freebsd-skq/lib/libc_r/uthread/uthread_mutex.c

1543 lines
39 KiB
C
Raw Normal View History

/*
* Copyright (c) 1995 John Birrell <jb@cimlogic.com.au>.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the author nor the names of any co-contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY JOHN BIRRELL AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
1999-08-28 00:22:10 +00:00
* $FreeBSD$
*/
#include <stdlib.h>
#include <errno.h>
#include <string.h>
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
#include <sys/param.h>
#include <sys/queue.h>
#include <pthread.h>
#include "pthread_private.h"
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
#if defined(_PTHREADS_INVARIANTS)
#define _MUTEX_INIT_LINK(m) do { \
(m)->m_qe.tqe_prev = NULL; \
(m)->m_qe.tqe_next = NULL; \
} while (0)
#define _MUTEX_ASSERT_IS_OWNED(m) do { \
if ((m)->m_qe.tqe_prev == NULL) \
PANIC("mutex is not on list"); \
} while (0)
#define _MUTEX_ASSERT_NOT_OWNED(m) do { \
if (((m)->m_qe.tqe_prev != NULL) || \
((m)->m_qe.tqe_next != NULL)) \
PANIC("mutex is on list"); \
} while (0)
#else
#define _MUTEX_INIT_LINK(m)
#define _MUTEX_ASSERT_IS_OWNED(m)
#define _MUTEX_ASSERT_NOT_OWNED(m)
#endif
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Prototypes
*/
static inline int mutex_self_trylock(pthread_mutex_t);
static inline int mutex_self_lock(pthread_mutex_t);
static inline int mutex_unlock_common(pthread_mutex_t *, int);
static void mutex_priority_adjust(pthread_mutex_t);
static void mutex_rescan_owned (pthread_t, pthread_mutex_t);
static inline pthread_t mutex_queue_deq(pthread_mutex_t);
static inline void mutex_queue_remove(pthread_mutex_t, pthread_t);
static inline void mutex_queue_enq(pthread_mutex_t, pthread_t);
static spinlock_t static_init_lock = _SPINLOCK_INITIALIZER;
static struct pthread_mutex_attr static_mutex_attr =
PTHREAD_MUTEXATTR_STATIC_INITIALIZER;
static pthread_mutexattr_t static_mattr = &static_mutex_attr;
/* Single underscore versions provided for libc internal usage: */
__weak_reference(__pthread_mutex_trylock, pthread_mutex_trylock);
__weak_reference(__pthread_mutex_lock, pthread_mutex_lock);
/* No difference between libc and application usage of these: */
__weak_reference(_pthread_mutex_init, pthread_mutex_init);
__weak_reference(_pthread_mutex_destroy, pthread_mutex_destroy);
__weak_reference(_pthread_mutex_unlock, pthread_mutex_unlock);
/*
* Reinitialize a private mutex; this is only used for internal mutexes.
*/
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
int
_mutex_reinit(pthread_mutex_t * mutex)
{
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret = 0;
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
if (mutex == NULL)
ret = EINVAL;
else if (*mutex == NULL)
ret = _pthread_mutex_init(mutex, NULL);
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
else {
/*
* Initialize the mutex structure:
*/
(*mutex)->m_type = PTHREAD_MUTEX_DEFAULT;
(*mutex)->m_protocol = PTHREAD_PRIO_NONE;
TAILQ_INIT(&(*mutex)->m_queue);
(*mutex)->m_owner = NULL;
(*mutex)->m_data.m_count = 0;
(*mutex)->m_flags |= MUTEX_FLAGS_INITED | MUTEX_FLAGS_PRIVATE;
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
(*mutex)->m_refcount = 0;
(*mutex)->m_prio = 0;
(*mutex)->m_saved_prio = 0;
_MUTEX_INIT_LINK(*mutex);
memset(&(*mutex)->lock, 0, sizeof((*mutex)->lock));
}
return (ret);
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
int
_pthread_mutex_init(pthread_mutex_t * mutex,
const pthread_mutexattr_t * mutex_attr)
{
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
enum pthread_mutextype type;
int protocol;
int ceiling;
int flags;
pthread_mutex_t pmutex;
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret = 0;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
if (mutex == NULL)
ret = EINVAL;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Check if default mutex attributes: */
if (mutex_attr == NULL || *mutex_attr == NULL) {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Default to a (error checking) POSIX mutex: */
type = PTHREAD_MUTEX_ERRORCHECK;
protocol = PTHREAD_PRIO_NONE;
ceiling = PTHREAD_MAX_PRIORITY;
flags = 0;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
/* Check mutex type: */
else if (((*mutex_attr)->m_type < PTHREAD_MUTEX_ERRORCHECK) ||
((*mutex_attr)->m_type >= PTHREAD_MUTEX_TYPE_MAX))
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Return an invalid argument error: */
ret = EINVAL;
/* Check mutex protocol: */
else if (((*mutex_attr)->m_protocol < PTHREAD_PRIO_NONE) ||
((*mutex_attr)->m_protocol > PTHREAD_MUTEX_RECURSIVE))
/* Return an invalid argument error: */
ret = EINVAL;
else {
/* Use the requested mutex type and protocol: */
type = (*mutex_attr)->m_type;
protocol = (*mutex_attr)->m_protocol;
ceiling = (*mutex_attr)->m_ceiling;
flags = (*mutex_attr)->m_flags;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
/* Check no errors so far: */
if (ret == 0) {
if ((pmutex = (pthread_mutex_t)
malloc(sizeof(struct pthread_mutex))) == NULL)
ret = ENOMEM;
else {
/* Set the mutex flags: */
pmutex->m_flags = flags;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Process according to mutex type: */
switch (type) {
/* case PTHREAD_MUTEX_DEFAULT: */
case PTHREAD_MUTEX_ERRORCHECK:
case PTHREAD_MUTEX_NORMAL:
/* Nothing to do here. */
break;
/* Single UNIX Spec 2 recursive mutex: */
case PTHREAD_MUTEX_RECURSIVE:
/* Reset the mutex count: */
pmutex->m_data.m_count = 0;
break;
/* Trap invalid mutex types: */
default:
/* Return an invalid argument error: */
ret = EINVAL;
break;
}
if (ret == 0) {
/* Initialise the rest of the mutex: */
TAILQ_INIT(&pmutex->m_queue);
pmutex->m_flags |= MUTEX_FLAGS_INITED;
pmutex->m_owner = NULL;
pmutex->m_type = type;
pmutex->m_protocol = protocol;
pmutex->m_refcount = 0;
if (protocol == PTHREAD_PRIO_PROTECT)
pmutex->m_prio = ceiling;
else
pmutex->m_prio = 0;
pmutex->m_saved_prio = 0;
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_INIT_LINK(pmutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
memset(&pmutex->lock, 0, sizeof(pmutex->lock));
*mutex = pmutex;
} else {
free(pmutex);
*mutex = NULL;
}
}
}
/* Return the completion status: */
return (ret);
}
int
_pthread_mutex_destroy(pthread_mutex_t * mutex)
{
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret = 0;
if (mutex == NULL || *mutex == NULL)
ret = EINVAL;
else {
/* Lock the mutex structure: */
_SPINLOCK(&(*mutex)->lock);
/*
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
* Check to see if this mutex is in use:
*/
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
if (((*mutex)->m_owner != NULL) ||
(TAILQ_FIRST(&(*mutex)->m_queue) != NULL) ||
((*mutex)->m_refcount != 0)) {
ret = EBUSY;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Unlock the mutex structure: */
_SPINUNLOCK(&(*mutex)->lock);
}
else {
/*
* Free the memory allocated for the mutex
* structure:
*/
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_ASSERT_NOT_OWNED(*mutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
free(*mutex);
/*
* Leave the caller's pointer NULL now that
* the mutex has been destroyed:
*/
*mutex = NULL;
}
}
/* Return the completion status: */
return (ret);
}
static int
init_static(pthread_mutex_t *mutex)
{
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret;
_SPINLOCK(&static_init_lock);
1998-08-02 23:07:25 +00:00
if (*mutex == NULL)
ret = _pthread_mutex_init(mutex, NULL);
else
ret = 0;
_SPINUNLOCK(&static_init_lock);
return (ret);
}
static int
init_static_private(pthread_mutex_t *mutex)
{
int ret;
_SPINLOCK(&static_init_lock);
if (*mutex == NULL)
ret = _pthread_mutex_init(mutex, &static_mattr);
else
ret = 0;
_SPINUNLOCK(&static_init_lock);
return (ret);
}
static int
mutex_trylock_common(pthread_mutex_t *mutex)
{
struct pthread *curthread = _get_curthread();
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret = 0;
PTHREAD_ASSERT((mutex != NULL) && (*mutex != NULL),
"Uninitialized mutex in pthread_mutex_trylock_basic");
/*
* Defer signals to protect the scheduling queues from
* access by the signal handler:
*/
_thread_kern_sig_defer();
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
/* Lock the mutex structure: */
_SPINLOCK(&(*mutex)->lock);
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
/*
* If the mutex was statically allocated, properly
* initialize the tail queue.
*/
if (((*mutex)->m_flags & MUTEX_FLAGS_INITED) == 0) {
TAILQ_INIT(&(*mutex)->m_queue);
_MUTEX_INIT_LINK(*mutex);
(*mutex)->m_flags |= MUTEX_FLAGS_INITED;
}
/* Process according to mutex type: */
switch ((*mutex)->m_protocol) {
/* Default POSIX mutex: */
case PTHREAD_PRIO_NONE:
/* Check if this mutex is not locked: */
if ((*mutex)->m_owner == NULL) {
/* Lock the mutex for the running thread: */
(*mutex)->m_owner = curthread;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Add to the list of owned mutexes: */
_MUTEX_ASSERT_NOT_OWNED(*mutex);
TAILQ_INSERT_TAIL(&curthread->mutexq,
(*mutex), m_qe);
} else if ((*mutex)->m_owner == curthread)
ret = mutex_self_trylock(*mutex);
else
/* Return a busy error: */
ret = EBUSY;
break;
/* POSIX priority inheritence mutex: */
case PTHREAD_PRIO_INHERIT:
/* Check if this mutex is not locked: */
if ((*mutex)->m_owner == NULL) {
/* Lock the mutex for the running thread: */
(*mutex)->m_owner = curthread;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Track number of priority mutexes owned: */
curthread->priority_mutex_count++;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* The mutex takes on the attributes of the
* running thread when there are no waiters.
*/
(*mutex)->m_prio = curthread->active_priority;
(*mutex)->m_saved_prio =
curthread->inherited_priority;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Add to the list of owned mutexes: */
_MUTEX_ASSERT_NOT_OWNED(*mutex);
TAILQ_INSERT_TAIL(&curthread->mutexq,
(*mutex), m_qe);
} else if ((*mutex)->m_owner == curthread)
ret = mutex_self_trylock(*mutex);
else
/* Return a busy error: */
ret = EBUSY;
break;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* POSIX priority protection mutex: */
case PTHREAD_PRIO_PROTECT:
/* Check for a priority ceiling violation: */
if (curthread->active_priority > (*mutex)->m_prio)
ret = EINVAL;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Check if this mutex is not locked: */
else if ((*mutex)->m_owner == NULL) {
/* Lock the mutex for the running thread: */
(*mutex)->m_owner = curthread;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Track number of priority mutexes owned: */
curthread->priority_mutex_count++;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* The running thread inherits the ceiling
* priority of the mutex and executes at that
* priority.
*/
curthread->active_priority = (*mutex)->m_prio;
(*mutex)->m_saved_prio =
curthread->inherited_priority;
curthread->inherited_priority =
(*mutex)->m_prio;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Add to the list of owned mutexes: */
_MUTEX_ASSERT_NOT_OWNED(*mutex);
TAILQ_INSERT_TAIL(&curthread->mutexq,
(*mutex), m_qe);
} else if ((*mutex)->m_owner == curthread)
ret = mutex_self_trylock(*mutex);
else
/* Return a busy error: */
ret = EBUSY;
break;
/* Trap invalid mutex types: */
default:
/* Return an invalid argument error: */
ret = EINVAL;
break;
}
/* Unlock the mutex structure: */
_SPINUNLOCK(&(*mutex)->lock);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Undefer and handle pending signals, yielding if
* necessary:
*/
_thread_kern_sig_undefer();
/* Return the completion status: */
return (ret);
}
int
__pthread_mutex_trylock(pthread_mutex_t *mutex)
{
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret = 0;
if (mutex == NULL)
ret = EINVAL;
/*
* If the mutex is statically initialized, perform the dynamic
* initialization:
*/
else if ((*mutex != NULL) || (ret = init_static(mutex)) == 0)
ret = mutex_trylock_common(mutex);
return (ret);
}
int
_pthread_mutex_trylock(pthread_mutex_t *mutex)
{
int ret = 0;
if (mutex == NULL)
ret = EINVAL;
/*
* If the mutex is statically initialized, perform the dynamic
* initialization marking the mutex private (delete safe):
*/
else if ((*mutex != NULL) || (ret = init_static_private(mutex)) == 0)
ret = mutex_trylock_common(mutex);
return (ret);
}
static int
mutex_lock_common(pthread_mutex_t * mutex)
{
struct pthread *curthread = _get_curthread();
int ret = 0;
PTHREAD_ASSERT((mutex != NULL) && (*mutex != NULL),
"Uninitialized mutex in pthread_mutex_trylock_basic");
/* Reset the interrupted flag: */
curthread->interrupted = 0;
/*
* Enter a loop waiting to become the mutex owner. We need a
* loop in case the waiting thread is interrupted by a signal
* to execute a signal handler. It is not (currently) possible
* to remain in the waiting queue while running a handler.
* Instead, the thread is interrupted and backed out of the
* waiting queue prior to executing the signal handler.
*/
do {
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
/*
* Defer signals to protect the scheduling queues from
* access by the signal handler:
*/
_thread_kern_sig_defer();
/* Lock the mutex structure: */
_SPINLOCK(&(*mutex)->lock);
/*
* If the mutex was statically allocated, properly
* initialize the tail queue.
*/
if (((*mutex)->m_flags & MUTEX_FLAGS_INITED) == 0) {
TAILQ_INIT(&(*mutex)->m_queue);
(*mutex)->m_flags |= MUTEX_FLAGS_INITED;
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_INIT_LINK(*mutex);
}
/* Process according to mutex type: */
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
switch ((*mutex)->m_protocol) {
/* Default POSIX mutex: */
case PTHREAD_PRIO_NONE:
if ((*mutex)->m_owner == NULL) {
/* Lock the mutex for this thread: */
(*mutex)->m_owner = curthread;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Add to the list of owned mutexes: */
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_ASSERT_NOT_OWNED(*mutex);
TAILQ_INSERT_TAIL(&curthread->mutexq,
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex), m_qe);
} else if ((*mutex)->m_owner == curthread)
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
ret = mutex_self_lock(*mutex);
else {
/*
* Join the queue of threads waiting to lock
* the mutex:
*/
mutex_queue_enq(*mutex, curthread);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Keep a pointer to the mutex this thread
* is waiting on:
*/
curthread->data.mutex = *mutex;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Unlock the mutex structure and schedule the
* next thread:
*/
_thread_kern_sched_state_unlock(PS_MUTEX_WAIT,
&(*mutex)->lock, __FILE__, __LINE__);
/* Lock the mutex structure again: */
_SPINLOCK(&(*mutex)->lock);
}
break;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* POSIX priority inheritence mutex: */
case PTHREAD_PRIO_INHERIT:
/* Check if this mutex is not locked: */
if ((*mutex)->m_owner == NULL) {
/* Lock the mutex for this thread: */
(*mutex)->m_owner = curthread;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Track number of priority mutexes owned: */
curthread->priority_mutex_count++;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* The mutex takes on attributes of the
* running thread when there are no waiters.
*/
(*mutex)->m_prio = curthread->active_priority;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_saved_prio =
curthread->inherited_priority;
curthread->inherited_priority =
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_prio;
/* Add to the list of owned mutexes: */
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_ASSERT_NOT_OWNED(*mutex);
TAILQ_INSERT_TAIL(&curthread->mutexq,
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex), m_qe);
} else if ((*mutex)->m_owner == curthread)
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
ret = mutex_self_lock(*mutex);
else {
/*
* Join the queue of threads waiting to lock
* the mutex:
*/
mutex_queue_enq(*mutex, curthread);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Keep a pointer to the mutex this thread
* is waiting on:
*/
curthread->data.mutex = *mutex;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
if (curthread->active_priority >
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_prio)
/* Adjust priorities: */
mutex_priority_adjust(*mutex);
/*
* Unlock the mutex structure and schedule the
* next thread:
*/
_thread_kern_sched_state_unlock(PS_MUTEX_WAIT,
&(*mutex)->lock, __FILE__, __LINE__);
/* Lock the mutex structure again: */
_SPINLOCK(&(*mutex)->lock);
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
break;
/* POSIX priority protection mutex: */
case PTHREAD_PRIO_PROTECT:
/* Check for a priority ceiling violation: */
if (curthread->active_priority > (*mutex)->m_prio)
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
ret = EINVAL;
/* Check if this mutex is not locked: */
else if ((*mutex)->m_owner == NULL) {
/*
* Lock the mutex for the running
* thread:
*/
(*mutex)->m_owner = curthread;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Track number of priority mutexes owned: */
curthread->priority_mutex_count++;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* The running thread inherits the ceiling
* priority of the mutex and executes at that
* priority:
*/
curthread->active_priority = (*mutex)->m_prio;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_saved_prio =
curthread->inherited_priority;
curthread->inherited_priority =
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_prio;
/* Add to the list of owned mutexes: */
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_ASSERT_NOT_OWNED(*mutex);
TAILQ_INSERT_TAIL(&curthread->mutexq,
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex), m_qe);
} else if ((*mutex)->m_owner == curthread)
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
ret = mutex_self_lock(*mutex);
else {
/*
* Join the queue of threads waiting to lock
* the mutex:
*/
mutex_queue_enq(*mutex, curthread);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Keep a pointer to the mutex this thread
* is waiting on:
*/
curthread->data.mutex = *mutex;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Clear any previous error: */
curthread->error = 0;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Unlock the mutex structure and schedule the
* next thread:
*/
_thread_kern_sched_state_unlock(PS_MUTEX_WAIT,
&(*mutex)->lock, __FILE__, __LINE__);
/* Lock the mutex structure again: */
_SPINLOCK(&(*mutex)->lock);
/*
* The threads priority may have changed while
* waiting for the mutex causing a ceiling
* violation.
*/
ret = curthread->error;
curthread->error = 0;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
break;
/* Trap invalid mutex types: */
default:
/* Return an invalid argument error: */
ret = EINVAL;
break;
}
/*
* Check to see if this thread was interrupted and
* is still in the mutex queue of waiting threads:
*/
if (curthread->interrupted != 0)
mutex_queue_remove(*mutex, curthread);
/* Unlock the mutex structure: */
_SPINUNLOCK(&(*mutex)->lock);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
* Undefer and handle pending signals, yielding if
* necessary:
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
*/
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_thread_kern_sig_undefer();
} while (((*mutex)->m_owner != curthread) && (ret == 0) &&
(curthread->interrupted == 0));
if (curthread->interrupted != 0 &&
curthread->continuation != NULL)
curthread->continuation((void *) curthread);
/* Return the completion status: */
return (ret);
}
int
__pthread_mutex_lock(pthread_mutex_t *mutex)
{
int ret = 0;
if (_thread_initial == NULL)
_thread_init();
if (mutex == NULL)
ret = EINVAL;
/*
* If the mutex is statically initialized, perform the dynamic
* initialization:
*/
else if ((*mutex != NULL) || ((ret = init_static(mutex)) == 0))
ret = mutex_lock_common(mutex);
return (ret);
}
int
_pthread_mutex_lock(pthread_mutex_t *mutex)
{
int ret = 0;
if (_thread_initial == NULL)
_thread_init();
if (mutex == NULL)
ret = EINVAL;
/*
* If the mutex is statically initialized, perform the dynamic
* initialization marking it private (delete safe):
*/
else if ((*mutex != NULL) || ((ret = init_static_private(mutex)) == 0))
ret = mutex_lock_common(mutex);
return (ret);
}
int
_pthread_mutex_unlock(pthread_mutex_t * mutex)
{
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
return (mutex_unlock_common(mutex, /* add reference */ 0));
}
int
_mutex_cv_unlock(pthread_mutex_t * mutex)
{
return (mutex_unlock_common(mutex, /* add reference */ 1));
}
int
_mutex_cv_lock(pthread_mutex_t * mutex)
{
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret;
if ((ret = _pthread_mutex_lock(mutex)) == 0)
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_refcount--;
return (ret);
}
static inline int
mutex_self_trylock(pthread_mutex_t mutex)
{
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret = 0;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
switch (mutex->m_type) {
/* case PTHREAD_MUTEX_DEFAULT: */
case PTHREAD_MUTEX_ERRORCHECK:
case PTHREAD_MUTEX_NORMAL:
/*
* POSIX specifies that mutexes should return EDEADLK if a
* recursive lock is detected.
*/
ret = EBUSY;
break;
case PTHREAD_MUTEX_RECURSIVE:
/* Increment the lock count: */
mutex->m_data.m_count++;
break;
default:
/* Trap invalid mutex types; */
ret = EINVAL;
}
return (ret);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
static inline int
mutex_self_lock(pthread_mutex_t mutex)
{
int ret = 0;
switch (mutex->m_type) {
/* case PTHREAD_MUTEX_DEFAULT: */
case PTHREAD_MUTEX_ERRORCHECK:
/*
* POSIX specifies that mutexes should return EDEADLK if a
* recursive lock is detected.
*/
ret = EDEADLK;
break;
case PTHREAD_MUTEX_NORMAL:
/*
* What SS2 define as a 'normal' mutex. Intentionally
* deadlock on attempts to get a lock you already own.
*/
_thread_kern_sched_state_unlock(PS_DEADLOCK,
&mutex->lock, __FILE__, __LINE__);
break;
case PTHREAD_MUTEX_RECURSIVE:
/* Increment the lock count: */
mutex->m_data.m_count++;
break;
default:
/* Trap invalid mutex types; */
ret = EINVAL;
}
return (ret);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
static inline int
mutex_unlock_common(pthread_mutex_t * mutex, int add_reference)
{
struct pthread *curthread = _get_curthread();
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
int ret = 0;
if (mutex == NULL || *mutex == NULL) {
ret = EINVAL;
} else {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
* Defer signals to protect the scheduling queues from
* access by the signal handler:
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
*/
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_thread_kern_sig_defer();
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Lock the mutex structure: */
_SPINLOCK(&(*mutex)->lock);
/* Process according to mutex type: */
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
switch ((*mutex)->m_protocol) {
/* Default POSIX mutex: */
case PTHREAD_PRIO_NONE:
/*
* Check if the running thread is not the owner of the
* mutex:
*/
if ((*mutex)->m_owner != curthread) {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Return an invalid argument error for no
* owner and a permission error otherwise:
*/
ret = (*mutex)->m_owner == NULL ? EINVAL : EPERM;
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
else if (((*mutex)->m_type == PTHREAD_MUTEX_RECURSIVE) &&
((*mutex)->m_data.m_count > 0)) {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Decrement the count: */
(*mutex)->m_data.m_count--;
} else {
/*
* Clear the count in case this is recursive
* mutex.
*/
(*mutex)->m_data.m_count = 0;
/* Remove the mutex from the threads queue. */
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_ASSERT_IS_OWNED(*mutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
TAILQ_REMOVE(&(*mutex)->m_owner->mutexq,
(*mutex), m_qe);
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_INIT_LINK(*mutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Get the next thread from the queue of
* threads waiting on the mutex:
*/
if (((*mutex)->m_owner =
mutex_queue_deq(*mutex)) != NULL) {
/* Make the new owner runnable: */
PTHREAD_NEW_STATE((*mutex)->m_owner,
PS_RUNNING);
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
/*
* Add the mutex to the threads list of
* owned mutexes:
*/
TAILQ_INSERT_TAIL(&(*mutex)->m_owner->mutexq,
(*mutex), m_qe);
/*
* The owner is no longer waiting for
* this mutex:
*/
(*mutex)->m_owner->data.mutex = NULL;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
}
break;
/* POSIX priority inheritence mutex: */
case PTHREAD_PRIO_INHERIT:
/*
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
* Check if the running thread is not the owner of the
* mutex:
*/
if ((*mutex)->m_owner != curthread) {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Return an invalid argument error for no
* owner and a permission error otherwise:
*/
ret = (*mutex)->m_owner == NULL ? EINVAL : EPERM;
}
else if (((*mutex)->m_type == PTHREAD_MUTEX_RECURSIVE) &&
((*mutex)->m_data.m_count > 0)) {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Decrement the count: */
(*mutex)->m_data.m_count--;
} else {
/*
* Clear the count in case this is recursive
* mutex.
*/
(*mutex)->m_data.m_count = 0;
/*
* Restore the threads inherited priority and
* recompute the active priority (being careful
* not to override changes in the threads base
* priority subsequent to locking the mutex).
*/
curthread->inherited_priority =
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_saved_prio;
curthread->active_priority =
MAX(curthread->inherited_priority,
curthread->base_priority);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* This thread now owns one less priority mutex.
*/
curthread->priority_mutex_count--;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Remove the mutex from the threads queue. */
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_ASSERT_IS_OWNED(*mutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
TAILQ_REMOVE(&(*mutex)->m_owner->mutexq,
(*mutex), m_qe);
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_INIT_LINK(*mutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Get the next thread from the queue of threads
* waiting on the mutex:
*/
if (((*mutex)->m_owner =
mutex_queue_deq(*mutex)) == NULL)
/* This mutex has no priority. */
(*mutex)->m_prio = 0;
else {
/*
* Track number of priority mutexes owned:
*/
(*mutex)->m_owner->priority_mutex_count++;
/*
* Add the mutex to the threads list
* of owned mutexes:
*/
TAILQ_INSERT_TAIL(&(*mutex)->m_owner->mutexq,
(*mutex), m_qe);
/*
* The owner is no longer waiting for
* this mutex:
*/
(*mutex)->m_owner->data.mutex = NULL;
/*
* Set the priority of the mutex. Since
* our waiting threads are in descending
* priority order, the priority of the
* mutex becomes the active priority of
* the thread we just dequeued.
*/
(*mutex)->m_prio =
(*mutex)->m_owner->active_priority;
/*
* Save the owning threads inherited
* priority:
*/
(*mutex)->m_saved_prio =
(*mutex)->m_owner->inherited_priority;
/*
* The owning threads inherited priority
* now becomes his active priority (the
* priority of the mutex).
*/
(*mutex)->m_owner->inherited_priority =
(*mutex)->m_prio;
/*
* Make the new owner runnable:
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
*/
PTHREAD_NEW_STATE((*mutex)->m_owner,
PS_RUNNING);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
}
break;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* POSIX priority ceiling mutex: */
case PTHREAD_PRIO_PROTECT:
/*
* Check if the running thread is not the owner of the
* mutex:
*/
if ((*mutex)->m_owner != curthread) {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Return an invalid argument error for no
* owner and a permission error otherwise:
*/
ret = (*mutex)->m_owner == NULL ? EINVAL : EPERM;
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
else if (((*mutex)->m_type == PTHREAD_MUTEX_RECURSIVE) &&
((*mutex)->m_data.m_count > 0)) {
/* Decrement the count: */
(*mutex)->m_data.m_count--;
} else {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Clear the count in case this is recursive
* mutex.
*/
(*mutex)->m_data.m_count = 0;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
* Restore the threads inherited priority and
* recompute the active priority (being careful
* not to override changes in the threads base
* priority subsequent to locking the mutex).
*/
curthread->inherited_priority =
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
(*mutex)->m_saved_prio;
curthread->active_priority =
MAX(curthread->inherited_priority,
curthread->base_priority);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* This thread now owns one less priority mutex.
*/
curthread->priority_mutex_count--;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/* Remove the mutex from the threads queue. */
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_MUTEX_ASSERT_IS_OWNED(*mutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
TAILQ_REMOVE(&(*mutex)->m_owner->mutexq,
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
(*mutex), m_qe);
_MUTEX_INIT_LINK(*mutex);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Enter a loop to find a waiting thread whose
* active priority will not cause a ceiling
* violation:
*/
while ((((*mutex)->m_owner =
mutex_queue_deq(*mutex)) != NULL) &&
((*mutex)->m_owner->active_priority >
(*mutex)->m_prio)) {
/*
* Either the mutex ceiling priority
* been lowered and/or this threads
* priority has been raised subsequent
* to this thread being queued on the
* waiting list.
*/
(*mutex)->m_owner->error = EINVAL;
PTHREAD_NEW_STATE((*mutex)->m_owner,
PS_RUNNING);
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
/*
* The thread is no longer waiting for
* this mutex:
*/
(*mutex)->m_owner->data.mutex = NULL;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
/* Check for a new owner: */
if ((*mutex)->m_owner != NULL) {
/*
* Track number of priority mutexes owned:
*/
(*mutex)->m_owner->priority_mutex_count++;
/*
* Add the mutex to the threads list
* of owned mutexes:
*/
TAILQ_INSERT_TAIL(&(*mutex)->m_owner->mutexq,
(*mutex), m_qe);
/*
* The owner is no longer waiting for
* this mutex:
*/
(*mutex)->m_owner->data.mutex = NULL;
/*
* Save the owning threads inherited
* priority:
*/
(*mutex)->m_saved_prio =
(*mutex)->m_owner->inherited_priority;
/*
* The owning thread inherits the
* ceiling priority of the mutex and
* executes at that priority:
*/
(*mutex)->m_owner->inherited_priority =
(*mutex)->m_prio;
(*mutex)->m_owner->active_priority =
(*mutex)->m_prio;
/*
* Make the new owner runnable:
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
*/
PTHREAD_NEW_STATE((*mutex)->m_owner,
PS_RUNNING);
}
}
break;
/* Trap invalid mutex types: */
default:
/* Return an invalid argument error: */
ret = EINVAL;
break;
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
if ((ret == 0) && (add_reference != 0)) {
/* Increment the reference count: */
(*mutex)->m_refcount++;
}
/* Unlock the mutex structure: */
_SPINUNLOCK(&(*mutex)->lock);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
* Undefer and handle pending signals, yielding if
* necessary:
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
*/
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
_thread_kern_sig_undefer();
}
/* Return the completion status: */
return (ret);
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
* This function is called when a change in base priority occurs for
* a thread that is holding or waiting for a priority protection or
* inheritence mutex. A change in a threads base priority can effect
* changes to active priorities of other threads and to the ordering
* of mutex locking by waiting threads.
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
*
* This must be called while thread scheduling is deferred.
*/
void
_mutex_notify_priochange(pthread_t pthread)
{
/* Adjust the priorites of any owned priority mutexes: */
if (pthread->priority_mutex_count > 0) {
/*
* Rescan the mutexes owned by this thread and correct
* their priorities to account for this threads change
* in priority. This has the side effect of changing
* the threads active priority.
*/
mutex_rescan_owned(pthread, /* rescan all owned */ NULL);
}
/*
* If this thread is waiting on a priority inheritence mutex,
* check for priority adjustments. A change in priority can
* also effect a ceiling violation(*) for a thread waiting on
* a priority protection mutex; we don't perform the check here
* as it is done in pthread_mutex_unlock.
*
* (*) It should be noted that a priority change to a thread
* _after_ taking and owning a priority ceiling mutex
* does not affect ownership of that mutex; the ceiling
* priority is only checked before mutex ownership occurs.
*/
if (pthread->state == PS_MUTEX_WAIT) {
/* Lock the mutex structure: */
_SPINLOCK(&pthread->data.mutex->lock);
/*
* Check to make sure this thread is still in the same state
* (the spinlock above can yield the CPU to another thread):
*/
if (pthread->state == PS_MUTEX_WAIT) {
/*
* Remove and reinsert this thread into the list of
* waiting threads to preserve decreasing priority
* order.
*/
mutex_queue_remove(pthread->data.mutex, pthread);
mutex_queue_enq(pthread->data.mutex, pthread);
if (pthread->data.mutex->m_protocol ==
PTHREAD_PRIO_INHERIT) {
/* Adjust priorities: */
mutex_priority_adjust(pthread->data.mutex);
}
}
/* Unlock the mutex structure: */
_SPINUNLOCK(&pthread->data.mutex->lock);
}
}
/*
* Called when a new thread is added to the mutex waiting queue or
* when a threads priority changes that is already in the mutex
* waiting queue.
*/
static void
mutex_priority_adjust(pthread_mutex_t mutex)
{
pthread_t pthread_next, pthread = mutex->m_owner;
int temp_prio;
pthread_mutex_t m = mutex;
/*
* Calculate the mutex priority as the maximum of the highest
* active priority of any waiting threads and the owning threads
* active priority(*).
*
* (*) Because the owning threads current active priority may
* reflect priority inherited from this mutex (and the mutex
* priority may have changed) we must recalculate the active
* priority based on the threads saved inherited priority
* and its base priority.
*/
pthread_next = TAILQ_FIRST(&m->m_queue); /* should never be NULL */
temp_prio = MAX(pthread_next->active_priority,
MAX(m->m_saved_prio, pthread->base_priority));
/* See if this mutex really needs adjusting: */
if (temp_prio == m->m_prio)
/* No need to propagate the priority: */
return;
/* Set new priority of the mutex: */
m->m_prio = temp_prio;
while (m != NULL) {
/*
* Save the threads priority before rescanning the
* owned mutexes:
*/
temp_prio = pthread->active_priority;
/*
* Fix the priorities for all the mutexes this thread has
* locked since taking this mutex. This also has a
* potential side-effect of changing the threads priority.
*/
mutex_rescan_owned(pthread, m);
/*
* If the thread is currently waiting on a mutex, check
* to see if the threads new priority has affected the
* priority of the mutex.
*/
if ((temp_prio != pthread->active_priority) &&
(pthread->state == PS_MUTEX_WAIT) &&
(pthread->data.mutex->m_protocol == PTHREAD_PRIO_INHERIT)) {
/* Grab the mutex this thread is waiting on: */
m = pthread->data.mutex;
/*
* The priority for this thread has changed. Remove
* and reinsert this thread into the list of waiting
* threads to preserve decreasing priority order.
*/
mutex_queue_remove(m, pthread);
mutex_queue_enq(m, pthread);
/* Grab the waiting thread with highest priority: */
pthread_next = TAILQ_FIRST(&m->m_queue);
/*
* Calculate the mutex priority as the maximum of the
* highest active priority of any waiting threads and
* the owning threads active priority.
*/
temp_prio = MAX(pthread_next->active_priority,
MAX(m->m_saved_prio, m->m_owner->base_priority));
if (temp_prio != m->m_prio) {
/*
* The priority needs to be propagated to the
* mutex this thread is waiting on and up to
* the owner of that mutex.
*/
m->m_prio = temp_prio;
pthread = m->m_owner;
}
else
/* We're done: */
m = NULL;
}
else
/* We're done: */
m = NULL;
}
}
static void
mutex_rescan_owned(pthread_t pthread, pthread_mutex_t mutex)
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
{
int active_prio, inherited_prio;
pthread_mutex_t m;
pthread_t pthread_next;
/*
* Start walking the mutexes the thread has taken since
* taking this mutex.
*/
if (mutex == NULL) {
/*
* A null mutex means start at the beginning of the owned
* mutex list.
*/
m = TAILQ_FIRST(&pthread->mutexq);
/* There is no inherited priority yet. */
inherited_prio = 0;
}
else {
/*
* The caller wants to start after a specific mutex. It
* is assumed that this mutex is a priority inheritence
* mutex and that its priority has been correctly
* calculated.
*/
m = TAILQ_NEXT(mutex, m_qe);
/* Start inheriting priority from the specified mutex. */
inherited_prio = mutex->m_prio;
}
active_prio = MAX(inherited_prio, pthread->base_priority);
while (m != NULL) {
/*
* We only want to deal with priority inheritence
* mutexes. This might be optimized by only placing
* priority inheritence mutexes into the owned mutex
* list, but it may prove to be useful having all
* owned mutexes in this list. Consider a thread
* exiting while holding mutexes...
*/
if (m->m_protocol == PTHREAD_PRIO_INHERIT) {
/*
* Fix the owners saved (inherited) priority to
* reflect the priority of the previous mutex.
*/
m->m_saved_prio = inherited_prio;
if ((pthread_next = TAILQ_FIRST(&m->m_queue)) != NULL)
/* Recalculate the priority of the mutex: */
m->m_prio = MAX(active_prio,
pthread_next->active_priority);
else
m->m_prio = active_prio;
/* Recalculate new inherited and active priorities: */
inherited_prio = m->m_prio;
active_prio = MAX(m->m_prio, pthread->base_priority);
}
/* Advance to the next mutex owned by this thread: */
m = TAILQ_NEXT(m, m_qe);
}
/*
* Fix the threads inherited priority and recalculate its
* active priority.
*/
pthread->inherited_priority = inherited_prio;
active_prio = MAX(inherited_prio, pthread->base_priority);
if (active_prio != pthread->active_priority) {
/*
* If this thread is in the priority queue, it must be
* removed and reinserted for its new priority.
*/
In the words of the author: 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>
1999-06-20 08:28:48 +00:00
if (pthread->flags & PTHREAD_FLAGS_IN_PRIOQ) {
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Remove the thread from the priority queue
* before changing its priority:
*/
PTHREAD_PRIOQ_REMOVE(pthread);
/*
* POSIX states that if the priority is being
* lowered, the thread must be inserted at the
* head of the queue for its priority if it owns
* any priority protection or inheritence mutexes.
*/
if ((active_prio < pthread->active_priority) &&
(pthread->priority_mutex_count > 0)) {
/* Set the new active priority. */
pthread->active_priority = active_prio;
PTHREAD_PRIOQ_INSERT_HEAD(pthread);
}
else {
/* Set the new active priority. */
pthread->active_priority = active_prio;
PTHREAD_PRIOQ_INSERT_TAIL(pthread);
}
}
else {
/* Set the new active priority. */
pthread->active_priority = active_prio;
}
}
}
void
_mutex_unlock_private(pthread_t pthread)
{
struct pthread_mutex *m, *m_next;
for (m = TAILQ_FIRST(&pthread->mutexq); m != NULL; m = m_next) {
m_next = TAILQ_NEXT(m, m_qe);
if ((m->m_flags & MUTEX_FLAGS_PRIVATE) != 0)
_pthread_mutex_unlock(&m);
}
}
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
void
_mutex_lock_backout(pthread_t pthread)
{
struct pthread_mutex *mutex;
/*
* Defer signals to protect the scheduling queues from
* access by the signal handler:
*/
_thread_kern_sig_defer();
if ((pthread->flags & PTHREAD_FLAGS_IN_MUTEXQ) != 0) {
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
mutex = pthread->data.mutex;
/* Lock the mutex structure: */
_SPINLOCK(&mutex->lock);
mutex_queue_remove(mutex, pthread);
/* This thread is no longer waiting for the mutex: */
pthread->data.mutex = NULL;
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
/* Unlock the mutex structure: */
_SPINUNLOCK(&mutex->lock);
}
/*
* Undefer and handle pending signals, yielding if
* necessary:
*/
_thread_kern_sig_undefer();
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* Dequeue a waiting thread from the head of a mutex queue in descending
* priority order.
*/
static inline pthread_t
mutex_queue_deq(pthread_mutex_t mutex)
{
pthread_t pthread;
while ((pthread = TAILQ_FIRST(&mutex->m_queue)) != NULL) {
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
TAILQ_REMOVE(&mutex->m_queue, pthread, sqe);
pthread->flags &= ~PTHREAD_FLAGS_IN_MUTEXQ;
/*
* Only exit the loop if the thread hasn't been
* cancelled.
*/
if (pthread->interrupted == 0)
break;
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
return (pthread);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
/*
* Remove a waiting thread from a mutex queue in descending priority order.
*/
static inline void
mutex_queue_remove(pthread_mutex_t mutex, pthread_t pthread)
{
if ((pthread->flags & PTHREAD_FLAGS_IN_MUTEXQ) != 0) {
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
TAILQ_REMOVE(&mutex->m_queue, pthread, sqe);
pthread->flags &= ~PTHREAD_FLAGS_IN_MUTEXQ;
}
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
/*
* Enqueue a waiting thread to a queue in descending priority order.
*/
static inline void
mutex_queue_enq(pthread_mutex_t mutex, pthread_t pthread)
{
pthread_t tid = TAILQ_LAST(&mutex->m_queue, mutex_head);
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
PTHREAD_ASSERT_NOT_IN_SYNCQ(pthread);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
/*
* For the common case of all threads having equal priority,
* we perform a quick check against the priority of the thread
* at the tail of the queue.
*/
if ((tid == NULL) || (pthread->active_priority <= tid->active_priority))
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
TAILQ_INSERT_TAIL(&mutex->m_queue, pthread, sqe);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
else {
tid = TAILQ_FIRST(&mutex->m_queue);
while (pthread->active_priority <= tid->active_priority)
Implement zero system call thread switching. Performance of 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
2000-10-13 22:12:32 +00:00
tid = TAILQ_NEXT(tid, sqe);
TAILQ_INSERT_BEFORE(tid, pthread, sqe);
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}
pthread->flags |= PTHREAD_FLAGS_IN_MUTEXQ;
[ The author's description... ] 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.
1999-03-23 05:07:56 +00:00
}