zosbuilder/scripts/lib/testing.sh

#!/bin/bash
# Testing with QEMU and cloud-hypervisor

# Source common functions
LIB_SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
source "${LIB_SCRIPT_DIR}/common.sh"

# Testing configuration
QEMU_MEMORY="${QEMU_MEMORY:-512M}"
QEMU_TIMEOUT="${QEMU_TIMEOUT:-60}"
CLOUD_HYPERVISOR_MEMORY="${CLOUD_HYPERVISOR_MEMORY:-512M}"

# Test kernel boot with QEMU
function testing_qemu_boot() {
    local kernel_file="$1"
    local test_type="${2:-basic}"
    local timeout="${3:-$QEMU_TIMEOUT}"
    
    section_header "Testing with QEMU"
    
    if [[ ! -f "$kernel_file" ]]; then
        log_error "Kernel file not found: ${kernel_file}"
        return 1
    fi
    
    # Check if QEMU is available
    if ! command_exists "qemu-system-x86_64"; then
        log_error "QEMU not found. Install with: apt-get install qemu-system-x86"
        return 1
    fi
    
    log_info "Kernel file: ${kernel_file}"
    log_info "Memory: ${QEMU_MEMORY}"
    log_info "Test type: ${test_type}"
    log_info "Timeout: ${timeout}s"
    
    case "$test_type" in
        "basic")
            testing_qemu_basic_boot "$kernel_file" "$timeout"
            ;;
        "serial")
            testing_qemu_serial_boot "$kernel_file" "$timeout"
            ;;
        "interactive")
            testing_qemu_interactive_boot "$kernel_file"
            ;;
        *)
            log_error "Unknown QEMU test type: ${test_type}"
            return 1
            ;;
    esac
}

# Basic QEMU boot test (automated)
function testing_qemu_basic_boot() {
    local kernel_file="$1"
    local timeout="$2"
    
    log_info "Running basic QEMU boot test (${timeout}s timeout)"
    
    # QEMU command for automated testing
    local qemu_cmd="qemu-system-x86_64 \
        -kernel $kernel_file \
        -m $QEMU_MEMORY \
        -nographic \
        -serial mon:stdio \
        -append 'console=ttyS0,115200 console=tty1 loglevel=7 panic=10' \
        -no-reboot"
    
    log_info "QEMU command: ${qemu_cmd}"
    
    # Run with timeout
    if timeout "$timeout" $qemu_cmd 2>&1 | tee /tmp/qemu-boot.log; then
        log_info "QEMU boot test completed"
    else
        local exit_code=$?
        if [[ $exit_code -eq 124 ]]; then
            log_info "QEMU boot test timed out (${timeout}s) - this may be normal"
        else
            log_error "QEMU boot test failed with exit code: ${exit_code}"
            return 1
        fi
    fi
    
    # Check boot log for success indicators
    testing_analyze_boot_log "/tmp/qemu-boot.log"
}

# QEMU serial console test
function testing_qemu_serial_boot() {
    local kernel_file="$1"
    local timeout="$2"
    
    log_info "Running QEMU serial console test"
    
    # QEMU command optimized for serial console
    local qemu_cmd="qemu-system-x86_64 \
        -kernel $kernel_file \
        -m $QEMU_MEMORY \
        -nographic \
        -serial stdio \
        -append 'console=ttyS0,115200n8 console=tty1 consoleblank=0 earlyprintk=serial,ttyS0,115200n8 loglevel=7'"
    
    log_info "QEMU serial command: ${qemu_cmd}"
    
    # Run with timeout and capture output
    timeout "$timeout" $qemu_cmd 2>&1 | tee /tmp/qemu-serial.log
    
    # Analyze serial output
    testing_analyze_boot_log "/tmp/qemu-serial.log"
}

# Interactive QEMU boot (no timeout)
function testing_qemu_interactive_boot() {
    local kernel_file="$1"
    
    log_info "Starting interactive QEMU session"
    log_info "Use Ctrl+A, X to exit QEMU"
    
    # Interactive QEMU command
    local qemu_cmd="qemu-system-x86_64 \
        -kernel $kernel_file \
        -m $QEMU_MEMORY \
        -nographic \
        -serial mon:stdio \
        -append 'console=ttyS0,115200 console=tty1 loglevel=7'"
    
    log_info "QEMU command: ${qemu_cmd}"
    
    # Run interactively
    safe_execute $qemu_cmd
}

# Test kernel boot with cloud-hypervisor
function testing_cloud_hypervisor_boot() {
    local kernel_file="$1"
    local test_type="${2:-basic}"
    local timeout="${3:-$QEMU_TIMEOUT}"
    
    section_header "Testing with cloud-hypervisor"
    
    if [[ ! -f "$kernel_file" ]]; then
        log_error "Kernel file not found: ${kernel_file}"
        return 1
    fi
    
    # Check if cloud-hypervisor is available
    if ! command_exists "cloud-hypervisor"; then
        log_error "cloud-hypervisor not found. Install from: https://github.com/cloud-hypervisor/cloud-hypervisor"
        return 1
    fi
    
    log_info "Kernel file: ${kernel_file}"
    log_info "Memory: ${CLOUD_HYPERVISOR_MEMORY}"
    log_info "Test type: ${test_type}"
    
    case "$test_type" in
        "basic")
            testing_cloud_hypervisor_basic "$kernel_file" "$timeout"
            ;;
        "serial")
            testing_cloud_hypervisor_serial "$kernel_file" "$timeout"
            ;;
        *)
            log_error "Unknown cloud-hypervisor test type: ${test_type}"
            return 1
            ;;
    esac
}

# Basic cloud-hypervisor test
function testing_cloud_hypervisor_basic() {
    local kernel_file="$1"
    local timeout="$2"
    
    log_info "Running basic cloud-hypervisor test"
    
    # cloud-hypervisor command
    local ch_cmd="cloud-hypervisor \
        --kernel $kernel_file \
        --memory size=${CLOUD_HYPERVISOR_MEMORY} \
        --serial tty \
        --console off \
        --cmdline 'console=ttyS0,115200 loglevel=7 panic=10'"
    
    log_info "cloud-hypervisor command: ${ch_cmd}"
    
    # Run with timeout
    if timeout "$timeout" $ch_cmd 2>&1 | tee /tmp/cloud-hypervisor-boot.log; then
        log_info "cloud-hypervisor test completed"
    else
        local exit_code=$?
        if [[ $exit_code -eq 124 ]]; then
            log_info "cloud-hypervisor test timed out (${timeout}s)"
        else
            log_error "cloud-hypervisor test failed with exit code: ${exit_code}"
            return 1
        fi
    fi
    
    # Analyze boot log
    testing_analyze_boot_log "/tmp/cloud-hypervisor-boot.log"
}

# cloud-hypervisor serial test
function testing_cloud_hypervisor_serial() {
    local kernel_file="$1"
    local timeout="$2"
    
    log_info "Running cloud-hypervisor serial test"
    
    # cloud-hypervisor command with serial focus
    local ch_cmd="cloud-hypervisor \
        --kernel $kernel_file \
        --memory size=${CLOUD_HYPERVISOR_MEMORY} \
        --serial tty \
        --console off \
        --cmdline 'console=ttyS0,115200n8 earlyprintk=serial loglevel=7'"
    
    log_info "cloud-hypervisor serial command: ${ch_cmd}"
    
    timeout "$timeout" $ch_cmd 2>&1 | tee /tmp/cloud-hypervisor-serial.log
    
    testing_analyze_boot_log "/tmp/cloud-hypervisor-serial.log"
}

# Analyze boot log for success/failure indicators
function testing_analyze_boot_log() {
    local log_file="$1"
    
    section_header "Analyzing Boot Log"
    
    if [[ ! -f "$log_file" ]]; then
        log_warn "Boot log file not found: ${log_file}"
        return 1
    fi
    
    local log_size=$(get_file_size "$log_file")
    log_info "Boot log size: ${log_size}"
    
    # Success indicators
    local success_patterns=(
        "zinit.*starting"
        "zinit.*initialized"
        "login:"
        "zero-os.*login:"
        "Alpine Linux"
        "Welcome to Alpine"
    )
    
    # Error indicators
    local error_patterns=(
        "Kernel panic"
        "kernel BUG"
        "Unable to mount root"
        "VFS: Cannot open root device"
        "not syncing"
        "Attempted to kill init"
    )
    
    local success_count=0
    local error_count=0
    
    # Check for success patterns
    for pattern in "${success_patterns[@]}"; do
        if grep -i "$pattern" "$log_file" >/dev/null 2>&1; then
            log_info "✓ Found success indicator: ${pattern}"
            ((success_count++))
        fi
    done
    
    # Check for error patterns
    for pattern in "${error_patterns[@]}"; do
        if grep -i "$pattern" "$log_file" >/dev/null 2>&1; then
            log_error "✗ Found error indicator: ${pattern}"
            ((error_count++))
        fi
    done
    
    # Summary
    log_info "Boot log analysis:"
    log_info "  Success indicators: ${success_count}"
    log_info "  Error indicators: ${error_count}"
    
    if [[ $error_count -gt 0 ]]; then
        log_error "Boot test failed - errors detected in log"
        log_info "Check full log at: ${log_file}"
        return 1
    elif [[ $success_count -gt 0 ]]; then
        log_info "Boot test successful - system appears to be working"
        return 0
    else
        log_warn "Boot test inconclusive - no clear success/error indicators"
        return 2
    fi
}

# Run comprehensive test suite
function testing_run_all() {
    local kernel_file="$1"
    local test_timeout="${2:-60}"
    
    section_header "Running Comprehensive Test Suite"
    
    local test_results=()
    
    # Test with QEMU
    log_info "Running QEMU tests..."
    if testing_qemu_boot "$kernel_file" "basic" "$test_timeout"; then
        test_results+=("QEMU-basic: PASS")
    else
        test_results+=("QEMU-basic: FAIL")
    fi
    
    if testing_qemu_boot "$kernel_file" "serial" "$test_timeout"; then
        test_results+=("QEMU-serial: PASS")
    else
        test_results+=("QEMU-serial: FAIL")
    fi
    
    # Test with cloud-hypervisor (if available)
    if command_exists "cloud-hypervisor"; then
        log_info "Running cloud-hypervisor tests..."
        if testing_cloud_hypervisor_boot "$kernel_file" "basic" "$test_timeout"; then
            test_results+=("cloud-hypervisor-basic: PASS")
        else
            test_results+=("cloud-hypervisor-basic: FAIL")
        fi
    else
        log_warn "cloud-hypervisor not available, skipping tests"
        test_results+=("cloud-hypervisor: SKIPPED")
    fi
    
    # Report results
    section_header "Test Results Summary"
    local passed=0
    local failed=0
    local skipped=0
    
    for result in "${test_results[@]}"; do
        if [[ "$result" =~ PASS ]]; then
            log_info "✓ $result"
            ((passed++))
        elif [[ "$result" =~ FAIL ]]; then
            log_error "✗ $result"
            ((failed++))
        else
            log_warn "⚠ $result"
            ((skipped++))
        fi
    done
    
    log_info "Test summary: ${passed} passed, ${failed} failed, ${skipped} skipped"
    
    if [[ $failed -eq 0 ]]; then
        log_info "All tests passed successfully"
        return 0
    else
        log_error "Some tests failed"
        return 1
    fi
}

# Export functions
export -f testing_qemu_boot testing_qemu_basic_boot testing_qemu_serial_boot testing_qemu_interactive_boot
export -f testing_cloud_hypervisor_boot testing_cloud_hypervisor_basic testing_cloud_hypervisor_serial
export -f testing_analyze_boot_log testing_run_all