Repository navigation
Sync omacom #177 (G15, T6030 dock, platform fixes) onto 7.3 - #395
Conversation
fwnode_typec_switch_get() and fwnode_typec_mux_get() pass a stack array to fwnode_connection_find_matches(). The array collects the devices found so far. 7.2's typec_switch_match() and typec_mux_match() scan every slot of that array to skip a device already matched, but the property walker only fills the slots it has written. So the scan reads uninitialised stack pointers. With CONFIG_INIT_STACK_NONE, as in our production configs, a stale pointer that happens to equal the device makes a real match look like a duplicate and drops it. Zero both arrays. Empty slots are then NULL and never equal a device, and the duplicate skip, its put_device() and the unavailable-ancestor semantics are unchanged. An MSan harness built from the actual callbacks, getters and property walkers reported use-of-uninitialized-value from mux_devs and sw_devs before this change. All 12 lookup cases are clean after it, including duplicate put_device balance and the absent, missing and disabled-ancestor cases. Generated-by: LLM (cherry picked from commit 9d6d38611817dd6944c07eae4093923de13425bc) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit bd4c585)
After a suspend with a headset playing there is no sound after resume. The ASoC resume work unmutes the DAIs that were active at suspend while the card is still below D0. At that point the PCM is suspended and the host may not be driving SCLK yet, but cs42l42_mute_stream() already moves the codec from the internal oscillator onto SCLK. Without a clock the codec stops answering on I2C, and the stream stays dead until the codec is reset. Do not do the switch while the card is not in D0. The unmute that follows the next prepare switches the clock once SCLK runs. A PLL that fails to lock leads to the same state, because the codec switched to the PLL output anyway after printing a warning. Stop the PLL, stay on the oscillator and return the error instead. Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de> (cherry picked from commit e05ef13) (cherry picked from commit dffa7599a7f8a3a46bfcef07f16f44f611aef122) (cherry picked from commit 30f3203)
Based on aurora-silicon/linux PR331 (e05ef13, "ASoC: cs42l42: keep the codec on the oscillator until SCLK runs", Thomas Glanzmann). That commit only changes cs42l42, and the Apple laptops use a CS42L84 for the headphone jack, so it does not affect them. This applies the same change to cs42l84_mute_stream(). ASoC resume sets the card to D2, unmutes the playback DAIs that were active at suspend and only then sets D0. At that point the PCM is suspended and the Apple MCA may not be driving BCLK yet, but the unmute already moves the codec from its internal oscillator onto BCLK or the PLL. Without a clock the codec stops answering on I2C. Do not touch the clocks while the card is not in D0. MCA does not advertise SNDRV_PCM_INFO_RESUME, so userspace prepares the stream again after resume, and the unmute that follows that prepare does the switch at D0. If the PLL does not lock, the codec used to switch to the PLL output anyway after a warning. Instead stop the PLL, clear REFCLK_EN again (the same order as the mute path, which first moves the codec back to the oscillator) and return the error, leaving the codec on the oscillator, the DAC muted and stream_use unchanged so the next unmute retries. The PLL lock error bit stays a warning. It is only checked after the lock bit has been seen set, and nothing in the driver says what it means beyond its name, so making it fatal could break setups that work today. The BCLK path without the PLL is unchanged apart from the D0 check. Generated-by: LLM (cherry picked from commit 3a2865f65fdb396e1f3e011127ede58b753b06cf) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 4b69aec)
Add a KUnit suite for the clock switching in cs42l42_mute_stream(). The tests call the driver's own cs42l42_pll_config() and cs42l42_mute_stream() on a component backed by a regmap whose callbacks log every register write, with the PLL lock status under test control and a sound card whose power state the test sets. - unmute_below_d0_does_nothing (D1, D2, D3hot, D3cold): an unmute while the card is below D0 makes no register access and leaves stream_use clear and the headphone muted. Catches the codec being moved onto SCLK by the ASoC resume work before the stream runs. - resume_unmute_then_d0_unmute: the resume order. The unmute at D2 does nothing and the unmute at D0 does the whole switch. - unmute_pll_locked: at D0 with a PLL clock configuration the PLL is started, the lock bit is polled, then the PLL output is selected, the codec moves to SCLK and the headphone is unmuted, in that order. - unmute_pll_lock_timeout: when the PLL never locks, the unmute returns -ETIMEDOUT, stops the PLL again and does not select the PLL output, switch to SCLK or unmute the headphone. Catches the old behaviour of warning and switching to the unlocked PLL anyway. - unmute_after_pll_lock_timeout: after a failed unmute stream_use is still clear, so the next unmute with a locking PLL does the whole switch. - mute_returns_to_osc: muting the last stream goes back to the oscillator before deselecting and stopping the PLL. - unmute_no_pll: with SCLK used directly as MCLK the codec moves to SCLK without touching or polling the PLL. The PLL cases run with an N = 2 and an N = 1 clock configuration. A full run takes well under a second because the lock timeout is short. Generated-by: LLM (cherry picked from commit 3822f19ae8831ce9544c4518374a1f51993c7ce2) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 072a552)
Add a KUnit suite for the clock switching in cs42l84_mute_stream(). The function and the PLL setup are static, so the test file is included at the end of cs42l84.c when CONFIG_SND_SOC_CS42L84_TEST is enabled. The tests set the clock with snd_soc_dai_set_sysclk() and cs42l84_pll_config(), and mute and unmute through snd_soc_dai_digital_mute() and the driver's own DAI ops. The component is backed by a regmap whose callbacks log every register write, with the PLL lock status under test control and a sound card whose power state the test sets. - unmute_below_d0_does_nothing (D1, D2, D3hot, D3cold): an unmute while the card is below D0 makes no register access and leaves stream_use clear, the DAC muted and CCM_CTL1 on the oscillator. Catches the codec being moved onto BCLK by the ASoC resume work before the stream runs. - resume_unmute_then_d0_unmute: the resume order. The unmute at D2 does nothing and the unmute at D0 does the whole switch. - unmute_pll_locked: at D0 with a PLL BCLK rate, REFCLK_EN and the PLL are enabled, the lock bit is polled, then CCM_CTL1 selects the PLL and the DAC is unmuted, in that order. - unmute_pll_lock_error_bit: the lock error bit next to a set lock bit is only reported, the switch still happens. - unmute_pll_lock_timeout: when the PLL never locks, the unmute returns -ETIMEDOUT, CCM_CTL1 is never written, PLL_CTL1_EN and then REFCLK_EN are cleared again, and the DAC stays muted with stream_use clear. Catches the old behaviour of warning and switching to the unlocked PLL anyway. - unmute_after_pll_lock_timeout: after a failed unmute the next unmute with a locking PLL does the whole switch. - mute_returns_to_rco: muting the last stream mutes the DAC, goes back to the oscillator, then stops the PLL and REFCLK. - unmute_bclk: with BCLK used directly as MCLK the codec moves to BCLK without touching or polling the PLL. The suite runs in well under a second. Generated-by: LLM (cherry picked from commit 7560c8b4747f79cf18d8ac90c11f6bec9f4e2eca) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit d6ec03d)
The optional CODECS trailer occupies the final 64 bytes of the firmware. A successful firmware request does not establish that minimum length, so forming and reading the trailer pointer can read before a truncated image. Check the image length before forming the pointer and only parse an available trailer. Short images keep the existing legacy variant-capability fallback, as do larger images without the CODECS marker. The supported codec intersection and normal firmware handling are unchanged. An actual-block ASan/UBSan harness reproduces invalid reads at firmware lengths 0, 1, 7 and 63 on the parent. The guarded code passes those limits, 64/128-byte trailer controls, legacy fallback and variant intersections. Fixes: 883ff44 ("media: apple: avd: Announce only codecs the firmware supports") Generated-by: LLM (cherry picked from commit 703cabfcbe35d99f8457ef1cc781a051e4a3ad3a) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit a9a9c3a)
Extract the production AVD codec-tail block and capability definitions into an ASan/UBSan host harness. Cover 0/1/7/63-byte firmware boundaries, 64/128-byte tails, all codec bits, the variant-capability intersection, unknown/no-marker trailers, and the legacy capability fallback. Keep a before/after control around runtime mutations that remove the length guard, the variant intersection, or the legacy fallback. Record source hashes, compiler commands and per-case output in an external evidence directory. --ref reproduces the missing-bound defect on the parent without changing the checked-out source. Both GCC 16.2.1 and Clang 22.1.8 pass all 20 cases in both controls, and all three mutants build and fail the intended runtime cases. Parent 10ff0fe fails exactly the four short-image cases on both compilers. These are host parser tests, not firmware loading or hardware tests. Generated-by: LLM (cherry picked from commit 3d480b94dbbb41d7ca140bf93fc461e18df0c3d4) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 51d20a0)
Linux 7.2 replaces the sound-card power_ref atomic and separate power_ref_sleep waitqueue with struct snd_refcount. Initialize the same power_ref object as snd_card_init(), while preserving the independent power_sleep waitqueue. This adapts only the two KUnit fixtures; codec behavior is unchanged. Reference: 2fa8d8b (ALSA power-reference conversion) Generated-by: LLM (cherry picked from commit 1b3f2ecf19448f66359efa1417bc760d129dba4d) (cherry picked from commit c51022a6cc32efa48096ad043d6aedda19f0d1e2) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit e3e8dd4)
A failed DPRX completion removes the tunnel from the connection manager, but the completion worker still owns a reference. Immediate DP pairing then tries to reuse the old output adapter's fixed HopIDs before the worker drops that reference and destroys its paths. Allocation fails and no pairing pass follows the final put. Queue pairing on the ordered domain workqueue so DPRX completion can release its paths first. Keep no port pointer in the queued work. Balance one domain reference per queued execution, including coalescing, and take runtime PM before the domain mutex. Recheck hotplug activity and the root router under that mutex so stopped domains do not recreate tunnels. Exercise the actual retirement, queue, repair and completion-tail bodies with modeled paths, workqueue and runtime PM. Cover reference balance, coalescing, stop before/after queueing and during resume, missing root, PM failure, requeue while running and actual probe initialization. Nine ASan/UBSan cases pass; nine valid runtime mutants are rejected. This does not claim real workqueue concurrency or hardware qualification. Generated-by: LLM (cherry picked from commit f8650f83f4c2d83bbc440d921b3428bd67f08273) Signed-off-by: Chris Kearney <chris@kearneymail.com> (cherry picked from commit 7fcb47d)
A queued pairing pass can observe hotplug_active cleared during freeze
or suspend and exit before it runs. Thaw and resume do not always pair
unused DP resources again. Unlike a true stop, temporary inactivity
must not discard the outstanding request.
Track the request under the domain mutex. Clear it immediately before
the pairing pass, retain it on inactive or failed-runtime-PM exits, and
requeue it from system resume, thaw and runtime resume. True domain stop
discards the request. Resume hooks already own the domain mutex, and the
queued worker still acquires runtime PM before that mutex.
Extend the host fixture with the real freeze/thaw bodies. Cover a pass
that drains while frozen then retries on thaw, and a failed PM resume
that succeeds on a later wake. Eleven ASan/UBSan cases pass; twelve valid
runtime mutants are rejected, including missing pending state, failure
to consume it, and omission of the thaw requeue. This models framework
services and does not claim real suspend or workqueue qualification.
Fixes: f8650f83f4c2 ("thunderbolt: defer DP pairing until DPRX retirement releases paths")
Generated-by: LLM
(cherry picked from commit b224b82dd5dbc62e7d292c9cc34fe732132797bc)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
(cherry picked from commit 3975d42)
A repair queued from resume_noirq or thaw_noirq can run while runtime
PM remains disabled. The domain can still be runtime-suspended, causing
pm_runtime_resume_and_get() to fail with -EACCES. The pending bit survives,
but generic hosts then have no completion-phase replay to service it.
Replay a pending request under the domain mutex from tb_complete(), after
the early-resume phase has enabled runtime PM. On hosts that already
restore DP resources and pair directly in complete, consume the prior
request before that direct pass. A new request raised during the pass
can still be queued afterwards. XDomain rescan keeps its separate scoped
lock and completes before the pairing guard is taken.
Exercise the actual completion body in the existing host fixture, with
modeled PM, queue and paths. Cover noirq PM refusal followed by completion
replay, direct Apple completion acknowledgment, no-request completion,
and XDomain rescan locking. Fifteen ASan/UBSan cases and fifteen runtime
mutants pass. Substituting only the old completion body fails exactly the
three intended completion cases. Critical Thunderbolt W=1 compilation
passes with no C diagnostics. These checks do not qualify real workqueue
concurrency or hardware suspend/resume.
Fixes: b224b82dd5db ("thunderbolt: preserve deferred DP pairing across sleep")
Generated-by: LLM
(cherry picked from commit 3303a5322b90eb0fb191416aa308c96d55c046a0)
Signed-off-by: Chris Kearney <chris@kearneymail.com>
(cherry picked from commit bc1b422)
Adapt the reference-order fixture to the current single-argument DPRX callback and extract the actual tb_dp_tunnel_active implementation. Keep the callback-before-tunnel-put-before-domain-put order explicit, and cover canceled callbacks without pairing or path retirement. Sixteen actual-body host cases pass with ASan/UBSan and all sixteen runtime mutations are caught. Register this harness in the existing host-test job instead of importing a SiliconForest workflow. The queue, PM and path services remain boundary models; these tests do not qualify kernel concurrency or hardware reconnects. Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 81399ec)
…P source Upstream 6a78b76 ("mm/mremap: reset unfaulted VMA page offset for MREMAP_DONTUNMAP") sets vm_pgoff = vm_start >> PAGE_SHIFT on a source VMA that MREMAP_DONTUNMAP left unfaulted. That restores the invariant that an unfaulted anonymous VMA's page offset matches its address. With user4k the page offset has two parts, the page index and the sub-page byte offset. do_mmap(), do_brk_flags(), the mmap_region path and insert_vm_struct() all set both parts from vm_start. Do the same here, via vma_set_page_offset(), so a source starting on a 4 KiB boundary that is not 16 KiB aligned gets the right sub-page offset too. The VMA has no anon_vma and no pages at this point. The offset only affects later merge eligibility and the index of future anonymous folios. Generated-by: LLM (cherry picked from commit 5ec69f61051ba0bad23a5f6fc05e936067976df7) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 0c4d483)
The maintained display fabric already looks up apple_atc_dp_tunnel_rate_source and allows external source 0 or 2 for DP IN0. Its fallback can drive only source 0 when the provider is absent. Supply the existing optional API in ATC and track the owning source, refusing stale rate/stop requests. Adapt the provider and focused controls from SiliconForest 13d2aff. Keep omacom's newer fabric decisions and optional-provider fallback unchanged; do not restore retired host harnesses. DP IN1 remains unsupported here and other generations retain their existing qualified clock paths. Based-on: 13d2aff Source: https://github.com/iconidentify/siliconforest-linux/commit/13d2affbc4d176b528d0c5ef5b17d639c6daab19 Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 13d2aff) [SiliconForest: adapt the missing provider without replacing current fabric policy.] Generated-by: LLM (cherry picked from commit 60b2346)
Carry the previously source-tested dual-stream implementation onto the SiliconForest 7.2.11 kernel. Keep per-stream clock/source ownership and native pipeline admission separate from the ordered M1/M2 fabric policy. Preserve the shared prerequisite clock while either stream remains active, including rate changes and shutdown failures. Based-on: 8753b9ff351eed4efa9686fece6493891b4f18c8 Generated-by: LLM (cherry picked from commit ad5a02b8da72502953d032538ef18217f35ec486) [SiliconForest: retain omacom's newer CRTC narrowing/release behavior and SAFE ownership checks; keep retired harnesses absent. Update the pure fabric clock map alongside the provider and test their parity. Make clock controls standalone and run clock/wiring controls under ASan/UBSan. No forced source, reversed input order or private diagnostics.] Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 873dae4)
… tile A Thunderbolt DPIN1 request goes through dcp_tb_split_locked() first while tiled_split is on, the default. With no DPIN0 tunnel route on the port it returns -ENODEV, to wait for the first half of a tiled display. On T6030 every Type-C pipeline is an external display processor, which never takes a tile: once DPIN0 is there, the same function returns -EOPNOTSUPP. A lone dock display on DPIN1 therefore waits for a DPIN0 that never comes. After the Thunderbolt side's 60 retries (30 s) the tunnel fails with -ETIMEDOUT, logged as "no display pipeline free". Before that wait, return -EOPNOTSUPP when every route of the port is an external processor. dcp->external comes from the DT compatible at probe, before the routes are registered, so it does not depend on which services are up. A port with no routes, or with any pipeline that is not external, keeps the wait, and an existing split is handled first, as before. Only T6030 sets dcp->external, so other SoCs are unchanged. Seen on J516S with the DP IN order of a diagnostic build (DPIN1 first), where a display alone on a TS3 Plus stayed dark. With the production order a lone display normally gets DPIN0, but the same wait is reached whenever DPIN0 is unavailable or attaches late. The host test tools/asahi/tiled-split-tests.py compiles the production dispatch, split and helper bodies with ASan/UBSan. It covers T6030 in both DPIN orders, before services are up, with an external owner, with tiled_split off and for an inactive request; tiled joins, tile-hint waits, not-yet-ready and mixed or empty route lists on other pipelines; and the existing split lifecycle. The parent's bodies fail the four lone-DPIN1 T6030 cases, and each of six mutants of the new gate is caught. CI runs it with the T6030 display tests. Generated-by: LLM (cherry picked from commit 9af410d0034808e37d99a2fea25a2003aafdc94d) Signed-off-by: Chris Kearney <chris@kearneymail.com> (cherry picked from commit 5cefcd6)
Allow a separately opted-in native T6030 GPU to use the PMP admitted by
the maintained display gate, without replacing the normal display or PMP
stack. Keep the capability hidden and off by default; GPU Kconfig can
select APPLE_T6030_GPU_PMP_LINK when the display gate is enabled.
Link only the unique enabled T6030 GPU and retain an existing matching
supplier. Reject malformed or unresolved pre-existing T6030 phandles.
The existing T8122 linking path keeps its original behavior.
When the T6030 supplier has no declared native report, select the unique
available T6030 report with its exact supplier phandle and a unique,
enabled reg-7 display request. Use the existing entry-ready consumer,
which owns startup/removal and checks READY plus its request ACK before
the serialized AGX power command. Propagate every readiness failure.
A declared report always takes the existing native25 path, even when it
is malformed; the T8122 missing-report behavior is unchanged.
Adapt the earlier pairing to current omacom instead of restoring the old
broader current14/T8122 helper or relaxing the PR148 readiness contract.
The new lookup validates property shape, duplicate entries and identity.
Based-on: 9333b0728dc4 ("drm/asahi: pair native AGX with the maintained display and PMP stack")
Based-on: 734cd754d3bd ("soc: apple: admit native T6030 GPU through acknowledged PMP report")
Source: https://github.com/iconidentify/siliconforest-linux/commit/9333b0728dc4fa5cfb0271f30314eed5f97d6b9b
Source: https://github.com/iconidentify/siliconforest-linux/commit/734cd754d3bda17647f7dc2cadabe2e588c9ed7d
Generated-by: LLM
(cherry picked from commit dd0670a078b7525c2efc60f38b020bfd61500b35)
Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com>
(cherry picked from commit 0c4e44f)
The maintained T6030 report does not own a PMP startup object. The managed-start consumer therefore rejects its bound report with ENODEV. Keep managed-start consumers unchanged and validate the existing passive T6030 report instead. Pin its platform device and take its device lock without blocking behind supplier removal while the command mutex is held. Require bound driver data for this report, the exact T6030 offset table and no managed startup state. Poll READY together with the display request acknowledgment at bit 7 before allowing the existing serialized power command. Propagate lock contention, absent binding and readiness failures. The existing display report entry seeds its always-on reg-7 request and polls its acknowledgment; this reader sends no request and starts no firmware. The host controls reproduce the old managed-consumer refusal with no startup object, and exercise the actual new reader with controlled MMIO and locking boundaries. No real teardown-concurrency or hardware qualification is claimed. Generated-by: LLM (cherry picked from commit 650a8bc818e4b5470538feaa4da202e22fe53279) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 5108272)
…diness Extract the real gate linkage, supplier dispatch, passive readiness reader, existing managed-start consumer and serialized power bridge. Verify default-off parity against the prior source, exact supplier and entry roles, READY plus DISP ACK, report binding and reference/lock cleanup. Keep the existing managed-start consumers byte-identical. The real passive report case has no managed startup object. The unchanged managed consumer refuses it as expected, while the new reader admits only a ready and acknowledged report. Host MMIO, device binding, OF storage and lock boundaries are modeled. Both config controls pass; 24 runtime mutations are caught, and the prior config-off control passes. Generated-by: LLM (cherry picked from commit 669556aad1703caaf82b9feeeb820de7d3f3e403) Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit da0d736)
Update the ATC expectations for the supported second DP input and check both pure clock-policy helpers against the same explicit valid and invalid source/input table. Correct the stale comment describing the public helper as absent. Generated-by: LLM (cherry picked from commit d4659833685052360e55c74038528bd9d4cf7430) Signed-off-by: Chris <303316+iconidentify@users.noreply.github.com> (cherry picked from commit c6b25d8)
Adapt the missing T8103-generation USB3 reliability work onto the newer PIPE state and fixed-hub lifecycle. Retry a live lock at most ten times, skip legacy host BIST at dispatch, and admit an early role from clean OFF only with lane tunables. Retain incomplete unlock and reconnect obligations until completion or a committed replacement. Keep generic success-only publication, stopped parking, DP ownership and T8122/J700 behavior. The generation gate also covers the T600x, T8112 and T602x fallback-compatible PHYs. Only J313 hardware qualification is planned; host and emulated tests do not qualify the other boards. Based-on: AsahiLinux/linux#503 by milijan Source: AsahiLinux/linux#503 Upstream-commit: 87a2ad1c3428762076ce8e6dcbdd3579b275ec2e Adapted-from: https://github.com/iconidentify/siliconforest-linux/commit/22b8c3ba2838ee58b600e654504e20b6d9959f22 Adapted-from: https://github.com/iconidentify/siliconforest-linux/commit/37cbda8089c9eae7e0b058aab08d71483afc71c4 Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 191c603)
Compile actual PIPE, role, reset and mux-policy bodies under ASan/UBSan. Check fallback-compatible hardware generation scope from the real DTs and match table. Cover current all-generation success-only state, fixed-hub non-USB3 completion, generic pending retirement, canceled transitions and reset release. Add 40 deliberate runtime regressions and before/after controls to the existing host CI manifest. The fixture models MMIO polls, BIST, lane configuration, power restoration and mutex boundaries. It makes no timing, electrical, real concurrency or hardware qualification claim. Source: https://github.com/iconidentify/siliconforest-linux/commit/37cbda8089c9eae7e0b058aab08d71483afc71c4 Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 951b351)
dcp_attributes_replace schedules a connector hotplug notification as well as its dimensions work. The deferred-attributes test used a zeroed work item, triggering a workqueue warning even though its assertions passed. Reuse the catalog observer, initialize and flush its callback, and assert one notification. Production code is unchanged. Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit ae83957)
Cover both host and device role callbacks after a Type-C mux has already configured the lanes, and an opposite-role callback on the same backend. Check that only later-generation host setup runs BIST and repeated routing does not move the PIPE again. Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 2790d5f)
…butes Carry the narrow optional platform hook used by the public native G15 adapter. The default remains false, preserving every existing driver's bind/unbind interface. G15 opts in because it cannot stop the firmware safely after startup; retaining DMA objects alone does not make a device power-domain detach safe. Based-on: Eryk Wieliczko's native G15 registration work Source: https://github.com/iconidentify/siliconforest-linux/commit/004d92a5ea8f Upstream-commit: 004d92a5ea8f Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 8b375a9)
Complete the G15 generation (T6030, M3 Pro) with what userspace sees, and build it behind the new DRM_ASAHI_G15 option (default n): - vm: user VMs, each holding an exclusive TTBAT slot (ASIDs 1..63) for its lifetime, with the shared graph's client aliases mapped into every one, in VA ranges that belong to the kernel in every VM. - queue, job, fence: logical queues own no firmware state. Each has one DRM scheduler with 16 credits; a SUBMIT is one job with one hardware fence, handed to the device-global engine, and fences signal in queue order with the VM status latched first. QUEUE_DESTROY drains the queue. - command, gate: SUBMIT parsing and validation, in order: the stream's structure, the timestamp handles, the header barriers and EXACT_PRIOR ordinals (the semantics of the G17 rules; the engine's total order satisfies every valid one), every command's mapped spans with the right access, and the CDM/VDM stream gates that admit only known records (g15_stream_gate). - timestamp: timestamp buffers mapped into the firmware's 64 MiB arena in the host window, where the firmware writes nanoseconds directly. - params: GET_PARAMS (feature word 0x3e, QUEUE_LIMITS, identity), VM_BIND batches of at most 1 << 20 operations. SOFT_FAULTS, FEW_PRIMITIVES and RENDER_RAY_QUERY are never advertised; FRAGMENT_BARRIERS only with g15_early_tiling. - mod: the generation's GPU, its shared state and device lock, and its platform driver for apple,agx-t6030. The driver is registered separately from the other generations and has no module alias: the M3 must start its GPU only after the native display is up, so the firmware boots when userspace loads the module afterwards. It suppresses its bind attributes, and a started session keeps the module and everything the firmware can reach until reboot. The module parameters g15_* carry the few tunables that stay tunable (load time only): deadlines, the repoll and completion intervals, the firmware idle standby time and confidence, recovery, the stream gates and early tiling. Tested on the J514S with Mesa's asahi Vulkan and Gallium drivers: a Minecraft Vulkan replay runs at about 161 FPS with identical frames, the deqp-vk memory model, compute, render and submit subsets show no failures, and induced compute and render hangs are ended and recovered without a firmware crash. Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja> (cherry picked from commit 13a93ec) [port to integration/aurora: the generation is a backend of the shared platform driver instead of a second platform driver without module alias: asahi.m3_backend gains the value g15, and auto starts it on the validated J514S/J516S boards when CONFIG_DRM_ASAHI_G15 is built (asahi.m3_backend=runtime keeps the M3 runtime; without the option auto keeps starting the runtime); AsahiDriver::probe_g15() keeps the device bound with no DRM device when the probe fails after the firmware was touched; drm_gpu::Backend::G15 adapts gpu::Gpu to the DrmGpu file layer, and file.rs keeps the M3 runtime's per-SoC feature rules off a G15 device; the command stream check requires the ts_frag_end_copies/ts_end_copies arrays of this tree's UAPI layout to be empty, as G17P does, since G15 implements no timestamp copies; the crate root is renamed asahi_core.rs so that g15/glue.o can be linked into asahi.o; the Kconfig text follows] Assisted-by: LLM Signed-off-by: Diljit Singh <hey@djdev.me> [Maintained-driver port: G15 subtree snapshot including the preceding firmware/engine/recovery commits a8b817e,39e7cf0bfdaa,01812cc4894f; no shared APIs or display/PMP policy imported.] Source: omacom/linux-aurora@04795f0 Upstream-commit: 04795f0 Generated-by: LLM (cherry picked from commit ead6240)
The T6031 GPU is the same firmware generation as the T6030's (RTKit 2419, the same initdata graph and canonical VAs) on a different SoC: four clusters of ten core slots, so the enabled-core mask is two words (+0xe01500 and +0xe01504), identity words 0x07041100/0x0021040a, revision 0x12, and the GPU is powered by its power domain alone (there is no PMP vote on this SoC). Its GPU node is in the kernel device tree, filled by the boot loader (the UAT regions, the loaded firmware segments as fw-text and fw-data with their VAs, and the operating points), so the regions do not come from an ADT export. Make the SoC-specific parts of the G15 driver a property of the configuration row instead of module constants: the machine check and the PMP vote (Platform), the source of the regions (the ADT export or the GPU node), the initdata placement that differs between firmware builds (Layout: root version, HardwareData VA, size, cached prefix and fourth view, the IOMapping table), the board images (Board) and the compute register values that scale with the cluster count (ComputeRegs). The identity read fills one core mask per cluster, up to four, and GET_PARAMS reports as many as the SoC has. Add the T6031 row matched by apple,agx-t6031. It carries no firmware pin and no board data yet, so admission prints the measured TEXT identity and stops before anything firmware-visible. The T6030 row keeps every value it had; its behaviour is unchanged. Signed-off-by: Naeem Malik <malik2015naeem@gmail.com> (cherry picked from commit 7dfc3c2) [port to integration/aurora: the row is selected by the shared platform driver (asahi.m3_backend=g15, or auto on the J516C when the generation is built; every other value and board keeps the T6031 runtime selection) instead of a second OF match table; both SoCs use the GPU node's single unnamed mailbox; the T6030 row keeps this tree's PMP vote through apple_pmp_set_device_power; the T6031 node is integration's static t6031-gpu.dtsi, which the boot loader fills with the same region and segment names this row reads] Assisted-by: LLM Signed-off-by: Diljit Singh <hey@djdev.me> [Maintained-driver port: G15 subtree only; shared layers and registration adapted separately.] Source: omacom/linux-aurora@d8422e0 Upstream-commit: d8422e0 Generated-by: LLM (cherry picked from commit fa787b7)
The GPU firmware TEXT a J516C's macOS 14.7 boot chain leaves loaded (iBoot-10151.140.19.700.2, RTKit-2419.140.12.release): TEXT 0x64000 and DATA 0x140000 bytes, the GKTS entropy marker at 0x5cf28, SHA-256 of TEXT with the 8 entropy bytes zeroed c26ec89c1150eb6549cf602e012d0c5f49eb756e896d6bb8342cb2c5d98b332c, measured on the loaded segment of a J516C. The same machine started through a macOS 14.8.x chain carries a different stub whose full digest has not been recorded. For that case, and for the next build of either stub, add g15_accept_text_prefix: an unpinned build whose digest starts with the given 8 bytes is admitted for that boot with a warning that prints its full identity, so it can be pinned by the next kernel without a second measurement boot. Off by default. Signed-off-by: Naeem Malik <malik2015naeem@gmail.com> (cherry picked from commit 9609d14) [port to integration/aurora: the module parameter is added to the renamed crate root asahi_core.rs] Assisted-by: LLM Signed-off-by: Diljit Singh <hey@djdev.me> [Maintained-driver port: G15 subtree only; shared layers and registration adapted separately.] Source: omacom/linux-aurora@7c28576 Upstream-commit: 7c28576 Generated-by: LLM (cherry picked from commit 9cb33dc)
The T6030 board templates do not fit the T6031: its HardwareData record is 0xb480 bytes (the fourth HardwareDataB view at +0xb400), its root version word is 0x0c08e21e84820490, its IOMapping table has 17 slots at the T6031 register blocks (Fender at sgx+0xd00000 with a 0x144000 window, eight 0xd8000 memory-cache instances, the clock generator at sgx+0xe1c000, the ANE doorbell at 0x3c945c000, ...), it has four per-cluster voltage tables, and hundreds of its words differ from the two-cluster image. Everything here was read on a J516C: the device tree the boot loader hands over, the SGX registers and that machine's initdata. Give the T6031 row its layout and IO table (VAs packed from 0xfffffc2068000000 as the T6030 table is) and take its three board images from one firmware file, asahi/t6031-j516c-board.bin, pinned by SHA-256 and built into the kernel with CONFIG_EXTRA_FIRMWARE: the images that machine's firmware accepted, as the earlier J516C driver uploaded them. The build checks the file's IOMapping records against the layout, then writes what it owns: the TTBAT address, the IOMapping VAs, the timestamp arena, the creation timestamp and the idle policy. The HardwareData object is placed 16-byte aligned (0xfffffc20406fb580) and mapped ASC-cached as a whole: the firmware's MTR alarm handler does a 64-bit atomic in HardwareDataA, which faulted at the T6030 placement on T8122. The T6030 path (templates, DVFS from the ADT, the S8 request words) is unchanged. Signed-off-by: Naeem Malik <malik2015naeem@gmail.com> (cherry picked from commit 04e10f0) [port to integration/aurora: applied unchanged] Assisted-by: LLM Signed-off-by: Diljit Singh <hey@djdev.me> [Maintained-driver port: G15 subtree only; shared layers and registration adapted separately.] Source: omacom/linux-aurora@0c98b7e Upstream-commit: 0c98b7e Generated-by: LLM (cherry picked from commit 0fb95eb)
The compute work image carries four cluster-sized register values that the two-cluster G15S values get wrong on the four-cluster G15C, where they fault the first job: the saved-state pointers into the preemption buffer at +0x2980 instead of +0x1480 (0xd80 plus 0x700 per cluster), UMA PTC (0xa599) 0x6000400020 instead of 0x13200400020, DUPM (0xd411) 0x3c000000140 instead of 0x200000001 (32 N | 96 N << 32 for N = 10 cores per cluster) and 0x1a0e9 = 0 instead of 8. The J516C's own driver also encodes a queue's first command with the saved-state pointers and the 0x12091/0x101d9 words zero; the G15C commands here are all encoded that way, which is how compute work first completed on a J516C. The values were read from 38 compute commands on a J516C and verified by the register-fault analysis of the earlier J516C driver's boots. The T6030 values are unchanged. The render (TA/FS) values for the G15C are not known yet and stay the G15S ones. Signed-off-by: Naeem Malik <malik2015naeem@gmail.com> (cherry picked from commit 9174c1e) [port to integration/aurora: applied unchanged] Assisted-by: LLM Signed-off-by: Diljit Singh <hey@djdev.me> [Maintained-driver port: G15 subtree only; shared layers and registration adapted separately.] Source: omacom/linux-aurora@d3283ae Upstream-commit: d3283ae Generated-by: LLM (cherry picked from commit fdebd36)
Allocating an orphan with KBox::new consumed the live session before the allocation could fail. Allocate uninitialized storage first and explicitly retain the still-owned session if that allocation fails. Allocate the VM registry inside the guarded backend construction, so its failure takes the same orphan path as the other post-READY allocations. Make the UAT Arc-clone accessor infallible and use c_char for the host handoff fixture. Register drop-count controls using the actual retention, backend and KBox ownership bodies: 23 cases retain every live owner and backing buffer; both original P1 bodies compile and fail at runtime after dropping five owners. The normal control passes before and after them. Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> Generated-by: LLM (cherry picked from commit caafb8e)
ASoC unmutes active playback DAIs at D2 before publishing D0. MCA takes SNDRV_PCM_INFO_RESUME from ADMAC, so an active PCM can then resume through TRIGGER_RESUME without another prepare. Discarding the D2 unmute therefore leaves the headphone output muted until the stream is restarted. Remove the speculative card-power-state gate added to CS42L84. No missing- BCLK I2C hang has been established on this codec. Preserve the bounded PLL lock check and its oscillator rollback; D0 is not evidence that BCLK runs. The direct-BCLK path returns to its behavior before the gate was added. Replace tests that assumed a second unmute after resume with a running stream, suspend mute, single D2 unmute and D0 notification sequence. Cover PLL and direct-BCLK configurations and refusal when the PLL cannot lock at D2. UML KUnit passes all 17 cases; restoring the old gate fails six of them. No hardware suspend/resume qualification is claimed. Fixes: 4b69aec ("ASoC: cs42l84: keep the codec on the oscillator until BCLK runs") Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 46420c4)
Use one checked layout for atomic admission and DCP surface construction. Bound tile metadata, wire fields, offsets and GEM extents using the same geometry that is programmed. Refuse fractional-bpp Interchange formats (P030/P230) until their compressed packing is established; keep packed linear formats and existing integer-bpp layouts. The format callback also honors the plane's modifier list, so linear-only planes cannot acquire Interchange through DRM's callback dispatch. Use the actual framebuffer qualification required by the DMA helper. Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> Generated-by: LLM (cherry picked from commit 8c26443)
Compile the actual plane layout, atomic check, surface update and DRM format-dispatch bodies with the exact DMA-helper prototype. Cover wire bounds, undersized objects, fractional Interchange refusal, valid packed linear geometry and linear-only plane restrictions. Register the standalone suite in the CI host manifest. Twenty controls pass and ten runtime mutants fail with valid before/after controls. An old-callback control reproduces the modifier-list bypass. These host checks do not qualify hardware rendering. Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> Generated-by: LLM (cherry picked from commit b0effd1)
ASoC resumes playback unmute at D2, before MCA may supply SCLK. Keep the observed base-M1 clock guard, but do not let PCM RESUME bypass the prepare that supplies the host clock and unmutes again. Clear only playback's SNDRV_PCM_INFO_RESUME in DAI prepare, after hw_params assigns runtime info. ALSA preflight then refuses RESUME before either DPCM side is triggered; userspace recovers through prepare. Capture and other flags are unchanged. Admit both parts through one immutable probe-time flag: base-M1 OF root, CS42L42/83 OF codec and non-SoundWire transport. That macaudio jack path permits one FE per BE, whose runtime is shared. This is not a generic shared-BE fix: other hosts, J700 and SoundWire retain their previous D2 unmute and resume capability. No machine-wide speaker downgrade is added. Exercise real ALSA RESUME/PREPARE ioctls and ASoC DAI dispatch, linked capture-first preflight, prepare/unmute recovery and a real-OF admission matrix. Both codec suites pass 52 UML KUnit cases without skips. Original policy and deliberate guard, capability and admission mutants fail with matched controls. SoundWire compiles. Hardware suspend/resume remains unqualified. Fixes: 30f3203 ("ASoC: cs42l42: keep the codec on the oscillator until SCLK runs") Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> Generated-by: LLM (cherry picked from commit 99e2571)
The T8103 PHY compatible is also a fallback on M1 Pro/Max and M2. Admit only an apple,t8103 machine with the T8103 hardware row for the added live handoff retries, host-BIST bypass and early USB3 setup. Keep the existing generation-specific register and clock programming unchanged. An early role callback may configure USB3 only after Type-C requested plain USB3. Preserve the existing retry obligation when the role arrives before that request, or lane configuration fails, so the later mux update can finish the backend route for host and device roles. A committed non-USB3 mode clears that obligation, including same-mode updates. Exercise actual production bodies in 36 host controls and 49 runtime mutants, including both role/mux orders, lane/backend failures and M1/M2 fallback chips. All controls pass and every mutant is detected. Add a real-OF KUnit case for the machine/hardware admission matrix; all 36 ATC KUnit cases pass in arm64 QEMU. Update the DCP host fixtures for the narrowed admission; all 56 controls pass. No M2 or Pro/Max USB3 hardware qualification is claimed. Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> Generated-by: LLM (cherry picked from commit aac315d)
The AV endpoint receive queue and the audio owner's system workqueue have separate lifetimes. Flushing endpoint RX does not wait for running service-start or Terminate work embedded in the devm audio owner. Stop the native endpoint initializer before looking up its owner during unbind, including the dead-session path, and during permanent teardown. Refuse new owner work and synchronously disable the initialized works while endpoint RX can still complete their RPCs. Track one owner disable so successful rebind enables each work once; failed initialization stays disarmed. Capture the exact owner and endpoint in an owner-associated devres action registered after the AFK context action and before endpoint start. Drain that owner before quiescing its captured endpoint, then clear publications only when their owner still matches. Handle absent and failed endpoints, and run the same owner-only cleanup on initialization failure without waiting on the native initializer from inside itself. Preserve the original initialization errno and guard ordinary shutdown against error pointers. The ARM64 QEMU KUnit control passes all 26 AV tests with KASAN and lockdep. Actual devres group release, running and pending owner work, producer and owner rebind balance, missing/error endpoints, the action-reset registrar failure boundary, and successor publication are covered. Eight production-body mutants compile and fail runtime assertions; the restored control passes without guest diagnostics. Update the J700 readiness host fixture for the new no-owner cleanup boundary; native audio remains forbidden there. Hardware teardown is not qualified here. Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit 9904393)
The stale-temperature test unlinks the fake SMC key list before creating its replacement FIFO. If the sampler opens the path between those calls, it correctly reports a missing key list instead of a stale sample. CI then fails the diagnostic assertion despite stopping the load safely. Create the FIFO at a sibling path and rename it over the key list. Reads already in flight can complete; later opens block on the FIFO. Keep the separate missing-key test and all load-stop and restoration assertions. No driver or installed tool changes. Widening FIFO creation by 250 ms reproduces the exact CI assertion on the parent and passes with this change. Generated-by: LLM Signed-off-by: Chris Kearney <303316+iconidentify@users.noreply.github.com> (cherry picked from commit d60121f)
Linux 7.3 passes the tunnel's callback data to the DP tunnel callback, and tb_tunnel_alloc_dp() gives tb_dp_tunnel_active() the domain as that data. Match the extracted DPRX completion tail against the two-argument call, and give the host model's tunnel the same callback shape and data. The test checks the same release order as before: the callback runs, then the tunnel and domain references are dropped. All 16 cases pass and all 16 mutants are caught.
The PMP report driver in this tree can hold power state requests in memory until the PMP runs, for SoCs whose offsets set defer_requests. The extracted probe and set_state bodies therefore use the request shadow, the deferral flag, a delayed work item and apple_pmp_report_running(). Give the host model these fields and extract apple_pmp_report_running() with the other production bodies. T6030 does not defer requests, so the delayed work stubs fail the test if they are ever reached. All 18 graph cases pass and every mutant is caught.
The native G15 backend is built only with DRM_ASAHI_G15, which is off by default, so the shipped configuration does not compile it. Port it to the 7.3 interfaces the rest of the driver already uses: - module parameters: ModuleParam::value() returns the value itself; drop the dereferences. - the scheduler job's owner is kernel::module::this_module::<crate::LocalModule>(), as for G17. - IdArray::new() takes a reference to each entry's info, so the OF match table holds &&Config. - of_device_table!() no longer takes a module table name. The T6030 autoload table carries no ID info and uses its @none form. - apple_rtkit_ops has the optional syslog_is_debug policy; G15 keeps the default log level. The parameter-buffer growth host test models module parameters the same way, by value. No functional change. With DRM_ASAHI_G15=y the asahi module builds and links on 7.3.
This tree refuses charge control thresholds outside 0..100 with -EINVAL before any SMC key is read or written (3c82528, "power: supply: macsmc-power: propagate threshold read errors"). The end threshold test expected -1 to clamp to the minimum and 150 to clear the limit, as on a tree without that check. Move -1, 101 and 150 to a refusal check: each returns -EINVAL, leaves CHLS as it was and makes no SMC access. The in-range cases are unchanged. All 11 macsmc-power-charge cases pass.
446aa24 to
18ba468
Compare
Keep the original sync commits intact and reconcile the native G15 backend with the current G16/G17 registration, memory and timestamp APIs. Preserve SoC-specific PHY and power-domain ownership rules, combine the host fixtures, and describe the T6030 audio and tunnel routes in bindings. Assisted-by: LLM Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
|
The public head is Qualification commit Both full LLVM WERROR recipes passed with the same configs and toolchain. Fedora completed after a preserved storage-guard interruption with unchanged inputs. Fresh arm64 QEMU KUnit passed 779 tests with one suite-defined skip, and asahi-g15 passed 61 tests with two ignored. M2 and then M1 Pro passed their candidate one-shots against 73x. Native GPU readback, codec binding, USB-C and Apple USB4 controllers, Wi-Fi, speaker protection and enforcing SELinux matched the baselines, with zero warning, BUG, SError, DART-fault or AOP-failure markers. Both are back on 73x with the test kernels, entries and fallback timers removed. Saved defaults, boot.bin and the 7.1 DTB set were preserved. Ready for merge at this head and qualification commit. One existing metadata note, the four original 7.3 adaptation commits |
Model the pinned KBox pointer used by the shared G16 registration and include its actual conditional alias and helper. Run the G15-only and combined G15/G16 configurations, including late G16 registration failure and an activation-order mutation. Assisted-by: LLM Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
Keep the existing Rust 1.93.1 default while allowing an explicit host compiler for the PB growth and stats drain runners. Record the compiler version. Allow the retention test to consume the pinned public pre-fix source when that historical file is unavailable locally, preserving its ownership-drop controls. Assisted-by: LLM Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
Declare property-presence lookup and the rejected T6030 readiness probe in the M4 supplier fixture. Keep declared-but-invalid report links on the report path, and check that absent reports cannot admit either of the fixture's non-T6030 suppliers through the legacy fallback. Assisted-by: LLM Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
The soft-fault check uses the fully qualified module path, so this import is unused and fails builds with warnings treated as errors. Signed-off-by: Ryan <ryan@murraytraining.com.au> Assisted-by: LLM
Resolve the Makefile conflict with aurora-silicon#395: keep the shared fwid.o and add the G15 glue object. G15 has its own identity and SHA-256 helper, so the move of the G16 identity into fwid does not touch it. Assisted-by: LLM Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
Resolve the conflict with #395 in apple_drm_uninit(): withdraw the T6030 display publication first, then keep the T6041 rule that an aborted takeover leaves its DCPs running. Assisted-by: LLM Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
Include the newly qualified #395 omacom #177 sync on Linux 7.3 before running consolidation gates. Keep the newer complete DWC3 host fixture instead of duplicating its role fields and runtime-PM helpers. Retain the WIP-only files and non-conflicting changes from the earlier normal merge. No history is rebased or force-updated. Signed-off-by: Ryan Murray <ryan@aurorasilicon.org> Assisted-by: LLM
Ports the Aurora work that omacom/linux-aurora
integration/auroramerged in omacom/linux-aurora#177 (42672ed79d) onto the 7.3 tree, based onintegration/auroraatd7fc14ef83. #388 covered omacom up to70d5dcedfa; this picks up from there. Commits were cherry-picked oldest first with-x, with their original messages, authors and author dates.omacom/linux-aurora#177's history holds 2,773 non-merge commits after
70d5dcedfabecause it carries Ryan's 7.2.11 update (omacom/linux-aurora#158). The new Aurora work is the 131 commits on top of that update. 123 of them are ported below. The rest of the history is covered under "Not ported".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 commits for 7.3, by DjDeveloperr:
e085a95c5048e11855fcfc13d05b1d18ba468d82Adapted (commit messages unchanged):
0c4d483cde(user4k,MREMAP_DONTUNMAP): 7.3 already resetsvm_pgoffand the anonymous page offset of an unfaulted source withvma_set_pgoff()andvma_set_anon_pgoff(). The sub-page offset is set next to them withvma_set_subpage_offset(), asinsert_vm_struct()does on 7.3, instead of withvma_set_page_offset().0f366074d0(shared G15 hooks): the newGpu::feature_flags()hook readsfault_controlby value, as 7.3'sModuleParam::value()returns it. File open keeps this tree's per-client GPU memory accounting next to the newuser_range()hook. On thed7fc14ef83base,gem_createalso passes its registration data, which this commit renames toreg_data, to the large-BO reclaim policy from drm/asahi: fail large G17 backing population without direct reclaim #397.gem.rskeeps both that policy'suser_pages_no_reclaim()and the newprepare_new_object().3f80156817(G15 adapter registration): 7.3'sof_device_table!()does not accept attributes on its entries. A small local macro builds the maintained OF table and adds the T6030 and T6031 entries only whenDRM_ASAHI_G15is off. The table is the same as on omacom in both configurations.81399ec579,951b351f57,ea34dfd32a,caafb8e7a1andb0effd1c9bleave out theirtools/aurora-ci/host-tests.jsonhunks. That file is part of omacom's CI, which this tree does not carry (as in Sync omacom integration/aurora work onto 7.3 #388).0c4e44fadb,5108272d51and4c2b41a0e3conflicted only on context inpmp-report.c, next to this tree's deferred PMP requests. Their changes apply unchanged.Not ported:
32856f39fe).a6b50a114e(t600x DCS sleep state) andc703d40182(t602x old PMGR misc driver binding) are workarounds for the pre-7.3 pmgr-misc driver. Unified 7.3 dropped them because the upstream driver replaces them.tools/aurora-ci:c19c6c5a66,d0a6c435fa,53b2ab4e38,bfceddad62,a1a72a5e40,6d31ee55c5,6bd40b57a2and03c3afe702.integration/auroraalready has in newer form.Config effects (package configuration plus
asahi.config, afterolddefconfig):RESET_APPLE_CIOchanges from m to y (669333d2fa).asahi.configalso drops its duplicateTYPEC_SN201202XandUSB4_APPLE_SOClines. Both stay m, and the twoolddefconfigoverride warnings go away.APPLE_SIO_T6030=yandDRM_APPLE_DCP_T6030_AUDIO=y. Both affect only T6030. DP/HDMI audio stays behind the existinghdmi_audioparameter.DRM_ASAHI_G15is new, off by default and not set byasahi.config. It selects the hiddenAPPLE_T6030_GPU_PMP_LINK. While it is off, T6030 and T6031 stay on the maintained GPU driver.Testing:
linux-aurorapackage configuration plusasahi.config,olddefconfigandCONFIG_RUST=y, giving 7.3.0-ARCH. Toolchain: clang/LLVM 18.1.8, rustc 1.85.0, and the tree's bindgen 0.73.2. The based7fc14ef83was built in full, then this branch incrementally in the same output directory.Image, modules (1,916, as on the base) and all 116 Apple DTBs (including the base's new J616s) build. There are no new warnings. The two existing Rust dead-code warnings (store.rs,t8140_aop_audio.rs) are unchanged, and the two config override warnings are gone.CONFIG_RESET_APPLE_CIO=yafterolddefconfig.DRM_ASAHI_G15=ybuilds and linksasahi.ko. The G15 code then reports 27 dead-code warnings (unused constants and helpers underg15/), which do not appear in the shipped configuration.S35_dual_retry_off), 0 fail. The first run failedmacsmc_charge_end_threshold_test;18ba468d82fixes it.avd-codec-tail-tests.py,dcp14-panel-tests.py(both variants),tunnel-route-tests.py,tb-dp-repair-tests.py(16 cases, 16 mutants caught),t6030-clock-tests.py,t6030-wiring-tests.py,tiled-split-tests.py,t6030-pmp-supplier-tests.py,t6030-pmp-graph-tests.py(18 cases, all mutants caught),t6030-display-admission-tests.py,t6030-display-linkage-tests.py,atc-usb3-tests.py,apple-dcp/plane-layout.py, the eightg15-*-tests.pysuites andasahi-g15/run.py.Known issues:
tools/testing/selftests/drivers/gpu/apple-dcp/run.sh) passes 51 of 57 cases, the same as on the base.service-rebind,cpuidle-s2idle,probe-mux-pm-dispatch,dwc3-teardown,same-handle-srcu-indexandmux-release-stagesalready fail on8c2774e57f. The commits since then do not touch the sources these fixtures use.tools/aurora-sep/tests/test_m3_max.py(its fixture changes ind60121ff24) runs 68 cases.SameAs122Test.test_every_board_runs_as_on_12_2fails for 4 to 7 boards per run, on this branch (6, then 4) and on the base (7), with a different set of boards each time. In each failure, a "Verification timed out" line appears in only one of the two runs being compared. The test is timing-dependent on this host; the kernel changes do not cause it.auto-dischargecharge behaviour. CHLS-only Macs gain acharge_behaviourattribute (auto,auto-discharge). Probe clears a forced-discharge bit that was left set without a limit. Theauto-dischargevalue follows the still-open power: supply: macsmc: add "auto-discharge" charge behaviour for CHLS AsahiLinux/linux#521 and may change.RESET_APPLE_CIOis built in on every Mac.🤖 Generated with Claude Code