Lecturee 21

• Lecturee 21
  • Video
  • X86 Virtualization
  • Full Virtualization
  • Paravirtualization
  • Xen I/O
  • I/O Rings
  • Xen Event Model
  • Communicationg with shared memory
  • Split Device Driver Model
  • Hardware Virtualization
  • Nested Page Tables

Video

link

X86 Virtualization

• bahut pehle bhi tha vvirtualization (classical)
• architecture used at that time was trap and emulate
• any priveledge instr was to generate trap
• then emulate it
• but x86 me critical instr bhi hote
• so trap and emulare hi nahi chalega
• so binary translate > trap > emulate (scan, patch, trap, emulate)

Full Virtualization

• OS is nto modified
• it is unaware of virtualization
• it assumes sab kuch hai uske pas

• sometime guest os wants to go according to real time (calendar time)
• to get that, it will be diff bcz u cannot see actual hardware
• natural time lapse hai
• so real and virtual time hai

Paravirtualization

• modifies guest os itself
• looks at source code, critical instructions ko patch kar dega
• xen comes up with hypercalls
• system calls 80 pe hote, hyper calls int 82 pe

• memory aspect
  • shadow tables dkehi thi
  • R->V, V->P, Shadow thi teen
  • Xen kya karta?
    • jab trap vagairha hota x86 me tlb flushed
    • but tlb hamein visible nahi, so kaise kya kare
    • xen me no shaddow page tables
    • hota mapping R->P hai direct, so diff os ek dusre ki page tables access kar sakte kya?
      • nahi, these are write protected
      • hypercall(trap) hogi, and xen will check location validity and then modify
• xen is allocated some top ki space in page table as linux me kernel hota

• Xen runs in level 0
• guest os in 1
• apps in 2
• apps cannot access guest os
• priveledged and critical instructions are replaced by hypercalls
• improves performance
• hypercall is like system call to vmm
• guest os has to call that, guest os has to be modified, but not possible hamesha (windows ka code nahi na)
• we can batch operations in hypercalls (ek saath kai page table write karle, instead of bar bar)

• Guest app system calls = normal system call
• results in trap, goes to hypervisor, given to kernel, goes to app
• kernel calls hypercall
• accelerated system call
  • cmm ko de ke fer kernel and ferse vmm and then kernel ki bajaye
  • app se kernel
• kernel me handlers de de routines for hypervisor calls

Xen I/O

• domain 0 guest
  • vm hai
  • special vm
  • priveledged kam karta kuch kuch
  • provides UI to xen (kuch config karni agar teko)
  • provides device drivers (so all I/O ko route through this)
  • has direct access to devices through hypercalls
  • hypercalls has to come through domain 0 only
• domain u
  • cannot issue hypercalls

I/O Rings

• shmem create kardeta xen
• guest VMs communicate to domain 0 and 0 does to h/w
• place request/data in shared buffer and can be picked by domain 0

Xen Event Model

• guest os registers some signal handler for event to finish

Communicationg with shared memory

Split Device Driver Model

Hardware Virtualization

• no need to run binary translation, etc, intel ne itne kardia
• vmm root mode and non root mode
  • vmm running in root0

• vmm program h/w ki kab trap bhejna, uske alaawa nahi

Nested Page Tables

• shadow page table is overhead
• how h/w support this