[DjDeveloperr]

Makes the T8140 suspend-to-RAM driver an Apple SoC sleep driver. It binds only on SoCs whose sleep has been measured and tested, which today means T8140 (MacBook Neo). Every other SoC keeps s2idle and doesn't change. Stacked on #1313 (fix/t8140-sleep-review, cc859e5).

What's generic now

• The driver. drivers/soc/apple/t8140-sleep.c is now drivers/soc/apple/sleep.c, built as apple-sleep under CONFIG_APPLE_SOC_SLEEP. That covers the whole path: CPU power ops, refusals, the coprocessors' last stops, the display off, the debug-link cut, ANS's PCIe block, the macOS match, the commit, the return and the rewake.
• The old symbol. CONFIG_APPLE_T8140_SLEEP stays as a Kconfig transitional symbol, so an old .config still turns the driver on.
• What moved into match data. Each SoC has a struct apple_sleep_soc, one per compatible. It holds:
  • the power words the sleep drives by name: the DCP and DCPEXT CPU, front end and system domains, SIO_CPU, SEP, the USB complex, AP_TMM, GPIO, ANS and ANS's PCIe block;
  • the ps-aon debug switch and the pwrgate words;
  • every list macOS was measured with: all words, on and off at the commit, no-auto, the wake order, polled-off, stray and gate-off;
  • the coprocessor gate list, by DT path;
  • the wake-source names and the mask of the wake block's last word;
  • the PCIe host the radio's PERST# is found on.
• Per-device books. The books the driver keeps for each word used to be sized by T8140's tables. They are now allocated for the SoC's lists at probe.
• What stays in shared code. Block-relative layouts and protocols, each found through its own reg-names entry: the PS word bits, ASC, SPMI/HPM, WDT, DCS, MCC, AMCC, DVFM, CLLT, cluster, PTD, clk-mux, the pin controller and the APCIe block. A SoC whose block differs gets a field for it when it's added.
• AIC. The apple,t8140-aic3 check is now an aic_info.fw_suspend flag in the AIC's own match data.
• DCP. The apple,t8140-dcp / apple,t8140-dcpext checks for the sleep and DPMS power cycle and the oslog echo are now apple_dcp_hw_data.s2r / .s2r_ext.
• Binding. The file, the compatible and the properties are unchanged. The compatible is now an enum, and the title and description are generic.

How a chip is added

1. Measure its sleep under aura, as T8140's was.
2. Add <soc>_sleep_soc and { .compatible = "apple,<soc>-sleep", .data = &<soc>_sleep_soc } to apple_sleep_of_match. Add the compatible to the binding's enum and the node to the SoC's dtsi.
3. Set fw_suspend in its AIC match data. Set s2r / s2r_ext in its DCP data if the DCP's sequence holds there.
4. Once it has passed on hardware, add the compatible to apple_sleep_test_tested[] in sleep-test.c. The KUnit suite apple-sleep-soc-data checks that every list names only words of the SoC's full lists, with no duplicates, and that the lists fit the books. A mistake there would otherwise be skipped at the wake without a word.

Why untested chips are unchanged

• No probe. The match table holds only apple,t8140-sleep, and only t8140.dtsi has that compatible. Without a match the driver never probes, so there's no suspend_set_ops() and no spin-table power ops, and mem_sleep stays s2idle exactly as before.
• Flags only on T8140. The AIC and DCP flags are set only in T8140's match entries, where the compatible checks matched before. No other DT names those compatibles, even as a fallback.
• Helpers unchanged. The exported apple_sleep_* helpers are unchanged and still answer as before when no sleep driver has probed.

No functional change on T8140

• (a) rename. .text is identical to T8140 sleep: fixes from review #1313's t8140-sleep.o. The only data change is the module name string. Git shows the file as a 100% rename.
• (b1) identifier rename. .text, .rodata, .data and relocations are identical. Only the driver name string changes (apple-sleep, same length).
• (b2) match data. The new object's t8140_sleep_soc was decoded from its DWARF layout and checked against the old object and source: every table, named word, pwrgate word, count, mask and string, and the of_match data pointer. 50 of 50 checks match.
• (c) AIC and DCP. Each flag is set exactly where its compatible check matched before.

Not generalized, on purpose

• rtkit.c oslog echo. It's an of_machine_is_compatible("apple,t8140") check, but it's general RTKit behaviour, not sleep.
• pmp-export.c apple_pmp_keep_bound(). This is a driver-wide suppress_bind_attrs, read before any device or match data exists. A helper that depends on the sleep config would change T8140 kernels built without it.
• dwc3-apple GCTL hibernation setup. It's controller reset programming, run at every reset, and the driver has no match data.
• ATC PHY has_t8140_dp. It's a PHY variant flag. Its sleep-in-place also needs apple_sleep_domain_check(), which fails without the sleep driver.
• macsmc's MBSW wake-reason bits. They use T8140's numbering, but they're reached only after a firmware suspend, so only on matched SoCs. A new SoC must check them.
• Already generic. Everything else the series touches is already keyed on DT properties (apple,sleep-ssps, apple,sleep-ddr, apple,sleep-driver-owned), on pm_suspend_via_firmware(), on the sleep driver's registration, or on its own match data (NVMe sleep_traced, GPU sleep_restart, SIO firmware_only, EDAC t8140_ni).

Visible renames

• The platform driver is now apple-sleep, at /sys/bus/platform/drivers/apple-sleep/, and its log prefix is apple-sleep.
• The dyndbg module name is now apple_sleep.
• The distro configs' symbol moves to CONFIG_APPLE_SOC_SLEEP: alarm-aurora#199 and nixos-aurora#203 get a follow-up commit, which also takes T8140 sleep follow-ups: RTC wake on rtc0, SEP restart after a refusal, commit audit, USB recovery #1333's CONFIG_RTC_DRV_APPLE_PMU=y. fedora-aurora#242 needs nothing, because it validates against the pinned asahi.config.

Tests

• CI on this PR passes: source, rust-build and kunit.
• Build. Asahi profile with Rust on: drivers/soc/apple/ drivers/irqchip/ drivers/gpu/drm/apple/ drivers/gpu/drm/asahi/ drivers/dma/ drivers/mfd/ drivers/nvme/host/ drivers/pci/controller/ drivers/phy/apple/ drivers/pmdomain/apple/ drivers/usb/dwc3/ drivers/usb/typec/tipd/ plus the t8140-j700, t6000-j314s, t8112-j413, t6041-j614c and t6040-j616s DTBs. Exit 0. The only warnings are the known apple.c thread-safety ones and rustc deprecation notes, none from this series. sleep.o is also clean with W=1, and checkpatch is clean.
• KUnit (QEMU arm64, the stack's config with CONFIG_APPLE_SOC_SLEEP=y and the new suite):
  • Run 1: 532 tests, 531 passed, 1 skipped (apple_g17_rx_util_min_test, which needs UCLAMP). That's T8140 sleep: fixes from review #1313's 528 plus the 4 cases of apple-sleep-soc-data.
  • Runs 2 and 3 ran under load average 19-22 from other lanes' kernel builds. Both oopsed in the upstream DRM scheduler mock tests (drm_sched_parallel_submit_test cleanup: mock_sched_cancel_job() -> hrtimer_cancel(), a NULL dereference). That broke the KUnit device bus for every later suite. This series doesn't touch drivers/gpu/drm/scheduler.
  • Run 5 ran every other suite of the same build by name glob: 490 tests, 489 passed, 1 skipped.
• Negative control. With a word added to T8140's wake order that its full list lacks, apple-sleep-soc-data fails on exactly that word.
• Host tests (tools/asahi/run-host-tests.py): apple-dcp 87/87 (through the clang wrapper with -Wno-unused-but-set-global, as on T8140 sleep: fixes from review #1313), display26 1/1, t6030-wiring 3/3, tunnel-route 1/1, tiled-split 2/2 (with mutations), t6030-clock 1/1, rtkit-oslog 6/6, atc-usb3 2/2 (with mutations), neo-stability 10/10, apple-dart-fw 1/1, apple-dart-range 1/1, t8140-external-bw 1/1. apple-dcp's resource-profile copies the DCP's hw data struct, so its mock gets the two new fields in the DCP commit.
• DT. dt_binding_check on the binding passes. dt-validate of t8140-j700.dtb gives the same lines with the old and the new binding, and none for the sleep node.

Merging with #1333 (sleep follow-ups)

#1333 (fix/t8140-sleep-followups, a078d77) is also stacked on #1313. The rename commit here is a pure 100% move, so git carries #1333's changes to t8140-sleep.c over to sleep.c. A trial merge (git merge-tree):

• One conflict hunk. The comment above *_enclave_tell(): take T8140 sleep follow-ups: RTC wake on rtc0, SEP restart after a refusal, commit audit, USB recovery #1333's comment and this branch's apple_sleep_enclave_tell(struct apple_sleep *s).
• One rename to fix after the merge. T8140 sleep follow-ups: RTC wake on rtc0, SEP restart after a refusal, commit audit, USB recovery #1333's new apple_sleep_commit_stops_coprocs() reads t8140_sleep. It merges cleanly but doesn't build until it reads apple_sleep.
• Merges cleanly. include/linux/soc/apple/sleep.h (its new declaration and stub land inside this branch's CONFIG_APPLE_SOC_SLEEP block) and asahi.config (RTC_DRV_APPLE_PMU=y and APPLE_SOC_SLEEP=y are separate lines).
• No overlap. T8140 sleep follow-ups: RTC wake on rtc0, SEP restart after a refusal, commit audit, USB recovery #1333's rtc, sep, pm, g17, atc, dwc3 and pmp-export changes touch no file this branch changes.
• Trial build. The trial merge, resolved that way, builds: soc/apple, rtc, dwc3, usb/host, phy/apple, drm/asahi, drm/apple, irqchip and the j700 DTB, exit 0, with no new warnings.

🤖 Generated with Claude Code

[DjDeveloperr]

Superseded by #1445 (merged as 2b05915); its changes are in integration/aurora. The generic driver is drivers/soc/apple/sleep.c.