Panic Screen and Crash Experience

What is it?

The panic screen is the full-screen blue crash display the kernel draws when a fault or an explicit bugcheck cannot be recovered. panic_screen() in panic.c renders the icon, exception name, fault registers, an RBP-chain stack trace and the crash log path straight to the framebuffer, with no dependency on the compositor. The richer crash experience in the roadmap (TrueType text, inline log context, a data-bearing QR code, crash statistics, panic modes, beep codes and keyboard recovery actions) is not built yet.

How does it work?

panic_screen() is a thin public wrapper around the static panic_screen_impl(), which does the work: it masks interrupts, decides through a compare-exchange token which CPU owns the crash record when several fault at once, and copies the caller's description and file strings into fixed local buffers so a corrupted pointer is never read twice. It then calls panic_collect_evidence() to write a cross-boot forensic record (struct panic_evidence, panic.h) to a fixed physical page before it touches the screen or the disk. That record carries its own CRC-32 and a magic and epoch publication word, so the next boot's panic_evidence_restore_early() can tell whether a prior crash record is valid, and panic_had_previous_crash() lets the desktop surface an "unexpected shutdown" notice.

Only after the evidence is captured does the screen change: the 128x128 icon (bsod_icon.h), the exception name from the 32-entry panic_exception_names table, the register dump, and a stack trace built by walking the RBP chain and resolving each return address with symtab_resolve() (symtab.c). All text goes through printk() and fb_set_color() in the console bitmap font; nothing in panic.c calls the TrueType renderer. When the VFS is up, the crash reason and stack trace are also written to C:\Impossible\System\crashdump.log.

The auto-restart countdown reads the AutoRestart DWORD under HKLM\SYSTEM\Recovery and falls back to DEFAULT_RESTART_SECS (30) when the value is missing or invalid. The runtime tunable panic.timeout (0 to 3600 seconds, 0 disables the restart) overrides the Registry value when it is set. There is no keyboard-cancel path, no crash-loop counter, and no NVRAM crash statistics.

What are its interfaces?

Interface	Purpose
`panic_screen(frame, error_code, description, file, line)`	Public entry point for a raw-fault panic (`panic.c`)
`KeBugCheckEx()`, `KeBugCheckExFrame()`	Bugcheck entry points that share the same renderer
`struct panic_evidence`, `panic_evidence_populate()`, `panic_evidence_restore()`	Cross-boot crash record with CRC-32 and publication word (`panic.h`)
`panic_evidence_restore_early()`, `panic_had_previous_crash()`	Early-boot recovery of the prior boot's record, and the flag the desktop reads
`symtab_resolve()`	Symbol resolution for the stack trace (`symtab.c`)
`bugcheck_name()`	STOP-code to name lookup over the `s_bugcheck_names` table
`HKLM\SYSTEM\Recovery\AutoRestart`, `panic.timeout`	Registry DWORD and runtime tunable for the countdown length in seconds; the tunable wins

How do I use it?

The panic screen needs no configuration; it renders whenever a fault reaches panic_screen() or a bugcheck calls into the shared renderer, provided the framebuffer is up (Phase 1 onward). A Phase 0 panic, before the framebuffer exists, is serial-only. The countdown is set by the AutoRestart Registry value or, taking precedence, the panic.timeout tunable; boot.conf has no panic-screen key. To see it on a debug build, press Ctrl+ScrollLock twice within two seconds, which calls KeBugCheckEx() with BUGCHECK_MANUALLY_INITIATED_CRASH when HKLM\SYSTEM\CrashControl\CrashOnCtrlScroll is set (see Crash Dump Generation).

bash scripts/test.sh SUITE=boot   # bugcheck names and crash capture suites

What is not implemented yet?

TrueType text. The screen renders in the console bitmap font through printk(); panic.c never calls the TrueType renderer (TTF Font Rendering in Panic Screen).
The redesigned layout. No margin, section or HiDPI scaling constants exist (Improved Layout and Visual Hierarchy).
Inline klog context. There is no klog_get_recent(), so the last log lines are not shown (Crash Context: Last 10 klog Entries Inline).
A data-bearing QR code. No QR encoder exists (Smart QR Code with Compressed Crash Data).
A cancellable countdown. The countdown has no keyboard-cancel check (Auto-Restart Countdown Improvements).
Per-exception hints, crash statistics and loop protection. No hints table, NVRAM statistics or consecutive-crash tracking exist (Crash Analysis Hints On-Screen, Crash Statistics & Loop Protection, Safe Mode Suggestion After Repeated Crashes).
A "What failed" module line. exec_find_module_by_pc() exists in exec.c but the panic path does not call it (What Failed Faulting Module Identification).
Panic modes, beep codes and keyboard recovery actions. None of the user, developer or QR modes, PC speaker codes or F-key actions exist (Panic Screen Modes (User / Developer / QR), Audio Crash Notification (PC Speaker Beep Codes), Keyboard-Driven Recovery Actions at Crash Screen).
Dump progress. There is no progress callback, and the minidump writer it would report on is not built (Dump Collection Progress Percentage).

How does it compare with Windows 11 and Linux?

Today's screen matches the basics both operating systems ship: a full-screen stop display, a register dump and a stack trace. Windows 11 24H2 moved to a sparse black screen without a QR code, and the Linux DRM panic handler added kmsg QR codes and several verbosity modes in 6.12. The roadmap aims past both with a structured QR payload, local crash statistics and audio feedback, but none of that exists yet, so the current screen is closer to a pre-24H2 Windows stop screen than to either target.