CS262A Midterm Solutions

[Yahel Ben-David:] I’ll presume the hardware used is the easier case of MIPS. The x86 architecture complicates things as some instructions may silently fail if not run under a real “ring-0” privileges. Before delving into the flow of control for handling a system-call (syscall) on a virtualized system, let’s briefly discuss what happens in a traditional setting (without virtualization): When the running application executes an instruction requiring access to restricted resources controlled by the OS it would make a syscall. In a syscall first the arguments will be pushed onto the stack and then a special trap/interrupt will be issued to transfer control to the kernel. The interrupt handler will transfer control to a well-defined location for the system call code (established during boot-time and thereafter is well known by the hardware as the target jump location for the interrupt). At the time of the interrupt the hardware also changes the processor mode to a protected kernel mode. Let’s list a simplified form of these steps and those that follow: 1. User-space process Executes instruction requiring privileged access. 2. User-space process – makes a system call which interrupts the OS. 3. Hardware/CPU – Switch to protected kernel mode and jump to syscall code. 4. OS running in protected kernel mode – execute the syscall code and return. 5. Hardware/CPU Switch to unprotected user mode and return to execute user code. 6. User-space process – resume execution of user code. On a virtualized platform the application does exactly the same thing as in steps 1 and 2, however since it’s the VMM that controls the machine, the interrupt handler implemented in the VMM will get executed. The VMM does not know how to handle the syscall (it can’t know that for every OS) but it does know the location for the OS’s interrupt handler code. It knows that because on boot-time, the OS tried to install its own interrupt handlers through use of privileged mode and hence got trapped by the VMM which then recorded all memory locations for interrupt handlers. Since the VMM is running in protected kernel mode, it now jumps to the OS’s interrupt handler and successfully executes that code as should have been done without virtualization. Midterm Histogram