Repository navigation
Carry 13 fixes from omacom/linux-aurora - #1989
Draft
RyanTheTide wants to merge 14 commits into
Draft
RyanTheTide wants to merge 14 commits into
RyanTheTide wants to merge 14 commits into
Conversation
The extra trailing zero places the PMU address outside the PMGR range on J613. Correcting it resolves the probe-time firmware translation fault on a J613 running firmware compatibility 14.7. Hardware tested as a module override on the original PR95 kernel, 7.1.12-2-20261013.95-sep-ARCH. Rebased integration source has separate hardware qualification pending. Link: omacom/linux-aurora#117 Assisted-by: Codex:GPT-6 Signed-off-by: Scott Jones <scottajones@gmail.com> Link: omacom/linux-aurora@6b34040
…zed tunnel ports apple_pcie_config_write() tells the port block about the root port's prefetchable window (PORT_PREFMEM_ENABLE, port register 0x994) when the PCI core programs PREF_MEMORY_BASE/LIMIT, but only if pcie->kernel_init is set. kernel_init is true for apple,t8103-pciec with apple,pciec-kernel-init. The T600x/T602x tunnel ports use pcie->tunnel_cold_init instead, so on those ports the hook never runs and the enable bit stays 0. Measured on a 16-inch M1 Max (T6001), Intel Arc Pro B60 in a Sonnet Breakaway Box on USB4 tunnel 2 (pcie@f30000000), kernel 7.1.12-2-20261013.18: - Every CPU read of a BAR behind the root port's prefetchable window returned all-ones, with no AER, abort or UR bits on any bridge. This held for the 64-bit window with the identity DT range, for the 32-bit prefetchable window, and with the BARs placed inside the working non-prefetchable CPU window. A function behind the same chain in the non-prefetchable window read correct data in every configuration. - A runtime bisect with setpci gave each bridge in turn its prefetchable window while the others forwarded through widened non-prefetchable windows: the windows of 04:01.0, 03:00.0, 02:00.0 and 01:00.0 each read data; only the root port's own prefetchable window (0001:00:00.0) read all-ones. - PORT_PREFMEM_ENABLE (phys 0xf21000994) read 0 on that boot (0x990 read 1). With the kernel's original bridge windows in place and the BARs reachable only through the prefetchable windows, reads returned all-ones; after one 32-bit write of 1 to the register (writel_relaxed, mb, isb, non-posted mapping) the same reads returned correct data (B60 BAR0 registers and VRAM). The register reads back 0 after the write, so it is a write strobe. Run the existing hook for tunnel_cold_init ports too. T6001 only until it is tested on T6021: on a 14-inch M2 Max (T6021) the prefetchable window forwarded once the DT range was identity. Tested: the register write above (direct, from a test module). The hook path itself has not been booted: a rebuilt kernel changes the BTF base of the modules. (cherry picked from commit 4466fe2ad955d47290fc79c13963c51c0b253706) [conflict: our config_write takes the cio_cfg_lock and runs the MPS timer hook first, so the kernel_init test moved below them; the change is applied to that test] Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com> Link: omacom/linux-aurora@bb2885b
…dler brcmf_pcie_quick_check_isr() masks the dongle with brcmf_pcie_intr_disable() whenever mailboxint reads non-zero. Since commit f04be34 ("wifi: brcmfmac: Do not service msgbuf IRQs until ready in MSI mode") that helper also clears irq_ready, and the threaded handler services the rings only while irq_ready is set. The thread therefore skips brcmf_proto_msgbuf_rx_trigger() for exactly the interrupts the hard handler caught with mailboxint still set, and then sets irq_ready again on its way out. In MSI mode the firmware clears mailboxint itself, so whether the hard handler sees it set is a race. When it does, completions that arrived with that interrupt stay in the D2H rings until a later interrupt. On firmware that sends mailbox data over the control ring (no BRCMF_PCIE_SHARED_USE_MAILBOX) that includes the D3 ACK, and ioctl responses: brcmf_msgbuf_query_dcmd_len() and brcmf_pcie_pm_enter_D3() then time out, the latter aborting system suspend. On a MacBook Air (M1) with BCM4378 this became frequent with PCIe ASPM L1.2 enabled. In INTx mode the hard handler always clears irq_ready first, so the rings are never serviced at all. Let the hard handler mask the dongle without touching irq_ready, and let the thread unmask without setting it, so irq_ready is controlled only by the code that sets the rings up and tears them down. Fixes: f04be34 ("wifi: brcmfmac: Do not service msgbuf IRQs until ready in MSI mode") Assisted-by: Claude:claude-opus-5-5 Signed-off-by: Naeem Malik <malik2015naeem@gmail.com> Link: omacom/linux-aurora@3a936ed
An AP in WPA/WPA2 mixed mode offers CCMP and TKIP pairwise ciphers with a TKIP group cipher. The driver registers only CCMP, so iwd refuses the BSS with NotSupported before any association attempt, and the failure looks like a wrong password. The firmware runs the handshake and owns the keys, and it joined such an AP and passed unicast and group traffic with no other change. Register TKIP and accept it as the group cipher for WPA2-PSK. Pairwise TKIP, WPA1, and a TKIP group with SAE or OWE stay refused. Tested on J700 against one AP's 5 GHz network: gateway and internet traffic, DNS, and mDNS discovery of other LAN devices through the group key. Link: omacom/linux-aurora#398 Assisted-by: Claude:claude-opus-5-5 [conflict: kept our 6 GHz band check and PSK_SHA256/AES-CMAC entries and our PMF-required PSK test row; the TKIP group rule follows the band check] Signed-off-by: zoan37 <104385984+zoan37@users.noreply.github.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Link: omacom/linux-aurora@eb6549c
The station security notes still say CCMP only. Describe what the driver accepts, and record that MIC failure reporting and TKIP countermeasures are left to firmware and have not been verified. Link: omacom/linux-aurora#398 Assisted-by: Claude:claude-opus-5-5 [conflict: our notes already list PSK-SHA256, OWE and BIP-CMAC-128; the TKIP text is added to them] Signed-off-by: zoan37 <104385984+zoan37@users.noreply.github.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Link: omacom/linux-aurora@a2b0f53
bindgen emits valid unformatted Rust when rustfmt is unavailable. Accept whitespace around the unconditional const assertion and reject both formatted and unformatted test attributes. Keep the C/Rust compilation and layout assertions unchanged. Assisted-by: LLM Signed-off-by: Kallisti <kallisti@kallisti.io> Link: omacom/linux-aurora@6ba4695
Before the module's final activate() step, the G15 probe returned -EPROBE_DEFER. That put the GPU on the driver core's deferred-probe list. activate() then runs the real probe through driver_attach(), but if that probe fails for good, for example because the firmware is rejected, the driver core does not take the device off the list. It stays in /sys/kernel/debug/devices_deferred until the deferred-probe work next runs, so anything that waits on that list keeps waiting. Return -ENODEV instead. activate() already reprobes every unbound device once MODULE_READY is set, and a device added later sees it set, so nothing needs the deferred list to retry. The display and PMP waits inside the real probe still return -EPROBE_DEFER. g15-probe-init-tests.py now checks that every probe before activation returns -ENODEV, and adds a mutant that defers again. Assisted-by: Claude:claude-opus-5-5 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> [conflict: our module also registers the G16 adapter, so the host test counts each attempted registration instead of assuming two, and the mutant table keeps our activation-before-g16-registration entry] Signed-off-by: Scott Jones <scottajones@gmail.com> Link: omacom/linux-aurora@6447458
J514C and J516C native display now starts on normal boots. The PMP
failure diagnostic still recommends omitting enable=1, which no longer
requests the boot-framebuffer fallback on those boards. Name the explicit
apple_t6031_display.enable=0 owner opt-out instead.
This changes only the diagnostic string. It does not change PMP startup,
matching, handoff checks, display defaults or GPU admission. Checkpatch
passes with sign-off checking disabled, and the existing production
selector/owner-opt-out host checks pass. Normal kernel CI will run on
the new head; the prior8c309ed2a18d full technical CI remains recorded.
[on this tree J514C is still opt-in and b55bcaaba520 is not present; the
string is still wrong on the J516C, whose native display is on by default,
and enable=0 matches the T6030 and T8122 diagnostics]
Fixes: b55bcaaba520 ("soc: apple: default J514C native display on normal boots")
Assisted-by: Codex:GPT-6
Signed-off-by: Scott Jones <scottajones@gmail.com>
Link: omacom/linux-aurora@10a8340
…e M3 runtime The 7.2 rebase replaced the T8122 14.x branch of apple_pmp_report_wait_supplier_ready() with a T6030-only one, so on an M3 MacBook Air every GPU power vote returns -ENODEV even with the PMP running. With the new hidden APPLE_T8122_GPU_PMP_LINK, which DRM_ASAHI_M3_T8122 selects (as DRM_ASAHI_G15 selects APPLE_T6030_GPU_PMP_LINK), wait for the unique T8122 report's display request (DISP, reg 7) to be ready and acknowledged before AGX, as on the 7.1.12 kernel the runtime was qualified with. The PMP graph and supplier host tests copy the dispatch, so they copy the T8122 wait too, built off there; their mutations still target the T6030 path. [conflict: our dispatch also selects the T8132 report; the T8122 14.x branch is added in front of it unchanged] Generated-by: LLM Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Scott Jones <scottajones@gmail.com> Link: omacom/linux-aurora@d8ecb2e
Document the display consumer's optional DMA aperture, which can differ from the defaults of individual scanout DARTs. Include the T8122 display subsystem compatible so its shared aperture is checked by this binding. Assisted-by: OpenAI Codex (gpt-6.1-sol) [conflict: our binding lists compatibles under oneOf and adds the J713 memory-region properties; the T8122 compatible and apple,dma-range are added to it] Signed-off-by: DjDeveloperr <hey@djdev.me> Link: omacom/linux-aurora@b53dc2b
Domain allocation already lets a consumer specify apple,dma-range, but of_xlate intersects the provider defaults before that override is read. A consumer with a common scanout alias and DARTs with disjoint defaults therefore fails configuration before its domain can be allocated. Apply the same consumer aperture while parsing each specifier, retaining validation of both explicit five-cell windows and the override. Share the alignment, overflow and alias-width checks. Consumers without an override keep their existing windows. Add KUnit cases for the J613 high scanout alias across internal/external DART defaults, independent low PIODMA, malformed overrides and invalid five-cell specifiers. An ASAN/UBSAN fixture executing the production parser and of_xlate reproduces the old rejection and accepts the common window. Assisted-by: OpenAI Codex (gpt-6.1-sol) Signed-off-by: DjDeveloperr <hey@djdev.me> Link: omacom/linux-aurora@21d3646
The display gate adds the external display DART to the shared display subsystem. Its low PIODMA default does not overlap the internal display's inherited 1 TiB alias, so their defaults cannot be intersected for scanout. Describe the subsystem's common scanout aperture explicitly, matching the internal DART. The external PIODMA consumer retains its existing low aperture and the external processor still requires the firmware-ready gate. This is a software-validated candidate; hardware external-display qualification remains pending. Assisted-by: OpenAI Codex (gpt-6.1-sol) [conflict: our dispext0 comment was shortened in 3b1f90e; the sentence about the display subsystem aperture is added to it] Signed-off-by: DjDeveloperr <hey@djdev.me> Link: omacom/linux-aurora@56e6f92
The 14.7 presenter scans out one unscaled image of the mode's size from the top-left of the framebuffer, and iomfb_v14_7_atomic_check() admitted only planes that describe exactly that. It compared the plane's raw src_* and crtc_* fields with the mode, so it rejected a plane whose destination reaches past the right or bottom edge of the CRTC with -EINVAL. The DRM uAPI allows such a plane: apple_plane_atomic_check() has already clipped it with drm_atomic_helper_check_plane_state() (can_position is true), and the part left on the mode is what the presenter shows. KWin puts up such a plane when it mirrors one output on another. On a J514S (M3 Pro), with the 3024x1964 panel at scale 1.7 mirrored onto an XREAL One (1920x1080 over DisplayPort alt mode), it centres a 1663-pixel wide picture and rounds the 128.5-pixel border up on both sides, so the external primary plane is 1921x1080 on a 1920x1080 mode. Every frame on the glasses then failed its commit, and they showed nothing. Once the glasses dropped and asserted HPD again, the connector's link-status went bad and KWin forced a modeset that included the panel. That commit failed on the same plane, and KWin retried it on every frame, so the panel froze too, with the pointer: around 350 failed commits a second until the mirror was turned off. Judge the plane by its clipped rectangles instead: the clipped source must start at the framebuffer's origin and, like the clipped destination, be of the mode's size. Hand the presenter the clipped source size on external pipes. Planes that start left of or above the mode, are offset into the framebuffer, are scaled or do not cover the mode are still rejected, and the panel still takes only a framebuffer of its mode's size, as before. A plane that the old check admitted has clipped rectangles equal to its raw ones, so it is judged and shown as before. The new KUnit suite runs drm_atomic_helper_check_plane_state() with the limits apple_plane_atomic_check() passes it, and checks which planes are admitted and, through the helper flush uses, what size the presenter is given. Fixes: 98d767e ("drm/apple, media: apple: avd: T6030 internal panel, external display and AVD fixes") Signed-off-by: Eryk Wieliczko <eryk@wieliczko.ninja> Assisted-by: Claude:claude-opus-5-5 Link: omacom/linux-aurora@b9ed257
Contributor
Author
|
Hardware status for b908da4: the M2 (T8112) rows pass: brcmfmac traffic and s2idle x3 on rtc1, the same DART bindings and 0 apple-dart errors through boot, Wi-Fi, s2idle, DPMS and GPU. Before merge: 1674013 applies a consumer's |
The T8122 14.x GPU readiness wait was gated three times: by a hidden per-machine Kconfig symbol (APPLE_T8122_GPU_PMP_LINK, selected by the M3 runtime), by the "apple,t8122" machine compatible, and by the hardware itself, an "apple,t8122-pmp-v2" supplier with exactly one available "apple,t8122-pmp-v2-report" node whose "apple,pmp" points back at it. The hardware match is enough. Drop the Kconfig symbol and both machine checks, and try the T8122 wait first for every supplier without a declared report: it returns -ENODEV unless the PMP and its report are there, and the T6030 wait follows as before. Suppliers on other SoCs get the same results as before. On a T8122 whose device tree has the report, the wait now also runs in kernels built without the M3 GPU runtime. The host tests lose their CONFIG_APPLE_T8122_GPU_PMP_LINK=0 define. Signed-off-by: Ryan <ryan@aurorasilicon.org>
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.
Carries 13 commits from omacom/linux-aurora
integration/aurora(39d5f88a, 2026-11-14) that this tree did not have. Each one keeps its original author, author date and message. ALink:to the source commit is added, and where a pick conflicted, a bracketed note says how it was resolved.Their branch forked from ours at 44d5afc (#139, 2026-10-20). It has 2792 commits since then. 2683 of them are already here, matched by author and date or by subject. None of the other 109 has a patch-id match here. Those 109 were reviewed one by one, and the 13 below are the ones kept.
Carried
enable=0, as T6030/T8122)apple,t8122-display-subsystemcompatible our DT already uses)Credit: Naeem Malik, zoan37, Joshua Warren, garymeyerca, Scott Jones, Kallisti, Chris Kearney, DjDeveloperr and Eryk Wieliczko wrote these in omacom/linux-aurora (https://github.com/omacom/linux-aurora). Issues cited in the commits: omacom/linux-aurora#117 and #398.
Checks
LLVM=1 ARCH=arm64,W=1), every touched directory, plus the Apple DTBs. No errors and no new warnings in the touched files; all 137 Apple DTBs build.apple_dart_test_consumer_windowandapple_dart_test_invalid_consumer_window) and mt7932-association (mt_connect_auth_testwith the TKIP rows).g15-probe-init-tests.py --mutationspasses, and all 11 mutants fail as intended, including the newpre-activation-defersmutant.t6030-pmp-graph-tests.pyandt6030-pmp-supplier-tests.py --mutationspass. The aurora-bindgen compile test passes with the changedtest.py.Not carried
apple_radio_sleep), where losing firmware across a lease returns -EUCLEAN.platform_set_drvdata().OwnerModule.Driver::MODULEdoes not exist here.apple,t6034-*display path. Here the T6034 is described as the T6031 die (3aa8b5f). The firmware identity pins in it are not carried.&sioon every T8103 board, but 5417d5b later put both back on J274 only. Restoring 3afa738 is the fix and goes separately.tools/aurora-cicommits, and thetools/aurora-sep/recipes/mesa-m3packaging commits: not part of this tree.Hardware rows owed
apple,dma-rangeon every consumer). Run the apple-dart KUnit.range-diff (source commit vs carried commit)
On top of the carried commits, 99b2d3a selects the T8122 GPU readiness wait by the
apple,t8122-pmp-v2PMP and its linked report node only, dropping theAPPLE_T8122_GPU_PMP_LINKKconfig symbol and the machine checks.Assisted-by: LLM (Claude:Opus 5.5)