board
file: the root node with the loader-filled simple framebuffer, the
keyboard-backlight LED, the MTP keyboard/trackpad stack with the SMC
afe-reset line and the K114c touch firmware name, and the SMC hwmon
fans with the T6050 F0md/F1md mode keys. NVMe and the SMC come from
t6050-nvme.dtsi/t6050-smc.dtsi through t6050.dtsi.

Compared with Edi Moldovan's t6050-j714c.dts this leaves out, with
reasons: the DCP display pipeline (the K114c OLED panel and DCP
firmware segment table are unmeasured; the console runs on the
loader's simple framebuffer), the GPU, the PMP (this machine's ADT
has no pmp node), the camera/ISP and the audio stack - none of them
needed to boot.

Values from the K114c ADT ioreg gather of 2026-11-10.

Based on work by Edi Moldovan (edimoldovan).
Include t6050-k114c-usb2.dtsi, the K114c copy of Edi Moldovan's
J714c file: the spmi Type-C controllers on nub-spmi-a0/a1, the
apple,t6050-usb2-phy instances and the generic-xhci hosts with their
dart-usbN IOMMUs, USB 2.0 host mode on all three USB-C ports (port 1
is the proxy cable, as on the J714c).

Every window and interrupt is verified against the K114c ADT ioreg
gather of 2026-11-10 and is identical to the J714c's: usb-drd0
0x2a2280000 with xHCI AIC 1419, dart-usb0 0x2a2f00000/0x2a2f80000 at
AIC 1423, usb-drd1 0x2aa280000 AIC 1451, dart-usb1 0x2aaf00000/
0x2aaf80000 AIC 1455, usb-drd2 0x382280000 AIC 1483, dart-usb2
0x382f00000/0x382f80000 AIC 1487; atc-phyN reg 0 (usb2phy), usb-drdN
reg 3 (pipe handler), and all the SuperSpeed port-1 windows the J714c
file names. The four ADT tunable groups carried under their own names
(tunable_ATC0AXI2AF, tunable_ATC0AXI2AF_LIOA, tunable_ATC_COMMON_CFG,
tunable_ATC_FABRIC) match Edi's values entry for entry on this
machine, so the window-rebased tunables are copied unchanged; the one
value difference is the sn201202x unlock key, the target type with its
first four characters reversed: K114c -> 411K.

Based on work by Edi Moldovan (edimoldovan).
Override the t6050-die0.dtsi pcie0 (apple,t6050-pcie) with the port 0
register windows, the port interrupt and the N1 endpoint on port00:
the Control (pci106b,1900) with the K114c FTAB firmware name and
platform id, the Wi-Fi (pci106b,1902) and Bluetooth (pci106b,1903)
functions, PERST# on pinctrl_ap pin 80 and the endpoint power line on
SMC GPIO 19; dart_apcie0 (AIC 1524) goes status okay, and kboot fills
the factory MAC/BD addresses through the wifi0/bluetooth0 aliases.

Every value is from the K114c ADT ioreg gather of 2026-11-10: the
register windows map one-to-one onto Edi Moldovan's J714c set
(ADT reg 0, 1, 16, 18, 19, 3, 5, 6, 2 and 4; config identity-mapped,
the rest +0x200000000), with the K114c's own sizes for axi2af
(0x1000000) and phy-window (0x800000); port0 is AIC 1523; pci-bridge0
owns MSI vectors 0-31 at apcie0 vector offset 1824 (msi-ranges and
apple,msi-vector-base already in die0); function-perst = GPIO 80,
function-reset = pKW4 gP13; bus-range 0-8 from the ADT (the J714c's
says 0-4); root board-id 0x04, so platform id = (0x04 << 16) | 0x6050
= 0x00046050. iBoot leaves the port untrained and the N1 off; the
kernel cold-trains it (apple_pcie_t6050_cold_train, entered when
adopt finds the link down and the port names pwren-gpios), and
apple,link-trained-by-bootloader stays because it is the adopt
path's port-validity flag.

t6050-k114c-pcie-tunables.dtsi carries the two tunable properties the
K114c ADT (2026-11-10) differs in from the J714c: cio3pllcore (one
value 0x94 vs 0x95, offsets re-based +0x2200 onto pcieclkgen as the
driver applies them there) and phy-ip-pll (three PLL trim values).
axi2af, pcieclkgen, phy and phy-ip-auspma are entry-for-entry
identical, so the J714c file stays included as the base.

Based on work by Edi Moldovan (edimoldovan).
k114c.config, derived from Edi Moldovan's j714c.config with the same
options, K114c-neutral: LOCALVERSION -k114c-ARCH. It keeps the PCIe
host (CONFIG_PCIE_APPLE), the N1 drivers (WLAN_VENDOR_APPLE,
APPLE_CENTAURI, APPLE_CENTAURI_WIFI, BT_HCICENTAURI), NVMe, USB 2.0
host, the SMC/keyboard/fan stack, iwd's crypto menu and the
regulatory.db EXTRA_FIRMWARE entries, plus the rest of the J714c
bring-up options so the two boards build from one fragment shape.

Use with KCONFIG_ALLCONFIG and allnoconfig, as j714c.config is.

Based on work by Edi Moldovan (edimoldovan).
apple_pcie_t6050_tune_host() returned -EINVAL without a message when a
tunable property named a reg window that the device tree lacks, or when
the property had a cell count that is not a multiple of 4. The probe then
failed with "probe failed with error -22" and nothing else on the console.
Name the property and the window, or the cell count.

Analysis by w86 (M5 boot 3, 2026-11-10). Based on work by Edi Moldovan
(edimoldovan).
The two -EINVAL returns at the top of the adopt_link branch of
apple_pcie_probe() gave no message either. Name the cause.

Based on work by Edi Moldovan (edimoldovan).
pcie-apple deferred the N1 root port forever: its reset-gpios names GPIO 80
on pinctrl_ap, which the SoC dtsi leaves disabled until a board needs it.
Enable it as the J714c does.

Based on work by Edi Moldovan (edimoldovan).
The multi-touch interface of the 14-inch M5 Max MacBook Pro (t6050)
declares its raw touch report as a single Report Size item of 0x80000
bits. hid-core rejects any Report Size above 16384 with "invalid
report_size 524288", so the trackpad never registered although the
firmware, the interface and the keyboard all came up.

hid-magicmouse consumes these reports in raw_event() and never uses the
parsed field layout. Clamp each Report Size item in the copied descriptor
to the hid-core limit before parsing, and log the clamp once per item.
The driver is built in, so it probes from the initramfs before the root
file system with the personalized ftab is mounted. The ROM boot then
failed once with -ENOENT and the N1 stayed down until a manual rebind.

Poll for the firmware every 500 ms, up to two minutes, while no DMA has
started, and log the error only when the wait runs out.
The trackpad analog front end reset is the SMC pcIO output 28, which the
driver selects by machine compatible. K114c was not in the list, so the
line stayed an input and dchid_open() got -EPERM from
gpiod_set_value_cansleep() on every open: the multi-touch interface never
started, the trackpad produced no events, and no message was logged.

The K114c ADT function-afe-reset is pKW8("pcIO", 0x10000, 0x11c0000), the
same words as J714c.

Based on work by Edi Moldovan (edimoldovan).
The K114c multi-touch descriptor has a Digitizer-page field under a Mouse
application. hid-input reads that as a pen and sets INPUT_PROP_DIRECT, and
udev then tags the trackpad as a touchscreen instead of a touchpad. The
pad is an indirect pointing device: clear the property.
The K114c MTP firmware (personality K2C1, "AppleMTPFirmwareMac-6320.86.1")
does not send the 0x75 report that magicmouse registers. It sends report
0x5a. hid_get_report() finds no report with that id, so hid_input_report()
returned -1 for every packet and the trackpad never produced an input
event, although the dockchannel interrupt fired on every touch.

Register 0x5a next to 0x75 and parse it on dockchannel devices. 0x75
devices keep the old parser.

Report layout. Offsets are bytes from the report id, little endian:

  0       report id, 0x5a
  1-47    firmware header: counter, timestamp. Not used.
  48      contact section tag, 0x02
  49      section header size, 8
  50      contact record size, 48
  51      section payload size, header size + count * record size
  53      number of contact records
  59      bit 4: physical button
  60      start of the first record
  total   113 + 48 * count bytes, so 113, 161, 209, 257 for 0 to 3 contacts

Contact record, 48 bytes:

  +1      tracking id, rolling
  +2      phase: 1 start, 2-3 approach, 4 touching, 5-7 lifting, 0 gone
  +5      f32 x, in micrometres
  +9      f32 y
  +13     f32 velocity x
  +17     f32 velocity y
  +21     f32 major axis
  +25     f32 minor axis
  +29     three f32 whose meaning is not known
  +41     f32 force, 0 to about 840 at a hard click
  +45     three bytes, not used

A contact counts in phases 3 and 4. The floats are converted by hand
because the kernel avoids the FPU. Positions and axes go out in tens of
micrometres, which fits the J314 ranges the input device already
advertises.

The layout was read from kprobe captures of hid_input_report() on a 14-inch
M5 Max MacBook Pro with one, two and three fingers and clicks. With the
parser loaded, the input device reported 929 frames, X -5896 to 5875, four
clicks and the two-finger tool bit. The button bit and the phase rule are
read from those captures, not from documentation.
…interface; claim kiwi

A power method 2 interface drove only the shared multi-touch AFE reset
line. The K114c touch-screen (kiwi) interface is power method 2 as well,
but its reset and supply are NUB GPIO pins (ADT function-reset and
function-power), so driving the AFE line for it would reset the trackpad.
Read apple,reset-gpios and apple,power-gpios from the interface node; an
interface without apple,reset-gpios keeps the AFE line. The supply is
driven high when the interface is created and again before the reset is
released.

Booted on a K114c with power method 1 (the default), kiwi receives its
firmware and then fails the comm reset command with 0xe00002c2 and never
reports ready. The ADT says PowerMethod 2 for it.

Also add hid-apple-mtp-kiwi. The kiwi HID device matches the wildcard
BUS_HOST Apple entries of hid-apple and hid-magicmouse, so hid-generic
stands off from it, and both drivers reject it in probe: no driver ever
bound. The new driver claims it by name and exposes the raw report pipe on
hidraw. The interface starts when hidraw is first opened.
…interface

The NUB (always-on die) GPIO controller of t6050: ADT nub-gpio0@881E4000,
61 pins, pmgr nub_gpio (always on). It is described without interrupts and
enabled only by the K114c, which uses pins 25 (reset) and 26 (power) for
the kiwi touch-screen interface. The kiwi child gets its firmware name and
power method 2 with those two lines.
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.
gpiolib refuses to set the value of a line that is an input and returns
-EPERM. The reset line of a power method 2 interface with its own line was
requested with GPIOD_ASIS, so the first start of the kiwi interface failed
with EPERM when hidraw was opened. Request it deasserted (low) instead.
…method 2 interface

The K114c kiwi touch screen needs its 24 MHz reference clock. The ADT
kiwi node lists function-clock_24MHz_enable on NUB GPIO pin 22. Without
it the chip never asserts HINT_L and the boot fails with "No HINT_L".

Take an optional apple,clock-gpios line. Request it high at probe and
drive it with the power transitions: high for ON, low for OFF. Other
interfaces have no such property and are unchanged.

The kiwi node gains apple,clock-gpios on NUB pin 22 (active high).
The clock_source pmgr gate (ADT CRST) is still not driven.
ps_gfx is an IRQ-safe pmgr domain. The report genpd was not, so the parent
link between them failed with -EINVAL and the GPU vote never reached the
power-on path. Mark it IRQ-safe so the link holds. The vote waits at most
50 ms.
…e maximum

The K114c MTP input descriptor is a raw buffer pipe: usage TouchPad with
REPORT_SIZE 0x80000 bits (64 KiB), one field, no report id; macOS reads
variable-length reports against it. hid core refuses any REPORT_SIZE
above 16384 bits, aborts the whole parse (the invalid report_size 524288
probe failure), and the device ends up with zero registered reports -
which is why finger reports raise the dockchannel interrupt yet nothing
reaches hidraw or the input device.

Walk the descriptor in report_fixup and rewrite any oversized
REPORT_SIZE item in place to 16384 bits, keeping the item length stable.
Real reports are far shorter, and hid core accepts reports shorter than
the declared size. With the parse completing, the mouse usage drop and
the C1FE-family report decode take over.

Checked against the real K114c ADT descriptor bytes: exactly one item is
clamped, the rest of the byte stream is unchanged.
…wi GPIO map

The k114c kiwi boot fails with No HINT_L: the chip never gets its
24 MHz reference clock. The ADT kiwi node carries four OIPG functions -
reset (gpio 25), power (gpio 26), clock_24MHz_enable (gpio 22) and
display_sync (gpio 11) - and the MTP announces them as GPIO requests
during interface init, but the driver stored a single GPIO per
interface and rejected the second announce outright, so the clock
enable line was never claimed.

Track up to four announced GPIOs per interface and request each
apple,<name>-gpios at start. The legacy iface->gpio keeps pointing at
the first announced line, and the power method 2 multi-touch AFE reset
special case is unchanged (the pm2 transitions drive it, and it stays
claimed from probe). Multi-touch still announces exactly one GPIO, so
its behaviour is untouched.

The kiwi DTS child gains the four GPIO lines on pinctrl_ap with the
ADT pin numbers, active high (polarity from the boot logs will
confirm; the ADT flag bytes 0101/0100 do not decode to a documented
polarity table).

The clock_source pmgr gate (ADT phandle 0x1ed, label CRST) has no
measured pmgr offset in the t6050 device tree yet; the kernelcache
pull decides whether kiwi additionally needs that gate tied as a power
domain. No reboot was performed; boot verification waits for w7X.
The GPIO-command handler still compared cmd->gpio against the removed
singular iface->gpio_id, so the tree did not build after the switch to
the four-entry gpio_ids[] array. Resolve the command against every
announced id and drive that GPIO descriptor: the k114c kiwi firmware
pulses HINT_L (its 24 MHz reference clock enable), which is one of the
four announced lines and not the legacy primary.
…commands

The pm2 afe-reset branch sets iface->gpio but leaves gpios[g] NULL, so a
firmware pulse on the afe-reset id now errors with "bad GPIO" where the
old code pulsed afe_reset. Set gpios[g] too.
The PMP firmware appends five crashlog sections the parser prints as
Unknown crashlog section. Decode the interrupted context (ELR, FAR, ESR
cross-checked against the Crg8 dump), the crashed thread stack, and
hexdump the opaque state blobs. The Csta bounds check covers the full
ESR read.
A section size that is below the 16 byte header, or past the end of the
crashlog, made the parser read the next header or beyond the buffer. Check
both before dispatch and pass the handlers the payload size, which excludes
the section header.
The mailbox handler reads a type at offset 16 and an index at offset 24 and
computes the entry count as (size - 28) / entry size. A section payload
under 28 bytes read past the section and wrapped the count. Warn and
return instead.
The Csta, Ccst, Crtk, Ccdp and Cpct decode printed one console line per
8 bytes with no cap. Sections run to several KB and the boot console is
slow, so an early coprocessor crash could stall the boot inside the dump.
Print four words per line, at most 32 words per section and 32 sections
per crashlog. Keep the last dumped ESR per RTKit instance instead of in a
file-scope static shared by every coprocessor.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

9 participants