Linux Overview

1
Linux Overview

• Anand Sivasubramaniam
• Dept. of Computer Science Eng.
• The Pennsylvania State University

2
Why Linux?

• Its free!
• Open Source (modifiability, extensibility, )
• Works on several platforms
• Robustness (after several revisions, and several
  people working on it)
• Widespread Usage
• Compatibility with several other platforms.

3
Coverage from

• M. Beck, H. Bohme, M. Dziadzka, U. Kunitz, R.
  Magnus, D. Verworner. Linux Kernel Internals, 2nd
  edition, Addison-Wesley.
• R. Card, E. Dumas, F. Mevel. The Linux Kernel
  book. John-Wiley.
• Mainly Linux 2.0

4
Linux Features

• Monolithic kernel (but well-defined interfaces)
• Multi-tasking
• Multi-user capability
• Multi-processing Support (since 2.0)
• Architecture Independence (PCs, Alpha, Sparc,)
• Demand loaded executables (on fork, shared
  address space, and copy-on-write)
• 4K Pages, demand-paging with memory protection
• Dynamic size for disk cache
• Shared Libraries (dll)
• Support for Posix standard
• Several Executables formats
• Several File Systems (Ext2)
• Several network protocols

5
/usr/src/linux hierarchy
6

• arch/ contains architecture specific code
• arch/i386/boot contains init code (in assembly)
  to initialize h/w, load the kernel, install ISRs,
  switch to protected mode and then calls
  start_kernel(void) in init/
• Kernel/ and arch/i386/kernel/ contain the core
  kernel code (fork,scheduler, timers, DMA and
  interrupt management, signal handling, switching
  protection modes)
• VM system, memory allocation, paging are in mm/
  and arch/i386/mm.
• The virtual file system interface is in /fs, and
  the subdirectories contain respective file
  systems
• drivers/ contains different drivers.
• ipc/ has sources for IPC (semaphores, shared
  memory, message queues)
• net/ has different protocol codes
• lib/ contains standard C libraries
• Include/ and asm/include have the necessary
  include files with /usr/include having links to
  these.

7
Building

• Make config (reads arch/i386/config.in file, to
  find out which components to include, which
  inturn consults config.in in other directories.
• Config.in contains directives on what packages to
  include.
• Make depend
• Make boot (gives a bootable kernel
  arch/i386/boot/zImage).
• Make bzlilo (copies bootable kernel to /vmlinuz
  and the kernel can then be installed using LILO)
• Make drivers (to just compile drivers)
• Make modules (file systems and drivers not linked
  in can be created as modules)
• Make modules_install (the created modules are
  installed in the /lib/modules/kernel_version
  directory)

8
Booting

• On power up, CPU is reset and PC is set to a
  certain value in the BIOS ROM.
• BIOS does hardware tests and initializations.
• BIOS tries to read first sector (boot sector) of
  disk (first floppy then HD)
• This sector is read into memory and is of a
  pre-determined format
• 0x000 Jmp xxx
• 0x003 Disk Params
• 0x03E Code
• 0x1BE Partition 1 entry
• 0x1CE Partition 2 entry
• 0x1DE Partition 3 entry
• 0x1EE Partition 4 entry
• 0x1FE Magic Number
• PC is then set to the first location, which then
  goes to the code that checks which is he active
  partition. That entry has the sector number of
  the boot block of that partition which looks
  similar and has the code to boot that OS (part of
  LILO for Linux).

9

• After loading the kernel, it jumps to
  arch/i386/boot/setup.S (start)
• This code initializes and establishes hardware,
  and then switches to Protected Mode by setting a
  bit in the machine Status Word.
• Executes jmp 0x1000, KERNEL_CS
• Goes to arch/i386/kernel/head.S (startup_32)
• This does MMU, coprocessor and interrupt
  descriptor initializations, and sets the
  environment for the kernel C functions.
  Subsequently start_kernel() in init/main.c is
  called.

10

• Start_kernel(void)
• Memory_start paging_init(memory_start,memory_end
  )
• Trap_init()
• Init_IRQ()
• Sched_init()
• Time_init()
• Parse_options(command_line)
• Init_modules()
• Memory_start console_init(memory_start,memory_en
  d)
• Memory_start pci_init(memory_start,memory_end)
• Memory_start kmalloc_init(memory_start,memory_en
  d)
• Sti()
• Memory_startinode_init(memory_start,memory_end)
• Memory_startfile_table_init(memory_start,memory_e
  nd)
• Memory_startname_cache_init(memory_start,memory_e
  nd)
• Mem_init(memory_start,memory_end)
• Buffer_init()
• Sock_init()
• Ipc_init()

11

• Process 0 is now running. It generates a kernel
  thread which executes the init function
• Kernel_thread(init,NULL,0)
• Process 0 executes idle process
• Cpu_idle(NULL)
• Init()
• kernel_thread(bdflush,NULL,0) // buffer cache
  sync daemon
• kernel_thread(kswapd,NULL,0) // swap daemon
• setup() // init file sys, and mount root
• open console, file descriptors 0, 1 and 2
• execve(../init,argv_init,envp_init) //
  getty runs on each tty

12
Adding a System Call

• show_mult(x, y, z)
• Each sys call has a name and a number
• Go to /asm/unistd.h
• define __NR_sched_r_get_interval 161
• define __NR_nanosleep 162
• define __NR_mremap 163
• define __NR_show_mult 164
• Go to arch/i386/kernel/entry.S
• .long SYMBOL_NAME(sys_nanosleep)
• .long SYMBOL_NAME(sys_mremap)
• .long SYMBOL_NAME(sys_showmult)
• .space (NR_syscalls-164)4 // padding

13

• Now to add the code for the syscall, say in
  /sys.c (which has other syscalls)
• asmlinkage int sys_show_mult(int x, int y, int
  res)
• int error, compute
• error verify_area(VERIFY_WRITE, res,
  sizeof(res))
• if (error) return error
• compute xy
• put_user(compute,res)
• printk(Value computed)
• return 0
• Compile kernel and reboot the machine.

14

• How do you use this? If it is already defined in
  a library, then fine.
• Else, you can use a macro definition
• _syscall3 (int, show_mult, int, x, int, y, int ,
  resul)
• Which expands as
• int show_mult(int x, int y, int resul)
• long __res
• __asm__ __volatile (int 0x80
• a (__res)
• 0 (164),
• b ((long) (x)),
• c ((long) (y)),
• d ((long) (resul)))
• if (__resgt0) return (int) __res
• errno -__res
• return 1
• Which places sys call in eax register,
  parameters in ebx, ecx and edx, and then invokes
  software interrupt 0x80.

15

• Upon this trap, the function system_call in
  arch/i386/kernel/entry.S is invoked.
• This uses the syscall (in eax) to index the
  table sys_call_table, to call the corresponding
  function.
• User program is as follows
• include ltstdio.hgt
• include ltstdlib.hgt
• include ltlinux/unistd.hgt
• _syscall3 (int, show_mult, int, x, int, y, int ,
  resul)
• main()
• int ret0
• show_mult(2, 5, ret)

16
What happens on a syscall?

• On the software interrupt (0x80), the control is
  transferred to system_call() in
  arch/i386/kernel/entry.S
• Here is what goes on inside this routine
• SAVE_ALL // saves registers
• sys_call_tablesys_call_num(sys_call_args)
• if (intr_count) goto exit_now // nested
  interrupts
• if (bh_mask bh_active)
• intr_count
• Sti()
• Do_bottom_half()
• --intr_count
• sti()

17

• if (need_resched)
• schedule() // return much later!
• goto ret_from_sys_call
• if (current-gtsignal current-gtblocked)
  do_signal()
• exit_now RESTORE ALL // return using iret

18
Basics of Interrupt Handling

• In arch/i386/kernel/irq.c and include/asm/irq.h
• Three types of interrupts Fast, Slow, System
  Calls (software)
• Slow Interrupts (typical), turn off interrupts
  only for a little while. E.g. timer
• SLOW_IRQ(intr_num, intr_controller, intr_mask)
• SAVE_ALL // macro in include/asm/irq.h
• ENTER_KERNEL // exclusive execn. In kernel
  for SMP
• ACK(intr_controller,intr_mask)
• intr_count // nesting depth
• Sti() // enable interrupts
• Do_IRQ(intr_num, Register) // do actual ISR
• Cli()
• UNBLK(intr_controller,intr_mask)
• --intr_count
• Ret_from_sys_call()

19

• FAST_IRQ(intr_num,intr_controller,intr_mask)
• SAVE_MOST // macro in include/asm/irq.h
• ENTER_KERNEL
• ACK(intr_controller,intr_mask)
• int_count
• Do_fast_IRQ(intr_num)
• UNBLK(intr_controller,intr_mask)
• --intr_count
• LEAVE_KERNEL
• RESTORE_MOST

20
Timer Interrupt

• 1 tick 10 ms, 100 interrupts every second