Threads, Introduction to synchronization

Download Report

Transcript Threads, Introduction to synchronization

W4118 Operating Systems
Instructor: Junfeng Yang
Logistics

Homework 2



Who has problems getting the VM?
This assignment requires changes to some assembly
code; we’ll update webpage to include the assembly
changes
Any questions?
Last lecture


Process (cont)

Process termination

Interprocess communication
• Zombie
• Orphan
• Message passing: signal, pipe
• Shared memory: shmget
Processes in Linux

Data structures

Process operations
• Process control block: task_struct
• kernel stack
• Connections between them, thread_info
• fork
• exit()
Today

Processes in Linux


Context switch on x86
Threads



Why and what?
Multithreading Models
Intro to synchronization
Context switch call chain





schedule() (kernel/sched.c) (talk about scheduling later)
context_switch()
swtich_mm (include/asm-i386/mmu_context.h)
switch address space
switch_to (include/asm-i386/system.h)
switch stack, regs, and %eip
__swtich_to (arch/i386/kernel/process.c)
architecture dependent setup
Context switch by stack swtich: the idea

Kernel stack captures process states



Registers
task_struct through thread_info through %esp
Changing the stack pointer changes the
process
esp
Task’s
process-descriptor
Task’s kernel-stack
Task’s thread-info
Context switch by stack switch: the
implementation (simplified)
P0 stack
eax
P1 stack
eax
…
…
ret_addr
thread_info
thread_info
esp
esp
eip
eip
swtich_to(p0,p1)
save registers on stack
p0->esp = %esp
p0->eip = ret_addr;
%esp = p1->esp;
push p1->eip;
ret
ret_addr:
pop registers from stack
include/asm-i386/system.h
p0->eip = ret_addr
esp
eip
eax
…
CPU
Today

Processes in Linux


Context switch on x86
Threads



Why and what?
Multithreading Models
Intro to synchronization
Why threads?

Express concurrency
 Web server (multiple requests), Browser (gui +
network I/O), …
for(;;) {
int fd = accept_client();
create_thread(process_request, fd);
}

Efficient communication
 Using a separate process for each task can be
heavyweight
• Costly: process creation, context switch, communication
• In general, having a separate kernel anything may be
heavyweight: must trap into kernel for service
Threads

Threads: separate streams of executions that
share an address space

Allows one process to have multiple point of
executions, can potentially use multiple CPUs

Thread control block (TCB): PC, regs, stack, …

Very similar to processes, but different
Single and Multithreaded Processes
Threads in one process share code, data, files, …
Threads vs. Processes







A thread has no data segment
or heap
A thread cannot live on its
own, it must live within a
process
There can be more than one
thread in a process, the first
thread calls main & has the
process’s stack
Inexpensive creation
Inexpensive context
switching
Efficient communication
If a thread dies, its stack is
reclaimed
•
•
A process has code/data/heap &
other segments
A process has at least one
thread
•
Threads within a process share
code/data/heap, share I/O, but
each has its own stack &
registers
•
•
Expensive creation
Expensive context switching
•
Interprocess communication can
be expressive
If a process dies, its resources
are reclaimed & all threads die
•
How to use threads?

Use thread library


E.g. pthread, Win32 thread
Common operations




create/terminate
suspend/resume
priorities and scheduling
synchronization
Example pthread functions

int pthread_create(pthread_t *thread, const pthread_attr_t
*attr, void *(*start_routine)(void*), void *arg);
 Create a new thread to run start_routine on arg
 thread holds the new thread’s id

int pthread_join(pthread_t thread, void **value_ptr);
 Wait for thread termination, and retrieve return value in
value_ptr

void pthread_exit(void *value_ptr);
 Terminates the calling thread, and returns value_ptr to
threads waiting in pthread_join
pthread creation example
void* thread_fn(void *arg)
{
int id = (int)arg;
printf("thread %d runs\n", id);
return NULL;
$ gcc –o threads threads.c –Wall –lpthread
}
$ threads
int main()
thread 1 runs
{
thread 2 runs
pthread_t t1, t2;
pthread_create(&t1, NULL, thread_fn, (void*)1);
pthread_create(&t2, NULL, thread_fn, (void*)2);
pthread_join(t1, NULL);
pthread_join(t2, NULL);
return 0;
One way to view threads: function calls,
}
except caller doesn’t wait for callee;
instead, both run concurrently
Today

Processes in Linux


Context switch on x86
Threads



Why and what?
Multithreading Models
Intro to synchronization
Multithreading Models



Where to support threads?
User threads: thread management done by
user-level threads library, typically without
knowledge of the kernel
Kernel threads: threads directly supported by
the kernel

Virtually all modern OS support kernel threads
User vs. Kernel Threads
Example from Tanenbaum, Modern Operating Systems 3 e,
(c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639
User vs. Kernel Threads (cont.)


Pros: fast, no system call for
creation, context switch
• Cons: slow, kernel does
creation, scheduling, etc
Cons: kernel unaware, so can’t
schedule  one thread
blocks, all blocks
• Pros: kernel knows, complete
flexibility  one thread
blocks, schedule another
No free lunch!
Multiplexing User-Level Threads

A thread library must map user threads to kernel threads

Big picture:



kernel thread: physical concurrency, how many cores?
User thread: application concurrency, how many tasks?
Different mappings exist, representing different tradeoffs
 Many-to-One: many user threads map to one kernel
thread, i.e. kernel sees a single process
 One-to-One: one user thread maps to one kernel thread
 Many-to-Many: many user threads map to many kernel
threads
Many-to-One

Many user-level threads
mapped to single kernel
thread

Why good?
• Fast - no system calls required
• Portable - few system
dependencies

Examples:
• Solaris Green Threads
• GNU Portable Threads

Problem
No parallel execution of threads
- can’t exploit multiple CPUs
- All threads block when one uses
synchronous I/O

One-to-One

Each user-level thread maps to kernel thread



Why good? More concurrency
• better multiprocessor performance
• When one blocks, others can run
Examples
• Windows NT/XP/2000
• Linux
• Solaris 9 and later
Problem: expensive



Each user thread requires creation of kernel thread
Each thread requires kernel resources; limits
number of total threads
Most thread operation are system calls
Many-to-Many Model

Allows many user level
threads to be mapped to
many kernel threads (U >= K)

Why good?
• Flexible. OS creates kernel
threads for physical
concurrency, and
programmers create user
threads for application
concurrency

Examples:
• Solaris prior to version 9
• Windows NT/2000 with the
ThreadFiber package

Problem? complex

Most use 1:1 mapping anyway
Two-level Model


Similar to M:M,
except that a user
thread may be
bound to kernel
thread
Examples




IRIX
HP-UX
Tru64 UNIX
Solaris 8 and
earlier
Thread Design Issues

Semantics of fork() and exec() system calls


Does fork() duplicate only the calling thread or
all threads?
Signal handling

Which thread to deliver it to?
Thread Pools


Recall web server example: create a thread for each
user request
Problem:

Thread creation also costly
• And, the created thread exits after serving a request


More user request  More threads
Solution: thread pool




Pre-create a number of threads waiting for work
Wake up thread to serve user request --- faster than
thread creation
When request done, don’t exit --- go back to pool
Limits the max number of threads
Today

Processes in Linux


Context switch on x86
Threads



Why and what?
Multithreading Models
Intro to synchronization
Banking example
int balance = 1000;
int main()
{
pthread_t t1, t2;
pthread_create(&t1, NULL, deposit, (void*)1);
pthread_create(&t2, NULL, withdraw, (void*)2);
pthread_join(t1, NULL);
pthread_join(t2, NULL);
printf(“all done: balance = %d\n”, balance);
return 0;
}
void* deposit(void *arg)
{
int i;
for(i=0; i<1e7; ++i)
++ balance;
}
void* withdraw(void *arg)
{
int i;
for(i=0; i<1e7; ++i)
-- balance;
}
Results of the banking example
$ gcc –Wall –lpthread –o bank bank.c
$ bank
all done: balance = 1000
$ bank
all done: balance = 140020
$ bank
all done: balance = -94304
$ bank
all done: balance = -191009
Why?
A closer look at the banking example
$ objdump –d bank
…
08048464 <deposit>:
…
8048473: a1 80 97 04 08
8048478: 83 c0 01
804847b: a3 80 97 04 08
…
0804849b <withdraw>:
…
80484aa: a1 80 97 04 08
80484af: 83 e8 01
80484b2: a3 80 97 04 08
…
// ++
mov
add
mov
balance
0x8049780,%eax
$0x1,%eax
%eax,0x8049780
// -- balance
mov 0x8049780,%eax
sub $0x1,%eax
mov %eax,0x8049780
One possible schedule
CPU 0
mov
time
CPU 1
0x8049780,%eax
balance: 1000
eax0: 1000
add
$0x1,%eax
mov
%eax,0x8049780
eax0: 1001
balance: 1001
eax1: 1001
eax1: 1000
mov
sub
0x8049780,%eax
$0x1,%eax
mov
balance: 1000
One deposit and one withdraw,
balance unchanged. Correct
%eax,0x8049780
Another possible schedule
CPU 0
mov
add
mov
time
CPU 1
0x8049780,%eax
balance: 1000
eax0: 1000
$0x1,%eax
eax0: 1001
%eax,0x8049780
mov
0x8049780,%eax
eax1: 1000
balance: 1001
eax1: 999
sub
mov
$0x1,%eax
%eax,0x8049780
balance: 999
One deposit and one withdraw,
balance becomes less. Wrong!
Race condition

Definition: a timing dependent error involving
shared state

Can be very bad
• “non-deterministic:” don’t know what the output will
be, and it is likely to be different across runs
• Hard to detect: too many possible schedules
– Typical: doesn’t show up in testing, but does show up in field
• Hard to debug: “heisenbug,” debugging changes timing
so hides bugs
– vs “bohr bug”

What to do???
Avoiding race conditions


Critical section: a segment of code that
accesses shared variable (or resource) and
must not be concurrently executed by more
than one thread
To implement critical sections, need atomic
operations


Atomic operations: no other instructions can be
interleaved, executed “as a unit” “all or none”
Example atomic operations


x86: load, store of words
test_and_set, compare_and_swap
Critical-Section Problem
1.
2.
Entry Section - Code
that requests
permission to enter its
critical section.
Exit Section - Code
that is run after exiting
the critical section
Critical Section Goals

Threads do some stuff but eventually might
try to access shared data
time
T1
CSEnter();
Critical section
CSExit();
T1
T2
CSEnter();
Critical section
CSExit();
T2
Critical Section Goals

Perhaps they loop (perhaps not!)
T1
CSEnter();
Critical section
CSExit();
T1
T2
CSEnter();
Critical section
CSExit();
T2
Critical Section Goals

We would like

Safety (aka mutual exclusion)
• No more than one thread can be in a critical section at any time.

Liveness (aka progress)
• A thread that is seeking to enter the critical section will eventually
succeed

Bounded waiting
• A bound must exist on the number of times that other threads are allowed
to enter their critical sections after a thread has made a request to enter
its critical section and before that request is granted
• Assume that each process executes at a nonzero speed
• No assumption concerning relative speed of the N processes

Ideally we would like fairness as well


If two threads are both trying to enter a critical section, they have
equal chances of success
… in practice, fairness is rarely guaranteed