freebsd-dev/lib/libc_r/uthread/uthread_kern.c

1139 lines
29 KiB
C
Raw Normal View History

/*
* Copyright (c) 1995-1998 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. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by John Birrell.
* 4. 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
* $FreeBSD$
1997-04-01 22:51:48 +00:00
*
*/
#include <errno.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
#include <poll.h>
#include <stdlib.h>
#include <stdarg.h>
#include <string.h>
#include <unistd.h>
#include <setjmp.h>
#include <sys/param.h>
#include <sys/types.h>
#include <sys/signalvar.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/socket.h>
#include <sys/uio.h>
#include <sys/syscall.h>
#include <fcntl.h>
#include <pthread.h>
#include "pthread_private.h"
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
/* #define DEBUG_THREAD_KERN */
#ifdef DEBUG_THREAD_KERN
#define DBG_MSG stdout_debug
#else
#define DBG_MSG(x...)
#endif
/* Static function prototype definitions: */
static void
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
thread_kern_poll(int wait_reqd);
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
static void
dequeue_signals(void);
[ 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 void
thread_run_switch_hook(pthread_t thread_out, pthread_t thread_in);
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
/* Static variables: */
static int last_tick = 0;
static int called_from_handler = 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
/*
* This is called when a signal handler finishes and wants to
* return to a previous frame.
*/
void
_thread_kern_sched_frame(struct pthread_signal_frame *psf)
{
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
/*
* Flag the pthread kernel as executing scheduler code
* to avoid a signal from interrupting this execution and
* corrupting the (soon-to-be) current frame.
*/
_thread_kern_in_sched = 1;
/* Restore the signal frame: */
_thread_sigframe_restore(curthread, psf);
/* The signal mask was restored; check for any pending signals: */
curthread->check_pending = 1;
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
/* Switch to the thread scheduler: */
___longjmp(_thread_kern_sched_jb, 1);
}
void
_thread_kern_sched(ucontext_t *ucp)
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
{
struct pthread *curthread = _get_curthread();
/*
* Flag the pthread kernel as executing scheduler code
* to avoid a scheduler signal from interrupting this
* execution and calling the scheduler again.
*/
_thread_kern_in_sched = 1;
/* Check if this function was called from the signal handler: */
if (ucp != NULL) {
called_from_handler = 1;
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
DBG_MSG("Entering scheduler due to signal\n");
}
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
/* Save the state of the current thread: */
if (_setjmp(curthread->ctx.jb) != 0) {
DBG_MSG("Returned from ___longjmp, thread %p\n",
curthread);
/*
* This point is reached when a longjmp() is called
* to restore the state of a thread.
*
* This is the normal way out of the scheduler.
*/
_thread_kern_in_sched = 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
if (curthread->sig_defer_count == 0) {
if (((curthread->cancelflags &
PTHREAD_AT_CANCEL_POINT) == 0) &&
((curthread->cancelflags &
PTHREAD_CANCEL_ASYNCHRONOUS) != 0))
/*
* Cancellations override signals.
*
* Stick a cancellation point at the
* start of each async-cancellable
* thread's resumption.
*
* We allow threads woken at cancel
* points to do their own checks.
*/
pthread_testcancel();
}
if (_sched_switch_hook != NULL) {
/* Run the installed switch hook: */
thread_run_switch_hook(_last_user_thread, curthread);
}
if (ucp == 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
return;
else {
/*
* Set the process signal mask in the context; it
* could have changed by the handler.
*/
ucp->uc_sigmask = _process_sigmask;
/* Resume the interrupted thread: */
__sys_sigreturn(ucp);
}
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
}
/* Switch to the thread scheduler: */
___longjmp(_thread_kern_sched_jb, 1);
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
}
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
_thread_kern_sched_sig(void)
{
struct pthread *curthread = _get_curthread();
curthread->check_pending = 1;
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
_thread_kern_sched(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
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
_thread_kern_scheduler(void)
{
struct timespec ts;
struct timeval tv;
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
pthread_t pthread, pthread_h;
unsigned int current_tick;
int add_to_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
/* If the currently running thread is a user thread, save it: */
if ((curthread->flags & PTHREAD_FLAGS_PRIVATE) == 0)
_last_user_thread = 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
if (called_from_handler != 0) {
called_from_handler = 0;
/*
* We were called from a signal handler; restore the process
* signal mask.
*/
if (__sys_sigprocmask(SIG_SETMASK,
&_process_sigmask, NULL) != 0)
PANIC("Unable to restore process mask after signal");
}
/*
[ 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 scheduling loop that finds the next thread that is
* ready to run. This loop completes when there are no more threads
* in the global list or when a thread has its state restored by
* either a sigreturn (if the state was saved as a sigcontext) or a
* longjmp (if the state was saved by a setjmp).
*/
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
while (!(TAILQ_EMPTY(&_thread_list))) {
/* Get the current time of day: */
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
GET_CURRENT_TOD(tv);
TIMEVAL_TO_TIMESPEC(&tv, &ts);
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
current_tick = _sched_ticks;
/*
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
* Protect the scheduling queues from access by the signal
* handler.
*/
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
_queue_signals = 1;
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
add_to_prioq = 0;
if (curthread != &_thread_kern_thread) {
[ 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 thread no longer needs to yield the CPU.
*/
curthread->yield_on_sig_undefer = 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 (curthread->state != PS_RUNNING) {
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
/*
* Save the current time as the time that the
* thread became inactive:
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
*/
curthread->last_inactive = (long)current_tick;
if (curthread->last_inactive <
curthread->last_active) {
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
/* Account for a rollover: */
curthread->last_inactive =+
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
UINT_MAX + 1;
}
}
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
/*
* Place the currently running thread into the
* appropriate queue(s).
*/
switch (curthread->state) {
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
case PS_DEAD:
case PS_STATE_MAX: /* to silence -Wall */
case PS_SUSPENDED:
[ 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
/*
* Dead and suspended threads are not placed
* in any queue:
[ 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
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
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
case 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
* Runnable threads can't be placed in the
* priority queue until after waiting threads
* are polled (to preserve round-robin
* scheduling).
*/
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
add_to_prioq = 1;
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
break;
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
/*
* States which do not depend on file descriptor I/O
* operations or timeouts:
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
*/
case PS_DEADLOCK:
case PS_FDLR_WAIT:
case PS_FDLW_WAIT:
case PS_FILE_WAIT:
case PS_JOIN:
case PS_MUTEX_WAIT:
case PS_SIGSUSPEND:
case PS_SIGTHREAD:
case PS_SIGWAIT:
case PS_WAIT_WAIT:
/* No timeouts for these states: */
curthread->wakeup_time.tv_sec = -1;
curthread->wakeup_time.tv_nsec = -1;
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
/* Restart the time slice: */
curthread->slice_usec = -1;
[ 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
/* Insert into the waiting queue: */
PTHREAD_WAITQ_INSERT(curthread);
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
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
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
/* States which can timeout: */
case PS_COND_WAIT:
case PS_SLEEP_WAIT:
/* Restart the time slice: */
curthread->slice_usec = -1;
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
/* Insert into the waiting queue: */
PTHREAD_WAITQ_INSERT(curthread);
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
break;
/* States that require periodic work: */
case PS_SPINBLOCK:
/* No timeouts for this state: */
curthread->wakeup_time.tv_sec = -1;
curthread->wakeup_time.tv_nsec = -1;
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
/* Increment spinblock count: */
_spinblock_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
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
/* FALLTHROUGH */
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
case PS_FDR_WAIT:
case PS_FDW_WAIT:
case PS_POLL_WAIT:
case PS_SELECT_WAIT:
/* Restart the time slice: */
curthread->slice_usec = -1;
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
/* Insert into the waiting queue: */
PTHREAD_WAITQ_INSERT(curthread);
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
/* Insert into the work queue: */
PTHREAD_WORKQ_INSERT(curthread);
break;
}
/*
* Are there pending signals for this thread?
*
* This check has to be performed after the thread
* has been placed in the queue(s) appropriate for
* its state. The process of adding pending signals
* can change a threads state, which in turn will
* attempt to add or remove the thread from any
* scheduling queue to which it belongs.
*/
if (curthread->check_pending != 0) {
curthread->check_pending = 0;
_thread_sig_check_pending(curthread);
}
}
/*
* Avoid polling file descriptors if there are none
* waiting:
*/
if (TAILQ_EMPTY(&_workq) != 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
/*
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
* Poll file descriptors only if a new scheduling signal
* has occurred or if we have no more runnable threads.
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 if (((current_tick = _sched_ticks) != last_tick) ||
((curthread->state != PS_RUNNING) &&
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_PRIOQ_FIRST() == NULL))) {
/* Unprotect the scheduling queues: */
_queue_signals = 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
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
/*
* Poll file descriptors to update the state of threads
* waiting on file I/O where data may be available:
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
*/
thread_kern_poll(0);
/* Protect the scheduling queues: */
_queue_signals = 1;
}
last_tick = current_tick;
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
/*
* Wake up threads that have timedout. This has to be
* done after polling in case a thread does a poll or
* select with zero time.
*/
PTHREAD_WAITQ_SETACTIVE();
while (((pthread = TAILQ_FIRST(&_waitingq)) != NULL) &&
(pthread->wakeup_time.tv_sec != -1) &&
(((pthread->wakeup_time.tv_sec == 0) &&
(pthread->wakeup_time.tv_nsec == 0)) ||
(pthread->wakeup_time.tv_sec < ts.tv_sec) ||
((pthread->wakeup_time.tv_sec == ts.tv_sec) &&
(pthread->wakeup_time.tv_nsec <= ts.tv_nsec)))) {
switch (pthread->state) {
case PS_POLL_WAIT:
case PS_SELECT_WAIT:
/* Return zero file descriptors ready: */
pthread->data.poll_data->nfds = 0;
/* FALLTHROUGH */
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
default:
/*
* Remove this thread from the waiting queue
* (and work queue if necessary) and place it
* in the ready queue.
*/
PTHREAD_WAITQ_CLEARACTIVE();
if (pthread->flags & PTHREAD_FLAGS_IN_WORKQ)
PTHREAD_WORKQ_REMOVE(pthread);
PTHREAD_NEW_STATE(pthread, PS_RUNNING);
PTHREAD_WAITQ_SETACTIVE();
break;
}
/*
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
* Flag the timeout in the thread 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
pthread->timeout = 1;
}
PTHREAD_WAITQ_CLEARACTIVE();
/*
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
* Check to see if the current thread needs to be added
* to the priority 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
*/
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
if (add_to_prioq != 0) {
/*
* Save the current time as the time that the
* thread became inactive:
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
*/
current_tick = _sched_ticks;
curthread->last_inactive = (long)current_tick;
if (curthread->last_inactive <
curthread->last_active) {
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
/* Account for a rollover: */
curthread->last_inactive =+ UINT_MAX + 1;
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
}
if ((curthread->slice_usec != -1) &&
(curthread->attr.sched_policy != SCHED_FIFO)) {
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
/*
* Accumulate the number of microseconds for
* which the current thread has run:
*/
curthread->slice_usec +=
(curthread->last_inactive -
curthread->last_active) *
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
(long)_clock_res_usec;
/* Check for time quantum exceeded: */
if (curthread->slice_usec > TIMESLICE_USEC)
curthread->slice_usec = -1;
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
}
if (curthread->slice_usec == -1) {
/*
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 exceeded its time
* quantum or it yielded the CPU;
* place it at the tail of the
* queue for its priority.
*/
PTHREAD_PRIOQ_INSERT_TAIL(curthread);
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 {
/*
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 hasn't exceeded its
* interval. Place it at the head
* of the queue for its priority.
*/
PTHREAD_PRIOQ_INSERT_HEAD(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
* Get the highest priority thread in the ready 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
pthread_h = PTHREAD_PRIOQ_FIRST();
/* Check if there are no threads ready to run: */
[ 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 (pthread_h == NULL) {
/*
* Lock the pthread kernel by changing the pointer to
* the running thread to point to the global kernel
* thread structure:
*/
_set_curthread(&_thread_kern_thread);
curthread = &_thread_kern_thread;
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
DBG_MSG("No runnable threads, using kernel thread %p\n",
curthread);
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
/* Unprotect the scheduling queues: */
_queue_signals = 0;
/*
* There are no threads ready to run, so wait until
* something happens that changes this condition:
*/
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
thread_kern_poll(1);
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
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
/*
* This process' usage will likely be very small
* while waiting in a poll. Since the scheduling
* clock is based on the profiling timer, it is
* unlikely that the profiling timer will fire
* and update the time of day. To account for this,
* get the time of day after polling with a timeout.
*/
gettimeofday((struct timeval *) &_sched_tod, NULL);
/* Check once more for a runnable thread: */
_queue_signals = 1;
pthread_h = PTHREAD_PRIOQ_FIRST();
_queue_signals = 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
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
if (pthread_h != NULL) {
/* Remove the thread from the ready queue: */
PTHREAD_PRIOQ_REMOVE(pthread_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
/* Unprotect the scheduling queues: */
_queue_signals = 0;
/*
* Check for signals queued while the scheduling
* queues were protected:
*/
while (_sigq_check_reqd != 0) {
/* Clear before handling queued signals: */
_sigq_check_reqd = 0;
/* Protect the scheduling queues again: */
_queue_signals = 1;
dequeue_signals();
/*
* Check for a higher priority thread that
* became runnable due to signal handling.
*/
if (((pthread = PTHREAD_PRIOQ_FIRST()) != NULL) &&
(pthread->active_priority > pthread_h->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
/* Remove the thread from the ready queue: */
PTHREAD_PRIOQ_REMOVE(pthread);
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
/*
* Insert the lower priority thread
* at the head of its priority list:
*/
PTHREAD_PRIOQ_INSERT_HEAD(pthread_h);
/* There's a new thread in town: */
pthread_h = pthread;
}
/* Unprotect the scheduling queues: */
_queue_signals = 0;
}
/* Make the selected thread the current thread: */
_set_curthread(pthread_h);
curthread = pthread_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
/*
* Save the current time as the time that the thread
* became active:
*/
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
current_tick = _sched_ticks;
curthread->last_active = (long) current_tick;
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
/*
* Check if this thread is running for the first time
* or running again after using its full time slice
* allocation:
*/
if (curthread->slice_usec == -1) {
/* Reset the accumulated time slice period: */
curthread->slice_usec = 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
* If we had a context switch, run any
* installed switch hooks.
*/
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
if ((_sched_switch_hook != NULL) &&
(_last_user_thread != 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
thread_run_switch_hook(_last_user_thread,
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
* Continue the thread at its current frame:
*/
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
#if NOT_YET
_setcontext(&curthread->ctx.uc);
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
#else
___longjmp(curthread->ctx.jb, 1);
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
#endif
/* This point should not be reached. */
PANIC("Thread has returned from sigreturn or longjmp");
}
}
/* There are no more threads, so exit this process: */
exit(0);
}
void
_thread_kern_sched_state(enum pthread_state state, char *fname, int lineno)
{
struct pthread *curthread = _get_curthread();
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
/*
* Flag the pthread kernel as executing scheduler code
* to avoid a scheduler signal from interrupting this
* execution and calling the scheduler again.
*/
_thread_kern_in_sched = 1;
/*
* Prevent the signal handler from fiddling with this thread
* before its state is set and is placed into the proper queue.
*/
_queue_signals = 1;
/* Change the state of the current thread: */
curthread->state = state;
curthread->fname = fname;
curthread->lineno = lineno;
/* Schedule the next thread that is ready: */
_thread_kern_sched(NULL);
}
void
_thread_kern_sched_state_unlock(enum pthread_state state,
spinlock_t *lock, char *fname, int lineno)
{
struct pthread *curthread = _get_curthread();
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
/*
* Flag the pthread kernel as executing scheduler code
* to avoid a scheduler signal from interrupting this
* execution and calling the scheduler again.
*/
_thread_kern_in_sched = 1;
/*
* Prevent the signal handler from fiddling with this thread
* before its state is set and it is placed into the proper
* queue(s).
*/
_queue_signals = 1;
/* Change the state of the current thread: */
curthread->state = state;
curthread->fname = fname;
curthread->lineno = lineno;
_SPINUNLOCK(lock);
/* Schedule the next thread that is ready: */
_thread_kern_sched(NULL);
}
static void
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
thread_kern_poll(int wait_reqd)
{
int count = 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
int i, found;
int kern_pipe_added = 0;
int nfds = 0;
int timeout_ms = 0;
struct pthread *pthread;
struct timespec ts;
struct timeval tv;
/* Check if the caller wants to wait: */
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 (wait_reqd == 0) {
timeout_ms = 0;
}
else {
/* Get the current time of day: */
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
GET_CURRENT_TOD(tv);
TIMEVAL_TO_TIMESPEC(&tv, &ts);
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
_queue_signals = 1;
pthread = TAILQ_FIRST(&_waitingq);
_queue_signals = 0;
if ((pthread == NULL) || (pthread->wakeup_time.tv_sec == -1)) {
/*
* Either there are no threads in the waiting queue,
* or there are no threads that can timeout.
*/
timeout_ms = INFTIM;
}
else if (pthread->wakeup_time.tv_sec - ts.tv_sec > 60000)
/* Limit maximum timeout to prevent rollover. */
timeout_ms = 60000;
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 {
/*
* Calculate the time left for the next thread to
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
* timeout:
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
*/
timeout_ms = ((pthread->wakeup_time.tv_sec - ts.tv_sec) *
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
1000) + ((pthread->wakeup_time.tv_nsec - ts.tv_nsec) /
1000000);
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
/*
* Don't allow negative timeouts:
*/
if (timeout_ms < 0)
timeout_ms = 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
/* Protect the scheduling queues: */
_queue_signals = 1;
/*
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
* Check to see if the signal queue needs to be walked to look
* for threads awoken by a signal while in the scheduler.
*/
if (_sigq_check_reqd != 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
/* Reset flag before handling queued signals: */
_sigq_check_reqd = 0;
dequeue_signals();
}
/*
* Check for a thread that became runnable due to a signal:
*/
if (PTHREAD_PRIOQ_FIRST() != NULL) {
1997-04-01 22:51:48 +00:00
/*
* Since there is at least one runnable thread,
* disable the wait.
1997-04-01 22:51:48 +00:00
*/
timeout_ms = 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
}
[ 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
/*
* Form the poll table:
*/
nfds = 0;
if (timeout_ms != 0) {
/* Add the kernel pipe to the poll table: */
_thread_pfd_table[nfds].fd = _thread_kern_pipe[0];
_thread_pfd_table[nfds].events = POLLRDNORM;
_thread_pfd_table[nfds].revents = 0;
nfds++;
kern_pipe_added = 1;
}
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
PTHREAD_WAITQ_SETACTIVE();
TAILQ_FOREACH(pthread, &_workq, qe) {
switch (pthread->state) {
case PS_SPINBLOCK:
/*
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 lock is available, let the thread run.
*/
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->data.spinlock->access_lock == 0) {
PTHREAD_WAITQ_CLEARACTIVE();
PTHREAD_WORKQ_REMOVE(pthread);
PTHREAD_NEW_STATE(pthread,PS_RUNNING);
PTHREAD_WAITQ_SETACTIVE();
/* One less thread in a spinblock state: */
_spinblock_count--;
/*
* Since there is at least one runnable
* thread, disable the wait.
*/
timeout_ms = 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
break;
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
/* File descriptor read wait: */
case PS_FDR_WAIT:
/* Limit number of polled files to table size: */
if (nfds < _thread_dtablesize) {
_thread_pfd_table[nfds].events = POLLRDNORM;
_thread_pfd_table[nfds].fd = pthread->data.fd.fd;
nfds++;
}
break;
1997-04-01 22:51:48 +00:00
/* File descriptor write wait: */
case PS_FDW_WAIT:
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
/* Limit number of polled files to table size: */
if (nfds < _thread_dtablesize) {
_thread_pfd_table[nfds].events = POLLWRNORM;
_thread_pfd_table[nfds].fd = pthread->data.fd.fd;
nfds++;
}
break;
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
/* File descriptor poll or select wait: */
case PS_POLL_WAIT:
case PS_SELECT_WAIT:
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
/* Limit number of polled files to table size: */
if (pthread->data.poll_data->nfds + nfds <
_thread_dtablesize) {
for (i = 0; i < pthread->data.poll_data->nfds; i++) {
_thread_pfd_table[nfds + i].fd =
pthread->data.poll_data->fds[i].fd;
_thread_pfd_table[nfds + i].events =
pthread->data.poll_data->fds[i].events;
}
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
nfds += pthread->data.poll_data->nfds;
}
break;
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
/* Other states do not depend on file I/O. */
default:
break;
}
}
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
PTHREAD_WAITQ_CLEARACTIVE();
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
/*
* Wait for a file descriptor to be ready for read, write, or
* an exception, or a timeout to occur:
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
*/
count = __sys_poll(_thread_pfd_table, nfds, timeout_ms);
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 (kern_pipe_added != 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
* Remove the pthread kernel pipe file descriptor
* from the pollfd table:
*/
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
nfds = 1;
else
nfds = 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
/*
* Check if it is possible that there are bytes in the kernel
* read pipe waiting to be read:
*/
if (count < 0 || ((kern_pipe_added != 0) &&
(_thread_pfd_table[0].revents & POLLRDNORM))) {
/*
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 kernel read pipe was included in the
* count:
*/
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 (count > 0) {
/* Decrement the count of file descriptors: */
count--;
}
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 (_sigq_check_reqd != 0) {
/* Reset flag before handling signals: */
_sigq_check_reqd = 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
dequeue_signals();
}
}
/*
* Check if any file descriptors are ready:
*/
if (count > 0) {
/*
* Enter a loop to look for threads waiting on file
* descriptors that are flagged as available by the
* _poll syscall:
*/
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
PTHREAD_WAITQ_SETACTIVE();
TAILQ_FOREACH(pthread, &_workq, qe) {
switch (pthread->state) {
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
case PS_SPINBLOCK:
[ 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
* If the lock is available, let the thread run.
[ 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
if (pthread->data.spinlock->access_lock == 0) {
PTHREAD_WAITQ_CLEARACTIVE();
PTHREAD_WORKQ_REMOVE(pthread);
PTHREAD_NEW_STATE(pthread,PS_RUNNING);
PTHREAD_WAITQ_SETACTIVE();
[ 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
* One less thread in a spinblock state:
*/
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
_spinblock_count--;
}
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
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
/* File descriptor read wait: */
case PS_FDR_WAIT:
if ((nfds < _thread_dtablesize) &&
(_thread_pfd_table[nfds].revents
& (POLLRDNORM|POLLERR|POLLHUP|POLLNVAL))
!= 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
PTHREAD_WAITQ_CLEARACTIVE();
PTHREAD_WORKQ_REMOVE(pthread);
PTHREAD_NEW_STATE(pthread,PS_RUNNING);
PTHREAD_WAITQ_SETACTIVE();
}
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
nfds++;
break;
1997-04-01 22:51:48 +00:00
/* File descriptor write wait: */
case PS_FDW_WAIT:
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 ((nfds < _thread_dtablesize) &&
(_thread_pfd_table[nfds].revents
& (POLLWRNORM|POLLERR|POLLHUP|POLLNVAL))
!= 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
PTHREAD_WAITQ_CLEARACTIVE();
PTHREAD_WORKQ_REMOVE(pthread);
PTHREAD_NEW_STATE(pthread,PS_RUNNING);
PTHREAD_WAITQ_SETACTIVE();
}
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
nfds++;
break;
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
/* File descriptor poll or select wait: */
case PS_POLL_WAIT:
case PS_SELECT_WAIT:
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->data.poll_data->nfds + nfds <
_thread_dtablesize) {
/*
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
* Enter a loop looking for I/O
* readiness:
*/
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
found = 0;
for (i = 0; i < pthread->data.poll_data->nfds; i++) {
if (_thread_pfd_table[nfds + i].revents != 0) {
pthread->data.poll_data->fds[i].revents =
_thread_pfd_table[nfds + i].revents;
found++;
}
}
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
/* Increment before destroying: */
nfds += pthread->data.poll_data->nfds;
if (found != 0) {
pthread->data.poll_data->nfds = found;
PTHREAD_WAITQ_CLEARACTIVE();
PTHREAD_WORKQ_REMOVE(pthread);
PTHREAD_NEW_STATE(pthread,PS_RUNNING);
PTHREAD_WAITQ_SETACTIVE();
}
}
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
nfds += pthread->data.poll_data->nfds;
break;
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
/* Other states do not depend on file I/O. */
default:
break;
}
}
PTHREAD_WAITQ_CLEARACTIVE();
}
else if (_spinblock_count != 0) {
/*
* Enter a loop to look for threads waiting on a spinlock
* that is now available.
*/
PTHREAD_WAITQ_SETACTIVE();
TAILQ_FOREACH(pthread, &_workq, qe) {
if (pthread->state == PS_SPINBLOCK) {
/*
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 lock is available, let the thread run.
*/
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->data.spinlock->access_lock == 0) {
PTHREAD_WAITQ_CLEARACTIVE();
PTHREAD_WORKQ_REMOVE(pthread);
PTHREAD_NEW_STATE(pthread,PS_RUNNING);
PTHREAD_WAITQ_SETACTIVE();
[ 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
* One less thread in a spinblock state:
[ 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
_spinblock_count--;
}
}
}
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
PTHREAD_WAITQ_CLEARACTIVE();
}
/* Unprotect the scheduling queues: */
_queue_signals = 0;
while (_sigq_check_reqd != 0) {
/* Handle queued signals: */
_sigq_check_reqd = 0;
/* Protect the scheduling queues: */
_queue_signals = 1;
dequeue_signals();
/* Unprotect the scheduling queues: */
_queue_signals = 0;
}
}
void
_thread_kern_set_timeout(const struct timespec * timeout)
{
struct pthread *curthread = _get_curthread();
struct timespec current_time;
struct timeval tv;
/* Reset the timeout flag for the running thread: */
curthread->timeout = 0;
/* Check if the thread is to wait forever: */
if (timeout == NULL) {
/*
* Set the wakeup time to something that can be recognised as
* different to an actual time of day:
*/
curthread->wakeup_time.tv_sec = -1;
curthread->wakeup_time.tv_nsec = -1;
}
/* Check if no waiting is required: */
else if (timeout->tv_sec == 0 && timeout->tv_nsec == 0) {
/* Set the wake up time to 'immediately': */
curthread->wakeup_time.tv_sec = 0;
curthread->wakeup_time.tv_nsec = 0;
} else {
/* Get the current time: */
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
GET_CURRENT_TOD(tv);
TIMEVAL_TO_TIMESPEC(&tv, &current_time);
/* Calculate the time for the current thread to wake up: */
curthread->wakeup_time.tv_sec = current_time.tv_sec + timeout->tv_sec;
curthread->wakeup_time.tv_nsec = current_time.tv_nsec + timeout->tv_nsec;
/* Check if the nanosecond field needs to wrap: */
if (curthread->wakeup_time.tv_nsec >= 1000000000) {
/* Wrap the nanosecond field: */
curthread->wakeup_time.tv_sec += 1;
curthread->wakeup_time.tv_nsec -= 1000000000;
}
}
}
[ 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
void
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(void)
[ 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
{
struct pthread *curthread = _get_curthread();
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
/* Allow signal deferral to be recursive. */
curthread->sig_defer_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
}
void
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(void)
[ 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
{
struct pthread *curthread = _get_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
/*
* Perform checks to yield only if we are about to undefer
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
* signals.
[ 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->sig_defer_count > 1) {
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
/* Decrement the signal deferral count. */
curthread->sig_defer_count--;
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 if (curthread->sig_defer_count == 1) {
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
/* Reenable signals: */
curthread->sig_defer_count = 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
[ 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
* Check if there are queued signals:
[ 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
*/
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
if (_sigq_check_reqd != 0)
_thread_kern_sched(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
/*
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
* Check for asynchronous cancellation before delivering any
* pending signals:
*/
if (((curthread->cancelflags & PTHREAD_AT_CANCEL_POINT) == 0) &&
((curthread->cancelflags & PTHREAD_CANCEL_ASYNCHRONOUS) != 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
pthread_testcancel();
[ 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
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
/*
* If there are pending signals or this thread has
* to yield the CPU, call the kernel scheduler:
*
* XXX - Come back and revisit the pending signal problem
*/
if ((curthread->yield_on_sig_undefer != 0) ||
SIGNOTEMPTY(curthread->sigpend)) {
curthread->yield_on_sig_undefer = 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
_thread_kern_sched(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
}
}
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 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
static void
dequeue_signals(void)
{
char bufr[128];
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 num;
[ 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
/*
* Enter a loop to clear the pthread kernel pipe:
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
*/
while (((num = __sys_read(_thread_kern_pipe[0], bufr,
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
sizeof(bufr))) > 0) || (num == -1 && errno == EINTR)) {
}
if ((num < 0) && (errno != EAGAIN)) {
/*
* The only error we should expect is if there is
* no data to read.
*/
PANIC("Unable to read from thread kernel pipe");
[ 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
}
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
/* Handle any pending signals: */
_thread_sig_handle_pending();
[ 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 void
thread_run_switch_hook(pthread_t thread_out, pthread_t thread_in)
{
pthread_t tid_out = thread_out;
pthread_t tid_in = thread_in;
if ((tid_out != NULL) &&
(tid_out->flags & PTHREAD_FLAGS_PRIVATE) != 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
tid_out = NULL;
if ((tid_in != NULL) &&
(tid_in->flags & PTHREAD_FLAGS_PRIVATE) != 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
tid_in = NULL;
if ((_sched_switch_hook != NULL) && (tid_out != tid_in)) {
/* Run the scheduler switch hook: */
_sched_switch_hook(tid_out, tid_in);
}
}
struct pthread *
_get_curthread(void)
{
if (_thread_initial == NULL)
_thread_init();
return (_thread_run);
}
void
_set_curthread(struct pthread *newthread)
{
_thread_run = newthread;
}