Hardware.md

Genesis From Hardware To Process

• John von Neumann invented von Neumann computer architecture
  • CPU
  • Memory Unit
  • I/O devices (eg. SSD and disks)
  • Programs are stored on storage devices
  • Programs copied to memory for execution
  • CPU reads each
• Fetch-execute algorithm
• The program counter is loaded with address of first instruction, during boot sequence
• The instruction register is loaded with the instruction from that address

Booting Sequence

• Address of first instruction is fixed
• Stored in Read Only Memory (ROM)
• On i386 machines, ROM stores a Basic Input/Output system (BIOS)
• This was replaced with the United Extended Firmware Interface (UEFI)
  • Needed for accessing storage greater than 2TB

Booting

• Performs Power-On Self Test (POST)
  • Checks memory and device for their presence and correct operations
  • Old computers you can hear memory counting (noises from hard drive and CDROM).
• The master boot record (MBR)
  • loaded from the boot device
  • stored at first logical sector on boot device
  • fits into single 4k byte disk sector (boot sector)
  • Describes physical layout of the disk
• After getting info on boot device
  • BIOS loads a more sophisticated loader from other sectors on disk
  • Loader loads OS

OS Loaders

• Old linux loader, LILO (Linux Loader)
  • Partly stored in MBR with the disk partition table
  • User can specify which disk partition and OS image to boot
  • After loading kernel image, LILO sets the kernel mode and jumps to entry point
• Linux uses GRUB now (GRand Unified Bootloader)

Booting Sequence Brief

• CPU Jumps to Fixed Address in ROM
• Loads the BIOS (UEFI)
• Performs POST
• Loads the MBR (GPT) from boot device
• Loads kernel image
• Sets kernel mode

Linux Initialization

• Trap table
• Interrupt handlers
• Scheduler
• Clock
• Kernel modules
• Process manager

Process 1

• Initialized from init (now systemmd for parallelism)
• ancestor of all processes
• Controls transitions between runlevels
• Executes startup and shutdown scripts for each runlevel

Runlevels

0. shutdown
1. single-user
2. multi-user
3. full multi-user
4. X11

Process Creation

• via fork system call family

System Calls

• Allow us processes running at user mode to access kernel functions that run under kernel mode
• Prevent processes from doing malicious operations
  • Halting system
  • Modifying MBR
• Implemented through trap instructions

Parent Process

pid_t pid;
if((pid = fork()) == 0) {
	while(1){
		print("childs return value %d: I want to play.\n", pid);
	}
} else {
	while(1) {
		print("parent's return value %d: After the project.\n", pid);
	}
}

Child Process

pid_t pid;
if((pid = fork()) == 0) {
	while(1){
		print("childs return value %d: I want to play.\n", pid);
	}
} else {
	while(1) {
		print("parent's return value %d: After the project.\n", pid);
	}
}

Output

childs return value 0: I want to play.\n
childs return value 0: I want to play.\n
childs return value 0: I want to play.\n
...// context switch
parent's return value 3218: After the project.
parent's return value 3218: After the project.
parent's return value 3218: After the project.
...// context switch
childs return value 0: I want to play.\n
childs return value 0: I want to play.\n
childs return value 0: I want to play.\n

Why Clone a Process?

• Simplifying parameter passing
  • Environment Variables, Permissions, etc
• Performance optimization
  • Copy on write
• A fork by itself is not very useful
• To make a process run a program we use exec system call
• exec starts a program by overwriting the current process

Thread Creation

• use pthread_create() instead of using fork()
• Newly created thread shares current process address space and resources
• Efficient sharing of states
• Potential corruption