Shutdown and reboot

Applications with unsaved state register a dedicated task with g_power_register(G_POWER_APPLICATION). Accept events only with sender G_TID_NONE, exactly sizeof(g_power_event) bytes and G_POWER_EVENT_MAGIC. Acknowledge the request token using g_power_acknowledge; a false reply vetoes the operation. Approval requires stopping changes to application state.

Callbacks have 30 seconds to finish. Cancellation events retire the matching request token. Applications without registration receive no callback.

The BSP coordinator asks applications to prepare and blocks new application launches. Storage producers stop accepting work and finish accepted administration jobs and controls. The coordinator then blocks filesystem mutations, drains existing mutations, flushes mounted filesystems and storage devices, prepares ACPI and disables PCI bus mastering. It parks secondary CPUs before the final power-off/reset writes. Cancellation restores PCI access, storage service and firmware events before reopening mutations.

Storage coordinators register with G_POWER_STORAGE_PRODUCER. They acknowledge readiness without blocking their serving task, keep new work excluded, and acknowledge cancellation after resuming service. Uncertain I/O vetoes shutdown; work that exceeds the preparation deadline cancels shutdown and continues running.

acpidriver runs uACPI at driver privilege in ring 3; it handles firmware methods, SCI and deferred work. libacpi exposes capability and S0–S5 queries. Orderly transitions support S5 power-off and reboot; querying a sleep state does not enable suspend support.

The kernel retains early APIC/HPET table discovery and the bounded final register sequence. No driver IPC runs after CPUs are parked and interrupts disabled. Reboot uses the FADT reset mechanism with an 8042 fallback. Both live and installed systems start ACPI after PCI and before device discovery.