This eliminates ping-ponging of locks, where the idle KSE wakes
up only to find the lock it needs is being held. This gives
little or no gain to M:N mode but greatly speeds up 1:1 mode.
Reviewed & Tested by: davidxu
handed-off/signaled to a higher priority thread. Note that when
there are idle KSEs that could run the higher priority thread,
we still add the preemption point because it seems to take the
kernel a while to schedule an idle KSE. The drawbacks are that
threads will be swapped more often between CPUs (KSEs) and
that there will be an extra userland context switch (the idle
KSE is still woken and will probably resume the preempted
thread). We'll revisit this if and when idle CPU/KSE wakeup
times improve.
Inspired by: Petri Helenius <pete@he.iki.fi>
Reviewed by: davidxu
path, making them suitable for direct use by the dynamic loader.
Register libpthread-specific locking API with rtld on startup.
This still has some rough edges with signals which should be
addresses later.
Approved by: re (scottl)
a lock is being waitied on.
Fix a races in join and cancellation.
When trying to wait on a CV and the library is not yet
threaded, make it threaded so that waiting actually works.
When trying to nanosleep() and we're not threaded, just
call the system call nanosleep instead of adding the thread
to the wait queue.
Clean up adding/removing new threads to the "all threads queue",
assigning them unique ids, and tracking how many active threads
there are. Do it all when the thread is added to the scheduling
queue instead of making pthread_create() know how to do it.
Fix a race where a thread could be marked for signal delivery
but it could be exited before we actually add the signal to it.
Other minor cleanups and bug fixes.
Submitted by: davidxu
Approved by: re@ (blanket for libpthread)
be external (initialize()!).
Remove cancellation points from _pthread_cond_wait and
_pthread_cond_timedwait (single underscore versions are
libc private functions). Point the weak reference(!) for
these functions to the versions with cancellation points.
Approved by: re@(blanket till 5/19)
Pointed out by: kan (cancellation point bug)
lock held (_thr_sched_switch_unlocked()) and use this to avoid
dropping the scheduler lock and having the scheduler retake the
same lock again.
Add a better way of detecting if a low-level lock is in use.
When switching out a thread due to blocking in the UTS, don't
switch to the KSE's scheduler stack only to switch back to
another thread. If possible switch to the new thread directly
from the old thread and avoid the overhead of the extra
context switch.
Check for pending signals on a thread when entering the scheduler
and add them to the threads signal frame. This includes some
other minor signal fixes.
Most of this was a joint effor between davidxu and myself.
Reviewed by: davidxu
Approved by: re@ (blanket for libpthread)
KSEs when it's thread exits; allow the GC handler to do that.
o Make spinlock/spinlock critical regions.
The following were submitted by davidxu
o Alow thr_switch() to take a null mailbox argument.
o Better protect cancellation checks.
o Don't set KSE specific data when creating new KSEs; rely on the
first upcall of the KSE to set it.
o Add the ability to set the maximum concurrency level and do this
automatically. We should have a way to enable/disable this with
some sort of tunable because some applications may not want this
to be the default.
o Hold the scheduling lock across thread switch calls.
o If scheduling of a thread fails, make sure to remove it from the list
of active threads.
o Better protect accesses to a joining threads when the target thread is
exited and detached.
o Remove some macro definitions that are now provided by <sys/kse.h>.
o Don't leave the library in threaded mode if creation of the initial
KSE fails.
o Wakeup idle KSEs when there are threads ready to run.
o Maintain the number of threads active in the priority queue.
lock level is 0. Thus far, the threads implementation doesn't use
mutexes or condition variables so the lock level should be 0.
Save the return value when trying to schedule a new thread and
use this to return an error from pthread_create().
Change the max sleep time for an idle KSE to 1 minute from 2 minutes.
Maintain a count of the number of KSEs within a KSEG.
With these changes scope system threads seem to work, but heavy
use of them crash the kernel (supposedly VM bugs).
to be instances where the kernel doesn't properly save and/or
restore it.
Use noupcall and nocompleted flags in the KSE mailbox. These
require kernel changes to work which will be committed sometime
later. Things still work without the changes.
Remove the general kse entry function and use two different
functions -- one for scope system threads and one for scope
process threads. The scope system function is not yet enabled
and we use the same function for all threads at the moment.
Keep a copy of the KSE stack for the case that a KSE runs
a scope system thread and uses the same stack as the thread
(no upcalls are generated, so a separate stack isn't needed).
This isn't enabled yet.
Use a separate field for the KSE waiting flag. It isn't
correct to use the mailbox flags field.
The following fixes were provided by David Xu:
o Initialize condition variable locks with thread versions
of the low-level locking functions instead of the kse versions.
o Enable threading before creating the first thread instead
of after.
o Don't enter critical regions when trying to malloc/free
or call functions that malloc/free.
o Take the scheduling lock when inheriting thread attributes.
o Check the attribute's stack pointer instead of the
attributes stack size for null when allocating a
thread's stack.
o Add a kseg reinit function so we don't have to destroy and
then recreate the same lock.
o Check the return value of kse_create() and return an
appropriate error if it fails.
o Don't forget to destroy a thread's locks when freeing it.
o Examine the correct flags word for checking to see if
a thread is in a synchronization queue.
Things should now work on an SMP kernel.
environment. This includes support for multiple KSEs and KSEGs.
The ability to create more than 1 KSE via pthread_setconcurrency()
is in the works as well as support for PTHREAD_SCOPE_SYSTEM threads.
Those should come shortly.
There are still some known issues which davidxu and I are working
on, but it'll make it easier for us by committing what we have.
This library now passes all of the ACE tests that libc_r passes
with the exception of one. It also seems to work OK with KDE
including konqueror, kwrite, etc. I haven't been able to get
mozilla to run due to lack of java plugin, so I'd be interested
to see how it works with that.
Reviewed by: davidxu
The new libpthread will provide POSIX threading support using KSE.
These files were previously repo-copied from src/lib/libc_r.
Reviewed by: deischen
Approved by: -arch
and pthread_resume_all_np(). These suspend and resume all threads except
the current thread, respectively. The existing functions pthread_single_np()
and pthread_multi_np(), which formerly had no effect, now exhibit the same
behaviour and pthread_suspend_all_np() and pthread_resume_all_np(). These
functions have been added mostly for the native java port.
Don't allow the uthread kernel pipe to use the same descriptors as
stdio. Mostily submitted by Oswald Buddenhagen <ossi@kde.org>.
Correct some minor style nits.
_foo - wrapped system call
foo - weak definition to _foo
and for cancellation points:
_foo - wrapped system call
__foo - enter cancellation point, call _foo(), leave
cancellation point
foo - weak definition to __foo
Change use of global _thread_run to call a function to get the
currently running thread.
Make all pthread_foo functions weak definitions to _pthread_foo,
where _pthread_foo is the implementation. This allows an application
to provide its own pthread functions.
Provide slightly different versions of pthread_mutex_lock and
pthread_mutex_init so that we can tell the difference between
a libc mutex and an application mutex. Threads holding mutexes
internal to libc should never be allowed to exit, call signal
handlers, or cancel.
Approved by: -arch
was not getting properly initialized in pthread_cond_signal()
and pthread_cond_broadcast(). Reportedly, this can cause
an application to die.
MFC candidate
Submitted by: ade
file descriptors needing to be polled (Doh!). Reported
by Dan Nelson <dnelson@emsphone.com>.
Don't install and start the scheduling timer until the
first thread is created. This prevents the overhead of
having a periodic scheduling signal in a single threaded
program. Reported by Dan Nelson <dnelson@emsphone.com>.
Allow builtin longjmps out of application installed
signal handlers without the need perform any post-handler
cleanup:
o Change signal handling to save the threads interrupted
context on the stack. The threads current context is
now always stored in the same place (in the pthread).
If and when a signal handler returns, the interrupted
context is copied back to the storage area in the pthread.
o Before calling invoking a signal handler for a thread,
back the thread out of any internal waiting queues
(mutex, CV, join, etc) to which it belongs.
Rework uthread_info.c a bit to make it easier to change
the format of a thread dump.
Use an alternal signal stack for the thread library's
signal handler. This allows us to fiddle with the main
threads stack without fear of it being in use.
Reviewed by: jasone
thread switches should be on par with that under scheduler
activations.
o Timing is achieved through the use of a fixed interval
timer (ITIMER_PROF) to count scheduling ticks instead
of retrieving the time-of-day upon every thread switch
and calculating elapsed real time.
o Polling for I/O readiness is performed once for each
scheduling tick instead of every thread switch.
o The non-signal saving/restoring versions of setjmp/longjmp
are used to save and restore thread contexts. This may
allow the removal of _THREAD_SAFE macros from setjmp()
and longjmp() - needs more investigation.
Change signal handling so that signals are handled in the
context of the thread that is receiving the signal. When
signals are dispatched to a thread, a special signal handling
frame is created on top of the target threads stack. The
frame contains the threads saved state information and a new
context in which the thread can run. The applications signal
handler is invoked through a wrapper routine that knows how
to restore the threads saved state and unwind to previous
frames.
Fix interruption of threads due to signals. Some states
were being improperly interrupted while other states were
not being interrupted. This should fix several PRs.
Signal handlers, which are invoked as a result of a process
signal (not by pthread_kill()), are now called with the
code (or siginfo_t if SA_SIGINFO was set in sa_flags) and
sigcontext_t as received from the process signal handler.
Modify the search for a thread to which a signal is delivered.
The search algorithm is now:
o First thread found in sigwait() with signal in wait mask.
o First thread found sigsuspend()'d on the signal.
o Current thread if signal is unmasked.
o First thread found with signal unmasked.
Collapse machine dependent support into macros defined in
pthread_private.h. These should probably eventually be moved
into separate MD files.
Change the range of settable priorities to be compliant with
POSIX (0-31). The threads library uses higher priorities
internally for real-time threads (not yet implemented) and
threads executing signal handlers. Real-time threads and
threads running signal handlers add 64 and 32, respectively,
to a threads base priority.
Some other small changes and cleanups.
PR: 17757 18559 21943
Reviewed by: jasone
pthread_cond_signal(), pthread_cond_broadcast(), and pthread_cond_timedwait().
Do not dump core in pthread_cond_timedwait() (due to a NULL pointer
dereference) if attempting to wait on an uninitialized condition variable.
PR: bin/18099
not allowed to return EINTR, but use of pthread_suspend_np() could cause
EINTR to be returned. To fix this, restructure pthread_suspend_np() so that
it does not interrupt a thread that is waiting on a mutex or condition, and
keep enough state around that pthread_resume_np() can fix things up
afterwards.
Reviewed by: deischen
thread waiting on an event (I/O, condvar, etc) will, when resumed using
pthread_resume_np, return with EINTR. For example, suspending and resuming
a thread blocked on read() will not requeue the thread for the read, but
will return -1 with errno = EINTR. If the suspended thread is in a critical
region, the thread is suspended as soon as it leaves the critical region.
Fix a bogon in pthread_kill() where a signal was being delivered twice
to threads waiting in sigwait().
Reported by (suspend/resume bug): jdp
Reviewed by: jasone
just use _foo() <-- foo(). In the case of a libpthread that doesn't do
call conversion (such as linuxthreads and our upcoming libpthread), this
is adequate. In the case of libc_r, we still need three names, which are
now _thread_sys_foo() <-- _foo() <-- foo().
Convert all internal libc usage of: aio_suspend(), close(), fsync(), msync(),
nanosleep(), open(), fcntl(), read(), and write() to _foo() instead of foo().
Remove all internal libc usage of: creat(), pause(), sleep(), system(),
tcdrain(), wait(), and waitpid().
Make thread cancellation fully POSIX-compliant.
Suggested by: deischen
signal handler. Explicitly check for jumps to anywhere other than the
current stack, since such jumps are undefined according to POSIX.
While we're at it, convert thread cancellation to use continuations, since
it's cleaner than the original cancellation code.
Avoid delivering a signal to a thread twice. This was a pre-existing bug,
but was likely unexposed until these other changes were made.
Defer signals generated by pthread_kill() so that they can be delivered on
the appropriate stack. deischen claims that this is unnecessary, which is
likely true, but without this change, pthread_kill() can cause undefined
priority queue states and/or PANICs in [sig|_]longjmp(), so I'm leaving
this in for now. To compile this code out and exercise the bug, define
the _NO_UNDISPATCH cpp macro. Defining _PTHREADS_INVARIANTS as well will
cause earlier crashes.
PR: kern/14685
Collaboration with: deischen
o Cancellation flags were not getting properly set/cleared.
o Loops waiting for internal locks were not being exited
correctly by a cancelled thread.
o Minor spelling (cancelation -> cancellation) and formatting
corrections (missing tab).
Found by: tg
Reviewed by: jasone
eischen (Daniel Eischen) added wrappers to protect against cancled
threads orphaning internal resources.
the cancelability code is still a bit fuzzy but works for test
programs of my own, OpenBSD's and some examples from ORA's books.
add readdir_r to both libc and libc_r
add some 'const' attributes to function parameters
Reviewed by: eischen, jasone
o The polling mechanism for I/O readiness was changed from
select() to poll(). In additon, a wrapped version of poll()
is now provided.
o The wrapped select routine now converts each fd_set to a
poll array so that the thread scheduler doesn't have to
perform a bitwise search for selected fds each time file
descriptors are polled for I/O readiness.
o The thread scheduler was modified to use a new queue (_workq)
for threads that need work. Threads waiting for I/O readiness
and spinblocks are added to the work queue in addition to the
waiting queue. This reduces the time spent forming/searching
the array of file descriptors being polled.
o The waiting queue (_waitingq) is now maintained in order of
thread wakeup time. This allows the thread scheduler to
find the nearest wakeup time by looking at the first thread
in the queue instead of searching the entire queue.
o Removed file descriptor locking for select/poll routines. An
application should not rely on the threads library for providing
this locking; if necessary, the application should use mutexes
to protect selecting/polling of file descriptors.
o Retrieve and use the kernel clock rate/resolution at startup
instead of hardcoding the clock resolution to 10 msec (tested
with kernel running at 1000 HZ).
o All queues have been changed to use queue.h macros. These
include the queues of all threads, dead threads, and threads
waiting for file descriptor locks.
o Added reinitialization of the GC mutex and condition variable
after a fork. Also prevented reallocation of the ready queue
after a fork.
o Prevented the wrapped close routine from closing the thread
kernel pipes.
o Initialized file descriptor table for stdio entries at thread
init.
o Provided additional flags to indicate to what queues threads
belong.
o Moved TAILQ initialization for statically allocated mutex and
condition variables to after the spinlock.
o Added dispatching of signals to pthread_kill. Removing the
dispatching of signals from thread activation broke sigsuspend
when pthread_kill was used to send a signal to a thread.
o Temporarily set the state of a thread to PS_SUSPENDED when it
is first created and placed in the list of threads so that it
will not be accidentally scheduled before becoming a member
of one of the scheduling queues.
o Change the signal handler to queue signals to the thread kernel
pipe if the scheduling queues are protected. When scheduling
queues are unprotected, signals are then dequeued and handled.
o Ensured that all installed signal handlers block the scheduling
signal and that the scheduling signal handler blocks all
other signals. This ensures that the signal handler is only
interruptible for and by non-scheduling signals. An atomic
lock is used to decide which instance of the signal handler
will handle pending signals.
o Removed _lock_thread_list and _unlock_thread_list as they are
no longer used to protect the thread list.
o Added missing RCS IDs to modified files.
o Added checks for appropriate queue membership and activity when
adding, removing, and searching the scheduling queues. These
checks add very little overhead and are enabled when compiled
with _PTHREADS_INVARIANTS defined. Suggested and implemented
by Tor Egge with some modification by me.
o Close a race condition in uthread_close. (Tor Egge)
o Protect the scheduling queues while modifying them in
pthread_cond_signal and _thread_fd_unlock. (Tor Egge)
o Ensure that when a thread gets a mutex, the mutex is on that
threads list of owned mutexes. (Tor Egge)
o Set the kernel-in-scheduler flag in _thread_kern_sched_state
and _thread_kern_sched_state_unlock to prevent a scheduling
signal from calling the scheduler again. (Tor Egge)
o Don't use TAILQ_FOREACH macro while searching the waiting
queue for threads in a sigwait state, because a change of
state destroys the TAILQ link. It is actually safe to do
so, though, because once a sigwaiting thread is found, the
loop ends and the function returns. (Tor Egge)
o When dispatching signals to threads, make the thread inherit
the signal deferral flag of the currently running thread.
(Tor Egge)
Submitted by: Daniel Eischen <eischen@vigrid.com> and
Tor Egge <Tor.Egge@fast.no>
o Runnable threads are now maintained in priority queues. The
implementation requires two things:
1.) The priority queues must be protected during insertion
and removal of threads. Since the kernel scheduler
must modify the priority queues, a spinlock for
protection cannot be used. The functions
_thread_kern_sched_defer() and _thread_kern_sched_undefer()
were added to {un}defer kernel scheduler activation.
2.) A thread (active) priority change can be performed only
when the thread is removed from the priority queue. The
implementation uses a threads active priority when
inserting it into the queue.
A by-product is that thread switches are much faster. A
separate queue is used for waiting and/or blocked threads,
and it is searched at most 2 times in the kernel scheduler
when there are active threads. It should be possible to
reduce this to once by combining polling of threads waiting
on I/O with the loop that looks for timed out threads and
the minimum timeout value.
o Functions to defer kernel scheduler activation were added. These
are _thread_kern_sched_defer() and _thread_kern_sched_undefer()
and may be called recursively. These routines do not block the
scheduling signal, but latch its occurrence. The signal handler
will not call the kernel scheduler when the running thread has
deferred scheduling, but it will be called when running thread
undefers scheduling.
o Added support for _POSIX_THREAD_PRIORITY_SCHEDULING. All the
POSIX routines required by this should now be implemented.
One note, SCHED_OTHER, SCHED_FIFO, and SCHED_RR are required
to be defined by including pthread.h. These defines are currently
in sched.h. I modified pthread.h to include sched.h but don't
know if this is the proper thing to do.
o Added support for priority protection and inheritence mutexes.
This allows definition of _POSIX_THREAD_PRIO_PROTECT and
_POSIX_THREAD_PRIO_INHERIT.
o Added additional error checks required by POSIX for mutexes and
condition variables.
o Provided a wrapper for sigpending which is marked as a hidden
syscall.
o Added a non-portable function as a debugging aid to allow an
application to monitor thread context switches. An application
can install a routine that gets called everytime a thread
(explicitly created by the application) gets context switched.
The routine gets passed the pthread IDs of the threads that are
being switched in and out.
Submitted by: Dan Eischen <eischen@vigrid.com>
Changes by me:
o Added a PS_SPINBLOCK state to deal with the priority inversion
problem most often (I think) seen by threads calling malloc/free/realloc.
o Dispatch signals to the running thread directly rather than at a
context switch to avoid the situation where the switch never occurs.
line number every time a file descriptor is locked.
This looks like a big change but it isn't. It should reduce the size
of libc_r and make it run slightly faster.
for the process, not a separate set for each thread). By default, the
process now only has signal handlers installed for SIGVTALRM, SIGINFO
and SIGCHLD. The thread kernel signal handler is installed for other
signals on demand. This means that SIG_IGN and SIG_DFL processing is now
left to the kernel, not the thread kernel.
Change the signal dispatch to no longer use a signal thread, and
call the signal handler using the stack of the thread that has the
signal pending.
Change the atomic lock method to use test-and-set asm code with
a yield if blocked. This introduces separate locks for each type
of object instead of blocking signals to prevent a context
switch. It was this blocking of signals that caused the performance
degradation the people have noted.
This is a *big* change!
functions would return -1 and set errno to indicate the specific error.
POSIX requires that the functions return the error code as the return
value of the function instead.
Here are the diffs for libc_r to get it one step closer to P1003.1c
These make most of the thread/mutex/condvar structures opaque to the
user. There are three functions which have been renamed with _np
suffixes because they are extensions to P1003.1c (I did them for JAVA,
which needs to suspend/resume threads and also start threads suspended).
I've created a new header (pthread_np.h) for the non-POSIX stuff.
The egrep tags stuff in /usr/src/lib/libc_r/Makefile that I uncommented
doesn't work. I think its best to delete it. I don't think libc_r needs
tags anyway, 'cause most of the source is in libc which does have tags.
also:
Here's the first batch of man pages for the thread functions.
The diff to /usr/src/lib/libc_r/Makefile removes some stuff that was
inherited from /usr/src/lib/libc/Makefile that should only be done with
libc.
also:
I should have sent this diff with the pthread(3) man page.
It allows people to type
make -DWANT_LIBC_R world
to get libc_r built with the rest of the world. I put this in the
pthread(3) man page. The default is still not to build libc_r.
also:
The diff attached adds a pthread(3) man page to /usr/src/share/man/man3.
The idea is that without libc_r installed, this man page will give people
enough info to know that they have to build libc_r.
