Resume from S3 (deep) causes SATA IDENTIFY timeout (ata5) → DBus/systemd collapse → black screen

[carl10101010]

Hardware:

• Motherboard: Gigabyte X570 AORUS MASTER
• CPU: Ryzen 5900X
• GPU: NVIDIA RTX 3090
• Storage:
  • NVMe root (Samsung 970 EVO Plus)
  • 2× SATA HDDs (Seagate 12TB + 4TB)
• Firmware: S3 deep sleep enabled

Software:

• CachyOS (Plasma Wayland)
• Kernel: 6.18.2-2-cachyos
• systemd: systemd 259 (259-1-arch)
• NVIDIA driver: 590.48.01

Problem:
Resuming from S3 (deep) intermittently fails. Displays remain black and system is unreachable via SSH, requiring hard power-off.

Observed behavior during resume:

• SATA device on ata5 fails to respond to IDENTIFY:
  ata5: link is slow to respond
  ata5.00: qc timeout (cmd 0xec)
  ata5.00: failed to IDENTIFY (I/O error)
• systemd logs "Got disconnect on API bus"
• DBus clients disconnect (wireplumber, NetworkManager, logind)
• NetworkManager exits → SSH unreachable
• KWin Wayland reports atomic modeset failures

Notes:

• SMART on all drives is clean (no CRC, no pending sectors)
• Root filesystem is NVMe (not affected)
• Issue disappears when forcing s2idle:
  echo s2idle > /sys/power/mem_sleep
• Issue mitigated by freezing user sessions during suspend:
  SYSTEMD_SLEEP_FREEZE_USER_SESSIONS=1

Expected behavior:
System resumes cleanly from S3 without userspace collapse.