Lecture 16

• Lecture 16
  • Video
  • Efficient VMMs : Popek and Gold Theorem 1
  • Hardware Support for Virtualization
  • Intel Virtualization Technology Evolution
  • Emulation
  • Definitions
  • Ways to carry out emulation
  • Interpreter State
  • Indirect threaded Interpretation

Video

link

Efficient VMMs : Popek and Gold Theorem 1

• emulation directly nahi kar sakte bcz off critical instructions
• 2 ways to address
  • binary translation
    • ek ek ko dekho and trap lagado
  • paravirtualization
    • guest os ko hi modify karde
    • and hypercalls
  • both involve a lot of overhead
    • har instruction ko scan karo
    • ya context switches in hypercall

Hardware Support for Virtualization

• we want guest os level code ko h/w pe execute karna
• so h/w create 2 modes, dekh user and superuser mode the (3 and 0), 2 additional modes hai
  • VMM mode
    • VMM directly executing on h/w
  • non VMM root mode
    • guest os is executing
    • executing in root mode but VMM can program by providing VMCB (Virtual Machine Control Block) ki agar aisa kuch instructions ho, then meko trap bhej dena
    • so agar guest os does those instructions, trap a jata VMM ko and it sees context and all
    • virtualbiox uses this support

Intel Virtualization Technology Evolution

• hardware assistance can be used by 1 hypervisor at a time.

Emulation

• same interface provide karna kisi tarah (emulate karna diff interface pe)
• putty use karra main
  • it emulates Linux terminal interface on windows
• Simulation?
  • does not respect interfaces
  • tries to create env where some principles are followed
  • eg NS2 and NS3 simulators
    • I cannot run browser or socket programming in those
    • emulation preserves the interfaces

• mininet is an emulator
  • respects interfaces
  • if program works on system, it will on mininet as well

Definitions

• source ISA is emulated to target ISA

Ways to carry out emulation

• interpretation and translation
• inter
  • touch source instruction
  • analyze it (ek particular format me hoga na instruction)
  • perform opern in host
  • all done in s/w
  • looks at individual instruc
• trans
  • look at block of source instructions
  • translate to target block
  • store for future purposes
  • loop me efficient, since translate kar chuka
    • interpret me bar bar translate karunga repititive part ko

Interpreter State

• methods of doing interpretation

• interpreter
  • read source code
  • emulate on target
  • every instruction tries to modify some reg value etc
  • so program ki state hogi
  • segment and regs ki values etc
  • so it will have to store in its own memory bcz os nahi kareg ana bhai
• Model of interpreter
  • Decode and Dispatch
    • take one instr
    • interpret it = decode it
    • call corresp subroutine for that opcode
    • 
    • 
    • execLoad
      • extract first and second operand
      • get address
      • loads memory into the register
      • increment the PC
      • executing on target platform but changes are being done in interpreter state maintained by interpreter

• Branch instructions
  • testing if we get a halt or interrupt generated for process
  • switch case requires indirect branching
  • branch to interpreter code
  • branch that terminates loop iteratin

• bahut branches mtlb bt, kuki we don’t know kaunsi branch lega
• overcame by following

Indirect threaded Interpretation

• remove while loop
• jumping to subroutine hai, but going back nahi
• 3 branches removed
  • switch nahi
  • while nahi
  • subroutine se coming back nahi
• inside same subroutine, extract next instruction and ek lookup array hoga which help to know kidhar jump next (subroutine address table)

• Advantages?
  • interpretation routines can be modified and relocated independently
    • bcz lookup table hai, not hardcoded anywhere
    • change table to change location
    • improve efficiency bcz eliminates branches