[DjDeveloperr]

After a Linux sleep, the host's debug-USB reboot VDM sometimes got no answer from the MacBook Neo (J700), and the host's PD controller stayed busy on it until the cable was replugged.

Cause. Before each sleep, t8140-sleep cuts the debug link with a cold reset of the debug port's controller (sn201202x 1-0c). tipd then rearms it on resume. Two things went wrong:

• The controller sometimes refused a rearm command (-EPERM). The sequence stopped right there, leaving the controller in its "DBMa" debug mode, unlocked (its mode register read DBMa instead of APP ). The Debug USB stayed down, and the reboot VDM went unanswered for the rest of the boot.
• In about 1 in 25 cuts, the partner never got a power contract back after the reset. It stayed attached with no PD, while the controller saw nothing attached. No rearm or VDM could get through until something made the port attach again.

Fix.

• tipd: after entering debug mode, wait for the mode register before enabling the debug VDM, as the boot loader does. Leave debug mode and lock again after a failed step too. Repeat a failed rearm every 250 ms for up to 10 s from a delayed work, and keep it pending across a sleep that comes first.
• t8140-sleep: after the cut's reset, poll the controller's power status for a PD contract (up to 5 s) and log how long it took. If no contract comes, reset again, up to 3 times.

Hardware (J700). Stock DTB, production command line, SEP active. Each sleep was RTC-woken; the first of a boot is deep, the rest keep DRAM.

• The final kit with Deep sleep (suspend-to-RAM) for the MacBook Neo (T8140) at EL2 #1208's head e544085 merged locally: 17 sleeps, all returned. NVMe, USB hub and keyboard, Wi-Fi, display and mics were OK after every wake.
• 8 VDM reboots, each sent after deep and/or DRAM-kept sleeps on these kits: every one reached the loader, Debug USB enumerated, and the next kit booted. 2 more worked after a host-side recovery (below).
• On the PR head's tree: 23 sleeps, and every cut's partner had its contract back at the first poll (1000 ms).
• With injected rearm failures (1, 3 and 8 tries; TEST kits only): each failed try left the controller in APP mode, and a later try rearmed the link.
• Before the fix, a no-contract cut wedged the host's controller. Its simulated unplug brought PD back, and the reboot VDM then worked from the APP -mode controller.

Not yet shown on hardware. The second reset in t8140-sleep hasn't met a real no-contract cut yet (none occurred in 35 cuts with it), so whether it brings the partner back is still open. The sn201202x rejects DISC (-EINVAL), so a simulated unplug from the Neo's side isn't an option. Separately, on the merged test kit only, from its 13th sleep on the cut's first SPMI commands timed out (the stop failed: -110). The link then stayed up through those sleeps, and they still returned. This wasn't seen in 20 sleeps on the PR's tree alone.

🤖 Generated with Claude Code

[DjDeveloperr]

Superseded by #1445 (merged as 2b05915); its changes are in integration/aurora. The debug-link rearm is sleep_hpm_wait_contract() in drivers/soc/apple/sleep.c; the tipd side is in drivers/usb/typec/tipd/.