Skip to content

soc: apple: mcc: let the T6050 system level cache allocate - #2518

Open
satchlj wants to merge 1 commit into
integration/aurorafrom
satchlj/mcc-t6050-slc-allocate-20261120
Open

satchlj wants to merge 1 commit into
integration/aurorafrom
satchlj/mcc-t6050-slc-allocate-20261120

Conversation

@satchlj

@satchlj satchlj commented Nov 20, 2026 •

Copy link
Copy Markdown
Member

Claude Opus 5.5 on @satchlj's behalf.

soc: apple: mcc: let the T6050 system level cache allocate. The binding and J775D's device tree already describe apple,t6050-mcc, but apple-mcc has no match for it, so the boot loader's state stays: every plane's ways are enabled and powered with allocation disabled, and the cache holds no data.

  • Match: apple,t6050-mcc, using the T8140 sequence unchanged.
  • MCC_MAX_PLANES goes from 32 to 64 (two dies × eight AMCCs × four planes). The binding already allows 64.
  • Boot DMA: allocation still waits for every device listed in apple,boot-dma-devices to bind. A disabled display node can still be scanning out the boot framebuffer. If the loader lists a disabled client, allocation defers until the DT correctly describes the handoff.

This is based on #896 by @funkzy, which was closed when the change moved into the J775D series branch. The J775c DT side, a die-0 MCC node with 32 planes, is in #2102.

Evidence

  • J775D (from soc: apple: mcc: let the T6050 system level cache allocate #896, @funkzy, 2026-11-03): "system level cache allocation enabled on 64 planes". A scratch-heavy 512×512 GPU render in 2×2 passes took 51.5–55.8 ms per submit with allocation, against 202–273 ms without.

  • J775c, an earlier private lobata kernel (2026-11-05), toggling the same register at run time: GPU read at 32 MiB went from 583 to 1014 GB/s and write from 551 to 942 GB/s. 64 MiB and larger working sets stay at DRAM speed.

  • J775c CPU memory latency, tinymembench random read:

    Block size Default kernel (has this change) Candidate, integration/aurora 26294e8 + 68, without it
    16 MiB 0.9 ns 6.6 ns
    32 MiB 8.0 ns 24.5 ns
    64 MiB 18.1 ns 63.3 ns

    That fits a system level cache that holds nothing. It is not yet a same-boot A/B. The default logs system level cache allocation enabled on 32 planes and also gives the GPU's DSID groups a quota, which this PR does not.

Testing

  • The revised head builds drivers/soc/apple/mcc.o for ARM64 in neo-build. Hardware validation is pending, including J775c allocation, per-plane readback, the state of boot scanout, and a regression boot on T604x or T8140.

Assisted-by: LLM (Claude:Opus 5.5, Codex:GPT-6.2)

@satchlj

satchlj commented Nov 20, 2026

Copy link
Copy Markdown
Member Author

Claude Opus 5.5 on @satchlj's behalf: J775c (Mac Studio, M5 Max) row for this head: the cache allocates, and the CPU rows return to within about 2% of the default.

Kernel: lobata's cand-3, lobata/candidate3-20261120 8c4c3a2d690c. That is integration/aurora 61295c26102d (0 commits behind, 2026-11-20) + 80 commits, including this PR at 163de139e5 and #2102's J775c DT with its die-0 MCC node (32 planes). Booted 2026-11-20 04:57Z, with no SD-reader guard.

  • Boot log: apple-mcc 2220000000.memory-controller: system level cache allocation enabled on 32 planes at 0.68 s. arm64: dts: apple: support M5 Max Mac Studio (J775c) #2102's display node is disabled, so the new skip is what lets it get there.
  • Same scripts, cross-boot comparison:
    • Previous candidate (cand-2d, 2026-11-20): 26294e82 + 68, without this PR.
    • Default: lobata's default kernel, which allocates on 32 planes and also gives the GPU's DSID groups a quota.
Row cand-2d, no SLC cand-3, this PR default
tinymembench random read, 16 MiB 6.6 ns 3.2 ns 0.9 ns
32 MiB 24.5 ns 8.7 ns 8.0 ns
64 MiB 63.3 ns 19.0 ns 18.1 ns
7-Zip 1T pinned to a P core (cpu0) 8349 9662 9925
7-Zip 1T unpinned 11171 12716 13009
7-Zip 18T 174017 185607 189043
7-Zip 1T on cpu0: IPC at ~4.38 GHz – 1.553 1.613

The remaining 2-4% is IPC at the same clock, and the 16 MiB latency is still higher than the default's. Rows still to come here: GPU bandwidth by working-set size, and a same-boot A/B (allocation toggled at run time).

🤖 Generated with Claude Code

@funkzy

funkzy commented Nov 20, 2026

Copy link
Copy Markdown
Contributor

Tested on the Mac Studio (M5 Ultra, J775D): this commit, unchanged, on top of integration with the J775D device tree from #1973, against the same tree without it. 27.2 firmware, booted through the display handoff.

  • system level cache allocation enabled on 64 planes appears about 5 ms after the display driver binds; without the commit there is no cache line.
  • On the J775D the display node is enabled by the boot loader, so the new skip does not fire here: the cache still waits for the display, as before.
  • Single light compute submit: 221–223 µs without, 149–150 µs with (three passes each).
  • Single render submit: 260 µs without, 174–176 µs with.
  • Clean boots on both, with no GPU errors.

The 1.5x on single submits matches what #896 measured on the J775D.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Blockers: 1.

  1. The commit message carries Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>. Drop that trailer and keep only the human Signed-off-by. The PR body's last line should be Assisted-by: LLM (Claude:Opus 5.5); it currently ends with the "Generated with Claude Code" footer.
  2. drivers/soc/apple/mcc.c:152: the skip treats status = "disabled" as "the boot loader left this device no DMA". The binding (apple,t8140-mcc.yaml:113) defines the list as devices that take over DMA the boot loader left running. A disabled node only means Linux won't drive the device, and the boot loader may still have it scanning out. On J775c the J775c row shows the cache allocating, but not whether the boot loader's display was still active when it did. Either show that it wasn't (no boot framebuffer or scanout on that port), or note in the commit message that allocation under a still-running boot DMA is harmless on this cache and why. The comment at line 149-151 should then say the same.

The rest checks out. MCC_MAX_PLANES only sizes planes[] and the probe bound (mcc.c:42, :281); no plane bitmask or u32 index caps at 32. of_for_each_phandle with continue drops the node reference correctly. The apple,t6050-mcc match on the block's own compatible is fine.

HW row: J775c, same candidate kernel. Boot log with the allocation line, the plane readback (allocation-disable word 0 on all 32), and whether a boot framebuffer was live at that point. One T604x or T8140 boot showing an unchanged log line where the boot-DMA node is enabled. 0 AVC, 0 failed units, 0 new kernel errors.

Assisted-by: LLM (Claude:Opus 5.5)

The T6050 memory cache controller has up to 64 planes: eight AMCCs
of four planes per die, on up to two dies. The binding already
describes this layout, but apple-mcc has no match for the controller.
The boot loader leaves the ways enabled and powered with allocation
disabled, so a miss in an agent's own caches goes to DRAM.

T6050 uses the T8140 sequence: check each plane's ways and flush
engine, clear the allocation-disable word and read it back. On
shutdown, removal or deep sleep, disable allocation and flush. Add
the match and raise MCC_MAX_PLANES to the binding's limit of 64.

Wait for a driver to take over every client the DT lists in
apple,boot-dma-devices. A disabled display node can still be scanning
out the boot framebuffer, so its status cannot establish that it is
safe to enable allocation. The loader must drop a client from the
list only once it knows that client's boot DMA is quiescent.

Based on #896 by Jeff Queisser, which measured the effect on J775D.

Signed-off-by: Satya Benson <satya@benson.earth>
@satchlj
satchlj force-pushed the satchlj/mcc-t6050-slc-allocate-20261120 branch from 163de13 to e6fdaf6 Compare November 21, 2026 12:34
@satchlj

satchlj commented Nov 21, 2026

Copy link
Copy Markdown
Member Author

Codex GPT-6.2 Sol on @satchlj's behalf

Kept the existing wait for every client listed in apple,boot-dma-devices and documented why in the driver and commit message. Lobata's earlier candidate reports allocation on 32 planes, but the logs do not establish that every boot-DMA client was idle at allocation. A disabled client still listed by the loader will defer MCC probe. The commit carries the human sign-off, the PR footer is updated, and the ARM64 mcc.o build passed in neo-build.

The revised head is untested on hardware. J775c needs the loader handoff and boot-scanout state checked alongside allocation and plane readback. The T604x or T8140 boot and regression counts also remain pending.

@satchlj

satchlj commented Nov 21, 2026 •

Copy link
Copy Markdown
Member Author

Codex GPT-6.2 Sol on @satchlj's behalf

J775c hardware row for exact head e6fdaf6d4b23 on 26294e82e9c8. Two 27.0 Aura one-shot boots used the J775c carrier DTB (682476cf49f1…) and the same Image (8dd129302637…). The trial config disabled PCI SDHCI.

  • Both boots logged system level cache allocation enabled on 32 planes. The repeat independently read all 32 allocation-disable registers at plane offset 0x82c: 32/32 were 0x00000000.
  • The live apple,boot-dma-devices list held one client, the enabled /display-subsystem bound to apple-drm. No disabled client was listed. There was no simple framebuffer node, and DCP logged no boot framebuffer. HDMI was active later under Linux. Physical scanout at the allocation instant was unmeasured.
  • Each boot had 0 AVC denials, 1 failed unit (ac-power-mode.service), and 3 boot error lines absent from Aura (two for Gen4 SPMI IRQ rejection and one for AOP probe -61). No MCC error was logged. The repeat hit an RCU stall during shutdown with CPU10 unresponsive to NMI. The first reboot and the repeat both returned to verified default Aura, with offline ESP checks and peer connectivity passing. These boots leave the shutdown stall's cause open.

T604x and T8140 boot-DMA handoffs remain untested here.

Scope follow-up: the exact tested integration base was 26294e82e9c87a0162a7973c5e23c599ceaea22f, 370 commits behind integration/aurora a84637dca209 when checked on Nov 21. Its changed files merge cleanly in a text-only check against that tip. The merged tip has no J775c runtime row here. Trial differences from the installed boot were PCI SDHCI off, arm64.nomte, console=tty0 and console=ttyDC0, with the J775c carrier DTB.

The shutdown stall was seen on 1/2 exact-head runs. The #2483 parent used the same integration base, config and DTB and reached Restarting system with no stall in its captured console. It used a different workload and shutdown timing. Attribution to the base or this PR remains open.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Re-review of e6fdaf6: Blockers: 1.

From our last comment:

  • Co-Authored-By trailer: resolved, but see 1 below.
  • Disabled boot-DMA node treated as quiescent: resolved. The skip is gone; drivers/soc/apple/mcc.c:144-157 now defers until every client in apple,boot-dma-devices is bound, whatever its status, and the comment at :150 and the commit message say why.
  • PR body footer: resolved.
  1. The commit message now ends with Assisted-by: LLM (Claude:Opus 5.5, Codex:GPT-6.2). Keep that line in the PR body only; the commit should carry just the human Signed-off-by.

The rest of the diff is unchanged from the version reviewed (match on the block's own apple,t6050-mcc, MCC_MAX_PLANES 64 at mcc.c:37). The J775c row covers allocation, the 32/32 allocation-disable readback and the handoff state.

Nit: the PR body's Testing section still says hardware validation is pending; point it at the J775c row.

HW row: the exact head now runs the shutdown flush on T6050 (apple_mcc_shutdown -> apple_mcc_disable_planes, mcc.c:178-193 and 319-325), which the parent never ran on J775c, and 1 of 2 runs hit an RCU stall at shutdown. Reboot this head and its parent five times each on J775c with the console captured. Pass: no stall and no "flush timed out" on this head, or the stall shows up on the parent too. Plus one T604x or T8140 boot where the boot-DMA client is enabled: the allocation line is unchanged, 0 AVC, 0 failed units and 0 new kernel errors against its parent.

Assisted-by: LLM (Claude:Opus 5.5)

@DjDeveloperr

Copy link
Copy Markdown
Contributor

J714s (MacBook Pro 14-inch, M5 Pro, T6050) hardware row for the 16-plane T6050 apple-mcc enable path:

  • Live ADT & pre-kernel AMCC state: /arm-io/mcc has amcc-count = 8, amcc-enable-mask = 0xf (4 enabled AMCCs at 0x2220000000, 0x2222000000, 0x2224000000, 0x2226000000 × 4 planes, stride 0x40000 = 16 planes, matching /chosen/lock-regs/amcc). All 16 planes read +0x82c = 0x00000001, +0x2800 = 0x8000000c, +0x2804 = 0x0c000c00 (12 data ways, 12 tag ways), +0x2844 = 0x00000000 before Linux boots.
  • Test setup: 7.3.0-m5-sship7 (CONFIG_APPLE_MCC=y; drivers/soc/apple/mcc.c runs the same 16-plane enable/flush path as this PR when matched via a temporary "apple,t8140-mcc" fallback alongside "apple,t6050-mcc" in test DTB 24cbf3f8517e derived from default c0440c9e), apple,boot-dma-devices = <&dcp> (dcp@48ae00000).
  • Boot & shutdown: apple-mcc 2220000000.memory-controller: system level cache allocation enabled on 16 planes. Note that without soc: apple: mcc: wait for display controllers to retire boot scanout #2538/drm/apple: publish completion of native display takeover #2539, apple-mcc enables right after simple-framebuffer / platform-device bind before apple-dcp finishes async RTKit startup. Shutdown reboot (apple_mcc_shutdown flush on 16 planes) completed cleanly with no stall or timeout.
  • Measured A/B (same machine, before vs after vs macOS 27.0 26A428):
    • P-core (cpu12) latency (ns/load): 12 MiB 29.3 -> 14.7 (macOS 14.0), 16 MiB 54.3 -> 28.7 (macOS 25.1), 24 MiB 101.0 -> 46.7 (macOS 44.4), 32 MiB 105.4 -> 56.7 (macOS 66.7), 48 MiB 110.6 -> 79.4 (macOS 83.4), 64 MiB 110.7 -> 91.7 (macOS 113.0).
    • M-core (cpu0) latency (ns/load): 12 MiB 106.6 -> 46.6, 16 MiB 104.2 -> 47.6, 24 MiB 110.2 -> 58.8, 32 MiB 109.2 -> 75.1.
    • P-core read bandwidth: 16–32 MiB 94.4–95.8 -> 105.7–107.7 GB/s (macOS 103.7–106.4 GB/s).
    • Borealis Vulkan gpubench-vk --test compute.light (depth 1, 3×100 ms): us_per_submit median 156.99 -> 104.82 µs (-33.2%, 1.50x).
  • Regression: 0 AVC denials, 0 new failed units (tailscaled.service baseline only), 0 new kernel errors (13 vs 17 baseline), 0 MCC warnings/errors, GPU live throughout.

Assisted-by: LLM (Claude:Opus 5.5, Google:Gemini 4 Neon)

@DjDeveloperr

Copy link
Copy Markdown
Contributor

J714s (MacBook Pro 14-inch, M5 Pro, T6050) deep-sleep and s2idle suspend/resume row for the 16-plane apple-mcc path (#2518 e6fdaf6d4b23 + #2708 3481265d9edd):

  • Setup: 7.3.0-m5-sship7 (CONFIG_APPLE_MCC=y, test DTB 24cbf3f8517e derived from c0440c9e + #2708 with temporary "apple,t8140-mcc" test fallback). Boot log: apple-mcc 2220000000.memory-controller: system level cache allocation enabled on 16 planes.
  • 2× real deep sleeps (mem_sleep=deep, +30 s PMU RTC alarm each):
    • Both returned rc=0 in the same boot_id (~28.4 s and ~28.6 s asleep, SMC wake MBSE 0x10650), with 0 apple-mcc warnings/errors (0 flush timeouts across all 16 planes, 0 -EBUSY / not ready / stays off after the sleep).
    • 128 MiB dirty-memory SHA-256 across each sleep (64 MiB anon + 32 MiB anon freshly dirtied on every 128 B cache line right before suspend + 32 MiB tmpfs): 3/3 MATCH on both wakes.
    • Post-wake SLC remains active: P-core (cpu12) 16 / 24 / 32 MiB latency 27.77 / 44.69 / 59.32 ns (pre) -> 24.86 / 45.10 / 60.13 ns (wake 1) -> 28.28 / 44.50 / 58.33 ns (wake 2) vs 54.27 / 100.96 / 105.41 ns SLC-off; M-core (cpu0) 16 / 24 / 32 MiB 47.54 / 58.08 / 73.80 ns -> 47.29 / 51.89 / 70.18 ns -> 47.53 / 58.00 / 77.78 ns; Borealis gpubench-vk compute.light 105.60 -> 105.37 -> 105.37 µs/submit (vs 156.99 µs SLC-off), GPU live on all checks.
  • s2idle check (mem_sleep=s2idle, +15 s PMU RTC alarm): entry aborted in pcie-apple 1c70000000.pcie (genpd_suspend_noirq -16 / -EBUSY, unrelated to apple-mcc, which no-ops on !pm_suspend_via_firmware()). 0 apple-mcc warnings/errors, 128 MiB SHA-256 3/3 MATCH, P-core 16 / 24 / 32 MiB stayed 28.04 / 44.55 / 60.50 ns.
  • Regression: 0 AVC denials, 0 new failed units (tailscaled.service baseline only), 0 apple-mcc warnings/errors across the boot. T6050 wake ROM restores all 12 ways on all 16 planes across deep sleep, so mcc->ways_off = false (matching ADT skip-mcc-off-to-s2r = 1) works without any resume way-restore change in mcc.c.

Assisted-by: LLM (Google:Gemini 4 Neon)

@satchlj

satchlj commented Nov 21, 2026

Copy link
Copy Markdown
Member Author

Claude Opus 5.5 on @satchlj's behalf

J775c (Mac Studio, M5 Max) row for #2518, at head e6fdaf6d, along the lines of the s2idle checksum row asked on #2708.

Setup.

Method.

  • 48 GiB of 62 GiB was held under SHA-256 as address-unique 64-bit words.
  • 4 GiB of it was rewritten right before each sleep.
  • Three real rtcwake -m freeze cycles of 300, 20 and 20 s were driven from a separate process while the holder blocked.
  • cpu0 random-read latency (tinymembench) was measured before and after.
with #2518 without
apple-mcc allocation enabled on 32 planes not bound
300 s cycle wall 303.9 s, RTC wake, 48 GiB match wall 304.3 s, RTC wake, 48 GiB match
two 20 s cycles match, match match, match
16 MiB latency, before → after 10.0 → 10.5 ns 24.6 → 24.8 ns
32 MiB latency, before → after 13.6 → 13.4 ns 31.9 → 31.8 ns
MCC warnings / SELinux denials / failed units 0 / 0 / 0 0 / 0 / 0

So on J775c the SLC allocation and its latency benefit survive real suspend-to-idle, and dirty data came back intact on every cycle.

Limits.

@satchlj

satchlj commented Nov 21, 2026

Copy link
Copy Markdown
Member Author

Claude Opus 5.5 on @satchlj's behalf

Corrections to my J775c row above (#2518 (comment)):

  • Command line. Not exactly the default. The trial cmdline was the installed default (rdinit=/init nr_cpus=18 maxcpus=18 root=UUID=…, which is what lobata's default kernel boots with) plus arm64.nomte, which our one-shot trial helper always appends.
  • The DTB. I wrote that "without it this stack doesn't return from RTC sleep" as if it were settled. It isn't: that claim is the open arm64: dts: apple: t6050: keep the ATC USB power domains on in system sleep #2592 question. What I have is that this stack with its stock DTB returned from 0 of 1 RTC cycles, and with apple,sleep-retained on the ATC USB domains from every cycle tried.
  • The two arms were separate Image builds of the same source and .config, differing only by this PR (with: built 14:23 EST; without: built 10:35 EST).
  • The 300 s cycle was a reliability check. The functional rows are the two 20 s cycles.

The numbers in the table are unchanged.

@funkzy

funkzy commented Nov 22, 2026

Copy link
Copy Markdown
Contributor

J775D (M5 Ultra Mac Studio) row for this PR at e6fdaf6.

Setup: integration/aurora e0ee46a with #1650 and #1973, which the J775D needs to run its GPU, plus local GPU firmware-build configuration that is not public yet, plus this PR. The same Mesa build and benchmark on every boot. 27.0 stub, read-only root, one-shot boots.

  • Allocation: system level cache allocation enabled on 64 planes at 4.38 s, about 1 s after both display controllers bound. The display node is the J775D's only apple,boot-dma-devices entry, so the new wait for it changes nothing visible here. The earlier head, 163de13, enabled at the same point (3.8-4.2 s).
  • GPU: a single-triangle render with one submit in flight takes 175.1-176.7 us per submit with this PR on die 0 and 174.6-177.2 us on die 1. On the same base without it (boots on 11-20 and 11-21) it took 259-261 us. 163de13 gave 173-176 us.
  • Regression: no failed unit or kernel error line comes from this PR. The two failed units are the read-only root's iwd and networkd persistent-storage units, as on every boot of this root. One usb 3-1 ... error -108 enumeration line recovered within 0.4 s; it also appears on boots without this PR.

Not run: a boot without the display handed over (the boot path here cannot switch that off), and suspend with the cache enabled.

Assisted-by: LLM (Claude:Opus 5.5)

@RyanTheTide

Copy link
Copy Markdown
Contributor

Merge-train adversarial review (MERGETRAIN-LINUX2) at e6fdaf6: NO BLOCKERS.

Code read only, nothing built here. Before the merge it still needs Dj's verdict at this head; the missing rows are listed below.

Full review

#2518 soc: apple: mcc: let the T6050 system level cache allocate @e6fdaf6d

VERDICT: NO BLOCKERS

Checked: MCC_MAX_PLANES 64 (mcc.c:37) only sizes planes[] and the probe bound (:271 base).
The KUnit fixture (mcc-test.c:50) uses it only as an out-of-range "reject none" sentinel, so it
is still valid. The match is on the block's own apple,t6050-mcc, which the binding allows
standalone (apple,t8140-mcc.yaml oneOf enum), so no SoC gate is added. The disabled-node skip
from earlier revisions is gone. The only change to apple_mcc_wait_boot_dma() is a comment
(:150), so it still defers until every listed client is bound. On 24dc422 the only DT with
the node is t6050-ultra.dtsi:443 (J775d, 64 planes, boot-dma-devices = <&display>).
J714s/J716c/J775c have no MCC node in the base (J775c's comes with #2102, still OPEN). So on
merge the PR is live on J775d only. Commit hygiene: only Signed-off-by: Satya Benson; the
last re-review's "Assisted-by in the commit" point is not true at this head (it is resolved).

Nits:

  • mcc.c apple_mcc_disable_planes(): readl_poll_timeout_atomic(..., MCC_FLUSH_TIMEOUT_US) per
    plane, sequentially. With 64 planes the worst case is 6.4 s of atomic busy-wait at shutdown
    or suspend_noirq. That is pre-existing code, but this PR doubles the bound. Consider one
    shared deadline across the planes.
  • The PR body "Testing" still says hardware validation is pending. It should point at the
    J775c/J775D/J714s rows.
  • Pre-existing, not added here: the base mcc.c comments name a macOS kext class and method
    ("AppleT8142MCC's sleep action" at :46, "mccEnableCacheMode" at :194). That should be a
    separate cleanup.

Missing rows:

  • J775c shutdown stall: 1 of 2 exact-head reboots hit an RCU stall (CPU10 unresponsive). Five
    reboots on this head and five on its parent, with the console captured, as Ryan asked. Still
    open, and still the main risk.
  • J775d (the only DT live on merge): funkzy's row is "this commit unchanged" on integration +
    arm64: dts: apple: t6050-j775d: Give the Mac Studio (M5 Ultra) a working GPU, and its M and P core capacities (754/1024) to the scheduler #1973, not this head on 24dc422. It needs a reboot/shutdown loop (the flush now runs on 64
    planes) and a deep-sleep row if J775d sleeps.
  • One T6040 (J616s) or T8140 (Neo) regression boot: allocation line unchanged, 0 AVC, 0 failed
    units, 0 new kernel errors.
  • J714s rows (DjDeveloperr) exist for boot, shutdown and 2x deep sleep, but through a temporary
    apple,t8140-mcc fallback compatible, not this match. They cover the code path, not the DT
    that will ship.

@DjDeveloperr DjDeveloperr left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pre-PR checklist review at e6fdaf6d4b23 (code, policy, attribution, CI: 0 blockers; holding APPROVE for §5 hardware rows).

Checklist (§1–§6)

  • §1 No model gating: Pass. Matches standalone apple,t6050-mcc in apple_mcc_of_match[] (matching Documentation/devicetree/bindings/soc/apple/apple,t8140-mcc.yaml:78); no chip/board/firmware gate.
  • §2 Production-ready / code review: Pass. Raising MCC_MAX_PLANES from 32 to 64 (drivers/soc/apple/mcc.c:37) only sizes planes[] and the probe upper bound (mcc.c:271), and mcc-test.c:50 uses it solely as an out-of-range sentinel (no 32-bit plane bitmasks exist). Removing the earlier disabled-node skip keeps apple_mcc_wait_boot_dma() (mcc.c:143-157) strictly waiting for every listed client, which is clean both standalone and as the base for #2538–#2540. apple_mcc_probe() leaves retained and ways_off false for apple,t6050-mcc, matching T6050 wake behavior.
  • §3 Safety: Pass. Plane readiness is checked before clearing MCC_ALLOC_DISABLE, failures unwind all touched planes, and boot-DMA handoff deferral is intact.
  • §4 Clean-room & attribution: Pass. Commit e6fdaf6d4b23 has human author/committer and human Signed-off-by: only (no AI trailers in commit). PR body ends with Assisted-by:.
  • §6 Naming: Pass.
  • CI: All 7 checks green (host-tests, source, rust-build, kunit, and all three DWC3/ATC build matrices).

Remaining §5 hardware rows before APPROVE

  1. J775c shutdown RCU stall isolation: 1/2 exact-head (e6fdaf6d4b23) reboots hit an RCU stall on shutdown (CPU10 unresponsive to NMI), where this PR newly runs apple_mcc_shutdown() -> apple_mcc_disable_planes() across 32 planes. Please post a 5× reboot loop on this head vs 5× on its parent with console capture showing no stall / no flush timed out attributable to this PR (and ideally a clean default-cmdline / stock-DTB regression count).
  2. Non-T6050 regression boot: One T6040, T8140, or T8132 boot confirming the existing apple-mcc path logs the unchanged allocation line with 0 AVC denials, 0 failed units, and 0 new kernel errors. (J775D 64-plane allocation/GPU speedup at e6fdaf6d and J714s 16-plane enable/shutdown/2× deep-sleep rows cover T6050 well; a multi-reboot check on J775D's 64-plane shutdown flush is also worth confirming since t6050-ultra.dtsi:443 is the live in-tree consumer.)

Nits (non-blocking)

  • apple_mcc_disable_planes() polls MCC_FLUSH_TIMEOUT_US (100 ms) sequentially per plane; at 64 planes the worst-case atomic busy-wait bound doubles to 6.4 s. Consider a shared deadline in a follow-up.
  • The PR description's ## Testing section still says hardware validation is pending; updating it to point at the J775c/J775D/J714s comment rows would keep the summary current.

Assisted-by: LLM (Gemini 4 Neon)

@funkzy

funkzy commented Nov 22, 2026

Copy link
Copy Markdown
Contributor

J775D (M5 Ultra Mac Studio) row for e6fdaf6 on 24dc422, the PR's own base: integration/aurora 24dc422 plus this commit only, nothing added to boot. Baseline: 24dc422 alone. Both one-shot boots with read-only root, on the kit's standard command line (arm64.nomte nr_cpus=30 maxcpus=30 idle=nop arm64.nowfxt, the dockchannel console) plus a watchdog dead-man (watchdog.open_timeout).

row 24dc422 24dc422 + e6fdaf6
mcc apple,t6050-mcc node (64 planes) unbound, no lines system level cache allocation enabled on 64 planes at 3.6 s, 1.0 s after both DCPs bound; no deferral or "not ready" lines
reboots 3 of 3 consecutive reboots back into this kernel, each reaching userspace; allocation enabled on 64 planes again every time
shutdown one poweroff, clean on the console
failed units 2 (iwd, systemd-networkd-persistent-storage) the same 2 on every boot
kernel err-level lines 4 3-4; 0 new against the baseline (the extra line on two boots is the USB enumeration retry the baseline shows too)
WARN/BUG/Oops 0 0 on every boot and in every shutdown

ALARM has no SELinux, so there is no AVC count.

Shutdown flush: I read the console of each outgoing kernel with initcall_debug set at run time, so that device_shutdown() names each device. In every captured reboot and in the poweroff, apple-mcc 2220000000.memory-controller: shutdown is followed by the next device's line within 1 ms, after allowing for a printk delay I set so the console kept up. No plane N flush timed out, RCU stall, hung task or timeout lines. The two complete reboot captures end in reboot: Restarting system, and the poweroff ends in reboot: Power down. The first reboot's console was cut at the reset after the mcc line, but that boot came back normally.

Deep sleep: no row on this base. rtc0 has no wakealarm on 24dc422, so there is no RTC wake source to arm. /sys/power/mem_sleep offers only [s2idle], and since the suspend hook acts only when pm_suspend_via_firmware(), the J775D doesn't reach the sleep flush path even with s2idle working.

Assisted-by: LLM (Claude:Opus 5.5)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants