4 ms·

From the first code example: > Yep, notifying outside of lock is usually BETTER. Otherwise the other thread would be released but will immediately run into our

by exDM69 10y ago

From the first code example: > Yep, notifying outside of lock is usually BETTER. Otherwise the other thread would be released but will immediately run into our own lock above our own lock above, causing unnecessary (and Very Expensive) context switch Is this true and if so, on which platform(s)? Aren't most mutex/condition variable implementations optimized to avoid this case by deferring the condition signal to the mutex unlock? Additionally, in the implementation of kill() in the latter examples, there should be notify_all() instead of notify_one().

pierrebai 10y ago
That would be relying on the OS to do the right thing. One problem is that on some platform or for some usage pattern, then mutex and notifying object will be two independent objects, so the OS would have no way of knowing that it should not immediately wake up the notified threads. Furthermore, it is a good rule of thumb anyway: hold the mutex for the shortest time possible. The corresponding pattern is to hold a mutex, copy the data you need to a local copy, release the mutex and then do the processing on the local copy. Sometimes, the processing can have unknown dependency and so it avoids potential deadlocks caused by some function called during processing trying to take a lock that is held be another thread. Basically, never hold a lock while calling functions. You never know what might be hidden behind them now and will be hidden in the future.
- exDM69 10y ago
  > One problem is that on some platform or for some usage pattern, then mutex and notifying object will be two independent objects, so the OS would have no way of knowing that it should not immediately wake up the notified threads. On all significant platforms I know, condition wait works like this: condition_wait(locked_mutex, condition_to_wait); For each thread in the wait list, the condition variable "knows" which mutex it is holding. When signalling a waiting thread, it's added to the wait list of the mutex (if mutex is locked), not actually woken up. Did you have a platform in mind that works differently from this? Additionally, when you unlock a mutex before signalling the condition, another thread may be scheduled in between that may invalidate the condition that was signalled. This may introduce a race condition. I signal my condition variables with mutexes locked because it's easier to guarantee correctness that way.
  - no-bugs 10y ago
    > Additionally, when you unlock a mutex before signalling the condition, another thread may be scheduled in between that may invalidate the condition that was signalled. This may introduce a race condition. And if you unlock the mutex after signalling the condition, another thread can be still scheduled in between the call to notify() and whatever-notified-thread-which-starts-to-run (and can still invalidate whatever-conditions-it-can-invalidate). There is absolutely no guarantee whatsoever that notified-thread gets to your mutex first (not that it really should matter for correctness in any sane program BTW). > I signal my condition variables with mutexes locked because it's easier to guarantee correctness that way. There is no difference in guarantees in practice whether you notify within or after; I'm not even sure that there is any difference in theory (exactly because the thread-which-was-notified is not guaranteed to get mutex first).
  - gpderetta 10y ago
    The optimization you are talking about is called wait morphing. Linux futexes have support for this optimization, and glibc pthread_cond_signal did support it at one point. I'm not sure wether they still do though. Unless you are writing hard realtime systems where you can make use of the strict scheduling guarantees of SCHED_FIFO, signaling while holding the lock is never necessary nor sufficient for correctness. The waiter must be checking the condition in a loop anyway.
no-bugs 10y ago
> Aren't most mutex/condition variable implementations optimized to avoid this case by deferring the condition signal to the mutex unlock? Yes, _some_ but not _all_ implementations are doing it; however, as it is not really guaranteed (and actually is a workaround for poorly written programs) - standing recommendation is still to notify after the lock (which can be better, can be the same, but won't be worse than doing it within), see for example http://en.cppreference.com/w/cpp/thread/condition_variable/notify_one http://en.cppreference.com/w/cpp/thread/condition_variable/n...: "The notifying thread does not need to hold the lock on the same mutex as the one held by the waiting thread(s); in fact doing so is a pessimization, since the notified thread would immediately block again, waiting for the notifying thread to release the lock. However, some implementations (in particular many implementations of pthreads) recognize this situation and avoid this "hurry up and wait" scenario by transferring the waiting thread from the condition variable's queue directly to the queue of the mutex within the notify call, without waking it up." > Additionally, in the implementation of kill() in the latter examples, there should be notify_all() instead of notify_one(). In MOST cases, it didn't matter (as there was only one blocking thread - the one reading), but yes, there was one case when it was indeed important. Fixed now, THANKS!
- scott_s 10y ago
  Pthreads on Linux encourage holding the lock during a signal or broadcast (http://linux.die.net/man/3/pthread_cond_signal http://linux.die.net/man/3/pthread_cond_signal): The pthread_cond_broadcast() or pthread_cond_signal() functions may be called by a thread whether or not it currently owns the mutex that threads calling pthread_cond_wait() or pthread_cond_timedwait() have associated with the condition variable during their waits; however, if predictable scheduling behavior is required, then that mutex shall be locked by the thread calling pthread_cond_broadcast() or pthread_cond_signal(). They don't define what exactly "predictable scheduling is", but the message that most programmers will probably come away with is, grabbing the lock is probably the better thing to do.
- exDM69 10y ago
  Since most platforms defer condition signalling to mutex unlocks, the way the code is now written will cause more spurious wakeups and context switching than necessary. The pathological case in scheduling goes like this: 1. Reader A enters pop_front on an empty queue, goes to wait on the condition variable 2. Writer W enters push_back, adds an element to list and releases mutex and get pre-empted (on line 25 first example) 3. Reader B enters pop_front, sees queue not empty, pops element and leaves 4. Writer W signals condition, wakes up Reader A 5. Reader A wakes up but the queue is empty, goes back to sleep causing unnecessary context switch and trashing the TLB The code is "correct" either way, but it's now optimized for the rare platforms that have a badly implemented condition variable. I would say it's usually better to signal your condition variables with mutexes locked unless you know you're running on a platform with flaky condition variables. Easier to guarantee correctness and avoid spurious wakeups that way. related note: cppreference.com is a terrible resource. It was worse 10 years ago, but it's not improved much. I would not trust the weasel wording ("some implementations" etc) in the link you posted, what it says about pthreads contradicts the pthread man pages (that encourage signal while holding lock)

by exDM69 10y ago

pierrebai 10y ago

That would be relying on the OS to do the right thing. One problem is that on some platform or for some usage pattern, then mutex and notifying object will be two independent objects, so the OS would have no way of knowing that it should not immediately wake up the notified threads. Furthermore, it is a good rule of thumb anyway: hold the mutex for the shortest time possible. The corresponding pattern is to hold a mutex, copy the data you need to a local copy, release the mutex and then do the processing on the local copy. Sometimes, the processing can have unknown dependency and so it avoids potential deadlocks caused by some function called during processing trying to take a lock that is held be another thread. Basically, never hold a lock while calling functions. You never know what might be hidden behind them now and will be hidden in the future.

exDM69 10y ago
> One problem is that on some platform or for some usage pattern, then mutex and notifying object will be two independent objects, so the OS would have no way of knowing that it should not immediately wake up the notified threads. On all significant platforms I know, condition wait works like this: condition_wait(locked_mutex, condition_to_wait); For each thread in the wait list, the condition variable "knows" which mutex it is holding. When signalling a waiting thread, it's added to the wait list of the mutex (if mutex is locked), not actually woken up. Did you have a platform in mind that works differently from this? Additionally, when you unlock a mutex before signalling the condition, another thread may be scheduled in between that may invalidate the condition that was signalled. This may introduce a race condition. I signal my condition variables with mutexes locked because it's easier to guarantee correctness that way.
- no-bugs 10y ago
  > Additionally, when you unlock a mutex before signalling the condition, another thread may be scheduled in between that may invalidate the condition that was signalled. This may introduce a race condition. And if you unlock the mutex after signalling the condition, another thread can be still scheduled in between the call to notify() and whatever-notified-thread-which-starts-to-run (and can still invalidate whatever-conditions-it-can-invalidate). There is absolutely no guarantee whatsoever that notified-thread gets to your mutex first (not that it really should matter for correctness in any sane program BTW). > I signal my condition variables with mutexes locked because it's easier to guarantee correctness that way. There is no difference in guarantees in practice whether you notify within or after; I'm not even sure that there is any difference in theory (exactly because the thread-which-was-notified is not guaranteed to get mutex first).
- gpderetta 10y ago
  The optimization you are talking about is called wait morphing. Linux futexes have support for this optimization, and glibc pthread_cond_signal did support it at one point. I'm not sure wether they still do though. Unless you are writing hard realtime systems where you can make use of the strict scheduling guarantees of SCHED_FIFO, signaling while holding the lock is never necessary nor sufficient for correctness. The waiter must be checking the condition in a loop anyway.

no-bugs 10y ago

> Aren't most mutex/condition variable implementations optimized to avoid this case by deferring the condition signal to the mutex unlock? Yes, _some_ but not _all_ implementations are doing it; however, as it is not really guaranteed (and actually is a workaround for poorly written programs) - standing recommendation is still to notify after the lock (which can be better, can be the same, but won't be worse than doing it within), see for example http://en.cppreference.com/w/cpp/thread/condition_variable/notify_one http://en.cppreference.com/w/cpp/thread/condition_variable/n...: "The notifying thread does not need to hold the lock on the same mutex as the one held by the waiting thread(s); in fact doing so is a pessimization, since the notified thread would immediately block again, waiting for the notifying thread to release the lock. However, some implementations (in particular many implementations of pthreads) recognize this situation and avoid this "hurry up and wait" scenario by transferring the waiting thread from the condition variable's queue directly to the queue of the mutex within the notify call, without waking it up." > Additionally, in the implementation of kill() in the latter examples, there should be notify_all() instead of notify_one(). In MOST cases, it didn't matter (as there was only one blocking thread - the one reading), but yes, there was one case when it was indeed important. Fixed now, THANKS!

scott_s 10y ago
Pthreads on Linux encourage holding the lock during a signal or broadcast (http://linux.die.net/man/3/pthread_cond_signal http://linux.die.net/man/3/pthread_cond_signal): The pthread_cond_broadcast() or pthread_cond_signal() functions may be called by a thread whether or not it currently owns the mutex that threads calling pthread_cond_wait() or pthread_cond_timedwait() have associated with the condition variable during their waits; however, if predictable scheduling behavior is required, then that mutex shall be locked by the thread calling pthread_cond_broadcast() or pthread_cond_signal(). They don't define what exactly "predictable scheduling is", but the message that most programmers will probably come away with is, grabbing the lock is probably the better thing to do.
exDM69 10y ago
Since most platforms defer condition signalling to mutex unlocks, the way the code is now written will cause more spurious wakeups and context switching than necessary. The pathological case in scheduling goes like this: 1. Reader A enters pop_front on an empty queue, goes to wait on the condition variable 2. Writer W enters push_back, adds an element to list and releases mutex and get pre-empted (on line 25 first example) 3. Reader B enters pop_front, sees queue not empty, pops element and leaves 4. Writer W signals condition, wakes up Reader A 5. Reader A wakes up but the queue is empty, goes back to sleep causing unnecessary context switch and trashing the TLB The code is "correct" either way, but it's now optimized for the rare platforms that have a badly implemented condition variable. I would say it's usually better to signal your condition variables with mutexes locked unless you know you're running on a platform with flaky condition variables. Easier to guarantee correctness and avoid spurious wakeups that way. related note: cppreference.com is a terrible resource. It was worse 10 years ago, but it's not improved much. I would not trust the weasel wording ("some implementations" etc) in the link you posted, what it says about pthreads contradicts the pthread man pages (that encourage signal while holding lock)