Before 2019, the entire ConditionVariable system was "giant"-locked:
that is, there was a single global lock that all ConditionVariable
and ConditionVariableEntry operations had to pass through. This of
course was not very performant on multicore systems and when
ConditionVariables see significant use, so I reworked it then to have
more granular locking.
Those patches took a number of attempts to get right, as having two
objects in separate threads that can each access the other not turn
into a deadlock or use-after-free is not easy to say the least,
and the ultimate solution I came up with erased most of the performance
gains I initially saw on the first (partially broken) patchsets.
So I have wanted to revisit this and see if there was a better way
even since then. Recently there have been a few reports of
ConditionVariable-related panics (apparently double unlocks),
notably #16894, and so that was reason enough to actually revisit
this code and see if a better solution could be found.
Well, I think I have come up with one: after this commit, Entries
no longer have their own lock, and instead accesses to Entry members
are almost always atomic; and there is now a case where we spin inside
Variable::_NotifyLocked as well as one in Entry::_RemoveFromVariable.
This leads to somewhat simpler code (no more lock/unlock dance in Notify),
though it is significantly more difficult to understand the nuances of it,
so I have left a sizable number of comments explaining the intricacies
of the new logic.
Note: I initially tried 1000 for "tries", but on a few instances I did see
the panic hit, strangely. I don't think the code that is waited on can
be reasonably reduced any further, so I have just increased the limit to
10000 (which is still well below what spinlocks use.) Hopefully this suffices.
Quick benchmark, x86, compiling HaikuDepot and the mime_db in VMware, 2 cores:
before:
real 0m23.627s
user 0m25.152s
sys 0m7.319s
after:
real 0m23.962s
user 0m25.229s
sys 0m7.330s
Though I occasionally I saw sys times as low as 7.171s, so this seems
to be at least not a regression if not a definitive improvement.
Change-Id: Id042947976885cd5c1433cc4290bdf41b01ed10e
Reviewed-on: https://review.haiku-os.org/c/haiku/+/4727
Tested-by: Commit checker robot <[email protected]>
Reviewed-by: Alex von Gluck IV <[email protected]>
It is no longer an error to destroy a ConditionVariableEntry
that is still attached to a ConditionVariable; it will
now be implicitly detached in that case.
This makes ConditionVariableEntrys much eaiser to use
from an API standpoint.
Change-Id: I03c676d3a198aa885de733d3e1729b15f80de031
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2301
Reviewed-by: waddlesplash <[email protected]>
This "race prevention" code does not seem to be really hit at all
in practice, at least from testing, so no need to do a full
unlock/lock universally for it.
I'm still not sure why the previous fixes here removed 80% of
the performance benefits of the original change; I need to
investigate that more.
This avoids a few potential race conditions mmlr pointed out on
the mailing list. See inline comments.
Change-Id: I605523c1d2683c749751599c417a68a20c70edea
Before this commit, *all* ConditionVariable operations (yes, all;
even Wait, Notify, etc.) went through a single spinlock, that also
protected the sConditionVariableHash. This obviously does not scale
so well with core count, to say the least!
With this commit, we add spinlocks to each Variable and Entry.
This makes locking somewhat more complicated (and nuanced; see
inline comment), but the trade-off seems completely worth it:
(compile HaikuDepot in VMware, 2 cores)
before
real 1m20.219s
user 1m5.619s
sys 0m40.724s
after
real 1m12.667s
user 0m57.684s
sys 0m37.251s
The more cores there are, the more of an optimization this will
likely prove to be. But 10%-across-the-board is not bad to say
the least.
Change-Id: I1e40a997fff58a79e987d7cdcafa8f7358e1115a
This way the static functions (used only in KDL) are below
the actual notification methods, which are now all grouped
together in one place.
No functional change.
* Thread::scheduler_lock protects thread state, priority, etc.
* sThreadCreationLock protects thread creation and removal and list of
threads in team.
* Team::signal_lock and Team::time_lock protect list of threads in team
as well.
* Scheduler uses its own internal locking.
* Most of this is incorrect printf format strings. Changed all strings
causing errors to use the B_PRI* format string definitions, which
means the strings should be correct across all platforms.
* Some other fixes for errors, casts required, etc.
* Reorganized the kernel locking related to threads and teams.
* We now discriminate correctly between process and thread signals. Signal
handlers have been moved to teams. Fixes#5679.
* Implemented real-time signal support, including signal queuing, SA_SIGINFO
support, sigqueue(), sigwaitinfo(), sigtimedwait(), waitid(), and the addition
of the real-time signal range. Closes#1935 and #2695.
* Gave SIGBUS a separate signal number. Fixes#6704.
* Implemented <time.h> clock and timer support, and fixed/completed alarm() and
[set]itimer(). Closes#5682.
* Implemented support for thread cancellation. Closes#5686.
* Moved send_signal() from <signal.h> to <OS.h>. Fixes#7554.
* Lots over smaller more or less related changes.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@42116 a95241bf-73f2-0310-859d-f6bbb57e9c96
of the condition variable and synchronization subsystem of the freebsd compat
layer which will be committed next.
* Also there was a discussion about adding these functions on the commit
mailing list. The mail in http://www.freelists.org/post/haiku-commits/r34395-in-haikutrunksrclibscompatfreebsd-network-compatsys,3
is a good sum up of it (need to scroll somewhat down, though).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@34458 a95241bf-73f2-0310-859d-f6bbb57e9c96
read out in the ConditionVariableEntry::WaitStatus(). That way you can notify
with a specific status that can be read out on the other end.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33718 a95241bf-73f2-0310-859d-f6bbb57e9c96
Tracker's OpenHashTable.h which it should eventually replace. We've renamed the
class to BOpenHashTable and changed the interface slightly so that HashTableLink
became superfluous.
Adapted all the code that used it. Since the OpenHashTables no longer clash,
this should fix the GCC4 build.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31791 a95241bf-73f2-0310-859d-f6bbb57e9c96
- Moved scheduler listening interface to <listeners.h> and added more
convenient to use templatized notification functions.
- Added a listener mechanism for the wait objects (semaphores, condition
variables, mutex, rw_lock).
* system profiler:
- Hopefully fixed locking issues related to notifying the profiler thread
for good. We still had an inconsistent locking order, since the scheduler
notification callbacks are invoked with the thread lock held and have to
acquire the object lock then, while the other callbacks acquired the object
lock first and as a side effect of ConditionVariable::NotifyOne() acquired
the thread lock. Now we make sure the object lock is the innermost lock.
- Track the number of dropped events due to a full buffer.
_user_system_profiler_next_buffer() returns this count now.
- When scheduling profiling events are requested also listen to wait objects
and generate the respective profiling events. We send those events lazily
and cache the infos to avoid resending an event for the same wait object.
- When starting profiling we do now generate "thread scheduled" events for
the already running threads.
- _user_system_profiler_start(): Check whether the parameters pointer is a
userland address at all.
- The system_profiler_team_added event does now also contain the team's name.
* Added a sem_get_name_unsafe() returning a semaphore's name. It is "unsafe",
since the caller has to ensure that the semaphore exists and continues to
exist as long as the returned name is used.
* Adjusted the "profile" and "scheduling_recorder" according to the system
profiling changes. The latter prints the number of dropped events, now.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@30345 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Renamed DoublyLinkedList::Size() to Count(), since it actually counts the
items (ie. O(n)).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@29979 a95241bf-73f2-0310-859d-f6bbb57e9c96
unscheduled was incorrect.
* Introduced _kern_analyze_scheduling() syscall. It requires scheduler
kernel tracing to be enabled. It uses the tracing entries for a given
period of time to do a similar analysis the "scheduler" debugger
command does (i.e. number of runs, run time, latencies, preemption
times) for each thread. Additionally the analysis includes for each
thread how long the thread waited on each locking primitive in total.
* Added kernel tracing for the creation of semaphores and initialization
of condition variables, mutexes, and rw locks. The enabling macro is
SCHEDULING_ANALYSIS_TRACING. The only purpose is to provide
_kern_analyze_scheduling() with more info on the locking primitives
(the name in particular).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@27304 a95241bf-73f2-0310-859d-f6bbb57e9c96
gTeamSpinlock.
* Renamed the static global variables in smp.c to match our style guide.
* Minor other cleanup.
* Removed superfluous white space.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26730 a95241bf-73f2-0310-859d-f6bbb57e9c96
don't check any longer whether the given number is the address of a
condition variable in the global hash table; we just assume it is a
valid condition variable pointer.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26674 a95241bf-73f2-0310-859d-f6bbb57e9c96
introduces the following relevant changes:
* VMCache:
- Renamed vm_cache to VMCache, merged it with vm_store and made it a
C++ class with virtual methods (replacing the store operations).
Turned the different store implementations into subclasses.
- Introduced MergeStore() callback, changed semantics of Commit().
- Changed locking and referencing semantics. A reference can only be
acquired/released with the cache locked. An unreferenced cache is
deleted and a mergeable cache merged when it is unlocked. This
removes the "busy" state of a cache and simplifies the page fault
code.
* Added VMAnonymousCache, which will implement swap support (work by
Zhao Shuai). It is not integrated and used yet, though.
* Enabled the mutex/recursive lock holder asserts.
* Fixed DoublyLinkedList::Swap().
* Generalized the low memory handler to a low resource handler. And made
semaphores and reserved memory handled resources. Made
vm_try_resource_memory() optionally wait (with timeout), and used that
feature to reserve memory for areas.
...
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26572 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Removed the superfluous "flags" parameter from ConditionVariable::Add()
that we forgot there when we moved the flags field from
ConditionVariableEntry::Add() to Wait().
* Using this method was therefore not a good idea - only UnixFifo did, though.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26454 a95241bf-73f2-0310-859d-f6bbb57e9c96
always returned B_OK) by a Init() method, which sets the initial size
and returns an error, if that fails.
* Adjusted code using the classes accordingly. Replaced a few
InitCheck() methods in the network code by Init().
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26127 a95241bf-73f2-0310-859d-f6bbb57e9c96
Changed condition variables so that it is allowed to block (e.g. lock
mutexes etc.) between Add() and Wait(). This fixes#2059, since the
block writer used them this way and could thusly fail to wait for a
condition variable, causing a temporary stack object to be used past its
lifetime.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25525 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added support for non-published condition variables. One has to call
Init() on those, and add entries directly with the
ConditionVariable::Add() method.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25110 a95241bf-73f2-0310-859d-f6bbb57e9c96
respective Private* base class.
* Changed sigwait() and sigsuspend() to use thread_block() instead of a
condition variable.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25100 a95241bf-73f2-0310-859d-f6bbb57e9c96
thread_block(), thread_unblock(),...) that allow a thread to wait for
something without needing a semaphore or condition variable. It can
simply block and another thread can unblock it. Supports timeouts and
interrupting. Both semaphores and condition variables use this
common mechanism, now.
* Semaphores:
- Some simplifications due to the thread blocking mechanism.
- Changed locking order to sem -> thread. It was the other way around
before and when introducing the wait_for_objects() support I had
also introduced a situation where the locking was reverse, which
could potentially cause a dead lock on SMP systems.
- Instead of queueing thread structures, a semaphore queues
queued_thread entries now, which are created on the stack. The
thread::sem structure could thus be removed.
- Added sem_entry::net_count, which is sem_entry::count plus the
acquisition count of all waiting threads. This number is needed in
remove_thread_from_sem() and instead of computing it there we
maintain it.
- Fixed remove_thread_from_sem(). It would not unblock threads, if
the sem count was <= 0.
- Made sem::last_acquirer unconditional. It is actually needed for
sem_info::latest_holder. Fixed fill_sem_info() accordingly.
- Added some optional tracing output, though only via ktrace_printf().
* Condition variables:
- Could be simplified significantly through the use of the thread
blocking mechanism. Removed a good deal of unnecessary code.
- Moved the ConditionVariableEntry "flags" parameter from Wait() to
Add(), and adjusted all places where condition variables are used
accordingly.
* snooze() uses thread_block_with_timeout() instead of a semaphore.
* Simplified thread interrupting in the signal and user debugger code.
Instead of separate functions for threads waiting on a semaphore or
condititon variable, we only have a single thread_interrupt(), now.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25099 a95241bf-73f2-0310-859d-f6bbb57e9c96
variable at once. It wasn't used anymore, and the code gets simpler.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25080 a95241bf-73f2-0310-859d-f6bbb57e9c96
interrupted early, i.e. when signals were already pending when we
started waiting. Fixes#1832.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24063 a95241bf-73f2-0310-859d-f6bbb57e9c96
Before starting to wait on a condition variable check for pending
signals first, if the call is interruptable.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23980 a95241bf-73f2-0310-859d-f6bbb57e9c96
This should be in line with all uses of scheduler_enqueue_in_run_queue() and
simplifies a few places where it is used.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23738 a95241bf-73f2-0310-859d-f6bbb57e9c96
and thread spinlock was reverse in Wait() and Notify(). The thread lock
is now the outer lock -- this way it is still possible to call Notify()
with the thread lock being held.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@22401 a95241bf-73f2-0310-859d-f6bbb57e9c96
Condition variables would never be interrupted.
* ConditionVariableEntry::Add() did not correctly insert the entry into
the per-thread list of entries (the next link of the previous entry
was not adjusted), which could leave the entry unnotified when the
previous entry was notified, thus leaving it in the respective
condition variable's list after the end of its life time. This should
fix a crashing bug I rarely encountered.
* Added debug checks in the PrivateConditionVariableEntry
constructor/destructor that should have helped me to find
forementioned bug hours earlier, had I been bright enough to realize
that I didn't include <debug.h> and those KDEBUG guarded checks were
never executed. :-/
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@22151 a95241bf-73f2-0310-859d-f6bbb57e9c96
removed from the variable, which could lead to crashes under certain
cicrumstances.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@22090 a95241bf-73f2-0310-859d-f6bbb57e9c96
exists and thread was notified), B_ENTRY_NOT_FOUND (condition variable
not found or Unpublish()ed while waiting), or B_INTERRUPTED
(interrupted by a signal).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@22083 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Removed left-over commented C implementation.
* It is now possible for a thread to wait for more than one condition
variable at a time.
* Made waiting for condition variables optionally interruptable.
* Renamed Notify() method to NotifyAll() and added a NotifyOne(), so
that it is now possible to wake up only one of the waiting threads.
Pretty much untested at the moment.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@22081 a95241bf-73f2-0310-859d-f6bbb57e9c96
cheap means to block threads until notified explicitely.
threads
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21874 a95241bf-73f2-0310-859d-f6bbb57e9c96
Augustin Cavalier