Repository navigation
Conversation
e8becc3 to
29d7657
Compare
29d7657 to
3f26aa3
Compare
|
Updated: rebased onto integration/aurora 6ea76a5 without conflicts (old head 29d7657). Two commits from Jens's K116s bundle of 2026-11-12 are added on top:
Every commit builds ( Jens's notes also suggest moving the shared |
RyanTheTide
left a comment
RyanTheTide
left a comment
There was a problem hiding this comment.
@sithglan dt-checks reports 9 new warnings on t6050-k116s.dtb. Two things are needed:
- The port tunables on pci@1,0 (apple,apcie-config-tunables and the apple,pcie-rc-*tunables) are not in apple,pcie.yaml. Please add them to the binding in a separate commit; J714s uses the same properties.
- gpu@2300000000 has only the "sgx" reg. Either add the reg entries the apple,agx binding expects or adjust the binding for the loader hand-off form, so the node validates.
Small things: please keep the apple,k116-macaudio entry sorted in apple,macaudio.yaml, and add your own Signed-off-by to 3f26aa3 since you edited its message.
|
[Joshua's agent] K114c (14-inch M5 Max PROLED) panel values, for alignment with this PR's K116s panel node. Read-only, from that machine's own firmware device tree (values only):
Cross-check against this PR: 345 x 223 mm at 3456 x 2234 is 254.5 / 254.6 ppi, so the K116s numbers are consistent with the same panel density. Our K114c node will be panel { compatible = "apple,panel-k114", "apple,panel"; width-mm = <302>; height-mm = <198>; }, with no adj-height-mm and no apple,panel-mini-led, using the [jk] binding pattern this PR adds. Our commit depends on this PR's binding change and is not booted yet; the mm values stay marked unverified until the display hand-off runs and the DCP reports its own size. |
When the T6050 driver brings a root port up itself (pwren-gpios), or restarts an adopted one, it applies five tunable lists from the port's node. The first, apple,apcie-config-tunables, goes to the port registers. Then apple,pcie-rc-tunables and the Gen3, Gen4 and Gen5 rate shadow lists go to the root complex configuration space. The J714s and K116s SD card ports carry all five, but the binding described none of them, so their pci@1,0 nodes fail validation. Describe them in the T6050 tunable format and keep them off the other SoCs. Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de>
Two allOf entries match every node with apple,agx-g17s: - The M5 Pro hand-off entry wants one "sgx" reg and no board operating points. - The J716c entry wants "asc" and "sgx" plus the operating-point tables. No node satisfies both. The J714s, K114c and K116s GPU nodes fail the J716c entry, and the J716c node fails the hand-off entry. Apply the hand-off entry to a node whose only compatible is apple,agx-g17s, and the J716c entry to apple,agx-g17c. Also drop the second apple,agx-g17s in the enum of the non-coprocessor entry, which dt_binding_check reports as non-unique. Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de>
…s N1 aura's T6050 hand-off reads the PCIe host's iommu-map and refuses the boot when it is absent, which the bare disabled pcie0 of these boards hit. Give the host its iommu-map and bus-range, and on the K116s describe port 0 with the N1's three functions as on J714s, with this board's platform ID (ADT /chosen board-id 0x16 << 16 | 0x6050), the PERST# pin 80 exposed and the wifi0/bluetooth0 aliases for the loader's factory addresses. The K114s keeps the host disabled until its board ID is known. Signed-off-by: Jens Axboe <axboe@kernel.dk>
Port 0 carries the N1 (ADT /arm-io/apcie0/pci-bridge0 with centauri-control, -alpha and -beta). Describe the port and the N1's three functions as on J714s, with this board's platform ID (ADT /chosen board-id 0x16 << 16 | 0x6050), the firmware table name, the PERST# pin 80 exposed and the wifi0 and bluetooth0 aliases through which the loader fills the factory addresses. The shared K114/K116 file keeps the host disabled with its iommu-map, which aura's T6050 hand-off requires, until the K114s board ID is known. Booted on a K116s: the loader trains the link, the N1 enumerates at 01:00.0-2, apple_centauri_t6050 boots the ticketed firmware table from the 27.2 IPSW and wlp1s0f0 associates and gets DHCP. Signed-off-by: Jens Axboe <axboe@kernel.dk>
The PROLED MacBook Pros are K boards (K114s/c, K116s/c). Like the J boards, their panel node can name the board before the generic compatible: the K116s uses "apple,panel-k116", "apple,panel". Accept apple,panel-k<n> where apple,panel-j<n> is accepted. Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de>
…set, DCP Bring the K116s board file up to what the J714s describes and what the display hand-off needs: - The two non-debug USB-C ports as on the J714s: the SN201202x port controllers hpm1/hpm2 on nub-spmi-a0/a1 (the a1 controller needs its own reg), ATC1 and ATC2 PHYs, mailboxes, dwc3 controllers and DARTs enabled, with atcphy1/atcphy2 aliases so the loader copies the combo PHY tunables from the ADT. - SMC hwmon with the fan, power, voltage and current keys, without the die temperature keys this SMC lacks. - The trackpad's AFE reset as SMC GPIO gp1c (active low) and a keyboard alias for the loader's layout, as the J714s does; the pcIO node goes. - The dcp hand-off nodes the loader edits: dcp with its bandwidth scratch and firmware clock ids, fixed clocks for them, the DARTs' DMA range, the display subsystem and a panel of 345x223 mm, all disabled until chosen.aurora,t6050-display arms them. dcp and disp0 aliases. - gpio-reserved-ranges split around the N1 PERST# pin. Signed-off-by: Jens Axboe <axboe@kernel.dk>
The K116s (16-inch M5 Pro OLED) has the J714s's audio: six SN012776 amps in two woofer pairs and two tweeters, the same slots, frame and V/I sense layout, and the CS42L84 jack codec, so it shares macaudio_j714_cfg under its own chassis compatible. Signed-off-by: Jens Axboe <axboe@kernel.dk> [thomas: keep apple,k116-macaudio sorted in the binding] Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de>
The J714s's sound description, every value checked against the K116s ADT, which matches the J714s in each audio property: six SN012776 amps on i2c1 and i2c3 with the shared shutdown line on AP pin 107, the fault interrupt on pin 106 and the V/I sense slots of the ADT speaker-config rows; the CS42L84 jack codec on i2c2 with reset on pin 139 and its interrupt on 134; the AOP with its setup mailbox, DART and the base-ns ADMAC; and the card as apple,k116-macaudio with the speaker-id pins 136 and 137. The reserved GPIO ranges now expose only the pins named here plus the N1's PERST#, since pin 112 takes an SError when read. The amps carry their ADT speaker-config rows until the boot loader fills them in; it overwrites them with the same bytes. Signed-off-by: Jens Axboe <axboe@kernel.dk>
…714s does The PMP with its DART, mailbox, report and the PMC's static voters, and a tunables file generated from the K116s ADT's /arm-io/pmp0/iop-pmp0-nub (every name the J714s file has; only the DVFS voltage table, the segment ranges and the UUID differ). The SEP through its alias, which the boot loader uses to reserve its firmware and publish its manifests, with its mailbox and power domain and the two fabric parents left to the firmware. Signed-off-by: Jens Axboe <axboe@kernel.dk>
…d-off The GPU half of the J714s description, every value checked against the K116s ADT: the G17S GPU on the PMP's gfx domain, its two ASC mailboxes, the GPX/AFR/GFX power states and the six reservations the boot loader fills from the ADT. All of it stays disabled until a loader with the G17S hand-off enables it. Signed-off-by: Jens Axboe <axboe@kernel.dk>
The MacBook Pro 16-inch (M5 Pro, OLED) SD card reader, a GL9755 on PCIe
port 1 (ADT apcie0/pci-bridge1/pcie-sdreader), was not described. The
boot loader trains only port 0 (the N1) and leaves port 1 down, so the
kernel trains it, as on the J714s. Every value is from the K116s ADT:
- port 1's registers: apcie0 reg[22] (port1, 0x211028000 +0x8000),
reg[24] (phy1, 0x217014000 +0x4000) and reg[25] (intr2axi1,
0x211024000 +0x4000), the entries six after port 0's reg[16], reg[18]
and reg[19], behind /arm-io's ranges (+0x200000000);
- the shared PHY's window, apcie0 reg[3] (0x217020000 +0x4000), which
the kernel's train reads to see that the boot loader left the PHY
running;
- port 1's interrupt, apcie0 interrupts[1] = 1531;
- PERST#: pci-bridge1 function-perst = gpio0 GPIO(0x51, 0), pin 81,
which the pin controller now exposes (82-105 stay reserved, as does
pin 112, which takes an SError when read);
- the reader's power: pcie-sdreader (and /pcie-sdreader-helper)
function-sd_pwr_en = smc-pmu pKW8('pcIO', 0x0, 0x190000), described as
SMC pcIO output 0, command 0x0019000000000000. The board deleted the
pcIO node as it drove none of its outputs (the trackpad's AFE reset
uses SMC GPIO gp1c); the node is back with this one output, which
replaces the three of t6050-k114-k116.dtsi;
- the link tunables: pci-bridge1's apcie-config-tunables,
pcie-rc-tunables and pcie-rc-gen{3,4,5}-shadow-tunables, decoded as
(offset, size, mask, value). They are the same as the J714s's;
- dart-apcie1 (0x211000000, interrupt 1532), which the ADT marks
manual-availability: it answers only once the root complex is up, so
the host does not wait for it (post-init-providers) and pcie-apple
enables it once the port is up. The node is apcie1_dart of
t6050-peripherals.dtsi;
- the reader's card detect and write protect, active high: the ADT's
card-detect-polarity and write-protect-polarity are 1.
One value is not the ADT's: the DART stream. mapper-apcie1 has no reg
(macOS has the bridge request a stream at run time, function-
dart_request_sid), so the reader gets stream 1, below the DART's first
bypass stream (16), as on the J714s. The driver routes the reader's
requester ID to it.
pci-bridge1's function-clkreq (gpio0 pin 0x4d) and the ADT's GL9755
settings (gl9755-sdr104-delay*, gl9755-ssc_*, ...) are not described:
no driver takes them.
Not tested on the hardware: the DTB builds, and the same description
works on the J714s.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ADT audio-hp-mic: three microphones on leap-s RX0, through the proxy ADMAC (Linux channel 1, as on J714s and J813). The AOP audio service captures them on a card of its own, AppleK116HPAI. The board file carried this as an "#if 0" block until the proxy ADMAC was described at its real address. Enable it. Checked against the K116s ADT: admac-hp-mic-ec-proxy reg 0x2d3200000 size 0x34000, dma-channel 0; admac-leap-s interrupt 638, nine RX channels; dart-aop mapper-admac-leap-s stream 11 with vm-base-11 0x10040000000 and vm-size-11 0xffff0000; leap-s-adma-rx-0 three channels, 32 bits, 48 kHz. These are the J714s values. Not tested on the hardware yet. Signed-off-by: Jens Axboe <axboe@kernel.dk>
The 16-inch M5 Pro OLED MacBook Pro (K116s) has the T6050 ISP that J716c and J714s describe, so take their description. Checked against the K116sAP ADT of macOS 27.2 (26B8125): - isp0: reg[0] 0x380000000 + 0x3ff0000 (0x580000000 through the arm-io range) holds the coproc, mailbox and GPIO windows; reg[1] is the pmgr window 0x280900000 + 0x3f4; interrupts 513..516 and 526; the EIC's mmio-wd-isp is 0x5c3b50000 + 0x8000; clock-ids 350 and 352. - dart-isp0: instances at 0x5828c0000, 0x5828e0000 and 0x582900000 (0xc000 each), the address filter at 0x5828d0000, interrupt 531, host streams 0 and 12, exclave streams 4/5 and 8/9, remap 1, 6, 7, 10, 11, DAPF stream mask 0xcc3. - pmgr: FAB9_SOC at 0x398, ISP_SYS at 0x3a0 (parent FAB9_SOC) and ISPSENS0 at 0x100 in the second register block. ISP_CPU, ISP_FE, DPRX, ISP_SECURE, ISP_BE and ISP_CLR are devices 0xdc, 0xdd and 0xe1..0xe4; the table gives them no register, the include's offsets 0x3b0, 0x3b8 and 0x3d8..0x3f0 are eight bytes per device from ISP_SYS (0xda) and are the range the DAPF opens to the ISP (0x2809003b0 to 0x2809003f3). - the two firmware segments are 0xc1c000 and 0x13d0000 bytes at DART address 0x10000000000, as on J716c and J714s. Two things differ from J716c and need nothing here. The ADT's address filter lists four DSID windows (0x2220004000, 0x2222004000, 0x2224004000 and 0x2226004000; active-mcc-bit-vector 0xf) where the driver's T6050 table has eight; J714s, the same M5 Pro, streams with that table. The sensor type is 0xbc and cam-connections-scheme 0x10, which the loader passes as apple,platform-id and apple,cam-connections-scheme. The ISP firmware of this build (uuid 980F533B-34EA-3FBA-878C-809F6AF57345, the IPSW's isp_bni/adc-nyx-sck11pm) has one connection table for platforms 0xb9..0xbc, with a single "FrontRGB" channel on its IMX958 sensor driver: the sensor J714s reports (0x958), which the driver runs with the 2224_01 setfile. As on the other two boards the ISP and its DARTs stay disabled until the boot loader has reserved the firmware. The include does change what the power-domain driver does without that: it turns ISP_SYS on and keeps it on, registers the six ISP power states and ISPSENS0, and marks FAB9_SOC as firmware owned. Not booted on the hardware. t6050-k116s.dtb builds without DTC warnings. Signed-off-by: Jens Axboe <axboe@kernel.dk>
The boot loader copies each amp's row from the ADT (aura kboot_audio); the rows carried here were a stop-gap for a loader without it. Signed-off-by: Jens Axboe <axboe@kernel.dk>
…SMC link The shared T6050 files gained a CPU PMU node (t6050-base.dtsi) and an apple,smc phandle on the CLPC (t6050-die0.dtsi), so both became active on every T6050 board, the K116s included. Neither has been tested on the K116s. It ran at EL1 under the boot loader's monitor when they appeared, and now runs Linux at EL2 (27.2, KVM working) with this change in place, so the PMU and the SMC link are still to be tested there. Until then keep its device tree as it was: disable the PMU node and drop the SMC link, without which the CLPC's low-power platform profile has only its frequency ceiling. With this the K116s DTB differs from the one built before the rebase only in nodes that are disabled. Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de>
3f26aa3 to
cd2c9f5
Compare
|
@RyanTheTide all four done. Force-pushed as cd2c9f5 (was 3f26aa3). There are two new binding commits at the start of the series, and the 13 commits are otherwise unchanged apart from the two amendments.
dt-checks, run locally with
A CI blind spot: dt-validate crashes on 29 of the 137 Apple DTBs, the J714s and J716c among them, with
@joshuaswarren thanks for the K114c values. They agree with the K116s panel density, and the binding pattern stays as it is in this PR. |
|
A note on the dcp node in this tree, from booting the K116s on public aura today (27.2 stub). Public aura's T6050 handoff ( What I tested (aurora-silicon/aura#226, which admits the 27.2 DCP image): a copy of this DTB with only these removed:
So the tree should lose those four entries to work with aura main. I'd do that as a follow-up once aura#226 is in rather than in this PR: without #226 aura refuses the 27.2 panel either way, and a tree without the entries no longer works with the loader the rest of my testing here was done on. Happy to send it as a commit on top of this branch, or as its own PR, whichever is easier. Separately, with the external engines handed off as well (not what #226 does), |
|
@axboe thanks, agreed: a follow-up once aura#226 is in. Please send it as its own PR then, based on integration once this one has landed (or stacked on this branch if it hasn't).
I added a line on it to the description's runtime dependencies. The 16 s deferred-probe delay from dcpext0 is noted. External displays on the K116s stay out of scope until its dependencies are sorted. |
Base: integration/aurora (6ea76a5). The K116s (16-inch MacBook Pro, M5 Pro, OLED; Mac17,12) board DT, brought up to what the J714s describes. Every value is checked against the K116s ADT (macOS 27.2, 26B8125). 13 commits: 12 by Jens and one binding commit by me.
chosen.aurora,t6050-displayarms them."apple,panel-k116", "apple,panel", and the binding allowed onlyapple,panel-j<n>.macaudio_j714_cfgunder its own compatible. (J714 itself is already in integration as ef6c144.)kboot_audio); the rows here were a stop-gap for a loader without it.t6050-base.dtsi) and give the CLPC anapple,smclink (t6050-die0.dtsi). Neither has been tested on the K116s yet, so its tree keeps both off until they are. With this the K116s DTB differs from the one built before the rebase only in disabled nodes.apple,k116-macaudiocard), the PMP and SEP, the GPU (disabled until a loader with the G17S hand-off enables it), the SD card reader (GL9755 on port 1, trained by the kernel), the microphone array and the camera capture resources.One change from Jens's version: the GPU node's compatible is only
"apple,agx-g17s". Jens's DT listed"apple,agx-g17s-k116"first, which only his local board-table patch uses. His DT comment says to drop it without that patch, and the J714 tables are then used (5 mV below this unit's ADT voltages at eight of the 14 states). That patch says itself that it is a local bring-up patch: the voltages are most likely per unit, and the real fix is a loader hand-off of the ADT tables. So it is not submitted. I also removed the local file paths from the GPU dtsi comment.Runtime dependencies, which the DT does not need to build: sound needs #1645 on the 27.2 firmware; the SEP needs the SEP profile commit in #1649; s2idle with the SD reader needs #1647; the GPU needs a loader with the G17S hand-off and #1650 (on #1173). The internal panel on public aura needs aura#226 plus a follow-up that drops
/clock-dcp-firmware-361,/clock-dcp-firmware-458and theclocksandapple,bw-scratchproperties in&dcp, because aura adds those from the ADT. That follow-up comes after aura#226 merges. As it stands, the tree works with the loader Jens tested on.Tested by Jens on the K116s, with the commits as listed in their messages: N1 Wi-Fi associates, USB-C, SMC sensors, panel, speakers (with speakersafetyd), jack codec probes, PMP and SEP, camera (IMX958), and the SD reader probes (no card tested). The microphone commit says "not tested on the hardware yet", but his status of 2026-11-11 lists the 3-mic array as working.
Checks:
make dtbsbuilds;dt_binding_checkpasses for apple,macaudio.yaml and apple,dcp.yaml.dtbs_checkon t6050-k116s.dtb leaves two kinds of warning. First, gpu@2300000000 reg/reg-names (only "sgx"), the same warning the K114c DTB in integration already has. Second, the port 1 tunables on pci@1,0 (apple,apcie-config-tunables,apple,pcie-rc-*tunables): pcie-apple reads them and the J714s DT carries the same properties, butapple,pcie.yamldoes not describe them yet.Authorship. Jens Axboe wrote twelve of these commits and runs them on his K116s (16-inch MacBook Pro, M5 Pro, OLED; Mac17,12; stub macOS 27.2 26B8125). They come from his K116s bundle of 2026-11-11 (integration/aurora 254c9bd plus 34 commits, kernel 7.3.0 running on the machine). I rebased them onto integration/aurora 6ea76a5 and removed the LLM co-author and session trailers. The speaker-config and PMU/CLPC commits come from his bundle of 2026-11-12; in the second I updated the sentence that described the K116s at EL1, since it now runs Linux at EL2 with KVM. Jens has given his OK: his twelve commits carry his Signed-off-by. The binding commit is mine and carries mine.