Scuola Superiore Sant’Anna Operating Systems - Introduction Giuseppe Lipari
Scuola Superiore Sant’Anna
Operating Systems IntroductionGiuseppe Lipari
OS Course - SSSA - 2007 2
Fundamentals
• Algorithm:– It is the logical procedure to solve a certain problem– Informally specified a a sequence of elementary steps that an
“execution machine” must follow to solve the problem– not necessarily expressed in a formal programming language!
• Program:– It is the implementation of an algorithm in a programming
language– Can be executed several times with different inputs
• Process:– An instance of a program that, given a set of inputs values,
produces a set of outputs
OS Course - SSSA - 2007 3
Operating System
• An operating system is a program that– Provides an “abstraction” of the physical machine
through a simple interface– Each part of the interface is a “service”
• An OS is also a resource manager– With the term “resource” we denote all physical
entities of a computing machine– The OS provides access to the physical resources– The OS provides abstract resources (for example, a
file, a virtual page in memory, etc.)
OS Course - SSSA - 2007 4
Levels of abstraction
Kim Lisa Bill
Main Board CPU
Keyboard
Video Card
Network Card
Printer
Printer
Hard disk
OperatingSystem
Interface (System API)
Virtual Memory Scheduler Virtual File Sys.
Device Driver
Device Driver
Device Driver
Device Driver
Device Driver
Device Driver
Web Browser
Shell Videogame PrinterDaemon
User Level
ProgrammerLevel
SystemLevel
HW Level
OS Course - SSSA - 2007 5
Abstraction mechanisms
• Why abstraction?– Programming the HW directly has several drawbacks
• It is difficult and errorprone• It is not portable
– Suppose you want to write a program that reads a text file from disk and outputs it on the screen
• Without a proper interface it is virtually impossible!
OS Course - SSSA - 2007 6
Abstraction Mechanisms
• Application programming interface (API)– Provides a convenient and uniform way to access to
one service so that• HW details are hidden to the high level programmer• Applications do not depend on the specific HW• The programmer can concentrate on higher level tasks
– Example• For reading a file, linux and many other unix OS provide the
open(), read() system calls that, given a “file name” allow to load the data from an external support
OS Course - SSSA - 2007 7
Historical Perspective
• In the beginning was the batch processor– Huge machines, not very powerful– Used mainly for scientific computation and military
applications– Program were executed one at time
• They were called jobs
– Program were simple sequential computations• Read the input• Compute• Produce output
– Noninteractive!
OS Course - SSSA - 2007 8
Batch processor
• Batch = noninteractive• The program could not be interrupted or suspended
(nonpreemptive)• Scheduling:
– Priority based (e.g. first the military...)– FIFO– Shortest job first (SJF)
CPUProgram PunchCards
Result
jobs
OS Course - SSSA - 2007 9
Drawbacks
• CPU was inactive for long intervals of time– While reading the punch cards, the CPU had to wait– The punch card reader was very slow
• Solution: spooling– Use a magnetic disk (a faster I/O device)– Job were grouped into “job pools”– While executing one job of a pool, read the next one
into disk– When a job finishes, load the next one from the disk – Spool = symultaneous peripheral operation online
OS Course - SSSA - 2007 10
Interactivity
• The need for interaction– For reading input from the keyboard during the
computation– For showing intermediate results– For saving intermediate result on magnetic support
• Input/output– It can be done with a technique called polling
• Wait until the device is ready and get/put the data• Handshaking
– Again, the CPU was inactive during I/O operations
OS Course - SSSA - 2007 11
Multiprogramming
• The natural evolution was “concurrency”– IDEA: while a job is reading/writing from/to a I/O
device, schedule another job to execute (preemption)
CPU Result
jobs
Preemption
OS Course - SSSA - 2007 12
Multiprogramming
• Multiprogramming is very common in reallife– Consider a lawyer that has many clients
• FIFO policy: serving one client at time, from the beginning until the court sentence
• In italy, a sentence can be given after more than 10 years. Imagine a poor lawyer trying to survive with one client for ten years!
• In reality, the lawyer adopts a TIME SHARING policy!
– All of us adopts a timesharing policy when doing many jobs at the same time!
OS Course - SSSA - 2007 13
The role of the Operating System
• Structure of a multiprogrammed system– Who decides when a job is suspended?– Who decided who is to be executed next?
• In the first computers, these tasks were carried out by the application itself
• Each job could suspend itself and pass the “turn” to the next job (coroutines)
• However, this is not very general or portable!
– Today, the OS provide the multiprogramming services
• The scheduler module chooses which job executes next depending on the status of the system
OS Course - SSSA - 2007 14
ProcessSwitch
Time sharing systems
• In time sharing systems– The time line is divided into “slots”, or “rounds”, each
one of maximum length equal to a fixed time quantum
– If the executing job blocks on a I/O operation, or if the quantum finishes, it is suspended to be executed later CPU
jobs
OS Course - SSSA - 2007 15
Time sharing systems
• In time sharing systems– Each process executes approximately as it were alone on a
slower processor– The OS (thanks to the scheduler) “virtualizes” the processor
• One single processor is seen as many (slower) parallel processors (one for each process)
– We will see that an OS can virtualize many HW resources• Memory, disk, network, etc
• Time sharing systems are not predicatable– The amount of execution time received by one process
depends on the number of processes in the system– If we want predictable behavior, we must use a RTOS
OS Course - SSSA - 2007 16
Multiuser systems
• The first computers were very powerful and very expensive– An university could afford only one mainframe, but
many people needed to access the same computer– Therefore, the mainframe would give simultanous
access to many users at the same time– This is an obvious extension of the multiprocess
system• One or more processes for each user
OS Course - SSSA - 2007 17
Multiuser systems
• The terminals had no computing power– A keyboard + a monitor + a serial line– Every computation was carried out in the mainframe– It is like having one computer with many keyboards and
videos
Mainframe
DumbTerminal
DumbTerminal
DumbTerminal
OS Course - SSSA - 2007 18
Multiuser system
• Another dimension was necessary– The concept of user and account was born– The first privacy concerns were raised
• Access rules• Passwords• Criptography was applied for the first time in a nonmilitary
environment!
– This makes the system more complex!
OS Course - SSSA - 2007 19
Distributed systems
• Finally, distribution was introduced– Thanks to the DARPA, the TCP/IP protocol was
developed and internet was born• The major universities in the USA connected their
mainframes• Mail, telnet, ftp, etc• The natural evolution was internet and the world wide web
– All of this was possible thanks to• The freedom of circulation of ideas• The “liberal” environment in universities• The need for communication and sharing information
OS Course - SSSA - 2007 20
Distributed systems
• More flexibility– Client/server architectures
• One server provides “services” to remote clients• Example: web, ftp, databases, etc
– It is possible to “distribute” an application• Different “parts” execute on different computers and then
communicate each other to exchange information and synchronise
• Massively parallel programs can be easily implemented– Migration
• Processes can “move” from one computer to another to carry out a certain service
• Examples: agents, videogames, applets, etc
OS Course - SSSA - 2007 21
Classification of Operating Systems
• The OS provides an abstraction of a physical machine– To allow portability– To make programmer’s life easier
• The level of abstraction depends on the application context– the kind of services an OS provides depend on which kind of
services the application requires• General purpouses OS should provide a wide range of services to
satisfy as many users as possible• Specialised OS provide only a limited set of specialised services
– OS can be classified depending on the application context• General purpouse (windows, linux, etc), servers, microkernel,
embedded OS, realtime OS
OS Course - SSSA - 2007 22
Services
• Virtual processor– An OS provides “concurrency” between processes
• Many processes are executed at the same time in the same system
• Each process executes for a fraction of the processor bandwidth (as it were on a dedicated slower processor)
– Provided by the scheduling subsystem– Provided by almost all OS, from nanokernels to
generalpurpouse systems
CPU CPU CPU CPU
OS Course - SSSA - 2007 23
Services
• Virtual memory– Physical memory is limited;– In old systems, the number of concurrent
processes was limited by the amount of physical memory
– IDEA: extend the physical memory by using a “fast” mass storage system (disk)
• Some of the processes stay in memory, some are temporarily saved on the disk
• When a process must be executed, if on the disk, it is first loaded in memory and then executed
• This technique is called “swapping”
OS Course - SSSA - 2007 24
Virtual memory and physical memory
• Virtual memory is very large (virtually infinite!)• The program functionality does not depend on the size of the memory• The program performance could be reduced by the swapping
mechanism
Process E
Process D
Process C
Process B
Process A
Process E
Process A
Process C
CPUACEBD
B
D
DiskPhysical memoryVirtual memory
A CEBDA CEBD
Process B A
Process B
Process A
OS Course - SSSA - 2007 25
Virtual Memory
• Advantages– Virtual infinite memory– The program is not limited by the size of the physical
memory
• Disadvantages– If we have too many programs, we spend most of the
time swapping back and forth– Performance degradation!– Not suitable for realtime systems
• It is not possible to guarantee a short response time because it depends on the program location
OS Course - SSSA - 2007 26
Virtual File System
• Basic concepts– File: sequence of data bytes
• It can be on a mass storage (hard disk, cdrom, etc.)• It can be on special virtual devices (i.e. RAM disks)• It can be on a remote system!
– Directory: list of files• Usually organised in a tree• Represents how files are organised on the mass storage system
• Virtualisation– In most OS, external serial devices (like the console or the
video terminal) can be seen as files (i.e. stdin, stout , stderr)
OS Course - SSSA - 2007 27
Virtual file system
• A good virtual file system provides additional features:– Buffering & caching
• For optimising I/O from block devices
– Transactions• For example the Reiser FS
– Fault tolerance capabilities• For example, the RAID system
• Virtual file system is not provided by all OS categories– Micro and nano kernels do not even provide a file system!
OS Course - SSSA - 2007 28
Privacy and access rules
• When many users are supported– We must avoid that nonauthorised users access
restricted information– Usually, there are two or more “classes” of users
• Supervisors• Normal users
– Each resource in the system can be customised with proper “access rules” that prevent access from nonauthorised users
• For example, the password file should be visible only to the system supervisor
Scuola Superiore Sant’Anna
Overview of hardware architectures
OS Course - SSSA - 2007 30
Basic blocks
CPU
BUS
MainMemory
Other I/O devices
Disk keyboard Video
OS Course - SSSA - 2007 31
The processor
• Set of registers– IP: instruction pointer– SP: stack pointer– A0A3: general registers– CR: control register
• Execution unit– Arithmetic unit– Fetching unit– Branch prediction unit
• Other components– Pipeline– Cache
CPU
IP
SP
R0
R1
R2
R3
ExecutionUnits
CR
OS Course - SSSA - 2007 32
Processor registers
• User visible registers– Used as temporary buffers for processor operations– Can be in any number
• RISC architectures: array of registers• CISC architectures: set of registers dedicated to specific operations
• Control and Status registers– IP Instruction pointer– SP Stack Pointer– CR Control Register (or PSW Program Status Word)
OS Course - SSSA - 2007 33
Modes of operation
• Many processors have at least two modes of operation– Supervisor mode
• All instructions are allowed • Kernel routines execute in supervisor mode• The OS must access all features of the system
– User mode• Not all instructions are allowed• User programs execute in user mode• Some instruction (for example, disabling interrupts) cannot be
invoked directly by user programs
• Switching– It is possible to switch from user mode to supervisor mode with
special instructions
OS Course - SSSA - 2007 34
Main Memory and bus
• The RAM– Sequence of data locations– Contains both instructions (TEXT) and data variables
• The bus– A set of “wires”
• Address wires• Data wires
– The number of data wires is the amount of bits that can be read with one memory access
• Current PC buses: 32 bit
OS Course - SSSA - 2007 35
Instruction execution
• We distinguish at least two phases– Fetching: the instruction is read from the memory– Execute: the instruction is executed
• Data processing instr. – the result is stored in registers• Load instr. – the data is loaded from main memory• Store – the data is stored in main memory• Control – the flow of execution may change (change IP)
– Some instruction may be the combination of different types
Start HaltFetch nextinstruction
Executeinstruction
OS Course - SSSA - 2007 36
StackFrame
Stack Frames
• The stack is used to– Save local variables– Implement function calling
• Every time a function is called– The parameters are saved on the stack– Call <address>: The current IP is saved on
the stack– The routine saves the registers that will be
modified on the stack– local variables are defined on the stack– When the function is over the stack is
cleaned and the RET instruction is called which restores IP
Stack
Parameters
IP
R0R1R2xy
BP
OS Course - SSSA - 2007 37
External devices
• I/O devices– Set of data registers – Set of control registers– mapped on certain
memory locations
D0 CR0
CR1
CR2
D1
D2
I/O device interface
BUS
CPUIP
SP
R0
R1
R2
R3CR
MemoryA3B0A3B2A3B4A3B6A3B8A3BAA3BC
…
FF00
FF02
FF04
FF06
FF08
FF0A
OS Course - SSSA - 2007 38
I/O operations
• Structure of an I/O operation– Phase 1: prepare the device for the operation
• In case of output, data is transferred to the data buffer registers
• The operation parameters are set with the control registers• The operation is triggered
– Phase 2: wait for the operation to be performed• Devices are much slower than the processor• It may take a while to get/put the data on the device
– Phase 3: complete the operation• Usually, cleaning up the control registers
OS Course - SSSA - 2007 39
Example of input operation
• Phase 1: nothing• Phase 2: wait until bit 0 of CR0 becomes 1• Phase 3: read data from D0 and reset bit 0 of CR0
BUS
CPUIP
SP
R0
R1
R2
R3CR
D0 CR0
CR1
CR2
D1
D2
I/O device interface
FF00
FF02
FF04
FF06
FF08
FF0A
CR0D0 CR0D0 R0
OS Course - SSSA - 2007 40
Example of output operation
• Phase 1: write data to D1 and set bit 0 of CR1• Phase 2: wait for bit 1 of CR1 to become 1• Phase 3: clean CR1
BUS
CPUIP
SP
R0
R1
R2
R3CR
D0 CR0
CR1
CR2
D1
D2
I/O device interface
FF00
FF02
FF04
FF06
FF08
FF0A
CR1D1D1
R0CR1
R0CR1
OS Course - SSSA - 2007 41
Temporal diagram
• Polling– This technique is called “polling” because the
processor “polls” the device until the operation is completed
– In general, it can be a waste of time– The processor can executed something useful
while the device is working– How the processor can know when the device
has completed the I/O operation?
OS Course - SSSA - 2007 42
Interrupts
• Every processor supports an interrupt mechanism– The processor has a special pin, called “interrupt request (IRQ)”– Upon reception of a signal on the IRQ pin,
• If interrupts are enabled, the processor suspends execution and invokes an “interrupt handler” routine
• If interrupts are disabled, the request is pending and will be served as soon as the interrupts are enabled
Start HaltFetch nextinstruction
Executeinstruction
Interrupts?Serve
Interrupt
OS Course - SSSA - 2007 43
Interrupt handling
• Every interrupt is associated one “handler”
• When the interrupt arrives– The processor suspend what is doing– Pushes CR on the stack– Calls the handler (pushes the IP on the
stack)– The handler saves the registers that will be
modified on the stack– Executes the interrupt handling code– Restores the registers– Executes IRET (restores IP and CR)
Stack
CR
IP
R0
R1
OS Course - SSSA - 2007 44
Input with interrupts
• Phase 1: do nothing• Phase 2: execute other code• Phase 3: upon reception of the interrupt, read data from
D0, clean CR0 and return to the interrupted codeBUS
CPUIP
SP
R0
R1
R2
R3CR
D0 CR0
CR1
CR2
D1
D2
I/O device interface
FF00
FF02
FF04
FF06
FF08
FF0A
CR0D0 CR0D0 R0
IRQ
OS Course - SSSA - 2007 45
Interrupts
• Let’s compare polling and interrupt
Normalcode
Interrupthandler Phase 1
Phase 2
Phase 3
Pollingcode
OS Course - SSSA - 2007 46
The meaning of phase 3
• Phase 3 is used to signal the device that the interrupt has been served– It is an handshake protocol
• The device signal the interrupt• The processor serves the interrupt and exchange the data• The processor signal the device that it has finished serving
the interrupt• Now a new interrupt from the same device can be raised
OS Course - SSSA - 2007 47
Interrupt disabling
• Two special instructions– STI: enables interrupts– CLI: disables interrupts– These instructions are privileged
• Can be done only in supervisor mode
– When an interrupt arrives the processor goes automatically in supervisor mode
Normalcode
Interrupthandler
CLI STI
PendingInterrupt
OS Course - SSSA - 2007 48
Many sources of interrupts
• Usually, processor have one single IRQ pin– However, there are several different I/O devices– Intel processors use an external Interrupt Controller
• 8 IRQ input lines, one output line
BUS
CPUIRQ
IRQ0IRQ1IRQ2IRQ3IRQ4IRQ5IRQ6IRQ7
I/ODevice
I/ODevice
Int e
rru
pt
Co
ntr
oll
er
OS Course - SSSA - 2007 49
Nesting interrupts
• Interrupt disabling– With CLI, all interrupts are disabled
• When an interrupt is raised, – before calling the interrupt handler, interrupts are
automatically disabled– However, it is possible to explicitely call STI to re
enable interrupts even during an interrupt handler– In this way, we can “nest interrupts”
• One interrupt handler can itself be interrupted by another interrupt
OS Course - SSSA - 2007 50
Interrupt controller
• Interrupts have priority– IRQ0 has the highest priority, IRQ7 the lowest
• When an interrupt from a I/O device is raised– If there are other interrupts pending
• If it is the highest priority interrupt, it is forwarded to the processor (raising the IRQ line)
• Otherwise, it remains pending, and it will be served when the processor finishes serving the current interrupt
OS Course - SSSA - 2007 51
Nesting interrupts
• Why nesting interrupts?– If interrupts are not nested, important services many
be delayed too much• For example, IRQ0 is the timer interrupt• The timer interrupt is used to set the time reference of the
system• If the timer interrupt is delayed too much, it can get lost (i.e.
another interrupt from the timer could arrive before the previous one is served)
• Losing a timer interrupt can cause losing the correct time reference in the OS
• Therefore, the timer interrupt has the highest priority and can interrupt everything, even another “slower” interrupt
OS Course - SSSA - 2007 52
Nested interrupts
Normalcode
Slow Interrupthandler
High priority Interrupt handler
OS Course - SSSA - 2007 53
Atomicity
• An hardware instruction is atomic if it cannot be “interleaved” with other instructions– Atomic operations are always sequentialized– Atomic operations cannot be interrupted
• They are safe operations• For example, transferring one word from memory to register
or viceversa
– Non atomic operations can be interrupted• They are not “safe” operations• Non elementary operations are not atomic
OS Course - SSSA - 2007 54
Non atomic operations
• Consider a “simple” operation like x = x+1;
In assembler LD R0, x
INC R0ST x,RO
A simple operation like incrementing a memory variable, may consist of three machine instructions
If the same operation is done inside an interrupt handler, an inconsistency can arise!
OS Course - SSSA - 2007 55
Interrupt on nonatomic operations
int x=0;
...x = x + 1;...
Normal codevoid handler(void){ ... x = x + 1; ....}
Handler code
...LD R0, xINC R0ST x, RO...
Save registers...LD R0, xINC R0ST x, RO...Restore registers
?R0
0x
CPU
memory
0
Saved registers
0
01
1
01
1
OS Course - SSSA - 2007 56
Solving the problem in single processor
• One possibility is to disable interrupts in “critical sections”
...CLILD R0, xINC R0ST x, ROSTI...
Save registers...LD R0, xINC R0ST x, RO...Restore registers
OS Course - SSSA - 2007 57
Multiprocessor systems
• Symmetric multiprocessors (SMP)– Identical processors– One shared memory
CPU 0 CPU 1 CPU 2 CPU 3
Memory
OS Course - SSSA - 2007 58
Multiprocessor systems
• Two typical organisations– Master / Slave
• The OS runs on one processor only (master), CPU0• When a process requires a OS service, sends a message to
CPU0
– Symmetric• One copy of the OS runs indipendentely on each processor• They must synchronise on common data structures• We will analyse this configuration later in the course
OS Course - SSSA - 2007 59
Low level synchronisation in SMP
• The atomicity problem cannot be solved by disabling the interrupts!– If we disable the interrupts, we protect the code from
interrupts. – It is not easy to protect from other processors...
LD R0, xINC R0ST x, RO...
...LD R0, xINC R0ST x, RO...
...LD R0, x (CPU 0)LD R0, x (CPU 1)INC R0 (CPU 0)INC R0 (CPU 1)ST x, R0 (CPU 0)ST x, R0 (CPU 1)...
CPU 0
CPU 1
OS Course - SSSA - 2007 60
XCH and TST are atomic!
Low level synchronisation in SMP
• Most processors support some special instruction– XCH Exchange register with memory location– TST If memory location = 0, set location to 1 and
return true (1), else return false (0)void xch(register R, memory x){
int tmp;tmp = R; R = x; x=tmp;
}
int tst(int x){
if (x == 1) return 0;else {
x=1;return 1;
}}
OS Course - SSSA - 2007 61
Locking in multiprocessors
• We define one variable s– If s == 0, then we can perform the critical operation– If s == 1, the must wait before performing the critical operation
• Using XCH or TST we can implement two functions:– lock() and unlock()void lock(int s)
{int a = 1;while (a==1) XCH (s,a);
}
void lock(int x){
while (TST (s) == 0);}
void unlock(int s){
s = 0;}
OS Course - SSSA - 2007 62
Locking in multiprocessorsL0: TST s
JZ L0LD R0, xINC R0ST x, R0LD R1, 0ST s, R1
...
TST s (CPU 0)TST s (CPU 1)JZ L0 (CPU 0)JZ L0 (CPU 1)LD R0, x (CPU 0)TST s (CPU 1)INC R0 (CPU 0)JZ L0 (CPU 1)ST x, R0 (CPU 0)TST s (CPU 1)LD R1, 0 (CPU 0)JZ L0 (CPU 1)ST s, R1 (CPU 0)TST s (CPU 1)... (CPU 0)JZ L0 (CPU 1)... (CPU 0)LD R0, x (CPU 1)
CPU 0
CPU 1
L0: TST sJZ L0LD R0, xINC R0ST x, ROLD R1, 0ST s, R1
...
Lock(s)
Unlock(s)
Unlock(s)
Lock(s)
x=x+1
x=x+1
OS Course - SSSA - 2007 63
Locking
• The lock / unlock operations are “safe”– No matter how you interleave the operations, there is
no possibility that the “critical parts interleave– However, lock() is an active wait and a possible wast
of time
• The problem of locking is very general• Solutions will be presented and analysed in
greater details later in the course