7. Helgrind: a thread error detector (original) (raw)

7. Helgrind: a thread error detector

To use this tool, you must specify--tool=helgrind on the Valgrind command line.

7.1. Overview

Helgrind is a Valgrind tool for detecting synchronisation errors in C, C++ and Fortran programs that use the POSIX pthreads threading primitives.

The main abstractions in POSIX pthreads are: a set of threads sharing a common address space, thread creation, thread joining, thread exit, mutexes (locks), condition variables (inter-thread event notifications), reader-writer locks, spinlocks, semaphores and barriers.

Helgrind can detect three classes of errors, which are discussed in detail in the next three sections:

1. Misuses of the POSIX pthreads API.
2. Potential deadlocks arising from lock ordering problems.
3. Data races -- accessing memory without adequate locking or synchronisation.

Problems like these often result in unreproducible, timing-dependent crashes, deadlocks and other misbehaviour, and can be difficult to find by other means.

Helgrind is aware of all the pthread abstractions and tracks their effects as accurately as it can. On x86 and amd64 platforms, it understands and partially handles implicit locking arising from the use of the LOCK instruction prefix. On PowerPC/POWER and ARM platforms, it partially handles implicit locking arising from load-linked and store-conditional instruction pairs.

Helgrind works best when your application uses only the POSIX pthreads API. However, if you want to use custom threading primitives, you can describe their behaviour to Helgrind using theANNOTATE_* macros defined in helgrind.h.

Helgrind also provides Execution Trees memory profiling using the command line option --xtree-memory and the monitor commandxtmemory.

Following those is a section containing hints and tips on how to get the best out of Helgrind.

Then there is asummary of command-line options.

Finally, there is a brief summary of areas in which Helgrind could be improved.

7.2. Detected errors: Misuses of the POSIX pthreads API

Helgrind intercepts calls to many POSIX pthreads functions, and is therefore able to report on various common problems. Although these are unglamourous errors, their presence can lead to undefined program behaviour and hard-to-find bugs later on. The detected errors are:

• unlocking an invalid mutex
• unlocking a not-locked mutex
• unlocking a mutex held by a different thread
• destroying an invalid or a locked mutex
• recursively locking a non-recursive mutex
• deallocation of memory that contains a locked mutex
• passing mutex arguments to functions expecting reader-writer lock arguments, and vice versa
• when a POSIX pthread function fails with an error code that must be handled
• when a thread exits whilst still holding locked locks
• calling pthread_cond_wait with a not-locked mutex, an invalid mutex, or one locked by a different thread
• inconsistent bindings between condition variables and their associated mutexes
• invalid or duplicate initialisation of a pthread barrier
• initialisation of a pthread barrier on which threads are still waiting
• destruction of a pthread barrier object which was never initialised, or on which threads are still waiting
• waiting on an uninitialised pthread barrier
• for all of the pthreads functions that Helgrind intercepts, an error is reported, along with a stack trace, if the system threading library routine returns an error code, even if Helgrind itself detected no error

Checks pertaining to the validity of mutexes are generally also performed for reader-writer locks.

Signalling or broadcasting a condition variable when the associated mutex is unlocked is not strictly an error. The resulting thread scheduling may be unpredictable if the mutex is not held. The option --check-cond-signal-mutex=yes|noturns on checking for this situation. This kind of error is categorised as 'dubious'. The check is not turned on by default because some standard C and C++ libraries use condition signals/broadcasts with the associated mutex unlocked.

Various kinds of this-can't-possibly-happen events are also reported. These usually indicate bugs in the system threading library.

Reported errors always contain a primary stack trace indicating where the error was detected. They may also contain auxiliary stack traces giving additional information. In particular, most errors relating to mutexes will also tell you where that mutex first came to Helgrind's attention (the "was first observed at" part), so you have a chance of figuring out which mutex it is referring to. For example:

Thread #1 unlocked a not-locked lock at 0x7FEFFFA90 at 0x4C2408D: pthread_mutex_unlock (hg_intercepts.c:492) by 0x40073A: nearly_main (tc09_bad_unlock.c:27) by 0x40079B: main (tc09_bad_unlock.c:50) Lock at 0x7FEFFFA90 was first observed at 0x4C25D01: pthread_mutex_init (hg_intercepts.c:326) by 0x40071F: nearly_main (tc09_bad_unlock.c:23) by 0x40079B: main (tc09_bad_unlock.c:50)

Helgrind has a way of summarising thread identities, as you see here with the text "Thread #1". This is so that it can speak about threads and sets of threads without overwhelming you with details. See belowfor more information on interpreting error messages.

7.3. Detected errors: Inconsistent Lock Orderings

In this section, and in general, to "acquire" a lock simply means to lock that lock, and to "release" a lock means to unlock it.

Helgrind monitors the order in which threads acquire locks. This allows it to detect potential deadlocks which could arise from the formation of cycles of locks. Detecting such inconsistencies is useful because, whilst actual deadlocks are fairly obvious, potential deadlocks may never be discovered during testing and could later lead to hard-to-diagnose in-service failures.

The simplest example of such a problem is as follows.

• Imagine some shared resource R, which, for whatever reason, is guarded by two locks, L1 and L2, which must both be held when R is accessed.
• Suppose a thread acquires L1, then L2, and proceeds to access R. The implication of this is that all threads in the program must acquire the two locks in the order first L1 then L2. Not doing so risks deadlock.
• The deadlock could happen if two threads -- call them T1 and T2 -- both want to access R. Suppose T1 acquires L1 first, and T2 acquires L2 first. Then T1 tries to acquire L2, and T2 tries to acquire L1, but those locks are both already held. So T1 and T2 become deadlocked.

Helgrind builds a directed graph indicating the order in which locks have been acquired in the past. When a thread acquires a new lock, the graph is updated, and then checked to see if it now contains a cycle. The presence of a cycle indicates a potential deadlock involving the locks in the cycle.

In general, Helgrind will choose two locks involved in the cycle and show you how their acquisition ordering has become inconsistent. It does this by showing the program points that first defined the ordering, and the program points which later violated it. Here is a simple example involving just two locks:

Thread #1: lock order "0x7FF0006D0 before 0x7FF0006A0" violated

Observed (incorrect) order is: acquisition of lock at 0x7FF0006A0 at 0x4C2BC62: pthread_mutex_lock (hg_intercepts.c:494) by 0x400825: main (tc13_laog1.c:23)

followed by a later acquisition of lock at 0x7FF0006D0 at 0x4C2BC62: pthread_mutex_lock (hg_intercepts.c:494) by 0x400853: main (tc13_laog1.c:24)

Required order was established by acquisition of lock at 0x7FF0006D0 at 0x4C2BC62: pthread_mutex_lock (hg_intercepts.c:494) by 0x40076D: main (tc13_laog1.c:17)

followed by a later acquisition of lock at 0x7FF0006A0 at 0x4C2BC62: pthread_mutex_lock (hg_intercepts.c:494) by 0x40079B: main (tc13_laog1.c:18)

When there are more than two locks in the cycle, the error is equally serious. However, at present Helgrind does not show the locks involved, sometimes because that information is not available, but also so as to avoid flooding you with information. For example, a naive implementation of the famous Dining Philosophers problem involves a cycle of five locks (see helgrind/tests/tc14_laog_dinphils.c). In this case Helgrind has detected that all 5 philosophers could simultaneously pick up their left fork and then deadlock whilst waiting to pick up their right forks.

Thread #6: lock order "0x80499A0 before 0x8049A00" violated

Observed (incorrect) order is: acquisition of lock at 0x8049A00 at 0x40085BC: pthread_mutex_lock (hg_intercepts.c:495) by 0x80485B4: dine (tc14_laog_dinphils.c:18) by 0x400BDA4: mythread_wrapper (hg_intercepts.c:219) by 0x39B924: start_thread (pthread_create.c:297) by 0x2F107D: clone (clone.S:130)

followed by a later acquisition of lock at 0x80499A0 at 0x40085BC: pthread_mutex_lock (hg_intercepts.c:495) by 0x80485CD: dine (tc14_laog_dinphils.c:19) by 0x400BDA4: mythread_wrapper (hg_intercepts.c:219) by 0x39B924: start_thread (pthread_create.c:297) by 0x2F107D: clone (clone.S:130)

7.4. Detected errors: Data Races

A data race happens, or could happen, when two threads access a shared memory location without using suitable locks or other synchronisation to ensure single-threaded access. Such missing locking can cause obscure timing dependent bugs. Ensuring programs are race-free is one of the central difficulties of threaded programming.

Reliably detecting races is a difficult problem, and most of Helgrind's internals are devoted to dealing with it. We begin with a simple example.

7.4.1. A Simple Data Race

About the simplest possible example of a race is as follows. In this program, it is impossible to know what the value of var is at the end of the program. Is it 2 ? Or 1 ?

#include <pthread.h>

int var = 0;

void* child_fn ( void* arg ) { var++; /* Unprotected relative to parent / / this is line 6 */ return NULL; }

int main ( void ) { pthread_t child; pthread_create(&child, NULL, child_fn, NULL); var++; /* Unprotected relative to child / / this is line 13 */ pthread_join(child, NULL); return 0; }

The problem is there is nothing to stop var being updated simultaneously by both threads. A correct program would protect var with a lock of typepthread_mutex_t, which is acquired before each access and released afterwards. Helgrind's output for this program is:

Thread #1 is the program's root thread

Thread #2 was created at 0x511C08E: clone (in /lib64/libc-2.8.so) by 0x4E333A4: do_clone (in /lib64/libpthread-2.8.so) by 0x4E33A30: pthread_create@@GLIBC_2.2.5 (in /lib64/libpthread-2.8.so) by 0x4C299D4: pthread_create@* (hg_intercepts.c:214) by 0x400605: main (simple_race.c:12)

Possible data race during read of size 4 at 0x601038 by thread #1 Locks held: none at 0x400606: main (simple_race.c:13)

This conflicts with a previous write of size 4 by thread #2 Locks held: none at 0x4005DC: child_fn (simple_race.c:6) by 0x4C29AFF: mythread_wrapper (hg_intercepts.c:194) by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) by 0x511C0CC: clone (in /lib64/libc-2.8.so)

Location 0x601038 is 0 bytes inside global var "var" declared at simple_race.c:3

This is quite a lot of detail for an apparently simple error. The last clause is the main error message. It says there is a race as a result of a read of size 4 (bytes), at 0x601038, which is the address of var, happening in function main at line 13 in the program.

Two important parts of the message are:

• Helgrind shows two stack traces for the error, not one. By definition, a race involves two different threads accessing the same location in such a way that the result depends on the relative speeds of the two threads.
  The first stack trace follows the text "Possible data race during read of size 4 ..." and the second trace follows the text "This conflicts with a previous write of size 4 ...". Helgrind is usually able to show both accesses involved in a race. At least one of these will be a write (since two concurrent, unsynchronised reads are harmless), and they will of course be from different threads.
  By examining your program at the two locations, you should be able to get at least some idea of what the root cause of the problem is. For each location, Helgrind shows the set of locks held at the time of the access. This often makes it clear which thread, if any, failed to take a required lock. In this example neither thread holds a lock during the access.
• For races which occur on global or stack variables, Helgrind tries to identify the name and defining point of the variable. Hence the text "Location 0x601038 is 0 bytes inside global var "var" declared at simple_race.c:3".
  Showing names of stack and global variables carries no run-time overhead once Helgrind has your program up and running. However, it does require Helgrind to spend considerable extra time and memory at program startup to read the relevant debug info. Hence this facility is disabled by default. To enable it, you need to give the --read-var-info=yes option to Helgrind.

The following section explains Helgrind's race detection algorithm in more detail.

7.4.2. Helgrind's Race Detection Algorithm

Most programmers think about threaded programming in terms of the basic functionality provided by the threading library (POSIX Pthreads): thread creation, thread joining, locks, condition variables, semaphores and barriers.

The effect of using these functions is to impose constraints upon the order in which memory accesses can happen. This implied ordering is generally known as the "happens-before relation". Once you understand the happens-before relation, it is easy to see how Helgrind finds races in your code. Fortunately, the happens-before relation is itself easy to understand, and is by itself a useful tool for reasoning about the behaviour of parallel programs. We now introduce it using a simple example.

Consider first the following buggy program:

Parent thread: Child thread:

int var;

// create child thread pthread_create(...)
var = 20; var = 10; exit

// wait for child pthread_join(...) printf("%d\n", var);

The parent thread creates a child. Both then write different values to some variable var, and the parent then waits for the child to exit.

What is the value of var at the end of the program, 10 or 20? We don't know. The program is considered buggy (it has a race) because the final value of var depends on the relative rates of progress of the parent and child threads. If the parent is fast and the child is slow, then the child's assignment may happen later, so the final value will be 10; and vice versa if the child is faster than the parent.

The relative rates of progress of parent vs child is not something the programmer can control, and will often change from run to run. It depends on factors such as the load on the machine, what else is running, the kernel's scheduling strategy, and many other factors.

The obvious fix is to use a lock to protect var. It is however instructive to consider a somewhat more abstract solution, which is to send a message from one thread to the other:

Parent thread: Child thread:

int var;

// create child thread pthread_create(...)
var = 20; // send message to child // wait for message to arrive var = 10; exit

// wait for child pthread_join(...) printf("%d\n", var);

Now the program reliably prints "10", regardless of the speed of the threads. Why? Because the child's assignment cannot happen until after it receives the message. And the message is not sent until after the parent's assignment is done.

The message transmission creates a "happens-before" dependency between the two assignments: var = 20;must now happen-before var = 10;. And so there is no longer a race on var.

Note that it's not significant that the parent sends a message to the child. Sending a message from the child (after its assignment) to the parent (before its assignment) would also fix the problem, causing the program to reliably print "20".

Helgrind's algorithm is (conceptually) very simple. It monitors all accesses to memory locations. If a location -- in this example, var, is accessed by two different threads, Helgrind checks to see if the two accesses are ordered by the happens-before relation. If so, that's fine; if not, it reports a race.

It is important to understand that the happens-before relation creates only a partial ordering, not a total ordering. An example of a total ordering is comparison of numbers: for any two numbers x andy, either x is less than, equal to, or greater thany. A partial ordering is like a total ordering, but it can also express the concept that two elements are neither equal, less or greater, but merely unordered with respect to each other.

In the fixed example above, we say that var = 20; "happens-before"var = 10;. But in the original version, they are unordered: we cannot say that either happens-before the other.

What does it mean to say that two accesses from different threads are ordered by the happens-before relation? It means that there is some chain of inter-thread synchronisation operations which cause those accesses to happen in a particular order, irrespective of the actual rates of progress of the individual threads. This is a required property for a reliable threaded program, which is why Helgrind checks for it.

The happens-before relations created by standard threading primitives are as follows:

• When a mutex is unlocked by thread T1 and later (or immediately) locked by thread T2, then the memory accesses in T1 prior to the unlock must happen-before those in T2 after it acquires the lock.
• The same idea applies to reader-writer locks, although with some complication so as to allow correct handling of reads vs writes.
• When a condition variable (CV) is signalled on by thread T1 and some other thread T2 is thereby released from a wait on the same CV, then the memory accesses in T1 prior to the signalling must happen-before those in T2 after it returns from the wait. If no thread was waiting on the CV then there is no effect.
• If instead T1 broadcasts on a CV, then all of the waiting threads, rather than just one of them, acquire a happens-before dependency on the broadcasting thread at the point it did the broadcast.
• A thread T2 that continues after completing sem_wait on a semaphore that thread T1 posts on, acquires a happens-before dependence on the posting thread, a bit like dependencies caused mutex unlock-lock pairs. However, since a semaphore can be posted on many times, it is unspecified from which of the post calls the wait call gets its happens-before dependency.
• For a group of threads T1 .. Tn which arrive at a barrier and then move on, each thread after the call has a happens-after dependency from all threads before the barrier.
• A newly-created child thread acquires an initial happens-after dependency on the point where its parent created it. That is, all memory accesses performed by the parent prior to creating the child are regarded as happening-before all the accesses of the child.
• Similarly, when an exiting thread is reaped via a call to pthread_join, once the call returns, the reaping thread acquires a happens-after dependency relative to all memory accesses made by the exiting thread.

In summary: Helgrind intercepts the above listed events, and builds a directed acyclic graph represented the collective happens-before dependencies. It also monitors all memory accesses.

If a location is accessed by two different threads, but Helgrind cannot find any path through the happens-before graph from one access to the other, then it reports a race.

There are a couple of caveats:

• Helgrind doesn't check for a race in the case where both accesses are reads. That would be silly, since concurrent reads are harmless.
• Two accesses are considered to be ordered by the happens-before dependency even through arbitrarily long chains of synchronisation events. For example, if T1 accesses some location L, and then pthread_cond_signals T2, which laterpthread_cond_signals T3, which then accesses L, then a suitable happens-before dependency exists between the first and second accesses, even though it involves two different inter-thread synchronisation events.

7.4.3. Interpreting Race Error Messages

Helgrind's race detection algorithm collects a lot of information, and tries to present it in a helpful way when a race is detected. Here's an example:

Thread #2 was created at 0x511C08E: clone (in /lib64/libc-2.8.so) by 0x4E333A4: do_clone (in /lib64/libpthread-2.8.so) by 0x4E33A30: pthread_create@@GLIBC_2.2.5 (in /lib64/libpthread-2.8.so) by 0x4C299D4: pthread_create@* (hg_intercepts.c:214) by 0x4008F2: main (tc21_pthonce.c:86)

Thread #3 was created at 0x511C08E: clone (in /lib64/libc-2.8.so) by 0x4E333A4: do_clone (in /lib64/libpthread-2.8.so) by 0x4E33A30: pthread_create@@GLIBC_2.2.5 (in /lib64/libpthread-2.8.so) by 0x4C299D4: pthread_create@* (hg_intercepts.c:214) by 0x4008F2: main (tc21_pthonce.c:86)

Possible data race during read of size 4 at 0x601070 by thread #3 Locks held: none at 0x40087A: child (tc21_pthonce.c:74) by 0x4C29AFF: mythread_wrapper (hg_intercepts.c:194) by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) by 0x511C0CC: clone (in /lib64/libc-2.8.so)

This conflicts with a previous write of size 4 by thread #2 Locks held: none at 0x400883: child (tc21_pthonce.c:74) by 0x4C29AFF: mythread_wrapper (hg_intercepts.c:194) by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) by 0x511C0CC: clone (in /lib64/libc-2.8.so)

Location 0x601070 is 0 bytes inside local var "unprotected2" declared at tc21_pthonce.c:51, in frame #0 of thread 3

Helgrind first announces the creation points of any threads referenced in the error message. This is so it can speak concisely about threads without repeatedly printing their creation point call stacks. Each thread is only ever announced once, the first time it appears in any Helgrind error message.

The main error message begins at the text "Possible data race during read". At the start is information you would expect to see -- address and size of the racing access, whether a read or a write, and the call stack at the point it was detected.

A second call stack is presented starting at the text "This conflicts with a previous write". This shows a previous access which also accessed the stated address, and which is believed to be racing against the access in the first call stack. Note that this second call stack is limited to a maximum of --history-backtrace-sizeentries with a default value of 8 to limit the memory usage.

Finally, Helgrind may attempt to give a description of the raced-on address in source level terms. In this example, it identifies it as a local variable, shows its name, declaration point, and in which frame (of the first call stack) it lives. Note that this information is only shown when --read-var-info=yesis specified on the command line. That's because reading the DWARF3 debug information in enough detail to capture variable type and location information makes Helgrind much slower at startup, and also requires considerable amounts of memory, for large programs.

Once you have your two call stacks, how do you find the root cause of the race?

The first thing to do is examine the source locations referred to by each call stack. They should both show an access to the same location, or variable.

Now figure out how how that location should have been made thread-safe:

• Perhaps the location was intended to be protected by a mutex? If so, you need to lock and unlock the mutex at both access points, even if one of the accesses is reported to be a read. Did you perhaps forget the locking at one or other of the accesses? To help you do this, Helgrind shows the set of locks held by each threads at the time they accessed the raced-on location.
• Alternatively, perhaps you intended to use a some other scheme to make it safe, such as signalling on a condition variable. In all such cases, try to find a synchronisation event (or a chain thereof) which separates the earlier-observed access (as shown in the second call stack) from the later-observed access (as shown in the first call stack). In other words, try to find evidence that the earlier access "happens-before" the later access. See the previous subsection for an explanation of the happens-before relation.
  The fact that Helgrind is reporting a race means it did not observe any happens-before relation between the two accesses. If Helgrind is working correctly, it should also be the case that you also cannot find any such relation, even on detailed inspection of the source code. Hopefully, though, your inspection of the code will show where the missing synchronisation operation(s) should have been.

7.5. Hints and Tips for Effective Use of Helgrind

Helgrind can be very helpful in finding and resolving threading-related problems. Like all sophisticated tools, it is most effective when you understand how to play to its strengths.

Helgrind will be less effective when you merely throw an existing threaded program at it and try to make sense of any reported errors. It will be more effective if you design threaded programs from the start in a way that helps Helgrind verify correctness. The same is true for finding memory errors with Memcheck, but applies more here, because thread checking is a harder problem. Consequently it is much easier to write a correct program for which Helgrind falsely reports (threading) errors than it is to write a correct program for which Memcheck falsely reports (memory) errors.

With that in mind, here are some tips, listed most important first, for getting reliable results and avoiding false errors. The first two are critical. Any violations of them will swamp you with huge numbers of false data-race errors.

1. Make sure your application, and all the libraries it uses, use the POSIX threading primitives. Helgrind needs to be able to see all events pertaining to thread creation, exit, locking and other synchronisation events. To do so it intercepts many POSIX pthreads functions.
   Do not roll your own threading primitives (mutexes, etc) from combinations of the Linux futex syscall, atomic counters, etc. These throw Helgrind's internal what's-going-on models way off course and will give bogus results.
   Also, do not reimplement existing POSIX abstractions using other POSIX abstractions. For example, don't build your own semaphore routines or reader-writer locks from POSIX mutexes and condition variables. Instead use POSIX reader-writer locks and semaphores directly, since Helgrind supports them directly.
   Helgrind directly supports the following POSIX threading abstractions: mutexes, reader-writer locks, condition variables (but see below), semaphores and barriers. Currently spinlocks are not supported, although they could be in future.
   At the time of writing, the following popular Linux packages are known to implement their own threading primitives:
   • Qt version 4.X. Qt 3.X is harmless in that it only uses POSIX pthreads primitives. Unfortunately Qt 4.X has its own implementation of mutexes (QMutex) and thread reaping. Helgrind 3.4.x contains direct support for Qt 4.X threading, which is experimental but is believed to work fairly well. A side effect of supporting Qt 4 directly is that Helgrind can be used to debug KDE4 applications. As this is an experimental feature, we would particularly appreciate feedback from folks who have used Helgrind to successfully debug Qt 4 and/or KDE4 applications.
   • Runtime support library for GNU OpenMP (part of GCC), at least for GCC versions 4.2 and 4.3. The GNU OpenMP runtime library (libgomp.so) constructs its own synchronisation primitives using combinations of atomic memory instructions and the futex syscall, which causes total chaos since in Helgrind since it cannot "see" those.
     Fortunately, this can be solved using a configuration-time option (for GCC). Rebuild GCC from source, and configure using--disable-linux-futex. This makes libgomp.so use the standard POSIX threading primitives instead. Note that this was tested using GCC 4.2.3 and has not been re-tested using more recent GCC versions. We would appreciate hearing about any successes or failures with more recent versions.
     If you must implement your own threading primitives, there are a set of client request macros in helgrind.h to help you describe your primitives to Helgrind. You should be able to mark up mutexes, condition variables, etc, without difficulty.
     It is also possible to mark up the effects of thread-safe reference counting using theANNOTATE_HAPPENS_BEFORE,ANNOTATE_HAPPENS_AFTER andANNOTATE_HAPPENS_BEFORE_FORGET_ALL, macros. Thread-safe reference counting using an atomically incremented/decremented refcount variable causes Helgrind problems because a one-to-zero transition of the reference count means the accessing thread has exclusive ownership of the associated resource (normally, a C++ object) and can therefore access it (normally, to run its destructor) without locking. Helgrind doesn't understand this, and markup is essential to avoid false positives.
     Here are recommended guidelines for marking up thread safe reference counting in C++. You only need to mark up your release methods -- the ones which decrement the reference count. Given a class like this:
     class MyClass {
     unsigned int mRefCount;
     void Release ( void ) {
     unsigned int newCount = atomic_decrement(&mRefCount);
     if (newCount == 0) {
     delete this;
     }
     }
     }
     the release method should be marked up as follows:
     void Release ( void ) {
     unsigned int newCount = atomic_decrement(&mRefCount);
     if (newCount == 0) {
     ANNOTATE_HAPPENS_AFTER(&mRefCount);
     ANNOTATE_HAPPENS_BEFORE_FORGET_ALL(&mRefCount);
     delete this;
     } else {
     ANNOTATE_HAPPENS_BEFORE(&mRefCount);
     }
     }
     There are a number of complex, mostly-theoretical objections to this scheme. From a theoretical standpoint it appears to be impossible to devise a markup scheme which is completely correct in the sense of guaranteeing to remove all false races. The proposed scheme however works well in practice.
2. Avoid memory recycling. If you can't avoid it, you must use tell Helgrind what is going on via theVALGRIND_HG_CLEAN_MEMORY client request (inhelgrind.h).
   Helgrind is aware of standard heap memory allocation and deallocation that occurs viamalloc/free/new/delete and from entry and exit of stack frames. In particular, when memory is deallocated via free, delete, or function exit, Helgrind considers that memory clean, so when it is eventually reallocated, its history is irrelevant.
   However, it is common practice to implement memory recycling schemes. In these, memory to be freed is not handed tofree/delete, but instead put into a pool of free buffers to be handed out again as required. The problem is that Helgrind has no way to know that such memory is logically no longer in use, and its history is irrelevant. Hence you must make that explicit, using the VALGRIND_HG_CLEAN_MEMORY client request to specify the relevant address ranges. It's easiest to put these requests into the pool manager code, and use them either when memory is returned to the pool, or is allocated from it.
3. Avoid POSIX condition variables. If you can, use POSIX semaphores (sem_t, sem_post,sem_wait) to do inter-thread event signalling. Semaphores with an initial value of zero are particularly useful for this.
   Helgrind only partially correctly handles POSIX condition variables. This is because Helgrind can see inter-thread dependencies between a pthread_cond_wait call and apthread_cond_signal/pthread_cond_broadcast call only if the waiting thread actually gets to the rendezvous first (so that it actually callspthread_cond_wait). It can't see dependencies between the threads if the signaller arrives first. In the latter case, POSIX guidelines imply that the associated boolean condition still provides an inter-thread synchronisation event, but one which is invisible to Helgrind.
   The result of Helgrind missing some inter-thread synchronisation events is to cause it to report false positives.
   The root cause of this synchronisation lossage is particularly hard to understand, so an example is helpful. It was discussed at length by Arndt Muehlenfeld ("Runtime Race Detection in Multi-Threaded Programs", Dissertation, TU Graz, Austria). The canonical POSIX-recommended usage scheme for condition variables is as follows:
   b is a Boolean condition, which is False most of the time
   cv is a condition variable
   mx is its associated mutex
   Signaller: Waiter:
   lock(mx) lock(mx)
   b = True while (b == False)
   signal(cv) wait(cv,mx)
   unlock(mx) unlock(mx)
   Assume b is False most of the time. If the waiter arrives at the rendezvous first, it enters its while-loop, waits for the signaller to signal, and eventually proceeds. Helgrind sees the signal, notes the dependency, and all is well.
   If the signaller arrives first, b is set to true, and the signal disappears into nowhere. When the waiter later arrives, it does not enter its while-loop and simply carries on. But even in this case, the waiter code following the while-loop cannot execute until the signaller sets b to True. Hence there is still the same inter-thread dependency, but this time it is through an arbitrary in-memory condition, and Helgrind cannot see it.
   By comparison, Helgrind's detection of inter-thread dependencies caused by semaphore operations is believed to be exactly correct.
   As far as I know, a solution to this problem that does not require source-level annotation of condition-variable wait loops is beyond the current state of the art.
4. Make sure you are using a supported Linux distribution. At present, Helgrind only properly supports glibc-2.3 or later. This in turn means we only support glibc's NPTL threading implementation. The old LinuxThreads implementation is not supported.
5. If your application is using thread local variables, helgrind might report false positive race conditions on these variables, despite being very probably race free. On Linux, you can use --sim-hints=deactivate-pthread-stack-cache-via-hack to avoid such false positive error messages (see --sim-hints).
6. Round up all finished threads usingpthread_join. Avoid detaching threads: don't create threads in the detached state, and don't call pthread_detach on existing threads.
   Using pthread_join to round up finished threads provides a clear synchronisation point that both Helgrind and programmers can see. If you don't callpthread_join on a thread, Helgrind has no way to know when it finishes, relative to any significant synchronisation points for other threads in the program. So it assumes that the thread lingers indefinitely and can potentially interfere indefinitely with the memory state of the program. It has every right to assume that -- after all, it might really be the case that, for scheduling reasons, the exiting thread did run very slowly in the last stages of its life.
7. Perform thread debugging (with Helgrind) and memory debugging (with Memcheck) together.
   Helgrind tracks the state of memory in detail, and memory management bugs in the application are liable to cause confusion. In extreme cases, applications which do many invalid reads and writes (particularly to freed memory) have been known to crash Helgrind. So, ideally, you should make your application Memcheck-clean before using Helgrind.
   It may be impossible to make your application Memcheck-clean unless you first remove threading bugs. In particular, it may be difficult to remove all reads and writes to freed memory in multithreaded C++ destructor sequences at program termination. So, ideally, you should make your application Helgrind-clean before using Memcheck.
   Since this circularity is obviously unresolvable, at least bear in mind that Memcheck and Helgrind are to some extent complementary, and you may need to use them together.
8. POSIX requires that implementations of standard I/O (printf, fprintf,fwrite, fread, etc) are thread safe. Unfortunately GNU libc implements this by using internal locking primitives that Helgrind is unable to intercept. Consequently Helgrind generates many false race reports when you use these functions.
   Helgrind attempts to hide these errors using the standard Valgrind error-suppression mechanism. So, at least for simple test cases, you don't see any. Nevertheless, some may slip through. Just something to be aware of.
9. Helgrind's error checks do not work properly inside the system threading library itself (libpthread.so), and it usually observes large numbers of (false) errors in there. Valgrind's suppression system then filters these out, so you should not see them.
   If you see any race errors reported where libpthread.so orld.so is the object associated with the innermost stack frame, please file a bug report athttp://www.valgrind.org/.

7.6. Helgrind Command-line Options

The following end-user options are available:

--free-is-write=no|yes [default: no]

When enabled (not the default), Helgrind treats freeing of heap memory as if the memory was written immediately before the free. This exposes races where memory is referenced by one thread, and freed by another, but there is no observable synchronisation event to ensure that the reference happens before the free.

This functionality is new in Valgrind 3.7.0, and is regarded as experimental. It is not enabled by default because its interaction with custom memory allocators is not well understood at present. User feedback is welcomed.

--track-lockorders=no|yes [default: yes]

When enabled (the default), Helgrind performs lock order consistency checking. For some buggy programs, the large number of lock order errors reported can become annoying, particularly if you're only interested in race errors. You may therefore find it helpful to disable lock order checking.

--history-level=none|approx|full [default: full]

--history-level=full (the default) causes Helgrind collects enough information about "old" accesses that it can produce two stack traces in a race report -- both the stack trace for the current access, and the trace for the older, conflicting access. To limit memory usage, "old" accesses stack traces are limited to a maximum of --history-backtrace-size entries (default 8) or to --num-callers value if this value is smaller.

Collecting such information is expensive in both speed and memory, particularly for programs that do many inter-thread synchronisation events (locks, unlocks, etc). Without such information, it is more difficult to track down the root causes of races. Nonetheless, you may not need it in situations where you just want to check for the presence or absence of races, for example, when doing regression testing of a previously race-free program.

--history-level=none is the opposite extreme. It causes Helgrind not to collect any information about previous accesses. This can be dramatically faster than --history-level=full.

--history-level=approx provides a compromise between these two extremes. It causes Helgrind to show a full trace for the later access, and approximate information regarding the earlier access. This approximate information consists of two stacks, and the earlier access is guaranteed to have occurred somewhere between program points denoted by the two stacks. This is not as useful as showing the exact stack for the previous access (as --history-level=full does), but it is better than nothing, and it is almost as fast as--history-level=none.

--history-backtrace-size=<number> [default: 8]

When --history-level=full is selected,--history-backtrace-size=number indicates how many entries to record in "old" accesses stack traces.

--delta-stacktrace=no|yes [default: yes on linux amd64/x86]

This flag only has any effect at --history-level=full.

--delta-stacktrace configures the way Helgrind captures the stacktraces for the option --history-level=full. Such a stacktrace is typically needed each time a new piece of memory is read or written in a basic block of instructions.

--delta-stacktrace=no causes Helgrind to compute a full history stacktrace from the unwind info each time a stacktrace is needed.

--delta-stacktrace=yes indicates to Helgrind to derive a new stacktrace from the previous stacktrace, as long as there was no call instruction, no return instruction, or any other instruction changing the call stack since the previous stacktrace was captured. If no such instruction was executed, the new stacktrace can be derived from the previous stacktrace by just changing the top frame to the current program counter. This option can speed up Helgrind by 25% when using --history-level=full.

The following aspects have to be considered when using --delta-stacktrace=yes :

• In some cases (for example in a function prologue), the valgrind unwinder might not properly unwind the stack, due to some limitations and/or due to wrong unwind info. When using --delta-stacktrace=yes, the wrong stack trace captured in the function prologue will be kept till the next call or return.
• On the other hand, --delta-stacktrace=yes sometimes helps to obtain a correct stacktrace, for example when the unwind info allows a correct stacktrace to be done in the beginning of the sequence, but not later on in the instruction sequence.
• Determining which instructions are changing the callstack is partially based on platform dependent heuristics, which have to be tuned/validated specifically for the platform. Also, unwinding in a function prologue must be good enough to allow using --delta-stacktrace=yes. Currently, the option --delta-stacktrace=yes has been reasonably validated only on linux x86 32 bits and linux amd64 64 bits. For more details about how to validate --delta-stacktrace=yes, see debug option --hg-sanity-flags and the function check_cached_rcec_ok in libhb_core.c.

--conflict-cache-size=N [default: 1000000]

This flag only has any effect at --history-level=full.

Information about "old" conflicting accesses is stored in a cache of limited size, with LRU-style management. This is necessary because it isn't practical to store a stack trace for every single memory access made by the program. Historical information on not recently accessed locations is periodically discarded, to free up space in the cache.

This option controls the size of the cache, in terms of the number of different memory addresses for which conflicting access information is stored. If you find that Helgrind is showing race errors with only one stack instead of the expected two stacks, try increasing this value.

The minimum value is 10,000 and the maximum is 30,000,000 (thirty times the default value). Increasing the value by 1 increases Helgrind's memory requirement by very roughly 100 bytes, so the maximum value will easily eat up three extra gigabytes or so of memory.

--check-stack-refs=no|yes [default: yes]

By default Helgrind checks all data memory accesses made by your program. This flag enables you to skip checking for accesses to thread stacks (local variables). This can improve performance, but comes at the cost of missing races on stack-allocated data.

--ignore-thread-creation=<yes|no> [default: no]

Controls whether all activities during thread creation should be ignored. By default enabled only on Solaris. Solaris provides higher throughput, parallelism and scalability than other operating systems, at the cost of more fine-grained locking activity. This means for example that when a thread is created under glibc, just one big lock is used for all thread setup. Solaris libc uses several fine-grained locks and the creator thread resumes its activities as soon as possible, leaving for example stack and TLS setup sequence to the created thread. This situation confuses Helgrind as it assumes there is some false ordering in place between creator and created thread; and therefore many types of race conditions in the application would not be reported. To prevent such false ordering, this command line option is set toyes by default on Solaris. All activity (loads, stores, client requests) is therefore ignored during:

• pthread_create() call in the creator thread
• thread creation phase (stack and TLS setup) in the created thread

Also new memory allocated during thread creation is untracked, that is race reporting is suppressed there. DRD does the same thing implicitly. This is necessary because Solaris libc caches many objects and reuses them for different threads and that confuses Helgrind.

7.7. Helgrind Monitor Commands

The Helgrind tool provides monitor commands handled by Valgrind's built-in gdbserver (see Monitor command handling by the Valgrind gdbserver). Valgrind python code provides GDB front end commands giving an easier usage of the helgrind monitor commands (seeGDB front end commands for Valgrind gdbserver monitor commands). To launch an helgrind monitor command via its GDB front end command, instead of prefixing the command with "monitor", you must use the GDB helgrind command (or the shorter aliases hg). Using the helgrind GDB front end command provide a more flexible usage, such as evaluation of address and length arguments by GDB. In GDB, you can use help helgrind to get help about the helgrind front end monitor commands and you can use apropos helgrind to get all the commands mentionning the word "helgrind" in their name or on-line help.

• info locks [lock_addr] shows the list of locks and their status. If lock_addr is given, only shows the lock located at this address.
  In the following example, helgrind knows about one lock. This lock is located at the guest address ga 0x8049a20. The lock kind is rdwr indicating a reader-writer lock. Other possible lock kinds are nonRec (simple mutex, non recursive) and mbRec (simple mutex, possibly recursive). The lock kind is then followed by the list of threads helding the lock. In the below example, R1:thread #6 tid 3 indicates that the helgrind thread #6 has acquired (once, as the counter following the letter R is one) the lock in read mode. The helgrind thread nr is incremented for each started thread. The presence of 'tid 3' indicates that the thread #6 is has not exited yet and is the valgrind tid 3. If a thread has terminated, then this is indicated with 'tid (exited)'.
  (gdb) monitor info locks
  Lock ga 0x8049a20 {
  kind rdwr
  { R1:thread #6 tid 3 }
  }
  (gdb)
  If you give the option --read-var-info=yes, then more information will be provided about the lock location, such as the global variable or the heap block that contains the lock:
  Lock ga 0x8049a20 {
  Location 0x8049a20 is 0 bytes inside global var "s_rwlock"
  declared at rwlock_race.c:17
  kind rdwr
  { R1:thread #3 tid 3 }
  }
  The GDB equivalent helgrind front end command helgrind info locks [ADDR] accept any address expression for its first ADDR argument.
• accesshistory <addr> [<len>] shows the access history recorded for (default 1) bytes starting at . For each recorded access that overlaps with the given range, accesshistory shows the operation type (read or write), the address and size read or written, the helgrind thread nr/valgrind tid number that did the operation and the locks held by the thread at the time of the operation. The oldest access is shown first, the most recent access is shown last.
  In the following example, we see first a recorded write of 4 bytes by thread #7 that has modified the given 2 bytes range. The second recorded write is the most recent recorded write : thread #9 modified the same 2 bytes as part of a 4 bytes write operation. The list of locks held by each thread at the time of the write operation are also shown.
  (gdb) monitor accesshistory 0x8049D8A 2
  write of size 4 at 0x8049D88 by thread #7 tid 3
  ==6319== Locks held: 2, at address 0x8049D8C (and 1 that can't be shown)
  ==6319== at 0x804865F: child_fn1 (locked_vs_unlocked2.c:29)
  ==6319== by 0x400AE61: mythread_wrapper (hg_intercepts.c:234)
  ==6319== by 0x39B924: start_thread (pthread_create.c:297)
  ==6319== by 0x2F107D: clone (clone.S:130)
  write of size 4 at 0x8049D88 by thread #9 tid 2
  ==6319== Locks held: 2, at addresses 0x8049DA4 0x8049DD4
  ==6319== at 0x804877B: child_fn2 (locked_vs_unlocked2.c:45)
  ==6319== by 0x400AE61: mythread_wrapper (hg_intercepts.c:234)
  ==6319== by 0x39B924: start_thread (pthread_create.c:297)
  ==6319== by 0x2F107D: clone (clone.S:130)
  The GDB equivalent helgrind front end command helgrind accesshistory ADDR [LEN] accept any address expression for its first ADDR argument. The second optional argument is any integer expression. Note that these 2 arguments must be separated by a space, like in the following example:
  (gdb) hg accesshistory &mx sizeof(mx)
  read of size 4 at 0x1130A8 by thread #2 tid (exited)
  ==302== Locks held: none
  ==302== at 0x1094AC: child8 (tc19_shadowmem.c:37)
  ==302== by 0x10A0DF: steer (tc19_shadowmem.c:288)
  ==302== by 0x48448A3: mythread_wrapper (hg_intercepts.c:406)
  ==302== by 0x4879EA6: start_thread (pthread_create.c:477)
  ==302== by 0x4990A2E: clone (clone.S:95)
• xtmemory [<filename> default xtmemory.kcg.%p.%n] requests Helgrind tool to produce an xtree heap memory report. See Execution Trees for a detailed explanation about execution trees.

7.8. Helgrind Client Requests

The following client requests are defined inhelgrind.h. See that file for exact details of their arguments.

• VALGRIND_HG_CLEAN_MEMORY
  This makes Helgrind forget everything it knows about a specified memory range. This is particularly useful for memory allocators that wish to recycle memory.
• ANNOTATE_HAPPENS_BEFORE
• ANNOTATE_HAPPENS_AFTER
• ANNOTATE_NEW_MEMORY
• ANNOTATE_RWLOCK_CREATE
• ANNOTATE_RWLOCK_DESTROY
• ANNOTATE_RWLOCK_ACQUIRED
• ANNOTATE_RWLOCK_RELEASED
  These are used to describe to Helgrind, the behaviour of custom (non-POSIX) synchronisation primitives, which it otherwise has no way to understand. See comments in helgrind.h for further documentation.

7.9. A To-Do List for Helgrind

The following is a list of loose ends which should be tidied up some time.

• For lock order errors, print the complete lock cycle, rather than only doing for size-2 cycles as at present.
• The conflicting access mechanism sometimes mysteriously fails to show the conflicting access' stack, even when provided with unbounded storage for conflicting access info. This should be investigated.
• Document races caused by GCC's thread-unsafe code generation for speculative stores. In the interim seehttp://gcc.gnu.org/ml/gcc/2007-10/msg00266.html and http://lkml.org/lkml/2007/10/24/673.
• Don't update the lock-order graph, and don't check for errors, when a "try"-style lock operation happens (e.g.pthread_mutex_trylock). Such calls do not add any real restrictions to the locking order, since they can always fail to acquire the lock, resulting in the caller going off and doing Plan B (presumably it will have a Plan B). Doing such checks could generate false lock-order errors and confuse users.
• Performance can be very poor. Slowdowns on the order of 100:1 are not unusual. There is limited scope for performance improvements.