Lab 4 Tutorial Instructor: Youngjin Kwon What weve done so far Lab - - PowerPoint PPT Presentation

Lab 4 Tutorial

Instructor: Youngjin Kwon

What we’ve done so far

• Lab 1: Booting OS from BIOS and initializing kernel
• Lab 2: Physical memory management and memory mapping (kernel)
• Lab 3: Defining user environment, handling interrupt/exception/system call

What you will be given for lab 4

• Lab 1: Booting OS from BIOS and initializing kernel
• Lab 2: Physical memory management and memory mapping (kernel)
• Lab 3: Defining user environment, handling interrupt/exception/system call
• (JOS LAB 4): Multi-process environments, scheduler, IPC primitives
• (JOS LAB 5): File system, read/write syscalls, shell
• (JOS LAB 6): Network stack, network driver (memory-mapped IO)

JOS OS architecture

JOS OS

Application (user environment) File system server (FS environment) Network server (NS environment)

Read/Write IPC fsipc()

Hardware

Direct hardware access

Not enabled in LAB 4

JOS VMM overview

JOS hOS + VMM

JOS gOS (guest OS) environment File system server (FS environment) Network server (NS environment)

Read/Write IPC fsipc()

Hardware

Direct hardware access

Not enabled in LAB 4

Let’s see demo “Run JOS on JOS”

Steps to run VMM

• JOS booting
• Launch FS server (fs_fs) and shell server (user_icode)
• Run application called “vmm” in shell
• vmm application: launching guest OS environment
• Declare itself as gOS (OS environment) to hOS
• Load gOS kernel to memory
• sys_yield(): control goes to hOS
• JOS hOS
• Turn hOS to hOS + VMM by enabling intel VT
• Execute the vmm environment as gOS

JOS hOS vmm (user env.) JOS hOS + VMM vmm (guest OS)

vmm: declare itself as guest OS

vmm (user environment) user/vmm.c hOS kern/syscall.c Exercise 1 Newly added member in struct Env Prepare env for guest OS type

vmm: load gOS kernel to memory

• Exercise 3
• Open GUEST_KERN (elf format)
• Load elf sections to guest memory with

map_in_guest()

• vmm env.: map_in_guest()
• Allocate a temp page (to where?)
• Read given elf section

(specified by fd and offset)

• Call sys_ept_map() to ask hOS to do the

mapping in EPT

• hOS: sys_ept_map()
• Do some checks for error conditions
• Call ept_map_hva2gpa()

vmm (user environment) hOS

vmm: load gOS kernel to memory

• Exercise 2: handling EPT
• Call ept_map_hva2gpa to do the

mapping

• ept_map_hva2gpa
• Map hva to gpa
• Use ept_lookup_gpa() to find ept

entry of given gpa

• ept_lookup_gpa
• Similar to page_lookup but it walks

ept and returns the leaf ept entry

Same address? Or not?

Host virtual and guest physical address

JOS hOS create env with 5GB 0x10000000 0x10000000 + 5G - 1

Guest OS

4GB

Guest physical Host Virtual Guest virtual

0x0 4GB - 1 Guest page table 0x50000000 Host page table

Host physical

Used for guest physical

Mov $rdx, addr

Non root mode Root mode

How to get host VA in JOS?

Kernel virtual address

Steps to run VMM

• JOS booting
• Launch FS server (fs_fs) and shell server (user_icode)
• Run application called “vmm” in shell
• vmm application: launching guest OS environment
• Declare itself as gOS to hOS
• Load gOS kernel to memory
• sys_yield(): control goes to hOS
• JOS hOS
• Turn hOS to hOS + VMM by Intel VT
• Execute the vmm environment as gOS

JOS hOS vmm (user env.) JOS hOS + VMM vmm (guest OS)

hOS: Turn hOS to hOS + VMM

Sched_yield (kern/sched.c)

hOS + VMM: execute vmm env. as gOS

• vmx_vmrun()
• Execute environment as guest
• peration system

Env_run() (kern/env.c)

Background: Intel-VT

When Guest OS executes VMX root-privileged instructions

Virtual-Machine Control data Structure (VMCS)

• VMCS data area
• Guest-state area
• Host-state area
• VM execution control field
• VM exit control field
• VM entry control field
• VM exit information field

Detailed layout of VMCS data area at vmm/vmx.h How to manipulate VMCS: vmcs_ctls_init()

VMCS control example

• Scenario: page fault validation
• VMM hijacks page faults happened in a guest OS
• Verify the page faults
• If it is a legal page fault, VMM injects the page faults

to the guest OS

Hijacking page faults

• Using exception bitmap: 32 bit
• If a bit of a certain position is set, the exception cause vm exit.

Otherwise CPU delivers the exception to guest OS IDT

• if 14 bit (exception vector 14 == page fault) is set, VMM takes a control

when PF happens in the guest OS

Vmcs_ctls_init() – vmm/vmx.c

injects page faults

• Using vm entry control field: 32 bit
• On vm entry, CPU delivers an event through the guest OS IDT

Vmcs_ctls_init() – vmm/vmx.c

Vmlaunch/vmresume and vmexit

• Exercise 4
• Write code for vmlaunch/vmresume
• How to determine vmlaunch or

vmresume?

• When vmlaunch/vmresume returns, it

means vmexit (guest OS completely stops)

asm_vmrun (vmm/vmx.c)

…

What causes vmexit?

Bugnion et al, Hardware and Software Support for Virtualization, Morgan & Calypool Publisher

vmcall

hOS + VMM gOS

Vmexit handler

• Exercise 5,6,7 (trap-and-emulate)
• gOS traps  vmexit()
• Find out vmexit reason (how?)
• Implement the corresponding vmexit handler
• Exercise 5,7: vmcall
• Exercise 6: cpuid instruction