[DjDeveloperr]

Stacked on #1310 (fix/t8140-debug-port-select). Two ways a sleep that keeps DRAM up (SLP_DDR) never came back on the Neo: the SoC reset 5 s after the alarm, inside the sleep engine's D->M wake, before aura or Linux ran. A deep sleep returns in both cases, so each now takes the deep path when it would stall.

After the camera ran. A capture since the last deep wake made the next SLP_DDR sleep stall 6 of 6 times: right after the capture, ~50 s after it, and whether the ISP firmware was stopped at the sleep or at the end of the stream. A failed camera start counts too. SLP_DDR wakes without camera use returned 70+ of 70+. The ISP's firmware start now clears the same flag the first deep wake sets, so the next sleep goes deep and SLP_DDR applies again from its wake. It costs one USB hub reset after a camera session.

After a coprocessor that runs through a sleep crashed. The AOP firmware sometimes crashes during a sleep (RTKit: Message to undiscovered endpoint 0x01, co-processor has crashed: crash log buffer ... not mapped). After one during a deep sleep, the next, DRAM-kept sleep stalled. SLP_DDR keeps DRAM up for the AOP, and nothing restarts it short of a reboot, so after such a crash every later sleep takes the deep path. RTKit now reports each firmware crash to the sleep controller. Coprocessors the controller stops and restarts around each sleep (DCP, ANS, GPU) are left alone.

Tested on the Neo (J700), #1310 + #1313 + this, stage2-14, production policy (first sleep deep), 30 s PMU RTC sleeps:

• Camera start before sleeps 2, 4, 6 and 8, none before 3, 5, 7 and 9: 9 of 9 returned. The rule fired before 2 and 4 (both deep), and 3 and 5 kept DRAM. The starts before 6 and 8 failed without reaching the ISP, see below, so 6 to 9 all kept DRAM and returned.
• The crash rule fired on hardware: an AOP crash during sleep 1 logged 310c00000.aop crashed: later sleeps take the deep path, and sleeps 2 and 3 went deep and returned.
• Without either rule, the same kernel: camera then SLP_DDR stalled 3 of 3, and the AOP crash then SLP_DDR stalled 1 of 1.

Not fixed here:

• The first sleep of a boot still has to go deep: a forced first SLP_DDR sleep still stalls, ISP off or not.
• The camera fails to start again (STREAMON ... Timer expired, then -EIO until reboot) after a capture, a deep sleep and a DRAM-kept sleep.
• The AOP crash itself.
• Stage2-15 (aura main + iMac · M3 family #117 + Moved · J414C · MacBook Pro 14-inch (M2 Max / T6021) #112's rewrite) stalled 2 of 2 SLP_DDR wakes on the same kernel, which returned 5 of 5 on stage2-14. Those two kits' first cycles also ran the camera, so this may be the same trigger: re-test before blaming aura.

🤖 Generated with Claude Code

[DjDeveloperr]

Superseded by #1445 (merged as 2b05915); its changes are in integration/aurora. apple_sleep_coproc_crashed() still sends later sleeps deep after a coprocessor crash. The ISP rule was replaced by #1393's DART restore.