I'm reading C++ Concurrency in Action by Anthony Williams, and don't understand its push implementation of the lock_free_stack class. Listing 7.12 to be precise
void push(T const& data)
{
counted_node_ptr new_node;
new_node.ptr=new node(data);
new_node.external_count=1;
new_node.ptr->next=head.load(std::memory_order_relaxed)
while(!head.compare_exchange_weak(new_node.ptr->next,new_node, std::memory_order_release, std::memory_order_relaxed));
}
So imagine 2 threads (A, B) calling push function. Both of them reach while loop but not start it. So they both read the same value from head.load(std::memory_order_relaxed).
Then we have the following things going on:
B thread gets swiped out for any reason
A thread starts the loop and obviously successfully adds a new node to the stack.
B thread gets back on track and also starts the loop.
And this is where it gets interesting as it seems to me.
Because there was a load operation with std::memory_order_relaxed and compare_exchange_weak(..., std::memory_order_release, ...) in case of success it looks like there is no synchronization between threads whatsoever.
I mean it's like std::memory_order_relaxed - std::memory_order_release and not std::memory_order_acquire - std::memory_order_release.
So B thread will simply add a new node to the stack but to its initial state when we had no nodes in the stack and reset head to this new node.
I was doing my research all around this subject and the best i could find was in this post Does exchange or compare_and_exchange reads last value in modification order?
So the question is, is it true? and all RMW functions see the last value in modification order? No matter what std::memory_order we used, if we use RMW operation it will synchronize with all threads (CPU and etc) and find the last value to be written to the atomic operation upon it is called?
So after some research and asking a bunch of people I believe I found the proper answer to this question, I hope it'll be a help to someone.
So the question is, is it true? and all RMW functions see the last
value in modification order?
Yes, it is true.
No matter what std::memory_order we used, if we use RMW operation it
will synchronize with all threads (CPU and etc) and find the last
value to be written to the atomic operation upon it is called?
Yes, it is also true, however there is something that needs to be highlighted.
RMW operation will synchronize only the atomic variable it works with. In our case, it is head.load
Perhaps you would like to ask why we need release - acquire semantics at all if RMW does the synchronization even with the relaxed memory order.
The answer is because RMW works only with the variable it synchronizes, but other operations which occurred before RMW might not be seen in the other thread.
lets look at the push function again:
void push(T const& data)
{
counted_node_ptr new_node;
new_node.ptr=new node(data);
new_node.external_count=1;
new_node.ptr->next=head.load(std::memory_order_relaxed)
while(!head.compare_exchange_weak(new_node.ptr->next,new_node, std::memory_order_release, std::memory_order_relaxed));
}
In this example, in case of using two push threads they won't be synchronized with each other to some extent, but it could be allowed here.
Both threads will always see the newest head because compare_exchange_weak
provides this. And a new node will be always added to the top of the stack.
However if we tried to get the value like this *(new_node.ptr->next) after this line new_node.ptr->next=head.load(std::memory_order_relaxed) things could easily turn ugly: empty pointer might be dereferenced.
This might happen because a processor can change the order of instructions and because there is no synchronization between threads the second thread could see the pointer to a top node even before that was initialized!
And this is exactly where release-acquire semantic comes to help. It ensures that all operations which happened before release operation will be seen in acquire part!
Check out and compare listings 5.5 and 5.8 in the book.
I also recommend you to read this article about how processors work, it might provide some essential information for better understanding.
memory barriers
Related
The following implementation from Wikipedia:
volatile unsigned int produceCount = 0, consumeCount = 0;
TokenType buffer[BUFFER_SIZE];
void producer(void) {
while (1) {
while (produceCount - consumeCount == BUFFER_SIZE)
sched_yield(); // buffer is full
buffer[produceCount % BUFFER_SIZE] = produceToken();
// a memory_barrier should go here, see the explanation above
++produceCount;
}
}
void consumer(void) {
while (1) {
while (produceCount - consumeCount == 0)
sched_yield(); // buffer is empty
consumeToken(buffer[consumeCount % BUFFER_SIZE]);
// a memory_barrier should go here, the explanation above still applies
++consumeCount;
}
}
says that a memory barrier must be used between the line that accesses the buffer and the line that updates the Count variable.
This is done to prevent the CPU from reordering the instructions above the fence along-with that below it. The Count variable shouldn't be incremented before it is used to index into the buffer.
If a fence is not used, won't this kind of reordering violate the correctness of code? The CPU shouldn't perform increment of Count before it is used to index into buffer. Does the CPU not take care of data dependency while instruction reordering?
Thanks
If a fence is not used, won't this kind of reordering violate the correctness of code? The CPU shouldn't perform increment of Count before it is used to index into buffer. Does the CPU not take care of data dependency while instruction reordering?
Good question.
In c++, unless some form of memory barrier is used (atomic, mutex, etc), the compiler assumes that the code is single-threaded. In which case, the as-if rule says that the compiler may emit whatever code it likes, provided that the overall observable effect is 'as if' your code was executed sequentially.
As mentioned in the comments, volatile does not necessarily alter this, being merely an implementation-defined hint that the variable may change between accesses (this is not the same as being modified by another thread).
So if you write multi-threaded code without memory barriers, you get no guarantees that changes to a variable in one thread will even be observed by another thread, because as far as the compiler is concerned that other thread should not be touching the same memory, ever.
What you will actually observe is undefined behaviour.
It seems, that your question is "can incrementing Count and assigment to buffer be reordered without changing code behavior?".
Consider following code tansformation:
int count1 = produceCount++;
buffer[count1 % BUFFER_SIZE] = produceToken();
Notice that code behaves exactly as original one: one read from volatile variable, one write to volatile, read happens before write, state of program is the same. However, other threads will see different picture regarding order of produceCount increment and buffer modifications.
Both compiler and CPU can do that transformation without memory fences, so you need to force those two operations to be in correct order.
If a fence is not used, won't this kind of reordering violate the correctness of code?
Nope. Can you construct any portable code that can tell the difference?
The CPU shouldn't perform increment of Count before it is used to index into buffer. Does the CPU not take care of data dependency while instruction reordering?
Why shouldn't it? What would the payoff be for the costs incurred? Things like write combining and speculative fetching are huge optimizations and disabling them is a non-starter.
If you're thinking that volatile alone should do it, that's simply not true. The volatile keyword has no defined thread synchronization semantics in C or C++. It might happen to work on some platforms and it might happen not to work on others. In Java, volatile does have defined thread synchronization semantics, but they don't include providing ordering for accesses to non-volatiles.
However, memory barriers do have well-defined thread synchronization semantics. We need to make sure that no thread can see that data is available before it sees that data. And we need to make sure that a thread that marks data as able to be overwritten is not seen before the thread is finished with that data.
I've always been told to puts locks around variables that multiple threads will access, I've always assumed that this was because you want to make sure that the value you are working with doesn't change before you write it back
i.e.
mutex.lock()
int a = sharedVar
a = someComplexOperation(a)
sharedVar = a
mutex.unlock()
And that makes sense that you would lock that. But in other cases I don't understand why I can't get away with not using Mutexes.
Thread A:
sharedVar = someFunction()
Thread B:
localVar = sharedVar
What could possibly go wrong in this instance? Especially if I don't care that Thread B reads any particular value that Thread A assigns.
It depends a lot on the type of sharedVar, the language you're using, any framework, and the platform. In many cases, it's possible that assigning a single value to sharedVar may take more than one instruction, in which case you may read a "half-set" copy of the value.
Even when that's not the case, and the assignment is atomic, you may not see the latest value without a memory barrier in place.
MSDN Magazine has a good explanation of different problems you may encounter in multithreaded code:
Forgotten Synchronization
Incorrect Granularity
Read and Write Tearing
Lock-Free Reordering
Lock Convoys
Two-Step Dance
Priority Inversion
The code in your question is particularly vulnerable to Read/Write Tearing. But your code, having neither locks nor memory barriers, is also subject to Lock-Free Reordering (which may include speculative writes in which thread B reads a value that thread A never stored) in which side-effects become visible to a second thread in a different order from how they appeared in your source code.
It goes on to describe some known design patterns which avoid these problems:
Immutability
Purity
Isolation
The article is available here
The main problem is that the assignment operator (operator= in C++) is not always guaranteed to be atomic (not even for primitive, built in types). In plain English, that means that assignment can take more than a single clock cycle to complete. If, in the middle of that, the thread gets interrupted, then the current value of the variable might be corrupted.
Let me build off of your example:
Lets say sharedVar is some object with operator= defined as this:
object& operator=(const object& other) {
ready = false;
doStuff(other);
if (other.value == true) {
value = true;
doOtherStuff();
} else {
value = false;
}
ready = true;
return *this;
}
If thread A from your example is interrupted in the middle of this function, ready will still be false when thread B starts to run. This could mean that the object is only partially copied over, or is in some intermediate, invalid state when thread B attempts to copy it into a local variable.
For a particularly nasty example of this, think of a data structure with a removed node being deleted, then interrupted before it could be set to NULL.
(For some more information regarding structures that don't need a lock (aka, are atomic), here is another question that talks a bit more about that.)
This could go wrong, because threads can be suspended and resumed by the thread scheduler, so you can't be sure about the order these instructions are executed. It might just as well be in this order:
Thread B:
localVar = sharedVar
Thread A:
sharedVar = someFunction()
In which case localvar will be null or 0 (or some completeley unexpected value in an unsafe language), probably not what you intended.
Mutexes actually won't fix this particular issue by the way. The example you supply does not lend itself well for parallelization.
I have this POSIX thread:
void subthread(void)
{
while(!quit_thread) {
// do something
...
// don't waste cpu cycles
if(!quit_thread) usleep(500);
}
// free resources
...
// tell main thread we're done
quit_thread = FALSE;
}
Now I want to terminate subthread() from my main thread. I've tried the following:
quit_thread = TRUE;
// wait until subthread() has cleaned its resources
while(quit_thread);
But it does not work! The while() clause does never exit although my subthread clearly sets quit_thread to FALSE after having freed its resources!
If I modify my shutdown code like this:
quit_thread = TRUE;
// wait until subthread() has cleaned its resources
while(quit_thread) usleep(10);
Then everything is working fine! Could someone explain to me why the first solution does not work and why the version with usleep(10) suddenly works? I know that this is not a pretty solution. I could use semaphores/signals for this but I'd like to learn something about multithreading, so I'd like to know why my first solution doesn't work.
Thanks!
Without a memory fence, there is no guarantee that values written in one thread will appear in another. Most of the pthread primitives introduce a barrier, as do several system calls such as usleep. Using a mutex around both the read and write introduces a barrier, and more generally prevents multi-byte values being visible in partially written state.
You also need to separate the idea of asking a thread to stop executing, and reporting that it has stopped, and appear to be using the same variable for both.
What's most likely to be happening is that your compiler is not aware that quit_thread can be changed by another thread (because C doesn't know about threads, at least at the time this question was asked). Because of that, it's optimising the while loop to an infinite loop.
In other words, it looks at this code:
quit_thread = TRUE;
while(quit_thread);
and thinks to itself, "Hah, nothing in that loop can ever change quit_thread to FALSE, so the coder obviously just meant to write while (TRUE);".
When you add the call to usleep, the compiler has another think about it and assumes that the function call may change the global, so it plays it safe and doesn't optimise it.
Normally you would mark the variable as volatile to stop the compiler from optimising it but, in this case, you should use the facilities provided by pthreads and join to the thread after setting the flag to true (and don't have the sub-thread reset it, do that in the main thread after the join if it's necessary). The reason for that is that a join is likely to be more efficient than a continuous loop waiting for a variable change since the thread doing the join will most likely not be executed until the join needs to be done.
In your spinning solution, the joining thread will most likely continue to run and suck up CPU grunt.
In other words, do something like:
Main thread Child thread
------------------- -------------------
fStop = false
start Child Initialise
Do some other stuff while not fStop:
fStop = true Do what you have to do
Finish up and exit
join to Child
Do yet more stuff
And, as an aside, you should technically protect shared variables with mutexes but this is one of the few cases where it's okay, one-way communication where half-changed values of a variable don't matter (false/not-false).
The reason you normally mutex-protect a variable is to stop one thread seeing it in a half-changed state. Let's say you have a two-byte integer for a count of some objects, and it's set to 0x00ff (255).
Let's further say that thread A tries to increment that count but it's not an atomic operation. It changes the top byte to 0x01 but, before it gets a chance to change the bottom byte to 0x00, thread B swoops in and reads it as 0x01ff.
Now that's not going to be very good if thread B want to do something with the last element counted by that value. It should be looking at 0x0100 but will instead try to look at 0x01ff, the effect of which will be wrong, if not catastrophic.
If the count variable were protected by a mutex, thread B wouldn't be looking at it until thread A had finished updating it, hence no problem would occur.
The reason that doesn't matter with one-way booleans is because any half state will also be considered as true or false so, if thread A was halfway between turning 0x0000 into 0x0001 (just the top byte), thread B would still see that as 0x0000 (false) and keep going (until thread A finishes its update next time around).
And if thread A was turning the boolean into 0xffff, the half state of 0xff00 would still be considered true by thread B so it would do its thing before thread A had finished updating the boolean.
Neither of those two possibilities is bad simply because, in both, thread A is in the process of changing the boolean and it will finish eventually. Whether thread B detects it a tiny bit earlier or a tiny bit later doesn't really matter.
The while(quite_thread); is using the value quit_thread was set to on the line before it. Calling a function (usleep) induces the compiler to reload the value on each test.
In any case, this is the wrong way to wait for a thread to complete. Use pthread_join instead.
You're "learning" multhithreading the wrong way. The right way is to learn to use mutexes and condition variables; any other solution will fail under some circumstances.
I'm attempting to learn Clojure from the API and documentation available on the site. I'm a bit unclear about mutable storage in Clojure and I want to make sure my understanding is correct. Please let me know if there are any ideas that I've gotten wrong.
Edit: I'm updating this as I receive comments on its correctness.
Disclaimer: All of this information is informal and potentially wrong. Do not use this post for gaining an understanding of how Clojure works.
Vars always contain a root binding and possibly a per-thread binding. They are comparable to regular variables in imperative languages and are not suited for sharing information between threads. (thanks Arthur Ulfeldt)
Refs are locations shared between threads that support atomic transactions that can change the state of any number of refs in a single transaction. Transactions are committed upon exiting sync expressions (dosync) and conflicts are resolved automatically with STM magic (rollbacks, queues, waits, etc.)
Agents are locations that enable information to be asynchronously shared between threads with minimal overhead by dispatching independent action functions to change the agent's state. Agents are returned immediately and are therefore non-blocking, although an agent's value isn't set until a dispatched function has completed.
Atoms are locations that can be synchronously shared between threads. They support safe manipulation between different threads.
Here's my friendly summary based on when to use these structures:
Vars are like regular old variables in imperative languages. (avoid when possible)
Atoms are like Vars but with thread-sharing safety that allows for immediate reading and safe setting. (thanks Martin)
An Agent is like an Atom but rather than blocking it spawns a new thread to calculate its value, only blocks if in the middle of changing a value, and can let other threads know that it's finished assigning.
Refs are shared locations that lock themselves in transactions. Instead of making the programmer decide what happens during race conditions for every piece of locked code, we just start up a transaction and let Clojure handle all the lock conditions between the refs in that transaction.
Also, a related concept is the function future. To me, it seems like a future object can be described as a synchronous Agent where the value can't be accessed at all until the calculation is completed. It can also be described as a non-blocking Atom. Are these accurate conceptions of future?
It sounds like you are really getting Clojure! good job :)
Vars have a "root binding" visible in all threads and each individual thread can change the value it sees with out affecting the other threads. If my understanding is correct a var cannot exist in just one thread with out a root binding that is visible to all and it cant be "rebound" until it has been defined with (def ... ) the first time.
Refs are committed at the end of the (dosync ... ) transaction that encloses the changes but only when the transaction was able to finish in a consistent state.
I think your conclusion about Atoms is wrong:
Atoms are like Vars but with thread-sharing safety that blocks until the value has changed
Atoms are changed with swap! or low-level with compare-and-set!. This never blocks anything. swap! works like a transaction with just one ref:
the old value is taken from the atom and stored thread-local
the function is applied to the old value to generate a new value
if this succeeds compare-and-set is called with old and new value; only if the value of the atom has not been changed by any other thread (still equals old value), the new value is written, otherwise the operation restarts at (1) until is succeeds eventually.
I've found two issues with your question.
You say:
If an agent is accessed while an action is occurring then the value isn't returned until the action has finished
http://clojure.org/agents says:
the state of an Agent is always immediately available for reading by any thread
I.e. you never have to wait to get the value of an agent (I assume the value changed by an action is proxied and changed atomically).
The code for the deref-method of an Agent looks like this (SVN revision 1382):
public Object deref() throws Exception{
if(errors != null)
{
throw new Exception("Agent has errors", (Exception) RT.first(errors));
}
return state;
}
No blocking is involved.
Also, I don't understand what you mean (in your Ref section) by
Transactions are committed on calls to deref
Transactions are committed when all actions of the dosync block have been completed, no exceptions have been thrown and nothing has caused the transaction to be retried. I think deref has nothing to do with it, but maybe I misunderstand your point.
Martin is right when he say that Atoms operation restarts at 1. until is succeeds eventually.
It is also called spin waiting.
While it is note really blocking on a lock the thread that did the operation is blocked until the operation succeeds so it is a blocking operation and not an asynchronously operation.
Also about Futures, Clojure 1.1 has added abstractions for promises and futures.
A promise is a synchronization construct that can be used to deliver a value from one thread to another. Until the value has been delivered, any attempt to dereference the promise will block.
(def a-promise (promise))
(deliver a-promise :fred)
Futures represent asynchronous computations. They are a way to get code to run in another thread, and obtain the result.
(def f (future (some-sexp)))
(deref f) ; blocks the thread that derefs f until value is available
Vars don't always have a root binding. It's legal to create a var without a binding using
(def x)
or
(declare x)
Attempting to evaluate x before it has a value will result in
Var user/x is unbound.
[Thrown class java.lang.IllegalStateException]
I understand about race conditions and how with multiple threads accessing the same variable, updates made by one can be ignored and overwritten by others, but what if each thread is writing the same value (not different values) to the same variable; can even this cause problems? Could this code:
GlobalVar.property = 11;
(assuming that property will never be assigned anything other than 11), cause problems if multiple threads execute it at the same time?
The problem comes when you read that state back, and do something about it. Writing is a red herring - it is true that as long as this is a single word most environments guarantee the write will be atomic, but that doesn't mean that a larger piece of code that includes this fragment is thread-safe. Firstly, presumably your global variable contained a different value to begin with - otherwise if you know it's always the same, why is it a variable? Second, presumably you eventually read this value back again?
The issue is that presumably, you are writing to this bit of shared state for a reason - to signal that something has occurred? This is where it falls down: when you have no locking constructs, there is no implied order of memory accesses at all. It's hard to point to what's wrong here because your example doesn't actually contain the use of the variable, so here's a trivialish example in neutral C-like syntax:
int x = 0, y = 0;
//thread A does:
x = 1;
y = 2;
if (y == 2)
print(x);
//thread B does, at the same time:
if (y == 2)
print(x);
Thread A will always print 1, but it's completely valid for thread B to print 0. The order of operations in thread A is only required to be observable from code executing in thread A - thread B is allowed to see any combination of the state. The writes to x and y may not actually happen in order.
This can happen even on single-processor systems, where most people do not expect this kind of reordering - your compiler may reorder it for you. On SMP even if the compiler doesn't reorder things, the memory writes may be reordered between the caches of the separate processors.
If that doesn't seem to answer it for you, include more detail of your example in the question. Without the use of the variable it's impossible to definitively say whether such a usage is safe or not.
It depends on the work actually done by that statement. There can still be some cases where Something Bad happens - for example, if a C++ class has overloaded the = operator, and does anything nontrivial within that statement.
I have accidentally written code that did something like this with POD types (builtin primitive types), and it worked fine -- however, it's definitely not good practice, and I'm not confident that it's dependable.
Why not just lock the memory around this variable when you use it? In fact, if you somehow "know" this is the only write statement that can occur at some point in your code, why not just use the value 11 directly, instead of writing it to a shared variable?
(edit: I guess it's better to use a constant name instead of the magic number 11 directly in the code, btw.)
If you're using this to figure out when at least one thread has reached this statement, you could use a semaphore that starts at 1, and is decremented by the first thread that hits it.
I would expect the result to be undetermined. As in it would vary from compiler to complier, langauge to language and OS to OS etc. So no, it is not safe
WHy would you want to do this though - adding in a line to obtain a mutex lock is only one or two lines of code (in most languages), and would remove any possibility of problem. If this is going to be two expensive then you need to find an alternate way of solving the problem
In General, this is not considered a safe thing to do unless your system provides for atomic operation (operations that are guaranteed to be executed in a single cycle).
The reason is that while the "C" statement looks simple, often there are a number of underlying assembly operations taking place.
Depending on your OS, there are a few things you could do:
Take a mutual exclusion semaphore (mutex) to protect access
in some OS, you can temporarily disable preemption, which guarantees your thread will not swap out.
Some OS provide a writer or reader semaphore which is more performant than a plain old mutex.
Here's my take on the question.
You have two or more threads running that write to a variable...like a status flag or something, where you only want to know if one or more of them was true. Then in another part of the code (after the threads complete) you want to check and see if at least on thread set that status... for example
bool flag = false
threadContainer tc
threadInputs inputs
check(input)
{
...do stuff to input
if(success)
flag = true
}
start multiple threads
foreach(i in inputs)
t = startthread(check, i)
tc.add(t) // Keep track of all the threads started
foreach(t in tc)
t.join( ) // Wait until each thread is done
if(flag)
print "One of the threads were successful"
else
print "None of the threads were successful"
I believe the above code would be OK, assuming you're fine with not knowing which thread set the status to true, and you can wait for all the multi-threaded stuff to finish before reading that flag. I could be wrong though.
If the operation is atomic, you should be able to get by just fine. But I wouldn't do that in practice. It is better just to acquire a lock on the object and write the value.
Assuming that property will never be assigned anything other than 11, then I don't see a reason for assigment in the first place. Just make it a constant then.
Assigment only makes sense when you intend to change the value unless the act of assigment itself has other side effects - like volatile writes have memory visibility side-effects in Java. And if you change state shared between multiple threads, then you need to synchronize or otherwise "handle" the problem of concurrency.
When you assign a value, without proper synchronization, to some state shared between multiple threads, then there's no guarantees for when the other threads will see that change. And no visibility guarantees means that it it possible that the other threads will never see the assignt.
Compilers, JITs, CPU caches. They're all trying to make your code run as fast as possible, and if you don't make any explicit requirements for memory visibility, then they will take advantage of that. If not on your machine, then somebody elses.