1. Paravirtual/HVM linux examples Argentina Software Development Center Software and Solutions Group 31 July 2008

2. INTEL CONFIDENTIAL void __init trap_init(void) -> for Xen Xen patch at /arch/i386/kernels/traps-xen.c static trap_info_t trap_table[] = { { 0, 0, __KERNEL_CS, (unsigned long)divide_error}, { 1, 0|4, __KERNEL_CS, (unsigned long)debug}, { 3, 3|4, __KERNEL_CS, (unsigned long)int3}, { 4, 3, __KERNEL_CS, (unsigned long)overflow}, { 5, 0, __KERNEL_CS, (unsigned long)bounds}, { 6, 0, __KERNEL_CS, (unsigned long)invalid_op}, { 7, 0|4, __KERNEL_CS, (unsigned long)device_not_available}, { 9, 0, __KERNEL_CS, (unsigned long)coprocessor_segment_overrun }, { 10, 0, __KERNEL_CS, (unsigned long)invalid_TSS}, { 11, 0, __KERNEL_CS, (unsigned long)segment_not_present}, { 12, 0, __KERNEL_CS, (unsigned long)stack_segment}, { 13, 0, __KERNEL_CS, (unsigned long)general_protection}, { 14, 0|4, __KERNEL_CS, (unsigned long)page_fault}, { 15, 0, __KERNEL_CS, (unsigned long)fixup_4gb_segment}, { 16, 0, __KERNEL_CS, (unsigned long)coprocessor_error}, { 17, 0, __KERNEL_CS, (unsigned long)alignment_check}, { 19, 0, __KERNEL_CS, (unsigned long)simd_coprocessor_error}, { SYSCALL_VECTOR, 3, __KERNEL_CS, (unsigned long)system_call}, { 0, 0, 0, 0} }; void __init trap_init(void) { HYPERVISOR_set_trap_table(trap_table);

3. INTEL CONFIDENTIAL void __init trap_init(void) -> for linux /arch/i386/kernels/traps-xen.c void __init trap_init(void) { #ifdef CONFIG_X86_LOCAL_APIC init_apic_mappings(); #endif set_trap_gate(0,&divide_error); set_intr_gate(1,&debug); set_intr_gate(2,&nmi); set_system_intr_gate(3, &int3); /* int3/4 can be called from all */ set_system_gate(4,&overflow); … static inline void _set_gate(int gate, unsigned int type, void *addr, unsigned short seg) { __u32 a, b; pack_gate(&a, &b, (unsigned long)addr, seg, type, 0); write_idt_entry(idt_table, gate, a, b); } LGDT, LIDT, LLDT, LTR, SGDT, SIDT, SLDT, STR. These instructions cause VM exits if the “descriptor- table exiting” VM-execution control is 1

4. INTEL CONFIDENTIAL void __init time_init (void) -> for Xen Xen patch at /arch/i386/kernels/time-xen.c void __init time_init(void) { … HYPERVISOR_vcpu_op(VCPUOP_set_periodic_timer, 0, &xen_set_periodic_tick); get_time_values_from_xen(0);

5. INTEL CONFIDENTIAL void __init time_init (void) -> for linux /arch/i386/kernels/time.c void __init time_init(void) { struct timespec ts; ts.tv_sec = get_cmos_time(); ts.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ); do_settimeofday(&ts); do_time_init(); } static inline int native_set_wallclock(unsigned long nowtime) /* time.h */ { int retval; if (efi_enabled) retval = efi_set_rtc_mmss(nowtime); else retval = mach_set_rtc_mmss(nowtime); return retval; } IN, INS/INSB/INSW/INSD, OUT, OUTS/OUTSB/OUTSW/OUTSD. The behavior of each of these instructions is determined by the settings of the “unconditional I/O exiting” and “use I/O bitmaps” VM-execution controls

6. INTEL CONFIDENTIAL void __cpuinit cpu_gdt_init(…) -> for Xen Xen patch at /arch/i386/kernels/cpu/common-xen.c void __cpuinit cpu_gdt_init(struct Xgt_desc_struct *gdt_descr) { unsigned long frames[16]; unsigned long va; int f; for (va = gdt_descr->address, f = 0; va address + gdt_descr->size; va += PAGE_SIZE, f++) { frames[f] = virt_to_mfn(va); make_lowmem_page_readonly( (void *)va, XENFEAT_writable_descriptor_tables); } if (HYPERVISOR_set_gdt(frames, gdt_descr->size / 8)) BUG(); }

7. INTEL CONFIDENTIAL void __cpuinit cpu_gdt_init(…) -> for linux /arch/i386/kernels/cpu/common.c void __cpuinit cpu_set_gdt(int cpu) { struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu); /* Reinit these anyway, even if they've already been done (on the boot CPU, this will transition from the boot gdt+pda to the real ones). */ load_gdt(cpu_gdt_descr); set_kernel_gs(); } #define load_gdt(dtr) __asm__ __volatile("lgdt %0"::"m" (*dtr)) LGDT, LIDT, LLDT, LTR, SGDT, SIDT, SLDT, STR. These instructions cause VM exits if the “descriptor-table exiting” VM-execution control is 1

8. INTEL CONFIDENTIAL Managing the VMCS Before entering VMX operation, the host VMM allocates a VMXON region A unique VMCS region is required for each virtual machine; a VMXON region is required for the VMM itself. A VMM determines the VMCS region size by reading IA32_VMX_BASIC MSR The address of the VMXON region for the VMM is provided as an operand to VMXON instruction Once in VMX root operation, the VMM needs to prepare data fields in the VMCS that control the execution of a VM upon a VM entry The VMM can make a VMCS the current VMCS by using the VMPTRLD instruction. VMCS data fields must be read or written only through VMREAD and VMWRITE commands respectively Every component of the VMCS is identified by a 32-bit encoding that is provided as an operand to VMREAD and VMWRITE Software must maintain the VMCS structures in cache-coherent memory

9. INTEL CONFIDENTIAL VMX Transitions and States of VMCS

10. INTEL CONFIDENTIAL Exercise 1: build your paravirtual Hello World

11. INTEL CONFIDENTIAL Exercise 2: write info in the xenstore

12. INTEL CONFIDENTIAL Exercise 3: basics for an HVM driver Determine whether the OS is running over Xen –Xen provides a standard string “XenVMMXenVMM” when calling CPUID with EAX = 40000000h –CPUID traps and Xen handles the VM Exit at vmexit.c:vmx_vmexit_handler –Response at traps.c:cpuid_hypervisor_leaves Define the initial address for the hypercalls page –Call CPUID with EAX = 40000002h –You’ll receive the # of hypercalls to be supported, use it to allocate a block to store callbacks