Repository navigation
Conversation
jannau
force-pushed
the
bits/000-devicetree
branch
from
August 23, 2026 22:01
128734c to
d7da162
Compare
Author
|
Hi @jannau , I've noticed that |
Member
|
yes, please rebase it. Github webui can't really handle force pushes. For merging it's not an issue as one can easily fetch the pull request and rebase/cherry-pick the commits but it is useless for leaving review comments. |
Ryan1202
force-pushed
the
bits-000-devicetree
branch
from
August 30, 2026 05:13
676a38a to
49b0c17
Compare
Author
|
@jannau I've rebased it |
Ryan1202
force-pushed
the
bits-000-devicetree
branch
from
August 30, 2026 05:22
49b0c17 to
59d106e
Compare
jannau
force-pushed
the
bits/000-devicetree
branch
from
September 8, 2026 18:57
5dd40d5 to
c3da9f3
Compare
jannau
force-pushed
the
bits/000-devicetree
branch
from
November 2, 2026 17:37
4bb4117 to
e2932d4
Compare
Ryan1202
force-pushed
the
bits-000-devicetree
branch
from
November 3, 2026 15:24
59d106e to
f1a4d7d
Compare
Author
|
Hi @jannau and @svenpeter42 , I've rebased onto the latest Would you have a chance to take a look? Thanks! |
jannau
force-pushed
the
bits/000-devicetree
branch
from
November 5, 2026 21:21
e2932d4 to
c46abf5
Compare
Add the SMC mailbox, SMC and SMC GPIO controller nodes for t8132. Tested on an M4 MacBook Pro (J604), where the SMC and GPIO controllers probe successfully. Signed-off-by: Jiajun Wang <ryan1202@foxmail.com>
Add the MTP, mailbox, DART, DockChannel and HID transport nodes for t8132. Unlike t8122, the MTP DART uses stream ID 0 on t8132. Signed-off-by: Jiajun Wang <ryan1202@foxmail.com>
Tested on an M4 MacBook Pro (J604), the built-in keyboard works. Signed-off-by: Jiajun Wang <ryan1202@foxmail.com>
Ryan1202
force-pushed
the
bits-000-devicetree
branch
from
November 7, 2026 11:46
f1a4d7d to
5aa41c9
Compare
3 tasks done
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 12, 2026
A scheduler of one credit left the engine idle while the host built the
next call. The firmware gives each process 32 call slots
(CREATE_PROCESS's pool; a PROCEDURE_CALL takes one before its 0x300)
and asserts on a 33rd without refusing it ("No resource for new
Procedure call", CAneProgramManagerH15.cpp:368; ane_fw_t603x_1483
sub_47484 0x476e4, sub_48580 0x489d8). The kext counts against
RESOURCE_INFO_GET's +0x10 and +0x14, 32 each, under its 64 overall
(AppleH11ANEInterface firmwareResourceIsAvailable 0xfffffe0009791830):
the scheduler's limit is 32, a firmware that reports less is logged,
and the lab's debugfs run, which bypasses the scheduler, waits for an
empty queue.
j514m, macOS 27.2's compiled models through anetest, 2026-11-12
(test kernels AsahiLinux#567 and AsahiLinux#568): big5's bursts of 200 calls at 2.27 ms a
call against 3.1 ms at one in flight (+37 %), conv16 1,913 and res4
1,825 calls a second (1,812-1,821 and 1,618-1,622 before). At a limit
of 64 big5 asserts the firmware as the RE says; conv16, whose calls end
sooner, never reaches 33.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 12, 2026
A call the firmware never answered left it in an assert's loop, and the restart failed: STOP and POWER_DOWN only waited out their timeouts, the fences signalled while the engine still ran, and the put that was to power-cycle it only queued its domains' idle (rpm_suspend_suppliers), so the next resume found ANE_CPU still powered, RUN did not restart it and the handshake read SCRATCH 0. In the kext's order now: a firmware that stopped answering gets no STOP or POWER_DOWN (ane_fw_kill), the engine is powered off before the lost calls' fences signal, each power domain is idled and checked suspended (pm_runtime_barrier, then pm_runtime_idle for one whose queued idle was cancelled), as DisableANEClocksAndPower does before each ANE_Init (AppleH11ANEInterface 0xfffffe00097867a0), and a boot with no handshake gets one more power cycle and boot. j514m, the scheduler's limit raised to 64 to force the firmware's assert with big5's bursts, then conv16 after each: with the domains idled but no barrier (test kernel AsahiLinux#567), pm_runtime_idle returned -EAGAIN on a domain still suspending and each restart needed its second boot; with the barrier (AsahiLinux#568, 2026-11-12), 3 asserts, 3 timeouts, 3 restarts at the first boot, no domain left on, conv16's 200 calls at 0.52 ms each after every restart.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 13, 2026
A scheduler of one credit left the engine idle while the host built the
next call. The firmware gives each process 32 call slots
(CREATE_PROCESS's pool; a PROCEDURE_CALL takes one before its 0x300)
and asserts on a 33rd without refusing it ("No resource for new
Procedure call", CAneProgramManagerH15.cpp:368; ane_fw_t603x_1483
sub_47484 0x476e4, sub_48580 0x489d8). The kext counts against
RESOURCE_INFO_GET's +0x10 and +0x14, 32 each, under its 64 overall
(AppleH11ANEInterface firmwareResourceIsAvailable 0xfffffe0009791830):
the scheduler's limit is 32, a firmware that reports less is logged,
and the lab's debugfs run, which bypasses the scheduler, waits for an
empty queue.
j514m, macOS 27.2's compiled models through anetest, 2026-11-12
(test kernels AsahiLinux#567 and AsahiLinux#568): big5's bursts of 200 calls at 2.27 ms a
call against 3.1 ms at one in flight (+37 %), conv16 1,913 and res4
1,825 calls a second (1,812-1,821 and 1,618-1,622 before). At a limit
of 64 big5 asserts the firmware as the RE says; conv16, whose calls end
sooner, never reaches 33.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 13, 2026
A call the firmware never answered left it in an assert's loop, and the restart failed: STOP and POWER_DOWN only waited out their timeouts, the fences signalled while the engine still ran, and the put that was to power-cycle it only queued its domains' idle (rpm_suspend_suppliers), so the next resume found ANE_CPU still powered, RUN did not restart it and the handshake read SCRATCH 0. In the kext's order now: a firmware that stopped answering gets no STOP or POWER_DOWN (ane_fw_kill), the engine is powered off before the lost calls' fences signal, each power domain is idled and checked suspended (pm_runtime_barrier, then pm_runtime_idle for one whose queued idle was cancelled), as DisableANEClocksAndPower does before each ANE_Init (AppleH11ANEInterface 0xfffffe00097867a0), and a boot with no handshake gets one more power cycle and boot. j514m, the scheduler's limit raised to 64 to force the firmware's assert with big5's bursts, then conv16 after each: with the domains idled but no barrier (test kernel AsahiLinux#567), pm_runtime_idle returned -EAGAIN on a domain still suspending and each restart needed its second boot; with the barrier (AsahiLinux#568, 2026-11-12), 3 asserts, 3 timeouts, 3 restarts at the first boot, no domain left on, conv16's 200 calls at 0.52 ms each after every restart.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 14, 2026
A scheduler of one credit left the engine idle while the host built the
next call. The firmware gives each process 32 call slots
(CREATE_PROCESS's pool; a PROCEDURE_CALL takes one before its 0x300)
and asserts on a 33rd without refusing it ("No resource for new
Procedure call", CAneProgramManagerH15.cpp:368; ane_fw_t603x_1483
sub_47484 0x476e4, sub_48580 0x489d8). The kext counts against
RESOURCE_INFO_GET's +0x10 and +0x14, 32 each, under its 64 overall
(AppleH11ANEInterface firmwareResourceIsAvailable 0xfffffe0009791830):
the scheduler's limit is 32, a firmware that reports less is logged,
and the lab's debugfs run, which bypasses the scheduler, waits for an
empty queue.
j514m, macOS 27.2's compiled models through anetest, 2026-11-12
(test kernels AsahiLinux#567 and AsahiLinux#568): big5's bursts of 200 calls at 2.27 ms a
call against 3.1 ms at one in flight (+37 %), conv16 1,913 and res4
1,825 calls a second (1,812-1,821 and 1,618-1,622 before). At a limit
of 64 big5 asserts the firmware as the RE says; conv16, whose calls end
sooner, never reaches 33.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 14, 2026
A call the firmware never answered left it in an assert's loop, and the restart failed: STOP and POWER_DOWN only waited out their timeouts, the fences signalled while the engine still ran, and the put that was to power-cycle it only queued its domains' idle (rpm_suspend_suppliers), so the next resume found ANE_CPU still powered, RUN did not restart it and the handshake read SCRATCH 0. In the kext's order now: a firmware that stopped answering gets no STOP or POWER_DOWN (ane_fw_kill), the engine is powered off before the lost calls' fences signal, each power domain is idled and checked suspended (pm_runtime_barrier, then pm_runtime_idle for one whose queued idle was cancelled), as DisableANEClocksAndPower does before each ANE_Init (AppleH11ANEInterface 0xfffffe00097867a0), and a boot with no handshake gets one more power cycle and boot. j514m, the scheduler's limit raised to 64 to force the firmware's assert with big5's bursts, then conv16 after each: with the domains idled but no barrier (test kernel AsahiLinux#567), pm_runtime_idle returned -EAGAIN on a domain still suspending and each restart needed its second boot; with the barrier (AsahiLinux#568, 2026-11-12), 3 asserts, 3 timeouts, 3 restarts at the first boot, no domain left on, conv16's 200 calls at 0.52 ms each after every restart.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 19, 2026
A scheduler of one credit left the engine idle while the host built the
next call. The firmware gives each process 32 call slots
(CREATE_PROCESS's pool; a PROCEDURE_CALL takes one before its 0x300)
and asserts on a 33rd without refusing it ("No resource for new
Procedure call", CAneProgramManagerH15.cpp:368; ane_fw_t603x_1483
sub_47484 0x476e4, sub_48580 0x489d8). The kext counts against
RESOURCE_INFO_GET's +0x10 and +0x14, 32 each, under its 64 overall
(AppleH11ANEInterface firmwareResourceIsAvailable 0xfffffe0009791830):
the scheduler's limit is 32, a firmware that reports less is logged,
and the lab's debugfs run, which bypasses the scheduler, waits for an
empty queue.
j514m, macOS 27.2's compiled models through anetest, 2026-11-12
(test kernels AsahiLinux#567 and AsahiLinux#568): big5's bursts of 200 calls at 2.27 ms a
call against 3.1 ms at one in flight (+37 %), conv16 1,913 and res4
1,825 calls a second (1,812-1,821 and 1,618-1,622 before). At a limit
of 64 big5 asserts the firmware as the RE says; conv16, whose calls end
sooner, never reaches 33.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 19, 2026
A call the firmware never answered left it in an assert's loop, and the restart failed: STOP and POWER_DOWN only waited out their timeouts, the fences signalled while the engine still ran, and the put that was to power-cycle it only queued its domains' idle (rpm_suspend_suppliers), so the next resume found ANE_CPU still powered, RUN did not restart it and the handshake read SCRATCH 0. In the kext's order now: a firmware that stopped answering gets no STOP or POWER_DOWN (ane_fw_kill), the engine is powered off before the lost calls' fences signal, each power domain is idled and checked suspended (pm_runtime_barrier, then pm_runtime_idle for one whose queued idle was cancelled), as DisableANEClocksAndPower does before each ANE_Init (AppleH11ANEInterface 0xfffffe00097867a0), and a boot with no handshake gets one more power cycle and boot. j514m, the scheduler's limit raised to 64 to force the firmware's assert with big5's bursts, then conv16 after each: with the domains idled but no barrier (test kernel AsahiLinux#567), pm_runtime_idle returned -EAGAIN on a domain still suspending and each restart needed its second boot; with the barrier (AsahiLinux#568, 2026-11-12), 3 asserts, 3 timeouts, 3 restarts at the first boot, no domain left on, conv16's 200 calls at 0.52 ms each after every restart.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 19, 2026
A scheduler of one credit left the engine idle while the host built the
next call. The firmware gives each process 32 call slots
(CREATE_PROCESS's pool; a PROCEDURE_CALL takes one before its 0x300)
and asserts on a 33rd without refusing it ("No resource for new
Procedure call", CAneProgramManagerH15.cpp:368; ane_fw_t603x_1483
sub_47484 0x476e4, sub_48580 0x489d8). The kext counts against
RESOURCE_INFO_GET's +0x10 and +0x14, 32 each, under its 64 overall
(AppleH11ANEInterface firmwareResourceIsAvailable 0xfffffe0009791830):
the scheduler's limit is 32, a firmware that reports less is logged,
and the lab's debugfs run, which bypasses the scheduler, waits for an
empty queue.
j514m, macOS 27.2's compiled models through anetest, 2026-11-12
(test kernels AsahiLinux#567 and AsahiLinux#568): big5's bursts of 200 calls at 2.27 ms a
call against 3.1 ms at one in flight (+37 %), conv16 1,913 and res4
1,825 calls a second (1,812-1,821 and 1,618-1,622 before). At a limit
of 64 big5 asserts the firmware as the RE says; conv16, whose calls end
sooner, never reaches 33.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 19, 2026
A call the firmware never answered left it in an assert's loop, and the restart failed: STOP and POWER_DOWN only waited out their timeouts, the fences signalled while the engine still ran, and the put that was to power-cycle it only queued its domains' idle (rpm_suspend_suppliers), so the next resume found ANE_CPU still powered, RUN did not restart it and the handshake read SCRATCH 0. In the kext's order now: a firmware that stopped answering gets no STOP or POWER_DOWN (ane_fw_kill), the engine is powered off before the lost calls' fences signal, each power domain is idled and checked suspended (pm_runtime_barrier, then pm_runtime_idle for one whose queued idle was cancelled), as DisableANEClocksAndPower does before each ANE_Init (AppleH11ANEInterface 0xfffffe00097867a0), and a boot with no handshake gets one more power cycle and boot. j514m, the scheduler's limit raised to 64 to force the firmware's assert with big5's bursts, then conv16 after each: with the domains idled but no barrier (test kernel AsahiLinux#567), pm_runtime_idle returned -EAGAIN on a domain still suspending and each restart needed its second boot; with the barrier (AsahiLinux#568, 2026-11-12), 3 asserts, 3 timeouts, 3 restarts at the first boot, no domain left on, conv16's 200 calls at 0.52 ms each after every restart.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 20, 2026
A scheduler of one credit left the engine idle while the host built the
next call. The firmware gives each process 32 call slots
(CREATE_PROCESS's pool; a PROCEDURE_CALL takes one before its 0x300)
and asserts on a 33rd without refusing it ("No resource for new
Procedure call", CAneProgramManagerH15.cpp:368; ane_fw_t603x_1483
sub_47484 0x476e4, sub_48580 0x489d8). The kext counts against
RESOURCE_INFO_GET's +0x10 and +0x14, 32 each, under its 64 overall
(AppleH11ANEInterface firmwareResourceIsAvailable 0xfffffe0009791830):
the scheduler's limit is 32, a firmware that reports less is logged,
and the lab's debugfs run, which bypasses the scheduler, waits for an
empty queue.
j514m, macOS 27.2's compiled models through anetest, 2026-11-12
(test kernels AsahiLinux#567 and AsahiLinux#568): big5's bursts of 200 calls at 2.27 ms a
call against 3.1 ms at one in flight (+37 %), conv16 1,913 and res4
1,825 calls a second (1,812-1,821 and 1,618-1,622 before). At a limit
of 64 big5 asserts the firmware as the RE says; conv16, whose calls end
sooner, never reaches 33.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 20, 2026
A call the firmware never answered left it in an assert's loop, and the restart failed: STOP and POWER_DOWN only waited out their timeouts, the fences signalled while the engine still ran, and the put that was to power-cycle it only queued its domains' idle (rpm_suspend_suppliers), so the next resume found ANE_CPU still powered, RUN did not restart it and the handshake read SCRATCH 0. In the kext's order now: a firmware that stopped answering gets no STOP or POWER_DOWN (ane_fw_kill), the engine is powered off before the lost calls' fences signal, each power domain is idled and checked suspended (pm_runtime_barrier, then pm_runtime_idle for one whose queued idle was cancelled), as DisableANEClocksAndPower does before each ANE_Init (AppleH11ANEInterface 0xfffffe00097867a0), and a boot with no handshake gets one more power cycle and boot. j514m, the scheduler's limit raised to 64 to force the firmware's assert with big5's bursts, then conv16 after each: with the domains idled but no barrier (test kernel AsahiLinux#567), pm_runtime_idle returned -EAGAIN on a domain still suspending and each restart needed its second boot; with the barrier (AsahiLinux#568, 2026-11-12), 3 asserts, 3 timeouts, 3 restarts at the first boot, no domain left on, conv16's 200 calls at 0.52 ms each after every restart.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 20, 2026
A scheduler of one credit left the engine idle while the host built the
next call. The firmware gives each process 32 call slots
(CREATE_PROCESS's pool; a PROCEDURE_CALL takes one before its 0x300)
and asserts on a 33rd without refusing it ("No resource for new
Procedure call", CAneProgramManagerH15.cpp:368; ane_fw_t603x_1483
sub_47484 0x476e4, sub_48580 0x489d8). The kext counts against
RESOURCE_INFO_GET's +0x10 and +0x14, 32 each, under its 64 overall
(AppleH11ANEInterface firmwareResourceIsAvailable 0xfffffe0009791830):
the scheduler's limit is 32, a firmware that reports less is logged,
and the lab's debugfs run, which bypasses the scheduler, waits for an
empty queue.
j514m, macOS 27.2's compiled models through anetest, 2026-11-12
(test kernels AsahiLinux#567 and AsahiLinux#568): big5's bursts of 200 calls at 2.27 ms a
call against 3.1 ms at one in flight (+37 %), conv16 1,913 and res4
1,825 calls a second (1,812-1,821 and 1,618-1,622 before). At a limit
of 64 big5 asserts the firmware as the RE says; conv16, whose calls end
sooner, never reaches 33.
hitori-chan
added a commit
to hitori-chan/linux
that referenced
this pull request
Nov 20, 2026
A call the firmware never answered left it in an assert's loop, and the restart failed: STOP and POWER_DOWN only waited out their timeouts, the fences signalled while the engine still ran, and the put that was to power-cycle it only queued its domains' idle (rpm_suspend_suppliers), so the next resume found ANE_CPU still powered, RUN did not restart it and the handshake read SCRATCH 0. In the kext's order now: a firmware that stopped answering gets no STOP or POWER_DOWN (ane_fw_kill), the engine is powered off before the lost calls' fences signal, each power domain is idled and checked suspended (pm_runtime_barrier, then pm_runtime_idle for one whose queued idle was cancelled), as DisableANEClocksAndPower does before each ANE_Init (AppleH11ANEInterface 0xfffffe00097867a0), and a boot with no handshake gets one more power cycle and boot. j514m, the scheduler's limit raised to 64 to force the firmware's assert with big5's bursts, then conv16 after each: with the domains idled but no barrier (test kernel AsahiLinux#567), pm_runtime_idle returned -EAGAIN on a domain still suspending and each restart needed its second boot; with the barrier (AsahiLinux#568, 2026-11-12), 3 asserts, 3 timeouts, 3 restarts at the first boot, no domain left on, conv16's 200 calls at 0.52 ms each after every restart.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add device tree support required for the built-in keyboard on the M4 MacBook Pro (J604), including SMC, MTP, DockChannel and DART.
I wrote it based on the t8122 device tree. But unlike t8122, t8132 uses stream ID 0 instead of 1. I guess it's related to the
sidfield in/arm-io/dart-mtpin ADT.The device tree changes were only tested as part of the
asahibranch on an M4 MacBook Pro (J604).