| name | boot-analysis |
| description | This skill should be used when analyzing the Breenix kernel boot sequence, verifying initialization order, timing boot stages, identifying boot failures, optimizing boot time, or understanding the boot process from bootloader handoff to kernel ready state. |
Boot Sequence Analysis for Breenix
Analyze and optimize the kernel boot process from bootloader to kernel ready.
Purpose
Understanding the boot sequence is critical for debugging initialization issues, optimizing boot time, and ensuring proper subsystem ordering. This skill provides tools for analyzing boot logs, verifying checkpoint progression, and identifying boot failures.
When to Use
- Boot failures: Kernel hangs or crashes during initialization
- Initialization order issues: Subsystems initialized in wrong order
- Boot time optimization: Reducing time from bootloader to ready
- Checkpoint verification: Confirming all subsystems initialize correctly
- Boot regression analysis: New code breaks boot sequence
- Understanding boot flow: Learning how kernel initialization works
Breenix Boot Sequence
Phase 1: Bootloader Handoff
What happens:
- Bootloader (bootloader crate) loads kernel
- Sets up initial page tables
- Provides memory map
- Transfers control to kernel entry point
Entry point: kernel/src/main.rs kernel_main()
Initial state:
// CPU in Long Mode (64-bit)
// Interrupts disabled
// Paging enabled (bootloader setup)
// Stack ready
// Physical memory mapped at offset
Typical log output:
[Bootloader messages]
Loading kernel...
Jumping to kernel entry point...
Phase 2: Early Initialization
Subsystems initialized (in order):
1. Logger
[ INFO] Breenix OS starting...
- Serial output configured
- Framebuffer initialized
- Log level set
2. GDT (Global Descriptor Table)
[ INFO] GDT initialized
- Kernel/user code segments
- Kernel/user data segments
- TSS (Task State Segment)
3. IDT (Interrupt Descriptor Table)
[ INFO] IDT initialized
- Exception handlers (divide by zero, page fault, etc.)
- Interrupt handlers (timer, keyboard)
- Double fault handler with IST stack
4. PIC (Programmable Interrupt Controller)
[ INFO] PIC initialized
- Remapped to avoid conflicts
- All interrupts masked initially
Phase 3: Memory Subsystem
5. Frame Allocator
[ INFO] Physical memory: 94 MiB usable
[DEBUG] Frame allocator initialized
- Reads bootloader memory map
- Identifies usable regions
- Initializes frame tracking
6. Heap Allocator
[ INFO] Heap: 1024 KiB
- Sets up kernel heap
- Enables dynamic allocation
- #[global_allocator] now functional
7. Virtual Memory
[DEBUG] Page table initialized
- Kernel page table setup
- Higher-half kernel mapping
- Recursive mapping if used
8. Kernel Stacks
[DEBUG] Kernel stack allocator initialized
- Stack bitmap allocator
- Guard pages configured
- IST stacks for exceptions
Phase 4: Device Drivers
9. Timer (PIT)
[ INFO] Timer initialized at 100 Hz
- Configures Programmable Interval Timer
- Sets interrupt frequency
- Starts tick counting
10. RTC (Real-Time Clock)
[ INFO] RTC initialized: 2025-10-23 12:34:56 UTC
- Reads hardware clock
- Caches boot time
- Enables wall-clock time APIs
11. Serial Input
[ INFO] Serial input interrupts enabled
- UART receive interrupts
- Input buffer ready
- Command processing available
12. Keyboard
[ INFO] Keyboard initialized
- PS/2 keyboard driver
- Scancode processing
- Key event generation
Phase 5: System Infrastructure
13. Interrupts Enabled
[ INFO] Enabling interrupts...
- Unmasks timer interrupt
- Unmasks keyboard interrupt
- System becomes responsive
14. System Calls
[ INFO] System call infrastructure initialized
- INT 0x80 handler registered
- Syscall dispatcher ready
- SWAPGS configured
15. Threading
[ INFO] Threading subsystem initialized
- Scheduler initialized
- Idle thread created
- Context switch infrastructure ready
16. Process Management
[ INFO] Process management initialized
- Process manager ready
- PID allocation working
- Fork/exec infrastructure initialized
Phase 6: Testing (if enabled)
17. POST (Power-On Self Test)
[ INFO] Running POST tests...
=== Memory Test ===
✅ MEMORY TEST COMPLETE
...
🎯 KERNEL_POST_TESTS_COMPLETE 🎯
- Validates subsystems
- Runs self-checks
- Confirms kernel health
18. Userspace Tests (if configured)
RING3_SMOKE: creating hello_time userspace process
[ INFO] Process created: PID 1
USERSPACE OUTPUT: Hello from userspace!
- Creates test processes
- Verifies userspace execution
- Tests system calls
Phase 7: Kernel Ready
Final state:
[ INFO] Kernel initialization complete
[ INFO] System ready
- All subsystems operational
- Ready for interactive use or more tests
- Idle loop or wait for input
Boot Analysis Techniques
Technique 1: Extract Boot Timeline
Using log-analysis skill:
# Get all initialization messages in order
grep "initialized\|INITIALIZED\|Initializing" logs/breenix_20251023_*.log
# Or more comprehensive
grep -E "INFO|WARN|ERROR" logs/latest.log | less
Expected sequence:
- GDT initialized
- IDT initialized
- PIC initialized
- Physical memory info
- Timer initialized
- RTC initialized
- Interrupts enabled
- Threading initialized
- Process management initialized
Technique 2: Find Boot Checkpoint Failures
Identify last successful checkpoint:
# Find last "initialized" message
grep "initialized" logs/breenix_*.log | tail -10
# Or find last successful operation
grep "SUCCESS\|✅\|complete" logs/breenix_*.log | tail -10
If boot hangs:
- Last checkpoint shows how far boot progressed
- Next subsystem is where hang occurs
- Focus debugging on that subsystem
Technique 3: Compare Boot Sequences
Working vs broken boot:
# Extract initialization sequence
grep "initialized\|Initializing" working.log > working_boot.txt
grep "initialized\|Initializing" broken.log > broken_boot.txt
# Compare
diff -u working_boot.txt broken_boot.txt
Look for:
- Missing initialization steps
- Different initialization order
- New error messages
- Stops at different point
Technique 4: Time Boot Stages
Add timing checkpoints:
let start = kernel::time::get_monotonic_ms();
// Initialize subsystem
gdt::init();
let elapsed = kernel::time::get_monotonic_ms() - start;
log::info!("GDT initialization took {}ms", elapsed);
Analyze timing:
- Which stages are slow?
- Where can we optimize?
- Any unexpected delays?
Technique 5: Verify Subsystem Dependencies
Check initialization order:
// Memory must be initialized before heap
assert!(frame_allocator.is_initialized());
heap::init(); // Safe now
// GDT must be before IDT
gdt::init();
idt::init(); // Can reference GDT segments
// Interrupts must be off during sensitive operations
assert!(!are_enabled());
Common Boot Issues
Issue 1: Boot Hang
Symptoms:
- Kernel boots partway then stops
- No error message, just hangs
- Some subsystems initialized, others not
Diagnosis:
# Find last successful operation
grep "initialized\|complete" logs/latest.log | tail -1
# Check if interrupts were enabled prematurely
grep "Enabling interrupts" logs/latest.log
# Look for infinite loops
grep "WARN\|ERROR" logs/latest.log
Common causes:
Interrupts enabled too early
- Timer interrupt fires before handler ready
- Solution: Ensure all handlers registered before enabling
Deadlock during initialization
- Lock acquired, never released
- Solution: Check lock usage during boot
Infinite loop in subsystem init
- Waiting for condition that never happens
- Solution: Add timeouts or debug why condition fails
Fix patterns:
// Add checkpoint logging
log::info!("About to initialize subsystem X");
subsystem_x::init();
log::info!("Subsystem X initialized successfully");
// If hangs between checkpoints, focus on subsystem_x::init()
Issue 2: Boot Panic
Symptoms:
PANIC: [message]
Stack trace: ...
Diagnosis:
# Get panic message
grep "PANIC" logs/latest.log
# Get context
grep -B20 "PANIC" logs/latest.log
Common causes:
Assertion failure
assert!(condition); // Failed during bootCheck if assertion is correct or if precondition not met
Unwrap on None/Err
let value = option.unwrap(); // Panic if NoneUse proper error handling during boot
Out of memory
allocation error: Layout { ... }Increase heap size or defer allocation
Issue 3: Wrong Initialization Order
Symptoms:
- Later subsystem fails
- "Not initialized" error
- Double fault or page fault
Example:
// BAD - heap used before initialization
use alloc::vec::Vec;
let v = Vec::new(); // Panic! Heap not initialized yet
heap::init();
// GOOD
heap::init();
use alloc::vec::Vec;
let v = Vec::new(); // OK
Diagnosis:
- Check dependency chain
- Verify order matches requirements
- Look for use-before-init patterns
Issue 4: Boot Regression
Symptoms:
- Kernel booted before, now doesn't
- Recent change broke boot
- Used to reach checkpoint X, now stops at Y
Diagnosis:
# Find which commit broke it
git bisect start
git bisect bad HEAD
git bisect good last_working_commit
# Test each commit
kernel-debug-loop/scripts/quick_debug.py \
--signal "KERNEL READY" \
--timeout 15
Fix:
- Identify the breaking commit
- Understand what it changed
- Fix or revert
Boot Optimization
Current Boot Time
Measure with kernel-debug-loop:
# Time to specific checkpoint
kernel-debug-loop/scripts/quick_debug.py \
--signal "Kernel initialization complete" \
--timeout 30
Typical times (approximate):
- Early init (GDT, IDT, PIC): ~10ms
- Memory subsystem: ~50ms
- Drivers (timer, keyboard): ~20ms
- Threading/processes: ~10ms
- POST tests: ~100ms (if enabled)
- Total boot: ~200-500ms to ready state
Optimization Strategies
1. Defer non-critical initialization
// Don't initialize during boot if not needed
// keyboard::init(); // Defer until first use
// Or lazy initialization
pub fn get_keyboard() -> &'static Keyboard {
static INIT: Once = Once::new();
INIT.call_once(|| {
keyboard::init();
});
&KEYBOARD
}
2. Parallelize independent operations
// Currently serial:
timer::init();
rtc::init();
keyboard::init();
// Could be parallel (if truly independent):
// Note: Difficult in kernel without threading during boot
3. Reduce logging verbosity
// Debug builds: verbose
#[cfg(debug_assertions)]
log::debug!("Detailed info");
// Release builds: minimal
#[cfg(not(debug_assertions))]
log::info!("Essential info only");
4. Optimize expensive operations
// Identify slow operations with timing
let start = time::get_monotonic_ms();
expensive_operation();
let elapsed = time::get_monotonic_ms() - start;
if elapsed > 10 {
log::warn!("Slow operation: {}ms", elapsed);
}
Integration with Other Skills
With kernel-debug-loop
# Fast iteration on boot fixes
kernel-debug-loop/scripts/quick_debug.py \
--signal "BOOT_CHECKPOINT" \
--timeout 10
With log-analysis
# Extract boot sequence
echo '"initialized"' > /tmp/log-query.txt
./scripts/find-in-logs
# Find boot failures
echo '"PANIC|FAULT|ERROR"' > /tmp/log-query.txt
./scripts/find-in-logs
With systematic-debugging
Document boot issues:
# Problem
Kernel hangs after "Enabling interrupts"
# Root Cause
Timer interrupt handler called before scheduler ready
# Solution
Initialize scheduler before enabling interrupts
# Evidence
Before: Hang
After: Boot completes successfully
Boot Checkpoints Reference
Essential checkpoints every boot should reach:
[✓] GDT initialized
[✓] IDT initialized
[✓] PIC initialized
[✓] Physical memory detected
[✓] Heap initialized
[✓] Timer initialized
[✓] Interrupts enabled
[✓] Threading initialized
[✓] Kernel ready
If boot stops before reaching a checkpoint, debug that subsystem.
Best Practices
- Log initialization: Every subsystem should log successful init
- Check prerequisites: Verify dependencies before initializing
- Fail fast: Panic early if critical init fails
- Add checkpoints: Mark progress through boot sequence
- Time operations: Identify bottlenecks
- Test changes: Verify boot still works after changes
- Document order: Comment why init order matters
Quick Reference
Key Boot Files
kernel/src/main.rs - Entry point, boot orchestration
kernel/src/gdt.rs - GDT initialization
kernel/src/interrupts/mod.rs - IDT initialization
kernel/src/memory/frame_allocator.rs - Physical memory
kernel/src/time/timer.rs - Timer initialization
kernel/src/time/rtc.rs - RTC initialization
Boot Signals to Watch
"GDT initialized"
"IDT initialized"
"Physical memory:"
"Heap:"
"Timer initialized"
"Enabling interrupts"
"Threading subsystem initialized"
"Kernel initialization complete"
Summary
Boot analysis requires:
- Understanding the complete boot sequence
- Identifying checkpoints and dependencies
- Using logs to diagnose failures
- Comparing working vs broken boots
- Optimizing slow operations
- Ensuring proper initialization order
A well-understood boot sequence makes debugging initialization issues straightforward.